1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 
  22 /*
  23  * Copyright (c) 1991, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Copyright (c) 1990 Mentat Inc.
  25  * Copyright (c) 2017 OmniTI Computer Consulting, Inc. All rights reserved.
  26  * Copyright (c) 2016 by Delphix. All rights reserved.
  27  * Copyright (c) 2019 Joyent, Inc. All rights reserved.
  28  */
  29 
  30 #include <sys/types.h>
  31 #include <sys/stream.h>
  32 #include <sys/dlpi.h>
  33 #include <sys/stropts.h>
  34 #include <sys/sysmacros.h>
  35 #include <sys/strsubr.h>
  36 #include <sys/strlog.h>
  37 #include <sys/strsun.h>
  38 #include <sys/zone.h>
  39 #define _SUN_TPI_VERSION 2
  40 #include <sys/tihdr.h>
  41 #include <sys/xti_inet.h>
  42 #include <sys/ddi.h>
  43 #include <sys/suntpi.h>
  44 #include <sys/cmn_err.h>
  45 #include <sys/debug.h>
  46 #include <sys/kobj.h>
  47 #include <sys/modctl.h>
  48 #include <sys/atomic.h>
  49 #include <sys/policy.h>
  50 #include <sys/priv.h>
  51 #include <sys/taskq.h>
  52 
  53 #include <sys/systm.h>
  54 #include <sys/param.h>
  55 #include <sys/kmem.h>
  56 #include <sys/sdt.h>
  57 #include <sys/socket.h>
  58 #include <sys/vtrace.h>
  59 #include <sys/isa_defs.h>
  60 #include <sys/mac.h>
  61 #include <net/if.h>
  62 #include <net/if_arp.h>
  63 #include <net/route.h>
  64 #include <sys/sockio.h>
  65 #include <netinet/in.h>
  66 #include <net/if_dl.h>
  67 
  68 #include <inet/common.h>
  69 #include <inet/mi.h>
  70 #include <inet/mib2.h>
  71 #include <inet/nd.h>
  72 #include <inet/arp.h>
  73 #include <inet/snmpcom.h>
  74 #include <inet/optcom.h>
  75 #include <inet/kstatcom.h>
  76 
  77 #include <netinet/igmp_var.h>
  78 #include <netinet/ip6.h>
  79 #include <netinet/icmp6.h>
  80 #include <netinet/sctp.h>
  81 
  82 #include <inet/ip.h>
  83 #include <inet/ip_impl.h>
  84 #include <inet/ip6.h>
  85 #include <inet/ip6_asp.h>
  86 #include <inet/tcp.h>
  87 #include <inet/tcp_impl.h>
  88 #include <inet/ip_multi.h>
  89 #include <inet/ip_if.h>
  90 #include <inet/ip_ire.h>
  91 #include <inet/ip_ftable.h>
  92 #include <inet/ip_rts.h>
  93 #include <inet/ip_ndp.h>
  94 #include <inet/ip_listutils.h>
  95 #include <netinet/igmp.h>
  96 #include <netinet/ip_mroute.h>
  97 #include <inet/ipp_common.h>
  98 #include <inet/cc.h>
  99 
 100 #include <net/pfkeyv2.h>
 101 #include <inet/sadb.h>
 102 #include <inet/ipsec_impl.h>
 103 #include <inet/iptun/iptun_impl.h>
 104 #include <inet/ipdrop.h>
 105 #include <inet/ip_netinfo.h>
 106 #include <inet/ilb_ip.h>
 107 
 108 #include <sys/ethernet.h>
 109 #include <net/if_types.h>
 110 #include <sys/cpuvar.h>
 111 
 112 #include <ipp/ipp.h>
 113 #include <ipp/ipp_impl.h>
 114 #include <ipp/ipgpc/ipgpc.h>
 115 
 116 #include <sys/pattr.h>
 117 #include <inet/ipclassifier.h>
 118 #include <inet/sctp_ip.h>
 119 #include <inet/sctp/sctp_impl.h>
 120 #include <inet/udp_impl.h>
 121 #include <inet/rawip_impl.h>
 122 #include <inet/rts_impl.h>
 123 
 124 #include <sys/tsol/label.h>
 125 #include <sys/tsol/tnet.h>
 126 
 127 #include <sys/squeue_impl.h>
 128 #include <inet/ip_arp.h>
 129 
 130 #include <sys/clock_impl.h>       /* For LBOLT_FASTPATH{,64} */
 131 
 132 /*
 133  * Values for squeue switch:
 134  * IP_SQUEUE_ENTER_NODRAIN: SQ_NODRAIN
 135  * IP_SQUEUE_ENTER: SQ_PROCESS
 136  * IP_SQUEUE_FILL: SQ_FILL
 137  */
 138 int ip_squeue_enter = IP_SQUEUE_ENTER;  /* Setable in /etc/system */
 139 
 140 int ip_squeue_flag;
 141 
 142 /*
 143  * Setable in /etc/system
 144  */
 145 int ip_poll_normal_ms = 100;
 146 int ip_poll_normal_ticks = 0;
 147 int ip_modclose_ackwait_ms = 3000;
 148 
 149 /*
 150  * It would be nice to have these present only in DEBUG systems, but the
 151  * current design of the global symbol checking logic requires them to be
 152  * unconditionally present.
 153  */
 154 uint_t ip_thread_data;                  /* TSD key for debug support */
 155 krwlock_t ip_thread_rwlock;
 156 list_t  ip_thread_list;
 157 
 158 /*
 159  * Structure to represent a linked list of msgblks. Used by ip_snmp_ functions.
 160  */
 161 
 162 struct listptr_s {
 163         mblk_t  *lp_head;       /* pointer to the head of the list */
 164         mblk_t  *lp_tail;       /* pointer to the tail of the list */
 165 };
 166 
 167 typedef struct listptr_s listptr_t;
 168 
 169 /*
 170  * This is used by ip_snmp_get_mib2_ip_route_media and
 171  * ip_snmp_get_mib2_ip6_route_media to carry the lists of return data.
 172  */
 173 typedef struct iproutedata_s {
 174         uint_t          ird_idx;
 175         uint_t          ird_flags;      /* see below */
 176         listptr_t       ird_route;      /* ipRouteEntryTable */
 177         listptr_t       ird_netmedia;   /* ipNetToMediaEntryTable */
 178         listptr_t       ird_attrs;      /* ipRouteAttributeTable */
 179 } iproutedata_t;
 180 
 181 /* Include ire_testhidden and IRE_IF_CLONE routes */
 182 #define IRD_REPORT_ALL  0x01
 183 
 184 /*
 185  * Cluster specific hooks. These should be NULL when booted as a non-cluster
 186  */
 187 
 188 /*
 189  * Hook functions to enable cluster networking
 190  * On non-clustered systems these vectors must always be NULL.
 191  *
 192  * Hook function to Check ip specified ip address is a shared ip address
 193  * in the cluster
 194  *
 195  */
 196 int (*cl_inet_isclusterwide)(netstackid_t stack_id, uint8_t protocol,
 197     sa_family_t addr_family, uint8_t *laddrp, void *args) = NULL;
 198 
 199 /*
 200  * Hook function to generate cluster wide ip fragment identifier
 201  */
 202 uint32_t (*cl_inet_ipident)(netstackid_t stack_id, uint8_t protocol,
 203     sa_family_t addr_family, uint8_t *laddrp, uint8_t *faddrp,
 204     void *args) = NULL;
 205 
 206 /*
 207  * Hook function to generate cluster wide SPI.
 208  */
 209 void (*cl_inet_getspi)(netstackid_t, uint8_t, uint8_t *, size_t,
 210     void *) = NULL;
 211 
 212 /*
 213  * Hook function to verify if the SPI is already utlized.
 214  */
 215 
 216 int (*cl_inet_checkspi)(netstackid_t, uint8_t, uint32_t, void *) = NULL;
 217 
 218 /*
 219  * Hook function to delete the SPI from the cluster wide repository.
 220  */
 221 
 222 void (*cl_inet_deletespi)(netstackid_t, uint8_t, uint32_t, void *) = NULL;
 223 
 224 /*
 225  * Hook function to inform the cluster when packet received on an IDLE SA
 226  */
 227 
 228 void (*cl_inet_idlesa)(netstackid_t, uint8_t, uint32_t, sa_family_t,
 229     in6_addr_t, in6_addr_t, void *) = NULL;
 230 
 231 /*
 232  * Synchronization notes:
 233  *
 234  * IP is a fully D_MP STREAMS module/driver. Thus it does not depend on any
 235  * MT level protection given by STREAMS. IP uses a combination of its own
 236  * internal serialization mechanism and standard Solaris locking techniques.
 237  * The internal serialization is per phyint.  This is used to serialize
 238  * plumbing operations, IPMP operations, most set ioctls, etc.
 239  *
 240  * Plumbing is a long sequence of operations involving message
 241  * exchanges between IP, ARP and device drivers. Many set ioctls are typically
 242  * involved in plumbing operations. A natural model is to serialize these
 243  * ioctls one per ill. For example plumbing of hme0 and qfe0 can go on in
 244  * parallel without any interference. But various set ioctls on hme0 are best
 245  * serialized, along with IPMP operations and processing of DLPI control
 246  * messages received from drivers on a per phyint basis. This serialization is
 247  * provided by the ipsq_t and primitives operating on this. Details can
 248  * be found in ip_if.c above the core primitives operating on ipsq_t.
 249  *
 250  * Lookups of an ipif or ill by a thread return a refheld ipif / ill.
 251  * Simiarly lookup of an ire by a thread also returns a refheld ire.
 252  * In addition ipif's and ill's referenced by the ire are also indirectly
 253  * refheld. Thus no ipif or ill can vanish as long as an ipif is refheld
 254  * directly or indirectly. For example an SIOCSLIFADDR ioctl that changes the
 255  * address of an ipif has to go through the ipsq_t. This ensures that only
 256  * one such exclusive operation proceeds at any time on the ipif. It then
 257  * waits for all refcnts
 258  * associated with this ipif to come down to zero. The address is changed
 259  * only after the ipif has been quiesced. Then the ipif is brought up again.
 260  * More details are described above the comment in ip_sioctl_flags.
 261  *
 262  * Packet processing is based mostly on IREs and are fully multi-threaded
 263  * using standard Solaris MT techniques.
 264  *
 265  * There are explicit locks in IP to handle:
 266  * - The ip_g_head list maintained by mi_open_link() and friends.
 267  *
 268  * - The reassembly data structures (one lock per hash bucket)
 269  *
 270  * - conn_lock is meant to protect conn_t fields. The fields actually
 271  *   protected by conn_lock are documented in the conn_t definition.
 272  *
 273  * - ire_lock to protect some of the fields of the ire, IRE tables
 274  *   (one lock per hash bucket). Refer to ip_ire.c for details.
 275  *
 276  * - ndp_g_lock and ncec_lock for protecting NCEs.
 277  *
 278  * - ill_lock protects fields of the ill and ipif. Details in ip.h
 279  *
 280  * - ill_g_lock: This is a global reader/writer lock. Protects the following
 281  *      * The AVL tree based global multi list of all ills.
 282  *      * The linked list of all ipifs of an ill
 283  *      * The <ipsq-xop> mapping
 284  *      * <ill-phyint> association
 285  *   Insertion/deletion of an ill in the system, insertion/deletion of an ipif
 286  *   into an ill, changing the <ipsq-xop> mapping of an ill, changing the
 287  *   <ill-phyint> assoc of an ill will all have to hold the ill_g_lock as
 288  *   writer for the actual duration of the insertion/deletion/change.
 289  *
 290  * - ill_lock:  This is a per ill mutex.
 291  *   It protects some members of the ill_t struct; see ip.h for details.
 292  *   It also protects the <ill-phyint> assoc.
 293  *   It also protects the list of ipifs hanging off the ill.
 294  *
 295  * - ipsq_lock: This is a per ipsq_t mutex lock.
 296  *   This protects some members of the ipsq_t struct; see ip.h for details.
 297  *   It also protects the <ipsq-ipxop> mapping
 298  *
 299  * - ipx_lock: This is a per ipxop_t mutex lock.
 300  *   This protects some members of the ipxop_t struct; see ip.h for details.
 301  *
 302  * - phyint_lock: This is a per phyint mutex lock. Protects just the
 303  *   phyint_flags
 304  *
 305  * - ip_addr_avail_lock: This is used to ensure the uniqueness of IP addresses.
 306  *   This lock is held in ipif_up_done and the ipif is marked IPIF_UP and the
 307  *   uniqueness check also done atomically.
 308  *
 309  * - ill_g_usesrc_lock: This readers/writer lock protects the usesrc
 310  *   group list linked by ill_usesrc_grp_next. It also protects the
 311  *   ill_usesrc_ifindex field. It is taken as a writer when a member of the
 312  *   group is being added or deleted.  This lock is taken as a reader when
 313  *   walking the list/group(eg: to get the number of members in a usesrc group).
 314  *   Note, it is only necessary to take this lock if the ill_usesrc_grp_next
 315  *   field is changing state i.e from NULL to non-NULL or vice-versa. For
 316  *   example, it is not necessary to take this lock in the initial portion
 317  *   of ip_sioctl_slifusesrc or at all in ip_sioctl_flags since these
 318  *   operations are executed exclusively and that ensures that the "usesrc
 319  *   group state" cannot change. The "usesrc group state" change can happen
 320  *   only in the latter part of ip_sioctl_slifusesrc and in ill_delete.
 321  *
 322  * Changing <ill-phyint>, <ipsq-xop> assocications:
 323  *
 324  * To change the <ill-phyint> association, the ill_g_lock must be held
 325  * as writer, and the ill_locks of both the v4 and v6 instance of the ill
 326  * must be held.
 327  *
 328  * To change the <ipsq-xop> association, the ill_g_lock must be held as
 329  * writer, the ipsq_lock must be held, and one must be writer on the ipsq.
 330  * This is only done when ills are added or removed from IPMP groups.
 331  *
 332  * To add or delete an ipif from the list of ipifs hanging off the ill,
 333  * ill_g_lock (writer) and ill_lock must be held and the thread must be
 334  * a writer on the associated ipsq.
 335  *
 336  * To add or delete an ill to the system, the ill_g_lock must be held as
 337  * writer and the thread must be a writer on the associated ipsq.
 338  *
 339  * To add or delete an ilm to an ill, the ill_lock must be held and the thread
 340  * must be a writer on the associated ipsq.
 341  *
 342  * Lock hierarchy
 343  *
 344  * Some lock hierarchy scenarios are listed below.
 345  *
 346  * ill_g_lock -> conn_lock -> ill_lock -> ipsq_lock -> ipx_lock
 347  * ill_g_lock -> ill_lock(s) -> phyint_lock
 348  * ill_g_lock -> ndp_g_lock -> ill_lock -> ncec_lock
 349  * ill_g_lock -> ip_addr_avail_lock
 350  * conn_lock -> irb_lock -> ill_lock -> ire_lock
 351  * ill_g_lock -> ip_g_nd_lock
 352  * ill_g_lock -> ips_ipmp_lock -> ill_lock -> nce_lock
 353  * ill_g_lock -> ndp_g_lock -> ill_lock -> ncec_lock -> nce_lock
 354  * arl_lock -> ill_lock
 355  * ips_ire_dep_lock -> irb_lock
 356  *
 357  * When more than 1 ill lock is needed to be held, all ill lock addresses
 358  * are sorted on address and locked starting from highest addressed lock
 359  * downward.
 360  *
 361  * Multicast scenarios
 362  * ips_ill_g_lock -> ill_mcast_lock
 363  * conn_ilg_lock -> ips_ill_g_lock -> ill_lock
 364  * ill_mcast_serializer -> ill_mcast_lock -> ips_ipmp_lock -> ill_lock
 365  * ill_mcast_serializer -> ill_mcast_lock -> connf_lock -> conn_lock
 366  * ill_mcast_serializer -> ill_mcast_lock -> conn_ilg_lock
 367  * ill_mcast_serializer -> ill_mcast_lock -> ips_igmp_timer_lock
 368  *
 369  * IPsec scenarios
 370  *
 371  * ipsa_lock -> ill_g_lock -> ill_lock
 372  * ill_g_usesrc_lock -> ill_g_lock -> ill_lock
 373  *
 374  * Trusted Solaris scenarios
 375  *
 376  * igsa_lock -> gcgrp_rwlock -> gcgrp_lock
 377  * igsa_lock -> gcdb_lock
 378  * gcgrp_rwlock -> ire_lock
 379  * gcgrp_rwlock -> gcdb_lock
 380  *
 381  * squeue(sq_lock), flow related (ft_lock, fe_lock) locking
 382  *
 383  * cpu_lock --> ill_lock --> sqset_lock --> sq_lock
 384  * sq_lock -> conn_lock -> QLOCK(q)
 385  * ill_lock -> ft_lock -> fe_lock
 386  *
 387  * Routing/forwarding table locking notes:
 388  *
 389  * Lock acquisition order: Radix tree lock, irb_lock.
 390  * Requirements:
 391  * i.  Walker must not hold any locks during the walker callback.
 392  * ii  Walker must not see a truncated tree during the walk because of any node
 393  *     deletion.
 394  * iii Existing code assumes ire_bucket is valid if it is non-null and is used
 395  *     in many places in the code to walk the irb list. Thus even if all the
 396  *     ires in a bucket have been deleted, we still can't free the radix node
 397  *     until the ires have actually been inactive'd (freed).
 398  *
 399  * Tree traversal - Need to hold the global tree lock in read mode.
 400  * Before dropping the global tree lock, need to either increment the ire_refcnt
 401  * to ensure that the radix node can't be deleted.
 402  *
 403  * Tree add - Need to hold the global tree lock in write mode to add a
 404  * radix node. To prevent the node from being deleted, increment the
 405  * irb_refcnt, after the node is added to the tree. The ire itself is
 406  * added later while holding the irb_lock, but not the tree lock.
 407  *
 408  * Tree delete - Need to hold the global tree lock and irb_lock in write mode.
 409  * All associated ires must be inactive (i.e. freed), and irb_refcnt
 410  * must be zero.
 411  *
 412  * Walker - Increment irb_refcnt before calling the walker callback. Hold the
 413  * global tree lock (read mode) for traversal.
 414  *
 415  * IRE dependencies - In some cases we hold ips_ire_dep_lock across ire_refrele
 416  * hence we will acquire irb_lock while holding ips_ire_dep_lock.
 417  *
 418  * IPsec notes :
 419  *
 420  * IP interacts with the IPsec code (AH/ESP) by storing IPsec attributes
 421  * in the ip_xmit_attr_t ip_recv_attr_t. For outbound datagrams, the
 422  * ip_xmit_attr_t has the
 423  * information used by the IPsec code for applying the right level of
 424  * protection. The information initialized by IP in the ip_xmit_attr_t
 425  * is determined by the per-socket policy or global policy in the system.
 426  * For inbound datagrams, the ip_recv_attr_t
 427  * starts out with nothing in it. It gets filled
 428  * with the right information if it goes through the AH/ESP code, which
 429  * happens if the incoming packet is secure. The information initialized
 430  * by AH/ESP, is later used by IP (during fanouts to ULP) to see whether
 431  * the policy requirements needed by per-socket policy or global policy
 432  * is met or not.
 433  *
 434  * For fully connected sockets i.e dst, src [addr, port] is known,
 435  * conn_policy_cached is set indicating that policy has been cached.
 436  * conn_in_enforce_policy may or may not be set depending on whether
 437  * there is a global policy match or per-socket policy match.
 438  * Policy inheriting happpens in ip_policy_set once the destination is known.
 439  * Once the right policy is set on the conn_t, policy cannot change for
 440  * this socket. This makes life simpler for TCP (UDP ?) where
 441  * re-transmissions go out with the same policy. For symmetry, policy
 442  * is cached for fully connected UDP sockets also. Thus if policy is cached,
 443  * it also implies that policy is latched i.e policy cannot change
 444  * on these sockets. As we have the right policy on the conn, we don't
 445  * have to lookup global policy for every outbound and inbound datagram
 446  * and thus serving as an optimization. Note that a global policy change
 447  * does not affect fully connected sockets if they have policy. If fully
 448  * connected sockets did not have any policy associated with it, global
 449  * policy change may affect them.
 450  *
 451  * IP Flow control notes:
 452  * ---------------------
 453  * Non-TCP streams are flow controlled by IP. The way this is accomplished
 454  * differs when ILL_CAPAB_DLD_DIRECT is enabled for that IP instance. When
 455  * ILL_DIRECT_CAPABLE(ill) is TRUE, IP can do direct function calls into
 456  * GLDv3. Otherwise packets are sent down to lower layers using STREAMS
 457  * functions.
 458  *
 459  * Per Tx ring udp flow control:
 460  * This is applicable only when ILL_CAPAB_DLD_DIRECT capability is set in
 461  * the ill (i.e. ILL_DIRECT_CAPABLE(ill) is true).
 462  *
 463  * The underlying link can expose multiple Tx rings to the GLDv3 mac layer.
 464  * To achieve best performance, outgoing traffic need to be fanned out among
 465  * these Tx ring. mac_tx() is called (via str_mdata_fastpath_put()) to send
 466  * traffic out of the NIC and it takes a fanout hint. UDP connections pass
 467  * the address of connp as fanout hint to mac_tx(). Under flow controlled
 468  * condition, mac_tx() returns a non-NULL cookie (ip_mac_tx_cookie_t). This
 469  * cookie points to a specific Tx ring that is blocked. The cookie is used to
 470  * hash into an idl_tx_list[] entry in idl_tx_list[] array. Each idl_tx_list_t
 471  * point to drain_lists (idl_t's). These drain list will store the blocked UDP
 472  * connp's. The drain list is not a single list but a configurable number of
 473  * lists.
 474  *
 475  * The diagram below shows idl_tx_list_t's and their drain_lists. ip_stack_t
 476  * has an array of idl_tx_list_t. The size of the array is TX_FANOUT_SIZE
 477  * which is equal to 128. This array in turn contains a pointer to idl_t[],
 478  * the ip drain list. The idl_t[] array size is MIN(max_ncpus, 8). The drain
 479  * list will point to the list of connp's that are flow controlled.
 480  *
 481  *                      ---------------   -------   -------   -------
 482  *                   |->|drain_list[0]|-->|connp|-->|connp|-->|connp|-->
 483  *                   |  ---------------   -------   -------   -------
 484  *                   |  ---------------   -------   -------   -------
 485  *                   |->|drain_list[1]|-->|connp|-->|connp|-->|connp|-->
 486  * ----------------  |  ---------------   -------   -------   -------
 487  * |idl_tx_list[0]|->|  ---------------   -------   -------   -------
 488  * ----------------  |->|drain_list[2]|-->|connp|-->|connp|-->|connp|-->
 489  *                   |  ---------------   -------   -------   -------
 490  *                   .        .              .         .         .
 491  *                   |  ---------------   -------   -------   -------
 492  *                   |->|drain_list[n]|-->|connp|-->|connp|-->|connp|-->
 493  *                      ---------------   -------   -------   -------
 494  *                      ---------------   -------   -------   -------
 495  *                   |->|drain_list[0]|-->|connp|-->|connp|-->|connp|-->
 496  *                   |  ---------------   -------   -------   -------
 497  *                   |  ---------------   -------   -------   -------
 498  * ----------------  |->|drain_list[1]|-->|connp|-->|connp|-->|connp|-->
 499  * |idl_tx_list[1]|->|  ---------------   -------   -------   -------
 500  * ----------------  |        .              .         .         .
 501  *                   |  ---------------   -------   -------   -------
 502  *                   |->|drain_list[n]|-->|connp|-->|connp|-->|connp|-->
 503  *                      ---------------   -------   -------   -------
 504  *     .....
 505  * ----------------
 506  * |idl_tx_list[n]|-> ...
 507  * ----------------
 508  *
 509  * When mac_tx() returns a cookie, the cookie is hashed into an index into
 510  * ips_idl_tx_list[], and conn_drain_insert() is called with the idl_tx_list
 511  * to insert the conn onto.  conn_drain_insert() asserts flow control for the
 512  * sockets via su_txq_full() (non-STREAMS) or QFULL on conn_wq (STREAMS).
 513  * Further, conn_blocked is set to indicate that the conn is blocked.
 514  *
 515  * GLDv3 calls ill_flow_enable() when flow control is relieved.  The cookie
 516  * passed in the call to ill_flow_enable() identifies the blocked Tx ring and
 517  * is again hashed to locate the appropriate idl_tx_list, which is then
 518  * drained via conn_walk_drain().  conn_walk_drain() goes through each conn in
 519  * the drain list and calls conn_drain_remove() to clear flow control (via
 520  * calling su_txq_full() or clearing QFULL), and remove the conn from the
 521  * drain list.
 522  *
 523  * Note that the drain list is not a single list but a (configurable) array of
 524  * lists (8 elements by default).  Synchronization between drain insertion and
 525  * flow control wakeup is handled by using idl_txl->txl_lock, and only
 526  * conn_drain_insert() and conn_drain_remove() manipulate the drain list.
 527  *
 528  * Flow control via STREAMS is used when ILL_DIRECT_CAPABLE() returns FALSE.
 529  * On the send side, if the packet cannot be sent down to the driver by IP
 530  * (canput() fails), ip_xmit() drops the packet and returns EWOULDBLOCK to the
 531  * caller, who may then invoke ixa_check_drain_insert() to insert the conn on
 532  * the 0'th drain list.  When ip_wsrv() runs on the ill_wq because flow
 533  * control has been relieved, the blocked conns in the 0'th drain list are
 534  * drained as in the non-STREAMS case.
 535  *
 536  * In both the STREAMS and non-STREAMS cases, the sockfs upcall to set QFULL
 537  * is done when the conn is inserted into the drain list (conn_drain_insert())
 538  * and cleared when the conn is removed from the it (conn_drain_remove()).
 539  *
 540  * IPQOS notes:
 541  *
 542  * IPQoS Policies are applied to packets using IPPF (IP Policy framework)
 543  * and IPQoS modules. IPPF includes hooks in IP at different control points
 544  * (callout positions) which direct packets to IPQoS modules for policy
 545  * processing. Policies, if present, are global.
 546  *
 547  * The callout positions are located in the following paths:
 548  *              o local_in (packets destined for this host)
 549  *              o local_out (packets orginating from this host )
 550  *              o fwd_in  (packets forwarded by this m/c - inbound)
 551  *              o fwd_out (packets forwarded by this m/c - outbound)
 552  * Hooks at these callout points can be enabled/disabled using the ndd variable
 553  * ip_policy_mask (a bit mask with the 4 LSB indicating the callout positions).
 554  * By default all the callout positions are enabled.
 555  *
 556  * Outbound (local_out)
 557  * Hooks are placed in ire_send_wire_v4 and ire_send_wire_v6.
 558  *
 559  * Inbound (local_in)
 560  * Hooks are placed in ip_fanout_v4 and ip_fanout_v6.
 561  *
 562  * Forwarding (in and out)
 563  * Hooks are placed in ire_recv_forward_v4/v6.
 564  *
 565  * IP Policy Framework processing (IPPF processing)
 566  * Policy processing for a packet is initiated by ip_process, which ascertains
 567  * that the classifier (ipgpc) is loaded and configured, failing which the
 568  * packet resumes normal processing in IP. If the clasifier is present, the
 569  * packet is acted upon by one or more IPQoS modules (action instances), per
 570  * filters configured in ipgpc and resumes normal IP processing thereafter.
 571  * An action instance can drop a packet in course of its processing.
 572  *
 573  * Zones notes:
 574  *
 575  * The partitioning rules for networking are as follows:
 576  * 1) Packets coming from a zone must have a source address belonging to that
 577  * zone.
 578  * 2) Packets coming from a zone can only be sent on a physical interface on
 579  * which the zone has an IP address.
 580  * 3) Between two zones on the same machine, packet delivery is only allowed if
 581  * there's a matching route for the destination and zone in the forwarding
 582  * table.
 583  * 4) The TCP and UDP port spaces are per-zone; that is, two processes in
 584  * different zones can bind to the same port with the wildcard address
 585  * (INADDR_ANY).
 586  *
 587  * The granularity of interface partitioning is at the logical interface level.
 588  * Therefore, every zone has its own IP addresses, and incoming packets can be
 589  * attributed to a zone unambiguously. A logical interface is placed into a zone
 590  * using the SIOCSLIFZONE ioctl; this sets the ipif_zoneid field in the ipif_t
 591  * structure. Rule (1) is implemented by modifying the source address selection
 592  * algorithm so that the list of eligible addresses is filtered based on the
 593  * sending process zone.
 594  *
 595  * The Internet Routing Entries (IREs) are either exclusive to a zone or shared
 596  * across all zones, depending on their type. Here is the break-up:
 597  *
 598  * IRE type                             Shared/exclusive
 599  * --------                             ----------------
 600  * IRE_BROADCAST                        Exclusive
 601  * IRE_DEFAULT (default routes)         Shared (*)
 602  * IRE_LOCAL                            Exclusive (x)
 603  * IRE_LOOPBACK                         Exclusive
 604  * IRE_PREFIX (net routes)              Shared (*)
 605  * IRE_IF_NORESOLVER (interface routes) Exclusive
 606  * IRE_IF_RESOLVER (interface routes)   Exclusive
 607  * IRE_IF_CLONE (interface routes)      Exclusive
 608  * IRE_HOST (host routes)               Shared (*)
 609  *
 610  * (*) A zone can only use a default or off-subnet route if the gateway is
 611  * directly reachable from the zone, that is, if the gateway's address matches
 612  * one of the zone's logical interfaces.
 613  *
 614  * (x) IRE_LOCAL are handled a bit differently.
 615  * When ip_restrict_interzone_loopback is set (the default),
 616  * ire_route_recursive restricts loopback using an IRE_LOCAL
 617  * between zone to the case when L2 would have conceptually looped the packet
 618  * back, i.e. the loopback which is required since neither Ethernet drivers
 619  * nor Ethernet hardware loops them back. This is the case when the normal
 620  * routes (ignoring IREs with different zoneids) would send out the packet on
 621  * the same ill as the ill with which is IRE_LOCAL is associated.
 622  *
 623  * Multiple zones can share a common broadcast address; typically all zones
 624  * share the 255.255.255.255 address. Incoming as well as locally originated
 625  * broadcast packets must be dispatched to all the zones on the broadcast
 626  * network. For directed broadcasts (e.g. 10.16.72.255) this is not trivial
 627  * since some zones may not be on the 10.16.72/24 network. To handle this, each
 628  * zone has its own set of IRE_BROADCAST entries; then, broadcast packets are
 629  * sent to every zone that has an IRE_BROADCAST entry for the destination
 630  * address on the input ill, see ip_input_broadcast().
 631  *
 632  * Applications in different zones can join the same multicast group address.
 633  * The same logic applies for multicast as for broadcast. ip_input_multicast
 634  * dispatches packets to all zones that have members on the physical interface.
 635  */
 636 
 637 /*
 638  * Squeue Fanout flags:
 639  *      0: No fanout.
 640  *      1: Fanout across all squeues
 641  */
 642 boolean_t       ip_squeue_fanout = 0;
 643 
 644 /*
 645  * Maximum dups allowed per packet.
 646  */
 647 uint_t ip_max_frag_dups = 10;
 648 
 649 static int      ip_open(queue_t *q, dev_t *devp, int flag, int sflag,
 650                     cred_t *credp, boolean_t isv6);
 651 static mblk_t   *ip_xmit_attach_llhdr(mblk_t *, nce_t *);
 652 
 653 static boolean_t icmp_inbound_verify_v4(mblk_t *, icmph_t *, ip_recv_attr_t *);
 654 static void     icmp_inbound_too_big_v4(icmph_t *, ip_recv_attr_t *);
 655 static void     icmp_inbound_error_fanout_v4(mblk_t *, icmph_t *,
 656     ip_recv_attr_t *);
 657 static void     icmp_options_update(ipha_t *);
 658 static void     icmp_param_problem(mblk_t *, uint8_t,  ip_recv_attr_t *);
 659 static void     icmp_pkt(mblk_t *, void *, size_t, ip_recv_attr_t *);
 660 static mblk_t   *icmp_pkt_err_ok(mblk_t *, ip_recv_attr_t *);
 661 static void     icmp_redirect_v4(mblk_t *mp, ipha_t *, icmph_t *,
 662     ip_recv_attr_t *);
 663 static void     icmp_send_redirect(mblk_t *, ipaddr_t, ip_recv_attr_t *);
 664 static void     icmp_send_reply_v4(mblk_t *, ipha_t *, icmph_t *,
 665     ip_recv_attr_t *);
 666 
 667 mblk_t          *ip_dlpi_alloc(size_t, t_uscalar_t);
 668 char            *ip_dot_addr(ipaddr_t, char *);
 669 mblk_t          *ip_carve_mp(mblk_t **, ssize_t);
 670 static char     *ip_dot_saddr(uchar_t *, char *);
 671 static int      ip_lrput(queue_t *, mblk_t *);
 672 ipaddr_t        ip_net_mask(ipaddr_t);
 673 char            *ip_nv_lookup(nv_t *, int);
 674 int             ip_rput(queue_t *, mblk_t *);
 675 static void     ip_rput_dlpi_writer(ipsq_t *dummy_sq, queue_t *q, mblk_t *mp,
 676                     void *dummy_arg);
 677 int             ip_snmp_get(queue_t *, mblk_t *, int, boolean_t);
 678 static mblk_t   *ip_snmp_get_mib2_ip(queue_t *, mblk_t *,
 679                     mib2_ipIfStatsEntry_t *, ip_stack_t *, boolean_t);
 680 static mblk_t   *ip_snmp_get_mib2_ip_traffic_stats(queue_t *, mblk_t *,
 681                     ip_stack_t *, boolean_t);
 682 static mblk_t   *ip_snmp_get_mib2_ip6(queue_t *, mblk_t *, ip_stack_t *,
 683                     boolean_t);
 684 static mblk_t   *ip_snmp_get_mib2_icmp(queue_t *, mblk_t *, ip_stack_t *ipst);
 685 static mblk_t   *ip_snmp_get_mib2_icmp6(queue_t *, mblk_t *, ip_stack_t *ipst);
 686 static mblk_t   *ip_snmp_get_mib2_igmp(queue_t *, mblk_t *, ip_stack_t *ipst);
 687 static mblk_t   *ip_snmp_get_mib2_multi(queue_t *, mblk_t *, ip_stack_t *ipst);
 688 static mblk_t   *ip_snmp_get_mib2_ip_addr(queue_t *, mblk_t *,
 689                     ip_stack_t *ipst, boolean_t);
 690 static mblk_t   *ip_snmp_get_mib2_ip6_addr(queue_t *, mblk_t *,
 691                     ip_stack_t *ipst, boolean_t);
 692 static mblk_t   *ip_snmp_get_mib2_ip_group_src(queue_t *, mblk_t *,
 693                     ip_stack_t *ipst);
 694 static mblk_t   *ip_snmp_get_mib2_ip6_group_src(queue_t *, mblk_t *,
 695                     ip_stack_t *ipst);
 696 static mblk_t   *ip_snmp_get_mib2_ip_group_mem(queue_t *, mblk_t *,
 697                     ip_stack_t *ipst);
 698 static mblk_t   *ip_snmp_get_mib2_ip6_group_mem(queue_t *, mblk_t *,
 699                     ip_stack_t *ipst);
 700 static mblk_t   *ip_snmp_get_mib2_virt_multi(queue_t *, mblk_t *,
 701                     ip_stack_t *ipst);
 702 static mblk_t   *ip_snmp_get_mib2_multi_rtable(queue_t *, mblk_t *,
 703                     ip_stack_t *ipst);
 704 static mblk_t   *ip_snmp_get_mib2_ip_route_media(queue_t *, mblk_t *, int,
 705                     ip_stack_t *ipst);
 706 static mblk_t   *ip_snmp_get_mib2_ip6_route_media(queue_t *, mblk_t *, int,
 707                     ip_stack_t *ipst);
 708 static void     ip_snmp_get2_v4(ire_t *, iproutedata_t *);
 709 static void     ip_snmp_get2_v6_route(ire_t *, iproutedata_t *);
 710 static void     ip_snmp_get2_v4_media(ncec_t *, void *);
 711 static void     ip_snmp_get2_v6_media(ncec_t *, void *);
 712 int             ip_snmp_set(queue_t *, int, int, uchar_t *, int);
 713 
 714 static mblk_t   *ip_fragment_copyhdr(uchar_t *, int, int, ip_stack_t *,
 715                     mblk_t *);
 716 
 717 static void     conn_drain_init(ip_stack_t *);
 718 static void     conn_drain_fini(ip_stack_t *);
 719 static void     conn_drain(conn_t *connp, boolean_t closing);
 720 
 721 static void     conn_walk_drain(ip_stack_t *, idl_tx_list_t *);
 722 static void     conn_walk_sctp(pfv_t, void *, zoneid_t, netstack_t *);
 723 
 724 static void     *ip_stack_init(netstackid_t stackid, netstack_t *ns);
 725 static void     ip_stack_shutdown(netstackid_t stackid, void *arg);
 726 static void     ip_stack_fini(netstackid_t stackid, void *arg);
 727 
 728 static int      ip_multirt_apply_membership(int (*fn)(conn_t *, boolean_t,
 729     const in6_addr_t *, ipaddr_t, uint_t, mcast_record_t, const in6_addr_t *),
 730     ire_t *, conn_t *, boolean_t, const in6_addr_t *,  mcast_record_t,
 731     const in6_addr_t *);
 732 
 733 static int      ip_squeue_switch(int);
 734 
 735 static void     *ip_kstat_init(netstackid_t, ip_stack_t *);
 736 static void     ip_kstat_fini(netstackid_t, kstat_t *);
 737 static int      ip_kstat_update(kstat_t *kp, int rw);
 738 static void     *icmp_kstat_init(netstackid_t);
 739 static void     icmp_kstat_fini(netstackid_t, kstat_t *);
 740 static int      icmp_kstat_update(kstat_t *kp, int rw);
 741 static void     *ip_kstat2_init(netstackid_t, ip_stat_t *);
 742 static void     ip_kstat2_fini(netstackid_t, kstat_t *);
 743 
 744 static void     ipobs_init(ip_stack_t *);
 745 static void     ipobs_fini(ip_stack_t *);
 746 
 747 static int      ip_tp_cpu_update(cpu_setup_t, int, void *);
 748 
 749 ipaddr_t        ip_g_all_ones = IP_HOST_MASK;
 750 
 751 static long ip_rput_pullups;
 752 int     dohwcksum = 1;  /* use h/w cksum if supported by the hardware */
 753 
 754 vmem_t *ip_minor_arena_sa; /* for minor nos. from INET_MIN_DEV+2 thru 2^^18-1 */
 755 vmem_t *ip_minor_arena_la; /* for minor nos. from 2^^18 thru 2^^32-1 */
 756 
 757 int     ip_debug;
 758 
 759 /*
 760  * Multirouting/CGTP stuff
 761  */
 762 int     ip_cgtp_filter_rev = CGTP_FILTER_REV;   /* CGTP hooks version */
 763 
 764 /*
 765  * IP tunables related declarations. Definitions are in ip_tunables.c
 766  */
 767 extern mod_prop_info_t ip_propinfo_tbl[];
 768 extern int ip_propinfo_count;
 769 
 770 /*
 771  * Table of IP ioctls encoding the various properties of the ioctl and
 772  * indexed based on the last byte of the ioctl command. Occasionally there
 773  * is a clash, and there is more than 1 ioctl with the same last byte.
 774  * In such a case 1 ioctl is encoded in the ndx table and the remaining
 775  * ioctls are encoded in the misc table. An entry in the ndx table is
 776  * retrieved by indexing on the last byte of the ioctl command and comparing
 777  * the ioctl command with the value in the ndx table. In the event of a
 778  * mismatch the misc table is then searched sequentially for the desired
 779  * ioctl command.
 780  *
 781  * Entry: <command> <copyin_size> <flags> <cmd_type> <function> <restart_func>
 782  */
 783 ip_ioctl_cmd_t ip_ndx_ioctl_table[] = {
 784         /* 000 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 785         /* 001 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 786         /* 002 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 787         /* 003 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 788         /* 004 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 789         /* 005 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 790         /* 006 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 791         /* 007 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 792         /* 008 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 793         /* 009 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 794 
 795         /* 010 */ { SIOCADDRT,  sizeof (struct rtentry), IPI_PRIV,
 796                         MISC_CMD, ip_siocaddrt, NULL },
 797         /* 011 */ { SIOCDELRT,  sizeof (struct rtentry), IPI_PRIV,
 798                         MISC_CMD, ip_siocdelrt, NULL },
 799 
 800         /* 012 */ { SIOCSIFADDR, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
 801                         IF_CMD, ip_sioctl_addr, ip_sioctl_addr_restart },
 802         /* 013 */ { SIOCGIFADDR, sizeof (struct ifreq), IPI_GET_CMD,
 803                         IF_CMD, ip_sioctl_get_addr, NULL },
 804 
 805         /* 014 */ { SIOCSIFDSTADDR, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
 806                         IF_CMD, ip_sioctl_dstaddr, ip_sioctl_dstaddr_restart },
 807         /* 015 */ { SIOCGIFDSTADDR, sizeof (struct ifreq),
 808                         IPI_GET_CMD, IF_CMD, ip_sioctl_get_dstaddr, NULL },
 809 
 810         /* 016 */ { SIOCSIFFLAGS, sizeof (struct ifreq),
 811                         IPI_PRIV | IPI_WR,
 812                         IF_CMD, ip_sioctl_flags, ip_sioctl_flags_restart },
 813         /* 017 */ { SIOCGIFFLAGS, sizeof (struct ifreq),
 814                         IPI_MODOK | IPI_GET_CMD,
 815                         IF_CMD, ip_sioctl_get_flags, NULL },
 816 
 817         /* 018 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 818         /* 019 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 819 
 820         /* copyin size cannot be coded for SIOCGIFCONF */
 821         /* 020 */ { O_SIOCGIFCONF, 0, IPI_GET_CMD,
 822                         MISC_CMD, ip_sioctl_get_ifconf, NULL },
 823 
 824         /* 021 */ { SIOCSIFMTU, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
 825                         IF_CMD, ip_sioctl_mtu, NULL },
 826         /* 022 */ { SIOCGIFMTU, sizeof (struct ifreq), IPI_GET_CMD,
 827                         IF_CMD, ip_sioctl_get_mtu, NULL },
 828         /* 023 */ { SIOCGIFBRDADDR, sizeof (struct ifreq),
 829                         IPI_GET_CMD, IF_CMD, ip_sioctl_get_brdaddr, NULL },
 830         /* 024 */ { SIOCSIFBRDADDR, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
 831                         IF_CMD, ip_sioctl_brdaddr, NULL },
 832         /* 025 */ { SIOCGIFNETMASK, sizeof (struct ifreq),
 833                         IPI_GET_CMD, IF_CMD, ip_sioctl_get_netmask, NULL },
 834         /* 026 */ { SIOCSIFNETMASK, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
 835                         IF_CMD, ip_sioctl_netmask, ip_sioctl_netmask_restart },
 836         /* 027 */ { SIOCGIFMETRIC, sizeof (struct ifreq),
 837                         IPI_GET_CMD, IF_CMD, ip_sioctl_get_metric, NULL },
 838         /* 028 */ { SIOCSIFMETRIC, sizeof (struct ifreq), IPI_PRIV,
 839                         IF_CMD, ip_sioctl_metric, NULL },
 840         /* 029 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 841 
 842         /* See 166-168 below for extended SIOC*XARP ioctls */
 843         /* 030 */ { SIOCSARP, sizeof (struct arpreq), IPI_PRIV | IPI_WR,
 844                         ARP_CMD, ip_sioctl_arp, NULL },
 845         /* 031 */ { SIOCGARP, sizeof (struct arpreq), IPI_GET_CMD,
 846                         ARP_CMD, ip_sioctl_arp, NULL },
 847         /* 032 */ { SIOCDARP, sizeof (struct arpreq), IPI_PRIV | IPI_WR,
 848                         ARP_CMD, ip_sioctl_arp, NULL },
 849 
 850         /* 033 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 851         /* 034 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 852         /* 035 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 853         /* 036 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 854         /* 037 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 855         /* 038 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 856         /* 039 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 857         /* 040 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 858         /* 041 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 859         /* 042 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 860         /* 043 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 861         /* 044 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 862         /* 045 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 863         /* 046 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 864         /* 047 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 865         /* 048 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 866         /* 049 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 867         /* 050 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 868         /* 051 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 869         /* 052 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 870         /* 053 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 871 
 872         /* 054 */ { IF_UNITSEL, sizeof (int), IPI_PRIV | IPI_WR | IPI_MODOK,
 873                         MISC_CMD, if_unitsel, if_unitsel_restart },
 874 
 875         /* 055 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 876         /* 056 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 877         /* 057 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 878         /* 058 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 879         /* 059 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 880         /* 060 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 881         /* 061 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 882         /* 062 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 883         /* 063 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 884         /* 064 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 885         /* 065 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 886         /* 066 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 887         /* 067 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 888         /* 068 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 889         /* 069 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 890         /* 070 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 891         /* 071 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 892         /* 072 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 893 
 894         /* 073 */ { SIOCSIFNAME, sizeof (struct ifreq),
 895                         IPI_PRIV | IPI_WR | IPI_MODOK,
 896                         IF_CMD, ip_sioctl_sifname, NULL },
 897 
 898         /* 074 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 899         /* 075 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 900         /* 076 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 901         /* 077 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 902         /* 078 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 903         /* 079 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 904         /* 080 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 905         /* 081 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 906         /* 082 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 907         /* 083 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 908         /* 084 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 909         /* 085 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 910         /* 086 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 911 
 912         /* 087 */ { SIOCGIFNUM, sizeof (int), IPI_GET_CMD,
 913                         MISC_CMD, ip_sioctl_get_ifnum, NULL },
 914         /* 088 */ { SIOCGIFMUXID, sizeof (struct ifreq), IPI_GET_CMD,
 915                         IF_CMD, ip_sioctl_get_muxid, NULL },
 916         /* 089 */ { SIOCSIFMUXID, sizeof (struct ifreq),
 917                         IPI_PRIV | IPI_WR, IF_CMD, ip_sioctl_muxid, NULL },
 918 
 919         /* Both if and lif variants share same func */
 920         /* 090 */ { SIOCGIFINDEX, sizeof (struct ifreq), IPI_GET_CMD,
 921                         IF_CMD, ip_sioctl_get_lifindex, NULL },
 922         /* Both if and lif variants share same func */
 923         /* 091 */ { SIOCSIFINDEX, sizeof (struct ifreq),
 924                         IPI_PRIV | IPI_WR, IF_CMD, ip_sioctl_slifindex, NULL },
 925 
 926         /* copyin size cannot be coded for SIOCGIFCONF */
 927         /* 092 */ { SIOCGIFCONF, 0, IPI_GET_CMD,
 928                         MISC_CMD, ip_sioctl_get_ifconf, NULL },
 929         /* 093 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 930         /* 094 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 931         /* 095 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 932         /* 096 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 933         /* 097 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 934         /* 098 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 935         /* 099 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 936         /* 100 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 937         /* 101 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 938         /* 102 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 939         /* 103 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 940         /* 104 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 941         /* 105 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 942         /* 106 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 943         /* 107 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 944         /* 108 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 945         /* 109 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 946 
 947         /* 110 */ { SIOCLIFREMOVEIF, sizeof (struct lifreq),
 948                         IPI_PRIV | IPI_WR, LIF_CMD, ip_sioctl_removeif,
 949                         ip_sioctl_removeif_restart },
 950         /* 111 */ { SIOCLIFADDIF, sizeof (struct lifreq),
 951                         IPI_GET_CMD | IPI_PRIV | IPI_WR,
 952                         LIF_CMD, ip_sioctl_addif, NULL },
 953 #define SIOCLIFADDR_NDX 112
 954         /* 112 */ { SIOCSLIFADDR, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
 955                         LIF_CMD, ip_sioctl_addr, ip_sioctl_addr_restart },
 956         /* 113 */ { SIOCGLIFADDR, sizeof (struct lifreq),
 957                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_addr, NULL },
 958         /* 114 */ { SIOCSLIFDSTADDR, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
 959                         LIF_CMD, ip_sioctl_dstaddr, ip_sioctl_dstaddr_restart },
 960         /* 115 */ { SIOCGLIFDSTADDR, sizeof (struct lifreq),
 961                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_dstaddr, NULL },
 962         /* 116 */ { SIOCSLIFFLAGS, sizeof (struct lifreq),
 963                         IPI_PRIV | IPI_WR,
 964                         LIF_CMD, ip_sioctl_flags, ip_sioctl_flags_restart },
 965         /* 117 */ { SIOCGLIFFLAGS, sizeof (struct lifreq),
 966                         IPI_GET_CMD | IPI_MODOK,
 967                         LIF_CMD, ip_sioctl_get_flags, NULL },
 968 
 969         /* 118 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 970         /* 119 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 971 
 972         /* 120 */ { O_SIOCGLIFCONF, 0, IPI_GET_CMD, MISC_CMD,
 973                         ip_sioctl_get_lifconf, NULL },
 974         /* 121 */ { SIOCSLIFMTU, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
 975                         LIF_CMD, ip_sioctl_mtu, NULL },
 976         /* 122 */ { SIOCGLIFMTU, sizeof (struct lifreq), IPI_GET_CMD,
 977                         LIF_CMD, ip_sioctl_get_mtu, NULL },
 978         /* 123 */ { SIOCGLIFBRDADDR, sizeof (struct lifreq),
 979                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_brdaddr, NULL },
 980         /* 124 */ { SIOCSLIFBRDADDR, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
 981                         LIF_CMD, ip_sioctl_brdaddr, NULL },
 982         /* 125 */ { SIOCGLIFNETMASK, sizeof (struct lifreq),
 983                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_netmask, NULL },
 984         /* 126 */ { SIOCSLIFNETMASK, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
 985                         LIF_CMD, ip_sioctl_netmask, ip_sioctl_netmask_restart },
 986         /* 127 */ { SIOCGLIFMETRIC, sizeof (struct lifreq),
 987                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_metric, NULL },
 988         /* 128 */ { SIOCSLIFMETRIC, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
 989                         LIF_CMD, ip_sioctl_metric, NULL },
 990         /* 129 */ { SIOCSLIFNAME, sizeof (struct lifreq),
 991                         IPI_PRIV | IPI_WR | IPI_MODOK,
 992                         LIF_CMD, ip_sioctl_slifname,
 993                         ip_sioctl_slifname_restart },
 994 
 995         /* 130 */ { SIOCGLIFNUM, sizeof (struct lifnum), IPI_GET_CMD,
 996                         MISC_CMD, ip_sioctl_get_lifnum, NULL },
 997         /* 131 */ { SIOCGLIFMUXID, sizeof (struct lifreq),
 998                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_muxid, NULL },
 999         /* 132 */ { SIOCSLIFMUXID, sizeof (struct lifreq),
1000                         IPI_PRIV | IPI_WR, LIF_CMD, ip_sioctl_muxid, NULL },
1001         /* 133 */ { SIOCGLIFINDEX, sizeof (struct lifreq),
1002                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_lifindex, 0 },
1003         /* 134 */ { SIOCSLIFINDEX, sizeof (struct lifreq),
1004                         IPI_PRIV | IPI_WR, LIF_CMD, ip_sioctl_slifindex, 0 },
1005         /* 135 */ { SIOCSLIFTOKEN, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
1006                         LIF_CMD, ip_sioctl_token, NULL },
1007         /* 136 */ { SIOCGLIFTOKEN, sizeof (struct lifreq),
1008                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_token, NULL },
1009         /* 137 */ { SIOCSLIFSUBNET, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
1010                         LIF_CMD, ip_sioctl_subnet, ip_sioctl_subnet_restart },
1011         /* 138 */ { SIOCGLIFSUBNET, sizeof (struct lifreq),
1012                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_subnet, NULL },
1013         /* 139 */ { SIOCSLIFLNKINFO, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
1014                         LIF_CMD, ip_sioctl_lnkinfo, NULL },
1015 
1016         /* 140 */ { SIOCGLIFLNKINFO, sizeof (struct lifreq),
1017                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_lnkinfo, NULL },
1018         /* 141 */ { SIOCLIFDELND, sizeof (struct lifreq), IPI_PRIV,
1019                         LIF_CMD, ip_siocdelndp_v6, NULL },
1020         /* 142 */ { SIOCLIFGETND, sizeof (struct lifreq), IPI_GET_CMD,
1021                         LIF_CMD, ip_siocqueryndp_v6, NULL },
1022         /* 143 */ { SIOCLIFSETND, sizeof (struct lifreq), IPI_PRIV,
1023                         LIF_CMD, ip_siocsetndp_v6, NULL },
1024         /* 144 */ { SIOCTMYADDR, sizeof (struct sioc_addrreq), IPI_GET_CMD,
1025                         MISC_CMD, ip_sioctl_tmyaddr, NULL },
1026         /* 145 */ { SIOCTONLINK, sizeof (struct sioc_addrreq), IPI_GET_CMD,
1027                         MISC_CMD, ip_sioctl_tonlink, NULL },
1028         /* 146 */ { SIOCTMYSITE, sizeof (struct sioc_addrreq), 0,
1029                         MISC_CMD, ip_sioctl_tmysite, NULL },
1030         /* 147 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1031         /* 148 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1032 
1033         /* Old *IPSECONFIG ioctls are now deprecated, now see spdsock.c */
1034         /* 149 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1035         /* 150 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1036         /* 151 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1037         /* 152 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1038 
1039         /* 153 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1040 
1041         /* 154 */ { SIOCGLIFBINDING, sizeof (struct lifreq), IPI_GET_CMD,
1042                         LIF_CMD, ip_sioctl_get_binding, NULL },
1043         /* 155 */ { SIOCSLIFGROUPNAME, sizeof (struct lifreq),
1044                         IPI_PRIV | IPI_WR,
1045                         LIF_CMD, ip_sioctl_groupname, ip_sioctl_groupname },
1046         /* 156 */ { SIOCGLIFGROUPNAME, sizeof (struct lifreq),
1047                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_groupname, NULL },
1048         /* 157 */ { SIOCGLIFGROUPINFO, sizeof (lifgroupinfo_t),
1049                         IPI_GET_CMD, MISC_CMD, ip_sioctl_groupinfo, NULL },
1050 
1051         /* Leave 158-160 unused; used to be SIOC*IFARP ioctls */
1052         /* 158 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1053         /* 159 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1054         /* 160 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1055 
1056         /* 161 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1057 
1058         /* These are handled in ip_sioctl_copyin_setup itself */
1059         /* 162 */ { SIOCGIP6ADDRPOLICY, 0, IPI_NULL_BCONT,
1060                         MISC_CMD, NULL, NULL },
1061         /* 163 */ { SIOCSIP6ADDRPOLICY, 0, IPI_PRIV | IPI_NULL_BCONT,
1062                         MISC_CMD, NULL, NULL },
1063         /* 164 */ { SIOCGDSTINFO, 0, IPI_GET_CMD, MISC_CMD, NULL, NULL },
1064 
1065         /* 165 */ { SIOCGLIFCONF, 0, IPI_GET_CMD, MISC_CMD,
1066                         ip_sioctl_get_lifconf, NULL },
1067 
1068         /* 166 */ { SIOCSXARP, sizeof (struct xarpreq), IPI_PRIV | IPI_WR,
1069                         XARP_CMD, ip_sioctl_arp, NULL },
1070         /* 167 */ { SIOCGXARP, sizeof (struct xarpreq), IPI_GET_CMD,
1071                         XARP_CMD, ip_sioctl_arp, NULL },
1072         /* 168 */ { SIOCDXARP, sizeof (struct xarpreq), IPI_PRIV | IPI_WR,
1073                         XARP_CMD, ip_sioctl_arp, NULL },
1074 
1075         /* SIOCPOPSOCKFS is not handled by IP */
1076         /* 169 */ { IPI_DONTCARE /* SIOCPOPSOCKFS */, 0, 0, 0, NULL, NULL },
1077 
1078         /* 170 */ { SIOCGLIFZONE, sizeof (struct lifreq),
1079                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_lifzone, NULL },
1080         /* 171 */ { SIOCSLIFZONE, sizeof (struct lifreq),
1081                         IPI_PRIV | IPI_WR, LIF_CMD, ip_sioctl_slifzone,
1082                         ip_sioctl_slifzone_restart },
1083         /* 172-174 are SCTP ioctls and not handled by IP */
1084         /* 172 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1085         /* 173 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1086         /* 174 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1087         /* 175 */ { SIOCGLIFUSESRC, sizeof (struct lifreq),
1088                         IPI_GET_CMD, LIF_CMD,
1089                         ip_sioctl_get_lifusesrc, 0 },
1090         /* 176 */ { SIOCSLIFUSESRC, sizeof (struct lifreq),
1091                         IPI_PRIV | IPI_WR,
1092                         LIF_CMD, ip_sioctl_slifusesrc,
1093                         NULL },
1094         /* 177 */ { SIOCGLIFSRCOF, 0, IPI_GET_CMD, MISC_CMD,
1095                         ip_sioctl_get_lifsrcof, NULL },
1096         /* 178 */ { SIOCGMSFILTER, sizeof (struct group_filter), IPI_GET_CMD,
1097                         MSFILT_CMD, ip_sioctl_msfilter, NULL },
1098         /* 179 */ { SIOCSMSFILTER, sizeof (struct group_filter), 0,
1099                         MSFILT_CMD, ip_sioctl_msfilter, NULL },
1100         /* 180 */ { SIOCGIPMSFILTER, sizeof (struct ip_msfilter), IPI_GET_CMD,
1101                         MSFILT_CMD, ip_sioctl_msfilter, NULL },
1102         /* 181 */ { SIOCSIPMSFILTER, sizeof (struct ip_msfilter), 0,
1103                         MSFILT_CMD, ip_sioctl_msfilter, NULL },
1104         /* 182 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1105         /* SIOCSENABLESDP is handled by SDP */
1106         /* 183 */ { IPI_DONTCARE /* SIOCSENABLESDP */, 0, 0, 0, NULL, NULL },
1107         /* 184 */ { IPI_DONTCARE /* SIOCSQPTR */, 0, 0, 0, NULL, NULL },
1108         /* 185 */ { SIOCGIFHWADDR, sizeof (struct ifreq), IPI_GET_CMD,
1109                         IF_CMD, ip_sioctl_get_ifhwaddr, NULL },
1110         /* 186 */ { IPI_DONTCARE /* SIOCGSTAMP */, 0, 0, 0, NULL, NULL },
1111         /* 187 */ { SIOCILB, 0, IPI_PRIV | IPI_GET_CMD, MISC_CMD,
1112                         ip_sioctl_ilb_cmd, NULL },
1113         /* 188 */ { SIOCGETPROP, 0, IPI_GET_CMD, 0, NULL, NULL },
1114         /* 189 */ { SIOCSETPROP, 0, IPI_PRIV | IPI_WR, 0, NULL, NULL},
1115         /* 190 */ { SIOCGLIFDADSTATE, sizeof (struct lifreq),
1116                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_dadstate, NULL },
1117         /* 191 */ { SIOCSLIFPREFIX, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
1118                         LIF_CMD, ip_sioctl_prefix, ip_sioctl_prefix_restart },
1119         /* 192 */ { SIOCGLIFHWADDR, sizeof (struct lifreq), IPI_GET_CMD,
1120                         LIF_CMD, ip_sioctl_get_lifhwaddr, NULL }
1121 };
1122 
1123 int ip_ndx_ioctl_count = sizeof (ip_ndx_ioctl_table) / sizeof (ip_ioctl_cmd_t);
1124 
1125 ip_ioctl_cmd_t ip_misc_ioctl_table[] = {
1126         { I_LINK,       0, IPI_PRIV | IPI_WR, 0, NULL, NULL },
1127         { I_UNLINK,     0, IPI_PRIV | IPI_WR, 0, NULL, NULL },
1128         { I_PLINK,      0, IPI_PRIV | IPI_WR, 0, NULL, NULL },
1129         { I_PUNLINK,    0, IPI_PRIV | IPI_WR, 0, NULL, NULL },
1130         { ND_GET,       0, 0, 0, NULL, NULL },
1131         { ND_SET,       0, IPI_PRIV | IPI_WR, 0, NULL, NULL },
1132         { IP_IOCTL,     0, 0, 0, NULL, NULL },
1133         { SIOCGETVIFCNT, sizeof (struct sioc_vif_req), IPI_GET_CMD,
1134                 MISC_CMD, mrt_ioctl},
1135         { SIOCGETSGCNT, sizeof (struct sioc_sg_req), IPI_GET_CMD,
1136                 MISC_CMD, mrt_ioctl},
1137         { SIOCGETLSGCNT, sizeof (struct sioc_lsg_req), IPI_GET_CMD,
1138                 MISC_CMD, mrt_ioctl}
1139 };
1140 
1141 int ip_misc_ioctl_count =
1142     sizeof (ip_misc_ioctl_table) / sizeof (ip_ioctl_cmd_t);
1143 
1144 int     conn_drain_nthreads;            /* Number of drainers reqd. */
1145                                         /* Settable in /etc/system */
1146 /* Defined in ip_ire.c */
1147 extern uint32_t ip_ire_max_bucket_cnt, ip6_ire_max_bucket_cnt;
1148 extern uint32_t ip_ire_min_bucket_cnt, ip6_ire_min_bucket_cnt;
1149 extern uint32_t ip_ire_mem_ratio, ip_ire_cpu_ratio;
1150 
1151 static nv_t     ire_nv_arr[] = {
1152         { IRE_BROADCAST, "BROADCAST" },
1153         { IRE_LOCAL, "LOCAL" },
1154         { IRE_LOOPBACK, "LOOPBACK" },
1155         { IRE_DEFAULT, "DEFAULT" },
1156         { IRE_PREFIX, "PREFIX" },
1157         { IRE_IF_NORESOLVER, "IF_NORESOL" },
1158         { IRE_IF_RESOLVER, "IF_RESOLV" },
1159         { IRE_IF_CLONE, "IF_CLONE" },
1160         { IRE_HOST, "HOST" },
1161         { IRE_MULTICAST, "MULTICAST" },
1162         { IRE_NOROUTE, "NOROUTE" },
1163         { 0 }
1164 };
1165 
1166 nv_t    *ire_nv_tbl = ire_nv_arr;
1167 
1168 /* Simple ICMP IP Header Template */
1169 static ipha_t icmp_ipha = {
1170         IP_SIMPLE_HDR_VERSION, 0, 0, 0, 0, 0, IPPROTO_ICMP
1171 };
1172 
1173 struct module_info ip_mod_info = {
1174         IP_MOD_ID, IP_MOD_NAME, IP_MOD_MINPSZ, IP_MOD_MAXPSZ, IP_MOD_HIWAT,
1175         IP_MOD_LOWAT
1176 };
1177 
1178 /*
1179  * Duplicate static symbols within a module confuses mdb; so we avoid the
1180  * problem by making the symbols here distinct from those in udp.c.
1181  */
1182 
1183 /*
1184  * Entry points for IP as a device and as a module.
1185  * We have separate open functions for the /dev/ip and /dev/ip6 devices.
1186  */
1187 static struct qinit iprinitv4 = {
1188         ip_rput, NULL, ip_openv4, ip_close, NULL, &ip_mod_info
1189 };
1190 
1191 struct qinit iprinitv6 = {
1192         ip_rput_v6, NULL, ip_openv6, ip_close, NULL, &ip_mod_info
1193 };
1194 
1195 static struct qinit ipwinit = {
1196         ip_wput_nondata, ip_wsrv, NULL, NULL, NULL, &ip_mod_info
1197 };
1198 
1199 static struct qinit iplrinit = {
1200         ip_lrput, NULL, ip_openv4, ip_close, NULL, &ip_mod_info
1201 };
1202 
1203 static struct qinit iplwinit = {
1204         ip_lwput, NULL, NULL, NULL, NULL, &ip_mod_info
1205 };
1206 
1207 /* For AF_INET aka /dev/ip */
1208 struct streamtab ipinfov4 = {
1209         &iprinitv4, &ipwinit, &iplrinit, &iplwinit
1210 };
1211 
1212 /* For AF_INET6 aka /dev/ip6 */
1213 struct streamtab ipinfov6 = {
1214         &iprinitv6, &ipwinit, &iplrinit, &iplwinit
1215 };
1216 
1217 #ifdef  DEBUG
1218 boolean_t skip_sctp_cksum = B_FALSE;
1219 #endif
1220 
1221 /*
1222  * Generate an ICMP fragmentation needed message.
1223  * When called from ip_output side a minimal ip_recv_attr_t needs to be
1224  * constructed by the caller.
1225  */
1226 void
1227 icmp_frag_needed(mblk_t *mp, int mtu, ip_recv_attr_t *ira)
1228 {
1229         icmph_t icmph;
1230         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
1231 
1232         mp = icmp_pkt_err_ok(mp, ira);
1233         if (mp == NULL)
1234                 return;
1235 
1236         bzero(&icmph, sizeof (icmph_t));
1237         icmph.icmph_type = ICMP_DEST_UNREACHABLE;
1238         icmph.icmph_code = ICMP_FRAGMENTATION_NEEDED;
1239         icmph.icmph_du_mtu = htons((uint16_t)mtu);
1240         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutFragNeeded);
1241         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDestUnreachs);
1242 
1243         icmp_pkt(mp, &icmph, sizeof (icmph_t), ira);
1244 }
1245 
1246 /*
1247  * icmp_inbound_v4 deals with ICMP messages that are handled by IP.
1248  * If the ICMP message is consumed by IP, i.e., it should not be delivered
1249  * to any IPPROTO_ICMP raw sockets, then it returns NULL.
1250  * Likewise, if the ICMP error is misformed (too short, etc), then it
1251  * returns NULL. The caller uses this to determine whether or not to send
1252  * to raw sockets.
1253  *
1254  * All error messages are passed to the matching transport stream.
1255  *
1256  * The following cases are handled by icmp_inbound:
1257  * 1) It needs to send a reply back and possibly delivering it
1258  *    to the "interested" upper clients.
1259  * 2) Return the mblk so that the caller can pass it to the RAW socket clients.
1260  * 3) It needs to change some values in IP only.
1261  * 4) It needs to change some values in IP and upper layers e.g TCP
1262  *    by delivering an error to the upper layers.
1263  *
1264  * We handle the above three cases in the context of IPsec in the
1265  * following way :
1266  *
1267  * 1) Send the reply back in the same way as the request came in.
1268  *    If it came in encrypted, it goes out encrypted. If it came in
1269  *    clear, it goes out in clear. Thus, this will prevent chosen
1270  *    plain text attack.
1271  * 2) The client may or may not expect things to come in secure.
1272  *    If it comes in secure, the policy constraints are checked
1273  *    before delivering it to the upper layers. If it comes in
1274  *    clear, ipsec_inbound_accept_clear will decide whether to
1275  *    accept this in clear or not. In both the cases, if the returned
1276  *    message (IP header + 8 bytes) that caused the icmp message has
1277  *    AH/ESP headers, it is sent up to AH/ESP for validation before
1278  *    sending up. If there are only 8 bytes of returned message, then
1279  *    upper client will not be notified.
1280  * 3) Check with global policy to see whether it matches the constaints.
1281  *    But this will be done only if icmp_accept_messages_in_clear is
1282  *    zero.
1283  * 4) If we need to change both in IP and ULP, then the decision taken
1284  *    while affecting the values in IP and while delivering up to TCP
1285  *    should be the same.
1286  *
1287  *      There are two cases.
1288  *
1289  *      a) If we reject data at the IP layer (ipsec_check_global_policy()
1290  *         failed), we will not deliver it to the ULP, even though they
1291  *         are *willing* to accept in *clear*. This is fine as our global
1292  *         disposition to icmp messages asks us reject the datagram.
1293  *
1294  *      b) If we accept data at the IP layer (ipsec_check_global_policy()
1295  *         succeeded or icmp_accept_messages_in_clear is 1), and not able
1296  *         to deliver it to ULP (policy failed), it can lead to
1297  *         consistency problems. The cases known at this time are
1298  *         ICMP_DESTINATION_UNREACHABLE  messages with following code
1299  *         values :
1300  *
1301  *         - ICMP_FRAGMENTATION_NEEDED : IP adapts to the new value
1302  *           and Upper layer rejects. Then the communication will
1303  *           come to a stop. This is solved by making similar decisions
1304  *           at both levels. Currently, when we are unable to deliver
1305  *           to the Upper Layer (due to policy failures) while IP has
1306  *           adjusted dce_pmtu, the next outbound datagram would
1307  *           generate a local ICMP_FRAGMENTATION_NEEDED message - which
1308  *           will be with the right level of protection. Thus the right
1309  *           value will be communicated even if we are not able to
1310  *           communicate when we get from the wire initially. But this
1311  *           assumes there would be at least one outbound datagram after
1312  *           IP has adjusted its dce_pmtu value. To make things
1313  *           simpler, we accept in clear after the validation of
1314  *           AH/ESP headers.
1315  *
1316  *         - Other ICMP ERRORS : We may not be able to deliver it to the
1317  *           upper layer depending on the level of protection the upper
1318  *           layer expects and the disposition in ipsec_inbound_accept_clear().
1319  *           ipsec_inbound_accept_clear() decides whether a given ICMP error
1320  *           should be accepted in clear when the Upper layer expects secure.
1321  *           Thus the communication may get aborted by some bad ICMP
1322  *           packets.
1323  */
1324 mblk_t *
1325 icmp_inbound_v4(mblk_t *mp, ip_recv_attr_t *ira)
1326 {
1327         icmph_t         *icmph;
1328         ipha_t          *ipha;          /* Outer header */
1329         int             ip_hdr_length;  /* Outer header length */
1330         boolean_t       interested;
1331         ipif_t          *ipif;
1332         uint32_t        ts;
1333         uint32_t        *tsp;
1334         timestruc_t     now;
1335         ill_t           *ill = ira->ira_ill;
1336         ip_stack_t      *ipst = ill->ill_ipst;
1337         zoneid_t        zoneid = ira->ira_zoneid;
1338         int             len_needed;
1339         mblk_t          *mp_ret = NULL;
1340 
1341         ipha = (ipha_t *)mp->b_rptr;
1342 
1343         BUMP_MIB(&ipst->ips_icmp_mib, icmpInMsgs);
1344 
1345         ip_hdr_length = ira->ira_ip_hdr_length;
1346         if ((mp->b_wptr - mp->b_rptr) < (ip_hdr_length + ICMPH_SIZE)) {
1347                 if (ira->ira_pktlen < (ip_hdr_length + ICMPH_SIZE)) {
1348                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInTruncatedPkts);
1349                         ip_drop_input("ipIfStatsInTruncatedPkts", mp, ill);
1350                         freemsg(mp);
1351                         return (NULL);
1352                 }
1353                 /* Last chance to get real. */
1354                 ipha = ip_pullup(mp, ip_hdr_length + ICMPH_SIZE, ira);
1355                 if (ipha == NULL) {
1356                         BUMP_MIB(&ipst->ips_icmp_mib, icmpInErrors);
1357                         freemsg(mp);
1358                         return (NULL);
1359                 }
1360         }
1361 
1362         /* The IP header will always be a multiple of four bytes */
1363         icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1364         ip2dbg(("icmp_inbound_v4: type %d code %d\n", icmph->icmph_type,
1365             icmph->icmph_code));
1366 
1367         /*
1368          * We will set "interested" to "true" if we should pass a copy to
1369          * the transport or if we handle the packet locally.
1370          */
1371         interested = B_FALSE;
1372         switch (icmph->icmph_type) {
1373         case ICMP_ECHO_REPLY:
1374                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInEchoReps);
1375                 break;
1376         case ICMP_DEST_UNREACHABLE:
1377                 if (icmph->icmph_code == ICMP_FRAGMENTATION_NEEDED)
1378                         BUMP_MIB(&ipst->ips_icmp_mib, icmpInFragNeeded);
1379                 interested = B_TRUE;    /* Pass up to transport */
1380                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInDestUnreachs);
1381                 break;
1382         case ICMP_SOURCE_QUENCH:
1383                 interested = B_TRUE;    /* Pass up to transport */
1384                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInSrcQuenchs);
1385                 break;
1386         case ICMP_REDIRECT:
1387                 if (!ipst->ips_ip_ignore_redirect)
1388                         interested = B_TRUE;
1389                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInRedirects);
1390                 break;
1391         case ICMP_ECHO_REQUEST:
1392                 /*
1393                  * Whether to respond to echo requests that come in as IP
1394                  * broadcasts or as IP multicast is subject to debate
1395                  * (what isn't?).  We aim to please, you pick it.
1396                  * Default is do it.
1397                  */
1398                 if (ira->ira_flags & IRAF_MULTICAST) {
1399                         /* multicast: respond based on tunable */
1400                         interested = ipst->ips_ip_g_resp_to_echo_mcast;
1401                 } else if (ira->ira_flags & IRAF_BROADCAST) {
1402                         /* broadcast: respond based on tunable */
1403                         interested = ipst->ips_ip_g_resp_to_echo_bcast;
1404                 } else {
1405                         /* unicast: always respond */
1406                         interested = B_TRUE;
1407                 }
1408                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInEchos);
1409                 if (!interested) {
1410                         /* We never pass these to RAW sockets */
1411                         freemsg(mp);
1412                         return (NULL);
1413                 }
1414 
1415                 /* Check db_ref to make sure we can modify the packet. */
1416                 if (mp->b_datap->db_ref > 1) {
1417                         mblk_t  *mp1;
1418 
1419                         mp1 = copymsg(mp);
1420                         freemsg(mp);
1421                         if (!mp1) {
1422                                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
1423                                 return (NULL);
1424                         }
1425                         mp = mp1;
1426                         ipha = (ipha_t *)mp->b_rptr;
1427                         icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1428                 }
1429                 icmph->icmph_type = ICMP_ECHO_REPLY;
1430                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutEchoReps);
1431                 icmp_send_reply_v4(mp, ipha, icmph, ira);
1432                 return (NULL);
1433 
1434         case ICMP_ROUTER_ADVERTISEMENT:
1435         case ICMP_ROUTER_SOLICITATION:
1436                 break;
1437         case ICMP_TIME_EXCEEDED:
1438                 interested = B_TRUE;    /* Pass up to transport */
1439                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInTimeExcds);
1440                 break;
1441         case ICMP_PARAM_PROBLEM:
1442                 interested = B_TRUE;    /* Pass up to transport */
1443                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInParmProbs);
1444                 break;
1445         case ICMP_TIME_STAMP_REQUEST:
1446                 /* Response to Time Stamp Requests is local policy. */
1447                 if (ipst->ips_ip_g_resp_to_timestamp) {
1448                         if (ira->ira_flags & IRAF_MULTIBROADCAST)
1449                                 interested =
1450                                     ipst->ips_ip_g_resp_to_timestamp_bcast;
1451                         else
1452                                 interested = B_TRUE;
1453                 }
1454                 if (!interested) {
1455                         /* We never pass these to RAW sockets */
1456                         freemsg(mp);
1457                         return (NULL);
1458                 }
1459 
1460                 /* Make sure we have enough of the packet */
1461                 len_needed = ip_hdr_length + ICMPH_SIZE +
1462                     3 * sizeof (uint32_t);
1463 
1464                 if (mp->b_wptr - mp->b_rptr < len_needed) {
1465                         ipha = ip_pullup(mp, len_needed, ira);
1466                         if (ipha == NULL) {
1467                                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1468                                 ip_drop_input("ipIfStatsInDiscards - ip_pullup",
1469                                     mp, ill);
1470                                 freemsg(mp);
1471                                 return (NULL);
1472                         }
1473                         /* Refresh following the pullup. */
1474                         icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1475                 }
1476                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInTimestamps);
1477                 /* Check db_ref to make sure we can modify the packet. */
1478                 if (mp->b_datap->db_ref > 1) {
1479                         mblk_t  *mp1;
1480 
1481                         mp1 = copymsg(mp);
1482                         freemsg(mp);
1483                         if (!mp1) {
1484                                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
1485                                 return (NULL);
1486                         }
1487                         mp = mp1;
1488                         ipha = (ipha_t *)mp->b_rptr;
1489                         icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1490                 }
1491                 icmph->icmph_type = ICMP_TIME_STAMP_REPLY;
1492                 tsp = (uint32_t *)&icmph[1];
1493                 tsp++;          /* Skip past 'originate time' */
1494                 /* Compute # of milliseconds since midnight */
1495                 gethrestime(&now);
1496                 ts = (now.tv_sec % (24 * 60 * 60)) * 1000 +
1497                     NSEC2MSEC(now.tv_nsec);
1498                 *tsp++ = htonl(ts);     /* Lay in 'receive time' */
1499                 *tsp++ = htonl(ts);     /* Lay in 'send time' */
1500                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutTimestampReps);
1501                 icmp_send_reply_v4(mp, ipha, icmph, ira);
1502                 return (NULL);
1503 
1504         case ICMP_TIME_STAMP_REPLY:
1505                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInTimestampReps);
1506                 break;
1507         case ICMP_INFO_REQUEST:
1508                 /* Per RFC 1122 3.2.2.7, ignore this. */
1509         case ICMP_INFO_REPLY:
1510                 break;
1511         case ICMP_ADDRESS_MASK_REQUEST:
1512                 if (ira->ira_flags & IRAF_MULTIBROADCAST) {
1513                         interested =
1514                             ipst->ips_ip_respond_to_address_mask_broadcast;
1515                 } else {
1516                         interested = B_TRUE;
1517                 }
1518                 if (!interested) {
1519                         /* We never pass these to RAW sockets */
1520                         freemsg(mp);
1521                         return (NULL);
1522                 }
1523                 len_needed = ip_hdr_length + ICMPH_SIZE + IP_ADDR_LEN;
1524                 if (mp->b_wptr - mp->b_rptr < len_needed) {
1525                         ipha = ip_pullup(mp, len_needed, ira);
1526                         if (ipha == NULL) {
1527                                 BUMP_MIB(ill->ill_ip_mib,
1528                                     ipIfStatsInTruncatedPkts);
1529                                 ip_drop_input("ipIfStatsInTruncatedPkts", mp,
1530                                     ill);
1531                                 freemsg(mp);
1532                                 return (NULL);
1533                         }
1534                         /* Refresh following the pullup. */
1535                         icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1536                 }
1537                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInAddrMasks);
1538                 /* Check db_ref to make sure we can modify the packet. */
1539                 if (mp->b_datap->db_ref > 1) {
1540                         mblk_t  *mp1;
1541 
1542                         mp1 = copymsg(mp);
1543                         freemsg(mp);
1544                         if (!mp1) {
1545                                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
1546                                 return (NULL);
1547                         }
1548                         mp = mp1;
1549                         ipha = (ipha_t *)mp->b_rptr;
1550                         icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1551                 }
1552                 /*
1553                  * Need the ipif with the mask be the same as the source
1554                  * address of the mask reply. For unicast we have a specific
1555                  * ipif. For multicast/broadcast we only handle onlink
1556                  * senders, and use the source address to pick an ipif.
1557                  */
1558                 ipif = ipif_lookup_addr(ipha->ipha_dst, ill, zoneid, ipst);
1559                 if (ipif == NULL) {
1560                         /* Broadcast or multicast */
1561                         ipif = ipif_lookup_remote(ill, ipha->ipha_src, zoneid);
1562                         if (ipif == NULL) {
1563                                 freemsg(mp);
1564                                 return (NULL);
1565                         }
1566                 }
1567                 icmph->icmph_type = ICMP_ADDRESS_MASK_REPLY;
1568                 bcopy(&ipif->ipif_net_mask, &icmph[1], IP_ADDR_LEN);
1569                 ipif_refrele(ipif);
1570                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutAddrMaskReps);
1571                 icmp_send_reply_v4(mp, ipha, icmph, ira);
1572                 return (NULL);
1573 
1574         case ICMP_ADDRESS_MASK_REPLY:
1575                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInAddrMaskReps);
1576                 break;
1577         default:
1578                 interested = B_TRUE;    /* Pass up to transport */
1579                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInUnknowns);
1580                 break;
1581         }
1582         /*
1583          * See if there is an ICMP client to avoid an extra copymsg/freemsg
1584          * if there isn't one.
1585          */
1586         if (ipst->ips_ipcl_proto_fanout_v4[IPPROTO_ICMP].connf_head != NULL) {
1587                 /* If there is an ICMP client and we want one too, copy it. */
1588 
1589                 if (!interested) {
1590                         /* Caller will deliver to RAW sockets */
1591                         return (mp);
1592                 }
1593                 mp_ret = copymsg(mp);
1594                 if (mp_ret == NULL) {
1595                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1596                         ip_drop_input("ipIfStatsInDiscards - copymsg", mp, ill);
1597                 }
1598         } else if (!interested) {
1599                 /* Neither we nor raw sockets are interested. Drop packet now */
1600                 freemsg(mp);
1601                 return (NULL);
1602         }
1603 
1604         /*
1605          * ICMP error or redirect packet. Make sure we have enough of
1606          * the header and that db_ref == 1 since we might end up modifying
1607          * the packet.
1608          */
1609         if (mp->b_cont != NULL) {
1610                 if (ip_pullup(mp, -1, ira) == NULL) {
1611                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1612                         ip_drop_input("ipIfStatsInDiscards - ip_pullup",
1613                             mp, ill);
1614                         freemsg(mp);
1615                         return (mp_ret);
1616                 }
1617         }
1618 
1619         if (mp->b_datap->db_ref > 1) {
1620                 mblk_t  *mp1;
1621 
1622                 mp1 = copymsg(mp);
1623                 if (mp1 == NULL) {
1624                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1625                         ip_drop_input("ipIfStatsInDiscards - copymsg", mp, ill);
1626                         freemsg(mp);
1627                         return (mp_ret);
1628                 }
1629                 freemsg(mp);
1630                 mp = mp1;
1631         }
1632 
1633         /*
1634          * In case mp has changed, verify the message before any further
1635          * processes.
1636          */
1637         ipha = (ipha_t *)mp->b_rptr;
1638         icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1639         if (!icmp_inbound_verify_v4(mp, icmph, ira)) {
1640                 freemsg(mp);
1641                 return (mp_ret);
1642         }
1643 
1644         switch (icmph->icmph_type) {
1645         case ICMP_REDIRECT:
1646                 icmp_redirect_v4(mp, ipha, icmph, ira);
1647                 break;
1648         case ICMP_DEST_UNREACHABLE:
1649                 if (icmph->icmph_code == ICMP_FRAGMENTATION_NEEDED) {
1650                         /* Update DCE and adjust MTU is icmp header if needed */
1651                         icmp_inbound_too_big_v4(icmph, ira);
1652                 }
1653                 /* FALLTHROUGH */
1654         default:
1655                 icmp_inbound_error_fanout_v4(mp, icmph, ira);
1656                 break;
1657         }
1658         return (mp_ret);
1659 }
1660 
1661 /*
1662  * Send an ICMP echo, timestamp or address mask reply.
1663  * The caller has already updated the payload part of the packet.
1664  * We handle the ICMP checksum, IP source address selection and feed
1665  * the packet into ip_output_simple.
1666  */
1667 static void
1668 icmp_send_reply_v4(mblk_t *mp, ipha_t *ipha, icmph_t *icmph,
1669     ip_recv_attr_t *ira)
1670 {
1671         uint_t          ip_hdr_length = ira->ira_ip_hdr_length;
1672         ill_t           *ill = ira->ira_ill;
1673         ip_stack_t      *ipst = ill->ill_ipst;
1674         ip_xmit_attr_t  ixas;
1675 
1676         /* Send out an ICMP packet */
1677         icmph->icmph_checksum = 0;
1678         icmph->icmph_checksum = IP_CSUM(mp, ip_hdr_length, 0);
1679         /* Reset time to live. */
1680         ipha->ipha_ttl = ipst->ips_ip_def_ttl;
1681         {
1682                 /* Swap source and destination addresses */
1683                 ipaddr_t tmp;
1684 
1685                 tmp = ipha->ipha_src;
1686                 ipha->ipha_src = ipha->ipha_dst;
1687                 ipha->ipha_dst = tmp;
1688         }
1689         ipha->ipha_ident = 0;
1690         if (!IS_SIMPLE_IPH(ipha))
1691                 icmp_options_update(ipha);
1692 
1693         bzero(&ixas, sizeof (ixas));
1694         ixas.ixa_flags = IXAF_BASIC_SIMPLE_V4;
1695         ixas.ixa_zoneid = ira->ira_zoneid;
1696         ixas.ixa_cred = kcred;
1697         ixas.ixa_cpid = NOPID;
1698         ixas.ixa_tsl = ira->ira_tsl; /* Behave as a multi-level responder */
1699         ixas.ixa_ifindex = 0;
1700         ixas.ixa_ipst = ipst;
1701         ixas.ixa_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1702 
1703         if (!(ira->ira_flags & IRAF_IPSEC_SECURE)) {
1704                 /*
1705                  * This packet should go out the same way as it
1706                  * came in i.e in clear, independent of the IPsec policy
1707                  * for transmitting packets.
1708                  */
1709                 ixas.ixa_flags |= IXAF_NO_IPSEC;
1710         } else {
1711                 if (!ipsec_in_to_out(ira, &ixas, mp, ipha, NULL)) {
1712                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1713                         /* Note: mp already consumed and ip_drop_packet done */
1714                         return;
1715                 }
1716         }
1717         if (ira->ira_flags & IRAF_MULTIBROADCAST) {
1718                 /*
1719                  * Not one or our addresses (IRE_LOCALs), thus we let
1720                  * ip_output_simple pick the source.
1721                  */
1722                 ipha->ipha_src = INADDR_ANY;
1723                 ixas.ixa_flags |= IXAF_SET_SOURCE;
1724         }
1725         /* Should we send with DF and use dce_pmtu? */
1726         if (ipst->ips_ipv4_icmp_return_pmtu) {
1727                 ixas.ixa_flags |= IXAF_PMTU_DISCOVERY;
1728                 ipha->ipha_fragment_offset_and_flags |= IPH_DF_HTONS;
1729         }
1730 
1731         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutMsgs);
1732 
1733         (void) ip_output_simple(mp, &ixas);
1734         ixa_cleanup(&ixas);
1735 }
1736 
1737 /*
1738  * Verify the ICMP messages for either for ICMP error or redirect packet.
1739  * The caller should have fully pulled up the message. If it's a redirect
1740  * packet, only basic checks on IP header will be done; otherwise, verify
1741  * the packet by looking at the included ULP header.
1742  *
1743  * Called before icmp_inbound_error_fanout_v4 is called.
1744  */
1745 static boolean_t
1746 icmp_inbound_verify_v4(mblk_t *mp, icmph_t *icmph, ip_recv_attr_t *ira)
1747 {
1748         ill_t           *ill = ira->ira_ill;
1749         int             hdr_length;
1750         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
1751         conn_t          *connp;
1752         ipha_t          *ipha;  /* Inner IP header */
1753 
1754         ipha = (ipha_t *)&icmph[1];
1755         if ((uchar_t *)ipha + IP_SIMPLE_HDR_LENGTH > mp->b_wptr)
1756                 goto truncated;
1757 
1758         hdr_length = IPH_HDR_LENGTH(ipha);
1759 
1760         if ((IPH_HDR_VERSION(ipha) != IPV4_VERSION))
1761                 goto discard_pkt;
1762 
1763         if (hdr_length < sizeof (ipha_t))
1764                 goto truncated;
1765 
1766         if ((uchar_t *)ipha + hdr_length > mp->b_wptr)
1767                 goto truncated;
1768 
1769         /*
1770          * Stop here for ICMP_REDIRECT.
1771          */
1772         if (icmph->icmph_type == ICMP_REDIRECT)
1773                 return (B_TRUE);
1774 
1775         /*
1776          * ICMP errors only.
1777          */
1778         switch (ipha->ipha_protocol) {
1779         case IPPROTO_UDP:
1780                 /*
1781                  * Verify we have at least ICMP_MIN_TP_HDR_LEN bytes of
1782                  * transport header.
1783                  */
1784                 if ((uchar_t *)ipha + hdr_length + ICMP_MIN_TP_HDR_LEN >
1785                     mp->b_wptr)
1786                         goto truncated;
1787                 break;
1788         case IPPROTO_TCP: {
1789                 tcpha_t         *tcpha;
1790 
1791                 /*
1792                  * Verify we have at least ICMP_MIN_TP_HDR_LEN bytes of
1793                  * transport header.
1794                  */
1795                 if ((uchar_t *)ipha + hdr_length + ICMP_MIN_TP_HDR_LEN >
1796                     mp->b_wptr)
1797                         goto truncated;
1798 
1799                 tcpha = (tcpha_t *)((uchar_t *)ipha + hdr_length);
1800                 connp = ipcl_tcp_lookup_reversed_ipv4(ipha, tcpha, TCPS_LISTEN,
1801                     ipst);
1802                 if (connp == NULL)
1803                         goto discard_pkt;
1804 
1805                 if ((connp->conn_verifyicmp != NULL) &&
1806                     !connp->conn_verifyicmp(connp, tcpha, icmph, NULL, ira)) {
1807                         CONN_DEC_REF(connp);
1808                         goto discard_pkt;
1809                 }
1810                 CONN_DEC_REF(connp);
1811                 break;
1812         }
1813         case IPPROTO_SCTP:
1814                 /*
1815                  * Verify we have at least ICMP_MIN_TP_HDR_LEN bytes of
1816                  * transport header.
1817                  */
1818                 if ((uchar_t *)ipha + hdr_length + ICMP_MIN_TP_HDR_LEN >
1819                     mp->b_wptr)
1820                         goto truncated;
1821                 break;
1822         case IPPROTO_ESP:
1823         case IPPROTO_AH:
1824                 break;
1825         case IPPROTO_ENCAP:
1826                 if ((uchar_t *)ipha + hdr_length + sizeof (ipha_t) >
1827                     mp->b_wptr)
1828                         goto truncated;
1829                 break;
1830         default:
1831                 break;
1832         }
1833 
1834         return (B_TRUE);
1835 
1836 discard_pkt:
1837         /* Bogus ICMP error. */
1838         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1839         return (B_FALSE);
1840 
1841 truncated:
1842         /* We pulled up everthing already. Must be truncated */
1843         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInTruncatedPkts);
1844         ip_drop_input("ipIfStatsInTruncatedPkts", mp, ill);
1845         return (B_FALSE);
1846 }
1847 
1848 /* Table from RFC 1191 */
1849 static int icmp_frag_size_table[] =
1850 { 32000, 17914, 8166, 4352, 2002, 1496, 1006, 508, 296, 68 };
1851 
1852 /*
1853  * Process received ICMP Packet too big.
1854  * Just handles the DCE create/update, including using the above table of
1855  * PMTU guesses. The caller is responsible for validating the packet before
1856  * passing it in and also to fanout the ICMP error to any matching transport
1857  * conns. Assumes the message has been fully pulled up and verified.
1858  *
1859  * Before getting here, the caller has called icmp_inbound_verify_v4()
1860  * that should have verified with ULP to prevent undoing the changes we're
1861  * going to make to DCE. For example, TCP might have verified that the packet
1862  * which generated error is in the send window.
1863  *
1864  * In some cases modified this MTU in the ICMP header packet; the caller
1865  * should pass to the matching ULP after this returns.
1866  */
1867 static void
1868 icmp_inbound_too_big_v4(icmph_t *icmph, ip_recv_attr_t *ira)
1869 {
1870         dce_t           *dce;
1871         int             old_mtu;
1872         int             mtu, orig_mtu;
1873         ipaddr_t        dst;
1874         boolean_t       disable_pmtud;
1875         ill_t           *ill = ira->ira_ill;
1876         ip_stack_t      *ipst = ill->ill_ipst;
1877         uint_t          hdr_length;
1878         ipha_t          *ipha;
1879 
1880         /* Caller already pulled up everything. */
1881         ipha = (ipha_t *)&icmph[1];
1882         ASSERT(icmph->icmph_type == ICMP_DEST_UNREACHABLE &&
1883             icmph->icmph_code == ICMP_FRAGMENTATION_NEEDED);
1884         ASSERT(ill != NULL);
1885 
1886         hdr_length = IPH_HDR_LENGTH(ipha);
1887 
1888         /*
1889          * We handle path MTU for source routed packets since the DCE
1890          * is looked up using the final destination.
1891          */
1892         dst = ip_get_dst(ipha);
1893 
1894         dce = dce_lookup_and_add_v4(dst, ipst);
1895         if (dce == NULL) {
1896                 /* Couldn't add a unique one - ENOMEM */
1897                 ip1dbg(("icmp_inbound_too_big_v4: no dce for 0x%x\n",
1898                     ntohl(dst)));
1899                 return;
1900         }
1901 
1902         /* Check for MTU discovery advice as described in RFC 1191 */
1903         mtu = ntohs(icmph->icmph_du_mtu);
1904         orig_mtu = mtu;
1905         disable_pmtud = B_FALSE;
1906 
1907         mutex_enter(&dce->dce_lock);
1908         if (dce->dce_flags & DCEF_PMTU)
1909                 old_mtu = dce->dce_pmtu;
1910         else
1911                 old_mtu = ill->ill_mtu;
1912 
1913         if (icmph->icmph_du_zero != 0 || mtu < ipst->ips_ip_pmtu_min) {
1914                 uint32_t length;
1915                 int     i;
1916 
1917                 /*
1918                  * Use the table from RFC 1191 to figure out
1919                  * the next "plateau" based on the length in
1920                  * the original IP packet.
1921                  */
1922                 length = ntohs(ipha->ipha_length);
1923                 DTRACE_PROBE2(ip4__pmtu__guess, dce_t *, dce,
1924                     uint32_t, length);
1925                 if (old_mtu <= length &&
1926                     old_mtu >= length - hdr_length) {
1927                         /*
1928                          * Handle broken BSD 4.2 systems that
1929                          * return the wrong ipha_length in ICMP
1930                          * errors.
1931                          */
1932                         ip1dbg(("Wrong mtu: sent %d, dce %d\n",
1933                             length, old_mtu));
1934                         length -= hdr_length;
1935                 }
1936                 for (i = 0; i < A_CNT(icmp_frag_size_table); i++) {
1937                         if (length > icmp_frag_size_table[i])
1938                                 break;
1939                 }
1940                 if (i == A_CNT(icmp_frag_size_table)) {
1941                         /* Smaller than IP_MIN_MTU! */
1942                         ip1dbg(("Too big for packet size %d\n",
1943                             length));
1944                         disable_pmtud = B_TRUE;
1945                         mtu = ipst->ips_ip_pmtu_min;
1946                 } else {
1947                         mtu = icmp_frag_size_table[i];
1948                         ip1dbg(("Calculated mtu %d, packet size %d, "
1949                             "before %d\n", mtu, length, old_mtu));
1950                         if (mtu < ipst->ips_ip_pmtu_min) {
1951                                 mtu = ipst->ips_ip_pmtu_min;
1952                                 disable_pmtud = B_TRUE;
1953                         }
1954                 }
1955         }
1956         if (disable_pmtud)
1957                 dce->dce_flags |= DCEF_TOO_SMALL_PMTU;
1958         else
1959                 dce->dce_flags &= ~DCEF_TOO_SMALL_PMTU;
1960 
1961         dce->dce_pmtu = MIN(old_mtu, mtu);
1962         /* Prepare to send the new max frag size for the ULP. */
1963         icmph->icmph_du_zero = 0;
1964         icmph->icmph_du_mtu =  htons((uint16_t)dce->dce_pmtu);
1965         DTRACE_PROBE4(ip4__pmtu__change, icmph_t *, icmph, dce_t *,
1966             dce, int, orig_mtu, int, mtu);
1967 
1968         /* We now have a PMTU for sure */
1969         dce->dce_flags |= DCEF_PMTU;
1970         dce->dce_last_change_time = TICK_TO_SEC(ddi_get_lbolt64());
1971         mutex_exit(&dce->dce_lock);
1972         /*
1973          * After dropping the lock the new value is visible to everyone.
1974          * Then we bump the generation number so any cached values reinspect
1975          * the dce_t.
1976          */
1977         dce_increment_generation(dce);
1978         dce_refrele(dce);
1979 }
1980 
1981 /*
1982  * If the packet in error is Self-Encapsulated, icmp_inbound_error_fanout_v4
1983  * calls this function.
1984  */
1985 static mblk_t *
1986 icmp_inbound_self_encap_error_v4(mblk_t *mp, ipha_t *ipha, ipha_t *in_ipha)
1987 {
1988         int length;
1989 
1990         ASSERT(mp->b_datap->db_type == M_DATA);
1991 
1992         /* icmp_inbound_v4 has already pulled up the whole error packet */
1993         ASSERT(mp->b_cont == NULL);
1994 
1995         /*
1996          * The length that we want to overlay is the inner header
1997          * and what follows it.
1998          */
1999         length = msgdsize(mp) - ((uchar_t *)in_ipha - mp->b_rptr);
2000 
2001         /*
2002          * Overlay the inner header and whatever follows it over the
2003          * outer header.
2004          */
2005         bcopy((uchar_t *)in_ipha, (uchar_t *)ipha, length);
2006 
2007         /* Adjust for what we removed */
2008         mp->b_wptr -= (uchar_t *)in_ipha - (uchar_t *)ipha;
2009         return (mp);
2010 }
2011 
2012 /*
2013  * Try to pass the ICMP message upstream in case the ULP cares.
2014  *
2015  * If the packet that caused the ICMP error is secure, we send
2016  * it to AH/ESP to make sure that the attached packet has a
2017  * valid association. ipha in the code below points to the
2018  * IP header of the packet that caused the error.
2019  *
2020  * For IPsec cases, we let the next-layer-up (which has access to
2021  * cached policy on the conn_t, or can query the SPD directly)
2022  * subtract out any IPsec overhead if they must.  We therefore make no
2023  * adjustments here for IPsec overhead.
2024  *
2025  * IFN could have been generated locally or by some router.
2026  *
2027  * LOCAL : ire_send_wire (before calling ipsec_out_process) can call
2028  * icmp_frag_needed/icmp_pkt2big_v6 to generated a local IFN.
2029  *          This happens because IP adjusted its value of MTU on an
2030  *          earlier IFN message and could not tell the upper layer,
2031  *          the new adjusted value of MTU e.g. Packet was encrypted
2032  *          or there was not enough information to fanout to upper
2033  *          layers. Thus on the next outbound datagram, ire_send_wire
2034  *          generates the IFN, where IPsec processing has *not* been
2035  *          done.
2036  *
2037  *          Note that we retain ixa_fragsize across IPsec thus once
2038  *          we have picking ixa_fragsize and entered ipsec_out_process we do
2039  *          no change the fragsize even if the path MTU changes before
2040  *          we reach ip_output_post_ipsec.
2041  *
2042  *          In the local case, IRAF_LOOPBACK will be set indicating
2043  *          that IFN was generated locally.
2044  *
2045  * ROUTER : IFN could be secure or non-secure.
2046  *
2047  *          * SECURE : We use the IPSEC_IN to fanout to AH/ESP if the
2048  *            packet in error has AH/ESP headers to validate the AH/ESP
2049  *            headers. AH/ESP will verify whether there is a valid SA or
2050  *            not and send it back. We will fanout again if we have more
2051  *            data in the packet.
2052  *
2053  *            If the packet in error does not have AH/ESP, we handle it
2054  *            like any other case.
2055  *
2056  *          * NON_SECURE : If the packet in error has AH/ESP headers, we send it
2057  *            up to AH/ESP for validation. AH/ESP will verify whether there is a
2058  *            valid SA or not and send it back. We will fanout again if
2059  *            we have more data in the packet.
2060  *
2061  *            If the packet in error does not have AH/ESP, we handle it
2062  *            like any other case.
2063  *
2064  * The caller must have called icmp_inbound_verify_v4.
2065  */
2066 static void
2067 icmp_inbound_error_fanout_v4(mblk_t *mp, icmph_t *icmph, ip_recv_attr_t *ira)
2068 {
2069         uint16_t        *up;    /* Pointer to ports in ULP header */
2070         uint32_t        ports;  /* reversed ports for fanout */
2071         ipha_t          ripha;  /* With reversed addresses */
2072         ipha_t          *ipha;  /* Inner IP header */
2073         uint_t          hdr_length; /* Inner IP header length */
2074         tcpha_t         *tcpha;
2075         conn_t          *connp;
2076         ill_t           *ill = ira->ira_ill;
2077         ip_stack_t      *ipst = ill->ill_ipst;
2078         ipsec_stack_t   *ipss = ipst->ips_netstack->netstack_ipsec;
2079         ill_t           *rill = ira->ira_rill;
2080 
2081         /* Caller already pulled up everything. */
2082         ipha = (ipha_t *)&icmph[1];
2083         ASSERT((uchar_t *)&ipha[1] <= mp->b_wptr);
2084         ASSERT(mp->b_cont == NULL);
2085 
2086         hdr_length = IPH_HDR_LENGTH(ipha);
2087         ira->ira_protocol = ipha->ipha_protocol;
2088 
2089         /*
2090          * We need a separate IP header with the source and destination
2091          * addresses reversed to do fanout/classification because the ipha in
2092          * the ICMP error is in the form we sent it out.
2093          */
2094         ripha.ipha_src = ipha->ipha_dst;
2095         ripha.ipha_dst = ipha->ipha_src;
2096         ripha.ipha_protocol = ipha->ipha_protocol;
2097         ripha.ipha_version_and_hdr_length = ipha->ipha_version_and_hdr_length;
2098 
2099         ip2dbg(("icmp_inbound_error_v4: proto %d %x to %x: %d/%d\n",
2100             ripha.ipha_protocol, ntohl(ipha->ipha_src),
2101             ntohl(ipha->ipha_dst),
2102             icmph->icmph_type, icmph->icmph_code));
2103 
2104         switch (ipha->ipha_protocol) {
2105         case IPPROTO_UDP:
2106                 up = (uint16_t *)((uchar_t *)ipha + hdr_length);
2107 
2108                 /* Attempt to find a client stream based on port. */
2109                 ip2dbg(("icmp_inbound_error_v4: UDP ports %d to %d\n",
2110                     ntohs(up[0]), ntohs(up[1])));
2111 
2112                 /* Note that we send error to all matches. */
2113                 ira->ira_flags |= IRAF_ICMP_ERROR;
2114                 ip_fanout_udp_multi_v4(mp, &ripha, up[0], up[1], ira);
2115                 ira->ira_flags &= ~IRAF_ICMP_ERROR;
2116                 return;
2117 
2118         case IPPROTO_TCP:
2119                 /*
2120                  * Find a TCP client stream for this packet.
2121                  * Note that we do a reverse lookup since the header is
2122                  * in the form we sent it out.
2123                  */
2124                 tcpha = (tcpha_t *)((uchar_t *)ipha + hdr_length);
2125                 connp = ipcl_tcp_lookup_reversed_ipv4(ipha, tcpha, TCPS_LISTEN,
2126                     ipst);
2127                 if (connp == NULL)
2128                         goto discard_pkt;
2129 
2130                 if (CONN_INBOUND_POLICY_PRESENT(connp, ipss) ||
2131                     (ira->ira_flags & IRAF_IPSEC_SECURE)) {
2132                         mp = ipsec_check_inbound_policy(mp, connp,
2133                             ipha, NULL, ira);
2134                         if (mp == NULL) {
2135                                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
2136                                 /* Note that mp is NULL */
2137                                 ip_drop_input("ipIfStatsInDiscards", mp, ill);
2138                                 CONN_DEC_REF(connp);
2139                                 return;
2140                         }
2141                 }
2142 
2143                 ira->ira_flags |= IRAF_ICMP_ERROR;
2144                 ira->ira_ill = ira->ira_rill = NULL;
2145                 if (IPCL_IS_TCP(connp)) {
2146                         SQUEUE_ENTER_ONE(connp->conn_sqp, mp,
2147                             connp->conn_recvicmp, connp, ira, SQ_FILL,
2148                             SQTAG_TCP_INPUT_ICMP_ERR);
2149                 } else {
2150                         /* Not TCP; must be SOCK_RAW, IPPROTO_TCP */
2151                         (connp->conn_recv)(connp, mp, NULL, ira);
2152                         CONN_DEC_REF(connp);
2153                 }
2154                 ira->ira_ill = ill;
2155                 ira->ira_rill = rill;
2156                 ira->ira_flags &= ~IRAF_ICMP_ERROR;
2157                 return;
2158 
2159         case IPPROTO_SCTP:
2160                 up = (uint16_t *)((uchar_t *)ipha + hdr_length);
2161                 /* Find a SCTP client stream for this packet. */
2162                 ((uint16_t *)&ports)[0] = up[1];
2163                 ((uint16_t *)&ports)[1] = up[0];
2164 
2165                 ira->ira_flags |= IRAF_ICMP_ERROR;
2166                 ip_fanout_sctp(mp, &ripha, NULL, ports, ira);
2167                 ira->ira_flags &= ~IRAF_ICMP_ERROR;
2168                 return;
2169 
2170         case IPPROTO_ESP:
2171         case IPPROTO_AH:
2172                 if (!ipsec_loaded(ipss)) {
2173                         ip_proto_not_sup(mp, ira);
2174                         return;
2175                 }
2176 
2177                 if (ipha->ipha_protocol == IPPROTO_ESP)
2178                         mp = ipsecesp_icmp_error(mp, ira);
2179                 else
2180                         mp = ipsecah_icmp_error(mp, ira);
2181                 if (mp == NULL)
2182                         return;
2183 
2184                 /* Just in case ipsec didn't preserve the NULL b_cont */
2185                 if (mp->b_cont != NULL) {
2186                         if (!pullupmsg(mp, -1))
2187                                 goto discard_pkt;
2188                 }
2189 
2190                 /*
2191                  * Note that ira_pktlen and ira_ip_hdr_length are no longer
2192                  * correct, but we don't use them any more here.
2193                  *
2194                  * If succesful, the mp has been modified to not include
2195                  * the ESP/AH header so we can fanout to the ULP's icmp
2196                  * error handler.
2197                  */
2198                 if (mp->b_wptr - mp->b_rptr < IP_SIMPLE_HDR_LENGTH)
2199                         goto truncated;
2200 
2201                 /* Verify the modified message before any further processes. */
2202                 ipha = (ipha_t *)mp->b_rptr;
2203                 hdr_length = IPH_HDR_LENGTH(ipha);
2204                 icmph = (icmph_t *)&mp->b_rptr[hdr_length];
2205                 if (!icmp_inbound_verify_v4(mp, icmph, ira)) {
2206                         freemsg(mp);
2207                         return;
2208                 }
2209 
2210                 icmp_inbound_error_fanout_v4(mp, icmph, ira);
2211                 return;
2212 
2213         case IPPROTO_ENCAP: {
2214                 /* Look for self-encapsulated packets that caused an error */
2215                 ipha_t *in_ipha;
2216 
2217                 /*
2218                  * Caller has verified that length has to be
2219                  * at least the size of IP header.
2220                  */
2221                 ASSERT(hdr_length >= sizeof (ipha_t));
2222                 /*
2223                  * Check the sanity of the inner IP header like
2224                  * we did for the outer header.
2225                  */
2226                 in_ipha = (ipha_t *)((uchar_t *)ipha + hdr_length);
2227                 if ((IPH_HDR_VERSION(in_ipha) != IPV4_VERSION)) {
2228                         goto discard_pkt;
2229                 }
2230                 if (IPH_HDR_LENGTH(in_ipha) < sizeof (ipha_t)) {
2231                         goto discard_pkt;
2232                 }
2233                 /* Check for Self-encapsulated tunnels */
2234                 if (in_ipha->ipha_src == ipha->ipha_src &&
2235                     in_ipha->ipha_dst == ipha->ipha_dst) {
2236 
2237                         mp = icmp_inbound_self_encap_error_v4(mp, ipha,
2238                             in_ipha);
2239                         if (mp == NULL)
2240                                 goto discard_pkt;
2241 
2242                         /*
2243                          * Just in case self_encap didn't preserve the NULL
2244                          * b_cont
2245                          */
2246                         if (mp->b_cont != NULL) {
2247                                 if (!pullupmsg(mp, -1))
2248                                         goto discard_pkt;
2249                         }
2250                         /*
2251                          * Note that ira_pktlen and ira_ip_hdr_length are no
2252                          * longer correct, but we don't use them any more here.
2253                          */
2254                         if (mp->b_wptr - mp->b_rptr < IP_SIMPLE_HDR_LENGTH)
2255                                 goto truncated;
2256 
2257                         /*
2258                          * Verify the modified message before any further
2259                          * processes.
2260                          */
2261                         ipha = (ipha_t *)mp->b_rptr;
2262                         hdr_length = IPH_HDR_LENGTH(ipha);
2263                         icmph = (icmph_t *)&mp->b_rptr[hdr_length];
2264                         if (!icmp_inbound_verify_v4(mp, icmph, ira)) {
2265                                 freemsg(mp);
2266                                 return;
2267                         }
2268 
2269                         /*
2270                          * The packet in error is self-encapsualted.
2271                          * And we are finding it further encapsulated
2272                          * which we could not have possibly generated.
2273                          */
2274                         if (ipha->ipha_protocol == IPPROTO_ENCAP) {
2275                                 goto discard_pkt;
2276                         }
2277                         icmp_inbound_error_fanout_v4(mp, icmph, ira);
2278                         return;
2279                 }
2280                 /* No self-encapsulated */
2281         }
2282         /* FALLTHROUGH */
2283         case IPPROTO_IPV6:
2284                 if ((connp = ipcl_iptun_classify_v4(&ripha.ipha_src,
2285                     &ripha.ipha_dst, ipst)) != NULL) {
2286                         ira->ira_flags |= IRAF_ICMP_ERROR;
2287                         connp->conn_recvicmp(connp, mp, NULL, ira);
2288                         CONN_DEC_REF(connp);
2289                         ira->ira_flags &= ~IRAF_ICMP_ERROR;
2290                         return;
2291                 }
2292                 /*
2293                  * No IP tunnel is interested, fallthrough and see
2294                  * if a raw socket will want it.
2295                  */
2296                 /* FALLTHROUGH */
2297         default:
2298                 ira->ira_flags |= IRAF_ICMP_ERROR;
2299                 ip_fanout_proto_v4(mp, &ripha, ira);
2300                 ira->ira_flags &= ~IRAF_ICMP_ERROR;
2301                 return;
2302         }
2303         /* NOTREACHED */
2304 discard_pkt:
2305         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
2306         ip1dbg(("icmp_inbound_error_fanout_v4: drop pkt\n"));
2307         ip_drop_input("ipIfStatsInDiscards", mp, ill);
2308         freemsg(mp);
2309         return;
2310 
2311 truncated:
2312         /* We pulled up everthing already. Must be truncated */
2313         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInTruncatedPkts);
2314         ip_drop_input("ipIfStatsInTruncatedPkts", mp, ill);
2315         freemsg(mp);
2316 }
2317 
2318 /*
2319  * Common IP options parser.
2320  *
2321  * Setup routine: fill in *optp with options-parsing state, then
2322  * tail-call ipoptp_next to return the first option.
2323  */
2324 uint8_t
2325 ipoptp_first(ipoptp_t *optp, ipha_t *ipha)
2326 {
2327         uint32_t totallen; /* total length of all options */
2328 
2329         totallen = ipha->ipha_version_and_hdr_length -
2330             (uint8_t)((IP_VERSION << 4) + IP_SIMPLE_HDR_LENGTH_IN_WORDS);
2331         totallen <<= 2;
2332         optp->ipoptp_next = (uint8_t *)(&ipha[1]);
2333         optp->ipoptp_end = optp->ipoptp_next + totallen;
2334         optp->ipoptp_flags = 0;
2335         return (ipoptp_next(optp));
2336 }
2337 
2338 /* Like above but without an ipha_t */
2339 uint8_t
2340 ipoptp_first2(ipoptp_t *optp, uint32_t totallen, uint8_t *opt)
2341 {
2342         optp->ipoptp_next = opt;
2343         optp->ipoptp_end = optp->ipoptp_next + totallen;
2344         optp->ipoptp_flags = 0;
2345         return (ipoptp_next(optp));
2346 }
2347 
2348 /*
2349  * Common IP options parser: extract next option.
2350  */
2351 uint8_t
2352 ipoptp_next(ipoptp_t *optp)
2353 {
2354         uint8_t *end = optp->ipoptp_end;
2355         uint8_t *cur = optp->ipoptp_next;
2356         uint8_t opt, len, pointer;
2357 
2358         /*
2359          * If cur > end already, then the ipoptp_end or ipoptp_next pointer
2360          * has been corrupted.
2361          */
2362         ASSERT(cur <= end);
2363 
2364         if (cur == end)
2365                 return (IPOPT_EOL);
2366 
2367         opt = cur[IPOPT_OPTVAL];
2368 
2369         /*
2370          * Skip any NOP options.
2371          */
2372         while (opt == IPOPT_NOP) {
2373                 cur++;
2374                 if (cur == end)
2375                         return (IPOPT_EOL);
2376                 opt = cur[IPOPT_OPTVAL];
2377         }
2378 
2379         if (opt == IPOPT_EOL)
2380                 return (IPOPT_EOL);
2381 
2382         /*
2383          * Option requiring a length.
2384          */
2385         if ((cur + 1) >= end) {
2386                 optp->ipoptp_flags |= IPOPTP_ERROR;
2387                 return (IPOPT_EOL);
2388         }
2389         len = cur[IPOPT_OLEN];
2390         if (len < 2) {
2391                 optp->ipoptp_flags |= IPOPTP_ERROR;
2392                 return (IPOPT_EOL);
2393         }
2394         optp->ipoptp_cur = cur;
2395         optp->ipoptp_len = len;
2396         optp->ipoptp_next = cur + len;
2397         if (cur + len > end) {
2398                 optp->ipoptp_flags |= IPOPTP_ERROR;
2399                 return (IPOPT_EOL);
2400         }
2401 
2402         /*
2403          * For the options which require a pointer field, make sure
2404          * its there, and make sure it points to either something
2405          * inside this option, or the end of the option.
2406          */
2407         switch (opt) {
2408         case IPOPT_RR:
2409         case IPOPT_TS:
2410         case IPOPT_LSRR:
2411         case IPOPT_SSRR:
2412                 if (len <= IPOPT_OFFSET) {
2413                         optp->ipoptp_flags |= IPOPTP_ERROR;
2414                         return (opt);
2415                 }
2416                 pointer = cur[IPOPT_OFFSET];
2417                 if (pointer - 1 > len) {
2418                         optp->ipoptp_flags |= IPOPTP_ERROR;
2419                         return (opt);
2420                 }
2421                 break;
2422         }
2423 
2424         /*
2425          * Sanity check the pointer field based on the type of the
2426          * option.
2427          */
2428         switch (opt) {
2429         case IPOPT_RR:
2430         case IPOPT_SSRR:
2431         case IPOPT_LSRR:
2432                 if (pointer < IPOPT_MINOFF_SR)
2433                         optp->ipoptp_flags |= IPOPTP_ERROR;
2434                 break;
2435         case IPOPT_TS:
2436                 if (pointer < IPOPT_MINOFF_IT)
2437                         optp->ipoptp_flags |= IPOPTP_ERROR;
2438                 /*
2439                  * Note that the Internet Timestamp option also
2440                  * contains two four bit fields (the Overflow field,
2441                  * and the Flag field), which follow the pointer
2442                  * field.  We don't need to check that these fields
2443                  * fall within the length of the option because this
2444                  * was implicitely done above.  We've checked that the
2445                  * pointer value is at least IPOPT_MINOFF_IT, and that
2446                  * it falls within the option.  Since IPOPT_MINOFF_IT >
2447                  * IPOPT_POS_OV_FLG, we don't need the explicit check.
2448                  */
2449                 ASSERT(len > IPOPT_POS_OV_FLG);
2450                 break;
2451         }
2452 
2453         return (opt);
2454 }
2455 
2456 /*
2457  * Use the outgoing IP header to create an IP_OPTIONS option the way
2458  * it was passed down from the application.
2459  *
2460  * This is compatible with BSD in that it returns
2461  * the reverse source route with the final destination
2462  * as the last entry. The first 4 bytes of the option
2463  * will contain the final destination.
2464  */
2465 int
2466 ip_opt_get_user(conn_t *connp, uchar_t *buf)
2467 {
2468         ipoptp_t        opts;
2469         uchar_t         *opt;
2470         uint8_t         optval;
2471         uint8_t         optlen;
2472         uint32_t        len = 0;
2473         uchar_t         *buf1 = buf;
2474         uint32_t        totallen;
2475         ipaddr_t        dst;
2476         ip_pkt_t        *ipp = &connp->conn_xmit_ipp;
2477 
2478         if (!(ipp->ipp_fields & IPPF_IPV4_OPTIONS))
2479                 return (0);
2480 
2481         totallen = ipp->ipp_ipv4_options_len;
2482         if (totallen & 0x3)
2483                 return (0);
2484 
2485         buf += IP_ADDR_LEN;     /* Leave room for final destination */
2486         len += IP_ADDR_LEN;
2487         bzero(buf1, IP_ADDR_LEN);
2488 
2489         dst = connp->conn_faddr_v4;
2490 
2491         for (optval = ipoptp_first2(&opts, totallen, ipp->ipp_ipv4_options);
2492             optval != IPOPT_EOL;
2493             optval = ipoptp_next(&opts)) {
2494                 int     off;
2495 
2496                 opt = opts.ipoptp_cur;
2497                 if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
2498                         break;
2499                 }
2500                 optlen = opts.ipoptp_len;
2501 
2502                 switch (optval) {
2503                 case IPOPT_SSRR:
2504                 case IPOPT_LSRR:
2505 
2506                         /*
2507                          * Insert destination as the first entry in the source
2508                          * route and move down the entries on step.
2509                          * The last entry gets placed at buf1.
2510                          */
2511                         buf[IPOPT_OPTVAL] = optval;
2512                         buf[IPOPT_OLEN] = optlen;
2513                         buf[IPOPT_OFFSET] = optlen;
2514 
2515                         off = optlen - IP_ADDR_LEN;
2516                         if (off < 0) {
2517                                 /* No entries in source route */
2518                                 break;
2519                         }
2520                         /* Last entry in source route if not already set */
2521                         if (dst == INADDR_ANY)
2522                                 bcopy(opt + off, buf1, IP_ADDR_LEN);
2523                         off -= IP_ADDR_LEN;
2524 
2525                         while (off > 0) {
2526                                 bcopy(opt + off,
2527                                     buf + off + IP_ADDR_LEN,
2528                                     IP_ADDR_LEN);
2529                                 off -= IP_ADDR_LEN;
2530                         }
2531                         /* ipha_dst into first slot */
2532                         bcopy(&dst, buf + off + IP_ADDR_LEN,
2533                             IP_ADDR_LEN);
2534                         buf += optlen;
2535                         len += optlen;
2536                         break;
2537 
2538                 default:
2539                         bcopy(opt, buf, optlen);
2540                         buf += optlen;
2541                         len += optlen;
2542                         break;
2543                 }
2544         }
2545 done:
2546         /* Pad the resulting options */
2547         while (len & 0x3) {
2548                 *buf++ = IPOPT_EOL;
2549                 len++;
2550         }
2551         return (len);
2552 }
2553 
2554 /*
2555  * Update any record route or timestamp options to include this host.
2556  * Reverse any source route option.
2557  * This routine assumes that the options are well formed i.e. that they
2558  * have already been checked.
2559  */
2560 static void
2561 icmp_options_update(ipha_t *ipha)
2562 {
2563         ipoptp_t        opts;
2564         uchar_t         *opt;
2565         uint8_t         optval;
2566         ipaddr_t        src;            /* Our local address */
2567         ipaddr_t        dst;
2568 
2569         ip2dbg(("icmp_options_update\n"));
2570         src = ipha->ipha_src;
2571         dst = ipha->ipha_dst;
2572 
2573         for (optval = ipoptp_first(&opts, ipha);
2574             optval != IPOPT_EOL;
2575             optval = ipoptp_next(&opts)) {
2576                 ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
2577                 opt = opts.ipoptp_cur;
2578                 ip2dbg(("icmp_options_update: opt %d, len %d\n",
2579                     optval, opts.ipoptp_len));
2580                 switch (optval) {
2581                         int off1, off2;
2582                 case IPOPT_SSRR:
2583                 case IPOPT_LSRR:
2584                         /*
2585                          * Reverse the source route.  The first entry
2586                          * should be the next to last one in the current
2587                          * source route (the last entry is our address).
2588                          * The last entry should be the final destination.
2589                          */
2590                         off1 = IPOPT_MINOFF_SR - 1;
2591                         off2 = opt[IPOPT_OFFSET] - IP_ADDR_LEN - 1;
2592                         if (off2 < 0) {
2593                                 /* No entries in source route */
2594                                 ip1dbg((
2595                                     "icmp_options_update: bad src route\n"));
2596                                 break;
2597                         }
2598                         bcopy((char *)opt + off2, &dst, IP_ADDR_LEN);
2599                         bcopy(&ipha->ipha_dst, (char *)opt + off2, IP_ADDR_LEN);
2600                         bcopy(&dst, &ipha->ipha_dst, IP_ADDR_LEN);
2601                         off2 -= IP_ADDR_LEN;
2602 
2603                         while (off1 < off2) {
2604                                 bcopy((char *)opt + off1, &src, IP_ADDR_LEN);
2605                                 bcopy((char *)opt + off2, (char *)opt + off1,
2606                                     IP_ADDR_LEN);
2607                                 bcopy(&src, (char *)opt + off2, IP_ADDR_LEN);
2608                                 off1 += IP_ADDR_LEN;
2609                                 off2 -= IP_ADDR_LEN;
2610                         }
2611                         opt[IPOPT_OFFSET] = IPOPT_MINOFF_SR;
2612                         break;
2613                 }
2614         }
2615 }
2616 
2617 /*
2618  * Process received ICMP Redirect messages.
2619  * Assumes the caller has verified that the headers are in the pulled up mblk.
2620  * Consumes mp.
2621  */
2622 static void
2623 icmp_redirect_v4(mblk_t *mp, ipha_t *ipha, icmph_t *icmph, ip_recv_attr_t *ira)
2624 {
2625         ire_t           *ire, *nire;
2626         ire_t           *prev_ire;
2627         ipaddr_t        src, dst, gateway;
2628         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
2629         ipha_t          *inner_ipha;    /* Inner IP header */
2630 
2631         /* Caller already pulled up everything. */
2632         inner_ipha = (ipha_t *)&icmph[1];
2633         src = ipha->ipha_src;
2634         dst = inner_ipha->ipha_dst;
2635         gateway = icmph->icmph_rd_gateway;
2636         /* Make sure the new gateway is reachable somehow. */
2637         ire = ire_ftable_lookup_v4(gateway, 0, 0, IRE_ONLINK, NULL,
2638             ALL_ZONES, NULL, MATCH_IRE_TYPE, 0, ipst, NULL);
2639         /*
2640          * Make sure we had a route for the dest in question and that
2641          * that route was pointing to the old gateway (the source of the
2642          * redirect packet.)
2643          * We do longest match and then compare ire_gateway_addr below.
2644          */
2645         prev_ire = ire_ftable_lookup_v4(dst, 0, 0, 0, NULL, ALL_ZONES,
2646             NULL, MATCH_IRE_DSTONLY, 0, ipst, NULL);
2647         /*
2648          * Check that
2649          *      the redirect was not from ourselves
2650          *      the new gateway and the old gateway are directly reachable
2651          */
2652         if (prev_ire == NULL || ire == NULL ||
2653             (prev_ire->ire_type & (IRE_LOCAL|IRE_LOOPBACK)) ||
2654             (prev_ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE)) ||
2655             !(ire->ire_type & IRE_IF_ALL) ||
2656             prev_ire->ire_gateway_addr != src) {
2657                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInBadRedirects);
2658                 ip_drop_input("icmpInBadRedirects - ire", mp, ira->ira_ill);
2659                 freemsg(mp);
2660                 if (ire != NULL)
2661                         ire_refrele(ire);
2662                 if (prev_ire != NULL)
2663                         ire_refrele(prev_ire);
2664                 return;
2665         }
2666 
2667         ire_refrele(prev_ire);
2668         ire_refrele(ire);
2669 
2670         /*
2671          * TODO: more precise handling for cases 0, 2, 3, the latter two
2672          * require TOS routing
2673          */
2674         switch (icmph->icmph_code) {
2675         case 0:
2676         case 1:
2677                 /* TODO: TOS specificity for cases 2 and 3 */
2678         case 2:
2679         case 3:
2680                 break;
2681         default:
2682                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInBadRedirects);
2683                 ip_drop_input("icmpInBadRedirects - code", mp, ira->ira_ill);
2684                 freemsg(mp);
2685                 return;
2686         }
2687         /*
2688          * Create a Route Association.  This will allow us to remember that
2689          * someone we believe told us to use the particular gateway.
2690          */
2691         ire = ire_create(
2692             (uchar_t *)&dst,                        /* dest addr */
2693             (uchar_t *)&ip_g_all_ones,              /* mask */
2694             (uchar_t *)&gateway,            /* gateway addr */
2695             IRE_HOST,
2696             NULL,                               /* ill */
2697             ALL_ZONES,
2698             (RTF_DYNAMIC | RTF_GATEWAY | RTF_HOST),
2699             NULL,                               /* tsol_gc_t */
2700             ipst);
2701 
2702         if (ire == NULL) {
2703                 freemsg(mp);
2704                 return;
2705         }
2706         nire = ire_add(ire);
2707         /* Check if it was a duplicate entry */
2708         if (nire != NULL && nire != ire) {
2709                 ASSERT(nire->ire_identical_ref > 1);
2710                 ire_delete(nire);
2711                 ire_refrele(nire);
2712                 nire = NULL;
2713         }
2714         ire = nire;
2715         if (ire != NULL) {
2716                 ire_refrele(ire);               /* Held in ire_add */
2717 
2718                 /* tell routing sockets that we received a redirect */
2719                 ip_rts_change(RTM_REDIRECT, dst, gateway, IP_HOST_MASK, 0, src,
2720                     (RTF_DYNAMIC | RTF_GATEWAY | RTF_HOST), 0,
2721                     (RTA_DST | RTA_GATEWAY | RTA_NETMASK | RTA_AUTHOR), ipst);
2722         }
2723 
2724         /*
2725          * Delete any existing IRE_HOST type redirect ires for this destination.
2726          * This together with the added IRE has the effect of
2727          * modifying an existing redirect.
2728          */
2729         prev_ire = ire_ftable_lookup_v4(dst, 0, src, IRE_HOST, NULL,
2730             ALL_ZONES, NULL, (MATCH_IRE_GW | MATCH_IRE_TYPE), 0, ipst, NULL);
2731         if (prev_ire != NULL) {
2732                 if (prev_ire ->ire_flags & RTF_DYNAMIC)
2733                         ire_delete(prev_ire);
2734                 ire_refrele(prev_ire);
2735         }
2736 
2737         freemsg(mp);
2738 }
2739 
2740 /*
2741  * Generate an ICMP parameter problem message.
2742  * When called from ip_output side a minimal ip_recv_attr_t needs to be
2743  * constructed by the caller.
2744  */
2745 static void
2746 icmp_param_problem(mblk_t *mp, uint8_t ptr, ip_recv_attr_t *ira)
2747 {
2748         icmph_t icmph;
2749         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
2750 
2751         mp = icmp_pkt_err_ok(mp, ira);
2752         if (mp == NULL)
2753                 return;
2754 
2755         bzero(&icmph, sizeof (icmph_t));
2756         icmph.icmph_type = ICMP_PARAM_PROBLEM;
2757         icmph.icmph_pp_ptr = ptr;
2758         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutParmProbs);
2759         icmp_pkt(mp, &icmph, sizeof (icmph_t), ira);
2760 }
2761 
2762 /*
2763  * Build and ship an IPv4 ICMP message using the packet data in mp, and
2764  * the ICMP header pointed to by "stuff".  (May be called as writer.)
2765  * Note: assumes that icmp_pkt_err_ok has been called to verify that
2766  * an icmp error packet can be sent.
2767  * Assigns an appropriate source address to the packet. If ipha_dst is
2768  * one of our addresses use it for source. Otherwise let ip_output_simple
2769  * pick the source address.
2770  */
2771 static void
2772 icmp_pkt(mblk_t *mp, void *stuff, size_t len, ip_recv_attr_t *ira)
2773 {
2774         ipaddr_t dst;
2775         icmph_t *icmph;
2776         ipha_t  *ipha;
2777         uint_t  len_needed;
2778         size_t  msg_len;
2779         mblk_t  *mp1;
2780         ipaddr_t src;
2781         ire_t   *ire;
2782         ip_xmit_attr_t ixas;
2783         ip_stack_t *ipst = ira->ira_ill->ill_ipst;
2784 
2785         ipha = (ipha_t *)mp->b_rptr;
2786 
2787         bzero(&ixas, sizeof (ixas));
2788         ixas.ixa_flags = IXAF_BASIC_SIMPLE_V4;
2789         ixas.ixa_zoneid = ira->ira_zoneid;
2790         ixas.ixa_ifindex = 0;
2791         ixas.ixa_ipst = ipst;
2792         ixas.ixa_cred = kcred;
2793         ixas.ixa_cpid = NOPID;
2794         ixas.ixa_tsl = ira->ira_tsl; /* Behave as a multi-level responder */
2795         ixas.ixa_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
2796 
2797         if (ira->ira_flags & IRAF_IPSEC_SECURE) {
2798                 /*
2799                  * Apply IPsec based on how IPsec was applied to
2800                  * the packet that had the error.
2801                  *
2802                  * If it was an outbound packet that caused the ICMP
2803                  * error, then the caller will have setup the IRA
2804                  * appropriately.
2805                  */
2806                 if (!ipsec_in_to_out(ira, &ixas, mp, ipha, NULL)) {
2807                         BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsOutDiscards);
2808                         /* Note: mp already consumed and ip_drop_packet done */
2809                         return;
2810                 }
2811         } else {
2812                 /*
2813                  * This is in clear. The icmp message we are building
2814                  * here should go out in clear, independent of our policy.
2815                  */
2816                 ixas.ixa_flags |= IXAF_NO_IPSEC;
2817         }
2818 
2819         /* Remember our eventual destination */
2820         dst = ipha->ipha_src;
2821 
2822         /*
2823          * If the packet was for one of our unicast addresses, make
2824          * sure we respond with that as the source. Otherwise
2825          * have ip_output_simple pick the source address.
2826          */
2827         ire = ire_ftable_lookup_v4(ipha->ipha_dst, 0, 0,
2828             (IRE_LOCAL|IRE_LOOPBACK), NULL, ira->ira_zoneid, NULL,
2829             MATCH_IRE_TYPE|MATCH_IRE_ZONEONLY, 0, ipst, NULL);
2830         if (ire != NULL) {
2831                 ire_refrele(ire);
2832                 src = ipha->ipha_dst;
2833         } else {
2834                 src = INADDR_ANY;
2835                 ixas.ixa_flags |= IXAF_SET_SOURCE;
2836         }
2837 
2838         /*
2839          * Check if we can send back more then 8 bytes in addition to
2840          * the IP header.  We try to send 64 bytes of data and the internal
2841          * header in the special cases of ipv4 encapsulated ipv4 or ipv6.
2842          */
2843         len_needed = IPH_HDR_LENGTH(ipha);
2844         if (ipha->ipha_protocol == IPPROTO_ENCAP ||
2845             ipha->ipha_protocol == IPPROTO_IPV6) {
2846                 if (!pullupmsg(mp, -1)) {
2847                         BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsOutDiscards);
2848                         ip_drop_output("ipIfStatsOutDiscards", mp, NULL);
2849                         freemsg(mp);
2850                         return;
2851                 }
2852                 ipha = (ipha_t *)mp->b_rptr;
2853 
2854                 if (ipha->ipha_protocol == IPPROTO_ENCAP) {
2855                         len_needed += IPH_HDR_LENGTH(((uchar_t *)ipha +
2856                             len_needed));
2857                 } else {
2858                         ip6_t *ip6h = (ip6_t *)((uchar_t *)ipha + len_needed);
2859 
2860                         ASSERT(ipha->ipha_protocol == IPPROTO_IPV6);
2861                         len_needed += ip_hdr_length_v6(mp, ip6h);
2862                 }
2863         }
2864         len_needed += ipst->ips_ip_icmp_return;
2865         msg_len = msgdsize(mp);
2866         if (msg_len > len_needed) {
2867                 (void) adjmsg(mp, len_needed - msg_len);
2868                 msg_len = len_needed;
2869         }
2870         mp1 = allocb(sizeof (icmp_ipha) + len, BPRI_MED);
2871         if (mp1 == NULL) {
2872                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutErrors);
2873                 freemsg(mp);
2874                 return;
2875         }
2876         mp1->b_cont = mp;
2877         mp = mp1;
2878 
2879         /*
2880          * Set IXAF_TRUSTED_ICMP so we can let the ICMP messages this
2881          * node generates be accepted in peace by all on-host destinations.
2882          * If we do NOT assume that all on-host destinations trust
2883          * self-generated ICMP messages, then rework here, ip6.c, and spd.c.
2884          * (Look for IXAF_TRUSTED_ICMP).
2885          */
2886         ixas.ixa_flags |= IXAF_TRUSTED_ICMP;
2887 
2888         ipha = (ipha_t *)mp->b_rptr;
2889         mp1->b_wptr = (uchar_t *)ipha + (sizeof (icmp_ipha) + len);
2890         *ipha = icmp_ipha;
2891         ipha->ipha_src = src;
2892         ipha->ipha_dst = dst;
2893         ipha->ipha_ttl = ipst->ips_ip_def_ttl;
2894         msg_len += sizeof (icmp_ipha) + len;
2895         if (msg_len > IP_MAXPACKET) {
2896                 (void) adjmsg(mp, IP_MAXPACKET - msg_len);
2897                 msg_len = IP_MAXPACKET;
2898         }
2899         ipha->ipha_length = htons((uint16_t)msg_len);
2900         icmph = (icmph_t *)&ipha[1];
2901         bcopy(stuff, icmph, len);
2902         icmph->icmph_checksum = 0;
2903         icmph->icmph_checksum = IP_CSUM(mp, (int32_t)sizeof (ipha_t), 0);
2904         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutMsgs);
2905 
2906         (void) ip_output_simple(mp, &ixas);
2907         ixa_cleanup(&ixas);
2908 }
2909 
2910 /*
2911  * Determine if an ICMP error packet can be sent given the rate limit.
2912  * The limit consists of an average frequency (icmp_pkt_err_interval measured
2913  * in milliseconds) and a burst size. Burst size number of packets can
2914  * be sent arbitrarely closely spaced.
2915  * The state is tracked using two variables to implement an approximate
2916  * token bucket filter:
2917  *      icmp_pkt_err_last - lbolt value when the last burst started
2918  *      icmp_pkt_err_sent - number of packets sent in current burst
2919  */
2920 boolean_t
2921 icmp_err_rate_limit(ip_stack_t *ipst)
2922 {
2923         clock_t now = TICK_TO_MSEC(ddi_get_lbolt());
2924         uint_t refilled; /* Number of packets refilled in tbf since last */
2925         /* Guard against changes by loading into local variable */
2926         uint_t err_interval = ipst->ips_ip_icmp_err_interval;
2927 
2928         if (err_interval == 0)
2929                 return (B_FALSE);
2930 
2931         if (ipst->ips_icmp_pkt_err_last > now) {
2932                 /* 100HZ lbolt in ms for 32bit arch wraps every 49.7 days */
2933                 ipst->ips_icmp_pkt_err_last = 0;
2934                 ipst->ips_icmp_pkt_err_sent = 0;
2935         }
2936         /*
2937          * If we are in a burst update the token bucket filter.
2938          * Update the "last" time to be close to "now" but make sure
2939          * we don't loose precision.
2940          */
2941         if (ipst->ips_icmp_pkt_err_sent != 0) {
2942                 refilled = (now - ipst->ips_icmp_pkt_err_last)/err_interval;
2943                 if (refilled > ipst->ips_icmp_pkt_err_sent) {
2944                         ipst->ips_icmp_pkt_err_sent = 0;
2945                 } else {
2946                         ipst->ips_icmp_pkt_err_sent -= refilled;
2947                         ipst->ips_icmp_pkt_err_last += refilled * err_interval;
2948                 }
2949         }
2950         if (ipst->ips_icmp_pkt_err_sent == 0) {
2951                 /* Start of new burst */
2952                 ipst->ips_icmp_pkt_err_last = now;
2953         }
2954         if (ipst->ips_icmp_pkt_err_sent < ipst->ips_ip_icmp_err_burst) {
2955                 ipst->ips_icmp_pkt_err_sent++;
2956                 ip1dbg(("icmp_err_rate_limit: %d sent in burst\n",
2957                     ipst->ips_icmp_pkt_err_sent));
2958                 return (B_FALSE);
2959         }
2960         ip1dbg(("icmp_err_rate_limit: dropped\n"));
2961         return (B_TRUE);
2962 }
2963 
2964 /*
2965  * Check if it is ok to send an IPv4 ICMP error packet in
2966  * response to the IPv4 packet in mp.
2967  * Free the message and return null if no
2968  * ICMP error packet should be sent.
2969  */
2970 static mblk_t *
2971 icmp_pkt_err_ok(mblk_t *mp, ip_recv_attr_t *ira)
2972 {
2973         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
2974         icmph_t *icmph;
2975         ipha_t  *ipha;
2976         uint_t  len_needed;
2977 
2978         if (!mp)
2979                 return (NULL);
2980         ipha = (ipha_t *)mp->b_rptr;
2981         if (ip_csum_hdr(ipha)) {
2982                 BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsInCksumErrs);
2983                 ip_drop_input("ipIfStatsInCksumErrs", mp, NULL);
2984                 freemsg(mp);
2985                 return (NULL);
2986         }
2987         if (ip_type_v4(ipha->ipha_dst, ipst) == IRE_BROADCAST ||
2988             ip_type_v4(ipha->ipha_src, ipst) == IRE_BROADCAST ||
2989             CLASSD(ipha->ipha_dst) ||
2990             CLASSD(ipha->ipha_src) ||
2991             (ntohs(ipha->ipha_fragment_offset_and_flags) & IPH_OFFSET)) {
2992                 /* Note: only errors to the fragment with offset 0 */
2993                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
2994                 freemsg(mp);
2995                 return (NULL);
2996         }
2997         if (ipha->ipha_protocol == IPPROTO_ICMP) {
2998                 /*
2999                  * Check the ICMP type.  RFC 1122 sez:  don't send ICMP
3000                  * errors in response to any ICMP errors.
3001                  */
3002                 len_needed = IPH_HDR_LENGTH(ipha) + ICMPH_SIZE;
3003                 if (mp->b_wptr - mp->b_rptr < len_needed) {
3004                         if (!pullupmsg(mp, len_needed)) {
3005                                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInErrors);
3006                                 freemsg(mp);
3007                                 return (NULL);
3008                         }
3009                         ipha = (ipha_t *)mp->b_rptr;
3010                 }
3011                 icmph = (icmph_t *)
3012                     (&((char *)ipha)[IPH_HDR_LENGTH(ipha)]);
3013                 switch (icmph->icmph_type) {
3014                 case ICMP_DEST_UNREACHABLE:
3015                 case ICMP_SOURCE_QUENCH:
3016                 case ICMP_TIME_EXCEEDED:
3017                 case ICMP_PARAM_PROBLEM:
3018                 case ICMP_REDIRECT:
3019                         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
3020                         freemsg(mp);
3021                         return (NULL);
3022                 default:
3023                         break;
3024                 }
3025         }
3026         /*
3027          * If this is a labeled system, then check to see if we're allowed to
3028          * send a response to this particular sender.  If not, then just drop.
3029          */
3030         if (is_system_labeled() && !tsol_can_reply_error(mp, ira)) {
3031                 ip2dbg(("icmp_pkt_err_ok: can't respond to packet\n"));
3032                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
3033                 freemsg(mp);
3034                 return (NULL);
3035         }
3036         if (icmp_err_rate_limit(ipst)) {
3037                 /*
3038                  * Only send ICMP error packets every so often.
3039                  * This should be done on a per port/source basis,
3040                  * but for now this will suffice.
3041                  */
3042                 freemsg(mp);
3043                 return (NULL);
3044         }
3045         return (mp);
3046 }
3047 
3048 /*
3049  * Called when a packet was sent out the same link that it arrived on.
3050  * Check if it is ok to send a redirect and then send it.
3051  */
3052 void
3053 ip_send_potential_redirect_v4(mblk_t *mp, ipha_t *ipha, ire_t *ire,
3054     ip_recv_attr_t *ira)
3055 {
3056         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
3057         ipaddr_t        src, nhop;
3058         mblk_t          *mp1;
3059         ire_t           *nhop_ire;
3060 
3061         /*
3062          * Check the source address to see if it originated
3063          * on the same logical subnet it is going back out on.
3064          * If so, we should be able to send it a redirect.
3065          * Avoid sending a redirect if the destination
3066          * is directly connected (i.e., we matched an IRE_ONLINK),
3067          * or if the packet was source routed out this interface.
3068          *
3069          * We avoid sending a redirect if the
3070          * destination is directly connected
3071          * because it is possible that multiple
3072          * IP subnets may have been configured on
3073          * the link, and the source may not
3074          * be on the same subnet as ip destination,
3075          * even though they are on the same
3076          * physical link.
3077          */
3078         if ((ire->ire_type & IRE_ONLINK) ||
3079             ip_source_routed(ipha, ipst))
3080                 return;
3081 
3082         nhop_ire = ire_nexthop(ire);
3083         if (nhop_ire == NULL)
3084                 return;
3085 
3086         nhop = nhop_ire->ire_addr;
3087 
3088         if (nhop_ire->ire_type & IRE_IF_CLONE) {
3089                 ire_t   *ire2;
3090 
3091                 /* Follow ire_dep_parent to find non-clone IRE_INTERFACE */
3092                 mutex_enter(&nhop_ire->ire_lock);
3093                 ire2 = nhop_ire->ire_dep_parent;
3094                 if (ire2 != NULL)
3095                         ire_refhold(ire2);
3096                 mutex_exit(&nhop_ire->ire_lock);
3097                 ire_refrele(nhop_ire);
3098                 nhop_ire = ire2;
3099         }
3100         if (nhop_ire == NULL)
3101                 return;
3102 
3103         ASSERT(!(nhop_ire->ire_type & IRE_IF_CLONE));
3104 
3105         src = ipha->ipha_src;
3106 
3107         /*
3108          * We look at the interface ire for the nexthop,
3109          * to see if ipha_src is in the same subnet
3110          * as the nexthop.
3111          */
3112         if ((src & nhop_ire->ire_mask) == (nhop & nhop_ire->ire_mask)) {
3113                 /*
3114                  * The source is directly connected.
3115                  */
3116                 mp1 = copymsg(mp);
3117                 if (mp1 != NULL) {
3118                         icmp_send_redirect(mp1, nhop, ira);
3119                 }
3120         }
3121         ire_refrele(nhop_ire);
3122 }
3123 
3124 /*
3125  * Generate an ICMP redirect message.
3126  */
3127 static void
3128 icmp_send_redirect(mblk_t *mp, ipaddr_t gateway, ip_recv_attr_t *ira)
3129 {
3130         icmph_t icmph;
3131         ip_stack_t *ipst = ira->ira_ill->ill_ipst;
3132 
3133         mp = icmp_pkt_err_ok(mp, ira);
3134         if (mp == NULL)
3135                 return;
3136 
3137         bzero(&icmph, sizeof (icmph_t));
3138         icmph.icmph_type = ICMP_REDIRECT;
3139         icmph.icmph_code = 1;
3140         icmph.icmph_rd_gateway = gateway;
3141         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutRedirects);
3142         icmp_pkt(mp, &icmph, sizeof (icmph_t), ira);
3143 }
3144 
3145 /*
3146  * Generate an ICMP time exceeded message.
3147  */
3148 void
3149 icmp_time_exceeded(mblk_t *mp, uint8_t code, ip_recv_attr_t *ira)
3150 {
3151         icmph_t icmph;
3152         ip_stack_t *ipst = ira->ira_ill->ill_ipst;
3153 
3154         mp = icmp_pkt_err_ok(mp, ira);
3155         if (mp == NULL)
3156                 return;
3157 
3158         bzero(&icmph, sizeof (icmph_t));
3159         icmph.icmph_type = ICMP_TIME_EXCEEDED;
3160         icmph.icmph_code = code;
3161         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutTimeExcds);
3162         icmp_pkt(mp, &icmph, sizeof (icmph_t), ira);
3163 }
3164 
3165 /*
3166  * Generate an ICMP unreachable message.
3167  * When called from ip_output side a minimal ip_recv_attr_t needs to be
3168  * constructed by the caller.
3169  */
3170 void
3171 icmp_unreachable(mblk_t *mp, uint8_t code, ip_recv_attr_t *ira)
3172 {
3173         icmph_t icmph;
3174         ip_stack_t *ipst = ira->ira_ill->ill_ipst;
3175 
3176         mp = icmp_pkt_err_ok(mp, ira);
3177         if (mp == NULL)
3178                 return;
3179 
3180         bzero(&icmph, sizeof (icmph_t));
3181         icmph.icmph_type = ICMP_DEST_UNREACHABLE;
3182         icmph.icmph_code = code;
3183         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDestUnreachs);
3184         icmp_pkt(mp, &icmph, sizeof (icmph_t), ira);
3185 }
3186 
3187 /*
3188  * Latch in the IPsec state for a stream based the policy in the listener
3189  * and the actions in the ip_recv_attr_t.
3190  * Called directly from TCP and SCTP.
3191  */
3192 boolean_t
3193 ip_ipsec_policy_inherit(conn_t *connp, conn_t *lconnp, ip_recv_attr_t *ira)
3194 {
3195         ASSERT(lconnp->conn_policy != NULL);
3196         ASSERT(connp->conn_policy == NULL);
3197 
3198         IPPH_REFHOLD(lconnp->conn_policy);
3199         connp->conn_policy = lconnp->conn_policy;
3200 
3201         if (ira->ira_ipsec_action != NULL) {
3202                 if (connp->conn_latch == NULL) {
3203                         connp->conn_latch = iplatch_create();
3204                         if (connp->conn_latch == NULL)
3205                                 return (B_FALSE);
3206                 }
3207                 ipsec_latch_inbound(connp, ira);
3208         }
3209         return (B_TRUE);
3210 }
3211 
3212 /*
3213  * Verify whether or not the IP address is a valid local address.
3214  * Could be a unicast, including one for a down interface.
3215  * If allow_mcbc then a multicast or broadcast address is also
3216  * acceptable.
3217  *
3218  * In the case of a broadcast/multicast address, however, the
3219  * upper protocol is expected to reset the src address
3220  * to zero when we return IPVL_MCAST/IPVL_BCAST so that
3221  * no packets are emitted with broadcast/multicast address as
3222  * source address (that violates hosts requirements RFC 1122)
3223  * The addresses valid for bind are:
3224  *      (1) - INADDR_ANY (0)
3225  *      (2) - IP address of an UP interface
3226  *      (3) - IP address of a DOWN interface
3227  *      (4) - valid local IP broadcast addresses. In this case
3228  *      the conn will only receive packets destined to
3229  *      the specified broadcast address.
3230  *      (5) - a multicast address. In this case
3231  *      the conn will only receive packets destined to
3232  *      the specified multicast address. Note: the
3233  *      application still has to issue an
3234  *      IP_ADD_MEMBERSHIP socket option.
3235  *
3236  * In all the above cases, the bound address must be valid in the current zone.
3237  * When the address is loopback, multicast or broadcast, there might be many
3238  * matching IREs so bind has to look up based on the zone.
3239  */
3240 ip_laddr_t
3241 ip_laddr_verify_v4(ipaddr_t src_addr, zoneid_t zoneid,
3242     ip_stack_t *ipst, boolean_t allow_mcbc)
3243 {
3244         ire_t *src_ire;
3245 
3246         ASSERT(src_addr != INADDR_ANY);
3247 
3248         src_ire = ire_ftable_lookup_v4(src_addr, 0, 0, 0,
3249             NULL, zoneid, NULL, MATCH_IRE_ZONEONLY, 0, ipst, NULL);
3250 
3251         /*
3252          * If an address other than in6addr_any is requested,
3253          * we verify that it is a valid address for bind
3254          * Note: Following code is in if-else-if form for
3255          * readability compared to a condition check.
3256          */
3257         if (src_ire != NULL && (src_ire->ire_type & (IRE_LOCAL|IRE_LOOPBACK))) {
3258                 /*
3259                  * (2) Bind to address of local UP interface
3260                  */
3261                 ire_refrele(src_ire);
3262                 return (IPVL_UNICAST_UP);
3263         } else if (src_ire != NULL && src_ire->ire_type & IRE_BROADCAST) {
3264                 /*
3265                  * (4) Bind to broadcast address
3266                  */
3267                 ire_refrele(src_ire);
3268                 if (allow_mcbc)
3269                         return (IPVL_BCAST);
3270                 else
3271                         return (IPVL_BAD);
3272         } else if (CLASSD(src_addr)) {
3273                 /* (5) bind to multicast address. */
3274                 if (src_ire != NULL)
3275                         ire_refrele(src_ire);
3276 
3277                 if (allow_mcbc)
3278                         return (IPVL_MCAST);
3279                 else
3280                         return (IPVL_BAD);
3281         } else {
3282                 ipif_t *ipif;
3283 
3284                 /*
3285                  * (3) Bind to address of local DOWN interface?
3286                  * (ipif_lookup_addr() looks up all interfaces
3287                  * but we do not get here for UP interfaces
3288                  * - case (2) above)
3289                  */
3290                 if (src_ire != NULL)
3291                         ire_refrele(src_ire);
3292 
3293                 ipif = ipif_lookup_addr(src_addr, NULL, zoneid, ipst);
3294                 if (ipif == NULL)
3295                         return (IPVL_BAD);
3296 
3297                 /* Not a useful source? */
3298                 if (ipif->ipif_flags & (IPIF_NOLOCAL | IPIF_ANYCAST)) {
3299                         ipif_refrele(ipif);
3300                         return (IPVL_BAD);
3301                 }
3302                 ipif_refrele(ipif);
3303                 return (IPVL_UNICAST_DOWN);
3304         }
3305 }
3306 
3307 /*
3308  * Insert in the bind fanout for IPv4 and IPv6.
3309  * The caller should already have used ip_laddr_verify_v*() before calling
3310  * this.
3311  */
3312 int
3313 ip_laddr_fanout_insert(conn_t *connp)
3314 {
3315         int             error;
3316 
3317         /*
3318          * Allow setting new policies. For example, disconnects result
3319          * in us being called. As we would have set conn_policy_cached
3320          * to B_TRUE before, we should set it to B_FALSE, so that policy
3321          * can change after the disconnect.
3322          */
3323         connp->conn_policy_cached = B_FALSE;
3324 
3325         error = ipcl_bind_insert(connp);
3326         if (error != 0) {
3327                 if (connp->conn_anon_port) {
3328                         (void) tsol_mlp_anon(crgetzone(connp->conn_cred),
3329                             connp->conn_mlp_type, connp->conn_proto,
3330                             ntohs(connp->conn_lport), B_FALSE);
3331                 }
3332                 connp->conn_mlp_type = mlptSingle;
3333         }
3334         return (error);
3335 }
3336 
3337 /*
3338  * Verify that both the source and destination addresses are valid. If
3339  * IPDF_VERIFY_DST is not set, then the destination address may be unreachable,
3340  * i.e. have no route to it.  Protocols like TCP want to verify destination
3341  * reachability, while tunnels do not.
3342  *
3343  * Determine the route, the interface, and (optionally) the source address
3344  * to use to reach a given destination.
3345  * Note that we allow connect to broadcast and multicast addresses when
3346  * IPDF_ALLOW_MCBC is set.
3347  * first_hop and dst_addr are normally the same, but if source routing
3348  * they will differ; in that case the first_hop is what we'll use for the
3349  * routing lookup but the dce and label checks will be done on dst_addr,
3350  *
3351  * If uinfo is set, then we fill in the best available information
3352  * we have for the destination. This is based on (in priority order) any
3353  * metrics and path MTU stored in a dce_t, route metrics, and finally the
3354  * ill_mtu/ill_mc_mtu.
3355  *
3356  * Tsol note: If we have a source route then dst_addr != firsthop. But we
3357  * always do the label check on dst_addr.
3358  */
3359 int
3360 ip_set_destination_v4(ipaddr_t *src_addrp, ipaddr_t dst_addr, ipaddr_t firsthop,
3361     ip_xmit_attr_t *ixa, iulp_t *uinfo, uint32_t flags, uint_t mac_mode)
3362 {
3363         ire_t           *ire = NULL;
3364         int             error = 0;
3365         ipaddr_t        setsrc;                         /* RTF_SETSRC */
3366         zoneid_t        zoneid = ixa->ixa_zoneid;    /* Honors SO_ALLZONES */
3367         ip_stack_t      *ipst = ixa->ixa_ipst;
3368         dce_t           *dce;
3369         uint_t          pmtu;
3370         uint_t          generation;
3371         nce_t           *nce;
3372         ill_t           *ill = NULL;
3373         boolean_t       multirt = B_FALSE;
3374 
3375         ASSERT(ixa->ixa_flags & IXAF_IS_IPV4);
3376 
3377         /*
3378          * We never send to zero; the ULPs map it to the loopback address.
3379          * We can't allow it since we use zero to mean unitialized in some
3380          * places.
3381          */
3382         ASSERT(dst_addr != INADDR_ANY);
3383 
3384         if (is_system_labeled()) {
3385                 ts_label_t *tsl = NULL;
3386 
3387                 error = tsol_check_dest(ixa->ixa_tsl, &dst_addr, IPV4_VERSION,
3388                     mac_mode, (flags & IPDF_ZONE_IS_GLOBAL) != 0, &tsl);
3389                 if (error != 0)
3390                         return (error);
3391                 if (tsl != NULL) {
3392                         /* Update the label */
3393                         ip_xmit_attr_replace_tsl(ixa, tsl);
3394                 }
3395         }
3396 
3397         setsrc = INADDR_ANY;
3398         /*
3399          * Select a route; For IPMP interfaces, we would only select
3400          * a "hidden" route (i.e., going through a specific under_ill)
3401          * if ixa_ifindex has been specified.
3402          */
3403         ire = ip_select_route_v4(firsthop, *src_addrp, ixa,
3404             &generation, &setsrc, &error, &multirt);
3405         ASSERT(ire != NULL);    /* IRE_NOROUTE if none found */
3406         if (error != 0)
3407                 goto bad_addr;
3408 
3409         /*
3410          * ire can't be a broadcast or multicast unless IPDF_ALLOW_MCBC is set.
3411          * If IPDF_VERIFY_DST is set, the destination must be reachable;
3412          * Otherwise the destination needn't be reachable.
3413          *
3414          * If we match on a reject or black hole, then we've got a
3415          * local failure.  May as well fail out the connect() attempt,
3416          * since it's never going to succeed.
3417          */
3418         if (ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE)) {
3419                 /*
3420                  * If we're verifying destination reachability, we always want
3421                  * to complain here.
3422                  *
3423                  * If we're not verifying destination reachability but the
3424                  * destination has a route, we still want to fail on the
3425                  * temporary address and broadcast address tests.
3426                  *
3427                  * In both cases do we let the code continue so some reasonable
3428                  * information is returned to the caller. That enables the
3429                  * caller to use (and even cache) the IRE. conn_ip_ouput will
3430                  * use the generation mismatch path to check for the unreachable
3431                  * case thereby avoiding any specific check in the main path.
3432                  */
3433                 ASSERT(generation == IRE_GENERATION_VERIFY);
3434                 if (flags & IPDF_VERIFY_DST) {
3435                         /*
3436                          * Set errno but continue to set up ixa_ire to be
3437                          * the RTF_REJECT|RTF_BLACKHOLE IRE.
3438                          * That allows callers to use ip_output to get an
3439                          * ICMP error back.
3440                          */
3441                         if (!(ire->ire_type & IRE_HOST))
3442                                 error = ENETUNREACH;
3443                         else
3444                                 error = EHOSTUNREACH;
3445                 }
3446         }
3447 
3448         if ((ire->ire_type & (IRE_BROADCAST|IRE_MULTICAST)) &&
3449             !(flags & IPDF_ALLOW_MCBC)) {
3450                 ire_refrele(ire);
3451                 ire = ire_reject(ipst, B_FALSE);
3452                 generation = IRE_GENERATION_VERIFY;
3453                 error = ENETUNREACH;
3454         }
3455 
3456         /* Cache things */
3457         if (ixa->ixa_ire != NULL)
3458                 ire_refrele_notr(ixa->ixa_ire);
3459 #ifdef DEBUG
3460         ire_refhold_notr(ire);
3461         ire_refrele(ire);
3462 #endif
3463         ixa->ixa_ire = ire;
3464         ixa->ixa_ire_generation = generation;
3465 
3466         /*
3467          * Ensure that ixa_dce is always set any time that ixa_ire is set,
3468          * since some callers will send a packet to conn_ip_output() even if
3469          * there's an error.
3470          */
3471         if (flags & IPDF_UNIQUE_DCE) {
3472                 /* Fallback to the default dce if allocation fails */
3473                 dce = dce_lookup_and_add_v4(dst_addr, ipst);
3474                 if (dce != NULL)
3475                         generation = dce->dce_generation;
3476                 else
3477                         dce = dce_lookup_v4(dst_addr, ipst, &generation);
3478         } else {
3479                 dce = dce_lookup_v4(dst_addr, ipst, &generation);
3480         }
3481         ASSERT(dce != NULL);
3482         if (ixa->ixa_dce != NULL)
3483                 dce_refrele_notr(ixa->ixa_dce);
3484 #ifdef DEBUG
3485         dce_refhold_notr(dce);
3486         dce_refrele(dce);
3487 #endif
3488         ixa->ixa_dce = dce;
3489         ixa->ixa_dce_generation = generation;
3490 
3491         /*
3492          * For multicast with multirt we have a flag passed back from
3493          * ire_lookup_multi_ill_v4 since we don't have an IRE for each
3494          * possible multicast address.
3495          * We also need a flag for multicast since we can't check
3496          * whether RTF_MULTIRT is set in ixa_ire for multicast.
3497          */
3498         if (multirt) {
3499                 ixa->ixa_postfragfn = ip_postfrag_multirt_v4;
3500                 ixa->ixa_flags |= IXAF_MULTIRT_MULTICAST;
3501         } else {
3502                 ixa->ixa_postfragfn = ire->ire_postfragfn;
3503                 ixa->ixa_flags &= ~IXAF_MULTIRT_MULTICAST;
3504         }
3505         if (!(ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE))) {
3506                 /* Get an nce to cache. */
3507                 nce = ire_to_nce(ire, firsthop, NULL);
3508                 if (nce == NULL) {
3509                         /* Allocation failure? */
3510                         ixa->ixa_ire_generation = IRE_GENERATION_VERIFY;
3511                 } else {
3512                         if (ixa->ixa_nce != NULL)
3513                                 nce_refrele(ixa->ixa_nce);
3514                         ixa->ixa_nce = nce;
3515                 }
3516         }
3517 
3518         /*
3519          * If the source address is a loopback address, the
3520          * destination had best be local or multicast.
3521          * If we are sending to an IRE_LOCAL using a loopback source then
3522          * it had better be the same zoneid.
3523          */
3524         if (*src_addrp == htonl(INADDR_LOOPBACK)) {
3525                 if ((ire->ire_type & IRE_LOCAL) && ire->ire_zoneid != zoneid) {
3526                         ire = NULL;     /* Stored in ixa_ire */
3527                         error = EADDRNOTAVAIL;
3528                         goto bad_addr;
3529                 }
3530                 if (!(ire->ire_type & (IRE_LOOPBACK|IRE_LOCAL|IRE_MULTICAST))) {
3531                         ire = NULL;     /* Stored in ixa_ire */
3532                         error = EADDRNOTAVAIL;
3533                         goto bad_addr;
3534                 }
3535         }
3536         if (ire->ire_type & IRE_BROADCAST) {
3537                 /*
3538                  * If the ULP didn't have a specified source, then we
3539                  * make sure we reselect the source when sending
3540                  * broadcasts out different interfaces.
3541                  */
3542                 if (flags & IPDF_SELECT_SRC)
3543                         ixa->ixa_flags |= IXAF_SET_SOURCE;
3544                 else
3545                         ixa->ixa_flags &= ~IXAF_SET_SOURCE;
3546         }
3547 
3548         /*
3549          * Does the caller want us to pick a source address?
3550          */
3551         if (flags & IPDF_SELECT_SRC) {
3552                 ipaddr_t        src_addr;
3553 
3554                 /*
3555                  * We use use ire_nexthop_ill to avoid the under ipmp
3556                  * interface for source address selection. Note that for ipmp
3557                  * probe packets, ixa_ifindex would have been specified, and
3558                  * the ip_select_route() invocation would have picked an ire
3559                  * will ire_ill pointing at an under interface.
3560                  */
3561                 ill = ire_nexthop_ill(ire);
3562 
3563                 /* If unreachable we have no ill but need some source */
3564                 if (ill == NULL) {
3565                         src_addr = htonl(INADDR_LOOPBACK);
3566                         /* Make sure we look for a better source address */
3567                         generation = SRC_GENERATION_VERIFY;
3568                 } else {
3569                         error = ip_select_source_v4(ill, setsrc, dst_addr,
3570                             ixa->ixa_multicast_ifaddr, zoneid,
3571                             ipst, &src_addr, &generation, NULL);
3572                         if (error != 0) {
3573                                 ire = NULL;     /* Stored in ixa_ire */
3574                                 goto bad_addr;
3575                         }
3576                 }
3577 
3578                 /*
3579                  * We allow the source address to to down.
3580                  * However, we check that we don't use the loopback address
3581                  * as a source when sending out on the wire.
3582                  */
3583                 if ((src_addr == htonl(INADDR_LOOPBACK)) &&
3584                     !(ire->ire_type & (IRE_LOCAL|IRE_LOOPBACK|IRE_MULTICAST)) &&
3585                     !(ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE))) {
3586                         ire = NULL;     /* Stored in ixa_ire */
3587                         error = EADDRNOTAVAIL;
3588                         goto bad_addr;
3589                 }
3590 
3591                 *src_addrp = src_addr;
3592                 ixa->ixa_src_generation = generation;
3593         }
3594 
3595         /*
3596          * Make sure we don't leave an unreachable ixa_nce in place
3597          * since ip_select_route is used when we unplumb i.e., remove
3598          * references on ixa_ire, ixa_nce, and ixa_dce.
3599          */
3600         nce = ixa->ixa_nce;
3601         if (nce != NULL && nce->nce_is_condemned) {
3602                 nce_refrele(nce);
3603                 ixa->ixa_nce = NULL;
3604                 ixa->ixa_ire_generation = IRE_GENERATION_VERIFY;
3605         }
3606 
3607         /*
3608          * The caller has set IXAF_PMTU_DISCOVERY if path MTU is desired.
3609          * However, we can't do it for IPv4 multicast or broadcast.
3610          */
3611         if (ire->ire_type & (IRE_BROADCAST|IRE_MULTICAST))
3612                 ixa->ixa_flags &= ~IXAF_PMTU_DISCOVERY;
3613 
3614         /*
3615          * Set initial value for fragmentation limit. Either conn_ip_output
3616          * or ULP might updates it when there are routing changes.
3617          * Handles a NULL ixa_ire->ire_ill or a NULL ixa_nce for RTF_REJECT.
3618          */
3619         pmtu = ip_get_pmtu(ixa);
3620         ixa->ixa_fragsize = pmtu;
3621         /* Make sure ixa_fragsize and ixa_pmtu remain identical */
3622         if (ixa->ixa_flags & IXAF_VERIFY_PMTU)
3623                 ixa->ixa_pmtu = pmtu;
3624 
3625         /*
3626          * Extract information useful for some transports.
3627          * First we look for DCE metrics. Then we take what we have in
3628          * the metrics in the route, where the offlink is used if we have
3629          * one.
3630          */
3631         if (uinfo != NULL) {
3632                 bzero(uinfo, sizeof (*uinfo));
3633 
3634                 if (dce->dce_flags & DCEF_UINFO)
3635                         *uinfo = dce->dce_uinfo;
3636 
3637                 rts_merge_metrics(uinfo, &ire->ire_metrics);
3638 
3639                 /* Allow ire_metrics to decrease the path MTU from above */
3640                 if (uinfo->iulp_mtu == 0 || uinfo->iulp_mtu > pmtu)
3641                         uinfo->iulp_mtu = pmtu;
3642 
3643                 uinfo->iulp_localnet = (ire->ire_type & IRE_ONLINK) != 0;
3644                 uinfo->iulp_loopback = (ire->ire_type & IRE_LOOPBACK) != 0;
3645                 uinfo->iulp_local = (ire->ire_type & IRE_LOCAL) != 0;
3646         }
3647 
3648         if (ill != NULL)
3649                 ill_refrele(ill);
3650 
3651         return (error);
3652 
3653 bad_addr:
3654         if (ire != NULL)
3655                 ire_refrele(ire);
3656 
3657         if (ill != NULL)
3658                 ill_refrele(ill);
3659 
3660         /*
3661          * Make sure we don't leave an unreachable ixa_nce in place
3662          * since ip_select_route is used when we unplumb i.e., remove
3663          * references on ixa_ire, ixa_nce, and ixa_dce.
3664          */
3665         nce = ixa->ixa_nce;
3666         if (nce != NULL && nce->nce_is_condemned) {
3667                 nce_refrele(nce);
3668                 ixa->ixa_nce = NULL;
3669                 ixa->ixa_ire_generation = IRE_GENERATION_VERIFY;
3670         }
3671 
3672         return (error);
3673 }
3674 
3675 
3676 /*
3677  * Get the base MTU for the case when path MTU discovery is not used.
3678  * Takes the MTU of the IRE into account.
3679  */
3680 uint_t
3681 ip_get_base_mtu(ill_t *ill, ire_t *ire)
3682 {
3683         uint_t mtu;
3684         uint_t iremtu = ire->ire_metrics.iulp_mtu;
3685 
3686         if (ire->ire_type & (IRE_MULTICAST|IRE_BROADCAST))
3687                 mtu = ill->ill_mc_mtu;
3688         else
3689                 mtu = ill->ill_mtu;
3690 
3691         if (iremtu != 0 && iremtu < mtu)
3692                 mtu = iremtu;
3693 
3694         return (mtu);
3695 }
3696 
3697 /*
3698  * Get the PMTU for the attributes. Handles both IPv4 and IPv6.
3699  * Assumes that ixa_ire, dce, and nce have already been set up.
3700  *
3701  * The caller has set IXAF_PMTU_DISCOVERY if path MTU discovery is desired.
3702  * We avoid path MTU discovery if it is disabled with ndd.
3703  * Furtermore, if the path MTU is too small, then we don't set DF for IPv4.
3704  *
3705  * NOTE: We also used to turn it off for source routed packets. That
3706  * is no longer required since the dce is per final destination.
3707  */
3708 uint_t
3709 ip_get_pmtu(ip_xmit_attr_t *ixa)
3710 {
3711         ip_stack_t      *ipst = ixa->ixa_ipst;
3712         dce_t           *dce;
3713         nce_t           *nce;
3714         ire_t           *ire;
3715         uint_t          pmtu;
3716 
3717         ire = ixa->ixa_ire;
3718         dce = ixa->ixa_dce;
3719         nce = ixa->ixa_nce;
3720 
3721         /*
3722          * If path MTU discovery has been turned off by ndd, then we ignore
3723          * any dce_pmtu and for IPv4 we will not set DF.
3724          */
3725         if (!ipst->ips_ip_path_mtu_discovery)
3726                 ixa->ixa_flags &= ~IXAF_PMTU_DISCOVERY;
3727 
3728         pmtu = IP_MAXPACKET;
3729         /*
3730          * Decide whether whether IPv4 sets DF
3731          * For IPv6 "no DF" means to use the 1280 mtu
3732          */
3733         if (ixa->ixa_flags & IXAF_PMTU_DISCOVERY) {
3734                 ixa->ixa_flags |= IXAF_PMTU_IPV4_DF;
3735         } else {
3736                 ixa->ixa_flags &= ~IXAF_PMTU_IPV4_DF;
3737                 if (!(ixa->ixa_flags & IXAF_IS_IPV4))
3738                         pmtu = IPV6_MIN_MTU;
3739         }
3740 
3741         /* Check if the PMTU is to old before we use it */
3742         if ((dce->dce_flags & DCEF_PMTU) &&
3743             TICK_TO_SEC(ddi_get_lbolt64()) - dce->dce_last_change_time >
3744             ipst->ips_ip_pathmtu_interval) {
3745                 /*
3746                  * Older than 20 minutes. Drop the path MTU information.
3747                  */
3748                 mutex_enter(&dce->dce_lock);
3749                 dce->dce_flags &= ~(DCEF_PMTU|DCEF_TOO_SMALL_PMTU);
3750                 dce->dce_last_change_time = TICK_TO_SEC(ddi_get_lbolt64());
3751                 mutex_exit(&dce->dce_lock);
3752                 dce_increment_generation(dce);
3753         }
3754 
3755         /* The metrics on the route can lower the path MTU */
3756         if (ire->ire_metrics.iulp_mtu != 0 &&
3757             ire->ire_metrics.iulp_mtu < pmtu)
3758                 pmtu = ire->ire_metrics.iulp_mtu;
3759 
3760         /*
3761          * If the path MTU is smaller than some minimum, we still use dce_pmtu
3762          * above (would be 576 for IPv4 and 1280 for IPv6), but we clear
3763          * IXAF_PMTU_IPV4_DF so that we avoid setting DF for IPv4.
3764          */
3765         if (ixa->ixa_flags & IXAF_PMTU_DISCOVERY) {
3766                 if (dce->dce_flags & DCEF_PMTU) {
3767                         if (dce->dce_pmtu < pmtu)
3768                                 pmtu = dce->dce_pmtu;
3769 
3770                         if (dce->dce_flags & DCEF_TOO_SMALL_PMTU) {
3771                                 ixa->ixa_flags |= IXAF_PMTU_TOO_SMALL;
3772                                 ixa->ixa_flags &= ~IXAF_PMTU_IPV4_DF;
3773                         } else {
3774                                 ixa->ixa_flags &= ~IXAF_PMTU_TOO_SMALL;
3775                                 ixa->ixa_flags |= IXAF_PMTU_IPV4_DF;
3776                         }
3777                 } else {
3778                         ixa->ixa_flags &= ~IXAF_PMTU_TOO_SMALL;
3779                         ixa->ixa_flags |= IXAF_PMTU_IPV4_DF;
3780                 }
3781         }
3782 
3783         /*
3784          * If we have an IRE_LOCAL we use the loopback mtu instead of
3785          * the ill for going out the wire i.e., IRE_LOCAL gets the same
3786          * mtu as IRE_LOOPBACK.
3787          */
3788         if (ire->ire_type & (IRE_LOCAL|IRE_LOOPBACK)) {
3789                 uint_t loopback_mtu;
3790 
3791                 loopback_mtu = (ire->ire_ipversion == IPV6_VERSION) ?
3792                     ip_loopback_mtu_v6plus : ip_loopback_mtuplus;
3793 
3794                 if (loopback_mtu < pmtu)
3795                         pmtu = loopback_mtu;
3796         } else if (nce != NULL) {
3797                 /*
3798                  * Make sure we don't exceed the interface MTU.
3799                  * In the case of RTF_REJECT or RTF_BLACKHOLE we might not have
3800                  * an ill. We'd use the above IP_MAXPACKET in that case just
3801                  * to tell the transport something larger than zero.
3802                  */
3803                 if (ire->ire_type & (IRE_MULTICAST|IRE_BROADCAST)) {
3804                         if (nce->nce_common->ncec_ill->ill_mc_mtu < pmtu)
3805                                 pmtu = nce->nce_common->ncec_ill->ill_mc_mtu;
3806                         if (nce->nce_common->ncec_ill != nce->nce_ill &&
3807                             nce->nce_ill->ill_mc_mtu < pmtu) {
3808                                 /*
3809                                  * for interfaces in an IPMP group, the mtu of
3810                                  * the nce_ill (under_ill) could be different
3811                                  * from the mtu of the ncec_ill, so we take the
3812                                  * min of the two.
3813                                  */
3814                                 pmtu = nce->nce_ill->ill_mc_mtu;
3815                         }
3816                 } else {
3817                         if (nce->nce_common->ncec_ill->ill_mtu < pmtu)
3818                                 pmtu = nce->nce_common->ncec_ill->ill_mtu;
3819                         if (nce->nce_common->ncec_ill != nce->nce_ill &&
3820                             nce->nce_ill->ill_mtu < pmtu) {
3821                                 /*
3822                                  * for interfaces in an IPMP group, the mtu of
3823                                  * the nce_ill (under_ill) could be different
3824                                  * from the mtu of the ncec_ill, so we take the
3825                                  * min of the two.
3826                                  */
3827                                 pmtu = nce->nce_ill->ill_mtu;
3828                         }
3829                 }
3830         }
3831 
3832         /*
3833          * Handle the IPV6_USE_MIN_MTU socket option or ancillary data.
3834          * Only applies to IPv6.
3835          */
3836         if (!(ixa->ixa_flags & IXAF_IS_IPV4)) {
3837                 if (ixa->ixa_flags & IXAF_USE_MIN_MTU) {
3838                         switch (ixa->ixa_use_min_mtu) {
3839                         case IPV6_USE_MIN_MTU_MULTICAST:
3840                                 if (ire->ire_type & IRE_MULTICAST)
3841                                         pmtu = IPV6_MIN_MTU;
3842                                 break;
3843                         case IPV6_USE_MIN_MTU_ALWAYS:
3844                                 pmtu = IPV6_MIN_MTU;
3845                                 break;
3846                         case IPV6_USE_MIN_MTU_NEVER:
3847                                 break;
3848                         }
3849                 } else {
3850                         /* Default is IPV6_USE_MIN_MTU_MULTICAST */
3851                         if (ire->ire_type & IRE_MULTICAST)
3852                                 pmtu = IPV6_MIN_MTU;
3853                 }
3854         }
3855 
3856         /*
3857          * For multirouted IPv6 packets, the IP layer will insert a 8-byte
3858          * fragment header in every packet. We compensate for those cases by
3859          * returning a smaller path MTU to the ULP.
3860          *
3861          * In the case of CGTP then ip_output will add a fragment header.
3862          * Make sure there is room for it by telling a smaller number
3863          * to the transport.
3864          *
3865          * When IXAF_IPV6_ADDR_FRAGHDR we subtract the frag hdr here
3866          * so the ULPs consistently see a iulp_pmtu and ip_get_pmtu()
3867          * which is the size of the packets it can send.
3868          */
3869         if (!(ixa->ixa_flags & IXAF_IS_IPV4)) {
3870                 if ((ire->ire_flags & RTF_MULTIRT) ||
3871                     (ixa->ixa_flags & IXAF_MULTIRT_MULTICAST)) {
3872                         pmtu -= sizeof (ip6_frag_t);
3873                         ixa->ixa_flags |= IXAF_IPV6_ADD_FRAGHDR;
3874                 }
3875         }
3876 
3877         return (pmtu);
3878 }
3879 
3880 /*
3881  * Carve "len" bytes out of an mblk chain, consuming any we empty, and duping
3882  * the final piece where we don't.  Return a pointer to the first mblk in the
3883  * result, and update the pointer to the next mblk to chew on.  If anything
3884  * goes wrong (i.e., dupb fails), we waste everything in sight and return a
3885  * NULL pointer.
3886  */
3887 mblk_t *
3888 ip_carve_mp(mblk_t **mpp, ssize_t len)
3889 {
3890         mblk_t  *mp0;
3891         mblk_t  *mp1;
3892         mblk_t  *mp2;
3893 
3894         if (!len || !mpp || !(mp0 = *mpp))
3895                 return (NULL);
3896         /* If we aren't going to consume the first mblk, we need a dup. */
3897         if (mp0->b_wptr - mp0->b_rptr > len) {
3898                 mp1 = dupb(mp0);
3899                 if (mp1) {
3900                         /* Partition the data between the two mblks. */
3901                         mp1->b_wptr = mp1->b_rptr + len;
3902                         mp0->b_rptr = mp1->b_wptr;
3903                         /*
3904                          * after adjustments if mblk not consumed is now
3905                          * unaligned, try to align it. If this fails free
3906                          * all messages and let upper layer recover.
3907                          */
3908                         if (!OK_32PTR(mp0->b_rptr)) {
3909                                 if (!pullupmsg(mp0, -1)) {
3910                                         freemsg(mp0);
3911                                         freemsg(mp1);
3912                                         *mpp = NULL;
3913                                         return (NULL);
3914                                 }
3915                         }
3916                 }
3917                 return (mp1);
3918         }
3919         /* Eat through as many mblks as we need to get len bytes. */
3920         len -= mp0->b_wptr - mp0->b_rptr;
3921         for (mp2 = mp1 = mp0; (mp2 = mp2->b_cont) != 0 && len; mp1 = mp2) {
3922                 if (mp2->b_wptr - mp2->b_rptr > len) {
3923                         /*
3924                          * We won't consume the entire last mblk.  Like
3925                          * above, dup and partition it.
3926                          */
3927                         mp1->b_cont = dupb(mp2);
3928                         mp1 = mp1->b_cont;
3929                         if (!mp1) {
3930                                 /*
3931                                  * Trouble.  Rather than go to a lot of
3932                                  * trouble to clean up, we free the messages.
3933                                  * This won't be any worse than losing it on
3934                                  * the wire.
3935                                  */
3936                                 freemsg(mp0);
3937                                 freemsg(mp2);
3938                                 *mpp = NULL;
3939                                 return (NULL);
3940                         }
3941                         mp1->b_wptr = mp1->b_rptr + len;
3942                         mp2->b_rptr = mp1->b_wptr;
3943                         /*
3944                          * after adjustments if mblk not consumed is now
3945                          * unaligned, try to align it. If this fails free
3946                          * all messages and let upper layer recover.
3947                          */
3948                         if (!OK_32PTR(mp2->b_rptr)) {
3949                                 if (!pullupmsg(mp2, -1)) {
3950                                         freemsg(mp0);
3951                                         freemsg(mp2);
3952                                         *mpp = NULL;
3953                                         return (NULL);
3954                                 }
3955                         }
3956                         *mpp = mp2;
3957                         return (mp0);
3958                 }
3959                 /* Decrement len by the amount we just got. */
3960                 len -= mp2->b_wptr - mp2->b_rptr;
3961         }
3962         /*
3963          * len should be reduced to zero now.  If not our caller has
3964          * screwed up.
3965          */
3966         if (len) {
3967                 /* Shouldn't happen! */
3968                 freemsg(mp0);
3969                 *mpp = NULL;
3970                 return (NULL);
3971         }
3972         /*
3973          * We consumed up to exactly the end of an mblk.  Detach the part
3974          * we are returning from the rest of the chain.
3975          */
3976         mp1->b_cont = NULL;
3977         *mpp = mp2;
3978         return (mp0);
3979 }
3980 
3981 /* The ill stream is being unplumbed. Called from ip_close */
3982 int
3983 ip_modclose(ill_t *ill)
3984 {
3985         boolean_t success;
3986         ipsq_t  *ipsq;
3987         ipif_t  *ipif;
3988         queue_t *q = ill->ill_rq;
3989         ip_stack_t      *ipst = ill->ill_ipst;
3990         int     i;
3991         arl_ill_common_t *ai = ill->ill_common;
3992 
3993         /*
3994          * The punlink prior to this may have initiated a capability
3995          * negotiation. But ipsq_enter will block until that finishes or
3996          * times out.
3997          */
3998         success = ipsq_enter(ill, B_FALSE, NEW_OP);
3999 
4000         /*
4001          * Open/close/push/pop is guaranteed to be single threaded
4002          * per stream by STREAMS. FS guarantees that all references
4003          * from top are gone before close is called. So there can't
4004          * be another close thread that has set CONDEMNED on this ill.
4005          * and cause ipsq_enter to return failure.
4006          */
4007         ASSERT(success);
4008         ipsq = ill->ill_phyint->phyint_ipsq;
4009 
4010         /*
4011          * Mark it condemned. No new reference will be made to this ill.
4012          * Lookup functions will return an error. Threads that try to
4013          * increment the refcnt must check for ILL_CAN_LOOKUP. This ensures
4014          * that the refcnt will drop down to zero.
4015          */
4016         mutex_enter(&ill->ill_lock);
4017         ill->ill_state_flags |= ILL_CONDEMNED;
4018         for (ipif = ill->ill_ipif; ipif != NULL;
4019             ipif = ipif->ipif_next) {
4020                 ipif->ipif_state_flags |= IPIF_CONDEMNED;
4021         }
4022         /*
4023          * Wake up anybody waiting to enter the ipsq. ipsq_enter
4024          * returns  error if ILL_CONDEMNED is set
4025          */
4026         cv_broadcast(&ill->ill_cv);
4027         mutex_exit(&ill->ill_lock);
4028 
4029         /*
4030          * Send all the deferred DLPI messages downstream which came in
4031          * during the small window right before ipsq_enter(). We do this
4032          * without waiting for the ACKs because all the ACKs for M_PROTO
4033          * messages are ignored in ip_rput() when ILL_CONDEMNED is set.
4034          */
4035         ill_dlpi_send_deferred(ill);
4036 
4037         /*
4038          * Shut down fragmentation reassembly.
4039          * ill_frag_timer won't start a timer again.
4040          * Now cancel any existing timer
4041          */
4042         (void) untimeout(ill->ill_frag_timer_id);
4043         (void) ill_frag_timeout(ill, 0);
4044 
4045         /*
4046          * Call ill_delete to bring down the ipifs, ilms and ill on
4047          * this ill. Then wait for the refcnts to drop to zero.
4048          * ill_is_freeable checks whether the ill is really quiescent.
4049          * Then make sure that threads that are waiting to enter the
4050          * ipsq have seen the error returned by ipsq_enter and have
4051          * gone away. Then we call ill_delete_tail which does the
4052          * DL_UNBIND_REQ with the driver and then qprocsoff.
4053          */
4054         ill_delete(ill);
4055         mutex_enter(&ill->ill_lock);
4056         while (!ill_is_freeable(ill))
4057                 cv_wait(&ill->ill_cv, &ill->ill_lock);
4058 
4059         while (ill->ill_waiters)
4060                 cv_wait(&ill->ill_cv, &ill->ill_lock);
4061 
4062         mutex_exit(&ill->ill_lock);
4063 
4064         /*
4065          * ill_delete_tail drops reference on ill_ipst, but we need to keep
4066          * it held until the end of the function since the cleanup
4067          * below needs to be able to use the ip_stack_t.
4068          */
4069         netstack_hold(ipst->ips_netstack);
4070 
4071         /* qprocsoff is done via ill_delete_tail */
4072         ill_delete_tail(ill);
4073         /*
4074          * synchronously wait for arp stream to unbind. After this, we
4075          * cannot get any data packets up from the driver.
4076          */
4077         arp_unbind_complete(ill);
4078         ASSERT(ill->ill_ipst == NULL);
4079 
4080         /*
4081          * Walk through all conns and qenable those that have queued data.
4082          * Close synchronization needs this to
4083          * be done to ensure that all upper layers blocked
4084          * due to flow control to the closing device
4085          * get unblocked.
4086          */
4087         ip1dbg(("ip_wsrv: walking\n"));
4088         for (i = 0; i < TX_FANOUT_SIZE; i++) {
4089                 conn_walk_drain(ipst, &ipst->ips_idl_tx_list[i]);
4090         }
4091 
4092         /*
4093          * ai can be null if this is an IPv6 ill, or if the IPv4
4094          * stream is being torn down before ARP was plumbed (e.g.,
4095          * /sbin/ifconfig plumbing a stream twice, and encountering
4096          * an error
4097          */
4098         if (ai != NULL) {
4099                 ASSERT(!ill->ill_isv6);
4100                 mutex_enter(&ai->ai_lock);
4101                 ai->ai_ill = NULL;
4102                 if (ai->ai_arl == NULL) {
4103                         mutex_destroy(&ai->ai_lock);
4104                         kmem_free(ai, sizeof (*ai));
4105                 } else {
4106                         cv_signal(&ai->ai_ill_unplumb_done);
4107                         mutex_exit(&ai->ai_lock);
4108                 }
4109         }
4110 
4111         mutex_enter(&ipst->ips_ip_mi_lock);
4112         mi_close_unlink(&ipst->ips_ip_g_head, (IDP)ill);
4113         mutex_exit(&ipst->ips_ip_mi_lock);
4114 
4115         /*
4116          * credp could be null if the open didn't succeed and ip_modopen
4117          * itself calls ip_close.
4118          */
4119         if (ill->ill_credp != NULL)
4120                 crfree(ill->ill_credp);
4121 
4122         mutex_destroy(&ill->ill_saved_ire_lock);
4123         mutex_destroy(&ill->ill_lock);
4124         rw_destroy(&ill->ill_mcast_lock);
4125         mutex_destroy(&ill->ill_mcast_serializer);
4126         list_destroy(&ill->ill_nce);
4127         cv_destroy(&ill->ill_dlpi_capab_cv);
4128         mutex_destroy(&ill->ill_dlpi_capab_lock);
4129 
4130         /*
4131          * Now we are done with the module close pieces that
4132          * need the netstack_t.
4133          */
4134         netstack_rele(ipst->ips_netstack);
4135 
4136         mi_close_free((IDP)ill);
4137         q->q_ptr = WR(q)->q_ptr = NULL;
4138 
4139         ipsq_exit(ipsq);
4140 
4141         return (0);
4142 }
4143 
4144 /*
4145  * This is called as part of close() for IP, UDP, ICMP, and RTS
4146  * in order to quiesce the conn.
4147  */
4148 void
4149 ip_quiesce_conn(conn_t *connp)
4150 {
4151         boolean_t       drain_cleanup_reqd = B_FALSE;
4152         boolean_t       conn_ioctl_cleanup_reqd = B_FALSE;
4153         boolean_t       ilg_cleanup_reqd = B_FALSE;
4154         ip_stack_t      *ipst;
4155 
4156         ASSERT(!IPCL_IS_TCP(connp));
4157         ipst = connp->conn_netstack->netstack_ip;
4158 
4159         /*
4160          * Mark the conn as closing, and this conn must not be
4161          * inserted in future into any list. Eg. conn_drain_insert(),
4162          * won't insert this conn into the conn_drain_list.
4163          *
4164          * conn_idl, and conn_ilg cannot get set henceforth.
4165          */
4166         mutex_enter(&connp->conn_lock);
4167         ASSERT(!(connp->conn_state_flags & CONN_QUIESCED));
4168         connp->conn_state_flags |= CONN_CLOSING;
4169         if (connp->conn_idl != NULL)
4170                 drain_cleanup_reqd = B_TRUE;
4171         if (connp->conn_oper_pending_ill != NULL)
4172                 conn_ioctl_cleanup_reqd = B_TRUE;
4173         if (connp->conn_dhcpinit_ill != NULL) {
4174                 ASSERT(connp->conn_dhcpinit_ill->ill_dhcpinit != 0);
4175                 atomic_dec_32(&connp->conn_dhcpinit_ill->ill_dhcpinit);
4176                 ill_set_inputfn(connp->conn_dhcpinit_ill);
4177                 connp->conn_dhcpinit_ill = NULL;
4178         }
4179         if (connp->conn_ilg != NULL)
4180                 ilg_cleanup_reqd = B_TRUE;
4181         mutex_exit(&connp->conn_lock);
4182 
4183         if (conn_ioctl_cleanup_reqd)
4184                 conn_ioctl_cleanup(connp);
4185 
4186         if (is_system_labeled() && connp->conn_anon_port) {
4187                 (void) tsol_mlp_anon(crgetzone(connp->conn_cred),
4188                     connp->conn_mlp_type, connp->conn_proto,
4189                     ntohs(connp->conn_lport), B_FALSE);
4190                 connp->conn_anon_port = 0;
4191         }
4192         connp->conn_mlp_type = mlptSingle;
4193 
4194         /*
4195          * Remove this conn from any fanout list it is on.
4196          * and then wait for any threads currently operating
4197          * on this endpoint to finish
4198          */
4199         ipcl_hash_remove(connp);
4200 
4201         /*
4202          * Remove this conn from the drain list, and do any other cleanup that
4203          * may be required.  (TCP conns are never flow controlled, and
4204          * conn_idl will be NULL.)
4205          */
4206         if (drain_cleanup_reqd && connp->conn_idl != NULL) {
4207                 idl_t *idl = connp->conn_idl;
4208 
4209                 mutex_enter(&idl->idl_lock);
4210                 conn_drain(connp, B_TRUE);
4211                 mutex_exit(&idl->idl_lock);
4212         }
4213 
4214         if (connp == ipst->ips_ip_g_mrouter)
4215                 (void) ip_mrouter_done(ipst);
4216 
4217         if (ilg_cleanup_reqd)
4218                 ilg_delete_all(connp);
4219 
4220         /*
4221          * Now conn refcnt can increase only thru CONN_INC_REF_LOCKED.
4222          * callers from write side can't be there now because close
4223          * is in progress. The only other caller is ipcl_walk
4224          * which checks for the condemned flag.
4225          */
4226         mutex_enter(&connp->conn_lock);
4227         connp->conn_state_flags |= CONN_CONDEMNED;
4228         while (connp->conn_ref != 1)
4229                 cv_wait(&connp->conn_cv, &connp->conn_lock);
4230         connp->conn_state_flags |= CONN_QUIESCED;
4231         mutex_exit(&connp->conn_lock);
4232 }
4233 
4234 /* ARGSUSED */
4235 int
4236 ip_close(queue_t *q, int flags, cred_t *credp __unused)
4237 {
4238         conn_t          *connp;
4239 
4240         /*
4241          * Call the appropriate delete routine depending on whether this is
4242          * a module or device.
4243          */
4244         if (WR(q)->q_next != NULL) {
4245                 /* This is a module close */
4246                 return (ip_modclose((ill_t *)q->q_ptr));
4247         }
4248 
4249         connp = q->q_ptr;
4250         ip_quiesce_conn(connp);
4251 
4252         qprocsoff(q);
4253 
4254         /*
4255          * Now we are truly single threaded on this stream, and can
4256          * delete the things hanging off the connp, and finally the connp.
4257          * We removed this connp from the fanout list, it cannot be
4258          * accessed thru the fanouts, and we already waited for the
4259          * conn_ref to drop to 0. We are already in close, so
4260          * there cannot be any other thread from the top. qprocsoff
4261          * has completed, and service has completed or won't run in
4262          * future.
4263          */
4264         ASSERT(connp->conn_ref == 1);
4265 
4266         inet_minor_free(connp->conn_minor_arena, connp->conn_dev);
4267 
4268         connp->conn_ref--;
4269         ipcl_conn_destroy(connp);
4270 
4271         q->q_ptr = WR(q)->q_ptr = NULL;
4272         return (0);
4273 }
4274 
4275 /*
4276  * Wapper around putnext() so that ip_rts_request can merely use
4277  * conn_recv.
4278  */
4279 /*ARGSUSED2*/
4280 static void
4281 ip_conn_input(void *arg1, mblk_t *mp, void *arg2, ip_recv_attr_t *ira)
4282 {
4283         conn_t *connp = (conn_t *)arg1;
4284 
4285         putnext(connp->conn_rq, mp);
4286 }
4287 
4288 /* Dummy in case ICMP error delivery is attempted to a /dev/ip instance */
4289 /* ARGSUSED */
4290 static void
4291 ip_conn_input_icmp(void *arg1, mblk_t *mp, void *arg2, ip_recv_attr_t *ira)
4292 {
4293         freemsg(mp);
4294 }
4295 
4296 /*
4297  * Called when the module is about to be unloaded
4298  */
4299 void
4300 ip_ddi_destroy(void)
4301 {
4302         /* This needs to be called before destroying any transports. */
4303         mutex_enter(&cpu_lock);
4304         unregister_cpu_setup_func(ip_tp_cpu_update, NULL);
4305         mutex_exit(&cpu_lock);
4306 
4307         tnet_fini();
4308 
4309         icmp_ddi_g_destroy();
4310         rts_ddi_g_destroy();
4311         udp_ddi_g_destroy();
4312         sctp_ddi_g_destroy();
4313         tcp_ddi_g_destroy();
4314         ilb_ddi_g_destroy();
4315         dce_g_destroy();
4316         ipsec_policy_g_destroy();
4317         ipcl_g_destroy();
4318         ip_net_g_destroy();
4319         ip_ire_g_fini();
4320         inet_minor_destroy(ip_minor_arena_sa);
4321 #if defined(_LP64)
4322         inet_minor_destroy(ip_minor_arena_la);
4323 #endif
4324 
4325 #ifdef DEBUG
4326         list_destroy(&ip_thread_list);
4327         rw_destroy(&ip_thread_rwlock);
4328         tsd_destroy(&ip_thread_data);
4329 #endif
4330 
4331         netstack_unregister(NS_IP);
4332 }
4333 
4334 /*
4335  * First step in cleanup.
4336  */
4337 /* ARGSUSED */
4338 static void
4339 ip_stack_shutdown(netstackid_t stackid, void *arg)
4340 {
4341         ip_stack_t *ipst = (ip_stack_t *)arg;
4342         kt_did_t ktid;
4343 
4344 #ifdef NS_DEBUG
4345         printf("ip_stack_shutdown(%p, stack %d)\n", (void *)ipst, stackid);
4346 #endif
4347 
4348         /*
4349          * Perform cleanup for special interfaces (loopback and IPMP).
4350          */
4351         ip_interface_cleanup(ipst);
4352 
4353         /*
4354          * The *_hook_shutdown()s start the process of notifying any
4355          * consumers that things are going away.... nothing is destroyed.
4356          */
4357         ipv4_hook_shutdown(ipst);
4358         ipv6_hook_shutdown(ipst);
4359         arp_hook_shutdown(ipst);
4360 
4361         mutex_enter(&ipst->ips_capab_taskq_lock);
4362         ktid = ipst->ips_capab_taskq_thread->t_did;
4363         ipst->ips_capab_taskq_quit = B_TRUE;
4364         cv_signal(&ipst->ips_capab_taskq_cv);
4365         mutex_exit(&ipst->ips_capab_taskq_lock);
4366 
4367         /*
4368          * In rare occurrences, particularly on virtual hardware where CPUs can
4369          * be de-scheduled, the thread that we just signaled will not run until
4370          * after we have gotten through parts of ip_stack_fini. If that happens
4371          * then we'll try to grab the ips_capab_taskq_lock as part of returning
4372          * from cv_wait which no longer exists.
4373          */
4374         thread_join(ktid);
4375 }
4376 
4377 /*
4378  * Free the IP stack instance.
4379  */
4380 static void
4381 ip_stack_fini(netstackid_t stackid, void *arg)
4382 {
4383         ip_stack_t *ipst = (ip_stack_t *)arg;
4384         int ret;
4385 
4386 #ifdef NS_DEBUG
4387         printf("ip_stack_fini(%p, stack %d)\n", (void *)ipst, stackid);
4388 #endif
4389         /*
4390          * At this point, all of the notifications that the events and
4391          * protocols are going away have been run, meaning that we can
4392          * now set about starting to clean things up.
4393          */
4394         ipobs_fini(ipst);
4395         ipv4_hook_destroy(ipst);
4396         ipv6_hook_destroy(ipst);
4397         arp_hook_destroy(ipst);
4398         ip_net_destroy(ipst);
4399 
4400         ipmp_destroy(ipst);
4401 
4402         ip_kstat_fini(stackid, ipst->ips_ip_mibkp);
4403         ipst->ips_ip_mibkp = NULL;
4404         icmp_kstat_fini(stackid, ipst->ips_icmp_mibkp);
4405         ipst->ips_icmp_mibkp = NULL;
4406         ip_kstat2_fini(stackid, ipst->ips_ip_kstat);
4407         ipst->ips_ip_kstat = NULL;
4408         bzero(&ipst->ips_ip_statistics, sizeof (ipst->ips_ip_statistics));
4409         ip6_kstat_fini(stackid, ipst->ips_ip6_kstat);
4410         ipst->ips_ip6_kstat = NULL;
4411         bzero(&ipst->ips_ip6_statistics, sizeof (ipst->ips_ip6_statistics));
4412 
4413         kmem_free(ipst->ips_propinfo_tbl,
4414             ip_propinfo_count * sizeof (mod_prop_info_t));
4415         ipst->ips_propinfo_tbl = NULL;
4416 
4417         dce_stack_destroy(ipst);
4418         ip_mrouter_stack_destroy(ipst);
4419 
4420         /*
4421          * Quiesce all of our timers. Note we set the quiesce flags before we
4422          * call untimeout. The slowtimers may actually kick off another instance
4423          * of the non-slow timers.
4424          */
4425         mutex_enter(&ipst->ips_igmp_timer_lock);
4426         ipst->ips_igmp_timer_quiesce = B_TRUE;
4427         mutex_exit(&ipst->ips_igmp_timer_lock);
4428 
4429         mutex_enter(&ipst->ips_mld_timer_lock);
4430         ipst->ips_mld_timer_quiesce = B_TRUE;
4431         mutex_exit(&ipst->ips_mld_timer_lock);
4432 
4433         mutex_enter(&ipst->ips_igmp_slowtimeout_lock);
4434         ipst->ips_igmp_slowtimeout_quiesce = B_TRUE;
4435         mutex_exit(&ipst->ips_igmp_slowtimeout_lock);
4436 
4437         mutex_enter(&ipst->ips_mld_slowtimeout_lock);
4438         ipst->ips_mld_slowtimeout_quiesce = B_TRUE;
4439         mutex_exit(&ipst->ips_mld_slowtimeout_lock);
4440 
4441         ret = untimeout(ipst->ips_igmp_timeout_id);
4442         if (ret == -1) {
4443                 ASSERT(ipst->ips_igmp_timeout_id == 0);
4444         } else {
4445                 ASSERT(ipst->ips_igmp_timeout_id != 0);
4446                 ipst->ips_igmp_timeout_id = 0;
4447         }
4448         ret = untimeout(ipst->ips_igmp_slowtimeout_id);
4449         if (ret == -1) {
4450                 ASSERT(ipst->ips_igmp_slowtimeout_id == 0);
4451         } else {
4452                 ASSERT(ipst->ips_igmp_slowtimeout_id != 0);
4453                 ipst->ips_igmp_slowtimeout_id = 0;
4454         }
4455         ret = untimeout(ipst->ips_mld_timeout_id);
4456         if (ret == -1) {
4457                 ASSERT(ipst->ips_mld_timeout_id == 0);
4458         } else {
4459                 ASSERT(ipst->ips_mld_timeout_id != 0);
4460                 ipst->ips_mld_timeout_id = 0;
4461         }
4462         ret = untimeout(ipst->ips_mld_slowtimeout_id);
4463         if (ret == -1) {
4464                 ASSERT(ipst->ips_mld_slowtimeout_id == 0);
4465         } else {
4466                 ASSERT(ipst->ips_mld_slowtimeout_id != 0);
4467                 ipst->ips_mld_slowtimeout_id = 0;
4468         }
4469 
4470         ip_ire_fini(ipst);
4471         ip6_asp_free(ipst);
4472         conn_drain_fini(ipst);
4473         ipcl_destroy(ipst);
4474 
4475         mutex_destroy(&ipst->ips_ndp4->ndp_g_lock);
4476         mutex_destroy(&ipst->ips_ndp6->ndp_g_lock);
4477         kmem_free(ipst->ips_ndp4, sizeof (ndp_g_t));
4478         ipst->ips_ndp4 = NULL;
4479         kmem_free(ipst->ips_ndp6, sizeof (ndp_g_t));
4480         ipst->ips_ndp6 = NULL;
4481 
4482         if (ipst->ips_loopback_ksp != NULL) {
4483                 kstat_delete_netstack(ipst->ips_loopback_ksp, stackid);
4484                 ipst->ips_loopback_ksp = NULL;
4485         }
4486 
4487         mutex_destroy(&ipst->ips_capab_taskq_lock);
4488         cv_destroy(&ipst->ips_capab_taskq_cv);
4489 
4490         rw_destroy(&ipst->ips_srcid_lock);
4491 
4492         mutex_destroy(&ipst->ips_ip_mi_lock);
4493         rw_destroy(&ipst->ips_ill_g_usesrc_lock);
4494 
4495         mutex_destroy(&ipst->ips_igmp_timer_lock);
4496         mutex_destroy(&ipst->ips_mld_timer_lock);
4497         mutex_destroy(&ipst->ips_igmp_slowtimeout_lock);
4498         mutex_destroy(&ipst->ips_mld_slowtimeout_lock);
4499         mutex_destroy(&ipst->ips_ip_addr_avail_lock);
4500         rw_destroy(&ipst->ips_ill_g_lock);
4501 
4502         kmem_free(ipst->ips_phyint_g_list, sizeof (phyint_list_t));
4503         ipst->ips_phyint_g_list = NULL;
4504         kmem_free(ipst->ips_ill_g_heads, sizeof (ill_g_head_t) * MAX_G_HEADS);
4505         ipst->ips_ill_g_heads = NULL;
4506 
4507         ldi_ident_release(ipst->ips_ldi_ident);
4508         kmem_free(ipst, sizeof (*ipst));
4509 }
4510 
4511 /*
4512  * This function is called from the TSD destructor, and is used to debug
4513  * reference count issues in IP. See block comment in <inet/ip_if.h> for
4514  * details.
4515  */
4516 static void
4517 ip_thread_exit(void *phash)
4518 {
4519         th_hash_t *thh = phash;
4520 
4521         rw_enter(&ip_thread_rwlock, RW_WRITER);
4522         list_remove(&ip_thread_list, thh);
4523         rw_exit(&ip_thread_rwlock);
4524         mod_hash_destroy_hash(thh->thh_hash);
4525         kmem_free(thh, sizeof (*thh));
4526 }
4527 
4528 /*
4529  * Called when the IP kernel module is loaded into the kernel
4530  */
4531 void
4532 ip_ddi_init(void)
4533 {
4534         ip_squeue_flag = ip_squeue_switch(ip_squeue_enter);
4535 
4536         /*
4537          * For IP and TCP the minor numbers should start from 2 since we have 4
4538          * initial devices: ip, ip6, tcp, tcp6.
4539          */
4540         /*
4541          * If this is a 64-bit kernel, then create two separate arenas -
4542          * one for TLIs in the range of INET_MIN_DEV+2 through 2^^18-1, and the
4543          * other for socket apps in the range 2^^18 through 2^^32-1.
4544          */
4545         ip_minor_arena_la = NULL;
4546         ip_minor_arena_sa = NULL;
4547 #if defined(_LP64)
4548         if ((ip_minor_arena_sa = inet_minor_create("ip_minor_arena_sa",
4549             INET_MIN_DEV + 2, MAXMIN32, KM_SLEEP)) == NULL) {
4550                 cmn_err(CE_PANIC,
4551                     "ip_ddi_init: ip_minor_arena_sa creation failed\n");
4552         }
4553         if ((ip_minor_arena_la = inet_minor_create("ip_minor_arena_la",
4554             MAXMIN32 + 1, MAXMIN64, KM_SLEEP)) == NULL) {
4555                 cmn_err(CE_PANIC,
4556                     "ip_ddi_init: ip_minor_arena_la creation failed\n");
4557         }
4558 #else
4559         if ((ip_minor_arena_sa = inet_minor_create("ip_minor_arena_sa",
4560             INET_MIN_DEV + 2, MAXMIN, KM_SLEEP)) == NULL) {
4561                 cmn_err(CE_PANIC,
4562                     "ip_ddi_init: ip_minor_arena_sa creation failed\n");
4563         }
4564 #endif
4565         ip_poll_normal_ticks = MSEC_TO_TICK_ROUNDUP(ip_poll_normal_ms);
4566 
4567         ipcl_g_init();
4568         ip_ire_g_init();
4569         ip_net_g_init();
4570 
4571 #ifdef DEBUG
4572         tsd_create(&ip_thread_data, ip_thread_exit);
4573         rw_init(&ip_thread_rwlock, NULL, RW_DEFAULT, NULL);
4574         list_create(&ip_thread_list, sizeof (th_hash_t),
4575             offsetof(th_hash_t, thh_link));
4576 #endif
4577         ipsec_policy_g_init();
4578         tcp_ddi_g_init();
4579         sctp_ddi_g_init();
4580         dce_g_init();
4581 
4582         /*
4583          * We want to be informed each time a stack is created or
4584          * destroyed in the kernel, so we can maintain the
4585          * set of udp_stack_t's.
4586          */
4587         netstack_register(NS_IP, ip_stack_init, ip_stack_shutdown,
4588             ip_stack_fini);
4589 
4590         tnet_init();
4591 
4592         udp_ddi_g_init();
4593         rts_ddi_g_init();
4594         icmp_ddi_g_init();
4595         ilb_ddi_g_init();
4596 
4597         /* This needs to be called after all transports are initialized. */
4598         mutex_enter(&cpu_lock);
4599         register_cpu_setup_func(ip_tp_cpu_update, NULL);
4600         mutex_exit(&cpu_lock);
4601 }
4602 
4603 /*
4604  * Initialize the IP stack instance.
4605  */
4606 static void *
4607 ip_stack_init(netstackid_t stackid, netstack_t *ns)
4608 {
4609         ip_stack_t      *ipst;
4610         size_t          arrsz;
4611         major_t         major;
4612 
4613 #ifdef NS_DEBUG
4614         printf("ip_stack_init(stack %d)\n", stackid);
4615 #endif
4616 
4617         ipst = (ip_stack_t *)kmem_zalloc(sizeof (*ipst), KM_SLEEP);
4618         ipst->ips_netstack = ns;
4619 
4620         ipst->ips_ill_g_heads = kmem_zalloc(sizeof (ill_g_head_t) * MAX_G_HEADS,
4621             KM_SLEEP);
4622         ipst->ips_phyint_g_list = kmem_zalloc(sizeof (phyint_list_t),
4623             KM_SLEEP);
4624         ipst->ips_ndp4 = kmem_zalloc(sizeof (ndp_g_t), KM_SLEEP);
4625         ipst->ips_ndp6 = kmem_zalloc(sizeof (ndp_g_t), KM_SLEEP);
4626         mutex_init(&ipst->ips_ndp4->ndp_g_lock, NULL, MUTEX_DEFAULT, NULL);
4627         mutex_init(&ipst->ips_ndp6->ndp_g_lock, NULL, MUTEX_DEFAULT, NULL);
4628 
4629         mutex_init(&ipst->ips_igmp_timer_lock, NULL, MUTEX_DEFAULT, NULL);
4630         ipst->ips_igmp_deferred_next = INFINITY;
4631         mutex_init(&ipst->ips_mld_timer_lock, NULL, MUTEX_DEFAULT, NULL);
4632         ipst->ips_mld_deferred_next = INFINITY;
4633         mutex_init(&ipst->ips_igmp_slowtimeout_lock, NULL, MUTEX_DEFAULT, NULL);
4634         mutex_init(&ipst->ips_mld_slowtimeout_lock, NULL, MUTEX_DEFAULT, NULL);
4635         mutex_init(&ipst->ips_ip_mi_lock, NULL, MUTEX_DEFAULT, NULL);
4636         mutex_init(&ipst->ips_ip_addr_avail_lock, NULL, MUTEX_DEFAULT, NULL);
4637         rw_init(&ipst->ips_ill_g_lock, NULL, RW_DEFAULT, NULL);
4638         rw_init(&ipst->ips_ill_g_usesrc_lock, NULL, RW_DEFAULT, NULL);
4639 
4640         ipcl_init(ipst);
4641         ip_ire_init(ipst);
4642         ip6_asp_init(ipst);
4643         ipif_init(ipst);
4644         conn_drain_init(ipst);
4645         ip_mrouter_stack_init(ipst);
4646         dce_stack_init(ipst);
4647 
4648         ipst->ips_ip_multirt_log_interval = 1000;
4649 
4650         ipst->ips_ill_index = 1;
4651 
4652         ipst->ips_saved_ip_forwarding = -1;
4653         ipst->ips_reg_vif_num = ALL_VIFS;    /* Index to Register vif */
4654 
4655         arrsz = ip_propinfo_count * sizeof (mod_prop_info_t);
4656         ipst->ips_propinfo_tbl = (mod_prop_info_t *)kmem_alloc(arrsz, KM_SLEEP);
4657         bcopy(ip_propinfo_tbl, ipst->ips_propinfo_tbl, arrsz);
4658 
4659         ipst->ips_ip_mibkp = ip_kstat_init(stackid, ipst);
4660         ipst->ips_icmp_mibkp = icmp_kstat_init(stackid);
4661         ipst->ips_ip_kstat = ip_kstat2_init(stackid, &ipst->ips_ip_statistics);
4662         ipst->ips_ip6_kstat =
4663             ip6_kstat_init(stackid, &ipst->ips_ip6_statistics);
4664 
4665         ipst->ips_ip_src_id = 1;
4666         rw_init(&ipst->ips_srcid_lock, NULL, RW_DEFAULT, NULL);
4667 
4668         ipst->ips_src_generation = SRC_GENERATION_INITIAL;
4669 
4670         ip_net_init(ipst, ns);
4671         ipv4_hook_init(ipst);
4672         ipv6_hook_init(ipst);
4673         arp_hook_init(ipst);
4674         ipmp_init(ipst);
4675         ipobs_init(ipst);
4676 
4677         /*
4678          * Create the taskq dispatcher thread and initialize related stuff.
4679          */
4680         mutex_init(&ipst->ips_capab_taskq_lock, NULL, MUTEX_DEFAULT, NULL);
4681         cv_init(&ipst->ips_capab_taskq_cv, NULL, CV_DEFAULT, NULL);
4682         ipst->ips_capab_taskq_thread = thread_create(NULL, 0,
4683             ill_taskq_dispatch, ipst, 0, &p0, TS_RUN, minclsyspri);
4684 
4685         major = mod_name_to_major(INET_NAME);
4686         (void) ldi_ident_from_major(major, &ipst->ips_ldi_ident);
4687         return (ipst);
4688 }
4689 
4690 /*
4691  * Allocate and initialize a DLPI template of the specified length.  (May be
4692  * called as writer.)
4693  */
4694 mblk_t *
4695 ip_dlpi_alloc(size_t len, t_uscalar_t prim)
4696 {
4697         mblk_t  *mp;
4698 
4699         mp = allocb(len, BPRI_MED);
4700         if (!mp)
4701                 return (NULL);
4702 
4703         /*
4704          * DLPIv2 says that DL_INFO_REQ and DL_TOKEN_REQ (the latter
4705          * of which we don't seem to use) are sent with M_PCPROTO, and
4706          * that other DLPI are M_PROTO.
4707          */
4708         if (prim == DL_INFO_REQ) {
4709                 mp->b_datap->db_type = M_PCPROTO;
4710         } else {
4711                 mp->b_datap->db_type = M_PROTO;
4712         }
4713 
4714         mp->b_wptr = mp->b_rptr + len;
4715         bzero(mp->b_rptr, len);
4716         ((dl_unitdata_req_t *)mp->b_rptr)->dl_primitive = prim;
4717         return (mp);
4718 }
4719 
4720 /*
4721  * Allocate and initialize a DLPI notification.  (May be called as writer.)
4722  */
4723 mblk_t *
4724 ip_dlnotify_alloc(uint_t notification, uint_t data)
4725 {
4726         dl_notify_ind_t *notifyp;
4727         mblk_t          *mp;
4728 
4729         if ((mp = ip_dlpi_alloc(DL_NOTIFY_IND_SIZE, DL_NOTIFY_IND)) == NULL)
4730                 return (NULL);
4731 
4732         notifyp = (dl_notify_ind_t *)mp->b_rptr;
4733         notifyp->dl_notification = notification;
4734         notifyp->dl_data = data;
4735         return (mp);
4736 }
4737 
4738 mblk_t *
4739 ip_dlnotify_alloc2(uint_t notification, uint_t data1, uint_t data2)
4740 {
4741         dl_notify_ind_t *notifyp;
4742         mblk_t          *mp;
4743 
4744         if ((mp = ip_dlpi_alloc(DL_NOTIFY_IND_SIZE, DL_NOTIFY_IND)) == NULL)
4745                 return (NULL);
4746 
4747         notifyp = (dl_notify_ind_t *)mp->b_rptr;
4748         notifyp->dl_notification = notification;
4749         notifyp->dl_data1 = data1;
4750         notifyp->dl_data2 = data2;
4751         return (mp);
4752 }
4753 
4754 /*
4755  * Debug formatting routine.  Returns a character string representation of the
4756  * addr in buf, of the form xxx.xxx.xxx.xxx.  This routine takes the address
4757  * in the form of a ipaddr_t and calls ip_dot_saddr with a pointer.
4758  *
4759  * Once the ndd table-printing interfaces are removed, this can be changed to
4760  * standard dotted-decimal form.
4761  */
4762 char *
4763 ip_dot_addr(ipaddr_t addr, char *buf)
4764 {
4765         uint8_t *ap = (uint8_t *)&addr;
4766 
4767         (void) mi_sprintf(buf, "%03d.%03d.%03d.%03d",
4768             ap[0] & 0xFF, ap[1] & 0xFF, ap[2] & 0xFF, ap[3] & 0xFF);
4769         return (buf);
4770 }
4771 
4772 /*
4773  * Write the given MAC address as a printable string in the usual colon-
4774  * separated format.
4775  */
4776 const char *
4777 mac_colon_addr(const uint8_t *addr, size_t alen, char *buf, size_t buflen)
4778 {
4779         char *bp;
4780 
4781         if (alen == 0 || buflen < 4)
4782                 return ("?");
4783         bp = buf;
4784         for (;;) {
4785                 /*
4786                  * If there are more MAC address bytes available, but we won't
4787                  * have any room to print them, then add "..." to the string
4788                  * instead.  See below for the 'magic number' explanation.
4789                  */
4790                 if ((alen == 2 && buflen < 6) || (alen > 2 && buflen < 7)) {
4791                         (void) strcpy(bp, "...");
4792                         break;
4793                 }
4794                 (void) sprintf(bp, "%02x", *addr++);
4795                 bp += 2;
4796                 if (--alen == 0)
4797                         break;
4798                 *bp++ = ':';
4799                 buflen -= 3;
4800                 /*
4801                  * At this point, based on the first 'if' statement above,
4802                  * either alen == 1 and buflen >= 3, or alen > 1 and
4803                  * buflen >= 4.  The first case leaves room for the final "xx"
4804                  * number and trailing NUL byte.  The second leaves room for at
4805                  * least "...".  Thus the apparently 'magic' numbers chosen for
4806                  * that statement.
4807                  */
4808         }
4809         return (buf);
4810 }
4811 
4812 /*
4813  * Called when it is conceptually a ULP that would sent the packet
4814  * e.g., port unreachable and protocol unreachable. Check that the packet
4815  * would have passed the IPsec global policy before sending the error.
4816  *
4817  * Send an ICMP error after patching up the packet appropriately.
4818  * Uses ip_drop_input and bumps the appropriate MIB.
4819  */
4820 void
4821 ip_fanout_send_icmp_v4(mblk_t *mp, uint_t icmp_type, uint_t icmp_code,
4822     ip_recv_attr_t *ira)
4823 {
4824         ipha_t          *ipha;
4825         boolean_t       secure;
4826         ill_t           *ill = ira->ira_ill;
4827         ip_stack_t      *ipst = ill->ill_ipst;
4828         netstack_t      *ns = ipst->ips_netstack;
4829         ipsec_stack_t   *ipss = ns->netstack_ipsec;
4830 
4831         secure = ira->ira_flags & IRAF_IPSEC_SECURE;
4832 
4833         /*
4834          * We are generating an icmp error for some inbound packet.
4835          * Called from all ip_fanout_(udp, tcp, proto) functions.
4836          * Before we generate an error, check with global policy
4837          * to see whether this is allowed to enter the system. As
4838          * there is no "conn", we are checking with global policy.
4839          */
4840         ipha = (ipha_t *)mp->b_rptr;
4841         if (secure || ipss->ipsec_inbound_v4_policy_present) {
4842                 mp = ipsec_check_global_policy(mp, NULL, ipha, NULL, ira, ns);
4843                 if (mp == NULL)
4844                         return;
4845         }
4846 
4847         /* We never send errors for protocols that we do implement */
4848         if (ira->ira_protocol == IPPROTO_ICMP ||
4849             ira->ira_protocol == IPPROTO_IGMP) {
4850                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
4851                 ip_drop_input("ip_fanout_send_icmp_v4", mp, ill);
4852                 freemsg(mp);
4853                 return;
4854         }
4855         /*
4856          * Have to correct checksum since
4857          * the packet might have been
4858          * fragmented and the reassembly code in ip_rput
4859          * does not restore the IP checksum.
4860          */
4861         ipha->ipha_hdr_checksum = 0;
4862         ipha->ipha_hdr_checksum = ip_csum_hdr(ipha);
4863 
4864         switch (icmp_type) {
4865         case ICMP_DEST_UNREACHABLE:
4866                 switch (icmp_code) {
4867                 case ICMP_PROTOCOL_UNREACHABLE:
4868                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInUnknownProtos);
4869                         ip_drop_input("ipIfStatsInUnknownProtos", mp, ill);
4870                         break;
4871                 case ICMP_PORT_UNREACHABLE:
4872                         BUMP_MIB(ill->ill_ip_mib, udpIfStatsNoPorts);
4873                         ip_drop_input("ipIfStatsNoPorts", mp, ill);
4874                         break;
4875                 }
4876 
4877                 icmp_unreachable(mp, icmp_code, ira);
4878                 break;
4879         default:
4880 #ifdef DEBUG
4881                 panic("ip_fanout_send_icmp_v4: wrong type");
4882                 /*NOTREACHED*/
4883 #else
4884                 freemsg(mp);
4885                 break;
4886 #endif
4887         }
4888 }
4889 
4890 /*
4891  * Used to send an ICMP error message when a packet is received for
4892  * a protocol that is not supported. The mblk passed as argument
4893  * is consumed by this function.
4894  */
4895 void
4896 ip_proto_not_sup(mblk_t *mp, ip_recv_attr_t *ira)
4897 {
4898         ipha_t          *ipha;
4899 
4900         ipha = (ipha_t *)mp->b_rptr;
4901         if (ira->ira_flags & IRAF_IS_IPV4) {
4902                 ASSERT(IPH_HDR_VERSION(ipha) == IP_VERSION);
4903                 ip_fanout_send_icmp_v4(mp, ICMP_DEST_UNREACHABLE,
4904                     ICMP_PROTOCOL_UNREACHABLE, ira);
4905         } else {
4906                 ASSERT(IPH_HDR_VERSION(ipha) == IPV6_VERSION);
4907                 ip_fanout_send_icmp_v6(mp, ICMP6_PARAM_PROB,
4908                     ICMP6_PARAMPROB_NEXTHEADER, ira);
4909         }
4910 }
4911 
4912 /*
4913  * Deliver a rawip packet to the given conn, possibly applying ipsec policy.
4914  * Handles IPv4 and IPv6.
4915  * We are responsible for disposing of mp, such as by freemsg() or putnext()
4916  * Caller is responsible for dropping references to the conn.
4917  */
4918 void
4919 ip_fanout_proto_conn(conn_t *connp, mblk_t *mp, ipha_t *ipha, ip6_t *ip6h,
4920     ip_recv_attr_t *ira)
4921 {
4922         ill_t           *ill = ira->ira_ill;
4923         ip_stack_t      *ipst = ill->ill_ipst;
4924         ipsec_stack_t   *ipss = ipst->ips_netstack->netstack_ipsec;
4925         boolean_t       secure;
4926         uint_t          protocol = ira->ira_protocol;
4927         iaflags_t       iraflags = ira->ira_flags;
4928         queue_t         *rq;
4929 
4930         secure = iraflags & IRAF_IPSEC_SECURE;
4931 
4932         rq = connp->conn_rq;
4933         if (IPCL_IS_NONSTR(connp) ? connp->conn_flow_cntrld : !canputnext(rq)) {
4934                 switch (protocol) {
4935                 case IPPROTO_ICMPV6:
4936                         BUMP_MIB(ill->ill_icmp6_mib, ipv6IfIcmpInOverflows);
4937                         break;
4938                 case IPPROTO_ICMP:
4939                         BUMP_MIB(&ipst->ips_icmp_mib, icmpInOverflows);
4940                         break;
4941                 default:
4942                         BUMP_MIB(ill->ill_ip_mib, rawipIfStatsInOverflows);
4943                         break;
4944                 }
4945                 freemsg(mp);
4946                 return;
4947         }
4948 
4949         ASSERT(!(IPCL_IS_IPTUN(connp)));
4950 
4951         if (((iraflags & IRAF_IS_IPV4) ?
4952             CONN_INBOUND_POLICY_PRESENT(connp, ipss) :
4953             CONN_INBOUND_POLICY_PRESENT_V6(connp, ipss)) ||
4954             secure) {
4955                 mp = ipsec_check_inbound_policy(mp, connp, ipha,
4956                     ip6h, ira);
4957                 if (mp == NULL) {
4958                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
4959                         /* Note that mp is NULL */
4960                         ip_drop_input("ipIfStatsInDiscards", mp, ill);
4961                         return;
4962                 }
4963         }
4964 
4965         if (iraflags & IRAF_ICMP_ERROR) {
4966                 (connp->conn_recvicmp)(connp, mp, NULL, ira);
4967         } else {
4968                 ill_t *rill = ira->ira_rill;
4969 
4970                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCInDelivers);
4971                 ira->ira_ill = ira->ira_rill = NULL;
4972                 /* Send it upstream */
4973                 (connp->conn_recv)(connp, mp, NULL, ira);
4974                 ira->ira_ill = ill;
4975                 ira->ira_rill = rill;
4976         }
4977 }
4978 
4979 /*
4980  * Handle protocols with which IP is less intimate.  There
4981  * can be more than one stream bound to a particular
4982  * protocol.  When this is the case, normally each one gets a copy
4983  * of any incoming packets.
4984  *
4985  * IPsec NOTE :
4986  *
4987  * Don't allow a secure packet going up a non-secure connection.
4988  * We don't allow this because
4989  *
4990  * 1) Reply might go out in clear which will be dropped at
4991  *    the sending side.
4992  * 2) If the reply goes out in clear it will give the
4993  *    adversary enough information for getting the key in
4994  *    most of the cases.
4995  *
4996  * Moreover getting a secure packet when we expect clear
4997  * implies that SA's were added without checking for
4998  * policy on both ends. This should not happen once ISAKMP
4999  * is used to negotiate SAs as SAs will be added only after
5000  * verifying the policy.
5001  *
5002  * Zones notes:
5003  * Earlier in ip_input on a system with multiple shared-IP zones we
5004  * duplicate the multicast and broadcast packets and send them up
5005  * with each explicit zoneid that exists on that ill.
5006  * This means that here we can match the zoneid with SO_ALLZONES being special.
5007  */
5008 void
5009 ip_fanout_proto_v4(mblk_t *mp, ipha_t *ipha, ip_recv_attr_t *ira)
5010 {
5011         mblk_t          *mp1;
5012         ipaddr_t        laddr;
5013         conn_t          *connp, *first_connp, *next_connp;
5014         connf_t         *connfp;
5015         ill_t           *ill = ira->ira_ill;
5016         ip_stack_t      *ipst = ill->ill_ipst;
5017 
5018         laddr = ipha->ipha_dst;
5019 
5020         connfp = &ipst->ips_ipcl_proto_fanout_v4[ira->ira_protocol];
5021         mutex_enter(&connfp->connf_lock);
5022         connp = connfp->connf_head;
5023         for (connp = connfp->connf_head; connp != NULL;
5024             connp = connp->conn_next) {
5025                 /* Note: IPCL_PROTO_MATCH includes conn_wantpacket */
5026                 if (IPCL_PROTO_MATCH(connp, ira, ipha) &&
5027                     (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5028                     tsol_receive_local(mp, &laddr, IPV4_VERSION, ira, connp))) {
5029                         break;
5030                 }
5031         }
5032 
5033         if (connp == NULL) {
5034                 /*
5035                  * No one bound to these addresses.  Is
5036                  * there a client that wants all
5037                  * unclaimed datagrams?
5038                  */
5039                 mutex_exit(&connfp->connf_lock);
5040                 ip_fanout_send_icmp_v4(mp, ICMP_DEST_UNREACHABLE,
5041                     ICMP_PROTOCOL_UNREACHABLE, ira);
5042                 return;
5043         }
5044 
5045         ASSERT(IPCL_IS_NONSTR(connp) || connp->conn_rq != NULL);
5046 
5047         CONN_INC_REF(connp);
5048         first_connp = connp;
5049         connp = connp->conn_next;
5050 
5051         for (;;) {
5052                 while (connp != NULL) {
5053                         /* Note: IPCL_PROTO_MATCH includes conn_wantpacket */
5054                         if (IPCL_PROTO_MATCH(connp, ira, ipha) &&
5055                             (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5056                             tsol_receive_local(mp, &laddr, IPV4_VERSION,
5057                             ira, connp)))
5058                                 break;
5059                         connp = connp->conn_next;
5060                 }
5061 
5062                 if (connp == NULL) {
5063                         /* No more interested clients */
5064                         connp = first_connp;
5065                         break;
5066                 }
5067                 if (((mp1 = dupmsg(mp)) == NULL) &&
5068                     ((mp1 = copymsg(mp)) == NULL)) {
5069                         /* Memory allocation failed */
5070                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
5071                         ip_drop_input("ipIfStatsInDiscards", mp, ill);
5072                         connp = first_connp;
5073                         break;
5074                 }
5075 
5076                 CONN_INC_REF(connp);
5077                 mutex_exit(&connfp->connf_lock);
5078 
5079                 ip_fanout_proto_conn(connp, mp1, (ipha_t *)mp1->b_rptr, NULL,
5080                     ira);
5081 
5082                 mutex_enter(&connfp->connf_lock);
5083                 /* Follow the next pointer before releasing the conn. */
5084                 next_connp = connp->conn_next;
5085                 CONN_DEC_REF(connp);
5086                 connp = next_connp;
5087         }
5088 
5089         /* Last one.  Send it upstream. */
5090         mutex_exit(&connfp->connf_lock);
5091 
5092         ip_fanout_proto_conn(connp, mp, ipha, NULL, ira);
5093 
5094         CONN_DEC_REF(connp);
5095 }
5096 
5097 /*
5098  * If we have a IPsec NAT-Traversal packet, strip the zero-SPI or
5099  * pass it along to ESP if the SPI is non-zero.  Returns the mblk if the mblk
5100  * is not consumed.
5101  *
5102  * One of three things can happen, all of which affect the passed-in mblk:
5103  *
5104  * 1.) The packet is stock UDP and gets its zero-SPI stripped.  Return mblk..
5105  *
5106  * 2.) The packet is ESP-in-UDP, gets transformed into an equivalent
5107  *     ESP packet, and is passed along to ESP for consumption.  Return NULL.
5108  *
5109  * 3.) The packet is an ESP-in-UDP Keepalive.  Drop it and return NULL.
5110  */
5111 mblk_t *
5112 zero_spi_check(mblk_t *mp, ip_recv_attr_t *ira)
5113 {
5114         int shift, plen, iph_len;
5115         ipha_t *ipha;
5116         udpha_t *udpha;
5117         uint32_t *spi;
5118         uint32_t esp_ports;
5119         uint8_t *orptr;
5120         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
5121         ipsec_stack_t   *ipss = ipst->ips_netstack->netstack_ipsec;
5122 
5123         ipha = (ipha_t *)mp->b_rptr;
5124         iph_len = ira->ira_ip_hdr_length;
5125         plen = ira->ira_pktlen;
5126 
5127         if (plen - iph_len - sizeof (udpha_t) < sizeof (uint32_t)) {
5128                 /*
5129                  * Most likely a keepalive for the benefit of an intervening
5130                  * NAT.  These aren't for us, per se, so drop it.
5131                  *
5132                  * RFC 3947/8 doesn't say for sure what to do for 2-3
5133                  * byte packets (keepalives are 1-byte), but we'll drop them
5134                  * also.
5135                  */
5136                 ip_drop_packet(mp, B_TRUE, ira->ira_ill,
5137                     DROPPER(ipss, ipds_esp_nat_t_ka), &ipss->ipsec_dropper);
5138                 return (NULL);
5139         }
5140 
5141         if (MBLKL(mp) < iph_len + sizeof (udpha_t) + sizeof (*spi)) {
5142                 /* might as well pull it all up - it might be ESP. */
5143                 if (!pullupmsg(mp, -1)) {
5144                         ip_drop_packet(mp, B_TRUE, ira->ira_ill,
5145                             DROPPER(ipss, ipds_esp_nomem),
5146                             &ipss->ipsec_dropper);
5147                         return (NULL);
5148                 }
5149 
5150                 ipha = (ipha_t *)mp->b_rptr;
5151         }
5152         spi = (uint32_t *)(mp->b_rptr + iph_len + sizeof (udpha_t));
5153         if (*spi == 0) {
5154                 /* UDP packet - remove 0-spi. */
5155                 shift = sizeof (uint32_t);
5156         } else {
5157                 /* ESP-in-UDP packet - reduce to ESP. */
5158                 ipha->ipha_protocol = IPPROTO_ESP;
5159                 shift = sizeof (udpha_t);
5160         }
5161 
5162         /* Fix IP header */
5163         ira->ira_pktlen = (plen - shift);
5164         ipha->ipha_length = htons(ira->ira_pktlen);
5165         ipha->ipha_hdr_checksum = 0;
5166 
5167         orptr = mp->b_rptr;
5168         mp->b_rptr += shift;
5169 
5170         udpha = (udpha_t *)(orptr + iph_len);
5171         if (*spi == 0) {
5172                 ASSERT((uint8_t *)ipha == orptr);
5173                 udpha->uha_length = htons(plen - shift - iph_len);
5174                 iph_len += sizeof (udpha_t);    /* For the call to ovbcopy(). */
5175                 esp_ports = 0;
5176         } else {
5177                 esp_ports = *((uint32_t *)udpha);
5178                 ASSERT(esp_ports != 0);
5179         }
5180         ovbcopy(orptr, orptr + shift, iph_len);
5181         if (esp_ports != 0) /* Punt up for ESP processing. */ {
5182                 ipha = (ipha_t *)(orptr + shift);
5183 
5184                 ira->ira_flags |= IRAF_ESP_UDP_PORTS;
5185                 ira->ira_esp_udp_ports = esp_ports;
5186                 ip_fanout_v4(mp, ipha, ira);
5187                 return (NULL);
5188         }
5189         return (mp);
5190 }
5191 
5192 /*
5193  * Deliver a udp packet to the given conn, possibly applying ipsec policy.
5194  * Handles IPv4 and IPv6.
5195  * We are responsible for disposing of mp, such as by freemsg() or putnext()
5196  * Caller is responsible for dropping references to the conn.
5197  */
5198 void
5199 ip_fanout_udp_conn(conn_t *connp, mblk_t *mp, ipha_t *ipha, ip6_t *ip6h,
5200     ip_recv_attr_t *ira)
5201 {
5202         ill_t           *ill = ira->ira_ill;
5203         ip_stack_t      *ipst = ill->ill_ipst;
5204         ipsec_stack_t   *ipss = ipst->ips_netstack->netstack_ipsec;
5205         boolean_t       secure;
5206         iaflags_t       iraflags = ira->ira_flags;
5207 
5208         secure = iraflags & IRAF_IPSEC_SECURE;
5209 
5210         if (IPCL_IS_NONSTR(connp) ? connp->conn_flow_cntrld :
5211             !canputnext(connp->conn_rq)) {
5212                 BUMP_MIB(ill->ill_ip_mib, udpIfStatsInOverflows);
5213                 freemsg(mp);
5214                 return;
5215         }
5216 
5217         if (((iraflags & IRAF_IS_IPV4) ?
5218             CONN_INBOUND_POLICY_PRESENT(connp, ipss) :
5219             CONN_INBOUND_POLICY_PRESENT_V6(connp, ipss)) ||
5220             secure) {
5221                 mp = ipsec_check_inbound_policy(mp, connp, ipha,
5222                     ip6h, ira);
5223                 if (mp == NULL) {
5224                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
5225                         /* Note that mp is NULL */
5226                         ip_drop_input("ipIfStatsInDiscards", mp, ill);
5227                         return;
5228                 }
5229         }
5230 
5231         /*
5232          * Since this code is not used for UDP unicast we don't need a NAT_T
5233          * check. Only ip_fanout_v4 has that check.
5234          */
5235         if (ira->ira_flags & IRAF_ICMP_ERROR) {
5236                 (connp->conn_recvicmp)(connp, mp, NULL, ira);
5237         } else {
5238                 ill_t *rill = ira->ira_rill;
5239 
5240                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCInDelivers);
5241                 ira->ira_ill = ira->ira_rill = NULL;
5242                 /* Send it upstream */
5243                 (connp->conn_recv)(connp, mp, NULL, ira);
5244                 ira->ira_ill = ill;
5245                 ira->ira_rill = rill;
5246         }
5247 }
5248 
5249 /*
5250  * Fanout for UDP packets that are multicast or broadcast, and ICMP errors.
5251  * (Unicast fanout is handled in ip_input_v4.)
5252  *
5253  * If SO_REUSEADDR is set all multicast and broadcast packets
5254  * will be delivered to all conns bound to the same port.
5255  *
5256  * If there is at least one matching AF_INET receiver, then we will
5257  * ignore any AF_INET6 receivers.
5258  * In the special case where an AF_INET socket binds to 0.0.0.0/<port> and an
5259  * AF_INET6 socket binds to ::/<port>, only the AF_INET socket receives the IPv4
5260  * packets.
5261  *
5262  * Zones notes:
5263  * Earlier in ip_input on a system with multiple shared-IP zones we
5264  * duplicate the multicast and broadcast packets and send them up
5265  * with each explicit zoneid that exists on that ill.
5266  * This means that here we can match the zoneid with SO_ALLZONES being special.
5267  */
5268 void
5269 ip_fanout_udp_multi_v4(mblk_t *mp, ipha_t *ipha, uint16_t lport, uint16_t fport,
5270     ip_recv_attr_t *ira)
5271 {
5272         ipaddr_t        laddr;
5273         in6_addr_t      v6faddr;
5274         conn_t          *connp;
5275         connf_t         *connfp;
5276         ipaddr_t        faddr;
5277         ill_t           *ill = ira->ira_ill;
5278         ip_stack_t      *ipst = ill->ill_ipst;
5279 
5280         ASSERT(ira->ira_flags & (IRAF_MULTIBROADCAST|IRAF_ICMP_ERROR));
5281 
5282         laddr = ipha->ipha_dst;
5283         faddr = ipha->ipha_src;
5284 
5285         connfp = &ipst->ips_ipcl_udp_fanout[IPCL_UDP_HASH(lport, ipst)];
5286         mutex_enter(&connfp->connf_lock);
5287         connp = connfp->connf_head;
5288 
5289         /*
5290          * If SO_REUSEADDR has been set on the first we send the
5291          * packet to all clients that have joined the group and
5292          * match the port.
5293          */
5294         while (connp != NULL) {
5295                 if ((IPCL_UDP_MATCH(connp, lport, laddr, fport, faddr)) &&
5296                     conn_wantpacket(connp, ira, ipha) &&
5297                     (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5298                     tsol_receive_local(mp, &laddr, IPV4_VERSION, ira, connp)))
5299                         break;
5300                 connp = connp->conn_next;
5301         }
5302 
5303         if (connp == NULL)
5304                 goto notfound;
5305 
5306         CONN_INC_REF(connp);
5307 
5308         if (connp->conn_reuseaddr) {
5309                 conn_t          *first_connp = connp;
5310                 conn_t          *next_connp;
5311                 mblk_t          *mp1;
5312 
5313                 connp = connp->conn_next;
5314                 for (;;) {
5315                         while (connp != NULL) {
5316                                 if (IPCL_UDP_MATCH(connp, lport, laddr,
5317                                     fport, faddr) &&
5318                                     conn_wantpacket(connp, ira, ipha) &&
5319                                     (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5320                                     tsol_receive_local(mp, &laddr, IPV4_VERSION,
5321                                     ira, connp)))
5322                                         break;
5323                                 connp = connp->conn_next;
5324                         }
5325                         if (connp == NULL) {
5326                                 /* No more interested clients */
5327                                 connp = first_connp;
5328                                 break;
5329                         }
5330                         if (((mp1 = dupmsg(mp)) == NULL) &&
5331                             ((mp1 = copymsg(mp)) == NULL)) {
5332                                 /* Memory allocation failed */
5333                                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
5334                                 ip_drop_input("ipIfStatsInDiscards", mp, ill);
5335                                 connp = first_connp;
5336                                 break;
5337                         }
5338                         CONN_INC_REF(connp);
5339                         mutex_exit(&connfp->connf_lock);
5340 
5341                         IP_STAT(ipst, ip_udp_fanmb);
5342                         ip_fanout_udp_conn(connp, mp1, (ipha_t *)mp1->b_rptr,
5343                             NULL, ira);
5344                         mutex_enter(&connfp->connf_lock);
5345                         /* Follow the next pointer before releasing the conn */
5346                         next_connp = connp->conn_next;
5347                         CONN_DEC_REF(connp);
5348                         connp = next_connp;
5349                 }
5350         }
5351 
5352         /* Last one.  Send it upstream. */
5353         mutex_exit(&connfp->connf_lock);
5354         IP_STAT(ipst, ip_udp_fanmb);
5355         ip_fanout_udp_conn(connp, mp, ipha, NULL, ira);
5356         CONN_DEC_REF(connp);
5357         return;
5358 
5359 notfound:
5360         mutex_exit(&connfp->connf_lock);
5361         /*
5362          * IPv6 endpoints bound to multicast IPv4-mapped addresses
5363          * have already been matched above, since they live in the IPv4
5364          * fanout tables. This implies we only need to
5365          * check for IPv6 in6addr_any endpoints here.
5366          * Thus we compare using ipv6_all_zeros instead of the destination
5367          * address, except for the multicast group membership lookup which
5368          * uses the IPv4 destination.
5369          */
5370         IN6_IPADDR_TO_V4MAPPED(ipha->ipha_src, &v6faddr);
5371         connfp = &ipst->ips_ipcl_udp_fanout[IPCL_UDP_HASH(lport, ipst)];
5372         mutex_enter(&connfp->connf_lock);
5373         connp = connfp->connf_head;
5374         /*
5375          * IPv4 multicast packet being delivered to an AF_INET6
5376          * in6addr_any endpoint.
5377          * Need to check conn_wantpacket(). Note that we use conn_wantpacket()
5378          * and not conn_wantpacket_v6() since any multicast membership is
5379          * for an IPv4-mapped multicast address.
5380          */
5381         while (connp != NULL) {
5382                 if (IPCL_UDP_MATCH_V6(connp, lport, ipv6_all_zeros,
5383                     fport, v6faddr) &&
5384                     conn_wantpacket(connp, ira, ipha) &&
5385                     (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5386                     tsol_receive_local(mp, &laddr, IPV4_VERSION, ira, connp)))
5387                         break;
5388                 connp = connp->conn_next;
5389         }
5390 
5391         if (connp == NULL) {
5392                 /*
5393                  * No one bound to this port.  Is
5394                  * there a client that wants all
5395                  * unclaimed datagrams?
5396                  */
5397                 mutex_exit(&connfp->connf_lock);
5398 
5399                 if (ipst->ips_ipcl_proto_fanout_v4[IPPROTO_UDP].connf_head !=
5400                     NULL) {
5401                         ASSERT(ira->ira_protocol == IPPROTO_UDP);
5402                         ip_fanout_proto_v4(mp, ipha, ira);
5403                 } else {
5404                         /*
5405                          * We used to attempt to send an icmp error here, but
5406                          * since this is known to be a multicast packet
5407                          * and we don't send icmp errors in response to
5408                          * multicast, just drop the packet and give up sooner.
5409                          */
5410                         BUMP_MIB(ill->ill_ip_mib, udpIfStatsNoPorts);
5411                         freemsg(mp);
5412                 }
5413                 return;
5414         }
5415         CONN_INC_REF(connp);
5416         ASSERT(IPCL_IS_NONSTR(connp) || connp->conn_rq != NULL);
5417 
5418         /*
5419          * If SO_REUSEADDR has been set on the first we send the
5420          * packet to all clients that have joined the group and
5421          * match the port.
5422          */
5423         if (connp->conn_reuseaddr) {
5424                 conn_t          *first_connp = connp;
5425                 conn_t          *next_connp;
5426                 mblk_t          *mp1;
5427 
5428                 connp = connp->conn_next;
5429                 for (;;) {
5430                         while (connp != NULL) {
5431                                 if (IPCL_UDP_MATCH_V6(connp, lport,
5432                                     ipv6_all_zeros, fport, v6faddr) &&
5433                                     conn_wantpacket(connp, ira, ipha) &&
5434                                     (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5435                                     tsol_receive_local(mp, &laddr, IPV4_VERSION,
5436                                     ira, connp)))
5437                                         break;
5438                                 connp = connp->conn_next;
5439                         }
5440                         if (connp == NULL) {
5441                                 /* No more interested clients */
5442                                 connp = first_connp;
5443                                 break;
5444                         }
5445                         if (((mp1 = dupmsg(mp)) == NULL) &&
5446                             ((mp1 = copymsg(mp)) == NULL)) {
5447                                 /* Memory allocation failed */
5448                                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
5449                                 ip_drop_input("ipIfStatsInDiscards", mp, ill);
5450                                 connp = first_connp;
5451                                 break;
5452                         }
5453                         CONN_INC_REF(connp);
5454                         mutex_exit(&connfp->connf_lock);
5455 
5456                         IP_STAT(ipst, ip_udp_fanmb);
5457                         ip_fanout_udp_conn(connp, mp1, (ipha_t *)mp1->b_rptr,
5458                             NULL, ira);
5459                         mutex_enter(&connfp->connf_lock);
5460                         /* Follow the next pointer before releasing the conn */
5461                         next_connp = connp->conn_next;
5462                         CONN_DEC_REF(connp);
5463                         connp = next_connp;
5464                 }
5465         }
5466 
5467         /* Last one.  Send it upstream. */
5468         mutex_exit(&connfp->connf_lock);
5469         IP_STAT(ipst, ip_udp_fanmb);
5470         ip_fanout_udp_conn(connp, mp, ipha, NULL, ira);
5471         CONN_DEC_REF(connp);
5472 }
5473 
5474 /*
5475  * Split an incoming packet's IPv4 options into the label and the other options.
5476  * If 'allocate' is set it does memory allocation for the ip_pkt_t, including
5477  * clearing out any leftover label or options.
5478  * Otherwise it just makes ipp point into the packet.
5479  *
5480  * Returns zero if ok; ENOMEM if the buffer couldn't be allocated.
5481  */
5482 int
5483 ip_find_hdr_v4(ipha_t *ipha, ip_pkt_t *ipp, boolean_t allocate)
5484 {
5485         uchar_t         *opt;
5486         uint32_t        totallen;
5487         uint32_t        optval;
5488         uint32_t        optlen;
5489 
5490         ipp->ipp_fields |= IPPF_HOPLIMIT | IPPF_TCLASS | IPPF_ADDR;
5491         ipp->ipp_hoplimit = ipha->ipha_ttl;
5492         ipp->ipp_type_of_service = ipha->ipha_type_of_service;
5493         IN6_IPADDR_TO_V4MAPPED(ipha->ipha_dst, &ipp->ipp_addr);
5494 
5495         /*
5496          * Get length (in 4 byte octets) of IP header options.
5497          */
5498         totallen = ipha->ipha_version_and_hdr_length -
5499             (uint8_t)((IP_VERSION << 4) + IP_SIMPLE_HDR_LENGTH_IN_WORDS);
5500 
5501         if (totallen == 0) {
5502                 if (!allocate)
5503                         return (0);
5504 
5505                 /* Clear out anything from a previous packet */
5506                 if (ipp->ipp_fields & IPPF_IPV4_OPTIONS) {
5507                         kmem_free(ipp->ipp_ipv4_options,
5508                             ipp->ipp_ipv4_options_len);
5509                         ipp->ipp_ipv4_options = NULL;
5510                         ipp->ipp_ipv4_options_len = 0;
5511                         ipp->ipp_fields &= ~IPPF_IPV4_OPTIONS;
5512                 }
5513                 if (ipp->ipp_fields & IPPF_LABEL_V4) {
5514                         kmem_free(ipp->ipp_label_v4, ipp->ipp_label_len_v4);
5515                         ipp->ipp_label_v4 = NULL;
5516                         ipp->ipp_label_len_v4 = 0;
5517                         ipp->ipp_fields &= ~IPPF_LABEL_V4;
5518                 }
5519                 return (0);
5520         }
5521 
5522         totallen <<= 2;
5523         opt = (uchar_t *)&ipha[1];
5524         if (!is_system_labeled()) {
5525 
5526         copyall:
5527                 if (!allocate) {
5528                         if (totallen != 0) {
5529                                 ipp->ipp_ipv4_options = opt;
5530                                 ipp->ipp_ipv4_options_len = totallen;
5531                                 ipp->ipp_fields |= IPPF_IPV4_OPTIONS;
5532                         }
5533                         return (0);
5534                 }
5535                 /* Just copy all of options */
5536                 if (ipp->ipp_fields & IPPF_IPV4_OPTIONS) {
5537                         if (totallen == ipp->ipp_ipv4_options_len) {
5538                                 bcopy(opt, ipp->ipp_ipv4_options, totallen);
5539                                 return (0);
5540                         }
5541                         kmem_free(ipp->ipp_ipv4_options,
5542                             ipp->ipp_ipv4_options_len);
5543                         ipp->ipp_ipv4_options = NULL;
5544                         ipp->ipp_ipv4_options_len = 0;
5545                         ipp->ipp_fields &= ~IPPF_IPV4_OPTIONS;
5546                 }
5547                 if (totallen == 0)
5548                         return (0);
5549 
5550                 ipp->ipp_ipv4_options = kmem_alloc(totallen, KM_NOSLEEP);
5551                 if (ipp->ipp_ipv4_options == NULL)
5552                         return (ENOMEM);
5553                 ipp->ipp_ipv4_options_len = totallen;
5554                 ipp->ipp_fields |= IPPF_IPV4_OPTIONS;
5555                 bcopy(opt, ipp->ipp_ipv4_options, totallen);
5556                 return (0);
5557         }
5558 
5559         if (allocate && (ipp->ipp_fields & IPPF_LABEL_V4)) {
5560                 kmem_free(ipp->ipp_label_v4, ipp->ipp_label_len_v4);
5561                 ipp->ipp_label_v4 = NULL;
5562                 ipp->ipp_label_len_v4 = 0;
5563                 ipp->ipp_fields &= ~IPPF_LABEL_V4;
5564         }
5565 
5566         /*
5567          * Search for CIPSO option.
5568          * We assume CIPSO is first in options if it is present.
5569          * If it isn't, then ipp_opt_ipv4_options will not include the options
5570          * prior to the CIPSO option.
5571          */
5572         while (totallen != 0) {
5573                 switch (optval = opt[IPOPT_OPTVAL]) {
5574                 case IPOPT_EOL:
5575                         return (0);
5576                 case IPOPT_NOP:
5577                         optlen = 1;
5578                         break;
5579                 default:
5580                         if (totallen <= IPOPT_OLEN)
5581                                 return (EINVAL);
5582                         optlen = opt[IPOPT_OLEN];
5583                         if (optlen < 2)
5584                                 return (EINVAL);
5585                 }
5586                 if (optlen > totallen)
5587                         return (EINVAL);
5588 
5589                 switch (optval) {
5590                 case IPOPT_COMSEC:
5591                         if (!allocate) {
5592                                 ipp->ipp_label_v4 = opt;
5593                                 ipp->ipp_label_len_v4 = optlen;
5594                                 ipp->ipp_fields |= IPPF_LABEL_V4;
5595                         } else {
5596                                 ipp->ipp_label_v4 = kmem_alloc(optlen,
5597                                     KM_NOSLEEP);
5598                                 if (ipp->ipp_label_v4 == NULL)
5599                                         return (ENOMEM);
5600                                 ipp->ipp_label_len_v4 = optlen;
5601                                 ipp->ipp_fields |= IPPF_LABEL_V4;
5602                                 bcopy(opt, ipp->ipp_label_v4, optlen);
5603                         }
5604                         totallen -= optlen;
5605                         opt += optlen;
5606 
5607                         /* Skip padding bytes until we get to a multiple of 4 */
5608                         while ((totallen & 3) != 0 && opt[0] == IPOPT_NOP) {
5609                                 totallen--;
5610                                 opt++;
5611                         }
5612                         /* Remaining as ipp_ipv4_options */
5613                         goto copyall;
5614                 }
5615                 totallen -= optlen;
5616                 opt += optlen;
5617         }
5618         /* No CIPSO found; return everything as ipp_ipv4_options */
5619         totallen = ipha->ipha_version_and_hdr_length -
5620             (uint8_t)((IP_VERSION << 4) + IP_SIMPLE_HDR_LENGTH_IN_WORDS);
5621         totallen <<= 2;
5622         opt = (uchar_t *)&ipha[1];
5623         goto copyall;
5624 }
5625 
5626 /*
5627  * Efficient versions of lookup for an IRE when we only
5628  * match the address.
5629  * For RTF_REJECT or BLACKHOLE we return IRE_NOROUTE.
5630  * Does not handle multicast addresses.
5631  */
5632 uint_t
5633 ip_type_v4(ipaddr_t addr, ip_stack_t *ipst)
5634 {
5635         ire_t *ire;
5636         uint_t result;
5637 
5638         ire = ire_ftable_lookup_simple_v4(addr, 0, ipst, NULL);
5639         ASSERT(ire != NULL);
5640         if (ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE))
5641                 result = IRE_NOROUTE;
5642         else
5643                 result = ire->ire_type;
5644         ire_refrele(ire);
5645         return (result);
5646 }
5647 
5648 /*
5649  * Efficient versions of lookup for an IRE when we only
5650  * match the address.
5651  * For RTF_REJECT or BLACKHOLE we return IRE_NOROUTE.
5652  * Does not handle multicast addresses.
5653  */
5654 uint_t
5655 ip_type_v6(const in6_addr_t *addr, ip_stack_t *ipst)
5656 {
5657         ire_t *ire;
5658         uint_t result;
5659 
5660         ire = ire_ftable_lookup_simple_v6(addr, 0, ipst, NULL);
5661         ASSERT(ire != NULL);
5662         if (ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE))
5663                 result = IRE_NOROUTE;
5664         else
5665                 result = ire->ire_type;
5666         ire_refrele(ire);
5667         return (result);
5668 }
5669 
5670 /*
5671  * Nobody should be sending
5672  * packets up this stream
5673  */
5674 static int
5675 ip_lrput(queue_t *q, mblk_t *mp)
5676 {
5677         switch (mp->b_datap->db_type) {
5678         case M_FLUSH:
5679                 /* Turn around */
5680                 if (*mp->b_rptr & FLUSHW) {
5681                         *mp->b_rptr &= ~FLUSHR;
5682                         qreply(q, mp);
5683                         return (0);
5684                 }
5685                 break;
5686         }
5687         freemsg(mp);
5688         return (0);
5689 }
5690 
5691 /* Nobody should be sending packets down this stream */
5692 /* ARGSUSED */
5693 int
5694 ip_lwput(queue_t *q, mblk_t *mp)
5695 {
5696         freemsg(mp);
5697         return (0);
5698 }
5699 
5700 /*
5701  * Move the first hop in any source route to ipha_dst and remove that part of
5702  * the source route.  Called by other protocols.  Errors in option formatting
5703  * are ignored - will be handled by ip_output_options. Return the final
5704  * destination (either ipha_dst or the last entry in a source route.)
5705  */
5706 ipaddr_t
5707 ip_massage_options(ipha_t *ipha, netstack_t *ns)
5708 {
5709         ipoptp_t        opts;
5710         uchar_t         *opt;
5711         uint8_t         optval;
5712         uint8_t         optlen;
5713         ipaddr_t        dst;
5714         int             i;
5715         ip_stack_t      *ipst = ns->netstack_ip;
5716 
5717         ip2dbg(("ip_massage_options\n"));
5718         dst = ipha->ipha_dst;
5719         for (optval = ipoptp_first(&opts, ipha);
5720             optval != IPOPT_EOL;
5721             optval = ipoptp_next(&opts)) {
5722                 opt = opts.ipoptp_cur;
5723                 switch (optval) {
5724                         uint8_t off;
5725                 case IPOPT_SSRR:
5726                 case IPOPT_LSRR:
5727                         if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
5728                                 ip1dbg(("ip_massage_options: bad src route\n"));
5729                                 break;
5730                         }
5731                         optlen = opts.ipoptp_len;
5732                         off = opt[IPOPT_OFFSET];
5733                         off--;
5734                 redo_srr:
5735                         if (optlen < IP_ADDR_LEN ||
5736                             off > optlen - IP_ADDR_LEN) {
5737                                 /* End of source route */
5738                                 ip1dbg(("ip_massage_options: end of SR\n"));
5739                                 break;
5740                         }
5741                         bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
5742                         ip1dbg(("ip_massage_options: next hop 0x%x\n",
5743                             ntohl(dst)));
5744                         /*
5745                          * Check if our address is present more than
5746                          * once as consecutive hops in source route.
5747                          * XXX verify per-interface ip_forwarding
5748                          * for source route?
5749                          */
5750                         if (ip_type_v4(dst, ipst) == IRE_LOCAL) {
5751                                 off += IP_ADDR_LEN;
5752                                 goto redo_srr;
5753                         }
5754                         if (dst == htonl(INADDR_LOOPBACK)) {
5755                                 ip1dbg(("ip_massage_options: loopback addr in "
5756                                     "source route!\n"));
5757                                 break;
5758                         }
5759                         /*
5760                          * Update ipha_dst to be the first hop and remove the
5761                          * first hop from the source route (by overwriting
5762                          * part of the option with NOP options).
5763                          */
5764                         ipha->ipha_dst = dst;
5765                         /* Put the last entry in dst */
5766                         off = ((optlen - IP_ADDR_LEN - 3) & ~(IP_ADDR_LEN-1)) +
5767                             3;
5768                         bcopy(&opt[off], &dst, IP_ADDR_LEN);
5769 
5770                         ip1dbg(("ip_massage_options: last hop 0x%x\n",
5771                             ntohl(dst)));
5772                         /* Move down and overwrite */
5773                         opt[IP_ADDR_LEN] = opt[0];
5774                         opt[IP_ADDR_LEN+1] = opt[IPOPT_OLEN] - IP_ADDR_LEN;
5775                         opt[IP_ADDR_LEN+2] = opt[IPOPT_OFFSET];
5776                         for (i = 0; i < IP_ADDR_LEN; i++)
5777                                 opt[i] = IPOPT_NOP;
5778                         break;
5779                 }
5780         }
5781         return (dst);
5782 }
5783 
5784 /*
5785  * Return the network mask
5786  * associated with the specified address.
5787  */
5788 ipaddr_t
5789 ip_net_mask(ipaddr_t addr)
5790 {
5791         uchar_t *up = (uchar_t *)&addr;
5792         ipaddr_t mask = 0;
5793         uchar_t *maskp = (uchar_t *)&mask;
5794 
5795 #if defined(__i386) || defined(__amd64)
5796 #define TOTALLY_BRAIN_DAMAGED_C_COMPILER
5797 #endif
5798 #ifdef  TOTALLY_BRAIN_DAMAGED_C_COMPILER
5799         maskp[0] = maskp[1] = maskp[2] = maskp[3] = 0;
5800 #endif
5801         if (CLASSD(addr)) {
5802                 maskp[0] = 0xF0;
5803                 return (mask);
5804         }
5805 
5806         /* We assume Class E default netmask to be 32 */
5807         if (CLASSE(addr))
5808                 return (0xffffffffU);
5809 
5810         if (addr == 0)
5811                 return (0);
5812         maskp[0] = 0xFF;
5813         if ((up[0] & 0x80) == 0)
5814                 return (mask);
5815 
5816         maskp[1] = 0xFF;
5817         if ((up[0] & 0xC0) == 0x80)
5818                 return (mask);
5819 
5820         maskp[2] = 0xFF;
5821         if ((up[0] & 0xE0) == 0xC0)
5822                 return (mask);
5823 
5824         /* Otherwise return no mask */
5825         return ((ipaddr_t)0);
5826 }
5827 
5828 /* Name/Value Table Lookup Routine */
5829 char *
5830 ip_nv_lookup(nv_t *nv, int value)
5831 {
5832         if (!nv)
5833                 return (NULL);
5834         for (; nv->nv_name; nv++) {
5835                 if (nv->nv_value == value)
5836                         return (nv->nv_name);
5837         }
5838         return ("unknown");
5839 }
5840 
5841 static int
5842 ip_wait_for_info_ack(ill_t *ill)
5843 {
5844         int err;
5845 
5846         mutex_enter(&ill->ill_lock);
5847         while (ill->ill_state_flags & ILL_LL_SUBNET_PENDING) {
5848                 /*
5849                  * Return value of 0 indicates a pending signal.
5850                  */
5851                 err = cv_wait_sig(&ill->ill_cv, &ill->ill_lock);
5852                 if (err == 0) {
5853                         mutex_exit(&ill->ill_lock);
5854                         return (EINTR);
5855                 }
5856         }
5857         mutex_exit(&ill->ill_lock);
5858         /*
5859          * ip_rput_other could have set an error  in ill_error on
5860          * receipt of M_ERROR.
5861          */
5862         return (ill->ill_error);
5863 }
5864 
5865 /*
5866  * This is a module open, i.e. this is a control stream for access
5867  * to a DLPI device.  We allocate an ill_t as the instance data in
5868  * this case.
5869  */
5870 static int
5871 ip_modopen(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
5872 {
5873         ill_t   *ill;
5874         int     err;
5875         zoneid_t zoneid;
5876         netstack_t *ns;
5877         ip_stack_t *ipst;
5878 
5879         /*
5880          * Prevent unprivileged processes from pushing IP so that
5881          * they can't send raw IP.
5882          */
5883         if (secpolicy_net_rawaccess(credp) != 0)
5884                 return (EPERM);
5885 
5886         ns = netstack_find_by_cred(credp);
5887         ASSERT(ns != NULL);
5888         ipst = ns->netstack_ip;
5889         ASSERT(ipst != NULL);
5890 
5891         /*
5892          * For exclusive stacks we set the zoneid to zero
5893          * to make IP operate as if in the global zone.
5894          */
5895         if (ipst->ips_netstack->netstack_stackid != GLOBAL_NETSTACKID)
5896                 zoneid = GLOBAL_ZONEID;
5897         else
5898                 zoneid = crgetzoneid(credp);
5899 
5900         ill = (ill_t *)mi_open_alloc_sleep(sizeof (ill_t));
5901         q->q_ptr = WR(q)->q_ptr = ill;
5902         ill->ill_ipst = ipst;
5903         ill->ill_zoneid = zoneid;
5904 
5905         /*
5906          * ill_init initializes the ill fields and then sends down
5907          * down a DL_INFO_REQ after calling qprocson.
5908          */
5909         err = ill_init(q, ill);
5910 
5911         if (err != 0) {
5912                 mi_free(ill);
5913                 netstack_rele(ipst->ips_netstack);
5914                 q->q_ptr = NULL;
5915                 WR(q)->q_ptr = NULL;
5916                 return (err);
5917         }
5918 
5919         /*
5920          * Wait for the DL_INFO_ACK if a DL_INFO_REQ was sent.
5921          *
5922          * ill_init initializes the ipsq marking this thread as
5923          * writer
5924          */
5925         ipsq_exit(ill->ill_phyint->phyint_ipsq);
5926         err = ip_wait_for_info_ack(ill);
5927         if (err == 0)
5928                 ill->ill_credp = credp;
5929         else
5930                 goto fail;
5931 
5932         crhold(credp);
5933 
5934         mutex_enter(&ipst->ips_ip_mi_lock);
5935         err = mi_open_link(&ipst->ips_ip_g_head, (IDP)q->q_ptr, devp, flag,
5936             sflag, credp);
5937         mutex_exit(&ipst->ips_ip_mi_lock);
5938 fail:
5939         if (err) {
5940                 (void) ip_close(q, 0, credp);
5941                 return (err);
5942         }
5943         return (0);
5944 }
5945 
5946 /* For /dev/ip aka AF_INET open */
5947 int
5948 ip_openv4(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
5949 {
5950         return (ip_open(q, devp, flag, sflag, credp, B_FALSE));
5951 }
5952 
5953 /* For /dev/ip6 aka AF_INET6 open */
5954 int
5955 ip_openv6(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
5956 {
5957         return (ip_open(q, devp, flag, sflag, credp, B_TRUE));
5958 }
5959 
5960 /* IP open routine. */
5961 int
5962 ip_open(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp,
5963     boolean_t isv6)
5964 {
5965         conn_t          *connp;
5966         major_t         maj;
5967         zoneid_t        zoneid;
5968         netstack_t      *ns;
5969         ip_stack_t      *ipst;
5970 
5971         /* Allow reopen. */
5972         if (q->q_ptr != NULL)
5973                 return (0);
5974 
5975         if (sflag & MODOPEN) {
5976                 /* This is a module open */
5977                 return (ip_modopen(q, devp, flag, sflag, credp));
5978         }
5979 
5980         if ((flag & ~(FKLYR)) == IP_HELPER_STR) {
5981                 /*
5982                  * Non streams based socket looking for a stream
5983                  * to access IP
5984                  */
5985                 return (ip_helper_stream_setup(q, devp, flag, sflag,
5986                     credp, isv6));
5987         }
5988 
5989         ns = netstack_find_by_cred(credp);
5990         ASSERT(ns != NULL);
5991         ipst = ns->netstack_ip;
5992         ASSERT(ipst != NULL);
5993 
5994         /*
5995          * For exclusive stacks we set the zoneid to zero
5996          * to make IP operate as if in the global zone.
5997          */
5998         if (ipst->ips_netstack->netstack_stackid != GLOBAL_NETSTACKID)
5999                 zoneid = GLOBAL_ZONEID;
6000         else
6001                 zoneid = crgetzoneid(credp);
6002 
6003         /*
6004          * We are opening as a device. This is an IP client stream, and we
6005          * allocate an conn_t as the instance data.
6006          */
6007         connp = ipcl_conn_create(IPCL_IPCCONN, KM_SLEEP, ipst->ips_netstack);
6008 
6009         /*
6010          * ipcl_conn_create did a netstack_hold. Undo the hold that was
6011          * done by netstack_find_by_cred()
6012          */
6013         netstack_rele(ipst->ips_netstack);
6014 
6015         connp->conn_ixa->ixa_flags |= IXAF_MULTICAST_LOOP | IXAF_SET_ULP_CKSUM;
6016         /* conn_allzones can not be set this early, hence no IPCL_ZONEID */
6017         connp->conn_ixa->ixa_zoneid = zoneid;
6018         connp->conn_zoneid = zoneid;
6019 
6020         connp->conn_rq = q;
6021         q->q_ptr = WR(q)->q_ptr = connp;
6022 
6023         /* Minor tells us which /dev entry was opened */
6024         if (isv6) {
6025                 connp->conn_family = AF_INET6;
6026                 connp->conn_ipversion = IPV6_VERSION;
6027                 connp->conn_ixa->ixa_flags &= ~IXAF_IS_IPV4;
6028                 connp->conn_ixa->ixa_src_preferences = IPV6_PREFER_SRC_DEFAULT;
6029         } else {
6030                 connp->conn_family = AF_INET;
6031                 connp->conn_ipversion = IPV4_VERSION;
6032                 connp->conn_ixa->ixa_flags |= IXAF_IS_IPV4;
6033         }
6034 
6035         if ((ip_minor_arena_la != NULL) && (flag & SO_SOCKSTR) &&
6036             ((connp->conn_dev = inet_minor_alloc(ip_minor_arena_la)) != 0)) {
6037                 connp->conn_minor_arena = ip_minor_arena_la;
6038         } else {
6039                 /*
6040                  * Either minor numbers in the large arena were exhausted
6041                  * or a non socket application is doing the open.
6042                  * Try to allocate from the small arena.
6043                  */
6044                 if ((connp->conn_dev =
6045                     inet_minor_alloc(ip_minor_arena_sa)) == 0) {
6046                         /* CONN_DEC_REF takes care of netstack_rele() */
6047                         q->q_ptr = WR(q)->q_ptr = NULL;
6048                         CONN_DEC_REF(connp);
6049                         return (EBUSY);
6050                 }
6051                 connp->conn_minor_arena = ip_minor_arena_sa;
6052         }
6053 
6054         maj = getemajor(*devp);
6055         *devp = makedevice(maj, (minor_t)connp->conn_dev);
6056 
6057         /*
6058          * connp->conn_cred is crfree()ed in ipcl_conn_destroy()
6059          */
6060         connp->conn_cred = credp;
6061         connp->conn_cpid = curproc->p_pid;
6062         /* Cache things in ixa without an extra refhold */
6063         ASSERT(!(connp->conn_ixa->ixa_free_flags & IXA_FREE_CRED));
6064         connp->conn_ixa->ixa_cred = connp->conn_cred;
6065         connp->conn_ixa->ixa_cpid = connp->conn_cpid;
6066         if (is_system_labeled())
6067                 connp->conn_ixa->ixa_tsl = crgetlabel(connp->conn_cred);
6068 
6069         /*
6070          * Handle IP_IOC_RTS_REQUEST and other ioctls which use conn_recv
6071          */
6072         connp->conn_recv = ip_conn_input;
6073         connp->conn_recvicmp = ip_conn_input_icmp;
6074 
6075         crhold(connp->conn_cred);
6076 
6077         /*
6078          * If the caller has the process-wide flag set, then default to MAC
6079          * exempt mode.  This allows read-down to unlabeled hosts.
6080          */
6081         if (getpflags(NET_MAC_AWARE, credp) != 0)
6082                 connp->conn_mac_mode = CONN_MAC_AWARE;
6083 
6084         connp->conn_zone_is_global = (crgetzoneid(credp) == GLOBAL_ZONEID);
6085 
6086         connp->conn_rq = q;
6087         connp->conn_wq = WR(q);
6088 
6089         /* Non-zero default values */
6090         connp->conn_ixa->ixa_flags |= IXAF_MULTICAST_LOOP;
6091 
6092         /*
6093          * Make the conn globally visible to walkers
6094          */
6095         ASSERT(connp->conn_ref == 1);
6096         mutex_enter(&connp->conn_lock);
6097         connp->conn_state_flags &= ~CONN_INCIPIENT;
6098         mutex_exit(&connp->conn_lock);
6099 
6100         qprocson(q);
6101 
6102         return (0);
6103 }
6104 
6105 /*
6106  * Set IPsec policy from an ipsec_req_t. If the req is not "zero" and valid,
6107  * all of them are copied to the conn_t. If the req is "zero", the policy is
6108  * zeroed out. A "zero" policy has zero ipsr_{ah,req,self_encap}_req
6109  * fields.
6110  * We keep only the latest setting of the policy and thus policy setting
6111  * is not incremental/cumulative.
6112  *
6113  * Requests to set policies with multiple alternative actions will
6114  * go through a different API.
6115  */
6116 int
6117 ipsec_set_req(cred_t *cr, conn_t *connp, ipsec_req_t *req)
6118 {
6119         uint_t ah_req = 0;
6120         uint_t esp_req = 0;
6121         uint_t se_req = 0;
6122         ipsec_act_t *actp = NULL;
6123         uint_t nact;
6124         ipsec_policy_head_t *ph;
6125         boolean_t is_pol_reset, is_pol_inserted = B_FALSE;
6126         int error = 0;
6127         netstack_t      *ns = connp->conn_netstack;
6128         ip_stack_t      *ipst = ns->netstack_ip;
6129         ipsec_stack_t   *ipss = ns->netstack_ipsec;
6130 
6131 #define REQ_MASK (IPSEC_PREF_REQUIRED|IPSEC_PREF_NEVER)
6132 
6133         /*
6134          * The IP_SEC_OPT option does not allow variable length parameters,
6135          * hence a request cannot be NULL.
6136          */
6137         if (req == NULL)
6138                 return (EINVAL);
6139 
6140         ah_req = req->ipsr_ah_req;
6141         esp_req = req->ipsr_esp_req;
6142         se_req = req->ipsr_self_encap_req;
6143 
6144         /* Don't allow setting self-encap without one or more of AH/ESP. */
6145         if (se_req != 0 && esp_req == 0 && ah_req == 0)
6146                 return (EINVAL);
6147 
6148         /*
6149          * Are we dealing with a request to reset the policy (i.e.
6150          * zero requests).
6151          */
6152         is_pol_reset = ((ah_req & REQ_MASK) == 0 &&
6153             (esp_req & REQ_MASK) == 0 &&
6154             (se_req & REQ_MASK) == 0);
6155 
6156         if (!is_pol_reset) {
6157                 /*
6158                  * If we couldn't load IPsec, fail with "protocol
6159                  * not supported".
6160                  * IPsec may not have been loaded for a request with zero
6161                  * policies, so we don't fail in this case.
6162                  */
6163                 mutex_enter(&ipss->ipsec_loader_lock);
6164                 if (ipss->ipsec_loader_state != IPSEC_LOADER_SUCCEEDED) {
6165                         mutex_exit(&ipss->ipsec_loader_lock);
6166                         return (EPROTONOSUPPORT);
6167                 }
6168                 mutex_exit(&ipss->ipsec_loader_lock);
6169 
6170                 /*
6171                  * Test for valid requests. Invalid algorithms
6172                  * need to be tested by IPsec code because new
6173                  * algorithms can be added dynamically.
6174                  */
6175                 if ((ah_req & ~(REQ_MASK|IPSEC_PREF_UNIQUE)) != 0 ||
6176                     (esp_req & ~(REQ_MASK|IPSEC_PREF_UNIQUE)) != 0 ||
6177                     (se_req & ~(REQ_MASK|IPSEC_PREF_UNIQUE)) != 0) {
6178                         return (EINVAL);
6179                 }
6180 
6181                 /*
6182                  * Only privileged users can issue these
6183                  * requests.
6184                  */
6185                 if (((ah_req & IPSEC_PREF_NEVER) ||
6186                     (esp_req & IPSEC_PREF_NEVER) ||
6187                     (se_req & IPSEC_PREF_NEVER)) &&
6188                     secpolicy_ip_config(cr, B_FALSE) != 0) {
6189                         return (EPERM);
6190                 }
6191 
6192                 /*
6193                  * The IPSEC_PREF_REQUIRED and IPSEC_PREF_NEVER
6194                  * are mutually exclusive.
6195                  */
6196                 if (((ah_req & REQ_MASK) == REQ_MASK) ||
6197                     ((esp_req & REQ_MASK) == REQ_MASK) ||
6198                     ((se_req & REQ_MASK) == REQ_MASK)) {
6199                         /* Both of them are set */
6200                         return (EINVAL);
6201                 }
6202         }
6203 
6204         ASSERT(MUTEX_HELD(&connp->conn_lock));
6205 
6206         /*
6207          * If we have already cached policies in conn_connect(), don't
6208          * let them change now. We cache policies for connections
6209          * whose src,dst [addr, port] is known.
6210          */
6211         if (connp->conn_policy_cached) {
6212                 return (EINVAL);
6213         }
6214 
6215         /*
6216          * We have a zero policies, reset the connection policy if already
6217          * set. This will cause the connection to inherit the
6218          * global policy, if any.
6219          */
6220         if (is_pol_reset) {
6221                 if (connp->conn_policy != NULL) {
6222                         IPPH_REFRELE(connp->conn_policy, ipst->ips_netstack);
6223                         connp->conn_policy = NULL;
6224                 }
6225                 connp->conn_in_enforce_policy = B_FALSE;
6226                 connp->conn_out_enforce_policy = B_FALSE;
6227                 return (0);
6228         }
6229 
6230         ph = connp->conn_policy = ipsec_polhead_split(connp->conn_policy,
6231             ipst->ips_netstack);
6232         if (ph == NULL)
6233                 goto enomem;
6234 
6235         ipsec_actvec_from_req(req, &actp, &nact, ipst->ips_netstack);
6236         if (actp == NULL)
6237                 goto enomem;
6238 
6239         /*
6240          * Always insert IPv4 policy entries, since they can also apply to
6241          * ipv6 sockets being used in ipv4-compat mode.
6242          */
6243         if (!ipsec_polhead_insert(ph, actp, nact, IPSEC_AF_V4,
6244             IPSEC_TYPE_INBOUND, ns))
6245                 goto enomem;
6246         is_pol_inserted = B_TRUE;
6247         if (!ipsec_polhead_insert(ph, actp, nact, IPSEC_AF_V4,
6248             IPSEC_TYPE_OUTBOUND, ns))
6249                 goto enomem;
6250 
6251         /*
6252          * We're looking at a v6 socket, also insert the v6-specific
6253          * entries.
6254          */
6255         if (connp->conn_family == AF_INET6) {
6256                 if (!ipsec_polhead_insert(ph, actp, nact, IPSEC_AF_V6,
6257                     IPSEC_TYPE_INBOUND, ns))
6258                         goto enomem;
6259                 if (!ipsec_polhead_insert(ph, actp, nact, IPSEC_AF_V6,
6260                     IPSEC_TYPE_OUTBOUND, ns))
6261                         goto enomem;
6262         }
6263 
6264         ipsec_actvec_free(actp, nact);
6265 
6266         /*
6267          * If the requests need security, set enforce_policy.
6268          * If the requests are IPSEC_PREF_NEVER, one should
6269          * still set conn_out_enforce_policy so that ip_set_destination
6270          * marks the ip_xmit_attr_t appropriatly. This is needed so that
6271          * for connections that we don't cache policy in at connect time,
6272          * if global policy matches in ip_output_attach_policy, we
6273          * don't wrongly inherit global policy. Similarly, we need
6274          * to set conn_in_enforce_policy also so that we don't verify
6275          * policy wrongly.
6276          */
6277         if ((ah_req & REQ_MASK) != 0 ||
6278             (esp_req & REQ_MASK) != 0 ||
6279             (se_req & REQ_MASK) != 0) {
6280                 connp->conn_in_enforce_policy = B_TRUE;
6281                 connp->conn_out_enforce_policy = B_TRUE;
6282         }
6283 
6284         return (error);
6285 #undef REQ_MASK
6286 
6287         /*
6288          * Common memory-allocation-failure exit path.
6289          */
6290 enomem:
6291         if (actp != NULL)
6292                 ipsec_actvec_free(actp, nact);
6293         if (is_pol_inserted)
6294                 ipsec_polhead_flush(ph, ns);
6295         return (ENOMEM);
6296 }
6297 
6298 /*
6299  * Set socket options for joining and leaving multicast groups.
6300  * Common to IPv4 and IPv6; inet6 indicates the type of socket.
6301  * The caller has already check that the option name is consistent with
6302  * the address family of the socket.
6303  */
6304 int
6305 ip_opt_set_multicast_group(conn_t *connp, t_scalar_t name,
6306     uchar_t *invalp, boolean_t inet6, boolean_t checkonly)
6307 {
6308         int             *i1 = (int *)invalp;
6309         int             error = 0;
6310         ip_stack_t      *ipst = connp->conn_netstack->netstack_ip;
6311         struct ip_mreq  *v4_mreqp;
6312         struct ipv6_mreq *v6_mreqp;
6313         struct group_req *greqp;
6314         ire_t *ire;
6315         boolean_t done = B_FALSE;
6316         ipaddr_t ifaddr;
6317         in6_addr_t v6group;
6318         uint_t ifindex;
6319         boolean_t mcast_opt = B_TRUE;
6320         mcast_record_t fmode;
6321         int (*optfn)(conn_t *, boolean_t, const in6_addr_t *,
6322             ipaddr_t, uint_t, mcast_record_t, const in6_addr_t *);
6323 
6324         switch (name) {
6325         case IP_ADD_MEMBERSHIP:
6326         case IPV6_JOIN_GROUP:
6327                 mcast_opt = B_FALSE;
6328                 /* FALLTHROUGH */
6329         case MCAST_JOIN_GROUP:
6330                 fmode = MODE_IS_EXCLUDE;
6331                 optfn = ip_opt_add_group;
6332                 break;
6333 
6334         case IP_DROP_MEMBERSHIP:
6335         case IPV6_LEAVE_GROUP:
6336                 mcast_opt = B_FALSE;
6337                 /* FALLTHROUGH */
6338         case MCAST_LEAVE_GROUP:
6339                 fmode = MODE_IS_INCLUDE;
6340                 optfn = ip_opt_delete_group;
6341                 break;
6342         default:
6343                 ASSERT(0);
6344         }
6345 
6346         if (mcast_opt) {
6347                 struct sockaddr_in *sin;
6348                 struct sockaddr_in6 *sin6;
6349 
6350                 greqp = (struct group_req *)i1;
6351                 if (greqp->gr_group.ss_family == AF_INET) {
6352                         sin = (struct sockaddr_in *)&(greqp->gr_group);
6353                         IN6_INADDR_TO_V4MAPPED(&sin->sin_addr, &v6group);
6354                 } else {
6355                         if (!inet6)
6356                                 return (EINVAL);        /* Not on INET socket */
6357 
6358                         sin6 = (struct sockaddr_in6 *)&(greqp->gr_group);
6359                         v6group = sin6->sin6_addr;
6360                 }
6361                 ifaddr = INADDR_ANY;
6362                 ifindex = greqp->gr_interface;
6363         } else if (inet6) {
6364                 v6_mreqp = (struct ipv6_mreq *)i1;
6365                 v6group = v6_mreqp->ipv6mr_multiaddr;
6366                 ifaddr = INADDR_ANY;
6367                 ifindex = v6_mreqp->ipv6mr_interface;
6368         } else {
6369                 v4_mreqp = (struct ip_mreq *)i1;
6370                 IN6_INADDR_TO_V4MAPPED(&v4_mreqp->imr_multiaddr, &v6group);
6371                 ifaddr = (ipaddr_t)v4_mreqp->imr_interface.s_addr;
6372                 ifindex = 0;
6373         }
6374 
6375         /*
6376          * In the multirouting case, we need to replicate
6377          * the request on all interfaces that will take part
6378          * in replication.  We do so because multirouting is
6379          * reflective, thus we will probably receive multi-
6380          * casts on those interfaces.
6381          * The ip_multirt_apply_membership() succeeds if
6382          * the operation succeeds on at least one interface.
6383          */
6384         if (IN6_IS_ADDR_V4MAPPED(&v6group)) {
6385                 ipaddr_t group;
6386 
6387                 IN6_V4MAPPED_TO_IPADDR(&v6group, group);
6388 
6389                 ire = ire_ftable_lookup_v4(group, IP_HOST_MASK, 0,
6390                     IRE_HOST | IRE_INTERFACE, NULL, ALL_ZONES, NULL,
6391                     MATCH_IRE_MASK | MATCH_IRE_TYPE, 0, ipst, NULL);
6392         } else {
6393                 ire = ire_ftable_lookup_v6(&v6group, &ipv6_all_ones, 0,
6394                     IRE_HOST | IRE_INTERFACE, NULL, ALL_ZONES, NULL,
6395                     MATCH_IRE_MASK | MATCH_IRE_TYPE, 0, ipst, NULL);
6396         }
6397         if (ire != NULL) {
6398                 if (ire->ire_flags & RTF_MULTIRT) {
6399                         error = ip_multirt_apply_membership(optfn, ire, connp,
6400                             checkonly, &v6group, fmode, &ipv6_all_zeros);
6401                         done = B_TRUE;
6402                 }
6403                 ire_refrele(ire);
6404         }
6405 
6406         if (!done) {
6407                 error = optfn(connp, checkonly, &v6group, ifaddr, ifindex,
6408                     fmode, &ipv6_all_zeros);
6409         }
6410         return (error);
6411 }
6412 
6413 /*
6414  * Set socket options for joining and leaving multicast groups
6415  * for specific sources.
6416  * Common to IPv4 and IPv6; inet6 indicates the type of socket.
6417  * The caller has already check that the option name is consistent with
6418  * the address family of the socket.
6419  */
6420 int
6421 ip_opt_set_multicast_sources(conn_t *connp, t_scalar_t name,
6422     uchar_t *invalp, boolean_t inet6, boolean_t checkonly)
6423 {
6424         int             *i1 = (int *)invalp;
6425         int             error = 0;
6426         ip_stack_t      *ipst = connp->conn_netstack->netstack_ip;
6427         struct ip_mreq_source *imreqp;
6428         struct group_source_req *gsreqp;
6429         in6_addr_t v6group, v6src;
6430         uint32_t ifindex;
6431         ipaddr_t ifaddr;
6432         boolean_t mcast_opt = B_TRUE;
6433         mcast_record_t fmode;
6434         ire_t *ire;
6435         boolean_t done = B_FALSE;
6436         int (*optfn)(conn_t *, boolean_t, const in6_addr_t *,
6437             ipaddr_t, uint_t, mcast_record_t, const in6_addr_t *);
6438 
6439         switch (name) {
6440         case IP_BLOCK_SOURCE:
6441                 mcast_opt = B_FALSE;
6442                 /* FALLTHROUGH */
6443         case MCAST_BLOCK_SOURCE:
6444                 fmode = MODE_IS_EXCLUDE;
6445                 optfn = ip_opt_add_group;
6446                 break;
6447 
6448         case IP_UNBLOCK_SOURCE:
6449                 mcast_opt = B_FALSE;
6450                 /* FALLTHROUGH */
6451         case MCAST_UNBLOCK_SOURCE:
6452                 fmode = MODE_IS_EXCLUDE;
6453                 optfn = ip_opt_delete_group;
6454                 break;
6455 
6456         case IP_ADD_SOURCE_MEMBERSHIP:
6457                 mcast_opt = B_FALSE;
6458                 /* FALLTHROUGH */
6459         case MCAST_JOIN_SOURCE_GROUP:
6460                 fmode = MODE_IS_INCLUDE;
6461                 optfn = ip_opt_add_group;
6462                 break;
6463 
6464         case IP_DROP_SOURCE_MEMBERSHIP:
6465                 mcast_opt = B_FALSE;
6466                 /* FALLTHROUGH */
6467         case MCAST_LEAVE_SOURCE_GROUP:
6468                 fmode = MODE_IS_INCLUDE;
6469                 optfn = ip_opt_delete_group;
6470                 break;
6471         default:
6472                 ASSERT(0);
6473         }
6474 
6475         if (mcast_opt) {
6476                 gsreqp = (struct group_source_req *)i1;
6477                 ifindex = gsreqp->gsr_interface;
6478                 if (gsreqp->gsr_group.ss_family == AF_INET) {
6479                         struct sockaddr_in *s;
6480                         s = (struct sockaddr_in *)&gsreqp->gsr_group;
6481                         IN6_INADDR_TO_V4MAPPED(&s->sin_addr, &v6group);
6482                         s = (struct sockaddr_in *)&gsreqp->gsr_source;
6483                         IN6_INADDR_TO_V4MAPPED(&s->sin_addr, &v6src);
6484                 } else {
6485                         struct sockaddr_in6 *s6;
6486 
6487                         if (!inet6)
6488                                 return (EINVAL);        /* Not on INET socket */
6489 
6490                         s6 = (struct sockaddr_in6 *)&gsreqp->gsr_group;
6491                         v6group = s6->sin6_addr;
6492                         s6 = (struct sockaddr_in6 *)&gsreqp->gsr_source;
6493                         v6src = s6->sin6_addr;
6494                 }
6495                 ifaddr = INADDR_ANY;
6496         } else {
6497                 imreqp = (struct ip_mreq_source *)i1;
6498                 IN6_INADDR_TO_V4MAPPED(&imreqp->imr_multiaddr, &v6group);
6499                 IN6_INADDR_TO_V4MAPPED(&imreqp->imr_sourceaddr, &v6src);
6500                 ifaddr = (ipaddr_t)imreqp->imr_interface.s_addr;
6501                 ifindex = 0;
6502         }
6503 
6504         /*
6505          * Handle src being mapped INADDR_ANY by changing it to unspecified.
6506          */
6507         if (IN6_IS_ADDR_V4MAPPED_ANY(&v6src))
6508                 v6src = ipv6_all_zeros;
6509 
6510         /*
6511          * In the multirouting case, we need to replicate
6512          * the request as noted in the mcast cases above.
6513          */
6514         if (IN6_IS_ADDR_V4MAPPED(&v6group)) {
6515                 ipaddr_t group;
6516 
6517                 IN6_V4MAPPED_TO_IPADDR(&v6group, group);
6518 
6519                 ire = ire_ftable_lookup_v4(group, IP_HOST_MASK, 0,
6520                     IRE_HOST | IRE_INTERFACE, NULL, ALL_ZONES, NULL,
6521                     MATCH_IRE_MASK | MATCH_IRE_TYPE, 0, ipst, NULL);
6522         } else {
6523                 ire = ire_ftable_lookup_v6(&v6group, &ipv6_all_ones, 0,
6524                     IRE_HOST | IRE_INTERFACE, NULL, ALL_ZONES, NULL,
6525                     MATCH_IRE_MASK | MATCH_IRE_TYPE, 0, ipst, NULL);
6526         }
6527         if (ire != NULL) {
6528                 if (ire->ire_flags & RTF_MULTIRT) {
6529                         error = ip_multirt_apply_membership(optfn, ire, connp,
6530                             checkonly, &v6group, fmode, &v6src);
6531                         done = B_TRUE;
6532                 }
6533                 ire_refrele(ire);
6534         }
6535         if (!done) {
6536                 error = optfn(connp, checkonly, &v6group, ifaddr, ifindex,
6537                     fmode, &v6src);
6538         }
6539         return (error);
6540 }
6541 
6542 /*
6543  * Given a destination address and a pointer to where to put the information
6544  * this routine fills in the mtuinfo.
6545  * The socket must be connected.
6546  * For sctp conn_faddr is the primary address.
6547  */
6548 int
6549 ip_fill_mtuinfo(conn_t *connp, ip_xmit_attr_t *ixa, struct ip6_mtuinfo *mtuinfo)
6550 {
6551         uint32_t        pmtu = IP_MAXPACKET;
6552         uint_t          scopeid;
6553 
6554         if (IN6_IS_ADDR_UNSPECIFIED(&connp->conn_faddr_v6))
6555                 return (-1);
6556 
6557         /* In case we never sent or called ip_set_destination_v4/v6 */
6558         if (ixa->ixa_ire != NULL)
6559                 pmtu = ip_get_pmtu(ixa);
6560 
6561         if (ixa->ixa_flags & IXAF_SCOPEID_SET)
6562                 scopeid = ixa->ixa_scopeid;
6563         else
6564                 scopeid = 0;
6565 
6566         bzero(mtuinfo, sizeof (*mtuinfo));
6567         mtuinfo->ip6m_addr.sin6_family = AF_INET6;
6568         mtuinfo->ip6m_addr.sin6_port = connp->conn_fport;
6569         mtuinfo->ip6m_addr.sin6_addr = connp->conn_faddr_v6;
6570         mtuinfo->ip6m_addr.sin6_scope_id = scopeid;
6571         mtuinfo->ip6m_mtu = pmtu;
6572 
6573         return (sizeof (struct ip6_mtuinfo));
6574 }
6575 
6576 /*
6577  * When the src multihoming is changed from weak to [strong, preferred]
6578  * ip_ire_rebind_walker is called to walk the list of all ire_t entries
6579  * and identify routes that were created by user-applications in the
6580  * unbound state (i.e., without RTA_IFP), and for which an ire_ill is not
6581  * currently defined. These routes are then 'rebound', i.e., their ire_ill
6582  * is selected by finding an interface route for the gateway.
6583  */
6584 /* ARGSUSED */
6585 void
6586 ip_ire_rebind_walker(ire_t *ire, void *notused)
6587 {
6588         if (!ire->ire_unbound || ire->ire_ill != NULL)
6589                 return;
6590         ire_rebind(ire);
6591         ire_delete(ire);
6592 }
6593 
6594 /*
6595  * When the src multihoming is changed from  [strong, preferred] to weak,
6596  * ip_ire_unbind_walker is called to walk the list of all ire_t entries, and
6597  * set any entries that were created by user-applications in the unbound state
6598  * (i.e., without RTA_IFP) back to having a NULL ire_ill.
6599  */
6600 /* ARGSUSED */
6601 void
6602 ip_ire_unbind_walker(ire_t *ire, void *notused)
6603 {
6604         ire_t *new_ire;
6605 
6606         if (!ire->ire_unbound || ire->ire_ill == NULL)
6607                 return;
6608         if (ire->ire_ipversion == IPV6_VERSION) {
6609                 new_ire = ire_create_v6(&ire->ire_addr_v6, &ire->ire_mask_v6,
6610                     &ire->ire_gateway_addr_v6, ire->ire_type, NULL,
6611                     ire->ire_zoneid, ire->ire_flags, NULL, ire->ire_ipst);
6612         } else {
6613                 new_ire = ire_create((uchar_t *)&ire->ire_addr,
6614                     (uchar_t *)&ire->ire_mask,
6615                     (uchar_t *)&ire->ire_gateway_addr, ire->ire_type, NULL,
6616                     ire->ire_zoneid, ire->ire_flags, NULL, ire->ire_ipst);
6617         }
6618         if (new_ire == NULL)
6619                 return;
6620         new_ire->ire_unbound = B_TRUE;
6621         /*
6622          * The bound ire must first be deleted so that we don't return
6623          * the existing one on the attempt to add the unbound new_ire.
6624          */
6625         ire_delete(ire);
6626         new_ire = ire_add(new_ire);
6627         if (new_ire != NULL)
6628                 ire_refrele(new_ire);
6629 }
6630 
6631 /*
6632  * When the settings of ip*_strict_src_multihoming tunables are changed,
6633  * all cached routes need to be recomputed. This recomputation needs to be
6634  * done when going from weaker to stronger modes so that the cached ire
6635  * for the connection does not violate the current ip*_strict_src_multihoming
6636  * setting. It also needs to be done when going from stronger to weaker modes,
6637  * so that we fall back to matching on the longest-matching-route (as opposed
6638  * to a shorter match that may have been selected in the strong mode
6639  * to satisfy src_multihoming settings).
6640  *
6641  * The cached ixa_ire entires for all conn_t entries are marked as
6642  * "verify" so that they will be recomputed for the next packet.
6643  */
6644 void
6645 conn_ire_revalidate(conn_t *connp, void *arg)
6646 {
6647         boolean_t isv6 = (boolean_t)arg;
6648 
6649         if ((isv6 && connp->conn_ipversion != IPV6_VERSION) ||
6650             (!isv6 && connp->conn_ipversion != IPV4_VERSION))
6651                 return;
6652         connp->conn_ixa->ixa_ire_generation = IRE_GENERATION_VERIFY;
6653 }
6654 
6655 /*
6656  * Handles both IPv4 and IPv6 reassembly - doing the out-of-order cases,
6657  * When an ipf is passed here for the first time, if
6658  * we already have in-order fragments on the queue, we convert from the fast-
6659  * path reassembly scheme to the hard-case scheme.  From then on, additional
6660  * fragments are reassembled here.  We keep track of the start and end offsets
6661  * of each piece, and the number of holes in the chain.  When the hole count
6662  * goes to zero, we are done!
6663  *
6664  * The ipf_count will be updated to account for any mblk(s) added (pointed to
6665  * by mp) or subtracted (freeb()ed dups), upon return the caller must update
6666  * ipfb_count and ill_frag_count by the difference of ipf_count before and
6667  * after the call to ip_reassemble().
6668  */
6669 int
6670 ip_reassemble(mblk_t *mp, ipf_t *ipf, uint_t start, boolean_t more, ill_t *ill,
6671     size_t msg_len)
6672 {
6673         uint_t  end;
6674         mblk_t  *next_mp;
6675         mblk_t  *mp1;
6676         uint_t  offset;
6677         boolean_t incr_dups = B_TRUE;
6678         boolean_t offset_zero_seen = B_FALSE;
6679         boolean_t pkt_boundary_checked = B_FALSE;
6680 
6681         /* If start == 0 then ipf_nf_hdr_len has to be set. */
6682         ASSERT(start != 0 || ipf->ipf_nf_hdr_len != 0);
6683 
6684         /* Add in byte count */
6685         ipf->ipf_count += msg_len;
6686         if (ipf->ipf_end) {
6687                 /*
6688                  * We were part way through in-order reassembly, but now there
6689                  * is a hole.  We walk through messages already queued, and
6690                  * mark them for hard case reassembly.  We know that up till
6691                  * now they were in order starting from offset zero.
6692                  */
6693                 offset = 0;
6694                 for (mp1 = ipf->ipf_mp->b_cont; mp1; mp1 = mp1->b_cont) {
6695                         IP_REASS_SET_START(mp1, offset);
6696                         if (offset == 0) {
6697                                 ASSERT(ipf->ipf_nf_hdr_len != 0);
6698                                 offset = -ipf->ipf_nf_hdr_len;
6699                         }
6700                         offset += mp1->b_wptr - mp1->b_rptr;
6701                         IP_REASS_SET_END(mp1, offset);
6702                 }
6703                 /* One hole at the end. */
6704                 ipf->ipf_hole_cnt = 1;
6705                 /* Brand it as a hard case, forever. */
6706                 ipf->ipf_end = 0;
6707         }
6708         /* Walk through all the new pieces. */
6709         do {
6710                 end = start + (mp->b_wptr - mp->b_rptr);
6711                 /*
6712                  * If start is 0, decrease 'end' only for the first mblk of
6713                  * the fragment. Otherwise 'end' can get wrong value in the
6714                  * second pass of the loop if first mblk is exactly the
6715                  * size of ipf_nf_hdr_len.
6716                  */
6717                 if (start == 0 && !offset_zero_seen) {
6718                         /* First segment */
6719                         ASSERT(ipf->ipf_nf_hdr_len != 0);
6720                         end -= ipf->ipf_nf_hdr_len;
6721                         offset_zero_seen = B_TRUE;
6722                 }
6723                 next_mp = mp->b_cont;
6724                 /*
6725                  * We are checking to see if there is any interesing data
6726                  * to process.  If there isn't and the mblk isn't the
6727                  * one which carries the unfragmentable header then we
6728                  * drop it.  It's possible to have just the unfragmentable
6729                  * header come through without any data.  That needs to be
6730                  * saved.
6731                  *
6732                  * If the assert at the top of this function holds then the
6733                  * term "ipf->ipf_nf_hdr_len != 0" isn't needed.  This code
6734                  * is infrequently traveled enough that the test is left in
6735                  * to protect against future code changes which break that
6736                  * invariant.
6737                  */
6738                 if (start == end && start != 0 && ipf->ipf_nf_hdr_len != 0) {
6739                         /* Empty.  Blast it. */
6740                         IP_REASS_SET_START(mp, 0);
6741                         IP_REASS_SET_END(mp, 0);
6742                         /*
6743                          * If the ipf points to the mblk we are about to free,
6744                          * update ipf to point to the next mblk (or NULL
6745                          * if none).
6746                          */
6747                         if (ipf->ipf_mp->b_cont == mp)
6748                                 ipf->ipf_mp->b_cont = next_mp;
6749                         freeb(mp);
6750                         continue;
6751                 }
6752                 mp->b_cont = NULL;
6753                 IP_REASS_SET_START(mp, start);
6754                 IP_REASS_SET_END(mp, end);
6755                 if (!ipf->ipf_tail_mp) {
6756                         ipf->ipf_tail_mp = mp;
6757                         ipf->ipf_mp->b_cont = mp;
6758                         if (start == 0 || !more) {
6759                                 ipf->ipf_hole_cnt = 1;
6760                                 /*
6761                                  * if the first fragment comes in more than one
6762                                  * mblk, this loop will be executed for each
6763                                  * mblk. Need to adjust hole count so exiting
6764                                  * this routine will leave hole count at 1.
6765                                  */
6766                                 if (next_mp)
6767                                         ipf->ipf_hole_cnt++;
6768                         } else
6769                                 ipf->ipf_hole_cnt = 2;
6770                         continue;
6771                 } else if (ipf->ipf_last_frag_seen && !more &&
6772                     !pkt_boundary_checked) {
6773                         /*
6774                          * We check datagram boundary only if this fragment
6775                          * claims to be the last fragment and we have seen a
6776                          * last fragment in the past too. We do this only
6777                          * once for a given fragment.
6778                          *
6779                          * start cannot be 0 here as fragments with start=0
6780                          * and MF=0 gets handled as a complete packet. These
6781                          * fragments should not reach here.
6782                          */
6783 
6784                         if (start + msgdsize(mp) !=
6785                             IP_REASS_END(ipf->ipf_tail_mp)) {
6786                                 /*
6787                                  * We have two fragments both of which claim
6788                                  * to be the last fragment but gives conflicting
6789                                  * information about the whole datagram size.
6790                                  * Something fishy is going on. Drop the
6791                                  * fragment and free up the reassembly list.
6792                                  */
6793                                 return (IP_REASS_FAILED);
6794                         }
6795 
6796                         /*
6797                          * We shouldn't come to this code block again for this
6798                          * particular fragment.
6799                          */
6800                         pkt_boundary_checked = B_TRUE;
6801                 }
6802 
6803                 /* New stuff at or beyond tail? */
6804                 offset = IP_REASS_END(ipf->ipf_tail_mp);
6805                 if (start >= offset) {
6806                         if (ipf->ipf_last_frag_seen) {
6807                                 /* current fragment is beyond last fragment */
6808                                 return (IP_REASS_FAILED);
6809                         }
6810                         /* Link it on end. */
6811                         ipf->ipf_tail_mp->b_cont = mp;
6812                         ipf->ipf_tail_mp = mp;
6813                         if (more) {
6814                                 if (start != offset)
6815                                         ipf->ipf_hole_cnt++;
6816                         } else if (start == offset && next_mp == NULL)
6817                                         ipf->ipf_hole_cnt--;
6818                         continue;
6819                 }
6820                 mp1 = ipf->ipf_mp->b_cont;
6821                 offset = IP_REASS_START(mp1);
6822                 /* New stuff at the front? */
6823                 if (start < offset) {
6824                         if (start == 0) {
6825                                 if (end >= offset) {
6826                                         /* Nailed the hole at the begining. */
6827                                         ipf->ipf_hole_cnt--;
6828                                 }
6829                         } else if (end < offset) {
6830                                 /*
6831                                  * A hole, stuff, and a hole where there used
6832                                  * to be just a hole.
6833                                  */
6834                                 ipf->ipf_hole_cnt++;
6835                         }
6836                         mp->b_cont = mp1;
6837                         /* Check for overlap. */
6838                         while (end > offset) {
6839                                 if (end < IP_REASS_END(mp1)) {
6840                                         mp->b_wptr -= end - offset;
6841                                         IP_REASS_SET_END(mp, offset);
6842                                         BUMP_MIB(ill->ill_ip_mib,
6843                                             ipIfStatsReasmPartDups);
6844                                         break;
6845                                 }
6846                                 /* Did we cover another hole? */
6847                                 if ((mp1->b_cont &&
6848                                     IP_REASS_END(mp1) !=
6849                                     IP_REASS_START(mp1->b_cont) &&
6850                                     end >= IP_REASS_START(mp1->b_cont)) ||
6851                                     (!ipf->ipf_last_frag_seen && !more)) {
6852                                         ipf->ipf_hole_cnt--;
6853                                 }
6854                                 /* Clip out mp1. */
6855                                 if ((mp->b_cont = mp1->b_cont) == NULL) {
6856                                         /*
6857                                          * After clipping out mp1, this guy
6858                                          * is now hanging off the end.
6859                                          */
6860                                         ipf->ipf_tail_mp = mp;
6861                                 }
6862                                 IP_REASS_SET_START(mp1, 0);
6863                                 IP_REASS_SET_END(mp1, 0);
6864                                 /* Subtract byte count */
6865                                 ipf->ipf_count -= mp1->b_datap->db_lim -
6866                                     mp1->b_datap->db_base;
6867                                 freeb(mp1);
6868                                 BUMP_MIB(ill->ill_ip_mib,
6869                                     ipIfStatsReasmPartDups);
6870                                 mp1 = mp->b_cont;
6871                                 if (!mp1)
6872                                         break;
6873                                 offset = IP_REASS_START(mp1);
6874                         }
6875                         ipf->ipf_mp->b_cont = mp;
6876                         continue;
6877                 }
6878                 /*
6879                  * The new piece starts somewhere between the start of the head
6880                  * and before the end of the tail.
6881                  */
6882                 for (; mp1; mp1 = mp1->b_cont) {
6883                         offset = IP_REASS_END(mp1);
6884                         if (start < offset) {
6885                                 if (end <= offset) {
6886                                         /* Nothing new. */
6887                                         IP_REASS_SET_START(mp, 0);
6888                                         IP_REASS_SET_END(mp, 0);
6889                                         /* Subtract byte count */
6890                                         ipf->ipf_count -= mp->b_datap->db_lim -
6891                                             mp->b_datap->db_base;
6892                                         if (incr_dups) {
6893                                                 ipf->ipf_num_dups++;
6894                                                 incr_dups = B_FALSE;
6895                                         }
6896                                         freeb(mp);
6897                                         BUMP_MIB(ill->ill_ip_mib,
6898                                             ipIfStatsReasmDuplicates);
6899                                         break;
6900                                 }
6901                                 /*
6902                                  * Trim redundant stuff off beginning of new
6903                                  * piece.
6904                                  */
6905                                 IP_REASS_SET_START(mp, offset);
6906                                 mp->b_rptr += offset - start;
6907                                 BUMP_MIB(ill->ill_ip_mib,
6908                                     ipIfStatsReasmPartDups);
6909                                 start = offset;
6910                                 if (!mp1->b_cont) {
6911                                         /*
6912                                          * After trimming, this guy is now
6913                                          * hanging off the end.
6914                                          */
6915                                         mp1->b_cont = mp;
6916                                         ipf->ipf_tail_mp = mp;
6917                                         if (!more) {
6918                                                 ipf->ipf_hole_cnt--;
6919                                         }
6920                                         break;
6921                                 }
6922                         }
6923                         if (start >= IP_REASS_START(mp1->b_cont))
6924                                 continue;
6925                         /* Fill a hole */
6926                         if (start > offset)
6927                                 ipf->ipf_hole_cnt++;
6928                         mp->b_cont = mp1->b_cont;
6929                         mp1->b_cont = mp;
6930                         mp1 = mp->b_cont;
6931                         offset = IP_REASS_START(mp1);
6932                         if (end >= offset) {
6933                                 ipf->ipf_hole_cnt--;
6934                                 /* Check for overlap. */
6935                                 while (end > offset) {
6936                                         if (end < IP_REASS_END(mp1)) {
6937                                                 mp->b_wptr -= end - offset;
6938                                                 IP_REASS_SET_END(mp, offset);
6939                                                 /*
6940                                                  * TODO we might bump
6941                                                  * this up twice if there is
6942                                                  * overlap at both ends.
6943                                                  */
6944                                                 BUMP_MIB(ill->ill_ip_mib,
6945                                                     ipIfStatsReasmPartDups);
6946                                                 break;
6947                                         }
6948                                         /* Did we cover another hole? */
6949                                         if ((mp1->b_cont &&
6950                                             IP_REASS_END(mp1)
6951                                             != IP_REASS_START(mp1->b_cont) &&
6952                                             end >=
6953                                             IP_REASS_START(mp1->b_cont)) ||
6954                                             (!ipf->ipf_last_frag_seen &&
6955                                             !more)) {
6956                                                 ipf->ipf_hole_cnt--;
6957                                         }
6958                                         /* Clip out mp1. */
6959                                         if ((mp->b_cont = mp1->b_cont) ==
6960                                             NULL) {
6961                                                 /*
6962                                                  * After clipping out mp1,
6963                                                  * this guy is now hanging
6964                                                  * off the end.
6965                                                  */
6966                                                 ipf->ipf_tail_mp = mp;
6967                                         }
6968                                         IP_REASS_SET_START(mp1, 0);
6969                                         IP_REASS_SET_END(mp1, 0);
6970                                         /* Subtract byte count */
6971                                         ipf->ipf_count -=
6972                                             mp1->b_datap->db_lim -
6973                                             mp1->b_datap->db_base;
6974                                         freeb(mp1);
6975                                         BUMP_MIB(ill->ill_ip_mib,
6976                                             ipIfStatsReasmPartDups);
6977                                         mp1 = mp->b_cont;
6978                                         if (!mp1)
6979                                                 break;
6980                                         offset = IP_REASS_START(mp1);
6981                                 }
6982                         }
6983                         break;
6984                 }
6985         } while (start = end, mp = next_mp);
6986 
6987         /* Fragment just processed could be the last one. Remember this fact */
6988         if (!more)
6989                 ipf->ipf_last_frag_seen = B_TRUE;
6990 
6991         /* Still got holes? */
6992         if (ipf->ipf_hole_cnt)
6993                 return (IP_REASS_PARTIAL);
6994         /* Clean up overloaded fields to avoid upstream disasters. */
6995         for (mp1 = ipf->ipf_mp->b_cont; mp1; mp1 = mp1->b_cont) {
6996                 IP_REASS_SET_START(mp1, 0);
6997                 IP_REASS_SET_END(mp1, 0);
6998         }
6999         return (IP_REASS_COMPLETE);
7000 }
7001 
7002 /*
7003  * Fragmentation reassembly.  Each ILL has a hash table for
7004  * queuing packets undergoing reassembly for all IPIFs
7005  * associated with the ILL.  The hash is based on the packet
7006  * IP ident field.  The ILL frag hash table was allocated
7007  * as a timer block at the time the ILL was created.  Whenever
7008  * there is anything on the reassembly queue, the timer will
7009  * be running.  Returns the reassembled packet if reassembly completes.
7010  */
7011 mblk_t *
7012 ip_input_fragment(mblk_t *mp, ipha_t *ipha, ip_recv_attr_t *ira)
7013 {
7014         uint32_t        frag_offset_flags;
7015         mblk_t          *t_mp;
7016         ipaddr_t        dst;
7017         uint8_t         proto = ipha->ipha_protocol;
7018         uint32_t        sum_val;
7019         uint16_t        sum_flags;
7020         ipf_t           *ipf;
7021         ipf_t           **ipfp;
7022         ipfb_t          *ipfb;
7023         uint16_t        ident;
7024         uint32_t        offset;
7025         ipaddr_t        src;
7026         uint_t          hdr_length;
7027         uint32_t        end;
7028         mblk_t          *mp1;
7029         mblk_t          *tail_mp;
7030         size_t          count;
7031         size_t          msg_len;
7032         uint8_t         ecn_info = 0;
7033         uint32_t        packet_size;
7034         boolean_t       pruned = B_FALSE;
7035         ill_t           *ill = ira->ira_ill;
7036         ip_stack_t      *ipst = ill->ill_ipst;
7037 
7038         /*
7039          * Drop the fragmented as early as possible, if
7040          * we don't have resource(s) to re-assemble.
7041          */
7042         if (ipst->ips_ip_reass_queue_bytes == 0) {
7043                 freemsg(mp);
7044                 return (NULL);
7045         }
7046 
7047         /* Check for fragmentation offset; return if there's none */
7048         if ((frag_offset_flags = ntohs(ipha->ipha_fragment_offset_and_flags) &
7049             (IPH_MF | IPH_OFFSET)) == 0)
7050                 return (mp);
7051 
7052         /*
7053          * We utilize hardware computed checksum info only for UDP since
7054          * IP fragmentation is a normal occurrence for the protocol.  In
7055          * addition, checksum offload support for IP fragments carrying
7056          * UDP payload is commonly implemented across network adapters.
7057          */
7058         ASSERT(ira->ira_rill != NULL);
7059         if (proto == IPPROTO_UDP && dohwcksum &&
7060             ILL_HCKSUM_CAPABLE(ira->ira_rill) &&
7061             (DB_CKSUMFLAGS(mp) & (HCK_FULLCKSUM | HCK_PARTIALCKSUM))) {
7062                 mblk_t *mp1 = mp->b_cont;
7063                 int32_t len;
7064 
7065                 /* Record checksum information from the packet */
7066                 sum_val = (uint32_t)DB_CKSUM16(mp);
7067                 sum_flags = DB_CKSUMFLAGS(mp);
7068 
7069                 /* IP payload offset from beginning of mblk */
7070                 offset = ((uchar_t *)ipha + IPH_HDR_LENGTH(ipha)) - mp->b_rptr;
7071 
7072                 if ((sum_flags & HCK_PARTIALCKSUM) &&
7073                     (mp1 == NULL || mp1->b_cont == NULL) &&
7074                     offset >= DB_CKSUMSTART(mp) &&
7075                     ((len = offset - DB_CKSUMSTART(mp)) & 1) == 0) {
7076                         uint32_t adj;
7077                         /*
7078                          * Partial checksum has been calculated by hardware
7079                          * and attached to the packet; in addition, any
7080                          * prepended extraneous data is even byte aligned.
7081                          * If any such data exists, we adjust the checksum;
7082                          * this would also handle any postpended data.
7083                          */
7084                         IP_ADJCKSUM_PARTIAL(mp->b_rptr + DB_CKSUMSTART(mp),
7085                             mp, mp1, len, adj);
7086 
7087                         /* One's complement subtract extraneous checksum */
7088                         if (adj >= sum_val)
7089                                 sum_val = ~(adj - sum_val) & 0xFFFF;
7090                         else
7091                                 sum_val -= adj;
7092                 }
7093         } else {
7094                 sum_val = 0;
7095                 sum_flags = 0;
7096         }
7097 
7098         /* Clear hardware checksumming flag */
7099         DB_CKSUMFLAGS(mp) = 0;
7100 
7101         ident = ipha->ipha_ident;
7102         offset = (frag_offset_flags << 3) & 0xFFFF;
7103         src = ipha->ipha_src;
7104         dst = ipha->ipha_dst;
7105         hdr_length = IPH_HDR_LENGTH(ipha);
7106         end = ntohs(ipha->ipha_length) - hdr_length;
7107 
7108         /* If end == 0 then we have a packet with no data, so just free it */
7109         if (end == 0) {
7110                 freemsg(mp);
7111                 return (NULL);
7112         }
7113 
7114         /* Record the ECN field info. */
7115         ecn_info = (ipha->ipha_type_of_service & 0x3);
7116         if (offset != 0) {
7117                 /*
7118                  * If this isn't the first piece, strip the header, and
7119                  * add the offset to the end value.
7120                  */
7121                 mp->b_rptr += hdr_length;
7122                 end += offset;
7123         }
7124 
7125         /* Handle vnic loopback of fragments */
7126         if (mp->b_datap->db_ref > 2)
7127                 msg_len = 0;
7128         else
7129                 msg_len = MBLKSIZE(mp);
7130 
7131         tail_mp = mp;
7132         while (tail_mp->b_cont != NULL) {
7133                 tail_mp = tail_mp->b_cont;
7134                 if (tail_mp->b_datap->db_ref <= 2)
7135                         msg_len += MBLKSIZE(tail_mp);
7136         }
7137 
7138         /* If the reassembly list for this ILL will get too big, prune it */
7139         if ((msg_len + sizeof (*ipf) + ill->ill_frag_count) >=
7140             ipst->ips_ip_reass_queue_bytes) {
7141                 DTRACE_PROBE3(ip_reass_queue_bytes, uint_t, msg_len,
7142                     uint_t, ill->ill_frag_count,
7143                     uint_t, ipst->ips_ip_reass_queue_bytes);
7144                 ill_frag_prune(ill,
7145                     (ipst->ips_ip_reass_queue_bytes < msg_len) ? 0 :
7146                     (ipst->ips_ip_reass_queue_bytes - msg_len));
7147                 pruned = B_TRUE;
7148         }
7149 
7150         ipfb = &ill->ill_frag_hash_tbl[ILL_FRAG_HASH(src, ident)];
7151         mutex_enter(&ipfb->ipfb_lock);
7152 
7153         ipfp = &ipfb->ipfb_ipf;
7154         /* Try to find an existing fragment queue for this packet. */
7155         for (;;) {
7156                 ipf = ipfp[0];
7157                 if (ipf != NULL) {
7158                         /*
7159                          * It has to match on ident and src/dst address.
7160                          */
7161                         if (ipf->ipf_ident == ident &&
7162                             ipf->ipf_src == src &&
7163                             ipf->ipf_dst == dst &&
7164                             ipf->ipf_protocol == proto) {
7165                                 /*
7166                                  * If we have received too many
7167                                  * duplicate fragments for this packet
7168                                  * free it.
7169                                  */
7170                                 if (ipf->ipf_num_dups > ip_max_frag_dups) {
7171                                         ill_frag_free_pkts(ill, ipfb, ipf, 1);
7172                                         freemsg(mp);
7173                                         mutex_exit(&ipfb->ipfb_lock);
7174                                         return (NULL);
7175                                 }
7176                                 /* Found it. */
7177                                 break;
7178                         }
7179                         ipfp = &ipf->ipf_hash_next;
7180                         continue;
7181                 }
7182 
7183                 /*
7184                  * If we pruned the list, do we want to store this new
7185                  * fragment?. We apply an optimization here based on the
7186                  * fact that most fragments will be received in order.
7187                  * So if the offset of this incoming fragment is zero,
7188                  * it is the first fragment of a new packet. We will
7189                  * keep it.  Otherwise drop the fragment, as we have
7190                  * probably pruned the packet already (since the
7191                  * packet cannot be found).
7192                  */
7193                 if (pruned && offset != 0) {
7194                         mutex_exit(&ipfb->ipfb_lock);
7195                         freemsg(mp);
7196                         return (NULL);
7197                 }
7198 
7199                 if (ipfb->ipfb_frag_pkts >= MAX_FRAG_PKTS(ipst))  {
7200                         /*
7201                          * Too many fragmented packets in this hash
7202                          * bucket. Free the oldest.
7203                          */
7204                         ill_frag_free_pkts(ill, ipfb, ipfb->ipfb_ipf, 1);
7205                 }
7206 
7207                 /* New guy.  Allocate a frag message. */
7208                 mp1 = allocb(sizeof (*ipf), BPRI_MED);
7209                 if (mp1 == NULL) {
7210                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
7211                         ip_drop_input("ipIfStatsInDiscards", mp, ill);
7212                         freemsg(mp);
7213 reass_done:
7214                         mutex_exit(&ipfb->ipfb_lock);
7215                         return (NULL);
7216                 }
7217 
7218                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsReasmReqds);
7219                 mp1->b_cont = mp;
7220 
7221                 /* Initialize the fragment header. */
7222                 ipf = (ipf_t *)mp1->b_rptr;
7223                 ipf->ipf_mp = mp1;
7224                 ipf->ipf_ptphn = ipfp;
7225                 ipfp[0] = ipf;
7226                 ipf->ipf_hash_next = NULL;
7227                 ipf->ipf_ident = ident;
7228                 ipf->ipf_protocol = proto;
7229                 ipf->ipf_src = src;
7230                 ipf->ipf_dst = dst;
7231                 ipf->ipf_nf_hdr_len = 0;
7232                 /* Record reassembly start time. */
7233                 ipf->ipf_timestamp = gethrestime_sec();
7234                 /* Record ipf generation and account for frag header */
7235                 ipf->ipf_gen = ill->ill_ipf_gen++;
7236                 ipf->ipf_count = MBLKSIZE(mp1);
7237                 ipf->ipf_last_frag_seen = B_FALSE;
7238                 ipf->ipf_ecn = ecn_info;
7239                 ipf->ipf_num_dups = 0;
7240                 ipfb->ipfb_frag_pkts++;
7241                 ipf->ipf_checksum = 0;
7242                 ipf->ipf_checksum_flags = 0;
7243 
7244                 /* Store checksum value in fragment header */
7245                 if (sum_flags != 0) {
7246                         sum_val = (sum_val & 0xFFFF) + (sum_val >> 16);
7247                         sum_val = (sum_val & 0xFFFF) + (sum_val >> 16);
7248                         ipf->ipf_checksum = sum_val;
7249                         ipf->ipf_checksum_flags = sum_flags;
7250                 }
7251 
7252                 /*
7253                  * We handle reassembly two ways.  In the easy case,
7254                  * where all the fragments show up in order, we do
7255                  * minimal bookkeeping, and just clip new pieces on
7256                  * the end.  If we ever see a hole, then we go off
7257                  * to ip_reassemble which has to mark the pieces and
7258                  * keep track of the number of holes, etc.  Obviously,
7259                  * the point of having both mechanisms is so we can
7260                  * handle the easy case as efficiently as possible.
7261                  */
7262                 if (offset == 0) {
7263                         /* Easy case, in-order reassembly so far. */
7264                         ipf->ipf_count += msg_len;
7265                         ipf->ipf_tail_mp = tail_mp;
7266                         /*
7267                          * Keep track of next expected offset in
7268                          * ipf_end.
7269                          */
7270                         ipf->ipf_end = end;
7271                         ipf->ipf_nf_hdr_len = hdr_length;
7272                 } else {
7273                         /* Hard case, hole at the beginning. */
7274                         ipf->ipf_tail_mp = NULL;
7275                         /*
7276                          * ipf_end == 0 means that we have given up
7277                          * on easy reassembly.
7278                          */
7279                         ipf->ipf_end = 0;
7280 
7281                         /* Forget checksum offload from now on */
7282                         ipf->ipf_checksum_flags = 0;
7283 
7284                         /*
7285                          * ipf_hole_cnt is set by ip_reassemble.
7286                          * ipf_count is updated by ip_reassemble.
7287                          * No need to check for return value here
7288                          * as we don't expect reassembly to complete
7289                          * or fail for the first fragment itself.
7290                          */
7291                         (void) ip_reassemble(mp, ipf,
7292                             (frag_offset_flags & IPH_OFFSET) << 3,
7293                             (frag_offset_flags & IPH_MF), ill, msg_len);
7294                 }
7295                 /* Update per ipfb and ill byte counts */
7296                 ipfb->ipfb_count += ipf->ipf_count;
7297                 ASSERT(ipfb->ipfb_count > 0);     /* Wraparound */
7298                 atomic_add_32(&ill->ill_frag_count, ipf->ipf_count);
7299                 /* If the frag timer wasn't already going, start it. */
7300                 mutex_enter(&ill->ill_lock);
7301                 ill_frag_timer_start(ill);
7302                 mutex_exit(&ill->ill_lock);
7303                 goto reass_done;
7304         }
7305 
7306         /*
7307          * If the packet's flag has changed (it could be coming up
7308          * from an interface different than the previous, therefore
7309          * possibly different checksum capability), then forget about
7310          * any stored checksum states.  Otherwise add the value to
7311          * the existing one stored in the fragment header.
7312          */
7313         if (sum_flags != 0 && sum_flags == ipf->ipf_checksum_flags) {
7314                 sum_val += ipf->ipf_checksum;
7315                 sum_val = (sum_val & 0xFFFF) + (sum_val >> 16);
7316                 sum_val = (sum_val & 0xFFFF) + (sum_val >> 16);
7317                 ipf->ipf_checksum = sum_val;
7318         } else if (ipf->ipf_checksum_flags != 0) {
7319                 /* Forget checksum offload from now on */
7320                 ipf->ipf_checksum_flags = 0;
7321         }
7322 
7323         /*
7324          * We have a new piece of a datagram which is already being
7325          * reassembled.  Update the ECN info if all IP fragments
7326          * are ECN capable.  If there is one which is not, clear
7327          * all the info.  If there is at least one which has CE
7328          * code point, IP needs to report that up to transport.
7329          */
7330         if (ecn_info != IPH_ECN_NECT && ipf->ipf_ecn != IPH_ECN_NECT) {
7331                 if (ecn_info == IPH_ECN_CE)
7332                         ipf->ipf_ecn = IPH_ECN_CE;
7333         } else {
7334                 ipf->ipf_ecn = IPH_ECN_NECT;
7335         }
7336         if (offset && ipf->ipf_end == offset) {
7337                 /* The new fragment fits at the end */
7338                 ipf->ipf_tail_mp->b_cont = mp;
7339                 /* Update the byte count */
7340                 ipf->ipf_count += msg_len;
7341                 /* Update per ipfb and ill byte counts */
7342                 ipfb->ipfb_count += msg_len;
7343                 ASSERT(ipfb->ipfb_count > 0);     /* Wraparound */
7344                 atomic_add_32(&ill->ill_frag_count, msg_len);
7345                 if (frag_offset_flags & IPH_MF) {
7346                         /* More to come. */
7347                         ipf->ipf_end = end;
7348                         ipf->ipf_tail_mp = tail_mp;
7349                         goto reass_done;
7350                 }
7351         } else {
7352                 /* Go do the hard cases. */
7353                 int ret;
7354 
7355                 if (offset == 0)
7356                         ipf->ipf_nf_hdr_len = hdr_length;
7357 
7358                 /* Save current byte count */
7359                 count = ipf->ipf_count;
7360                 ret = ip_reassemble(mp, ipf,
7361                     (frag_offset_flags & IPH_OFFSET) << 3,
7362                     (frag_offset_flags & IPH_MF), ill, msg_len);
7363                 /* Count of bytes added and subtracted (freeb()ed) */
7364                 count = ipf->ipf_count - count;
7365                 if (count) {
7366                         /* Update per ipfb and ill byte counts */
7367                         ipfb->ipfb_count += count;
7368                         ASSERT(ipfb->ipfb_count > 0); /* Wraparound */
7369                         atomic_add_32(&ill->ill_frag_count, count);
7370                 }
7371                 if (ret == IP_REASS_PARTIAL) {
7372                         goto reass_done;
7373                 } else if (ret == IP_REASS_FAILED) {
7374                         /* Reassembly failed. Free up all resources */
7375                         ill_frag_free_pkts(ill, ipfb, ipf, 1);
7376                         for (t_mp = mp; t_mp != NULL; t_mp = t_mp->b_cont) {
7377                                 IP_REASS_SET_START(t_mp, 0);
7378                                 IP_REASS_SET_END(t_mp, 0);
7379                         }
7380                         freemsg(mp);
7381                         goto reass_done;
7382                 }
7383                 /* We will reach here iff 'ret' is IP_REASS_COMPLETE */
7384         }
7385         /*
7386          * We have completed reassembly.  Unhook the frag header from
7387          * the reassembly list.
7388          *
7389          * Before we free the frag header, record the ECN info
7390          * to report back to the transport.
7391          */
7392         ecn_info = ipf->ipf_ecn;
7393         BUMP_MIB(ill->ill_ip_mib, ipIfStatsReasmOKs);
7394         ipfp = ipf->ipf_ptphn;
7395 
7396         /* We need to supply these to caller */
7397         if ((sum_flags = ipf->ipf_checksum_flags) != 0)
7398                 sum_val = ipf->ipf_checksum;
7399         else
7400                 sum_val = 0;
7401 
7402         mp1 = ipf->ipf_mp;
7403         count = ipf->ipf_count;
7404         ipf = ipf->ipf_hash_next;
7405         if (ipf != NULL)
7406                 ipf->ipf_ptphn = ipfp;
7407         ipfp[0] = ipf;
7408         atomic_add_32(&ill->ill_frag_count, -count);
7409         ASSERT(ipfb->ipfb_count >= count);
7410         ipfb->ipfb_count -= count;
7411         ipfb->ipfb_frag_pkts--;
7412         mutex_exit(&ipfb->ipfb_lock);
7413         /* Ditch the frag header. */
7414         mp = mp1->b_cont;
7415 
7416         freeb(mp1);
7417 
7418         /* Restore original IP length in header. */
7419         packet_size = (uint32_t)msgdsize(mp);
7420         if (packet_size > IP_MAXPACKET) {
7421                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7422                 ip_drop_input("Reassembled packet too large", mp, ill);
7423                 freemsg(mp);
7424                 return (NULL);
7425         }
7426 
7427         if (DB_REF(mp) > 1) {
7428                 mblk_t *mp2 = copymsg(mp);
7429 
7430                 if (mp2 == NULL) {
7431                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
7432                         ip_drop_input("ipIfStatsInDiscards", mp, ill);
7433                         freemsg(mp);
7434                         return (NULL);
7435                 }
7436                 freemsg(mp);
7437                 mp = mp2;
7438         }
7439         ipha = (ipha_t *)mp->b_rptr;
7440 
7441         ipha->ipha_length = htons((uint16_t)packet_size);
7442         /* We're now complete, zip the frag state */
7443         ipha->ipha_fragment_offset_and_flags = 0;
7444         /* Record the ECN info. */
7445         ipha->ipha_type_of_service &= 0xFC;
7446         ipha->ipha_type_of_service |= ecn_info;
7447 
7448         /* Update the receive attributes */
7449         ira->ira_pktlen = packet_size;
7450         ira->ira_ip_hdr_length = IPH_HDR_LENGTH(ipha);
7451 
7452         /* Reassembly is successful; set checksum information in packet */
7453         DB_CKSUM16(mp) = (uint16_t)sum_val;
7454         DB_CKSUMFLAGS(mp) = sum_flags;
7455         DB_CKSUMSTART(mp) = ira->ira_ip_hdr_length;
7456 
7457         return (mp);
7458 }
7459 
7460 /*
7461  * Pullup function that should be used for IP input in order to
7462  * ensure we do not loose the L2 source address; we need the l2 source
7463  * address for IP_RECVSLLA and for ndp_input.
7464  *
7465  * We return either NULL or b_rptr.
7466  */
7467 void *
7468 ip_pullup(mblk_t *mp, ssize_t len, ip_recv_attr_t *ira)
7469 {
7470         ill_t           *ill = ira->ira_ill;
7471 
7472         if (ip_rput_pullups++ == 0) {
7473                 (void) mi_strlog(ill->ill_rq, 1, SL_ERROR|SL_TRACE,
7474                     "ip_pullup: %s forced us to "
7475                     " pullup pkt, hdr len %ld, hdr addr %p",
7476                     ill->ill_name, len, (void *)mp->b_rptr);
7477         }
7478         if (!(ira->ira_flags & IRAF_L2SRC_SET))
7479                 ip_setl2src(mp, ira, ira->ira_rill);
7480         ASSERT(ira->ira_flags & IRAF_L2SRC_SET);
7481         if (!pullupmsg(mp, len))
7482                 return (NULL);
7483         else
7484                 return (mp->b_rptr);
7485 }
7486 
7487 /*
7488  * Make sure ira_l2src has an address. If we don't have one fill with zeros.
7489  * When called from the ULP ira_rill will be NULL hence the caller has to
7490  * pass in the ill.
7491  */
7492 /* ARGSUSED */
7493 void
7494 ip_setl2src(mblk_t *mp, ip_recv_attr_t *ira, ill_t *ill)
7495 {
7496         const uchar_t *addr;
7497         int alen;
7498 
7499         if (ira->ira_flags & IRAF_L2SRC_SET)
7500                 return;
7501 
7502         ASSERT(ill != NULL);
7503         alen = ill->ill_phys_addr_length;
7504         ASSERT(alen <= sizeof (ira->ira_l2src));
7505         if (ira->ira_mhip != NULL &&
7506             (addr = ira->ira_mhip->mhi_saddr) != NULL) {
7507                 bcopy(addr, ira->ira_l2src, alen);
7508         } else if ((ira->ira_flags & IRAF_L2SRC_LOOPBACK) &&
7509             (addr = ill->ill_phys_addr) != NULL) {
7510                 bcopy(addr, ira->ira_l2src, alen);
7511         } else {
7512                 bzero(ira->ira_l2src, alen);
7513         }
7514         ira->ira_flags |= IRAF_L2SRC_SET;
7515 }
7516 
7517 /*
7518  * check ip header length and align it.
7519  */
7520 mblk_t *
7521 ip_check_and_align_header(mblk_t *mp, uint_t min_size, ip_recv_attr_t *ira)
7522 {
7523         ill_t   *ill = ira->ira_ill;
7524         ssize_t len;
7525 
7526         len = MBLKL(mp);
7527 
7528         if (!OK_32PTR(mp->b_rptr))
7529                 IP_STAT(ill->ill_ipst, ip_notaligned);
7530         else
7531                 IP_STAT(ill->ill_ipst, ip_recv_pullup);
7532 
7533         /* Guard against bogus device drivers */
7534         if (len < 0) {
7535                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7536                 ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7537                 freemsg(mp);
7538                 return (NULL);
7539         }
7540 
7541         if (len == 0) {
7542                 /* GLD sometimes sends up mblk with b_rptr == b_wptr! */
7543                 mblk_t *mp1 = mp->b_cont;
7544 
7545                 if (!(ira->ira_flags & IRAF_L2SRC_SET))
7546                         ip_setl2src(mp, ira, ira->ira_rill);
7547                 ASSERT(ira->ira_flags & IRAF_L2SRC_SET);
7548 
7549                 freeb(mp);
7550                 mp = mp1;
7551                 if (mp == NULL)
7552                         return (NULL);
7553 
7554                 if (OK_32PTR(mp->b_rptr) && MBLKL(mp) >= min_size)
7555                         return (mp);
7556         }
7557         if (ip_pullup(mp, min_size, ira) == NULL) {
7558                 if (msgdsize(mp) < min_size) {
7559                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7560                         ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7561                 } else {
7562                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
7563                         ip_drop_input("ipIfStatsInDiscards", mp, ill);
7564                 }
7565                 freemsg(mp);
7566                 return (NULL);
7567         }
7568         return (mp);
7569 }
7570 
7571 /*
7572  * Common code for IPv4 and IPv6 to check and pullup multi-mblks
7573  */
7574 mblk_t *
7575 ip_check_length(mblk_t *mp, uchar_t *rptr, ssize_t len, uint_t pkt_len,
7576     uint_t min_size, ip_recv_attr_t *ira)
7577 {
7578         ill_t   *ill = ira->ira_ill;
7579 
7580         /*
7581          * Make sure we have data length consistent
7582          * with the IP header.
7583          */
7584         if (mp->b_cont == NULL) {
7585                 /* pkt_len is based on ipha_len, not the mblk length */
7586                 if (pkt_len < min_size) {
7587                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7588                         ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7589                         freemsg(mp);
7590                         return (NULL);
7591                 }
7592                 if (len < 0) {
7593                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInTruncatedPkts);
7594                         ip_drop_input("ipIfStatsInTruncatedPkts", mp, ill);
7595                         freemsg(mp);
7596                         return (NULL);
7597                 }
7598                 /* Drop any pad */
7599                 mp->b_wptr = rptr + pkt_len;
7600         } else if ((len += msgdsize(mp->b_cont)) != 0) {
7601                 ASSERT(pkt_len >= min_size);
7602                 if (pkt_len < min_size) {
7603                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7604                         ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7605                         freemsg(mp);
7606                         return (NULL);
7607                 }
7608                 if (len < 0) {
7609                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInTruncatedPkts);
7610                         ip_drop_input("ipIfStatsInTruncatedPkts", mp, ill);
7611                         freemsg(mp);
7612                         return (NULL);
7613                 }
7614                 /* Drop any pad */
7615                 (void) adjmsg(mp, -len);
7616                 /*
7617                  * adjmsg may have freed an mblk from the chain, hence
7618                  * invalidate any hw checksum here. This will force IP to
7619                  * calculate the checksum in sw, but only for this packet.
7620                  */
7621                 DB_CKSUMFLAGS(mp) = 0;
7622                 IP_STAT(ill->ill_ipst, ip_multimblk);
7623         }
7624         return (mp);
7625 }
7626 
7627 /*
7628  * Check that the IPv4 opt_len is consistent with the packet and pullup
7629  * the options.
7630  */
7631 mblk_t *
7632 ip_check_optlen(mblk_t *mp, ipha_t *ipha, uint_t opt_len, uint_t pkt_len,
7633     ip_recv_attr_t *ira)
7634 {
7635         ill_t   *ill = ira->ira_ill;
7636         ssize_t len;
7637 
7638         /* Assume no IPv6 packets arrive over the IPv4 queue */
7639         if (IPH_HDR_VERSION(ipha) != IPV4_VERSION) {
7640                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7641                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInWrongIPVersion);
7642                 ip_drop_input("IPvN packet on IPv4 ill", mp, ill);
7643                 freemsg(mp);
7644                 return (NULL);
7645         }
7646 
7647         if (opt_len > (15 - IP_SIMPLE_HDR_LENGTH_IN_WORDS)) {
7648                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7649                 ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7650                 freemsg(mp);
7651                 return (NULL);
7652         }
7653         /*
7654          * Recompute complete header length and make sure we
7655          * have access to all of it.
7656          */
7657         len = ((size_t)opt_len + IP_SIMPLE_HDR_LENGTH_IN_WORDS) << 2;
7658         if (len > (mp->b_wptr - mp->b_rptr)) {
7659                 if (len > pkt_len) {
7660                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7661                         ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7662                         freemsg(mp);
7663                         return (NULL);
7664                 }
7665                 if (ip_pullup(mp, len, ira) == NULL) {
7666                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
7667                         ip_drop_input("ipIfStatsInDiscards", mp, ill);
7668                         freemsg(mp);
7669                         return (NULL);
7670                 }
7671         }
7672         return (mp);
7673 }
7674 
7675 /*
7676  * Returns a new ire, or the same ire, or NULL.
7677  * If a different IRE is returned, then it is held; the caller
7678  * needs to release it.
7679  * In no case is there any hold/release on the ire argument.
7680  */
7681 ire_t *
7682 ip_check_multihome(void *addr, ire_t *ire, ill_t *ill)
7683 {
7684         ire_t           *new_ire;
7685         ill_t           *ire_ill;
7686         uint_t          ifindex;
7687         ip_stack_t      *ipst = ill->ill_ipst;
7688         boolean_t       strict_check = B_FALSE;
7689 
7690         /*
7691          * IPMP common case: if IRE and ILL are in the same group, there's no
7692          * issue (e.g. packet received on an underlying interface matched an
7693          * IRE_LOCAL on its associated group interface).
7694          */
7695         ASSERT(ire->ire_ill != NULL);
7696         if (IS_IN_SAME_ILLGRP(ill, ire->ire_ill))
7697                 return (ire);
7698 
7699         /*
7700          * Do another ire lookup here, using the ingress ill, to see if the
7701          * interface is in a usesrc group.
7702          * As long as the ills belong to the same group, we don't consider
7703          * them to be arriving on the wrong interface. Thus, if the switch
7704          * is doing inbound load spreading, we won't drop packets when the
7705          * ip*_strict_dst_multihoming switch is on.
7706          * We also need to check for IPIF_UNNUMBERED point2point interfaces
7707          * where the local address may not be unique. In this case we were
7708          * at the mercy of the initial ire lookup and the IRE_LOCAL it
7709          * actually returned. The new lookup, which is more specific, should
7710          * only find the IRE_LOCAL associated with the ingress ill if one
7711          * exists.
7712          */
7713         if (ire->ire_ipversion == IPV4_VERSION) {
7714                 if (ipst->ips_ip_strict_dst_multihoming)
7715                         strict_check = B_TRUE;
7716                 new_ire = ire_ftable_lookup_v4(*((ipaddr_t *)addr), 0, 0,
7717                     IRE_LOCAL, ill, ALL_ZONES, NULL,
7718                     (MATCH_IRE_TYPE|MATCH_IRE_ILL), 0, ipst, NULL);
7719         } else {
7720                 ASSERT(!IN6_IS_ADDR_MULTICAST((in6_addr_t *)addr));
7721                 if (ipst->ips_ipv6_strict_dst_multihoming)
7722                         strict_check = B_TRUE;
7723                 new_ire = ire_ftable_lookup_v6((in6_addr_t *)addr, NULL, NULL,
7724                     IRE_LOCAL, ill, ALL_ZONES, NULL,
7725                     (MATCH_IRE_TYPE|MATCH_IRE_ILL), 0, ipst, NULL);
7726         }
7727         /*
7728          * If the same ire that was returned in ip_input() is found then this
7729          * is an indication that usesrc groups are in use. The packet
7730          * arrived on a different ill in the group than the one associated with
7731          * the destination address.  If a different ire was found then the same
7732          * IP address must be hosted on multiple ills. This is possible with
7733          * unnumbered point2point interfaces. We switch to use this new ire in
7734          * order to have accurate interface statistics.
7735          */
7736         if (new_ire != NULL) {
7737                 /* Note: held in one case but not the other? Caller handles */
7738                 if (new_ire != ire)
7739                         return (new_ire);
7740                 /* Unchanged */
7741                 ire_refrele(new_ire);
7742                 return (ire);
7743         }
7744 
7745         /*
7746          * Chase pointers once and store locally.
7747          */
7748         ASSERT(ire->ire_ill != NULL);
7749         ire_ill = ire->ire_ill;
7750         ifindex = ill->ill_usesrc_ifindex;
7751 
7752         /*
7753          * Check if it's a legal address on the 'usesrc' interface.
7754          * For IPMP data addresses the IRE_LOCAL is the upper, hence we
7755          * can just check phyint_ifindex.
7756          */
7757         if (ifindex != 0 && ifindex == ire_ill->ill_phyint->phyint_ifindex) {
7758                 return (ire);
7759         }
7760 
7761         /*
7762          * If the ip*_strict_dst_multihoming switch is on then we can
7763          * only accept this packet if the interface is marked as routing.
7764          */
7765         if (!(strict_check))
7766                 return (ire);
7767 
7768         if ((ill->ill_flags & ire->ire_ill->ill_flags & ILLF_ROUTER) != 0) {
7769                 return (ire);
7770         }
7771         return (NULL);
7772 }
7773 
7774 /*
7775  * This function is used to construct a mac_header_info_s from a
7776  * DL_UNITDATA_IND message.
7777  * The address fields in the mhi structure points into the message,
7778  * thus the caller can't use those fields after freeing the message.
7779  *
7780  * We determine whether the packet received is a non-unicast packet
7781  * and in doing so, determine whether or not it is broadcast vs multicast.
7782  * For it to be a broadcast packet, we must have the appropriate mblk_t
7783  * hanging off the ill_t.  If this is either not present or doesn't match
7784  * the destination mac address in the DL_UNITDATA_IND, the packet is deemed
7785  * to be multicast.  Thus NICs that have no broadcast address (or no
7786  * capability for one, such as point to point links) cannot return as
7787  * the packet being broadcast.
7788  */
7789 void
7790 ip_dlur_to_mhi(ill_t *ill, mblk_t *mb, struct mac_header_info_s *mhip)
7791 {
7792         dl_unitdata_ind_t *ind = (dl_unitdata_ind_t *)mb->b_rptr;
7793         mblk_t *bmp;
7794         uint_t extra_offset;
7795 
7796         bzero(mhip, sizeof (struct mac_header_info_s));
7797 
7798         mhip->mhi_dsttype = MAC_ADDRTYPE_UNICAST;
7799 
7800         if (ill->ill_sap_length < 0)
7801                 extra_offset = 0;
7802         else
7803                 extra_offset = ill->ill_sap_length;
7804 
7805         mhip->mhi_daddr = (uchar_t *)ind + ind->dl_dest_addr_offset +
7806             extra_offset;
7807         mhip->mhi_saddr = (uchar_t *)ind + ind->dl_src_addr_offset +
7808             extra_offset;
7809 
7810         if (!ind->dl_group_address)
7811                 return;
7812 
7813         /* Multicast or broadcast */
7814         mhip->mhi_dsttype = MAC_ADDRTYPE_MULTICAST;
7815 
7816         if (ind->dl_dest_addr_offset > sizeof (*ind) &&
7817             ind->dl_dest_addr_offset + ind->dl_dest_addr_length < MBLKL(mb) &&
7818             (bmp = ill->ill_bcast_mp) != NULL) {
7819                 dl_unitdata_req_t *dlur;
7820                 uint8_t *bphys_addr;
7821 
7822                 dlur = (dl_unitdata_req_t *)bmp->b_rptr;
7823                 bphys_addr = (uchar_t *)dlur + dlur->dl_dest_addr_offset +
7824                     extra_offset;
7825 
7826                 if (bcmp(mhip->mhi_daddr, bphys_addr,
7827                     ind->dl_dest_addr_length) == 0)
7828                         mhip->mhi_dsttype = MAC_ADDRTYPE_BROADCAST;
7829         }
7830 }
7831 
7832 /*
7833  * This function is used to construct a mac_header_info_s from a
7834  * M_DATA fastpath message from a DLPI driver.
7835  * The address fields in the mhi structure points into the message,
7836  * thus the caller can't use those fields after freeing the message.
7837  *
7838  * We determine whether the packet received is a non-unicast packet
7839  * and in doing so, determine whether or not it is broadcast vs multicast.
7840  * For it to be a broadcast packet, we must have the appropriate mblk_t
7841  * hanging off the ill_t.  If this is either not present or doesn't match
7842  * the destination mac address in the DL_UNITDATA_IND, the packet is deemed
7843  * to be multicast.  Thus NICs that have no broadcast address (or no
7844  * capability for one, such as point to point links) cannot return as
7845  * the packet being broadcast.
7846  */
7847 void
7848 ip_mdata_to_mhi(ill_t *ill, mblk_t *mp, struct mac_header_info_s *mhip)
7849 {
7850         mblk_t *bmp;
7851         struct ether_header *pether;
7852 
7853         bzero(mhip, sizeof (struct mac_header_info_s));
7854 
7855         mhip->mhi_dsttype = MAC_ADDRTYPE_UNICAST;
7856 
7857         pether = (struct ether_header *)((char *)mp->b_rptr
7858             - sizeof (struct ether_header));
7859 
7860         /*
7861          * Make sure the interface is an ethernet type, since we don't
7862          * know the header format for anything but Ethernet. Also make
7863          * sure we are pointing correctly above db_base.
7864          */
7865         if (ill->ill_type != IFT_ETHER)
7866                 return;
7867 
7868 retry:
7869         if ((uchar_t *)pether < mp->b_datap->db_base)
7870                 return;
7871 
7872         /* Is there a VLAN tag? */
7873         if (ill->ill_isv6) {
7874                 if (pether->ether_type != htons(ETHERTYPE_IPV6)) {
7875                         pether = (struct ether_header *)((char *)pether - 4);
7876                         goto retry;
7877                 }
7878         } else {
7879                 if (pether->ether_type != htons(ETHERTYPE_IP)) {
7880                         pether = (struct ether_header *)((char *)pether - 4);
7881                         goto retry;
7882                 }
7883         }
7884         mhip->mhi_daddr = (uchar_t *)&pether->ether_dhost;
7885         mhip->mhi_saddr = (uchar_t *)&pether->ether_shost;
7886 
7887         if (!(mhip->mhi_daddr[0] & 0x01))
7888                 return;
7889 
7890         /* Multicast or broadcast */
7891         mhip->mhi_dsttype = MAC_ADDRTYPE_MULTICAST;
7892 
7893         if ((bmp = ill->ill_bcast_mp) != NULL) {
7894                 dl_unitdata_req_t *dlur;
7895                 uint8_t *bphys_addr;
7896                 uint_t  addrlen;
7897 
7898                 dlur = (dl_unitdata_req_t *)bmp->b_rptr;
7899                 addrlen = dlur->dl_dest_addr_length;
7900                 if (ill->ill_sap_length < 0) {
7901                         bphys_addr = (uchar_t *)dlur +
7902                             dlur->dl_dest_addr_offset;
7903                         addrlen += ill->ill_sap_length;
7904                 } else {
7905                         bphys_addr = (uchar_t *)dlur +
7906                             dlur->dl_dest_addr_offset +
7907                             ill->ill_sap_length;
7908                         addrlen -= ill->ill_sap_length;
7909                 }
7910                 if (bcmp(mhip->mhi_daddr, bphys_addr, addrlen) == 0)
7911                         mhip->mhi_dsttype = MAC_ADDRTYPE_BROADCAST;
7912         }
7913 }
7914 
7915 /*
7916  * Handle anything but M_DATA messages
7917  * We see the DL_UNITDATA_IND which are part
7918  * of the data path, and also the other messages from the driver.
7919  */
7920 void
7921 ip_rput_notdata(ill_t *ill, mblk_t *mp)
7922 {
7923         mblk_t          *first_mp;
7924         struct iocblk   *iocp;
7925         struct mac_header_info_s mhi;
7926 
7927         switch (DB_TYPE(mp)) {
7928         case M_PROTO:
7929         case M_PCPROTO: {
7930                 if (((dl_unitdata_ind_t *)mp->b_rptr)->dl_primitive !=
7931                     DL_UNITDATA_IND) {
7932                         /* Go handle anything other than data elsewhere. */
7933                         ip_rput_dlpi(ill, mp);
7934                         return;
7935                 }
7936 
7937                 first_mp = mp;
7938                 mp = first_mp->b_cont;
7939                 first_mp->b_cont = NULL;
7940 
7941                 if (mp == NULL) {
7942                         freeb(first_mp);
7943                         return;
7944                 }
7945                 ip_dlur_to_mhi(ill, first_mp, &mhi);
7946                 if (ill->ill_isv6)
7947                         ip_input_v6(ill, NULL, mp, &mhi);
7948                 else
7949                         ip_input(ill, NULL, mp, &mhi);
7950 
7951                 /* Ditch the DLPI header. */
7952                 freeb(first_mp);
7953                 return;
7954         }
7955         case M_IOCACK:
7956                 iocp = (struct iocblk *)mp->b_rptr;
7957                 switch (iocp->ioc_cmd) {
7958                 case DL_IOC_HDR_INFO:
7959                         ill_fastpath_ack(ill, mp);
7960                         return;
7961                 default:
7962                         putnext(ill->ill_rq, mp);
7963                         return;
7964                 }
7965                 /* FALLTHROUGH */
7966         case M_ERROR:
7967         case M_HANGUP:
7968                 mutex_enter(&ill->ill_lock);
7969                 if (ill->ill_state_flags & ILL_CONDEMNED) {
7970                         mutex_exit(&ill->ill_lock);
7971                         freemsg(mp);
7972                         return;
7973                 }
7974                 ill_refhold_locked(ill);
7975                 mutex_exit(&ill->ill_lock);
7976                 qwriter_ip(ill, ill->ill_rq, mp, ip_rput_other, CUR_OP,
7977                     B_FALSE);
7978                 return;
7979         case M_CTL:
7980                 putnext(ill->ill_rq, mp);
7981                 return;
7982         case M_IOCNAK:
7983                 ip1dbg(("got iocnak "));
7984                 iocp = (struct iocblk *)mp->b_rptr;
7985                 switch (iocp->ioc_cmd) {
7986                 case DL_IOC_HDR_INFO:
7987                         ip_rput_other(NULL, ill->ill_rq, mp, NULL);
7988                         return;
7989                 default:
7990                         break;
7991                 }
7992                 /* FALLTHROUGH */
7993         default:
7994                 putnext(ill->ill_rq, mp);
7995                 return;
7996         }
7997 }
7998 
7999 /* Read side put procedure.  Packets coming from the wire arrive here. */
8000 int
8001 ip_rput(queue_t *q, mblk_t *mp)
8002 {
8003         ill_t   *ill;
8004         union DL_primitives *dl;
8005 
8006         ill = (ill_t *)q->q_ptr;
8007 
8008         if (ill->ill_state_flags & (ILL_CONDEMNED | ILL_LL_SUBNET_PENDING)) {
8009                 /*
8010                  * If things are opening or closing, only accept high-priority
8011                  * DLPI messages.  (On open ill->ill_ipif has not yet been
8012                  * created; on close, things hanging off the ill may have been
8013                  * freed already.)
8014                  */
8015                 dl = (union DL_primitives *)mp->b_rptr;
8016                 if (DB_TYPE(mp) != M_PCPROTO ||
8017                     dl->dl_primitive == DL_UNITDATA_IND) {
8018                         inet_freemsg(mp);
8019                         return (0);
8020                 }
8021         }
8022         if (DB_TYPE(mp) == M_DATA) {
8023                 struct mac_header_info_s mhi;
8024 
8025                 ip_mdata_to_mhi(ill, mp, &mhi);
8026                 ip_input(ill, NULL, mp, &mhi);
8027         } else {
8028                 ip_rput_notdata(ill, mp);
8029         }
8030         return (0);
8031 }
8032 
8033 /*
8034  * Move the information to a copy.
8035  */
8036 mblk_t *
8037 ip_fix_dbref(mblk_t *mp, ip_recv_attr_t *ira)
8038 {
8039         mblk_t          *mp1;
8040         ill_t           *ill = ira->ira_ill;
8041         ip_stack_t      *ipst = ill->ill_ipst;
8042 
8043         IP_STAT(ipst, ip_db_ref);
8044 
8045         /* Make sure we have ira_l2src before we loose the original mblk */
8046         if (!(ira->ira_flags & IRAF_L2SRC_SET))
8047                 ip_setl2src(mp, ira, ira->ira_rill);
8048 
8049         mp1 = copymsg(mp);
8050         if (mp1 == NULL) {
8051                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
8052                 ip_drop_input("ipIfStatsInDiscards", mp, ill);
8053                 freemsg(mp);
8054                 return (NULL);
8055         }
8056         /* preserve the hardware checksum flags and data, if present */
8057         if (DB_CKSUMFLAGS(mp) != 0) {
8058                 DB_CKSUMFLAGS(mp1) = DB_CKSUMFLAGS(mp);
8059                 DB_CKSUMSTART(mp1) = DB_CKSUMSTART(mp);
8060                 DB_CKSUMSTUFF(mp1) = DB_CKSUMSTUFF(mp);
8061                 DB_CKSUMEND(mp1) = DB_CKSUMEND(mp);
8062                 DB_CKSUM16(mp1) = DB_CKSUM16(mp);
8063         }
8064         freemsg(mp);
8065         return (mp1);
8066 }
8067 
8068 static void
8069 ip_dlpi_error(ill_t *ill, t_uscalar_t prim, t_uscalar_t dl_err,
8070     t_uscalar_t err)
8071 {
8072         if (dl_err == DL_SYSERR) {
8073                 (void) mi_strlog(ill->ill_rq, 1, SL_CONSOLE|SL_ERROR|SL_TRACE,
8074                     "%s: %s failed: DL_SYSERR (errno %u)\n",
8075                     ill->ill_name, dl_primstr(prim), err);
8076                 return;
8077         }
8078 
8079         (void) mi_strlog(ill->ill_rq, 1, SL_CONSOLE|SL_ERROR|SL_TRACE,
8080             "%s: %s failed: %s\n", ill->ill_name, dl_primstr(prim),
8081             dl_errstr(dl_err));
8082 }
8083 
8084 /*
8085  * ip_rput_dlpi is called by ip_rput to handle all DLPI messages other
8086  * than DL_UNITDATA_IND messages. If we need to process this message
8087  * exclusively, we call qwriter_ip, in which case we also need to call
8088  * ill_refhold before that, since qwriter_ip does an ill_refrele.
8089  */
8090 void
8091 ip_rput_dlpi(ill_t *ill, mblk_t *mp)
8092 {
8093         dl_ok_ack_t     *dloa = (dl_ok_ack_t *)mp->b_rptr;
8094         dl_error_ack_t  *dlea = (dl_error_ack_t *)dloa;
8095         queue_t         *q = ill->ill_rq;
8096         t_uscalar_t     prim = dloa->dl_primitive;
8097         t_uscalar_t     reqprim = DL_PRIM_INVAL;
8098 
8099         DTRACE_PROBE3(ill__dlpi, char *, "ip_rput_dlpi",
8100             char *, dl_primstr(prim), ill_t *, ill);
8101         ip1dbg(("ip_rput_dlpi"));
8102 
8103         /*
8104          * If we received an ACK but didn't send a request for it, then it
8105          * can't be part of any pending operation; discard up-front.
8106          */
8107         switch (prim) {
8108         case DL_ERROR_ACK:
8109                 reqprim = dlea->dl_error_primitive;
8110                 ip2dbg(("ip_rput_dlpi(%s): DL_ERROR_ACK for %s (0x%x): %s "
8111                     "(0x%x), unix %u\n", ill->ill_name, dl_primstr(reqprim),
8112                     reqprim, dl_errstr(dlea->dl_errno), dlea->dl_errno,
8113                     dlea->dl_unix_errno));
8114                 break;
8115         case DL_OK_ACK:
8116                 reqprim = dloa->dl_correct_primitive;
8117                 break;
8118         case DL_INFO_ACK:
8119                 reqprim = DL_INFO_REQ;
8120                 break;
8121         case DL_BIND_ACK:
8122                 reqprim = DL_BIND_REQ;
8123                 break;
8124         case DL_PHYS_ADDR_ACK:
8125                 reqprim = DL_PHYS_ADDR_REQ;
8126                 break;
8127         case DL_NOTIFY_ACK:
8128                 reqprim = DL_NOTIFY_REQ;
8129                 break;
8130         case DL_CAPABILITY_ACK:
8131                 reqprim = DL_CAPABILITY_REQ;
8132                 break;
8133         }
8134 
8135         if (prim != DL_NOTIFY_IND) {
8136                 if (reqprim == DL_PRIM_INVAL ||
8137                     !ill_dlpi_pending(ill, reqprim)) {
8138                         /* Not a DLPI message we support or expected */
8139                         freemsg(mp);
8140                         return;
8141                 }
8142                 ip1dbg(("ip_rput: received %s for %s\n", dl_primstr(prim),
8143                     dl_primstr(reqprim)));
8144         }
8145 
8146         switch (reqprim) {
8147         case DL_UNBIND_REQ:
8148                 /*
8149                  * NOTE: we mark the unbind as complete even if we got a
8150                  * DL_ERROR_ACK, since there's not much else we can do.
8151                  */
8152                 mutex_enter(&ill->ill_lock);
8153                 ill->ill_state_flags &= ~ILL_DL_UNBIND_IN_PROGRESS;
8154                 cv_signal(&ill->ill_cv);
8155                 mutex_exit(&ill->ill_lock);
8156                 break;
8157 
8158         case DL_ENABMULTI_REQ:
8159                 if (prim == DL_OK_ACK) {
8160                         if (ill->ill_dlpi_multicast_state == IDS_INPROGRESS)
8161                                 ill->ill_dlpi_multicast_state = IDS_OK;
8162                 }
8163                 break;
8164         }
8165 
8166         /*
8167          * The message is one we're waiting for (or DL_NOTIFY_IND), but we
8168          * need to become writer to continue to process it.  Because an
8169          * exclusive operation doesn't complete until replies to all queued
8170          * DLPI messages have been received, we know we're in the middle of an
8171          * exclusive operation and pass CUR_OP (except for DL_NOTIFY_IND).
8172          *
8173          * As required by qwriter_ip(), we refhold the ill; it will refrele.
8174          * Since this is on the ill stream we unconditionally bump up the
8175          * refcount without doing ILL_CAN_LOOKUP().
8176          */
8177         ill_refhold(ill);
8178         if (prim == DL_NOTIFY_IND)
8179                 qwriter_ip(ill, q, mp, ip_rput_dlpi_writer, NEW_OP, B_FALSE);
8180         else
8181                 qwriter_ip(ill, q, mp, ip_rput_dlpi_writer, CUR_OP, B_FALSE);
8182 }
8183 
8184 /*
8185  * Handling of DLPI messages that require exclusive access to the ipsq.
8186  *
8187  * Need to do ipsq_pending_mp_get on ioctl completion, which could
8188  * happen here. (along with mi_copy_done)
8189  */
8190 /* ARGSUSED */
8191 static void
8192 ip_rput_dlpi_writer(ipsq_t *ipsq, queue_t *q, mblk_t *mp, void *dummy_arg)
8193 {
8194         dl_ok_ack_t     *dloa = (dl_ok_ack_t *)mp->b_rptr;
8195         dl_error_ack_t  *dlea = (dl_error_ack_t *)dloa;
8196         int             err = 0;
8197         ill_t           *ill = (ill_t *)q->q_ptr;
8198         ipif_t          *ipif = NULL;
8199         mblk_t          *mp1 = NULL;
8200         conn_t          *connp = NULL;
8201         t_uscalar_t     paddrreq;
8202         mblk_t          *mp_hw;
8203         boolean_t       ioctl_aborted = B_FALSE;
8204         boolean_t       log = B_TRUE;
8205 
8206         DTRACE_PROBE3(ill__dlpi, char *, "ip_rput_dlpi_writer",
8207             char *, dl_primstr(dloa->dl_primitive), ill_t *, ill);
8208 
8209         ip1dbg(("ip_rput_dlpi_writer .."));
8210         ASSERT(ipsq->ipsq_xop == ill->ill_phyint->phyint_ipsq->ipsq_xop);
8211         ASSERT(IAM_WRITER_ILL(ill));
8212 
8213         ipif = ipsq->ipsq_xop->ipx_pending_ipif;
8214         /*
8215          * The current ioctl could have been aborted by the user and a new
8216          * ioctl to bring up another ill could have started. We could still
8217          * get a response from the driver later.
8218          */
8219         if (ipif != NULL && ipif->ipif_ill != ill)
8220                 ioctl_aborted = B_TRUE;
8221 
8222         switch (dloa->dl_primitive) {
8223         case DL_ERROR_ACK:
8224                 ip1dbg(("ip_rput_dlpi_writer: got DL_ERROR_ACK for %s\n",
8225                     dl_primstr(dlea->dl_error_primitive)));
8226 
8227                 DTRACE_PROBE3(ill__dlpi, char *, "ip_rput_dlpi_writer error",
8228                     char *, dl_primstr(dlea->dl_error_primitive),
8229                     ill_t *, ill);
8230 
8231                 switch (dlea->dl_error_primitive) {
8232                 case DL_DISABMULTI_REQ:
8233                         ill_dlpi_done(ill, dlea->dl_error_primitive);
8234                         break;
8235                 case DL_PROMISCON_REQ:
8236                 case DL_PROMISCOFF_REQ:
8237                 case DL_UNBIND_REQ:
8238                 case DL_ATTACH_REQ:
8239                 case DL_INFO_REQ:
8240                         ill_dlpi_done(ill, dlea->dl_error_primitive);
8241                         break;
8242                 case DL_NOTIFY_REQ:
8243                         ill_dlpi_done(ill, DL_NOTIFY_REQ);
8244                         log = B_FALSE;
8245                         break;
8246                 case DL_PHYS_ADDR_REQ:
8247                         /*
8248                          * For IPv6 only, there are two additional
8249                          * phys_addr_req's sent to the driver to get the
8250                          * IPv6 token and lla. This allows IP to acquire
8251                          * the hardware address format for a given interface
8252                          * without having built in knowledge of the hardware
8253                          * address. ill_phys_addr_pend keeps track of the last
8254                          * DL_PAR sent so we know which response we are
8255                          * dealing with. ill_dlpi_done will update
8256                          * ill_phys_addr_pend when it sends the next req.
8257                          * We don't complete the IOCTL until all three DL_PARs
8258                          * have been attempted, so set *_len to 0 and break.
8259                          */
8260                         paddrreq = ill->ill_phys_addr_pend;
8261                         ill_dlpi_done(ill, DL_PHYS_ADDR_REQ);
8262                         if (paddrreq == DL_IPV6_TOKEN) {
8263                                 ill->ill_token_length = 0;
8264                                 log = B_FALSE;
8265                                 break;
8266                         } else if (paddrreq == DL_IPV6_LINK_LAYER_ADDR) {
8267                                 ill->ill_nd_lla_len = 0;
8268                                 log = B_FALSE;
8269                                 break;
8270                         }
8271                         /*
8272                          * Something went wrong with the DL_PHYS_ADDR_REQ.
8273                          * We presumably have an IOCTL hanging out waiting
8274                          * for completion. Find it and complete the IOCTL
8275                          * with the error noted.
8276                          * However, ill_dl_phys was called on an ill queue
8277                          * (from SIOCSLIFNAME), thus conn_pending_ill is not
8278                          * set. But the ioctl is known to be pending on ill_wq.
8279                          */
8280                         if (!ill->ill_ifname_pending)
8281                                 break;
8282                         ill->ill_ifname_pending = 0;
8283                         if (!ioctl_aborted)
8284                                 mp1 = ipsq_pending_mp_get(ipsq, &connp);
8285                         if (mp1 != NULL) {
8286                                 /*
8287                                  * This operation (SIOCSLIFNAME) must have
8288                                  * happened on the ill. Assert there is no conn
8289                                  */
8290                                 ASSERT(connp == NULL);
8291                                 q = ill->ill_wq;
8292                         }
8293                         break;
8294                 case DL_BIND_REQ:
8295                         ill_dlpi_done(ill, DL_BIND_REQ);
8296                         if (ill->ill_ifname_pending)
8297                                 break;
8298                         mutex_enter(&ill->ill_lock);
8299                         ill->ill_state_flags &= ~ILL_DOWN_IN_PROGRESS;
8300                         mutex_exit(&ill->ill_lock);
8301                         /*
8302                          * Something went wrong with the bind. If this was the
8303                          * result of a DL_NOTE_REPLUMB, then we presumably
8304                          * have an IOCTL hanging out waiting for completion.
8305                          * Find it, take down the interface that was coming
8306                          * up, and complete the IOCTL with the error noted.
8307                          */
8308                         if (!ioctl_aborted)
8309                                 mp1 = ipsq_pending_mp_get(ipsq, &connp);
8310                         if (mp1 != NULL) {
8311                                 /*
8312                                  * This might be a result of a DL_NOTE_REPLUMB
8313                                  * notification. In that case, connp is NULL.
8314                                  */
8315                                 if (connp != NULL)
8316                                         q = CONNP_TO_WQ(connp);
8317 
8318                                 (void) ipif_down(ipif, NULL, NULL);
8319                                 /* error is set below the switch */
8320                         } else {
8321                                 /*
8322                                  * There's no pending IOCTL, so the bind was
8323                                  * most likely started by ill_dl_up(). We save
8324                                  * the error and let it take care of responding
8325                                  * to the IOCTL.
8326                                  */
8327                                 ill->ill_dl_bind_err = dlea->dl_unix_errno ?
8328                                     dlea->dl_unix_errno : ENXIO;
8329                         }
8330                         break;
8331                 case DL_ENABMULTI_REQ:
8332                         ill_dlpi_done(ill, DL_ENABMULTI_REQ);
8333 
8334                         if (ill->ill_dlpi_multicast_state == IDS_INPROGRESS)
8335                                 ill->ill_dlpi_multicast_state = IDS_FAILED;
8336                         if (ill->ill_dlpi_multicast_state == IDS_FAILED) {
8337 
8338                                 printf("ip: joining multicasts failed (%d)"
8339                                     " on %s - will use link layer "
8340                                     "broadcasts for multicast\n",
8341                                     dlea->dl_errno, ill->ill_name);
8342 
8343                                 /*
8344                                  * Set up for multi_bcast; We are the
8345                                  * writer, so ok to access ill->ill_ipif
8346                                  * without any lock.
8347                                  */
8348                                 mutex_enter(&ill->ill_phyint->phyint_lock);
8349                                 ill->ill_phyint->phyint_flags |=
8350                                     PHYI_MULTI_BCAST;
8351                                 mutex_exit(&ill->ill_phyint->phyint_lock);
8352 
8353                         }
8354                         freemsg(mp);    /* Don't want to pass this up */
8355                         return;
8356                 case DL_CAPABILITY_REQ:
8357                         ip1dbg(("ip_rput_dlpi_writer: got DL_ERROR_ACK for "
8358                             "DL_CAPABILITY REQ\n"));
8359                         if (ill->ill_dlpi_capab_state == IDCS_PROBE_SENT)
8360                                 ill->ill_dlpi_capab_state = IDCS_FAILED;
8361                         ill_capability_done(ill);
8362                         freemsg(mp);
8363                         return;
8364                 }
8365                 /*
8366                  * Note the error for IOCTL completion (mp1 is set when
8367                  * ready to complete ioctl). If ill_ifname_pending_err is
8368                  * set, an error occured during plumbing (ill_ifname_pending),
8369                  * so we want to report that error.
8370                  *
8371                  * NOTE: there are two addtional DL_PHYS_ADDR_REQ's
8372                  * (DL_IPV6_TOKEN and DL_IPV6_LINK_LAYER_ADDR) that are
8373                  * expected to get errack'd if the driver doesn't support
8374                  * these flags (e.g. ethernet). log will be set to B_FALSE
8375                  * if these error conditions are encountered.
8376                  */
8377                 if (mp1 != NULL) {
8378                         if (ill->ill_ifname_pending_err != 0)  {
8379                                 err = ill->ill_ifname_pending_err;
8380                                 ill->ill_ifname_pending_err = 0;
8381                         } else {
8382                                 err = dlea->dl_unix_errno ?
8383                                     dlea->dl_unix_errno : ENXIO;
8384                         }
8385                 /*
8386                  * If we're plumbing an interface and an error hasn't already
8387                  * been saved, set ill_ifname_pending_err to the error passed
8388                  * up. Ignore the error if log is B_FALSE (see comment above).
8389                  */
8390                 } else if (log && ill->ill_ifname_pending &&
8391                     ill->ill_ifname_pending_err == 0) {
8392                         ill->ill_ifname_pending_err = dlea->dl_unix_errno ?
8393                             dlea->dl_unix_errno : ENXIO;
8394                 }
8395 
8396                 if (log)
8397                         ip_dlpi_error(ill, dlea->dl_error_primitive,
8398                             dlea->dl_errno, dlea->dl_unix_errno);
8399                 break;
8400         case DL_CAPABILITY_ACK:
8401                 ill_capability_ack(ill, mp);
8402                 /*
8403                  * The message has been handed off to ill_capability_ack
8404                  * and must not be freed below
8405                  */
8406                 mp = NULL;
8407                 break;
8408 
8409         case DL_INFO_ACK:
8410                 /* Call a routine to handle this one. */
8411                 ill_dlpi_done(ill, DL_INFO_REQ);
8412                 ip_ll_subnet_defaults(ill, mp);
8413                 ASSERT(!MUTEX_HELD(&ill->ill_phyint->phyint_ipsq->ipsq_lock));
8414                 return;
8415         case DL_BIND_ACK:
8416                 /*
8417                  * We should have an IOCTL waiting on this unless
8418                  * sent by ill_dl_phys, in which case just return
8419                  */
8420                 ill_dlpi_done(ill, DL_BIND_REQ);
8421 
8422                 if (ill->ill_ifname_pending) {
8423                         DTRACE_PROBE2(ip__rput__dlpi__ifname__pending,
8424                             ill_t *, ill, mblk_t *, mp);
8425                         break;
8426                 }
8427                 mutex_enter(&ill->ill_lock);
8428                 ill->ill_dl_up = 1;
8429                 ill->ill_state_flags &= ~ILL_DOWN_IN_PROGRESS;
8430                 mutex_exit(&ill->ill_lock);
8431 
8432                 if (!ioctl_aborted)
8433                         mp1 = ipsq_pending_mp_get(ipsq, &connp);
8434                 if (mp1 == NULL) {
8435                         DTRACE_PROBE1(ip__rput__dlpi__no__mblk, ill_t *, ill);
8436                         break;
8437                 }
8438                 /*
8439                  * mp1 was added by ill_dl_up(). if that is a result of
8440                  * a DL_NOTE_REPLUMB notification, connp could be NULL.
8441                  */
8442                 if (connp != NULL)
8443                         q = CONNP_TO_WQ(connp);
8444                 /*
8445                  * We are exclusive. So nothing can change even after
8446                  * we get the pending mp.
8447                  */
8448                 ip1dbg(("ip_rput_dlpi: bind_ack %s\n", ill->ill_name));
8449                 DTRACE_PROBE1(ip__rput__dlpi__bind__ack, ill_t *, ill);
8450                 ill_nic_event_dispatch(ill, 0, NE_UP, NULL, 0);
8451 
8452                 if (ill->ill_up_ipifs) {
8453                         err = ill_up_ipifs(ill, q, mp1);
8454                         if (err == EINPROGRESS) {
8455                                 freemsg(mp);
8456                                 return;
8457                         }
8458                 }
8459 
8460                 break;
8461 
8462         case DL_NOTIFY_IND: {
8463                 dl_notify_ind_t *notify = (dl_notify_ind_t *)mp->b_rptr;
8464                 uint_t orig_mtu, orig_mc_mtu;
8465 
8466                 switch (notify->dl_notification) {
8467                 case DL_NOTE_PHYS_ADDR:
8468                         err = ill_set_phys_addr(ill, mp);
8469                         break;
8470 
8471                 case DL_NOTE_REPLUMB:
8472                         /*
8473                          * Directly return after calling ill_replumb().
8474                          * Note that we should not free mp as it is reused
8475                          * in the ill_replumb() function.
8476                          */
8477                         err = ill_replumb(ill, mp);
8478                         return;
8479 
8480                 case DL_NOTE_FASTPATH_FLUSH:
8481                         nce_flush(ill, B_FALSE);
8482                         break;
8483 
8484                 case DL_NOTE_SDU_SIZE:
8485                 case DL_NOTE_SDU_SIZE2:
8486                         /*
8487                          * The dce and fragmentation code can cope with
8488                          * this changing while packets are being sent.
8489                          * When packets are sent ip_output will discover
8490                          * a change.
8491                          *
8492                          * Change the MTU size of the interface.
8493                          */
8494                         mutex_enter(&ill->ill_lock);
8495                         orig_mtu = ill->ill_mtu;
8496                         orig_mc_mtu = ill->ill_mc_mtu;
8497                         switch (notify->dl_notification) {
8498                         case DL_NOTE_SDU_SIZE:
8499                                 ill->ill_current_frag =
8500                                     (uint_t)notify->dl_data;
8501                                 ill->ill_mc_mtu = (uint_t)notify->dl_data;
8502                                 break;
8503                         case DL_NOTE_SDU_SIZE2:
8504                                 ill->ill_current_frag =
8505                                     (uint_t)notify->dl_data1;
8506                                 ill->ill_mc_mtu = (uint_t)notify->dl_data2;
8507                                 break;
8508                         }
8509                         if (ill->ill_current_frag > ill->ill_max_frag)
8510                                 ill->ill_max_frag = ill->ill_current_frag;
8511 
8512                         if (!(ill->ill_flags & ILLF_FIXEDMTU)) {
8513                                 ill->ill_mtu = ill->ill_current_frag;
8514 
8515                                 /*
8516                                  * If ill_user_mtu was set (via
8517                                  * SIOCSLIFLNKINFO), clamp ill_mtu at it.
8518                                  */
8519                                 if (ill->ill_user_mtu != 0 &&
8520                                     ill->ill_user_mtu < ill->ill_mtu)
8521                                         ill->ill_mtu = ill->ill_user_mtu;
8522 
8523                                 if (ill->ill_user_mtu != 0 &&
8524                                     ill->ill_user_mtu < ill->ill_mc_mtu)
8525                                         ill->ill_mc_mtu = ill->ill_user_mtu;
8526 
8527                                 if (ill->ill_isv6) {
8528                                         if (ill->ill_mtu < IPV6_MIN_MTU)
8529                                                 ill->ill_mtu = IPV6_MIN_MTU;
8530                                         if (ill->ill_mc_mtu < IPV6_MIN_MTU)
8531                                                 ill->ill_mc_mtu = IPV6_MIN_MTU;
8532                                 } else {
8533                                         if (ill->ill_mtu < IP_MIN_MTU)
8534                                                 ill->ill_mtu = IP_MIN_MTU;
8535                                         if (ill->ill_mc_mtu < IP_MIN_MTU)
8536                                                 ill->ill_mc_mtu = IP_MIN_MTU;
8537                                 }
8538                         } else if (ill->ill_mc_mtu > ill->ill_mtu) {
8539                                 ill->ill_mc_mtu = ill->ill_mtu;
8540                         }
8541 
8542                         mutex_exit(&ill->ill_lock);
8543                         /*
8544                          * Make sure all dce_generation checks find out
8545                          * that ill_mtu/ill_mc_mtu has changed.
8546                          */
8547                         if (orig_mtu != ill->ill_mtu ||
8548                             orig_mc_mtu != ill->ill_mc_mtu) {
8549                                 dce_increment_all_generations(ill->ill_isv6,
8550                                     ill->ill_ipst);
8551                         }
8552 
8553                         /*
8554                          * Refresh IPMP meta-interface MTU if necessary.
8555                          */
8556                         if (IS_UNDER_IPMP(ill))
8557                                 ipmp_illgrp_refresh_mtu(ill->ill_grp);
8558                         break;
8559 
8560                 case DL_NOTE_LINK_UP:
8561                 case DL_NOTE_LINK_DOWN: {
8562                         /*
8563                          * We are writer. ill / phyint / ipsq assocs stable.
8564                          * The RUNNING flag reflects the state of the link.
8565                          */
8566                         phyint_t *phyint = ill->ill_phyint;
8567                         uint64_t new_phyint_flags;
8568                         boolean_t changed = B_FALSE;
8569                         boolean_t went_up;
8570 
8571                         went_up = notify->dl_notification == DL_NOTE_LINK_UP;
8572                         mutex_enter(&phyint->phyint_lock);
8573 
8574                         new_phyint_flags = went_up ?
8575                             phyint->phyint_flags | PHYI_RUNNING :
8576                             phyint->phyint_flags & ~PHYI_RUNNING;
8577 
8578                         if (IS_IPMP(ill)) {
8579                                 new_phyint_flags = went_up ?
8580                                     new_phyint_flags & ~PHYI_FAILED :
8581                                     new_phyint_flags | PHYI_FAILED;
8582                         }
8583 
8584                         if (new_phyint_flags != phyint->phyint_flags) {
8585                                 phyint->phyint_flags = new_phyint_flags;
8586                                 changed = B_TRUE;
8587                         }
8588                         mutex_exit(&phyint->phyint_lock);
8589                         /*
8590                          * ill_restart_dad handles the DAD restart and routing
8591                          * socket notification logic.
8592                          */
8593                         if (changed) {
8594                                 ill_restart_dad(phyint->phyint_illv4, went_up);
8595                                 ill_restart_dad(phyint->phyint_illv6, went_up);
8596                         }
8597                         break;
8598                 }
8599                 case DL_NOTE_PROMISC_ON_PHYS: {
8600                         phyint_t *phyint = ill->ill_phyint;
8601 
8602                         mutex_enter(&phyint->phyint_lock);
8603                         phyint->phyint_flags |= PHYI_PROMISC;
8604                         mutex_exit(&phyint->phyint_lock);
8605                         break;
8606                 }
8607                 case DL_NOTE_PROMISC_OFF_PHYS: {
8608                         phyint_t *phyint = ill->ill_phyint;
8609 
8610                         mutex_enter(&phyint->phyint_lock);
8611                         phyint->phyint_flags &= ~PHYI_PROMISC;
8612                         mutex_exit(&phyint->phyint_lock);
8613                         break;
8614                 }
8615                 case DL_NOTE_CAPAB_RENEG:
8616                         /*
8617                          * Something changed on the driver side.
8618                          * It wants us to renegotiate the capabilities
8619                          * on this ill. One possible cause is the aggregation
8620                          * interface under us where a port got added or
8621                          * went away.
8622                          *
8623                          * If the capability negotiation is already done
8624                          * or is in progress, reset the capabilities and
8625                          * mark the ill's ill_capab_reneg to be B_TRUE,
8626                          * so that when the ack comes back, we can start
8627                          * the renegotiation process.
8628                          *
8629                          * Note that if ill_capab_reneg is already B_TRUE
8630                          * (ill_dlpi_capab_state is IDS_UNKNOWN in this case),
8631                          * the capability resetting request has been sent
8632                          * and the renegotiation has not been started yet;
8633                          * nothing needs to be done in this case.
8634                          */
8635                         ipsq_current_start(ipsq, ill->ill_ipif, 0);
8636                         ill_capability_reset(ill, B_TRUE);
8637                         ipsq_current_finish(ipsq);
8638                         break;
8639 
8640                 case DL_NOTE_ALLOWED_IPS:
8641                         ill_set_allowed_ips(ill, mp);
8642                         break;
8643                 default:
8644                         ip0dbg(("ip_rput_dlpi_writer: unknown notification "
8645                             "type 0x%x for DL_NOTIFY_IND\n",
8646                             notify->dl_notification));
8647                         break;
8648                 }
8649 
8650                 /*
8651                  * As this is an asynchronous operation, we
8652                  * should not call ill_dlpi_done
8653                  */
8654                 break;
8655         }
8656         case DL_NOTIFY_ACK: {
8657                 dl_notify_ack_t *noteack = (dl_notify_ack_t *)mp->b_rptr;
8658 
8659                 if (noteack->dl_notifications & DL_NOTE_LINK_UP)
8660                         ill->ill_note_link = 1;
8661                 ill_dlpi_done(ill, DL_NOTIFY_REQ);
8662                 break;
8663         }
8664         case DL_PHYS_ADDR_ACK: {
8665                 /*
8666                  * As part of plumbing the interface via SIOCSLIFNAME,
8667                  * ill_dl_phys() will queue a series of DL_PHYS_ADDR_REQs,
8668                  * whose answers we receive here.  As each answer is received,
8669                  * we call ill_dlpi_done() to dispatch the next request as
8670                  * we're processing the current one.  Once all answers have
8671                  * been received, we use ipsq_pending_mp_get() to dequeue the
8672                  * outstanding IOCTL and reply to it.  (Because ill_dl_phys()
8673                  * is invoked from an ill queue, conn_oper_pending_ill is not
8674                  * available, but we know the ioctl is pending on ill_wq.)
8675                  */
8676                 uint_t  paddrlen, paddroff;
8677                 uint8_t *addr;
8678 
8679                 paddrreq = ill->ill_phys_addr_pend;
8680                 paddrlen = ((dl_phys_addr_ack_t *)mp->b_rptr)->dl_addr_length;
8681                 paddroff = ((dl_phys_addr_ack_t *)mp->b_rptr)->dl_addr_offset;
8682                 addr = mp->b_rptr + paddroff;
8683 
8684                 ill_dlpi_done(ill, DL_PHYS_ADDR_REQ);
8685                 if (paddrreq == DL_IPV6_TOKEN) {
8686                         /*
8687                          * bcopy to low-order bits of ill_token
8688                          *
8689                          * XXX Temporary hack - currently, all known tokens
8690                          * are 64 bits, so I'll cheat for the moment.
8691                          */
8692                         bcopy(addr, &ill->ill_token.s6_addr32[2], paddrlen);
8693                         ill->ill_token_length = paddrlen;
8694                         break;
8695                 } else if (paddrreq == DL_IPV6_LINK_LAYER_ADDR) {
8696                         ASSERT(ill->ill_nd_lla_mp == NULL);
8697                         ill_set_ndmp(ill, mp, paddroff, paddrlen);
8698                         mp = NULL;
8699                         break;
8700                 } else if (paddrreq == DL_CURR_DEST_ADDR) {
8701                         ASSERT(ill->ill_dest_addr_mp == NULL);
8702                         ill->ill_dest_addr_mp = mp;
8703                         ill->ill_dest_addr = addr;
8704                         mp = NULL;
8705                         if (ill->ill_isv6) {
8706                                 ill_setdesttoken(ill);
8707                                 ipif_setdestlinklocal(ill->ill_ipif);
8708                         }
8709                         break;
8710                 }
8711 
8712                 ASSERT(paddrreq == DL_CURR_PHYS_ADDR);
8713                 ASSERT(ill->ill_phys_addr_mp == NULL);
8714                 if (!ill->ill_ifname_pending)
8715                         break;
8716                 ill->ill_ifname_pending = 0;
8717                 if (!ioctl_aborted)
8718                         mp1 = ipsq_pending_mp_get(ipsq, &connp);
8719                 if (mp1 != NULL) {
8720                         ASSERT(connp == NULL);
8721                         q = ill->ill_wq;
8722                 }
8723                 /*
8724                  * If any error acks received during the plumbing sequence,
8725                  * ill_ifname_pending_err will be set. Break out and send up
8726                  * the error to the pending ioctl.
8727                  */
8728                 if (ill->ill_ifname_pending_err != 0) {
8729                         err = ill->ill_ifname_pending_err;
8730                         ill->ill_ifname_pending_err = 0;
8731                         break;
8732                 }
8733 
8734                 ill->ill_phys_addr_mp = mp;
8735                 ill->ill_phys_addr = (paddrlen == 0 ? NULL : addr);
8736                 mp = NULL;
8737 
8738                 /*
8739                  * If paddrlen or ill_phys_addr_length is zero, the DLPI
8740                  * provider doesn't support physical addresses.  We check both
8741                  * paddrlen and ill_phys_addr_length because sppp (PPP) does
8742                  * not have physical addresses, but historically adversises a
8743                  * physical address length of 0 in its DL_INFO_ACK, but 6 in
8744                  * its DL_PHYS_ADDR_ACK.
8745                  */
8746                 if (paddrlen == 0 || ill->ill_phys_addr_length == 0) {
8747                         ill->ill_phys_addr = NULL;
8748                 } else if (paddrlen != ill->ill_phys_addr_length) {
8749                         ip0dbg(("DL_PHYS_ADDR_ACK: got addrlen %d, expected %d",
8750                             paddrlen, ill->ill_phys_addr_length));
8751                         err = EINVAL;
8752                         break;
8753                 }
8754 
8755                 if (ill->ill_nd_lla_mp == NULL) {
8756                         if ((mp_hw = copyb(ill->ill_phys_addr_mp)) == NULL) {
8757                                 err = ENOMEM;
8758                                 break;
8759                         }
8760                         ill_set_ndmp(ill, mp_hw, paddroff, paddrlen);
8761                 }
8762 
8763                 if (ill->ill_isv6) {
8764                         ill_setdefaulttoken(ill);
8765                         ipif_setlinklocal(ill->ill_ipif);
8766                 }
8767                 break;
8768         }
8769         case DL_OK_ACK:
8770                 ip2dbg(("DL_OK_ACK %s (0x%x)\n",
8771                     dl_primstr((int)dloa->dl_correct_primitive),
8772                     dloa->dl_correct_primitive));
8773                 DTRACE_PROBE3(ill__dlpi, char *, "ip_rput_dlpi_writer ok",
8774                     char *, dl_primstr(dloa->dl_correct_primitive),
8775                     ill_t *, ill);
8776 
8777                 switch (dloa->dl_correct_primitive) {
8778                 case DL_ENABMULTI_REQ:
8779                 case DL_DISABMULTI_REQ:
8780                         ill_dlpi_done(ill, dloa->dl_correct_primitive);
8781                         break;
8782                 case DL_PROMISCON_REQ:
8783                 case DL_PROMISCOFF_REQ:
8784                 case DL_UNBIND_REQ:
8785                 case DL_ATTACH_REQ:
8786                         ill_dlpi_done(ill, dloa->dl_correct_primitive);
8787                         break;
8788                 }
8789                 break;
8790         default:
8791                 break;
8792         }
8793 
8794         freemsg(mp);
8795         if (mp1 == NULL)
8796                 return;
8797 
8798         /*
8799          * The operation must complete without EINPROGRESS since
8800          * ipsq_pending_mp_get() has removed the mblk (mp1).  Otherwise,
8801          * the operation will be stuck forever inside the IPSQ.
8802          */
8803         ASSERT(err != EINPROGRESS);
8804 
8805         DTRACE_PROBE4(ipif__ioctl, char *, "ip_rput_dlpi_writer finish",
8806             int, ipsq->ipsq_xop->ipx_current_ioctl, ill_t *, ill,
8807             ipif_t *, NULL);
8808 
8809         switch (ipsq->ipsq_xop->ipx_current_ioctl) {
8810         case 0:
8811                 ipsq_current_finish(ipsq);
8812                 break;
8813 
8814         case SIOCSLIFNAME:
8815         case IF_UNITSEL: {
8816                 ill_t *ill_other = ILL_OTHER(ill);
8817 
8818                 /*
8819                  * If SIOCSLIFNAME or IF_UNITSEL is about to succeed, and the
8820                  * ill has a peer which is in an IPMP group, then place ill
8821                  * into the same group.  One catch: although ifconfig plumbs
8822                  * the appropriate IPMP meta-interface prior to plumbing this
8823                  * ill, it is possible for multiple ifconfig applications to
8824                  * race (or for another application to adjust plumbing), in
8825                  * which case the IPMP meta-interface we need will be missing.
8826                  * If so, kick the phyint out of the group.
8827                  */
8828                 if (err == 0 && ill_other != NULL && IS_UNDER_IPMP(ill_other)) {
8829                         ipmp_grp_t      *grp = ill->ill_phyint->phyint_grp;
8830                         ipmp_illgrp_t   *illg;
8831 
8832                         illg = ill->ill_isv6 ? grp->gr_v6 : grp->gr_v4;
8833                         if (illg == NULL)
8834                                 ipmp_phyint_leave_grp(ill->ill_phyint);
8835                         else
8836                                 ipmp_ill_join_illgrp(ill, illg);
8837                 }
8838 
8839                 if (ipsq->ipsq_xop->ipx_current_ioctl == IF_UNITSEL)
8840                         ip_ioctl_finish(q, mp1, err, NO_COPYOUT, ipsq);
8841                 else
8842                         ip_ioctl_finish(q, mp1, err, COPYOUT, ipsq);
8843                 break;
8844         }
8845         case SIOCLIFADDIF:
8846                 ip_ioctl_finish(q, mp1, err, COPYOUT, ipsq);
8847                 break;
8848 
8849         default:
8850                 ip_ioctl_finish(q, mp1, err, NO_COPYOUT, ipsq);
8851                 break;
8852         }
8853 }
8854 
8855 /*
8856  * ip_rput_other is called by ip_rput to handle messages modifying the global
8857  * state in IP.  If 'ipsq' is non-NULL, caller is writer on it.
8858  */
8859 /* ARGSUSED */
8860 void
8861 ip_rput_other(ipsq_t *ipsq, queue_t *q, mblk_t *mp, void *dummy_arg)
8862 {
8863         ill_t           *ill = q->q_ptr;
8864         struct iocblk   *iocp;
8865 
8866         ip1dbg(("ip_rput_other "));
8867         if (ipsq != NULL) {
8868                 ASSERT(IAM_WRITER_IPSQ(ipsq));
8869                 ASSERT(ipsq->ipsq_xop ==
8870                     ill->ill_phyint->phyint_ipsq->ipsq_xop);
8871         }
8872 
8873         switch (mp->b_datap->db_type) {
8874         case M_ERROR:
8875         case M_HANGUP:
8876                 /*
8877                  * The device has a problem.  We force the ILL down.  It can
8878                  * be brought up again manually using SIOCSIFFLAGS (via
8879                  * ifconfig or equivalent).
8880                  */
8881                 ASSERT(ipsq != NULL);
8882                 if (mp->b_rptr < mp->b_wptr)
8883                         ill->ill_error = (int)(*mp->b_rptr & 0xFF);
8884                 if (ill->ill_error == 0)
8885                         ill->ill_error = ENXIO;
8886                 if (!ill_down_start(q, mp))
8887                         return;
8888                 ipif_all_down_tail(ipsq, q, mp, NULL);
8889                 break;
8890         case M_IOCNAK: {
8891                 iocp = (struct iocblk *)mp->b_rptr;
8892 
8893                 ASSERT(iocp->ioc_cmd == DL_IOC_HDR_INFO);
8894                 /*
8895                  * If this was the first attempt, turn off the fastpath
8896                  * probing.
8897                  */
8898                 mutex_enter(&ill->ill_lock);
8899                 if (ill->ill_dlpi_fastpath_state == IDS_INPROGRESS) {
8900                         ill->ill_dlpi_fastpath_state = IDS_FAILED;
8901                         mutex_exit(&ill->ill_lock);
8902                         /*
8903                          * don't flush the nce_t entries: we use them
8904                          * as an index to the ncec itself.
8905                          */
8906                         ip1dbg(("ip_rput: DLPI fastpath off on interface %s\n",
8907                             ill->ill_name));
8908                 } else {
8909                         mutex_exit(&ill->ill_lock);
8910                 }
8911                 freemsg(mp);
8912                 break;
8913         }
8914         default:
8915                 ASSERT(0);
8916                 break;
8917         }
8918 }
8919 
8920 /*
8921  * Update any source route, record route or timestamp options
8922  * When it fails it has consumed the message and BUMPed the MIB.
8923  */
8924 boolean_t
8925 ip_forward_options(mblk_t *mp, ipha_t *ipha, ill_t *dst_ill,
8926     ip_recv_attr_t *ira)
8927 {
8928         ipoptp_t        opts;
8929         uchar_t         *opt;
8930         uint8_t         optval;
8931         uint8_t         optlen;
8932         ipaddr_t        dst;
8933         ipaddr_t        ifaddr;
8934         uint32_t        ts;
8935         timestruc_t     now;
8936         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
8937 
8938         ip2dbg(("ip_forward_options\n"));
8939         dst = ipha->ipha_dst;
8940         for (optval = ipoptp_first(&opts, ipha);
8941             optval != IPOPT_EOL;
8942             optval = ipoptp_next(&opts)) {
8943                 ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
8944                 opt = opts.ipoptp_cur;
8945                 optlen = opts.ipoptp_len;
8946                 ip2dbg(("ip_forward_options: opt %d, len %d\n",
8947                     optval, opts.ipoptp_len));
8948                 switch (optval) {
8949                         uint32_t off;
8950                 case IPOPT_SSRR:
8951                 case IPOPT_LSRR:
8952                         /* Check if adminstratively disabled */
8953                         if (!ipst->ips_ip_forward_src_routed) {
8954                                 BUMP_MIB(dst_ill->ill_ip_mib,
8955                                     ipIfStatsForwProhibits);
8956                                 ip_drop_input("ICMP_SOURCE_ROUTE_FAILED",
8957                                     mp, dst_ill);
8958                                 icmp_unreachable(mp, ICMP_SOURCE_ROUTE_FAILED,
8959                                     ira);
8960                                 return (B_FALSE);
8961                         }
8962                         if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
8963                                 /*
8964                                  * Must be partial since ip_input_options
8965                                  * checked for strict.
8966                                  */
8967                                 break;
8968                         }
8969                         off = opt[IPOPT_OFFSET];
8970                         off--;
8971                 redo_srr:
8972                         if (optlen < IP_ADDR_LEN ||
8973                             off > optlen - IP_ADDR_LEN) {
8974                                 /* End of source route */
8975                                 ip1dbg((
8976                                     "ip_forward_options: end of SR\n"));
8977                                 break;
8978                         }
8979                         /* Pick a reasonable address on the outbound if */
8980                         ASSERT(dst_ill != NULL);
8981                         if (ip_select_source_v4(dst_ill, INADDR_ANY, dst,
8982                             INADDR_ANY, ALL_ZONES, ipst, &ifaddr, NULL,
8983                             NULL) != 0) {
8984                                 /* No source! Shouldn't happen */
8985                                 ifaddr = INADDR_ANY;
8986                         }
8987                         bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
8988                         bcopy(&ifaddr, (char *)opt + off, IP_ADDR_LEN);
8989                         ip1dbg(("ip_forward_options: next hop 0x%x\n",
8990                             ntohl(dst)));
8991 
8992                         /*
8993                          * Check if our address is present more than
8994                          * once as consecutive hops in source route.
8995                          */
8996                         if (ip_type_v4(dst, ipst) == IRE_LOCAL) {
8997                                 off += IP_ADDR_LEN;
8998                                 opt[IPOPT_OFFSET] += IP_ADDR_LEN;
8999                                 goto redo_srr;
9000                         }
9001                         ipha->ipha_dst = dst;
9002                         opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9003                         break;
9004                 case IPOPT_RR:
9005                         off = opt[IPOPT_OFFSET];
9006                         off--;
9007                         if (optlen < IP_ADDR_LEN ||
9008                             off > optlen - IP_ADDR_LEN) {
9009                                 /* No more room - ignore */
9010                                 ip1dbg((
9011                                     "ip_forward_options: end of RR\n"));
9012                                 break;
9013                         }
9014                         /* Pick a reasonable address on the outbound if */
9015                         ASSERT(dst_ill != NULL);
9016                         if (ip_select_source_v4(dst_ill, INADDR_ANY, dst,
9017                             INADDR_ANY, ALL_ZONES, ipst, &ifaddr, NULL,
9018                             NULL) != 0) {
9019                                 /* No source! Shouldn't happen */
9020                                 ifaddr = INADDR_ANY;
9021                         }
9022                         bcopy(&ifaddr, (char *)opt + off, IP_ADDR_LEN);
9023                         opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9024                         break;
9025                 case IPOPT_TS:
9026                         /* Insert timestamp if there is room */
9027                         switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
9028                         case IPOPT_TS_TSONLY:
9029                                 off = IPOPT_TS_TIMELEN;
9030                                 break;
9031                         case IPOPT_TS_PRESPEC:
9032                         case IPOPT_TS_PRESPEC_RFC791:
9033                                 /* Verify that the address matched */
9034                                 off = opt[IPOPT_OFFSET] - 1;
9035                                 bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
9036                                 if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
9037                                         /* Not for us */
9038                                         break;
9039                                 }
9040                                 /* FALLTHROUGH */
9041                         case IPOPT_TS_TSANDADDR:
9042                                 off = IP_ADDR_LEN + IPOPT_TS_TIMELEN;
9043                                 break;
9044                         default:
9045                                 /*
9046                                  * ip_*put_options should have already
9047                                  * dropped this packet.
9048                                  */
9049                                 cmn_err(CE_PANIC, "ip_forward_options: "
9050                                     "unknown IT - bug in ip_input_options?\n");
9051                         }
9052                         if (opt[IPOPT_OFFSET] - 1 + off > optlen) {
9053                                 /* Increase overflow counter */
9054                                 off = (opt[IPOPT_POS_OV_FLG] >> 4) + 1;
9055                                 opt[IPOPT_POS_OV_FLG] =
9056                                     (uint8_t)((opt[IPOPT_POS_OV_FLG] & 0x0F) |
9057                                     (off << 4));
9058                                 break;
9059                         }
9060                         off = opt[IPOPT_OFFSET] - 1;
9061                         switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
9062                         case IPOPT_TS_PRESPEC:
9063                         case IPOPT_TS_PRESPEC_RFC791:
9064                         case IPOPT_TS_TSANDADDR:
9065                                 /* Pick a reasonable addr on the outbound if */
9066                                 ASSERT(dst_ill != NULL);
9067                                 if (ip_select_source_v4(dst_ill, INADDR_ANY,
9068                                     dst, INADDR_ANY, ALL_ZONES, ipst, &ifaddr,
9069                                     NULL, NULL) != 0) {
9070                                         /* No source! Shouldn't happen */
9071                                         ifaddr = INADDR_ANY;
9072                                 }
9073                                 bcopy(&ifaddr, (char *)opt + off, IP_ADDR_LEN);
9074                                 opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9075                                 /* FALLTHROUGH */
9076                         case IPOPT_TS_TSONLY:
9077                                 off = opt[IPOPT_OFFSET] - 1;
9078                                 /* Compute # of milliseconds since midnight */
9079                                 gethrestime(&now);
9080                                 ts = (now.tv_sec % (24 * 60 * 60)) * 1000 +
9081                                     NSEC2MSEC(now.tv_nsec);
9082                                 bcopy(&ts, (char *)opt + off, IPOPT_TS_TIMELEN);
9083                                 opt[IPOPT_OFFSET] += IPOPT_TS_TIMELEN;
9084                                 break;
9085                         }
9086                         break;
9087                 }
9088         }
9089         return (B_TRUE);
9090 }
9091 
9092 /*
9093  * Call ill_frag_timeout to do garbage collection. ill_frag_timeout
9094  * returns 'true' if there are still fragments left on the queue, in
9095  * which case we restart the timer.
9096  */
9097 void
9098 ill_frag_timer(void *arg)
9099 {
9100         ill_t   *ill = (ill_t *)arg;
9101         boolean_t frag_pending;
9102         ip_stack_t *ipst = ill->ill_ipst;
9103         time_t  timeout;
9104 
9105         mutex_enter(&ill->ill_lock);
9106         ASSERT(!ill->ill_fragtimer_executing);
9107         if (ill->ill_state_flags & ILL_CONDEMNED) {
9108                 ill->ill_frag_timer_id = 0;
9109                 mutex_exit(&ill->ill_lock);
9110                 return;
9111         }
9112         ill->ill_fragtimer_executing = 1;
9113         mutex_exit(&ill->ill_lock);
9114 
9115         timeout = (ill->ill_isv6 ? ipst->ips_ipv6_reassembly_timeout :
9116             ipst->ips_ip_reassembly_timeout);
9117 
9118         frag_pending = ill_frag_timeout(ill, timeout);
9119 
9120         /*
9121          * Restart the timer, if we have fragments pending or if someone
9122          * wanted us to be scheduled again.
9123          */
9124         mutex_enter(&ill->ill_lock);
9125         ill->ill_fragtimer_executing = 0;
9126         ill->ill_frag_timer_id = 0;
9127         if (frag_pending || ill->ill_fragtimer_needrestart)
9128                 ill_frag_timer_start(ill);
9129         mutex_exit(&ill->ill_lock);
9130 }
9131 
9132 void
9133 ill_frag_timer_start(ill_t *ill)
9134 {
9135         ip_stack_t *ipst = ill->ill_ipst;
9136         clock_t timeo_ms;
9137 
9138         ASSERT(MUTEX_HELD(&ill->ill_lock));
9139 
9140         /* If the ill is closing or opening don't proceed */
9141         if (ill->ill_state_flags & ILL_CONDEMNED)
9142                 return;
9143 
9144         if (ill->ill_fragtimer_executing) {
9145                 /*
9146                  * ill_frag_timer is currently executing. Just record the
9147                  * the fact that we want the timer to be restarted.
9148                  * ill_frag_timer will post a timeout before it returns,
9149                  * ensuring it will be called again.
9150                  */
9151                 ill->ill_fragtimer_needrestart = 1;
9152                 return;
9153         }
9154 
9155         if (ill->ill_frag_timer_id == 0) {
9156                 timeo_ms = (ill->ill_isv6 ? ipst->ips_ipv6_reassembly_timeout :
9157                     ipst->ips_ip_reassembly_timeout) * SECONDS;
9158 
9159                 /*
9160                  * The timer is neither running nor is the timeout handler
9161                  * executing. Post a timeout so that ill_frag_timer will be
9162                  * called
9163                  */
9164                 ill->ill_frag_timer_id = timeout(ill_frag_timer, ill,
9165                     MSEC_TO_TICK(timeo_ms >> 1));
9166                 ill->ill_fragtimer_needrestart = 0;
9167         }
9168 }
9169 
9170 /*
9171  * Update any source route, record route or timestamp options.
9172  * Check that we are at end of strict source route.
9173  * The options have already been checked for sanity in ip_input_options().
9174  */
9175 boolean_t
9176 ip_input_local_options(mblk_t *mp, ipha_t *ipha, ip_recv_attr_t *ira)
9177 {
9178         ipoptp_t        opts;
9179         uchar_t         *opt;
9180         uint8_t         optval;
9181         uint8_t         optlen;
9182         ipaddr_t        dst;
9183         ipaddr_t        ifaddr;
9184         uint32_t        ts;
9185         timestruc_t     now;
9186         ill_t           *ill = ira->ira_ill;
9187         ip_stack_t      *ipst = ill->ill_ipst;
9188 
9189         ip2dbg(("ip_input_local_options\n"));
9190 
9191         for (optval = ipoptp_first(&opts, ipha);
9192             optval != IPOPT_EOL;
9193             optval = ipoptp_next(&opts)) {
9194                 ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
9195                 opt = opts.ipoptp_cur;
9196                 optlen = opts.ipoptp_len;
9197                 ip2dbg(("ip_input_local_options: opt %d, len %d\n",
9198                     optval, optlen));
9199                 switch (optval) {
9200                         uint32_t off;
9201                 case IPOPT_SSRR:
9202                 case IPOPT_LSRR:
9203                         off = opt[IPOPT_OFFSET];
9204                         off--;
9205                         if (optlen < IP_ADDR_LEN ||
9206                             off > optlen - IP_ADDR_LEN) {
9207                                 /* End of source route */
9208                                 ip1dbg(("ip_input_local_options: end of SR\n"));
9209                                 break;
9210                         }
9211                         /*
9212                          * This will only happen if two consecutive entries
9213                          * in the source route contains our address or if
9214                          * it is a packet with a loose source route which
9215                          * reaches us before consuming the whole source route
9216                          */
9217                         ip1dbg(("ip_input_local_options: not end of SR\n"));
9218                         if (optval == IPOPT_SSRR) {
9219                                 goto bad_src_route;
9220                         }
9221                         /*
9222                          * Hack: instead of dropping the packet truncate the
9223                          * source route to what has been used by filling the
9224                          * rest with IPOPT_NOP.
9225                          */
9226                         opt[IPOPT_OLEN] = (uint8_t)off;
9227                         while (off < optlen) {
9228                                 opt[off++] = IPOPT_NOP;
9229                         }
9230                         break;
9231                 case IPOPT_RR:
9232                         off = opt[IPOPT_OFFSET];
9233                         off--;
9234                         if (optlen < IP_ADDR_LEN ||
9235                             off > optlen - IP_ADDR_LEN) {
9236                                 /* No more room - ignore */
9237                                 ip1dbg((
9238                                     "ip_input_local_options: end of RR\n"));
9239                                 break;
9240                         }
9241                         /* Pick a reasonable address on the outbound if */
9242                         if (ip_select_source_v4(ill, INADDR_ANY, ipha->ipha_dst,
9243                             INADDR_ANY, ALL_ZONES, ipst, &ifaddr, NULL,
9244                             NULL) != 0) {
9245                                 /* No source! Shouldn't happen */
9246                                 ifaddr = INADDR_ANY;
9247                         }
9248                         bcopy(&ifaddr, (char *)opt + off, IP_ADDR_LEN);
9249                         opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9250                         break;
9251                 case IPOPT_TS:
9252                         /* Insert timestamp if there is romm */
9253                         switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
9254                         case IPOPT_TS_TSONLY:
9255                                 off = IPOPT_TS_TIMELEN;
9256                                 break;
9257                         case IPOPT_TS_PRESPEC:
9258                         case IPOPT_TS_PRESPEC_RFC791:
9259                                 /* Verify that the address matched */
9260                                 off = opt[IPOPT_OFFSET] - 1;
9261                                 bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
9262                                 if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
9263                                         /* Not for us */
9264                                         break;
9265                                 }
9266                                 /* FALLTHROUGH */
9267                         case IPOPT_TS_TSANDADDR:
9268                                 off = IP_ADDR_LEN + IPOPT_TS_TIMELEN;
9269                                 break;
9270                         default:
9271                                 /*
9272                                  * ip_*put_options should have already
9273                                  * dropped this packet.
9274                                  */
9275                                 cmn_err(CE_PANIC, "ip_input_local_options: "
9276                                     "unknown IT - bug in ip_input_options?\n");
9277                         }
9278                         if (opt[IPOPT_OFFSET] - 1 + off > optlen) {
9279                                 /* Increase overflow counter */
9280                                 off = (opt[IPOPT_POS_OV_FLG] >> 4) + 1;
9281                                 opt[IPOPT_POS_OV_FLG] =
9282                                     (uint8_t)((opt[IPOPT_POS_OV_FLG] & 0x0F) |
9283                                     (off << 4));
9284                                 break;
9285                         }
9286                         off = opt[IPOPT_OFFSET] - 1;
9287                         switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
9288                         case IPOPT_TS_PRESPEC:
9289                         case IPOPT_TS_PRESPEC_RFC791:
9290                         case IPOPT_TS_TSANDADDR:
9291                                 /* Pick a reasonable addr on the outbound if */
9292                                 if (ip_select_source_v4(ill, INADDR_ANY,
9293                                     ipha->ipha_dst, INADDR_ANY, ALL_ZONES, ipst,
9294                                     &ifaddr, NULL, NULL) != 0) {
9295                                         /* No source! Shouldn't happen */
9296                                         ifaddr = INADDR_ANY;
9297                                 }
9298                                 bcopy(&ifaddr, (char *)opt + off, IP_ADDR_LEN);
9299                                 opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9300                                 /* FALLTHROUGH */
9301                         case IPOPT_TS_TSONLY:
9302                                 off = opt[IPOPT_OFFSET] - 1;
9303                                 /* Compute # of milliseconds since midnight */
9304                                 gethrestime(&now);
9305                                 ts = (now.tv_sec % (24 * 60 * 60)) * 1000 +
9306                                     NSEC2MSEC(now.tv_nsec);
9307                                 bcopy(&ts, (char *)opt + off, IPOPT_TS_TIMELEN);
9308                                 opt[IPOPT_OFFSET] += IPOPT_TS_TIMELEN;
9309                                 break;
9310                         }
9311                         break;
9312                 }
9313         }
9314         return (B_TRUE);
9315 
9316 bad_src_route:
9317         /* make sure we clear any indication of a hardware checksum */
9318         DB_CKSUMFLAGS(mp) = 0;
9319         ip_drop_input("ICMP_SOURCE_ROUTE_FAILED", mp, ill);
9320         icmp_unreachable(mp, ICMP_SOURCE_ROUTE_FAILED, ira);
9321         return (B_FALSE);
9322 
9323 }
9324 
9325 /*
9326  * Process IP options in an inbound packet.  Always returns the nexthop.
9327  * Normally this is the passed in nexthop, but if there is an option
9328  * that effects the nexthop (such as a source route) that will be returned.
9329  * Sets *errorp if there is an error, in which case an ICMP error has been sent
9330  * and mp freed.
9331  */
9332 ipaddr_t
9333 ip_input_options(ipha_t *ipha, ipaddr_t dst, mblk_t *mp,
9334     ip_recv_attr_t *ira, int *errorp)
9335 {
9336         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
9337         ipoptp_t        opts;
9338         uchar_t         *opt;
9339         uint8_t         optval;
9340         uint8_t         optlen;
9341         intptr_t        code = 0;
9342         ire_t           *ire;
9343 
9344         ip2dbg(("ip_input_options\n"));
9345         *errorp = 0;
9346         for (optval = ipoptp_first(&opts, ipha);
9347             optval != IPOPT_EOL;
9348             optval = ipoptp_next(&opts)) {
9349                 opt = opts.ipoptp_cur;
9350                 optlen = opts.ipoptp_len;
9351                 ip2dbg(("ip_input_options: opt %d, len %d\n",
9352                     optval, optlen));
9353                 /*
9354                  * Note: we need to verify the checksum before we
9355                  * modify anything thus this routine only extracts the next
9356                  * hop dst from any source route.
9357                  */
9358                 switch (optval) {
9359                         uint32_t off;
9360                 case IPOPT_SSRR:
9361                 case IPOPT_LSRR:
9362                         if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
9363                                 if (optval == IPOPT_SSRR) {
9364                                         ip1dbg(("ip_input_options: not next"
9365                                             " strict source route 0x%x\n",
9366                                             ntohl(dst)));
9367                                         code = (char *)&ipha->ipha_dst -
9368                                             (char *)ipha;
9369                                         goto param_prob; /* RouterReq's */
9370                                 }
9371                                 ip2dbg(("ip_input_options: "
9372                                     "not next source route 0x%x\n",
9373                                     ntohl(dst)));
9374                                 break;
9375                         }
9376 
9377                         if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
9378                                 ip1dbg((
9379                                     "ip_input_options: bad option offset\n"));
9380                                 code = (char *)&opt[IPOPT_OLEN] -
9381                                     (char *)ipha;
9382                                 goto param_prob;
9383                         }
9384                         off = opt[IPOPT_OFFSET];
9385                         off--;
9386                 redo_srr:
9387                         if (optlen < IP_ADDR_LEN ||
9388                             off > optlen - IP_ADDR_LEN) {
9389                                 /* End of source route */
9390                                 ip1dbg(("ip_input_options: end of SR\n"));
9391                                 break;
9392                         }
9393                         bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
9394                         ip1dbg(("ip_input_options: next hop 0x%x\n",
9395                             ntohl(dst)));
9396 
9397                         /*
9398                          * Check if our address is present more than
9399                          * once as consecutive hops in source route.
9400                          * XXX verify per-interface ip_forwarding
9401                          * for source route?
9402                          */
9403                         if (ip_type_v4(dst, ipst) == IRE_LOCAL) {
9404                                 off += IP_ADDR_LEN;
9405                                 goto redo_srr;
9406                         }
9407 
9408                         if (dst == htonl(INADDR_LOOPBACK)) {
9409                                 ip1dbg(("ip_input_options: loopback addr in "
9410                                     "source route!\n"));
9411                                 goto bad_src_route;
9412                         }
9413                         /*
9414                          * For strict: verify that dst is directly
9415                          * reachable.
9416                          */
9417                         if (optval == IPOPT_SSRR) {
9418                                 ire = ire_ftable_lookup_v4(dst, 0, 0,
9419                                     IRE_INTERFACE, NULL, ALL_ZONES,
9420                                     ira->ira_tsl,
9421                                     MATCH_IRE_TYPE | MATCH_IRE_SECATTR, 0, ipst,
9422                                     NULL);
9423                                 if (ire == NULL) {
9424                                         ip1dbg(("ip_input_options: SSRR not "
9425                                             "directly reachable: 0x%x\n",
9426                                             ntohl(dst)));
9427                                         goto bad_src_route;
9428                                 }
9429                                 ire_refrele(ire);
9430                         }
9431                         /*
9432                          * Defer update of the offset and the record route
9433                          * until the packet is forwarded.
9434                          */
9435                         break;
9436                 case IPOPT_RR:
9437                         if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
9438                                 ip1dbg((
9439                                     "ip_input_options: bad option offset\n"));
9440                                 code = (char *)&opt[IPOPT_OLEN] -
9441                                     (char *)ipha;
9442                                 goto param_prob;
9443                         }
9444                         break;
9445                 case IPOPT_TS:
9446                         /*
9447                          * Verify that length >= 5 and that there is either
9448                          * room for another timestamp or that the overflow
9449                          * counter is not maxed out.
9450                          */
9451                         code = (char *)&opt[IPOPT_OLEN] - (char *)ipha;
9452                         if (optlen < IPOPT_MINLEN_IT) {
9453                                 goto param_prob;
9454                         }
9455                         if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
9456                                 ip1dbg((
9457                                     "ip_input_options: bad option offset\n"));
9458                                 code = (char *)&opt[IPOPT_OFFSET] -
9459                                     (char *)ipha;
9460                                 goto param_prob;
9461                         }
9462                         switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
9463                         case IPOPT_TS_TSONLY:
9464                                 off = IPOPT_TS_TIMELEN;
9465                                 break;
9466                         case IPOPT_TS_TSANDADDR:
9467                         case IPOPT_TS_PRESPEC:
9468                         case IPOPT_TS_PRESPEC_RFC791:
9469                                 off = IP_ADDR_LEN + IPOPT_TS_TIMELEN;
9470                                 break;
9471                         default:
9472                                 code = (char *)&opt[IPOPT_POS_OV_FLG] -
9473                                     (char *)ipha;
9474                                 goto param_prob;
9475                         }
9476                         if (opt[IPOPT_OFFSET] - 1 + off > optlen &&
9477                             (opt[IPOPT_POS_OV_FLG] & 0xF0) == 0xF0) {
9478                                 /*
9479                                  * No room and the overflow counter is 15
9480                                  * already.
9481                                  */
9482                                 goto param_prob;
9483                         }
9484                         break;
9485                 }
9486         }
9487 
9488         if ((opts.ipoptp_flags & IPOPTP_ERROR) == 0) {
9489                 return (dst);
9490         }
9491 
9492         ip1dbg(("ip_input_options: error processing IP options."));
9493         code = (char *)&opt[IPOPT_OFFSET] - (char *)ipha;
9494 
9495 param_prob:
9496         /* make sure we clear any indication of a hardware checksum */
9497         DB_CKSUMFLAGS(mp) = 0;
9498         ip_drop_input("ICMP_PARAM_PROBLEM", mp, ira->ira_ill);
9499         icmp_param_problem(mp, (uint8_t)code, ira);
9500         *errorp = -1;
9501         return (dst);
9502 
9503 bad_src_route:
9504         /* make sure we clear any indication of a hardware checksum */
9505         DB_CKSUMFLAGS(mp) = 0;
9506         ip_drop_input("ICMP_SOURCE_ROUTE_FAILED", mp, ira->ira_ill);
9507         icmp_unreachable(mp, ICMP_SOURCE_ROUTE_FAILED, ira);
9508         *errorp = -1;
9509         return (dst);
9510 }
9511 
9512 /*
9513  * IP & ICMP info in >=14 msg's ...
9514  *  - ip fixed part (mib2_ip_t)
9515  *  - icmp fixed part (mib2_icmp_t)
9516  *  - ipAddrEntryTable (ip 20)          all IPv4 ipifs
9517  *  - ipRouteEntryTable (ip 21)         all IPv4 IREs
9518  *  - ipNetToMediaEntryTable (ip 22)    all IPv4 Neighbor Cache entries
9519  *  - ipRouteAttributeTable (ip 102)    labeled routes
9520  *  - ip multicast membership (ip_member_t)
9521  *  - ip multicast source filtering (ip_grpsrc_t)
9522  *  - igmp fixed part (struct igmpstat)
9523  *  - multicast routing stats (struct mrtstat)
9524  *  - multicast routing vifs (array of struct vifctl)
9525  *  - multicast routing routes (array of struct mfcctl)
9526  *  - ip6 fixed part (mib2_ipv6IfStatsEntry_t)
9527  *                                      One per ill plus one generic
9528  *  - icmp6 fixed part (mib2_ipv6IfIcmpEntry_t)
9529  *                                      One per ill plus one generic
9530  *  - ipv6RouteEntry                    all IPv6 IREs
9531  *  - ipv6RouteAttributeTable (ip6 102) labeled routes
9532  *  - ipv6NetToMediaEntry               all IPv6 Neighbor Cache entries
9533  *  - ipv6AddrEntry                     all IPv6 ipifs
9534  *  - ipv6 multicast membership (ipv6_member_t)
9535  *  - ipv6 multicast source filtering (ipv6_grpsrc_t)
9536  *
9537  * NOTE: original mpctl is copied for msg's 2..N, since its ctl part is
9538  * already filled in by the caller.
9539  * If legacy_req is true then MIB structures needs to be truncated to their
9540  * legacy sizes before being returned.
9541  * Return value of 0 indicates that no messages were sent and caller
9542  * should free mpctl.
9543  */
9544 int
9545 ip_snmp_get(queue_t *q, mblk_t *mpctl, int level, boolean_t legacy_req)
9546 {
9547         ip_stack_t *ipst;
9548         sctp_stack_t *sctps;
9549 
9550         if (q->q_next != NULL) {
9551                 ipst = ILLQ_TO_IPST(q);
9552         } else {
9553                 ipst = CONNQ_TO_IPST(q);
9554         }
9555         ASSERT(ipst != NULL);
9556         sctps = ipst->ips_netstack->netstack_sctp;
9557 
9558         if (mpctl == NULL || mpctl->b_cont == NULL) {
9559                 return (0);
9560         }
9561 
9562         /*
9563          * For the purposes of the (broken) packet shell use
9564          * of the level we make sure MIB2_TCP/MIB2_UDP can be used
9565          * to make TCP and UDP appear first in the list of mib items.
9566          * TBD: We could expand this and use it in netstat so that
9567          * the kernel doesn't have to produce large tables (connections,
9568          * routes, etc) when netstat only wants the statistics or a particular
9569          * table.
9570          */
9571         if (!(level == MIB2_TCP || level == MIB2_UDP)) {
9572                 if ((mpctl = icmp_snmp_get(q, mpctl)) == NULL) {
9573                         return (1);
9574                 }
9575         }
9576 
9577         if (level != MIB2_TCP) {
9578                 if ((mpctl = udp_snmp_get(q, mpctl, legacy_req)) == NULL) {
9579                         return (1);
9580                 }
9581                 if (level == MIB2_UDP) {
9582                         goto done;
9583                 }
9584         }
9585 
9586         if (level != MIB2_UDP) {
9587                 if ((mpctl = tcp_snmp_get(q, mpctl, legacy_req)) == NULL) {
9588                         return (1);
9589                 }
9590                 if (level == MIB2_TCP) {
9591                         goto done;
9592                 }
9593         }
9594 
9595         if ((mpctl = ip_snmp_get_mib2_ip_traffic_stats(q, mpctl,
9596             ipst, legacy_req)) == NULL) {
9597                 return (1);
9598         }
9599 
9600         if ((mpctl = ip_snmp_get_mib2_ip6(q, mpctl, ipst,
9601             legacy_req)) == NULL) {
9602                 return (1);
9603         }
9604 
9605         if ((mpctl = ip_snmp_get_mib2_icmp(q, mpctl, ipst)) == NULL) {
9606                 return (1);
9607         }
9608 
9609         if ((mpctl = ip_snmp_get_mib2_icmp6(q, mpctl, ipst)) == NULL) {
9610                 return (1);
9611         }
9612 
9613         if ((mpctl = ip_snmp_get_mib2_igmp(q, mpctl, ipst)) == NULL) {
9614                 return (1);
9615         }
9616 
9617         if ((mpctl = ip_snmp_get_mib2_multi(q, mpctl, ipst)) == NULL) {
9618                 return (1);
9619         }
9620 
9621         if ((mpctl = ip_snmp_get_mib2_ip_addr(q, mpctl, ipst,
9622             legacy_req)) == NULL) {
9623                 return (1);
9624         }
9625 
9626         if ((mpctl = ip_snmp_get_mib2_ip6_addr(q, mpctl, ipst,
9627             legacy_req)) == NULL) {
9628                 return (1);
9629         }
9630 
9631         if ((mpctl = ip_snmp_get_mib2_ip_group_mem(q, mpctl, ipst)) == NULL) {
9632                 return (1);
9633         }
9634 
9635         if ((mpctl = ip_snmp_get_mib2_ip6_group_mem(q, mpctl, ipst)) == NULL) {
9636                 return (1);
9637         }
9638 
9639         if ((mpctl = ip_snmp_get_mib2_ip_group_src(q, mpctl, ipst)) == NULL) {
9640                 return (1);
9641         }
9642 
9643         if ((mpctl = ip_snmp_get_mib2_ip6_group_src(q, mpctl, ipst)) == NULL) {
9644                 return (1);
9645         }
9646 
9647         if ((mpctl = ip_snmp_get_mib2_virt_multi(q, mpctl, ipst)) == NULL) {
9648                 return (1);
9649         }
9650 
9651         if ((mpctl = ip_snmp_get_mib2_multi_rtable(q, mpctl, ipst)) == NULL) {
9652                 return (1);
9653         }
9654 
9655         mpctl = ip_snmp_get_mib2_ip_route_media(q, mpctl, level, ipst);
9656         if (mpctl == NULL)
9657                 return (1);
9658 
9659         mpctl = ip_snmp_get_mib2_ip6_route_media(q, mpctl, level, ipst);
9660         if (mpctl == NULL)
9661                 return (1);
9662 
9663         if ((mpctl = sctp_snmp_get_mib2(q, mpctl, sctps)) == NULL) {
9664                 return (1);
9665         }
9666         if ((mpctl = ip_snmp_get_mib2_ip_dce(q, mpctl, ipst)) == NULL) {
9667                 return (1);
9668         }
9669 done:
9670         freemsg(mpctl);
9671         return (1);
9672 }
9673 
9674 /* Get global (legacy) IPv4 statistics */
9675 static mblk_t *
9676 ip_snmp_get_mib2_ip(queue_t *q, mblk_t *mpctl, mib2_ipIfStatsEntry_t *ipmib,
9677     ip_stack_t *ipst, boolean_t legacy_req)
9678 {
9679         mib2_ip_t               old_ip_mib;
9680         struct opthdr           *optp;
9681         mblk_t                  *mp2ctl;
9682         mib2_ipAddrEntry_t      mae;
9683 
9684         /*
9685          * make a copy of the original message
9686          */
9687         mp2ctl = copymsg(mpctl);
9688 
9689         /* fixed length IP structure... */
9690         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
9691         optp->level = MIB2_IP;
9692         optp->name = 0;
9693         SET_MIB(old_ip_mib.ipForwarding,
9694             (WE_ARE_FORWARDING(ipst) ? 1 : 2));
9695         SET_MIB(old_ip_mib.ipDefaultTTL,
9696             (uint32_t)ipst->ips_ip_def_ttl);
9697         SET_MIB(old_ip_mib.ipReasmTimeout,
9698             ipst->ips_ip_reassembly_timeout);
9699         SET_MIB(old_ip_mib.ipAddrEntrySize,
9700             (legacy_req) ? LEGACY_MIB_SIZE(&mae, mib2_ipAddrEntry_t) :
9701             sizeof (mib2_ipAddrEntry_t));
9702         SET_MIB(old_ip_mib.ipRouteEntrySize,
9703             sizeof (mib2_ipRouteEntry_t));
9704         SET_MIB(old_ip_mib.ipNetToMediaEntrySize,
9705             sizeof (mib2_ipNetToMediaEntry_t));
9706         SET_MIB(old_ip_mib.ipMemberEntrySize, sizeof (ip_member_t));
9707         SET_MIB(old_ip_mib.ipGroupSourceEntrySize, sizeof (ip_grpsrc_t));
9708         SET_MIB(old_ip_mib.ipRouteAttributeSize,
9709             sizeof (mib2_ipAttributeEntry_t));
9710         SET_MIB(old_ip_mib.transportMLPSize, sizeof (mib2_transportMLPEntry_t));
9711         SET_MIB(old_ip_mib.ipDestEntrySize, sizeof (dest_cache_entry_t));
9712 
9713         /*
9714          * Grab the statistics from the new IP MIB
9715          */
9716         SET_MIB(old_ip_mib.ipInReceives,
9717             (uint32_t)ipmib->ipIfStatsHCInReceives);
9718         SET_MIB(old_ip_mib.ipInHdrErrors, ipmib->ipIfStatsInHdrErrors);
9719         SET_MIB(old_ip_mib.ipInAddrErrors, ipmib->ipIfStatsInAddrErrors);
9720         SET_MIB(old_ip_mib.ipForwDatagrams,
9721             (uint32_t)ipmib->ipIfStatsHCOutForwDatagrams);
9722         SET_MIB(old_ip_mib.ipInUnknownProtos,
9723             ipmib->ipIfStatsInUnknownProtos);
9724         SET_MIB(old_ip_mib.ipInDiscards, ipmib->ipIfStatsInDiscards);
9725         SET_MIB(old_ip_mib.ipInDelivers,
9726             (uint32_t)ipmib->ipIfStatsHCInDelivers);
9727         SET_MIB(old_ip_mib.ipOutRequests,
9728             (uint32_t)ipmib->ipIfStatsHCOutRequests);
9729         SET_MIB(old_ip_mib.ipOutDiscards, ipmib->ipIfStatsOutDiscards);
9730         SET_MIB(old_ip_mib.ipOutNoRoutes, ipmib->ipIfStatsOutNoRoutes);
9731         SET_MIB(old_ip_mib.ipReasmReqds, ipmib->ipIfStatsReasmReqds);
9732         SET_MIB(old_ip_mib.ipReasmOKs, ipmib->ipIfStatsReasmOKs);
9733         SET_MIB(old_ip_mib.ipReasmFails, ipmib->ipIfStatsReasmFails);
9734         SET_MIB(old_ip_mib.ipFragOKs, ipmib->ipIfStatsOutFragOKs);
9735         SET_MIB(old_ip_mib.ipFragFails, ipmib->ipIfStatsOutFragFails);
9736         SET_MIB(old_ip_mib.ipFragCreates, ipmib->ipIfStatsOutFragCreates);
9737 
9738         /* ipRoutingDiscards is not being used */
9739         SET_MIB(old_ip_mib.ipRoutingDiscards, 0);
9740         SET_MIB(old_ip_mib.tcpInErrs, ipmib->tcpIfStatsInErrs);
9741         SET_MIB(old_ip_mib.udpNoPorts, ipmib->udpIfStatsNoPorts);
9742         SET_MIB(old_ip_mib.ipInCksumErrs, ipmib->ipIfStatsInCksumErrs);
9743         SET_MIB(old_ip_mib.ipReasmDuplicates,
9744             ipmib->ipIfStatsReasmDuplicates);
9745         SET_MIB(old_ip_mib.ipReasmPartDups, ipmib->ipIfStatsReasmPartDups);
9746         SET_MIB(old_ip_mib.ipForwProhibits, ipmib->ipIfStatsForwProhibits);
9747         SET_MIB(old_ip_mib.udpInCksumErrs, ipmib->udpIfStatsInCksumErrs);
9748         SET_MIB(old_ip_mib.udpInOverflows, ipmib->udpIfStatsInOverflows);
9749         SET_MIB(old_ip_mib.rawipInOverflows,
9750             ipmib->rawipIfStatsInOverflows);
9751 
9752         SET_MIB(old_ip_mib.ipsecInSucceeded, ipmib->ipsecIfStatsInSucceeded);
9753         SET_MIB(old_ip_mib.ipsecInFailed, ipmib->ipsecIfStatsInFailed);
9754         SET_MIB(old_ip_mib.ipInIPv6, ipmib->ipIfStatsInWrongIPVersion);
9755         SET_MIB(old_ip_mib.ipOutIPv6, ipmib->ipIfStatsOutWrongIPVersion);
9756         SET_MIB(old_ip_mib.ipOutSwitchIPv6,
9757             ipmib->ipIfStatsOutSwitchIPVersion);
9758 
9759         if (!snmp_append_data(mpctl->b_cont, (char *)&old_ip_mib,
9760             (int)sizeof (old_ip_mib))) {
9761                 ip1dbg(("ip_snmp_get_mib2_ip: failed to allocate %u bytes\n",
9762                     (uint_t)sizeof (old_ip_mib)));
9763         }
9764 
9765         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
9766         ip3dbg(("ip_snmp_get_mib2_ip: level %d, name %d, len %d\n",
9767             (int)optp->level, (int)optp->name, (int)optp->len));
9768         qreply(q, mpctl);
9769         return (mp2ctl);
9770 }
9771 
9772 /* Per interface IPv4 statistics */
9773 static mblk_t *
9774 ip_snmp_get_mib2_ip_traffic_stats(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst,
9775     boolean_t legacy_req)
9776 {
9777         struct opthdr           *optp;
9778         mblk_t                  *mp2ctl;
9779         ill_t                   *ill;
9780         ill_walk_context_t      ctx;
9781         mblk_t                  *mp_tail = NULL;
9782         mib2_ipIfStatsEntry_t   global_ip_mib;
9783         mib2_ipAddrEntry_t      mae;
9784 
9785         /*
9786          * Make a copy of the original message
9787          */
9788         mp2ctl = copymsg(mpctl);
9789 
9790         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
9791         optp->level = MIB2_IP;
9792         optp->name = MIB2_IP_TRAFFIC_STATS;
9793         /* Include "unknown interface" ip_mib */
9794         ipst->ips_ip_mib.ipIfStatsIPVersion = MIB2_INETADDRESSTYPE_ipv4;
9795         ipst->ips_ip_mib.ipIfStatsIfIndex =
9796             MIB2_UNKNOWN_INTERFACE; /* Flag to netstat */
9797         SET_MIB(ipst->ips_ip_mib.ipIfStatsForwarding,
9798             (ipst->ips_ip_forwarding ? 1 : 2));
9799         SET_MIB(ipst->ips_ip_mib.ipIfStatsDefaultTTL,
9800             (uint32_t)ipst->ips_ip_def_ttl);
9801         SET_MIB(ipst->ips_ip_mib.ipIfStatsEntrySize,
9802             sizeof (mib2_ipIfStatsEntry_t));
9803         SET_MIB(ipst->ips_ip_mib.ipIfStatsAddrEntrySize,
9804             sizeof (mib2_ipAddrEntry_t));
9805         SET_MIB(ipst->ips_ip_mib.ipIfStatsRouteEntrySize,
9806             sizeof (mib2_ipRouteEntry_t));
9807         SET_MIB(ipst->ips_ip_mib.ipIfStatsNetToMediaEntrySize,
9808             sizeof (mib2_ipNetToMediaEntry_t));
9809         SET_MIB(ipst->ips_ip_mib.ipIfStatsMemberEntrySize,
9810             sizeof (ip_member_t));
9811         SET_MIB(ipst->ips_ip_mib.ipIfStatsGroupSourceEntrySize,
9812             sizeof (ip_grpsrc_t));
9813 
9814         bcopy(&ipst->ips_ip_mib, &global_ip_mib, sizeof (global_ip_mib));
9815 
9816         if (legacy_req) {
9817                 SET_MIB(global_ip_mib.ipIfStatsAddrEntrySize,
9818                     LEGACY_MIB_SIZE(&mae, mib2_ipAddrEntry_t));
9819         }
9820 
9821         if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
9822             (char *)&global_ip_mib, (int)sizeof (global_ip_mib))) {
9823                 ip1dbg(("ip_snmp_get_mib2_ip_traffic_stats: "
9824                     "failed to allocate %u bytes\n",
9825                     (uint_t)sizeof (global_ip_mib)));
9826         }
9827 
9828         rw_enter(&ipst->ips_ill_g_lock, RW_READER);
9829         ill = ILL_START_WALK_V4(&ctx, ipst);
9830         for (; ill != NULL; ill = ill_next(&ctx, ill)) {
9831                 ill->ill_ip_mib->ipIfStatsIfIndex =
9832                     ill->ill_phyint->phyint_ifindex;
9833                 SET_MIB(ill->ill_ip_mib->ipIfStatsForwarding,
9834                     (ipst->ips_ip_forwarding ? 1 : 2));
9835                 SET_MIB(ill->ill_ip_mib->ipIfStatsDefaultTTL,
9836                     (uint32_t)ipst->ips_ip_def_ttl);
9837 
9838                 ip_mib2_add_ip_stats(&global_ip_mib, ill->ill_ip_mib);
9839                 if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
9840                     (char *)ill->ill_ip_mib,
9841                     (int)sizeof (*ill->ill_ip_mib))) {
9842                         ip1dbg(("ip_snmp_get_mib2_ip_traffic_stats: "
9843                             "failed to allocate %u bytes\n",
9844                             (uint_t)sizeof (*ill->ill_ip_mib)));
9845                 }
9846         }
9847         rw_exit(&ipst->ips_ill_g_lock);
9848 
9849         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
9850         ip3dbg(("ip_snmp_get_mib2_ip_traffic_stats: "
9851             "level %d, name %d, len %d\n",
9852             (int)optp->level, (int)optp->name, (int)optp->len));
9853         qreply(q, mpctl);
9854 
9855         if (mp2ctl == NULL)
9856                 return (NULL);
9857 
9858         return (ip_snmp_get_mib2_ip(q, mp2ctl, &global_ip_mib, ipst,
9859             legacy_req));
9860 }
9861 
9862 /* Global IPv4 ICMP statistics */
9863 static mblk_t *
9864 ip_snmp_get_mib2_icmp(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
9865 {
9866         struct opthdr           *optp;
9867         mblk_t                  *mp2ctl;
9868 
9869         /*
9870          * Make a copy of the original message
9871          */
9872         mp2ctl = copymsg(mpctl);
9873 
9874         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
9875         optp->level = MIB2_ICMP;
9876         optp->name = 0;
9877         if (!snmp_append_data(mpctl->b_cont, (char *)&ipst->ips_icmp_mib,
9878             (int)sizeof (ipst->ips_icmp_mib))) {
9879                 ip1dbg(("ip_snmp_get_mib2_icmp: failed to allocate %u bytes\n",
9880                     (uint_t)sizeof (ipst->ips_icmp_mib)));
9881         }
9882         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
9883         ip3dbg(("ip_snmp_get_mib2_icmp: level %d, name %d, len %d\n",
9884             (int)optp->level, (int)optp->name, (int)optp->len));
9885         qreply(q, mpctl);
9886         return (mp2ctl);
9887 }
9888 
9889 /* Global IPv4 IGMP statistics */
9890 static mblk_t *
9891 ip_snmp_get_mib2_igmp(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
9892 {
9893         struct opthdr           *optp;
9894         mblk_t                  *mp2ctl;
9895 
9896         /*
9897          * make a copy of the original message
9898          */
9899         mp2ctl = copymsg(mpctl);
9900 
9901         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
9902         optp->level = EXPER_IGMP;
9903         optp->name = 0;
9904         if (!snmp_append_data(mpctl->b_cont, (char *)&ipst->ips_igmpstat,
9905             (int)sizeof (ipst->ips_igmpstat))) {
9906                 ip1dbg(("ip_snmp_get_mib2_igmp: failed to allocate %u bytes\n",
9907                     (uint_t)sizeof (ipst->ips_igmpstat)));
9908         }
9909         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
9910         ip3dbg(("ip_snmp_get_mib2_igmp: level %d, name %d, len %d\n",
9911             (int)optp->level, (int)optp->name, (int)optp->len));
9912         qreply(q, mpctl);
9913         return (mp2ctl);
9914 }
9915 
9916 /* Global IPv4 Multicast Routing statistics */
9917 static mblk_t *
9918 ip_snmp_get_mib2_multi(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
9919 {
9920         struct opthdr           *optp;
9921         mblk_t                  *mp2ctl;
9922 
9923         /*
9924          * make a copy of the original message
9925          */
9926         mp2ctl = copymsg(mpctl);
9927 
9928         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
9929         optp->level = EXPER_DVMRP;
9930         optp->name = 0;
9931         if (!ip_mroute_stats(mpctl->b_cont, ipst)) {
9932                 ip0dbg(("ip_mroute_stats: failed\n"));
9933         }
9934         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
9935         ip3dbg(("ip_snmp_get_mib2_multi: level %d, name %d, len %d\n",
9936             (int)optp->level, (int)optp->name, (int)optp->len));
9937         qreply(q, mpctl);
9938         return (mp2ctl);
9939 }
9940 
9941 /* IPv4 address information */
9942 static mblk_t *
9943 ip_snmp_get_mib2_ip_addr(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst,
9944     boolean_t legacy_req)
9945 {
9946         struct opthdr           *optp;
9947         mblk_t                  *mp2ctl;
9948         mblk_t                  *mp_tail = NULL;
9949         ill_t                   *ill;
9950         ipif_t                  *ipif;
9951         uint_t                  bitval;
9952         mib2_ipAddrEntry_t      mae;
9953         size_t                  mae_size;
9954         zoneid_t                zoneid;
9955         ill_walk_context_t      ctx;
9956 
9957         /*
9958          * make a copy of the original message
9959          */
9960         mp2ctl = copymsg(mpctl);
9961 
9962         mae_size = (legacy_req) ? LEGACY_MIB_SIZE(&mae, mib2_ipAddrEntry_t) :
9963             sizeof (mib2_ipAddrEntry_t);
9964 
9965         /* ipAddrEntryTable */
9966 
9967         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
9968         optp->level = MIB2_IP;
9969         optp->name = MIB2_IP_ADDR;
9970         zoneid = Q_TO_CONN(q)->conn_zoneid;
9971 
9972         rw_enter(&ipst->ips_ill_g_lock, RW_READER);
9973         ill = ILL_START_WALK_V4(&ctx, ipst);
9974         for (; ill != NULL; ill = ill_next(&ctx, ill)) {
9975                 for (ipif = ill->ill_ipif; ipif != NULL;
9976                     ipif = ipif->ipif_next) {
9977                         if (ipif->ipif_zoneid != zoneid &&
9978                             ipif->ipif_zoneid != ALL_ZONES)
9979                                 continue;
9980                         /* Sum of count from dead IRE_LO* and our current */
9981                         mae.ipAdEntInfo.ae_ibcnt = ipif->ipif_ib_pkt_count;
9982                         if (ipif->ipif_ire_local != NULL) {
9983                                 mae.ipAdEntInfo.ae_ibcnt +=
9984                                     ipif->ipif_ire_local->ire_ib_pkt_count;
9985                         }
9986                         mae.ipAdEntInfo.ae_obcnt = 0;
9987                         mae.ipAdEntInfo.ae_focnt = 0;
9988 
9989                         ipif_get_name(ipif, mae.ipAdEntIfIndex.o_bytes,
9990                             OCTET_LENGTH);
9991                         mae.ipAdEntIfIndex.o_length =
9992                             mi_strlen(mae.ipAdEntIfIndex.o_bytes);
9993                         mae.ipAdEntAddr = ipif->ipif_lcl_addr;
9994                         mae.ipAdEntNetMask = ipif->ipif_net_mask;
9995                         mae.ipAdEntInfo.ae_subnet = ipif->ipif_subnet;
9996                         mae.ipAdEntInfo.ae_subnet_len =
9997                             ip_mask_to_plen(ipif->ipif_net_mask);
9998                         mae.ipAdEntInfo.ae_src_addr = ipif->ipif_lcl_addr;
9999                         for (bitval = 1;
10000                             bitval &&
10001                             !(bitval & ipif->ipif_brd_addr);
10002                             bitval <<= 1)
10003                                 noop;
10004                         mae.ipAdEntBcastAddr = bitval;
10005                         mae.ipAdEntReasmMaxSize = IP_MAXPACKET;
10006                         mae.ipAdEntInfo.ae_mtu = ipif->ipif_ill->ill_mtu;
10007                         mae.ipAdEntInfo.ae_metric  = ipif->ipif_ill->ill_metric;
10008                         mae.ipAdEntInfo.ae_broadcast_addr =
10009                             ipif->ipif_brd_addr;
10010                         mae.ipAdEntInfo.ae_pp_dst_addr =
10011                             ipif->ipif_pp_dst_addr;
10012                         mae.ipAdEntInfo.ae_flags = ipif->ipif_flags |
10013                             ill->ill_flags | ill->ill_phyint->phyint_flags;
10014                         mae.ipAdEntRetransmitTime =
10015                             ill->ill_reachable_retrans_time;
10016 
10017                         if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10018                             (char *)&mae, (int)mae_size)) {
10019                                 ip1dbg(("ip_snmp_get_mib2_ip_addr: failed to "
10020                                     "allocate %u bytes\n", (uint_t)mae_size));
10021                         }
10022                 }
10023         }
10024         rw_exit(&ipst->ips_ill_g_lock);
10025 
10026         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10027         ip3dbg(("ip_snmp_get_mib2_ip_addr: level %d, name %d, len %d\n",
10028             (int)optp->level, (int)optp->name, (int)optp->len));
10029         qreply(q, mpctl);
10030         return (mp2ctl);
10031 }
10032 
10033 /* IPv6 address information */
10034 static mblk_t *
10035 ip_snmp_get_mib2_ip6_addr(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst,
10036     boolean_t legacy_req)
10037 {
10038         struct opthdr           *optp;
10039         mblk_t                  *mp2ctl;
10040         mblk_t                  *mp_tail = NULL;
10041         ill_t                   *ill;
10042         ipif_t                  *ipif;
10043         mib2_ipv6AddrEntry_t    mae6;
10044         size_t                  mae6_size;
10045         zoneid_t                zoneid;
10046         ill_walk_context_t      ctx;
10047 
10048         /*
10049          * make a copy of the original message
10050          */
10051         mp2ctl = copymsg(mpctl);
10052 
10053         mae6_size = (legacy_req) ?
10054             LEGACY_MIB_SIZE(&mae6, mib2_ipv6AddrEntry_t) :
10055             sizeof (mib2_ipv6AddrEntry_t);
10056 
10057         /* ipv6AddrEntryTable */
10058 
10059         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10060         optp->level = MIB2_IP6;
10061         optp->name = MIB2_IP6_ADDR;
10062         zoneid = Q_TO_CONN(q)->conn_zoneid;
10063 
10064         rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10065         ill = ILL_START_WALK_V6(&ctx, ipst);
10066         for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10067                 for (ipif = ill->ill_ipif; ipif != NULL;
10068                     ipif = ipif->ipif_next) {
10069                         if (ipif->ipif_zoneid != zoneid &&
10070                             ipif->ipif_zoneid != ALL_ZONES)
10071                                 continue;
10072                         /* Sum of count from dead IRE_LO* and our current */
10073                         mae6.ipv6AddrInfo.ae_ibcnt = ipif->ipif_ib_pkt_count;
10074                         if (ipif->ipif_ire_local != NULL) {
10075                                 mae6.ipv6AddrInfo.ae_ibcnt +=
10076                                     ipif->ipif_ire_local->ire_ib_pkt_count;
10077                         }
10078                         mae6.ipv6AddrInfo.ae_obcnt = 0;
10079                         mae6.ipv6AddrInfo.ae_focnt = 0;
10080 
10081                         ipif_get_name(ipif, mae6.ipv6AddrIfIndex.o_bytes,
10082                             OCTET_LENGTH);
10083                         mae6.ipv6AddrIfIndex.o_length =
10084                             mi_strlen(mae6.ipv6AddrIfIndex.o_bytes);
10085                         mae6.ipv6AddrAddress = ipif->ipif_v6lcl_addr;
10086                         mae6.ipv6AddrPfxLength =
10087                             ip_mask_to_plen_v6(&ipif->ipif_v6net_mask);
10088                         mae6.ipv6AddrInfo.ae_subnet = ipif->ipif_v6subnet;
10089                         mae6.ipv6AddrInfo.ae_subnet_len =
10090                             mae6.ipv6AddrPfxLength;
10091                         mae6.ipv6AddrInfo.ae_src_addr = ipif->ipif_v6lcl_addr;
10092 
10093                         /* Type: stateless(1), stateful(2), unknown(3) */
10094                         if (ipif->ipif_flags & IPIF_ADDRCONF)
10095                                 mae6.ipv6AddrType = 1;
10096                         else
10097                                 mae6.ipv6AddrType = 2;
10098                         /* Anycast: true(1), false(2) */
10099                         if (ipif->ipif_flags & IPIF_ANYCAST)
10100                                 mae6.ipv6AddrAnycastFlag = 1;
10101                         else
10102                                 mae6.ipv6AddrAnycastFlag = 2;
10103 
10104                         /*
10105                          * Address status: preferred(1), deprecated(2),
10106                          * invalid(3), inaccessible(4), unknown(5)
10107                          */
10108                         if (ipif->ipif_flags & IPIF_NOLOCAL)
10109                                 mae6.ipv6AddrStatus = 3;
10110                         else if (ipif->ipif_flags & IPIF_DEPRECATED)
10111                                 mae6.ipv6AddrStatus = 2;
10112                         else
10113                                 mae6.ipv6AddrStatus = 1;
10114                         mae6.ipv6AddrInfo.ae_mtu = ipif->ipif_ill->ill_mtu;
10115                         mae6.ipv6AddrInfo.ae_metric  =
10116                             ipif->ipif_ill->ill_metric;
10117                         mae6.ipv6AddrInfo.ae_pp_dst_addr =
10118                             ipif->ipif_v6pp_dst_addr;
10119                         mae6.ipv6AddrInfo.ae_flags = ipif->ipif_flags |
10120                             ill->ill_flags | ill->ill_phyint->phyint_flags;
10121                         mae6.ipv6AddrReasmMaxSize = IP_MAXPACKET;
10122                         mae6.ipv6AddrIdentifier = ill->ill_token;
10123                         mae6.ipv6AddrIdentifierLen = ill->ill_token_length;
10124                         mae6.ipv6AddrReachableTime = ill->ill_reachable_time;
10125                         mae6.ipv6AddrRetransmitTime =
10126                             ill->ill_reachable_retrans_time;
10127                         if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10128                             (char *)&mae6, (int)mae6_size)) {
10129                                 ip1dbg(("ip_snmp_get_mib2_ip6_addr: failed to "
10130                                     "allocate %u bytes\n",
10131                                     (uint_t)mae6_size));
10132                         }
10133                 }
10134         }
10135         rw_exit(&ipst->ips_ill_g_lock);
10136 
10137         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10138         ip3dbg(("ip_snmp_get_mib2_ip6_addr: level %d, name %d, len %d\n",
10139             (int)optp->level, (int)optp->name, (int)optp->len));
10140         qreply(q, mpctl);
10141         return (mp2ctl);
10142 }
10143 
10144 /* IPv4 multicast group membership. */
10145 static mblk_t *
10146 ip_snmp_get_mib2_ip_group_mem(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10147 {
10148         struct opthdr           *optp;
10149         mblk_t                  *mp2ctl;
10150         ill_t                   *ill;
10151         ipif_t                  *ipif;
10152         ilm_t                   *ilm;
10153         ip_member_t             ipm;
10154         mblk_t                  *mp_tail = NULL;
10155         ill_walk_context_t      ctx;
10156         zoneid_t                zoneid;
10157 
10158         /*
10159          * make a copy of the original message
10160          */
10161         mp2ctl = copymsg(mpctl);
10162         zoneid = Q_TO_CONN(q)->conn_zoneid;
10163 
10164         /* ipGroupMember table */
10165         optp = (struct opthdr *)&mpctl->b_rptr[
10166             sizeof (struct T_optmgmt_ack)];
10167         optp->level = MIB2_IP;
10168         optp->name = EXPER_IP_GROUP_MEMBERSHIP;
10169 
10170         rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10171         ill = ILL_START_WALK_V4(&ctx, ipst);
10172         for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10173                 /* Make sure the ill isn't going away. */
10174                 if (!ill_check_and_refhold(ill))
10175                         continue;
10176                 rw_exit(&ipst->ips_ill_g_lock);
10177                 rw_enter(&ill->ill_mcast_lock, RW_READER);
10178                 for (ilm = ill->ill_ilm; ilm; ilm = ilm->ilm_next) {
10179                         if (ilm->ilm_zoneid != zoneid &&
10180                             ilm->ilm_zoneid != ALL_ZONES)
10181                                 continue;
10182 
10183                         /* Is there an ipif for ilm_ifaddr? */
10184                         for (ipif = ill->ill_ipif; ipif != NULL;
10185                             ipif = ipif->ipif_next) {
10186                                 if (!IPIF_IS_CONDEMNED(ipif) &&
10187                                     ipif->ipif_lcl_addr == ilm->ilm_ifaddr &&
10188                                     ilm->ilm_ifaddr != INADDR_ANY)
10189                                         break;
10190                         }
10191                         if (ipif != NULL) {
10192                                 ipif_get_name(ipif,
10193                                     ipm.ipGroupMemberIfIndex.o_bytes,
10194                                     OCTET_LENGTH);
10195                         } else {
10196                                 ill_get_name(ill,
10197                                     ipm.ipGroupMemberIfIndex.o_bytes,
10198                                     OCTET_LENGTH);
10199                         }
10200                         ipm.ipGroupMemberIfIndex.o_length =
10201                             mi_strlen(ipm.ipGroupMemberIfIndex.o_bytes);
10202 
10203                         ipm.ipGroupMemberAddress = ilm->ilm_addr;
10204                         ipm.ipGroupMemberRefCnt = ilm->ilm_refcnt;
10205                         ipm.ipGroupMemberFilterMode = ilm->ilm_fmode;
10206                         if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10207                             (char *)&ipm, (int)sizeof (ipm))) {
10208                                 ip1dbg(("ip_snmp_get_mib2_ip_group: "
10209                                     "failed to allocate %u bytes\n",
10210                                     (uint_t)sizeof (ipm)));
10211                         }
10212                 }
10213                 rw_exit(&ill->ill_mcast_lock);
10214                 ill_refrele(ill);
10215                 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10216         }
10217         rw_exit(&ipst->ips_ill_g_lock);
10218         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10219         ip3dbg(("ip_snmp_get: level %d, name %d, len %d\n",
10220             (int)optp->level, (int)optp->name, (int)optp->len));
10221         qreply(q, mpctl);
10222         return (mp2ctl);
10223 }
10224 
10225 /* IPv6 multicast group membership. */
10226 static mblk_t *
10227 ip_snmp_get_mib2_ip6_group_mem(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10228 {
10229         struct opthdr           *optp;
10230         mblk_t                  *mp2ctl;
10231         ill_t                   *ill;
10232         ilm_t                   *ilm;
10233         ipv6_member_t           ipm6;
10234         mblk_t                  *mp_tail = NULL;
10235         ill_walk_context_t      ctx;
10236         zoneid_t                zoneid;
10237 
10238         /*
10239          * make a copy of the original message
10240          */
10241         mp2ctl = copymsg(mpctl);
10242         zoneid = Q_TO_CONN(q)->conn_zoneid;
10243 
10244         /* ip6GroupMember table */
10245         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10246         optp->level = MIB2_IP6;
10247         optp->name = EXPER_IP6_GROUP_MEMBERSHIP;
10248 
10249         rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10250         ill = ILL_START_WALK_V6(&ctx, ipst);
10251         for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10252                 /* Make sure the ill isn't going away. */
10253                 if (!ill_check_and_refhold(ill))
10254                         continue;
10255                 rw_exit(&ipst->ips_ill_g_lock);
10256                 /*
10257                  * Normally we don't have any members on under IPMP interfaces.
10258                  * We report them as a debugging aid.
10259                  */
10260                 rw_enter(&ill->ill_mcast_lock, RW_READER);
10261                 ipm6.ipv6GroupMemberIfIndex = ill->ill_phyint->phyint_ifindex;
10262                 for (ilm = ill->ill_ilm; ilm; ilm = ilm->ilm_next) {
10263                         if (ilm->ilm_zoneid != zoneid &&
10264                             ilm->ilm_zoneid != ALL_ZONES)
10265                                 continue;       /* not this zone */
10266                         ipm6.ipv6GroupMemberAddress = ilm->ilm_v6addr;
10267                         ipm6.ipv6GroupMemberRefCnt = ilm->ilm_refcnt;
10268                         ipm6.ipv6GroupMemberFilterMode = ilm->ilm_fmode;
10269                         if (!snmp_append_data2(mpctl->b_cont,
10270                             &mp_tail,
10271                             (char *)&ipm6, (int)sizeof (ipm6))) {
10272                                 ip1dbg(("ip_snmp_get_mib2_ip6_group: "
10273                                     "failed to allocate %u bytes\n",
10274                                     (uint_t)sizeof (ipm6)));
10275                         }
10276                 }
10277                 rw_exit(&ill->ill_mcast_lock);
10278                 ill_refrele(ill);
10279                 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10280         }
10281         rw_exit(&ipst->ips_ill_g_lock);
10282 
10283         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10284         ip3dbg(("ip_snmp_get: level %d, name %d, len %d\n",
10285             (int)optp->level, (int)optp->name, (int)optp->len));
10286         qreply(q, mpctl);
10287         return (mp2ctl);
10288 }
10289 
10290 /* IP multicast filtered sources */
10291 static mblk_t *
10292 ip_snmp_get_mib2_ip_group_src(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10293 {
10294         struct opthdr           *optp;
10295         mblk_t                  *mp2ctl;
10296         ill_t                   *ill;
10297         ipif_t                  *ipif;
10298         ilm_t                   *ilm;
10299         ip_grpsrc_t             ips;
10300         mblk_t                  *mp_tail = NULL;
10301         ill_walk_context_t      ctx;
10302         zoneid_t                zoneid;
10303         int                     i;
10304         slist_t                 *sl;
10305 
10306         /*
10307          * make a copy of the original message
10308          */
10309         mp2ctl = copymsg(mpctl);
10310         zoneid = Q_TO_CONN(q)->conn_zoneid;
10311 
10312         /* ipGroupSource table */
10313         optp = (struct opthdr *)&mpctl->b_rptr[
10314             sizeof (struct T_optmgmt_ack)];
10315         optp->level = MIB2_IP;
10316         optp->name = EXPER_IP_GROUP_SOURCES;
10317 
10318         rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10319         ill = ILL_START_WALK_V4(&ctx, ipst);
10320         for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10321                 /* Make sure the ill isn't going away. */
10322                 if (!ill_check_and_refhold(ill))
10323                         continue;
10324                 rw_exit(&ipst->ips_ill_g_lock);
10325                 rw_enter(&ill->ill_mcast_lock, RW_READER);
10326                 for (ilm = ill->ill_ilm; ilm; ilm = ilm->ilm_next) {
10327                         sl = ilm->ilm_filter;
10328                         if (ilm->ilm_zoneid != zoneid &&
10329                             ilm->ilm_zoneid != ALL_ZONES)
10330                                 continue;
10331                         if (SLIST_IS_EMPTY(sl))
10332                                 continue;
10333 
10334                         /* Is there an ipif for ilm_ifaddr? */
10335                         for (ipif = ill->ill_ipif; ipif != NULL;
10336                             ipif = ipif->ipif_next) {
10337                                 if (!IPIF_IS_CONDEMNED(ipif) &&
10338                                     ipif->ipif_lcl_addr == ilm->ilm_ifaddr &&
10339                                     ilm->ilm_ifaddr != INADDR_ANY)
10340                                         break;
10341                         }
10342                         if (ipif != NULL) {
10343                                 ipif_get_name(ipif,
10344                                     ips.ipGroupSourceIfIndex.o_bytes,
10345                                     OCTET_LENGTH);
10346                         } else {
10347                                 ill_get_name(ill,
10348                                     ips.ipGroupSourceIfIndex.o_bytes,
10349                                     OCTET_LENGTH);
10350                         }
10351                         ips.ipGroupSourceIfIndex.o_length =
10352                             mi_strlen(ips.ipGroupSourceIfIndex.o_bytes);
10353 
10354                         ips.ipGroupSourceGroup = ilm->ilm_addr;
10355                         for (i = 0; i < sl->sl_numsrc; i++) {
10356                                 if (!IN6_IS_ADDR_V4MAPPED(&sl->sl_addr[i]))
10357                                         continue;
10358                                 IN6_V4MAPPED_TO_IPADDR(&sl->sl_addr[i],
10359                                     ips.ipGroupSourceAddress);
10360                                 if (snmp_append_data2(mpctl->b_cont, &mp_tail,
10361                                     (char *)&ips, (int)sizeof (ips)) == 0) {
10362                                         ip1dbg(("ip_snmp_get_mib2_ip_group_src:"
10363                                             " failed to allocate %u bytes\n",
10364                                             (uint_t)sizeof (ips)));
10365                                 }
10366                         }
10367                 }
10368                 rw_exit(&ill->ill_mcast_lock);
10369                 ill_refrele(ill);
10370                 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10371         }
10372         rw_exit(&ipst->ips_ill_g_lock);
10373         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10374         ip3dbg(("ip_snmp_get: level %d, name %d, len %d\n",
10375             (int)optp->level, (int)optp->name, (int)optp->len));
10376         qreply(q, mpctl);
10377         return (mp2ctl);
10378 }
10379 
10380 /* IPv6 multicast filtered sources. */
10381 static mblk_t *
10382 ip_snmp_get_mib2_ip6_group_src(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10383 {
10384         struct opthdr           *optp;
10385         mblk_t                  *mp2ctl;
10386         ill_t                   *ill;
10387         ilm_t                   *ilm;
10388         ipv6_grpsrc_t           ips6;
10389         mblk_t                  *mp_tail = NULL;
10390         ill_walk_context_t      ctx;
10391         zoneid_t                zoneid;
10392         int                     i;
10393         slist_t                 *sl;
10394 
10395         /*
10396          * make a copy of the original message
10397          */
10398         mp2ctl = copymsg(mpctl);
10399         zoneid = Q_TO_CONN(q)->conn_zoneid;
10400 
10401         /* ip6GroupMember table */
10402         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10403         optp->level = MIB2_IP6;
10404         optp->name = EXPER_IP6_GROUP_SOURCES;
10405 
10406         rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10407         ill = ILL_START_WALK_V6(&ctx, ipst);
10408         for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10409                 /* Make sure the ill isn't going away. */
10410                 if (!ill_check_and_refhold(ill))
10411                         continue;
10412                 rw_exit(&ipst->ips_ill_g_lock);
10413                 /*
10414                  * Normally we don't have any members on under IPMP interfaces.
10415                  * We report them as a debugging aid.
10416                  */
10417                 rw_enter(&ill->ill_mcast_lock, RW_READER);
10418                 ips6.ipv6GroupSourceIfIndex = ill->ill_phyint->phyint_ifindex;
10419                 for (ilm = ill->ill_ilm; ilm; ilm = ilm->ilm_next) {
10420                         sl = ilm->ilm_filter;
10421                         if (ilm->ilm_zoneid != zoneid &&
10422                             ilm->ilm_zoneid != ALL_ZONES)
10423                                 continue;
10424                         if (SLIST_IS_EMPTY(sl))
10425                                 continue;
10426                         ips6.ipv6GroupSourceGroup = ilm->ilm_v6addr;
10427                         for (i = 0; i < sl->sl_numsrc; i++) {
10428                                 ips6.ipv6GroupSourceAddress = sl->sl_addr[i];
10429                                 if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10430                                     (char *)&ips6, (int)sizeof (ips6))) {
10431                                         ip1dbg(("ip_snmp_get_mib2_ip6_"
10432                                             "group_src: failed to allocate "
10433                                             "%u bytes\n",
10434                                             (uint_t)sizeof (ips6)));
10435                                 }
10436                         }
10437                 }
10438                 rw_exit(&ill->ill_mcast_lock);
10439                 ill_refrele(ill);
10440                 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10441         }
10442         rw_exit(&ipst->ips_ill_g_lock);
10443 
10444         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10445         ip3dbg(("ip_snmp_get: level %d, name %d, len %d\n",
10446             (int)optp->level, (int)optp->name, (int)optp->len));
10447         qreply(q, mpctl);
10448         return (mp2ctl);
10449 }
10450 
10451 /* Multicast routing virtual interface table. */
10452 static mblk_t *
10453 ip_snmp_get_mib2_virt_multi(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10454 {
10455         struct opthdr           *optp;
10456         mblk_t                  *mp2ctl;
10457 
10458         /*
10459          * make a copy of the original message
10460          */
10461         mp2ctl = copymsg(mpctl);
10462 
10463         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10464         optp->level = EXPER_DVMRP;
10465         optp->name = EXPER_DVMRP_VIF;
10466         if (!ip_mroute_vif(mpctl->b_cont, ipst)) {
10467                 ip0dbg(("ip_mroute_vif: failed\n"));
10468         }
10469         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10470         ip3dbg(("ip_snmp_get_mib2_virt_multi: level %d, name %d, len %d\n",
10471             (int)optp->level, (int)optp->name, (int)optp->len));
10472         qreply(q, mpctl);
10473         return (mp2ctl);
10474 }
10475 
10476 /* Multicast routing table. */
10477 static mblk_t *
10478 ip_snmp_get_mib2_multi_rtable(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10479 {
10480         struct opthdr           *optp;
10481         mblk_t                  *mp2ctl;
10482 
10483         /*
10484          * make a copy of the original message
10485          */
10486         mp2ctl = copymsg(mpctl);
10487 
10488         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10489         optp->level = EXPER_DVMRP;
10490         optp->name = EXPER_DVMRP_MRT;
10491         if (!ip_mroute_mrt(mpctl->b_cont, ipst)) {
10492                 ip0dbg(("ip_mroute_mrt: failed\n"));
10493         }
10494         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10495         ip3dbg(("ip_snmp_get_mib2_multi_rtable: level %d, name %d, len %d\n",
10496             (int)optp->level, (int)optp->name, (int)optp->len));
10497         qreply(q, mpctl);
10498         return (mp2ctl);
10499 }
10500 
10501 /*
10502  * Return ipRouteEntryTable, ipNetToMediaEntryTable, and ipRouteAttributeTable
10503  * in one IRE walk.
10504  */
10505 static mblk_t *
10506 ip_snmp_get_mib2_ip_route_media(queue_t *q, mblk_t *mpctl, int level,
10507     ip_stack_t *ipst)
10508 {
10509         struct opthdr   *optp;
10510         mblk_t          *mp2ctl;        /* Returned */
10511         mblk_t          *mp3ctl;        /* nettomedia */
10512         mblk_t          *mp4ctl;        /* routeattrs */
10513         iproutedata_t   ird;
10514         zoneid_t        zoneid;
10515 
10516         /*
10517          * make copies of the original message
10518          *      - mp2ctl is returned unchanged to the caller for its use
10519          *      - mpctl is sent upstream as ipRouteEntryTable
10520          *      - mp3ctl is sent upstream as ipNetToMediaEntryTable
10521          *      - mp4ctl is sent upstream as ipRouteAttributeTable
10522          */
10523         mp2ctl = copymsg(mpctl);
10524         mp3ctl = copymsg(mpctl);
10525         mp4ctl = copymsg(mpctl);
10526         if (mp3ctl == NULL || mp4ctl == NULL) {
10527                 freemsg(mp4ctl);
10528                 freemsg(mp3ctl);
10529                 freemsg(mp2ctl);
10530                 freemsg(mpctl);
10531                 return (NULL);
10532         }
10533 
10534         bzero(&ird, sizeof (ird));
10535 
10536         ird.ird_route.lp_head = mpctl->b_cont;
10537         ird.ird_netmedia.lp_head = mp3ctl->b_cont;
10538         ird.ird_attrs.lp_head = mp4ctl->b_cont;
10539         /*
10540          * If the level has been set the special EXPER_IP_AND_ALL_IRES value,
10541          * then also include ire_testhidden IREs and IRE_IF_CLONE.  This is
10542          * intended a temporary solution until a proper MIB API is provided
10543          * that provides complete filtering/caller-opt-in.
10544          */
10545         if (level == EXPER_IP_AND_ALL_IRES)
10546                 ird.ird_flags |= IRD_REPORT_ALL;
10547 
10548         zoneid = Q_TO_CONN(q)->conn_zoneid;
10549         ire_walk_v4(ip_snmp_get2_v4, &ird, zoneid, ipst);
10550 
10551         /* ipRouteEntryTable in mpctl */
10552         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10553         optp->level = MIB2_IP;
10554         optp->name = MIB2_IP_ROUTE;
10555         optp->len = msgdsize(ird.ird_route.lp_head);
10556         ip3dbg(("ip_snmp_get_mib2_ip_route_media: level %d, name %d, len %d\n",
10557             (int)optp->level, (int)optp->name, (int)optp->len));
10558         qreply(q, mpctl);
10559 
10560         /* ipNetToMediaEntryTable in mp3ctl */
10561         ncec_walk(NULL, ip_snmp_get2_v4_media, &ird, ipst);
10562 
10563         optp = (struct opthdr *)&mp3ctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10564         optp->level = MIB2_IP;
10565         optp->name = MIB2_IP_MEDIA;
10566         optp->len = msgdsize(ird.ird_netmedia.lp_head);
10567         ip3dbg(("ip_snmp_get_mib2_ip_route_media: level %d, name %d, len %d\n",
10568             (int)optp->level, (int)optp->name, (int)optp->len));
10569         qreply(q, mp3ctl);
10570 
10571         /* ipRouteAttributeTable in mp4ctl */
10572         optp = (struct opthdr *)&mp4ctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10573         optp->level = MIB2_IP;
10574         optp->name = EXPER_IP_RTATTR;
10575         optp->len = msgdsize(ird.ird_attrs.lp_head);
10576         ip3dbg(("ip_snmp_get_mib2_ip_route_media: level %d, name %d, len %d\n",
10577             (int)optp->level, (int)optp->name, (int)optp->len));
10578         if (optp->len == 0)
10579                 freemsg(mp4ctl);
10580         else
10581                 qreply(q, mp4ctl);
10582 
10583         return (mp2ctl);
10584 }
10585 
10586 /*
10587  * Return ipv6RouteEntryTable and ipv6RouteAttributeTable in one IRE walk, and
10588  * ipv6NetToMediaEntryTable in an NDP walk.
10589  */
10590 static mblk_t *
10591 ip_snmp_get_mib2_ip6_route_media(queue_t *q, mblk_t *mpctl, int level,
10592     ip_stack_t *ipst)
10593 {
10594         struct opthdr   *optp;
10595         mblk_t          *mp2ctl;        /* Returned */
10596         mblk_t          *mp3ctl;        /* nettomedia */
10597         mblk_t          *mp4ctl;        /* routeattrs */
10598         iproutedata_t   ird;
10599         zoneid_t        zoneid;
10600 
10601         /*
10602          * make copies of the original message
10603          *      - mp2ctl is returned unchanged to the caller for its use
10604          *      - mpctl is sent upstream as ipv6RouteEntryTable
10605          *      - mp3ctl is sent upstream as ipv6NetToMediaEntryTable
10606          *      - mp4ctl is sent upstream as ipv6RouteAttributeTable
10607          */
10608         mp2ctl = copymsg(mpctl);
10609         mp3ctl = copymsg(mpctl);
10610         mp4ctl = copymsg(mpctl);
10611         if (mp3ctl == NULL || mp4ctl == NULL) {
10612                 freemsg(mp4ctl);
10613                 freemsg(mp3ctl);
10614                 freemsg(mp2ctl);
10615                 freemsg(mpctl);
10616                 return (NULL);
10617         }
10618 
10619         bzero(&ird, sizeof (ird));
10620 
10621         ird.ird_route.lp_head = mpctl->b_cont;
10622         ird.ird_netmedia.lp_head = mp3ctl->b_cont;
10623         ird.ird_attrs.lp_head = mp4ctl->b_cont;
10624         /*
10625          * If the level has been set the special EXPER_IP_AND_ALL_IRES value,
10626          * then also include ire_testhidden IREs and IRE_IF_CLONE.  This is
10627          * intended a temporary solution until a proper MIB API is provided
10628          * that provides complete filtering/caller-opt-in.
10629          */
10630         if (level == EXPER_IP_AND_ALL_IRES)
10631                 ird.ird_flags |= IRD_REPORT_ALL;
10632 
10633         zoneid = Q_TO_CONN(q)->conn_zoneid;
10634         ire_walk_v6(ip_snmp_get2_v6_route, &ird, zoneid, ipst);
10635 
10636         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10637         optp->level = MIB2_IP6;
10638         optp->name = MIB2_IP6_ROUTE;
10639         optp->len = msgdsize(ird.ird_route.lp_head);
10640         ip3dbg(("ip_snmp_get_mib2_ip6_route_media: level %d, name %d, len %d\n",
10641             (int)optp->level, (int)optp->name, (int)optp->len));
10642         qreply(q, mpctl);
10643 
10644         /* ipv6NetToMediaEntryTable in mp3ctl */
10645         ncec_walk(NULL, ip_snmp_get2_v6_media, &ird, ipst);
10646 
10647         optp = (struct opthdr *)&mp3ctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10648         optp->level = MIB2_IP6;
10649         optp->name = MIB2_IP6_MEDIA;
10650         optp->len = msgdsize(ird.ird_netmedia.lp_head);
10651         ip3dbg(("ip_snmp_get_mib2_ip6_route_media: level %d, name %d, len %d\n",
10652             (int)optp->level, (int)optp->name, (int)optp->len));
10653         qreply(q, mp3ctl);
10654 
10655         /* ipv6RouteAttributeTable in mp4ctl */
10656         optp = (struct opthdr *)&mp4ctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10657         optp->level = MIB2_IP6;
10658         optp->name = EXPER_IP_RTATTR;
10659         optp->len = msgdsize(ird.ird_attrs.lp_head);
10660         ip3dbg(("ip_snmp_get_mib2_ip6_route_media: level %d, name %d, len %d\n",
10661             (int)optp->level, (int)optp->name, (int)optp->len));
10662         if (optp->len == 0)
10663                 freemsg(mp4ctl);
10664         else
10665                 qreply(q, mp4ctl);
10666 
10667         return (mp2ctl);
10668 }
10669 
10670 /*
10671  * IPv6 mib: One per ill
10672  */
10673 static mblk_t *
10674 ip_snmp_get_mib2_ip6(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst,
10675     boolean_t legacy_req)
10676 {
10677         struct opthdr           *optp;
10678         mblk_t                  *mp2ctl;
10679         ill_t                   *ill;
10680         ill_walk_context_t      ctx;
10681         mblk_t                  *mp_tail = NULL;
10682         mib2_ipv6AddrEntry_t    mae6;
10683         mib2_ipIfStatsEntry_t   *ise;
10684         size_t                  ise_size, iae_size;
10685 
10686         /*
10687          * Make a copy of the original message
10688          */
10689         mp2ctl = copymsg(mpctl);
10690 
10691         /* fixed length IPv6 structure ... */
10692 
10693         if (legacy_req) {
10694                 ise_size = LEGACY_MIB_SIZE(&ipst->ips_ip6_mib,
10695                     mib2_ipIfStatsEntry_t);
10696                 iae_size = LEGACY_MIB_SIZE(&mae6, mib2_ipv6AddrEntry_t);
10697         } else {
10698                 ise_size = sizeof (mib2_ipIfStatsEntry_t);
10699                 iae_size = sizeof (mib2_ipv6AddrEntry_t);
10700         }
10701 
10702         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10703         optp->level = MIB2_IP6;
10704         optp->name = 0;
10705         /* Include "unknown interface" ip6_mib */
10706         ipst->ips_ip6_mib.ipIfStatsIPVersion = MIB2_INETADDRESSTYPE_ipv6;
10707         ipst->ips_ip6_mib.ipIfStatsIfIndex =
10708             MIB2_UNKNOWN_INTERFACE; /* Flag to netstat */
10709         SET_MIB(ipst->ips_ip6_mib.ipIfStatsForwarding,
10710             ipst->ips_ipv6_forwarding ? 1 : 2);
10711         SET_MIB(ipst->ips_ip6_mib.ipIfStatsDefaultHopLimit,
10712             ipst->ips_ipv6_def_hops);
10713         SET_MIB(ipst->ips_ip6_mib.ipIfStatsEntrySize,
10714             sizeof (mib2_ipIfStatsEntry_t));
10715         SET_MIB(ipst->ips_ip6_mib.ipIfStatsAddrEntrySize,
10716             sizeof (mib2_ipv6AddrEntry_t));
10717         SET_MIB(ipst->ips_ip6_mib.ipIfStatsRouteEntrySize,
10718             sizeof (mib2_ipv6RouteEntry_t));
10719         SET_MIB(ipst->ips_ip6_mib.ipIfStatsNetToMediaEntrySize,
10720             sizeof (mib2_ipv6NetToMediaEntry_t));
10721         SET_MIB(ipst->ips_ip6_mib.ipIfStatsMemberEntrySize,
10722             sizeof (ipv6_member_t));
10723         SET_MIB(ipst->ips_ip6_mib.ipIfStatsGroupSourceEntrySize,
10724             sizeof (ipv6_grpsrc_t));
10725 
10726         /*
10727          * Synchronize 64- and 32-bit counters
10728          */
10729         SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsInReceives,
10730             ipIfStatsHCInReceives);
10731         SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsInDelivers,
10732             ipIfStatsHCInDelivers);
10733         SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsOutRequests,
10734             ipIfStatsHCOutRequests);
10735         SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsOutForwDatagrams,
10736             ipIfStatsHCOutForwDatagrams);
10737         SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsOutMcastPkts,
10738             ipIfStatsHCOutMcastPkts);
10739         SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsInMcastPkts,
10740             ipIfStatsHCInMcastPkts);
10741 
10742         if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10743             (char *)&ipst->ips_ip6_mib, (int)ise_size)) {
10744                 ip1dbg(("ip_snmp_get_mib2_ip6: failed to allocate %u bytes\n",
10745                     (uint_t)ise_size));
10746         } else if (legacy_req) {
10747                 /* Adjust the EntrySize fields for legacy requests. */
10748                 ise =
10749                     (mib2_ipIfStatsEntry_t *)(mp_tail->b_wptr - (int)ise_size);
10750                 SET_MIB(ise->ipIfStatsEntrySize, ise_size);
10751                 SET_MIB(ise->ipIfStatsAddrEntrySize, iae_size);
10752         }
10753 
10754         rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10755         ill = ILL_START_WALK_V6(&ctx, ipst);
10756         for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10757                 ill->ill_ip_mib->ipIfStatsIfIndex =
10758                     ill->ill_phyint->phyint_ifindex;
10759                 SET_MIB(ill->ill_ip_mib->ipIfStatsForwarding,
10760                     ipst->ips_ipv6_forwarding ? 1 : 2);
10761                 SET_MIB(ill->ill_ip_mib->ipIfStatsDefaultHopLimit,
10762                     ill->ill_max_hops);
10763 
10764                 /*
10765                  * Synchronize 64- and 32-bit counters
10766                  */
10767                 SYNC32_MIB(ill->ill_ip_mib, ipIfStatsInReceives,
10768                     ipIfStatsHCInReceives);
10769                 SYNC32_MIB(ill->ill_ip_mib, ipIfStatsInDelivers,
10770                     ipIfStatsHCInDelivers);
10771                 SYNC32_MIB(ill->ill_ip_mib, ipIfStatsOutRequests,
10772                     ipIfStatsHCOutRequests);
10773                 SYNC32_MIB(ill->ill_ip_mib, ipIfStatsOutForwDatagrams,
10774                     ipIfStatsHCOutForwDatagrams);
10775                 SYNC32_MIB(ill->ill_ip_mib, ipIfStatsOutMcastPkts,
10776                     ipIfStatsHCOutMcastPkts);
10777                 SYNC32_MIB(ill->ill_ip_mib, ipIfStatsInMcastPkts,
10778                     ipIfStatsHCInMcastPkts);
10779 
10780                 if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10781                     (char *)ill->ill_ip_mib, (int)ise_size)) {
10782                         ip1dbg(("ip_snmp_get_mib2_ip6: failed to allocate "
10783                         "%u bytes\n", (uint_t)ise_size));
10784                 } else if (legacy_req) {
10785                         /* Adjust the EntrySize fields for legacy requests. */
10786                         ise = (mib2_ipIfStatsEntry_t *)(mp_tail->b_wptr -
10787                             (int)ise_size);
10788                         SET_MIB(ise->ipIfStatsEntrySize, ise_size);
10789                         SET_MIB(ise->ipIfStatsAddrEntrySize, iae_size);
10790                 }
10791         }
10792         rw_exit(&ipst->ips_ill_g_lock);
10793 
10794         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10795         ip3dbg(("ip_snmp_get_mib2_ip6: level %d, name %d, len %d\n",
10796             (int)optp->level, (int)optp->name, (int)optp->len));
10797         qreply(q, mpctl);
10798         return (mp2ctl);
10799 }
10800 
10801 /*
10802  * ICMPv6 mib: One per ill
10803  */
10804 static mblk_t *
10805 ip_snmp_get_mib2_icmp6(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10806 {
10807         struct opthdr           *optp;
10808         mblk_t                  *mp2ctl;
10809         ill_t                   *ill;
10810         ill_walk_context_t      ctx;
10811         mblk_t                  *mp_tail = NULL;
10812         /*
10813          * Make a copy of the original message
10814          */
10815         mp2ctl = copymsg(mpctl);
10816 
10817         /* fixed length ICMPv6 structure ... */
10818 
10819         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10820         optp->level = MIB2_ICMP6;
10821         optp->name = 0;
10822         /* Include "unknown interface" icmp6_mib */
10823         ipst->ips_icmp6_mib.ipv6IfIcmpIfIndex =
10824             MIB2_UNKNOWN_INTERFACE; /* netstat flag */
10825         ipst->ips_icmp6_mib.ipv6IfIcmpEntrySize =
10826             sizeof (mib2_ipv6IfIcmpEntry_t);
10827         if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10828             (char *)&ipst->ips_icmp6_mib,
10829             (int)sizeof (ipst->ips_icmp6_mib))) {
10830                 ip1dbg(("ip_snmp_get_mib2_icmp6: failed to allocate %u bytes\n",
10831                     (uint_t)sizeof (ipst->ips_icmp6_mib)));
10832         }
10833 
10834         rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10835         ill = ILL_START_WALK_V6(&ctx, ipst);
10836         for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10837                 ill->ill_icmp6_mib->ipv6IfIcmpIfIndex =
10838                     ill->ill_phyint->phyint_ifindex;
10839                 if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10840                     (char *)ill->ill_icmp6_mib,
10841                     (int)sizeof (*ill->ill_icmp6_mib))) {
10842                         ip1dbg(("ip_snmp_get_mib2_icmp6: failed to allocate "
10843                             "%u bytes\n",
10844                             (uint_t)sizeof (*ill->ill_icmp6_mib)));
10845                 }
10846         }
10847         rw_exit(&ipst->ips_ill_g_lock);
10848 
10849         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10850         ip3dbg(("ip_snmp_get_mib2_icmp6: level %d, name %d, len %d\n",
10851             (int)optp->level, (int)optp->name, (int)optp->len));
10852         qreply(q, mpctl);
10853         return (mp2ctl);
10854 }
10855 
10856 /*
10857  * ire_walk routine to create both ipRouteEntryTable and
10858  * ipRouteAttributeTable in one IRE walk
10859  */
10860 static void
10861 ip_snmp_get2_v4(ire_t *ire, iproutedata_t *ird)
10862 {
10863         ill_t                           *ill;
10864         mib2_ipRouteEntry_t             *re;
10865         mib2_ipAttributeEntry_t         iaes;
10866         tsol_ire_gw_secattr_t           *attrp;
10867         tsol_gc_t                       *gc = NULL;
10868         tsol_gcgrp_t                    *gcgrp = NULL;
10869         ip_stack_t                      *ipst = ire->ire_ipst;
10870 
10871         ASSERT(ire->ire_ipversion == IPV4_VERSION);
10872 
10873         if (!(ird->ird_flags & IRD_REPORT_ALL)) {
10874                 if (ire->ire_testhidden)
10875                         return;
10876                 if (ire->ire_type & IRE_IF_CLONE)
10877                         return;
10878         }
10879 
10880         if ((re = kmem_zalloc(sizeof (*re), KM_NOSLEEP)) == NULL)
10881                 return;
10882 
10883         if ((attrp = ire->ire_gw_secattr) != NULL) {
10884                 mutex_enter(&attrp->igsa_lock);
10885                 if ((gc = attrp->igsa_gc) != NULL) {
10886                         gcgrp = gc->gc_grp;
10887                         ASSERT(gcgrp != NULL);
10888                         rw_enter(&gcgrp->gcgrp_rwlock, RW_READER);
10889                 }
10890                 mutex_exit(&attrp->igsa_lock);
10891         }
10892         /*
10893          * Return all IRE types for route table... let caller pick and choose
10894          */
10895         re->ipRouteDest = ire->ire_addr;
10896         ill = ire->ire_ill;
10897         re->ipRouteIfIndex.o_length = 0;
10898         if (ill != NULL) {
10899                 ill_get_name(ill, re->ipRouteIfIndex.o_bytes, OCTET_LENGTH);
10900                 re->ipRouteIfIndex.o_length =
10901                     mi_strlen(re->ipRouteIfIndex.o_bytes);
10902         }
10903         re->ipRouteMetric1 = -1;
10904         re->ipRouteMetric2 = -1;
10905         re->ipRouteMetric3 = -1;
10906         re->ipRouteMetric4 = -1;
10907 
10908         re->ipRouteNextHop = ire->ire_gateway_addr;
10909         /* indirect(4), direct(3), or invalid(2) */
10910         if (ire->ire_flags & (RTF_REJECT | RTF_BLACKHOLE))
10911                 re->ipRouteType = 2;
10912         else if (ire->ire_type & IRE_ONLINK)
10913                 re->ipRouteType = 3;
10914         else
10915                 re->ipRouteType = 4;
10916 
10917         re->ipRouteProto = -1;
10918         re->ipRouteAge = gethrestime_sec() - ire->ire_create_time;
10919         re->ipRouteMask = ire->ire_mask;
10920         re->ipRouteMetric5 = -1;
10921         re->ipRouteInfo.re_max_frag = ire->ire_metrics.iulp_mtu;
10922         if (ire->ire_ill != NULL && re->ipRouteInfo.re_max_frag == 0)
10923                 re->ipRouteInfo.re_max_frag = ire->ire_ill->ill_mtu;
10924 
10925         re->ipRouteInfo.re_frag_flag = 0;
10926         re->ipRouteInfo.re_rtt               = 0;
10927         re->ipRouteInfo.re_src_addr  = 0;
10928         re->ipRouteInfo.re_ref               = ire->ire_refcnt;
10929         re->ipRouteInfo.re_obpkt     = ire->ire_ob_pkt_count;
10930         re->ipRouteInfo.re_ibpkt     = ire->ire_ib_pkt_count;
10931         re->ipRouteInfo.re_flags     = ire->ire_flags;
10932 
10933         /* Add the IRE_IF_CLONE's counters to their parent IRE_INTERFACE */
10934         if (ire->ire_type & IRE_INTERFACE) {
10935                 ire_t *child;
10936 
10937                 rw_enter(&ipst->ips_ire_dep_lock, RW_READER);
10938                 child = ire->ire_dep_children;
10939                 while (child != NULL) {
10940                         re->ipRouteInfo.re_obpkt += child->ire_ob_pkt_count;
10941                         re->ipRouteInfo.re_ibpkt += child->ire_ib_pkt_count;
10942                         child = child->ire_dep_sib_next;
10943                 }
10944                 rw_exit(&ipst->ips_ire_dep_lock);
10945         }
10946 
10947         if (ire->ire_flags & RTF_DYNAMIC) {
10948                 re->ipRouteInfo.re_ire_type  = IRE_HOST_REDIRECT;
10949         } else {
10950                 re->ipRouteInfo.re_ire_type  = ire->ire_type;
10951         }
10952 
10953         if (!snmp_append_data2(ird->ird_route.lp_head, &ird->ird_route.lp_tail,
10954             (char *)re, (int)sizeof (*re))) {
10955                 ip1dbg(("ip_snmp_get2_v4: failed to allocate %u bytes\n",
10956                     (uint_t)sizeof (*re)));
10957         }
10958 
10959         if (gc != NULL) {
10960                 iaes.iae_routeidx = ird->ird_idx;
10961                 iaes.iae_doi = gc->gc_db->gcdb_doi;
10962                 iaes.iae_slrange = gc->gc_db->gcdb_slrange;
10963 
10964                 if (!snmp_append_data2(ird->ird_attrs.lp_head,
10965                     &ird->ird_attrs.lp_tail, (char *)&iaes, sizeof (iaes))) {
10966                         ip1dbg(("ip_snmp_get2_v4: failed to allocate %u "
10967                             "bytes\n", (uint_t)sizeof (iaes)));
10968                 }
10969         }
10970 
10971         /* bump route index for next pass */
10972         ird->ird_idx++;
10973 
10974         kmem_free(re, sizeof (*re));
10975         if (gcgrp != NULL)
10976                 rw_exit(&gcgrp->gcgrp_rwlock);
10977 }
10978 
10979 /*
10980  * ire_walk routine to create ipv6RouteEntryTable and ipRouteEntryTable.
10981  */
10982 static void
10983 ip_snmp_get2_v6_route(ire_t *ire, iproutedata_t *ird)
10984 {
10985         ill_t                           *ill;
10986         mib2_ipv6RouteEntry_t           *re;
10987         mib2_ipAttributeEntry_t         iaes;
10988         tsol_ire_gw_secattr_t           *attrp;
10989         tsol_gc_t                       *gc = NULL;
10990         tsol_gcgrp_t                    *gcgrp = NULL;
10991         ip_stack_t                      *ipst = ire->ire_ipst;
10992 
10993         ASSERT(ire->ire_ipversion == IPV6_VERSION);
10994 
10995         if (!(ird->ird_flags & IRD_REPORT_ALL)) {
10996                 if (ire->ire_testhidden)
10997                         return;
10998                 if (ire->ire_type & IRE_IF_CLONE)
10999                         return;
11000         }
11001 
11002         if ((re = kmem_zalloc(sizeof (*re), KM_NOSLEEP)) == NULL)
11003                 return;
11004 
11005         if ((attrp = ire->ire_gw_secattr) != NULL) {
11006                 mutex_enter(&attrp->igsa_lock);
11007                 if ((gc = attrp->igsa_gc) != NULL) {
11008                         gcgrp = gc->gc_grp;
11009                         ASSERT(gcgrp != NULL);
11010                         rw_enter(&gcgrp->gcgrp_rwlock, RW_READER);
11011                 }
11012                 mutex_exit(&attrp->igsa_lock);
11013         }
11014         /*
11015          * Return all IRE types for route table... let caller pick and choose
11016          */
11017         re->ipv6RouteDest = ire->ire_addr_v6;
11018         re->ipv6RoutePfxLength = ip_mask_to_plen_v6(&ire->ire_mask_v6);
11019         re->ipv6RouteIndex = 0;      /* Unique when multiple with same dest/plen */
11020         re->ipv6RouteIfIndex.o_length = 0;
11021         ill = ire->ire_ill;
11022         if (ill != NULL) {
11023                 ill_get_name(ill, re->ipv6RouteIfIndex.o_bytes, OCTET_LENGTH);
11024                 re->ipv6RouteIfIndex.o_length =
11025                     mi_strlen(re->ipv6RouteIfIndex.o_bytes);
11026         }
11027 
11028         ASSERT(!(ire->ire_type & IRE_BROADCAST));
11029 
11030         mutex_enter(&ire->ire_lock);
11031         re->ipv6RouteNextHop = ire->ire_gateway_addr_v6;
11032         mutex_exit(&ire->ire_lock);
11033 
11034         /* remote(4), local(3), or discard(2) */
11035         if (ire->ire_flags & (RTF_REJECT | RTF_BLACKHOLE))
11036                 re->ipv6RouteType = 2;
11037         else if (ire->ire_type & IRE_ONLINK)
11038                 re->ipv6RouteType = 3;
11039         else
11040                 re->ipv6RouteType = 4;
11041 
11042         re->ipv6RouteProtocol        = -1;
11043         re->ipv6RoutePolicy  = 0;
11044         re->ipv6RouteAge     = gethrestime_sec() - ire->ire_create_time;
11045         re->ipv6RouteNextHopRDI      = 0;
11046         re->ipv6RouteWeight  = 0;
11047         re->ipv6RouteMetric  = 0;
11048         re->ipv6RouteInfo.re_max_frag = ire->ire_metrics.iulp_mtu;
11049         if (ire->ire_ill != NULL && re->ipv6RouteInfo.re_max_frag == 0)
11050                 re->ipv6RouteInfo.re_max_frag = ire->ire_ill->ill_mtu;
11051 
11052         re->ipv6RouteInfo.re_frag_flag       = 0;
11053         re->ipv6RouteInfo.re_rtt     = 0;
11054         re->ipv6RouteInfo.re_src_addr        = ipv6_all_zeros;
11055         re->ipv6RouteInfo.re_obpkt   = ire->ire_ob_pkt_count;
11056         re->ipv6RouteInfo.re_ibpkt   = ire->ire_ib_pkt_count;
11057         re->ipv6RouteInfo.re_ref     = ire->ire_refcnt;
11058         re->ipv6RouteInfo.re_flags   = ire->ire_flags;
11059 
11060         /* Add the IRE_IF_CLONE's counters to their parent IRE_INTERFACE */
11061         if (ire->ire_type & IRE_INTERFACE) {
11062                 ire_t *child;
11063 
11064                 rw_enter(&ipst->ips_ire_dep_lock, RW_READER);
11065                 child = ire->ire_dep_children;
11066                 while (child != NULL) {
11067                         re->ipv6RouteInfo.re_obpkt += child->ire_ob_pkt_count;
11068                         re->ipv6RouteInfo.re_ibpkt += child->ire_ib_pkt_count;
11069                         child = child->ire_dep_sib_next;
11070                 }
11071                 rw_exit(&ipst->ips_ire_dep_lock);
11072         }
11073         if (ire->ire_flags & RTF_DYNAMIC) {
11074                 re->ipv6RouteInfo.re_ire_type        = IRE_HOST_REDIRECT;
11075         } else {
11076                 re->ipv6RouteInfo.re_ire_type        = ire->ire_type;
11077         }
11078 
11079         if (!snmp_append_data2(ird->ird_route.lp_head, &ird->ird_route.lp_tail,
11080             (char *)re, (int)sizeof (*re))) {
11081                 ip1dbg(("ip_snmp_get2_v6: failed to allocate %u bytes\n",
11082                     (uint_t)sizeof (*re)));
11083         }
11084 
11085         if (gc != NULL) {
11086                 iaes.iae_routeidx = ird->ird_idx;
11087                 iaes.iae_doi = gc->gc_db->gcdb_doi;
11088                 iaes.iae_slrange = gc->gc_db->gcdb_slrange;
11089 
11090                 if (!snmp_append_data2(ird->ird_attrs.lp_head,
11091                     &ird->ird_attrs.lp_tail, (char *)&iaes, sizeof (iaes))) {
11092                         ip1dbg(("ip_snmp_get2_v6: failed to allocate %u "
11093                             "bytes\n", (uint_t)sizeof (iaes)));
11094                 }
11095         }
11096 
11097         /* bump route index for next pass */
11098         ird->ird_idx++;
11099 
11100         kmem_free(re, sizeof (*re));
11101         if (gcgrp != NULL)
11102                 rw_exit(&gcgrp->gcgrp_rwlock);
11103 }
11104 
11105 /*
11106  * ncec_walk routine to create ipv6NetToMediaEntryTable
11107  */
11108 static void
11109 ip_snmp_get2_v6_media(ncec_t *ncec, void *ptr)
11110 {
11111         iproutedata_t *ird              = ptr;
11112         ill_t                           *ill;
11113         mib2_ipv6NetToMediaEntry_t      ntme;
11114 
11115         ill = ncec->ncec_ill;
11116         /* skip arpce entries, and loopback ncec entries */
11117         if (ill->ill_isv6 == B_FALSE || ill->ill_net_type == IRE_LOOPBACK)
11118                 return;
11119         /*
11120          * Neighbor cache entry attached to IRE with on-link
11121          * destination.
11122          * We report all IPMP groups on ncec_ill which is normally the upper.
11123          */
11124         ntme.ipv6NetToMediaIfIndex = ill->ill_phyint->phyint_ifindex;
11125         ntme.ipv6NetToMediaNetAddress = ncec->ncec_addr;
11126         ntme.ipv6NetToMediaPhysAddress.o_length = ill->ill_phys_addr_length;
11127         if (ncec->ncec_lladdr != NULL) {
11128                 bcopy(ncec->ncec_lladdr, ntme.ipv6NetToMediaPhysAddress.o_bytes,
11129                     ntme.ipv6NetToMediaPhysAddress.o_length);
11130         }
11131         /*
11132          * Note: Returns ND_* states. Should be:
11133          * reachable(1), stale(2), delay(3), probe(4),
11134          * invalid(5), unknown(6)
11135          */
11136         ntme.ipv6NetToMediaState = ncec->ncec_state;
11137         ntme.ipv6NetToMediaLastUpdated = 0;
11138 
11139         /* other(1), dynamic(2), static(3), local(4) */
11140         if (NCE_MYADDR(ncec)) {
11141                 ntme.ipv6NetToMediaType = 4;
11142         } else if (ncec->ncec_flags & NCE_F_PUBLISH) {
11143                 ntme.ipv6NetToMediaType = 1; /* proxy */
11144         } else if (ncec->ncec_flags & NCE_F_STATIC) {
11145                 ntme.ipv6NetToMediaType = 3;
11146         } else if (ncec->ncec_flags & (NCE_F_MCAST|NCE_F_BCAST)) {
11147                 ntme.ipv6NetToMediaType = 1;
11148         } else {
11149                 ntme.ipv6NetToMediaType = 2;
11150         }
11151 
11152         if (!snmp_append_data2(ird->ird_netmedia.lp_head,
11153             &ird->ird_netmedia.lp_tail, (char *)&ntme, sizeof (ntme))) {
11154                 ip1dbg(("ip_snmp_get2_v6_media: failed to allocate %u bytes\n",
11155                     (uint_t)sizeof (ntme)));
11156         }
11157 }
11158 
11159 int
11160 nce2ace(ncec_t *ncec)
11161 {
11162         int flags = 0;
11163 
11164         if (NCE_ISREACHABLE(ncec))
11165                 flags |= ACE_F_RESOLVED;
11166         if (ncec->ncec_flags & NCE_F_AUTHORITY)
11167                 flags |= ACE_F_AUTHORITY;
11168         if (ncec->ncec_flags & NCE_F_PUBLISH)
11169                 flags |= ACE_F_PUBLISH;
11170         if ((ncec->ncec_flags & NCE_F_NONUD) != 0)
11171                 flags |= ACE_F_PERMANENT;
11172         if (NCE_MYADDR(ncec))
11173                 flags |= (ACE_F_MYADDR | ACE_F_AUTHORITY);
11174         if (ncec->ncec_flags & NCE_F_UNVERIFIED)
11175                 flags |= ACE_F_UNVERIFIED;
11176         if (ncec->ncec_flags & NCE_F_AUTHORITY)
11177                 flags |= ACE_F_AUTHORITY;
11178         if (ncec->ncec_flags & NCE_F_DELAYED)
11179                 flags |= ACE_F_DELAYED;
11180         return (flags);
11181 }
11182 
11183 /*
11184  * ncec_walk routine to create ipNetToMediaEntryTable
11185  */
11186 static void
11187 ip_snmp_get2_v4_media(ncec_t *ncec, void *ptr)
11188 {
11189         iproutedata_t *ird              = ptr;
11190         ill_t                           *ill;
11191         mib2_ipNetToMediaEntry_t        ntme;
11192         const char                      *name = "unknown";
11193         ipaddr_t                        ncec_addr;
11194 
11195         ill = ncec->ncec_ill;
11196         if (ill->ill_isv6 || (ncec->ncec_flags & NCE_F_BCAST) ||
11197             ill->ill_net_type == IRE_LOOPBACK)
11198                 return;
11199 
11200         /* We report all IPMP groups on ncec_ill which is normally the upper. */
11201         name = ill->ill_name;
11202         /* Based on RFC 4293: other(1), inval(2), dyn(3), stat(4) */
11203         if (NCE_MYADDR(ncec)) {
11204                 ntme.ipNetToMediaType = 4;
11205         } else if (ncec->ncec_flags & (NCE_F_MCAST|NCE_F_BCAST|NCE_F_PUBLISH)) {
11206                 ntme.ipNetToMediaType = 1;
11207         } else {
11208                 ntme.ipNetToMediaType = 3;
11209         }
11210         ntme.ipNetToMediaIfIndex.o_length = MIN(OCTET_LENGTH, strlen(name));
11211         bcopy(name, ntme.ipNetToMediaIfIndex.o_bytes,
11212             ntme.ipNetToMediaIfIndex.o_length);
11213 
11214         IN6_V4MAPPED_TO_IPADDR(&ncec->ncec_addr, ncec_addr);
11215         bcopy(&ncec_addr, &ntme.ipNetToMediaNetAddress, sizeof (ncec_addr));
11216 
11217         ntme.ipNetToMediaInfo.ntm_mask.o_length = sizeof (ipaddr_t);
11218         ncec_addr = INADDR_BROADCAST;
11219         bcopy(&ncec_addr, ntme.ipNetToMediaInfo.ntm_mask.o_bytes,
11220             sizeof (ncec_addr));
11221         /*
11222          * map all the flags to the ACE counterpart.
11223          */
11224         ntme.ipNetToMediaInfo.ntm_flags = nce2ace(ncec);
11225 
11226         ntme.ipNetToMediaPhysAddress.o_length =
11227             MIN(OCTET_LENGTH, ill->ill_phys_addr_length);
11228 
11229         if (!NCE_ISREACHABLE(ncec))
11230                 ntme.ipNetToMediaPhysAddress.o_length = 0;
11231         else {
11232                 if (ncec->ncec_lladdr != NULL) {
11233                         bcopy(ncec->ncec_lladdr,
11234                             ntme.ipNetToMediaPhysAddress.o_bytes,
11235                             ntme.ipNetToMediaPhysAddress.o_length);
11236                 }
11237         }
11238 
11239         if (!snmp_append_data2(ird->ird_netmedia.lp_head,
11240             &ird->ird_netmedia.lp_tail, (char *)&ntme, sizeof (ntme))) {
11241                 ip1dbg(("ip_snmp_get2_v4_media: failed to allocate %u bytes\n",
11242                     (uint_t)sizeof (ntme)));
11243         }
11244 }
11245 
11246 /*
11247  * return (0) if invalid set request, 1 otherwise, including non-tcp requests
11248  */
11249 /* ARGSUSED */
11250 int
11251 ip_snmp_set(queue_t *q, int level, int name, uchar_t *ptr, int len)
11252 {
11253         switch (level) {
11254         case MIB2_IP:
11255         case MIB2_ICMP:
11256                 switch (name) {
11257                 default:
11258                         break;
11259                 }
11260                 return (1);
11261         default:
11262                 return (1);
11263         }
11264 }
11265 
11266 /*
11267  * When there exists both a 64- and 32-bit counter of a particular type
11268  * (i.e., InReceives), only the 64-bit counters are added.
11269  */
11270 void
11271 ip_mib2_add_ip_stats(mib2_ipIfStatsEntry_t *o1, mib2_ipIfStatsEntry_t *o2)
11272 {
11273         UPDATE_MIB(o1, ipIfStatsInHdrErrors, o2->ipIfStatsInHdrErrors);
11274         UPDATE_MIB(o1, ipIfStatsInTooBigErrors, o2->ipIfStatsInTooBigErrors);
11275         UPDATE_MIB(o1, ipIfStatsInNoRoutes, o2->ipIfStatsInNoRoutes);
11276         UPDATE_MIB(o1, ipIfStatsInAddrErrors, o2->ipIfStatsInAddrErrors);
11277         UPDATE_MIB(o1, ipIfStatsInUnknownProtos, o2->ipIfStatsInUnknownProtos);
11278         UPDATE_MIB(o1, ipIfStatsInTruncatedPkts, o2->ipIfStatsInTruncatedPkts);
11279         UPDATE_MIB(o1, ipIfStatsInDiscards, o2->ipIfStatsInDiscards);
11280         UPDATE_MIB(o1, ipIfStatsOutDiscards, o2->ipIfStatsOutDiscards);
11281         UPDATE_MIB(o1, ipIfStatsOutFragOKs, o2->ipIfStatsOutFragOKs);
11282         UPDATE_MIB(o1, ipIfStatsOutFragFails, o2->ipIfStatsOutFragFails);
11283         UPDATE_MIB(o1, ipIfStatsOutFragCreates, o2->ipIfStatsOutFragCreates);
11284         UPDATE_MIB(o1, ipIfStatsReasmReqds, o2->ipIfStatsReasmReqds);
11285         UPDATE_MIB(o1, ipIfStatsReasmOKs, o2->ipIfStatsReasmOKs);
11286         UPDATE_MIB(o1, ipIfStatsReasmFails, o2->ipIfStatsReasmFails);
11287         UPDATE_MIB(o1, ipIfStatsOutNoRoutes, o2->ipIfStatsOutNoRoutes);
11288         UPDATE_MIB(o1, ipIfStatsReasmDuplicates, o2->ipIfStatsReasmDuplicates);
11289         UPDATE_MIB(o1, ipIfStatsReasmPartDups, o2->ipIfStatsReasmPartDups);
11290         UPDATE_MIB(o1, ipIfStatsForwProhibits, o2->ipIfStatsForwProhibits);
11291         UPDATE_MIB(o1, udpInCksumErrs, o2->udpInCksumErrs);
11292         UPDATE_MIB(o1, udpInOverflows, o2->udpInOverflows);
11293         UPDATE_MIB(o1, rawipInOverflows, o2->rawipInOverflows);
11294         UPDATE_MIB(o1, ipIfStatsInWrongIPVersion,
11295             o2->ipIfStatsInWrongIPVersion);
11296         UPDATE_MIB(o1, ipIfStatsOutWrongIPVersion,
11297             o2->ipIfStatsInWrongIPVersion);
11298         UPDATE_MIB(o1, ipIfStatsOutSwitchIPVersion,
11299             o2->ipIfStatsOutSwitchIPVersion);
11300         UPDATE_MIB(o1, ipIfStatsHCInReceives, o2->ipIfStatsHCInReceives);
11301         UPDATE_MIB(o1, ipIfStatsHCInOctets, o2->ipIfStatsHCInOctets);
11302         UPDATE_MIB(o1, ipIfStatsHCInForwDatagrams,
11303             o2->ipIfStatsHCInForwDatagrams);
11304         UPDATE_MIB(o1, ipIfStatsHCInDelivers, o2->ipIfStatsHCInDelivers);
11305         UPDATE_MIB(o1, ipIfStatsHCOutRequests, o2->ipIfStatsHCOutRequests);
11306         UPDATE_MIB(o1, ipIfStatsHCOutForwDatagrams,
11307             o2->ipIfStatsHCOutForwDatagrams);
11308         UPDATE_MIB(o1, ipIfStatsOutFragReqds, o2->ipIfStatsOutFragReqds);
11309         UPDATE_MIB(o1, ipIfStatsHCOutTransmits, o2->ipIfStatsHCOutTransmits);
11310         UPDATE_MIB(o1, ipIfStatsHCOutOctets, o2->ipIfStatsHCOutOctets);
11311         UPDATE_MIB(o1, ipIfStatsHCInMcastPkts, o2->ipIfStatsHCInMcastPkts);
11312         UPDATE_MIB(o1, ipIfStatsHCInMcastOctets, o2->ipIfStatsHCInMcastOctets);
11313         UPDATE_MIB(o1, ipIfStatsHCOutMcastPkts, o2->ipIfStatsHCOutMcastPkts);
11314         UPDATE_MIB(o1, ipIfStatsHCOutMcastOctets,
11315             o2->ipIfStatsHCOutMcastOctets);
11316         UPDATE_MIB(o1, ipIfStatsHCInBcastPkts, o2->ipIfStatsHCInBcastPkts);
11317         UPDATE_MIB(o1, ipIfStatsHCOutBcastPkts, o2->ipIfStatsHCOutBcastPkts);
11318         UPDATE_MIB(o1, ipsecInSucceeded, o2->ipsecInSucceeded);
11319         UPDATE_MIB(o1, ipsecInFailed, o2->ipsecInFailed);
11320         UPDATE_MIB(o1, ipInCksumErrs, o2->ipInCksumErrs);
11321         UPDATE_MIB(o1, tcpInErrs, o2->tcpInErrs);
11322         UPDATE_MIB(o1, udpNoPorts, o2->udpNoPorts);
11323 }
11324 
11325 void
11326 ip_mib2_add_icmp6_stats(mib2_ipv6IfIcmpEntry_t *o1, mib2_ipv6IfIcmpEntry_t *o2)
11327 {
11328         UPDATE_MIB(o1, ipv6IfIcmpInMsgs, o2->ipv6IfIcmpInMsgs);
11329         UPDATE_MIB(o1, ipv6IfIcmpInErrors, o2->ipv6IfIcmpInErrors);
11330         UPDATE_MIB(o1, ipv6IfIcmpInDestUnreachs, o2->ipv6IfIcmpInDestUnreachs);
11331         UPDATE_MIB(o1, ipv6IfIcmpInAdminProhibs, o2->ipv6IfIcmpInAdminProhibs);
11332         UPDATE_MIB(o1, ipv6IfIcmpInTimeExcds, o2->ipv6IfIcmpInTimeExcds);
11333         UPDATE_MIB(o1, ipv6IfIcmpInParmProblems, o2->ipv6IfIcmpInParmProblems);
11334         UPDATE_MIB(o1, ipv6IfIcmpInPktTooBigs, o2->ipv6IfIcmpInPktTooBigs);
11335         UPDATE_MIB(o1, ipv6IfIcmpInEchos, o2->ipv6IfIcmpInEchos);
11336         UPDATE_MIB(o1, ipv6IfIcmpInEchoReplies, o2->ipv6IfIcmpInEchoReplies);
11337         UPDATE_MIB(o1, ipv6IfIcmpInRouterSolicits,
11338             o2->ipv6IfIcmpInRouterSolicits);
11339         UPDATE_MIB(o1, ipv6IfIcmpInRouterAdvertisements,
11340             o2->ipv6IfIcmpInRouterAdvertisements);
11341         UPDATE_MIB(o1, ipv6IfIcmpInNeighborSolicits,
11342             o2->ipv6IfIcmpInNeighborSolicits);
11343         UPDATE_MIB(o1, ipv6IfIcmpInNeighborAdvertisements,
11344             o2->ipv6IfIcmpInNeighborAdvertisements);
11345         UPDATE_MIB(o1, ipv6IfIcmpInRedirects, o2->ipv6IfIcmpInRedirects);
11346         UPDATE_MIB(o1, ipv6IfIcmpInGroupMembQueries,
11347             o2->ipv6IfIcmpInGroupMembQueries);
11348         UPDATE_MIB(o1, ipv6IfIcmpInGroupMembResponses,
11349             o2->ipv6IfIcmpInGroupMembResponses);
11350         UPDATE_MIB(o1, ipv6IfIcmpInGroupMembReductions,
11351             o2->ipv6IfIcmpInGroupMembReductions);
11352         UPDATE_MIB(o1, ipv6IfIcmpOutMsgs, o2->ipv6IfIcmpOutMsgs);
11353         UPDATE_MIB(o1, ipv6IfIcmpOutErrors, o2->ipv6IfIcmpOutErrors);
11354         UPDATE_MIB(o1, ipv6IfIcmpOutDestUnreachs,
11355             o2->ipv6IfIcmpOutDestUnreachs);
11356         UPDATE_MIB(o1, ipv6IfIcmpOutAdminProhibs,
11357             o2->ipv6IfIcmpOutAdminProhibs);
11358         UPDATE_MIB(o1, ipv6IfIcmpOutTimeExcds, o2->ipv6IfIcmpOutTimeExcds);
11359         UPDATE_MIB(o1, ipv6IfIcmpOutParmProblems,
11360             o2->ipv6IfIcmpOutParmProblems);
11361         UPDATE_MIB(o1, ipv6IfIcmpOutPktTooBigs, o2->ipv6IfIcmpOutPktTooBigs);
11362         UPDATE_MIB(o1, ipv6IfIcmpOutEchos, o2->ipv6IfIcmpOutEchos);
11363         UPDATE_MIB(o1, ipv6IfIcmpOutEchoReplies, o2->ipv6IfIcmpOutEchoReplies);
11364         UPDATE_MIB(o1, ipv6IfIcmpOutRouterSolicits,
11365             o2->ipv6IfIcmpOutRouterSolicits);
11366         UPDATE_MIB(o1, ipv6IfIcmpOutRouterAdvertisements,
11367             o2->ipv6IfIcmpOutRouterAdvertisements);
11368         UPDATE_MIB(o1, ipv6IfIcmpOutNeighborSolicits,
11369             o2->ipv6IfIcmpOutNeighborSolicits);
11370         UPDATE_MIB(o1, ipv6IfIcmpOutNeighborAdvertisements,
11371             o2->ipv6IfIcmpOutNeighborAdvertisements);
11372         UPDATE_MIB(o1, ipv6IfIcmpOutRedirects, o2->ipv6IfIcmpOutRedirects);
11373         UPDATE_MIB(o1, ipv6IfIcmpOutGroupMembQueries,
11374             o2->ipv6IfIcmpOutGroupMembQueries);
11375         UPDATE_MIB(o1, ipv6IfIcmpOutGroupMembResponses,
11376             o2->ipv6IfIcmpOutGroupMembResponses);
11377         UPDATE_MIB(o1, ipv6IfIcmpOutGroupMembReductions,
11378             o2->ipv6IfIcmpOutGroupMembReductions);
11379         UPDATE_MIB(o1, ipv6IfIcmpInOverflows, o2->ipv6IfIcmpInOverflows);
11380         UPDATE_MIB(o1, ipv6IfIcmpBadHoplimit, o2->ipv6IfIcmpBadHoplimit);
11381         UPDATE_MIB(o1, ipv6IfIcmpInBadNeighborAdvertisements,
11382             o2->ipv6IfIcmpInBadNeighborAdvertisements);
11383         UPDATE_MIB(o1, ipv6IfIcmpInBadNeighborSolicitations,
11384             o2->ipv6IfIcmpInBadNeighborSolicitations);
11385         UPDATE_MIB(o1, ipv6IfIcmpInBadRedirects, o2->ipv6IfIcmpInBadRedirects);
11386         UPDATE_MIB(o1, ipv6IfIcmpInGroupMembTotal,
11387             o2->ipv6IfIcmpInGroupMembTotal);
11388         UPDATE_MIB(o1, ipv6IfIcmpInGroupMembBadQueries,
11389             o2->ipv6IfIcmpInGroupMembBadQueries);
11390         UPDATE_MIB(o1, ipv6IfIcmpInGroupMembBadReports,
11391             o2->ipv6IfIcmpInGroupMembBadReports);
11392         UPDATE_MIB(o1, ipv6IfIcmpInGroupMembOurReports,
11393             o2->ipv6IfIcmpInGroupMembOurReports);
11394 }
11395 
11396 /*
11397  * Called before the options are updated to check if this packet will
11398  * be source routed from here.
11399  * This routine assumes that the options are well formed i.e. that they
11400  * have already been checked.
11401  */
11402 boolean_t
11403 ip_source_routed(ipha_t *ipha, ip_stack_t *ipst)
11404 {
11405         ipoptp_t        opts;
11406         uchar_t         *opt;
11407         uint8_t         optval;
11408         uint8_t         optlen;
11409         ipaddr_t        dst;
11410 
11411         if (IS_SIMPLE_IPH(ipha)) {
11412                 ip2dbg(("not source routed\n"));
11413                 return (B_FALSE);
11414         }
11415         dst = ipha->ipha_dst;
11416         for (optval = ipoptp_first(&opts, ipha);
11417             optval != IPOPT_EOL;
11418             optval = ipoptp_next(&opts)) {
11419                 ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
11420                 opt = opts.ipoptp_cur;
11421                 optlen = opts.ipoptp_len;
11422                 ip2dbg(("ip_source_routed: opt %d, len %d\n",
11423                     optval, optlen));
11424                 switch (optval) {
11425                         uint32_t off;
11426                 case IPOPT_SSRR:
11427                 case IPOPT_LSRR:
11428                         /*
11429                          * If dst is one of our addresses and there are some
11430                          * entries left in the source route return (true).
11431                          */
11432                         if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
11433                                 ip2dbg(("ip_source_routed: not next"
11434                                     " source route 0x%x\n",
11435                                     ntohl(dst)));
11436                                 return (B_FALSE);
11437                         }
11438                         off = opt[IPOPT_OFFSET];
11439                         off--;
11440                         if (optlen < IP_ADDR_LEN ||
11441                             off > optlen - IP_ADDR_LEN) {
11442                                 /* End of source route */
11443                                 ip1dbg(("ip_source_routed: end of SR\n"));
11444                                 return (B_FALSE);
11445                         }
11446                         return (B_TRUE);
11447                 }
11448         }
11449         ip2dbg(("not source routed\n"));
11450         return (B_FALSE);
11451 }
11452 
11453 /*
11454  * ip_unbind is called by the transports to remove a conn from
11455  * the fanout table.
11456  */
11457 void
11458 ip_unbind(conn_t *connp)
11459 {
11460 
11461         ASSERT(!MUTEX_HELD(&connp->conn_lock));
11462 
11463         if (is_system_labeled() && connp->conn_anon_port) {
11464                 (void) tsol_mlp_anon(crgetzone(connp->conn_cred),
11465                     connp->conn_mlp_type, connp->conn_proto,
11466                     ntohs(connp->conn_lport), B_FALSE);
11467                 connp->conn_anon_port = 0;
11468         }
11469         connp->conn_mlp_type = mlptSingle;
11470 
11471         ipcl_hash_remove(connp);
11472 }
11473 
11474 /*
11475  * Used for deciding the MSS size for the upper layer. Thus
11476  * we need to check the outbound policy values in the conn.
11477  */
11478 int
11479 conn_ipsec_length(conn_t *connp)
11480 {
11481         ipsec_latch_t *ipl;
11482 
11483         ipl = connp->conn_latch;
11484         if (ipl == NULL)
11485                 return (0);
11486 
11487         if (connp->conn_ixa->ixa_ipsec_policy == NULL)
11488                 return (0);
11489 
11490         return (connp->conn_ixa->ixa_ipsec_policy->ipsp_act->ipa_ovhd);
11491 }
11492 
11493 /*
11494  * Returns an estimate of the IPsec headers size. This is used if
11495  * we don't want to call into IPsec to get the exact size.
11496  */
11497 int
11498 ipsec_out_extra_length(ip_xmit_attr_t *ixa)
11499 {
11500         ipsec_action_t *a;
11501 
11502         if (!(ixa->ixa_flags & IXAF_IPSEC_SECURE))
11503                 return (0);
11504 
11505         a = ixa->ixa_ipsec_action;
11506         if (a == NULL) {
11507                 ASSERT(ixa->ixa_ipsec_policy != NULL);
11508                 a = ixa->ixa_ipsec_policy->ipsp_act;
11509         }
11510         ASSERT(a != NULL);
11511 
11512         return (a->ipa_ovhd);
11513 }
11514 
11515 /*
11516  * If there are any source route options, return the true final
11517  * destination. Otherwise, return the destination.
11518  */
11519 ipaddr_t
11520 ip_get_dst(ipha_t *ipha)
11521 {
11522         ipoptp_t        opts;
11523         uchar_t         *opt;
11524         uint8_t         optval;
11525         uint8_t         optlen;
11526         ipaddr_t        dst;
11527         uint32_t off;
11528 
11529         dst = ipha->ipha_dst;
11530 
11531         if (IS_SIMPLE_IPH(ipha))
11532                 return (dst);
11533 
11534         for (optval = ipoptp_first(&opts, ipha);
11535             optval != IPOPT_EOL;
11536             optval = ipoptp_next(&opts)) {
11537                 opt = opts.ipoptp_cur;
11538                 optlen = opts.ipoptp_len;
11539                 ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
11540                 switch (optval) {
11541                 case IPOPT_SSRR:
11542                 case IPOPT_LSRR:
11543                         off = opt[IPOPT_OFFSET];
11544                         /*
11545                          * If one of the conditions is true, it means
11546                          * end of options and dst already has the right
11547                          * value.
11548                          */
11549                         if (!(optlen < IP_ADDR_LEN || off > optlen - 3)) {
11550                                 off = optlen - IP_ADDR_LEN;
11551                                 bcopy(&opt[off], &dst, IP_ADDR_LEN);
11552                         }
11553                         return (dst);
11554                 default:
11555                         break;
11556                 }
11557         }
11558 
11559         return (dst);
11560 }
11561 
11562 /*
11563  * Outbound IP fragmentation routine.
11564  * Assumes the caller has checked whether or not fragmentation should
11565  * be allowed. Here we copy the DF bit from the header to all the generated
11566  * fragments.
11567  */
11568 int
11569 ip_fragment_v4(mblk_t *mp_orig, nce_t *nce, iaflags_t ixaflags,
11570     uint_t pkt_len, uint32_t max_frag, uint32_t xmit_hint, zoneid_t szone,
11571     zoneid_t nolzid, pfirepostfrag_t postfragfn, uintptr_t *ixa_cookie)
11572 {
11573         int             i1;
11574         int             hdr_len;
11575         mblk_t          *hdr_mp;
11576         ipha_t          *ipha;
11577         int             ip_data_end;
11578         int             len;
11579         mblk_t          *mp = mp_orig;
11580         int             offset;
11581         ill_t           *ill = nce->nce_ill;
11582         ip_stack_t      *ipst = ill->ill_ipst;
11583         mblk_t          *carve_mp;
11584         uint32_t        frag_flag;
11585         uint_t          priority = mp->b_band;
11586         int             error = 0;
11587 
11588         BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragReqds);
11589 
11590         if (pkt_len != msgdsize(mp)) {
11591                 ip0dbg(("Packet length mismatch: %d, %ld\n",
11592                     pkt_len, msgdsize(mp)));
11593                 freemsg(mp);
11594                 return (EINVAL);
11595         }
11596 
11597         if (max_frag == 0) {
11598                 ip1dbg(("ip_fragment_v4: max_frag is zero. Dropping packet\n"));
11599                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11600                 ip_drop_output("FragFails: zero max_frag", mp, ill);
11601                 freemsg(mp);
11602                 return (EINVAL);
11603         }
11604 
11605         ASSERT(MBLKL(mp) >= sizeof (ipha_t));
11606         ipha = (ipha_t *)mp->b_rptr;
11607         ASSERT(ntohs(ipha->ipha_length) == pkt_len);
11608         frag_flag = ntohs(ipha->ipha_fragment_offset_and_flags) & IPH_DF;
11609 
11610         /*
11611          * Establish the starting offset.  May not be zero if we are fragging
11612          * a fragment that is being forwarded.
11613          */
11614         offset = ntohs(ipha->ipha_fragment_offset_and_flags) & IPH_OFFSET;
11615 
11616         /* TODO why is this test needed? */
11617         if (((max_frag - ntohs(ipha->ipha_length)) & ~7) < 8) {
11618                 /* TODO: notify ulp somehow */
11619                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11620                 ip_drop_output("FragFails: bad starting offset", mp, ill);
11621                 freemsg(mp);
11622                 return (EINVAL);
11623         }
11624 
11625         hdr_len = IPH_HDR_LENGTH(ipha);
11626         ipha->ipha_hdr_checksum = 0;
11627 
11628         /*
11629          * Establish the number of bytes maximum per frag, after putting
11630          * in the header.
11631          */
11632         len = (max_frag - hdr_len) & ~7;
11633 
11634         /* Get a copy of the header for the trailing frags */
11635         hdr_mp = ip_fragment_copyhdr((uchar_t *)ipha, hdr_len, offset, ipst,
11636             mp);
11637         if (hdr_mp == NULL) {
11638                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11639                 ip_drop_output("FragFails: no hdr_mp", mp, ill);
11640                 freemsg(mp);
11641                 return (ENOBUFS);
11642         }
11643 
11644         /* Store the starting offset, with the MoreFrags flag. */
11645         i1 = offset | IPH_MF | frag_flag;
11646         ipha->ipha_fragment_offset_and_flags = htons((uint16_t)i1);
11647 
11648         /* Establish the ending byte offset, based on the starting offset. */
11649         offset <<= 3;
11650         ip_data_end = offset + ntohs(ipha->ipha_length) - hdr_len;
11651 
11652         /* Store the length of the first fragment in the IP header. */
11653         i1 = len + hdr_len;
11654         ASSERT(i1 <= IP_MAXPACKET);
11655         ipha->ipha_length = htons((uint16_t)i1);
11656 
11657         /*
11658          * Compute the IP header checksum for the first frag.  We have to
11659          * watch out that we stop at the end of the header.
11660          */
11661         ipha->ipha_hdr_checksum = ip_csum_hdr(ipha);
11662 
11663         /*
11664          * Now carve off the first frag.  Note that this will include the
11665          * original IP header.
11666          */
11667         if (!(mp = ip_carve_mp(&mp_orig, i1))) {
11668                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11669                 ip_drop_output("FragFails: could not carve mp", mp_orig, ill);
11670                 freeb(hdr_mp);
11671                 freemsg(mp_orig);
11672                 return (ENOBUFS);
11673         }
11674 
11675         BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragCreates);
11676 
11677         error = postfragfn(mp, nce, ixaflags, i1, xmit_hint, szone, nolzid,
11678             ixa_cookie);
11679         if (error != 0 && error != EWOULDBLOCK) {
11680                 /* No point in sending the other fragments */
11681                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11682                 ip_drop_output("FragFails: postfragfn failed", mp_orig, ill);
11683                 freeb(hdr_mp);
11684                 freemsg(mp_orig);
11685                 return (error);
11686         }
11687 
11688         /* No need to redo state machine in loop */
11689         ixaflags &= ~IXAF_REACH_CONF;
11690 
11691         /* Advance the offset to the second frag starting point. */
11692         offset += len;
11693         /*
11694          * Update hdr_len from the copied header - there might be less options
11695          * in the later fragments.
11696          */
11697         hdr_len = IPH_HDR_LENGTH(hdr_mp->b_rptr);
11698         /* Loop until done. */
11699         for (;;) {
11700                 uint16_t        offset_and_flags;
11701                 uint16_t        ip_len;
11702 
11703                 if (ip_data_end - offset > len) {
11704                         /*
11705                          * Carve off the appropriate amount from the original
11706                          * datagram.
11707                          */
11708                         if (!(carve_mp = ip_carve_mp(&mp_orig, len))) {
11709                                 mp = NULL;
11710                                 break;
11711                         }
11712                         /*
11713                          * More frags after this one.  Get another copy
11714                          * of the header.
11715                          */
11716                         if (carve_mp->b_datap->db_ref == 1 &&
11717                             hdr_mp->b_wptr - hdr_mp->b_rptr <
11718                             carve_mp->b_rptr - carve_mp->b_datap->db_base) {
11719                                 /* Inline IP header */
11720                                 carve_mp->b_rptr -= hdr_mp->b_wptr -
11721                                     hdr_mp->b_rptr;
11722                                 bcopy(hdr_mp->b_rptr, carve_mp->b_rptr,
11723                                     hdr_mp->b_wptr - hdr_mp->b_rptr);
11724                                 mp = carve_mp;
11725                         } else {
11726                                 if (!(mp = copyb(hdr_mp))) {
11727                                         freemsg(carve_mp);
11728                                         break;
11729                                 }
11730                                 /* Get priority marking, if any. */
11731                                 mp->b_band = priority;
11732                                 mp->b_cont = carve_mp;
11733                         }
11734                         ipha = (ipha_t *)mp->b_rptr;
11735                         offset_and_flags = IPH_MF;
11736                 } else {
11737                         /*
11738                          * Last frag.  Consume the header. Set len to
11739                          * the length of this last piece.
11740                          */
11741                         len = ip_data_end - offset;
11742 
11743                         /*
11744                          * Carve off the appropriate amount from the original
11745                          * datagram.
11746                          */
11747                         if (!(carve_mp = ip_carve_mp(&mp_orig, len))) {
11748                                 mp = NULL;
11749                                 break;
11750                         }
11751                         if (carve_mp->b_datap->db_ref == 1 &&
11752                             hdr_mp->b_wptr - hdr_mp->b_rptr <
11753                             carve_mp->b_rptr - carve_mp->b_datap->db_base) {
11754                                 /* Inline IP header */
11755                                 carve_mp->b_rptr -= hdr_mp->b_wptr -
11756                                     hdr_mp->b_rptr;
11757                                 bcopy(hdr_mp->b_rptr, carve_mp->b_rptr,
11758                                     hdr_mp->b_wptr - hdr_mp->b_rptr);
11759                                 mp = carve_mp;
11760                                 freeb(hdr_mp);
11761                                 hdr_mp = mp;
11762                         } else {
11763                                 mp = hdr_mp;
11764                                 /* Get priority marking, if any. */
11765                                 mp->b_band = priority;
11766                                 mp->b_cont = carve_mp;
11767                         }
11768                         ipha = (ipha_t *)mp->b_rptr;
11769                         /* A frag of a frag might have IPH_MF non-zero */
11770                         offset_and_flags =
11771                             ntohs(ipha->ipha_fragment_offset_and_flags) &
11772                             IPH_MF;
11773                 }
11774                 offset_and_flags |= (uint16_t)(offset >> 3);
11775                 offset_and_flags |= (uint16_t)frag_flag;
11776                 /* Store the offset and flags in the IP header. */
11777                 ipha->ipha_fragment_offset_and_flags = htons(offset_and_flags);
11778 
11779                 /* Store the length in the IP header. */
11780                 ip_len = (uint16_t)(len + hdr_len);
11781                 ipha->ipha_length = htons(ip_len);
11782 
11783                 /*
11784                  * Set the IP header checksum.  Note that mp is just
11785                  * the header, so this is easy to pass to ip_csum.
11786                  */
11787                 ipha->ipha_hdr_checksum = ip_csum_hdr(ipha);
11788 
11789                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragCreates);
11790 
11791                 error = postfragfn(mp, nce, ixaflags, ip_len, xmit_hint, szone,
11792                     nolzid, ixa_cookie);
11793                 /* All done if we just consumed the hdr_mp. */
11794                 if (mp == hdr_mp) {
11795                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragOKs);
11796                         return (error);
11797                 }
11798                 if (error != 0 && error != EWOULDBLOCK) {
11799                         DTRACE_PROBE2(ip__xmit__frag__fail, ill_t *, ill,
11800                             mblk_t *, hdr_mp);
11801                         /* No point in sending the other fragments */
11802                         break;
11803                 }
11804 
11805                 /* Otherwise, advance and loop. */
11806                 offset += len;
11807         }
11808         /* Clean up following allocation failure. */
11809         BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11810         ip_drop_output("FragFails: loop ended", NULL, ill);
11811         if (mp != hdr_mp)
11812                 freeb(hdr_mp);
11813         if (mp != mp_orig)
11814                 freemsg(mp_orig);
11815         return (error);
11816 }
11817 
11818 /*
11819  * Copy the header plus those options which have the copy bit set
11820  */
11821 static mblk_t *
11822 ip_fragment_copyhdr(uchar_t *rptr, int hdr_len, int offset, ip_stack_t *ipst,
11823     mblk_t *src)
11824 {
11825         mblk_t  *mp;
11826         uchar_t *up;
11827 
11828         /*
11829          * Quick check if we need to look for options without the copy bit
11830          * set
11831          */
11832         mp = allocb_tmpl(ipst->ips_ip_wroff_extra + hdr_len, src);
11833         if (!mp)
11834                 return (mp);
11835         mp->b_rptr += ipst->ips_ip_wroff_extra;
11836         if (hdr_len == IP_SIMPLE_HDR_LENGTH || offset != 0) {
11837                 bcopy(rptr, mp->b_rptr, hdr_len);
11838                 mp->b_wptr += hdr_len + ipst->ips_ip_wroff_extra;
11839                 return (mp);
11840         }
11841         up  = mp->b_rptr;
11842         bcopy(rptr, up, IP_SIMPLE_HDR_LENGTH);
11843         up += IP_SIMPLE_HDR_LENGTH;
11844         rptr += IP_SIMPLE_HDR_LENGTH;
11845         hdr_len -= IP_SIMPLE_HDR_LENGTH;
11846         while (hdr_len > 0) {
11847                 uint32_t optval;
11848                 uint32_t optlen;
11849 
11850                 optval = *rptr;
11851                 if (optval == IPOPT_EOL)
11852                         break;
11853                 if (optval == IPOPT_NOP)
11854                         optlen = 1;
11855                 else
11856                         optlen = rptr[1];
11857                 if (optval & IPOPT_COPY) {
11858                         bcopy(rptr, up, optlen);
11859                         up += optlen;
11860                 }
11861                 rptr += optlen;
11862                 hdr_len -= optlen;
11863         }
11864         /*
11865          * Make sure that we drop an even number of words by filling
11866          * with EOL to the next word boundary.
11867          */
11868         for (hdr_len = up - (mp->b_rptr + IP_SIMPLE_HDR_LENGTH);
11869             hdr_len & 0x3; hdr_len++)
11870                 *up++ = IPOPT_EOL;
11871         mp->b_wptr = up;
11872         /* Update header length */
11873         mp->b_rptr[0] = (uint8_t)((IP_VERSION << 4) | ((up - mp->b_rptr) >> 2));
11874         return (mp);
11875 }
11876 
11877 /*
11878  * Update any source route, record route, or timestamp options when
11879  * sending a packet back to ourselves.
11880  * Check that we are at end of strict source route.
11881  * The options have been sanity checked by ip_output_options().
11882  */
11883 void
11884 ip_output_local_options(ipha_t *ipha, ip_stack_t *ipst)
11885 {
11886         ipoptp_t        opts;
11887         uchar_t         *opt;
11888         uint8_t         optval;
11889         uint8_t         optlen;
11890         ipaddr_t        dst;
11891         uint32_t        ts;
11892         timestruc_t     now;
11893 
11894         for (optval = ipoptp_first(&opts, ipha);
11895             optval != IPOPT_EOL;
11896             optval = ipoptp_next(&opts)) {
11897                 opt = opts.ipoptp_cur;
11898                 optlen = opts.ipoptp_len;
11899                 ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
11900                 switch (optval) {
11901                         uint32_t off;
11902                 case IPOPT_SSRR:
11903                 case IPOPT_LSRR:
11904                         off = opt[IPOPT_OFFSET];
11905                         off--;
11906                         if (optlen < IP_ADDR_LEN ||
11907                             off > optlen - IP_ADDR_LEN) {
11908                                 /* End of source route */
11909                                 break;
11910                         }
11911                         /*
11912                          * This will only happen if two consecutive entries
11913                          * in the source route contains our address or if
11914                          * it is a packet with a loose source route which
11915                          * reaches us before consuming the whole source route
11916                          */
11917 
11918                         if (optval == IPOPT_SSRR) {
11919                                 return;
11920                         }
11921                         /*
11922                          * Hack: instead of dropping the packet truncate the
11923                          * source route to what has been used by filling the
11924                          * rest with IPOPT_NOP.
11925                          */
11926                         opt[IPOPT_OLEN] = (uint8_t)off;
11927                         while (off < optlen) {
11928                                 opt[off++] = IPOPT_NOP;
11929                         }
11930                         break;
11931                 case IPOPT_RR:
11932                         off = opt[IPOPT_OFFSET];
11933                         off--;
11934                         if (optlen < IP_ADDR_LEN ||
11935                             off > optlen - IP_ADDR_LEN) {
11936                                 /* No more room - ignore */
11937                                 ip1dbg((
11938                                     "ip_output_local_options: end of RR\n"));
11939                                 break;
11940                         }
11941                         dst = htonl(INADDR_LOOPBACK);
11942                         bcopy(&dst, (char *)opt + off, IP_ADDR_LEN);
11943                         opt[IPOPT_OFFSET] += IP_ADDR_LEN;
11944                         break;
11945                 case IPOPT_TS:
11946                         /* Insert timestamp if there is romm */
11947                         switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
11948                         case IPOPT_TS_TSONLY:
11949                                 off = IPOPT_TS_TIMELEN;
11950                                 break;
11951                         case IPOPT_TS_PRESPEC:
11952                         case IPOPT_TS_PRESPEC_RFC791:
11953                                 /* Verify that the address matched */
11954                                 off = opt[IPOPT_OFFSET] - 1;
11955                                 bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
11956                                 if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
11957                                         /* Not for us */
11958                                         break;
11959                                 }
11960                                 /* FALLTHROUGH */
11961                         case IPOPT_TS_TSANDADDR:
11962                                 off = IP_ADDR_LEN + IPOPT_TS_TIMELEN;
11963                                 break;
11964                         default:
11965                                 /*
11966                                  * ip_*put_options should have already
11967                                  * dropped this packet.
11968                                  */
11969                                 cmn_err(CE_PANIC, "ip_output_local_options: "
11970                                     "unknown IT - bug in ip_output_options?\n");
11971                         }
11972                         if (opt[IPOPT_OFFSET] - 1 + off > optlen) {
11973                                 /* Increase overflow counter */
11974                                 off = (opt[IPOPT_POS_OV_FLG] >> 4) + 1;
11975                                 opt[IPOPT_POS_OV_FLG] = (uint8_t)
11976                                     (opt[IPOPT_POS_OV_FLG] & 0x0F) |
11977                                     (off << 4);
11978                                 break;
11979                         }
11980                         off = opt[IPOPT_OFFSET] - 1;
11981                         switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
11982                         case IPOPT_TS_PRESPEC:
11983                         case IPOPT_TS_PRESPEC_RFC791:
11984                         case IPOPT_TS_TSANDADDR:
11985                                 dst = htonl(INADDR_LOOPBACK);
11986                                 bcopy(&dst, (char *)opt + off, IP_ADDR_LEN);
11987                                 opt[IPOPT_OFFSET] += IP_ADDR_LEN;
11988                                 /* FALLTHROUGH */
11989                         case IPOPT_TS_TSONLY:
11990                                 off = opt[IPOPT_OFFSET] - 1;
11991                                 /* Compute # of milliseconds since midnight */
11992                                 gethrestime(&now);
11993                                 ts = (now.tv_sec % (24 * 60 * 60)) * 1000 +
11994                                     NSEC2MSEC(now.tv_nsec);
11995                                 bcopy(&ts, (char *)opt + off, IPOPT_TS_TIMELEN);
11996                                 opt[IPOPT_OFFSET] += IPOPT_TS_TIMELEN;
11997                                 break;
11998                         }
11999                         break;
12000                 }
12001         }
12002 }
12003 
12004 /*
12005  * Prepend an M_DATA fastpath header, and if none present prepend a
12006  * DL_UNITDATA_REQ. Frees the mblk on failure.
12007  *
12008  * nce_dlur_mp and nce_fp_mp can not disappear once they have been set.
12009  * If there is a change to them, the nce will be deleted (condemned) and
12010  * a new nce_t will be created when packets are sent. Thus we need no locks
12011  * to access those fields.
12012  *
12013  * We preserve b_band to support IPQoS. If a DL_UNITDATA_REQ is prepended
12014  * we place b_band in dl_priority.dl_max.
12015  */
12016 static mblk_t *
12017 ip_xmit_attach_llhdr(mblk_t *mp, nce_t *nce)
12018 {
12019         uint_t  hlen;
12020         mblk_t *mp1;
12021         uint_t  priority;
12022         uchar_t *rptr;
12023 
12024         rptr = mp->b_rptr;
12025 
12026         ASSERT(DB_TYPE(mp) == M_DATA);
12027         priority = mp->b_band;
12028 
12029         ASSERT(nce != NULL);
12030         if ((mp1 = nce->nce_fp_mp) != NULL) {
12031                 hlen = MBLKL(mp1);
12032                 /*
12033                  * Check if we have enough room to prepend fastpath
12034                  * header
12035                  */
12036                 if (hlen != 0 && (rptr - mp->b_datap->db_base) >= hlen) {
12037                         rptr -= hlen;
12038                         bcopy(mp1->b_rptr, rptr, hlen);
12039                         /*
12040                          * Set the b_rptr to the start of the link layer
12041                          * header
12042                          */
12043                         mp->b_rptr = rptr;
12044                         return (mp);
12045                 }
12046                 mp1 = copyb(mp1);
12047                 if (mp1 == NULL) {
12048                         ill_t *ill = nce->nce_ill;
12049 
12050                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
12051                         ip_drop_output("ipIfStatsOutDiscards", mp, ill);
12052                         freemsg(mp);
12053                         return (NULL);
12054                 }
12055                 mp1->b_band = priority;
12056                 mp1->b_cont = mp;
12057                 DB_CKSUMSTART(mp1) = DB_CKSUMSTART(mp);
12058                 DB_CKSUMSTUFF(mp1) = DB_CKSUMSTUFF(mp);
12059                 DB_CKSUMEND(mp1) = DB_CKSUMEND(mp);
12060                 DB_CKSUMFLAGS(mp1) = DB_CKSUMFLAGS(mp);
12061                 DB_LSOMSS(mp1) = DB_LSOMSS(mp);
12062                 DTRACE_PROBE1(ip__xmit__copyb, (mblk_t *), mp1);
12063                 /*
12064                  * XXX disable ICK_VALID and compute checksum
12065                  * here; can happen if nce_fp_mp changes and
12066                  * it can't be copied now due to insufficient
12067                  * space. (unlikely, fp mp can change, but it
12068                  * does not increase in length)
12069                  */
12070                 return (mp1);
12071         }
12072         mp1 = copyb(nce->nce_dlur_mp);
12073 
12074         if (mp1 == NULL) {
12075                 ill_t *ill = nce->nce_ill;
12076 
12077                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
12078                 ip_drop_output("ipIfStatsOutDiscards", mp, ill);
12079                 freemsg(mp);
12080                 return (NULL);
12081         }
12082         mp1->b_cont = mp;
12083         if (priority != 0) {
12084                 mp1->b_band = priority;
12085                 ((dl_unitdata_req_t *)(mp1->b_rptr))->dl_priority.dl_max =
12086                     priority;
12087         }
12088         return (mp1);
12089 }
12090 
12091 /*
12092  * Finish the outbound IPsec processing. This function is called from
12093  * ipsec_out_process() if the IPsec packet was processed
12094  * synchronously, or from {ah,esp}_kcf_callback_outbound() if it was processed
12095  * asynchronously.
12096  *
12097  * This is common to IPv4 and IPv6.
12098  */
12099 int
12100 ip_output_post_ipsec(mblk_t *mp, ip_xmit_attr_t *ixa)
12101 {
12102         iaflags_t       ixaflags = ixa->ixa_flags;
12103         uint_t          pktlen;
12104 
12105 
12106         /* AH/ESP don't update ixa_pktlen when they modify the packet */
12107         if (ixaflags & IXAF_IS_IPV4) {
12108                 ipha_t          *ipha = (ipha_t *)mp->b_rptr;
12109 
12110                 ASSERT(IPH_HDR_VERSION(ipha) == IPV4_VERSION);
12111                 pktlen = ntohs(ipha->ipha_length);
12112         } else {
12113                 ip6_t           *ip6h = (ip6_t *)mp->b_rptr;
12114 
12115                 ASSERT(IPH_HDR_VERSION(mp->b_rptr) == IPV6_VERSION);
12116                 pktlen = ntohs(ip6h->ip6_plen) + IPV6_HDR_LEN;
12117         }
12118 
12119         /*
12120          * We release any hard reference on the SAs here to make
12121          * sure the SAs can be garbage collected. ipsr_sa has a soft reference
12122          * on the SAs.
12123          * If in the future we want the hard latching of the SAs in the
12124          * ip_xmit_attr_t then we should remove this.
12125          */
12126         if (ixa->ixa_ipsec_esp_sa != NULL) {
12127                 IPSA_REFRELE(ixa->ixa_ipsec_esp_sa);
12128                 ixa->ixa_ipsec_esp_sa = NULL;
12129         }
12130         if (ixa->ixa_ipsec_ah_sa != NULL) {
12131                 IPSA_REFRELE(ixa->ixa_ipsec_ah_sa);
12132                 ixa->ixa_ipsec_ah_sa = NULL;
12133         }
12134 
12135         /* Do we need to fragment? */
12136         if ((ixa->ixa_flags & IXAF_IPV6_ADD_FRAGHDR) ||
12137             pktlen > ixa->ixa_fragsize) {
12138                 if (ixaflags & IXAF_IS_IPV4) {
12139                         ASSERT(!(ixa->ixa_flags & IXAF_IPV6_ADD_FRAGHDR));
12140                         /*
12141                          * We check for the DF case in ipsec_out_process
12142                          * hence this only handles the non-DF case.
12143                          */
12144                         return (ip_fragment_v4(mp, ixa->ixa_nce, ixa->ixa_flags,
12145                             pktlen, ixa->ixa_fragsize,
12146                             ixa->ixa_xmit_hint, ixa->ixa_zoneid,
12147                             ixa->ixa_no_loop_zoneid, ixa->ixa_postfragfn,
12148                             &ixa->ixa_cookie));
12149                 } else {
12150                         mp = ip_fraghdr_add_v6(mp, ixa->ixa_ident, ixa);
12151                         if (mp == NULL) {
12152                                 /* MIB and ip_drop_output already done */
12153                                 return (ENOMEM);
12154                         }
12155                         pktlen += sizeof (ip6_frag_t);
12156                         if (pktlen > ixa->ixa_fragsize) {
12157                                 return (ip_fragment_v6(mp, ixa->ixa_nce,
12158                                     ixa->ixa_flags, pktlen,
12159                                     ixa->ixa_fragsize, ixa->ixa_xmit_hint,
12160                                     ixa->ixa_zoneid, ixa->ixa_no_loop_zoneid,
12161                                     ixa->ixa_postfragfn, &ixa->ixa_cookie));
12162                         }
12163                 }
12164         }
12165         return ((ixa->ixa_postfragfn)(mp, ixa->ixa_nce, ixa->ixa_flags,
12166             pktlen, ixa->ixa_xmit_hint, ixa->ixa_zoneid,
12167             ixa->ixa_no_loop_zoneid, NULL));
12168 }
12169 
12170 /*
12171  * Finish the inbound IPsec processing. This function is called from
12172  * ipsec_out_process() if the IPsec packet was processed
12173  * synchronously, or from {ah,esp}_kcf_callback_outbound() if it was processed
12174  * asynchronously.
12175  *
12176  * This is common to IPv4 and IPv6.
12177  */
12178 void
12179 ip_input_post_ipsec(mblk_t *mp, ip_recv_attr_t *ira)
12180 {
12181         iaflags_t       iraflags = ira->ira_flags;
12182 
12183         /* Length might have changed */
12184         if (iraflags & IRAF_IS_IPV4) {
12185                 ipha_t          *ipha = (ipha_t *)mp->b_rptr;
12186 
12187                 ASSERT(IPH_HDR_VERSION(ipha) == IPV4_VERSION);
12188                 ira->ira_pktlen = ntohs(ipha->ipha_length);
12189                 ira->ira_ip_hdr_length = IPH_HDR_LENGTH(ipha);
12190                 ira->ira_protocol = ipha->ipha_protocol;
12191 
12192                 ip_fanout_v4(mp, ipha, ira);
12193         } else {
12194                 ip6_t           *ip6h = (ip6_t *)mp->b_rptr;
12195                 uint8_t         *nexthdrp;
12196 
12197                 ASSERT(IPH_HDR_VERSION(mp->b_rptr) == IPV6_VERSION);
12198                 ira->ira_pktlen = ntohs(ip6h->ip6_plen) + IPV6_HDR_LEN;
12199                 if (!ip_hdr_length_nexthdr_v6(mp, ip6h, &ira->ira_ip_hdr_length,
12200                     &nexthdrp)) {
12201                         /* Malformed packet */
12202                         BUMP_MIB(ira->ira_ill->ill_ip_mib, ipIfStatsInDiscards);
12203                         ip_drop_input("ipIfStatsInDiscards", mp, ira->ira_ill);
12204                         freemsg(mp);
12205                         return;
12206                 }
12207                 ira->ira_protocol = *nexthdrp;
12208                 ip_fanout_v6(mp, ip6h, ira);
12209         }
12210 }
12211 
12212 /*
12213  * Select which AH & ESP SA's to use (if any) for the outbound packet.
12214  *
12215  * If this function returns B_TRUE, the requested SA's have been filled
12216  * into the ixa_ipsec_*_sa pointers.
12217  *
12218  * If the function returns B_FALSE, the packet has been "consumed", most
12219  * likely by an ACQUIRE sent up via PF_KEY to a key management daemon.
12220  *
12221  * The SA references created by the protocol-specific "select"
12222  * function will be released in ip_output_post_ipsec.
12223  */
12224 static boolean_t
12225 ipsec_out_select_sa(mblk_t *mp, ip_xmit_attr_t *ixa)
12226 {
12227         boolean_t need_ah_acquire = B_FALSE, need_esp_acquire = B_FALSE;
12228         ipsec_policy_t *pp;
12229         ipsec_action_t *ap;
12230 
12231         ASSERT(ixa->ixa_flags & IXAF_IPSEC_SECURE);
12232         ASSERT((ixa->ixa_ipsec_policy != NULL) ||
12233             (ixa->ixa_ipsec_action != NULL));
12234 
12235         ap = ixa->ixa_ipsec_action;
12236         if (ap == NULL) {
12237                 pp = ixa->ixa_ipsec_policy;
12238                 ASSERT(pp != NULL);
12239                 ap = pp->ipsp_act;
12240                 ASSERT(ap != NULL);
12241         }
12242 
12243         /*
12244          * We have an action.  now, let's select SA's.
12245          * A side effect of setting ixa_ipsec_*_sa is that it will
12246          * be cached in the conn_t.
12247          */
12248         if (ap->ipa_want_esp) {
12249                 if (ixa->ixa_ipsec_esp_sa == NULL) {
12250                         need_esp_acquire = !ipsec_outbound_sa(mp, ixa,
12251                             IPPROTO_ESP);
12252                 }
12253                 ASSERT(need_esp_acquire || ixa->ixa_ipsec_esp_sa != NULL);
12254         }
12255 
12256         if (ap->ipa_want_ah) {
12257                 if (ixa->ixa_ipsec_ah_sa == NULL) {
12258                         need_ah_acquire = !ipsec_outbound_sa(mp, ixa,
12259                             IPPROTO_AH);
12260                 }
12261                 ASSERT(need_ah_acquire || ixa->ixa_ipsec_ah_sa != NULL);
12262                 /*
12263                  * The ESP and AH processing order needs to be preserved
12264                  * when both protocols are required (ESP should be applied
12265                  * before AH for an outbound packet). Force an ESP ACQUIRE
12266                  * when both ESP and AH are required, and an AH ACQUIRE
12267                  * is needed.
12268                  */
12269                 if (ap->ipa_want_esp && need_ah_acquire)
12270                         need_esp_acquire = B_TRUE;
12271         }
12272 
12273         /*
12274          * Send an ACQUIRE (extended, regular, or both) if we need one.
12275          * Release SAs that got referenced, but will not be used until we
12276          * acquire _all_ of the SAs we need.
12277          */
12278         if (need_ah_acquire || need_esp_acquire) {
12279                 if (ixa->ixa_ipsec_ah_sa != NULL) {
12280                         IPSA_REFRELE(ixa->ixa_ipsec_ah_sa);
12281                         ixa->ixa_ipsec_ah_sa = NULL;
12282                 }
12283                 if (ixa->ixa_ipsec_esp_sa != NULL) {
12284                         IPSA_REFRELE(ixa->ixa_ipsec_esp_sa);
12285                         ixa->ixa_ipsec_esp_sa = NULL;
12286                 }
12287 
12288                 sadb_acquire(mp, ixa, need_ah_acquire, need_esp_acquire);
12289                 return (B_FALSE);
12290         }
12291 
12292         return (B_TRUE);
12293 }
12294 
12295 /*
12296  * Handle IPsec output processing.
12297  * This function is only entered once for a given packet.
12298  * We try to do things synchronously, but if we need to have user-level
12299  * set up SAs, or ESP or AH uses asynchronous kEF, then the operation
12300  * will be completed
12301  *  - when the SAs are added in esp_add_sa_finish/ah_add_sa_finish
12302  *  - when asynchronous ESP is done it will do AH
12303  *
12304  * In all cases we come back in ip_output_post_ipsec() to fragment and
12305  * send out the packet.
12306  */
12307 int
12308 ipsec_out_process(mblk_t *mp, ip_xmit_attr_t *ixa)
12309 {
12310         ill_t           *ill = ixa->ixa_nce->nce_ill;
12311         ip_stack_t      *ipst = ixa->ixa_ipst;
12312         ipsec_stack_t   *ipss;
12313         ipsec_policy_t  *pp;
12314         ipsec_action_t  *ap;
12315 
12316         ASSERT(ixa->ixa_flags & IXAF_IPSEC_SECURE);
12317 
12318         ASSERT((ixa->ixa_ipsec_policy != NULL) ||
12319             (ixa->ixa_ipsec_action != NULL));
12320 
12321         ipss = ipst->ips_netstack->netstack_ipsec;
12322         if (!ipsec_loaded(ipss)) {
12323                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
12324                 ip_drop_packet(mp, B_TRUE, ill,
12325                     DROPPER(ipss, ipds_ip_ipsec_not_loaded),
12326                     &ipss->ipsec_dropper);
12327                 return (ENOTSUP);
12328         }
12329 
12330         ap = ixa->ixa_ipsec_action;
12331         if (ap == NULL) {
12332                 pp = ixa->ixa_ipsec_policy;
12333                 ASSERT(pp != NULL);
12334                 ap = pp->ipsp_act;
12335                 ASSERT(ap != NULL);
12336         }
12337 
12338         /* Handle explicit drop action and bypass. */
12339         switch (ap->ipa_act.ipa_type) {
12340         case IPSEC_ACT_DISCARD:
12341         case IPSEC_ACT_REJECT:
12342                 ip_drop_packet(mp, B_FALSE, ill,
12343                     DROPPER(ipss, ipds_spd_explicit), &ipss->ipsec_spd_dropper);
12344                 return (EHOSTUNREACH);  /* IPsec policy failure */
12345         case IPSEC_ACT_BYPASS:
12346                 return (ip_output_post_ipsec(mp, ixa));
12347         }
12348 
12349         /*
12350          * The order of processing is first insert a IP header if needed.
12351          * Then insert the ESP header and then the AH header.
12352          */
12353         if ((ixa->ixa_flags & IXAF_IS_IPV4) && ap->ipa_want_se) {
12354                 /*
12355                  * First get the outer IP header before sending
12356                  * it to ESP.
12357                  */
12358                 ipha_t *oipha, *iipha;
12359                 mblk_t *outer_mp, *inner_mp;
12360 
12361                 if ((outer_mp = allocb(sizeof (ipha_t), BPRI_HI)) == NULL) {
12362                         (void) mi_strlog(ill->ill_rq, 0,
12363                             SL_ERROR|SL_TRACE|SL_CONSOLE,
12364                             "ipsec_out_process: "
12365                             "Self-Encapsulation failed: Out of memory\n");
12366                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
12367                         ip_drop_output("ipIfStatsOutDiscards", mp, ill);
12368                         freemsg(mp);
12369                         return (ENOBUFS);
12370                 }
12371                 inner_mp = mp;
12372                 ASSERT(inner_mp->b_datap->db_type == M_DATA);
12373                 oipha = (ipha_t *)outer_mp->b_rptr;
12374                 iipha = (ipha_t *)inner_mp->b_rptr;
12375                 *oipha = *iipha;
12376                 outer_mp->b_wptr += sizeof (ipha_t);
12377                 oipha->ipha_length = htons(ntohs(iipha->ipha_length) +
12378                     sizeof (ipha_t));
12379                 oipha->ipha_protocol = IPPROTO_ENCAP;
12380                 oipha->ipha_version_and_hdr_length =
12381                     IP_SIMPLE_HDR_VERSION;
12382                 oipha->ipha_hdr_checksum = 0;
12383                 oipha->ipha_hdr_checksum = ip_csum_hdr(oipha);
12384                 outer_mp->b_cont = inner_mp;
12385                 mp = outer_mp;
12386 
12387                 ixa->ixa_flags |= IXAF_IPSEC_TUNNEL;
12388         }
12389 
12390         /* If we need to wait for a SA then we can't return any errno */
12391         if (((ap->ipa_want_ah && (ixa->ixa_ipsec_ah_sa == NULL)) ||
12392             (ap->ipa_want_esp && (ixa->ixa_ipsec_esp_sa == NULL))) &&
12393             !ipsec_out_select_sa(mp, ixa))
12394                 return (0);
12395 
12396         /*
12397          * By now, we know what SA's to use.  Toss over to ESP & AH
12398          * to do the heavy lifting.
12399          */
12400         if (ap->ipa_want_esp) {
12401                 ASSERT(ixa->ixa_ipsec_esp_sa != NULL);
12402 
12403                 mp = ixa->ixa_ipsec_esp_sa->ipsa_output_func(mp, ixa);
12404                 if (mp == NULL) {
12405                         /*
12406                          * Either it failed or is pending. In the former case
12407                          * ipIfStatsInDiscards was increased.
12408                          */
12409                         return (0);
12410                 }
12411         }
12412 
12413         if (ap->ipa_want_ah) {
12414                 ASSERT(ixa->ixa_ipsec_ah_sa != NULL);
12415 
12416                 mp = ixa->ixa_ipsec_ah_sa->ipsa_output_func(mp, ixa);
12417                 if (mp == NULL) {
12418                         /*
12419                          * Either it failed or is pending. In the former case
12420                          * ipIfStatsInDiscards was increased.
12421                          */
12422                         return (0);
12423                 }
12424         }
12425         /*
12426          * We are done with IPsec processing. Send it over
12427          * the wire.
12428          */
12429         return (ip_output_post_ipsec(mp, ixa));
12430 }
12431 
12432 /*
12433  * ioctls that go through a down/up sequence may need to wait for the down
12434  * to complete. This involves waiting for the ire and ipif refcnts to go down
12435  * to zero. Subsequently the ioctl is restarted from ipif_ill_refrele_tail.
12436  */
12437 /* ARGSUSED */
12438 void
12439 ip_reprocess_ioctl(ipsq_t *ipsq, queue_t *q, mblk_t *mp, void *dummy_arg)
12440 {
12441         struct iocblk *iocp;
12442         mblk_t *mp1;
12443         ip_ioctl_cmd_t *ipip;
12444         int err;
12445         sin_t   *sin;
12446         struct lifreq *lifr;
12447         struct ifreq *ifr;
12448 
12449         iocp = (struct iocblk *)mp->b_rptr;
12450         ASSERT(ipsq != NULL);
12451         /* Existence of mp1 verified in ip_wput_nondata */
12452         mp1 = mp->b_cont->b_cont;
12453         ipip = ip_sioctl_lookup(iocp->ioc_cmd);
12454         if (ipip->ipi_cmd == SIOCSLIFNAME || ipip->ipi_cmd == IF_UNITSEL) {
12455                 /*
12456                  * Special case where ipx_current_ipif is not set:
12457                  * ill_phyint_reinit merged the v4 and v6 into a single ipsq.
12458                  * We are here as were not able to complete the operation in
12459                  * ipif_set_values because we could not become exclusive on
12460                  * the new ipsq.
12461                  */
12462                 ill_t *ill = q->q_ptr;
12463                 ipsq_current_start(ipsq, ill->ill_ipif, ipip->ipi_cmd);
12464         }
12465         ASSERT(ipsq->ipsq_xop->ipx_current_ipif != NULL);
12466 
12467         if (ipip->ipi_cmd_type == IF_CMD) {
12468                 /* This a old style SIOC[GS]IF* command */
12469                 ifr = (struct ifreq *)mp1->b_rptr;
12470                 sin = (sin_t *)&ifr->ifr_addr;
12471         } else if (ipip->ipi_cmd_type == LIF_CMD) {
12472                 /* This a new style SIOC[GS]LIF* command */
12473                 lifr = (struct lifreq *)mp1->b_rptr;
12474                 sin = (sin_t *)&lifr->lifr_addr;
12475         } else {
12476                 sin = NULL;
12477         }
12478 
12479         err = (*ipip->ipi_func_restart)(ipsq->ipsq_xop->ipx_current_ipif, sin,
12480             q, mp, ipip, mp1->b_rptr);
12481 
12482         DTRACE_PROBE4(ipif__ioctl, char *, "ip_reprocess_ioctl finish",
12483             int, ipip->ipi_cmd,
12484             ill_t *, ipsq->ipsq_xop->ipx_current_ipif->ipif_ill,
12485             ipif_t *, ipsq->ipsq_xop->ipx_current_ipif);
12486 
12487         ip_ioctl_finish(q, mp, err, IPI2MODE(ipip), ipsq);
12488 }
12489 
12490 /*
12491  * ioctl processing
12492  *
12493  * ioctl processing starts with ip_sioctl_copyin_setup(), which looks up
12494  * the ioctl command in the ioctl tables, determines the copyin data size
12495  * from the ipi_copyin_size field, and does an mi_copyin() of that size.
12496  *
12497  * ioctl processing then continues when the M_IOCDATA makes its way down to
12498  * ip_wput_nondata().  The ioctl is looked up again in the ioctl table, its
12499  * associated 'conn' is refheld till the end of the ioctl and the general
12500  * ioctl processing function ip_process_ioctl() is called to extract the
12501  * arguments and process the ioctl.  To simplify extraction, ioctl commands
12502  * are "typed" based on the arguments they take (e.g., LIF_CMD which takes a
12503  * `struct lifreq'), and a common extract function (e.g., ip_extract_lifreq())
12504  * is used to extract the ioctl's arguments.
12505  *
12506  * ip_process_ioctl determines if the ioctl needs to be serialized, and if
12507  * so goes thru the serialization primitive ipsq_try_enter. Then the
12508  * appropriate function to handle the ioctl is called based on the entry in
12509  * the ioctl table. ioctl completion is encapsulated in ip_ioctl_finish
12510  * which also refreleases the 'conn' that was refheld at the start of the
12511  * ioctl. Finally ipsq_exit is called if needed to exit the ipsq.
12512  *
12513  * Many exclusive ioctls go thru an internal down up sequence as part of
12514  * the operation. For example an attempt to change the IP address of an
12515  * ipif entails ipif_down, set address, ipif_up. Bringing down the interface
12516  * does all the cleanup such as deleting all ires that use this address.
12517  * Then we need to wait till all references to the interface go away.
12518  */
12519 void
12520 ip_process_ioctl(ipsq_t *ipsq, queue_t *q, mblk_t *mp, void *arg)
12521 {
12522         struct iocblk *iocp = (struct iocblk *)mp->b_rptr;
12523         ip_ioctl_cmd_t *ipip = arg;
12524         ip_extract_func_t *extract_funcp;
12525         ill_t *ill;
12526         cmd_info_t ci;
12527         int err;
12528         boolean_t entered_ipsq = B_FALSE;
12529 
12530         ip3dbg(("ip_process_ioctl: ioctl %X\n", iocp->ioc_cmd));
12531 
12532         if (ipip == NULL)
12533                 ipip = ip_sioctl_lookup(iocp->ioc_cmd);
12534 
12535         /*
12536          * SIOCLIFADDIF needs to go thru a special path since the
12537          * ill may not exist yet. This happens in the case of lo0
12538          * which is created using this ioctl.
12539          */
12540         if (ipip->ipi_cmd == SIOCLIFADDIF) {
12541                 err = ip_sioctl_addif(NULL, NULL, q, mp, NULL, NULL);
12542                 DTRACE_PROBE4(ipif__ioctl, char *, "ip_process_ioctl finish",
12543                     int, ipip->ipi_cmd, ill_t *, NULL, ipif_t *, NULL);
12544                 ip_ioctl_finish(q, mp, err, IPI2MODE(ipip), NULL);
12545                 return;
12546         }
12547 
12548         ci.ci_ipif = NULL;
12549         switch (ipip->ipi_cmd_type) {
12550         case MISC_CMD:
12551         case MSFILT_CMD:
12552                 /*
12553                  * All MISC_CMD ioctls come in here -- e.g. SIOCGLIFCONF.
12554                  */
12555                 if (ipip->ipi_cmd == IF_UNITSEL) {
12556                         /* ioctl comes down the ill */
12557                         ci.ci_ipif = ((ill_t *)q->q_ptr)->ill_ipif;
12558                         ipif_refhold(ci.ci_ipif);
12559                 }
12560                 err = 0;
12561                 ci.ci_sin = NULL;
12562                 ci.ci_sin6 = NULL;
12563                 ci.ci_lifr = NULL;
12564                 extract_funcp = NULL;
12565                 break;
12566 
12567         case IF_CMD:
12568         case LIF_CMD:
12569                 extract_funcp = ip_extract_lifreq;
12570                 break;
12571 
12572         case ARP_CMD:
12573         case XARP_CMD:
12574                 extract_funcp = ip_extract_arpreq;
12575                 break;
12576 
12577         default:
12578                 ASSERT(0);
12579         }
12580 
12581         if (extract_funcp != NULL) {
12582                 err = (*extract_funcp)(q, mp, ipip, &ci);
12583                 if (err != 0) {
12584                         DTRACE_PROBE4(ipif__ioctl,
12585                             char *, "ip_process_ioctl finish err",
12586                             int, ipip->ipi_cmd, ill_t *, NULL, ipif_t *, NULL);
12587                         ip_ioctl_finish(q, mp, err, IPI2MODE(ipip), NULL);
12588                         return;
12589                 }
12590 
12591                 /*
12592                  * All of the extraction functions return a refheld ipif.
12593                  */
12594                 ASSERT(ci.ci_ipif != NULL);
12595         }
12596 
12597         if (!(ipip->ipi_flags & IPI_WR)) {
12598                 /*
12599                  * A return value of EINPROGRESS means the ioctl is
12600                  * either queued and waiting for some reason or has
12601                  * already completed.
12602                  */
12603                 err = (*ipip->ipi_func)(ci.ci_ipif, ci.ci_sin, q, mp, ipip,
12604                     ci.ci_lifr);
12605                 if (ci.ci_ipif != NULL) {
12606                         DTRACE_PROBE4(ipif__ioctl,
12607                             char *, "ip_process_ioctl finish RD",
12608                             int, ipip->ipi_cmd, ill_t *, ci.ci_ipif->ipif_ill,
12609                             ipif_t *, ci.ci_ipif);
12610                         ipif_refrele(ci.ci_ipif);
12611                 } else {
12612                         DTRACE_PROBE4(ipif__ioctl,
12613                             char *, "ip_process_ioctl finish RD",
12614                             int, ipip->ipi_cmd, ill_t *, NULL, ipif_t *, NULL);
12615                 }
12616                 ip_ioctl_finish(q, mp, err, IPI2MODE(ipip), NULL);
12617                 return;
12618         }
12619 
12620         ASSERT(ci.ci_ipif != NULL);
12621 
12622         /*
12623          * If ipsq is non-NULL, we are already being called exclusively
12624          */
12625         ASSERT(ipsq == NULL || IAM_WRITER_IPSQ(ipsq));
12626         if (ipsq == NULL) {
12627                 ipsq = ipsq_try_enter(ci.ci_ipif, NULL, q, mp, ip_process_ioctl,
12628                     NEW_OP, B_TRUE);
12629                 if (ipsq == NULL) {
12630                         ipif_refrele(ci.ci_ipif);
12631                         return;
12632                 }
12633                 entered_ipsq = B_TRUE;
12634         }
12635         /*
12636          * Release the ipif so that ipif_down and friends that wait for
12637          * references to go away are not misled about the current ipif_refcnt
12638          * values. We are writer so we can access the ipif even after releasing
12639          * the ipif.
12640          */
12641         ipif_refrele(ci.ci_ipif);
12642 
12643         ipsq_current_start(ipsq, ci.ci_ipif, ipip->ipi_cmd);
12644 
12645         /*
12646          * We need to cache the ill_t that we're going to use as the argument
12647          * to the ipif-ioctl DTrace probe (below) because the ci_ipif can be
12648          * blown away by calling ipi_func.
12649          */
12650         ill = ci.ci_ipif == NULL ? NULL : ci.ci_ipif->ipif_ill;
12651 
12652         /*
12653          * A return value of EINPROGRESS means the ioctl is
12654          * either queued and waiting for some reason or has
12655          * already completed.
12656          */
12657         err = (*ipip->ipi_func)(ci.ci_ipif, ci.ci_sin, q, mp, ipip, ci.ci_lifr);
12658 
12659         DTRACE_PROBE4(ipif__ioctl, char *, "ip_process_ioctl finish WR",
12660             int, ipip->ipi_cmd, ill_t *, ill, ipif_t *, ci.ci_ipif);
12661         ip_ioctl_finish(q, mp, err, IPI2MODE(ipip), ipsq);
12662 
12663         if (entered_ipsq)
12664                 ipsq_exit(ipsq);
12665 }
12666 
12667 /*
12668  * Complete the ioctl. Typically ioctls use the mi package and need to
12669  * do mi_copyout/mi_copy_done.
12670  */
12671 void
12672 ip_ioctl_finish(queue_t *q, mblk_t *mp, int err, int mode, ipsq_t *ipsq)
12673 {
12674         conn_t  *connp = NULL;
12675 
12676         if (err == EINPROGRESS)
12677                 return;
12678 
12679         if (CONN_Q(q)) {
12680                 connp = Q_TO_CONN(q);
12681                 ASSERT(connp->conn_ref >= 2);
12682         }
12683 
12684         switch (mode) {
12685         case COPYOUT:
12686                 if (err == 0)
12687                         mi_copyout(q, mp);
12688                 else
12689                         mi_copy_done(q, mp, err);
12690                 break;
12691 
12692         case NO_COPYOUT:
12693                 mi_copy_done(q, mp, err);
12694                 break;
12695 
12696         default:
12697                 ASSERT(mode == CONN_CLOSE);     /* aborted through CONN_CLOSE */
12698                 break;
12699         }
12700 
12701         /*
12702          * The conn refhold and ioctlref placed on the conn at the start of the
12703          * ioctl are released here.
12704          */
12705         if (connp != NULL) {
12706                 CONN_DEC_IOCTLREF(connp);
12707                 CONN_OPER_PENDING_DONE(connp);
12708         }
12709 
12710         if (ipsq != NULL)
12711                 ipsq_current_finish(ipsq);
12712 }
12713 
12714 /* Handles all non data messages */
12715 int
12716 ip_wput_nondata(queue_t *q, mblk_t *mp)
12717 {
12718         mblk_t          *mp1;
12719         struct iocblk   *iocp;
12720         ip_ioctl_cmd_t  *ipip;
12721         conn_t          *connp;
12722         cred_t          *cr;
12723         char            *proto_str;
12724 
12725         if (CONN_Q(q))
12726                 connp = Q_TO_CONN(q);
12727         else
12728                 connp = NULL;
12729 
12730         switch (DB_TYPE(mp)) {
12731         case M_IOCTL:
12732                 /*
12733                  * IOCTL processing begins in ip_sioctl_copyin_setup which
12734                  * will arrange to copy in associated control structures.
12735                  */
12736                 ip_sioctl_copyin_setup(q, mp);
12737                 return (0);
12738         case M_IOCDATA:
12739                 /*
12740                  * Ensure that this is associated with one of our trans-
12741                  * parent ioctls.  If it's not ours, discard it if we're
12742                  * running as a driver, or pass it on if we're a module.
12743                  */
12744                 iocp = (struct iocblk *)mp->b_rptr;
12745                 ipip = ip_sioctl_lookup(iocp->ioc_cmd);
12746                 if (ipip == NULL) {
12747                         if (q->q_next == NULL) {
12748                                 goto nak;
12749                         } else {
12750                                 putnext(q, mp);
12751                         }
12752                         return (0);
12753                 }
12754                 if ((q->q_next != NULL) && !(ipip->ipi_flags & IPI_MODOK)) {
12755                         /*
12756                          * The ioctl is one we recognise, but is not consumed
12757                          * by IP as a module and we are a module, so we drop
12758                          */
12759                         goto nak;
12760                 }
12761 
12762                 /* IOCTL continuation following copyin or copyout. */
12763                 if (mi_copy_state(q, mp, NULL) == -1) {
12764                         /*
12765                          * The copy operation failed.  mi_copy_state already
12766                          * cleaned up, so we're out of here.
12767                          */
12768                         return (0);
12769                 }
12770                 /*
12771                  * If we just completed a copy in, we become writer and
12772                  * continue processing in ip_sioctl_copyin_done.  If it
12773                  * was a copy out, we call mi_copyout again.  If there is
12774                  * nothing more to copy out, it will complete the IOCTL.
12775                  */
12776                 if (MI_COPY_DIRECTION(mp) == MI_COPY_IN) {
12777                         if (!(mp1 = mp->b_cont) || !(mp1 = mp1->b_cont)) {
12778                                 mi_copy_done(q, mp, EPROTO);
12779                                 return (0);
12780                         }
12781                         /*
12782                          * Check for cases that need more copying.  A return
12783                          * value of 0 means a second copyin has been started,
12784                          * so we return; a return value of 1 means no more
12785                          * copying is needed, so we continue.
12786                          */
12787                         if (ipip->ipi_cmd_type == MSFILT_CMD &&
12788                             MI_COPY_COUNT(mp) == 1) {
12789                                 if (ip_copyin_msfilter(q, mp) == 0)
12790                                         return (0);
12791                         }
12792                         /*
12793                          * Refhold the conn, till the ioctl completes. This is
12794                          * needed in case the ioctl ends up in the pending mp
12795                          * list. Every mp in the ipx_pending_mp list must have
12796                          * a refhold on the conn to resume processing. The
12797                          * refhold is released when the ioctl completes
12798                          * (whether normally or abnormally). An ioctlref is also
12799                          * placed on the conn to prevent TCP from removing the
12800                          * queue needed to send the ioctl reply back.
12801                          * In all cases ip_ioctl_finish is called to finish
12802                          * the ioctl and release the refholds.
12803                          */
12804                         if (connp != NULL) {
12805                                 /* This is not a reentry */
12806                                 CONN_INC_REF(connp);
12807                                 CONN_INC_IOCTLREF(connp);
12808                         } else {
12809                                 if (!(ipip->ipi_flags & IPI_MODOK)) {
12810                                         mi_copy_done(q, mp, EINVAL);
12811                                         return (0);
12812                                 }
12813                         }
12814 
12815                         ip_process_ioctl(NULL, q, mp, ipip);
12816 
12817                 } else {
12818                         mi_copyout(q, mp);
12819                 }
12820                 return (0);
12821 
12822         case M_IOCNAK:
12823                 /*
12824                  * The only way we could get here is if a resolver didn't like
12825                  * an IOCTL we sent it.  This shouldn't happen.
12826                  */
12827                 (void) mi_strlog(q, 1, SL_ERROR|SL_TRACE,
12828                     "ip_wput_nondata: unexpected M_IOCNAK, ioc_cmd 0x%x",
12829                     ((struct iocblk *)mp->b_rptr)->ioc_cmd);
12830                 freemsg(mp);
12831                 return (0);
12832         case M_IOCACK:
12833                 /* /dev/ip shouldn't see this */
12834                 goto nak;
12835         case M_FLUSH:
12836                 if (*mp->b_rptr & FLUSHW)
12837                         flushq(q, FLUSHALL);
12838                 if (q->q_next) {
12839                         putnext(q, mp);
12840                         return (0);
12841                 }
12842                 if (*mp->b_rptr & FLUSHR) {
12843                         *mp->b_rptr &= ~FLUSHW;
12844                         qreply(q, mp);
12845                         return (0);
12846                 }
12847                 freemsg(mp);
12848                 return (0);
12849         case M_CTL:
12850                 break;
12851         case M_PROTO:
12852         case M_PCPROTO:
12853                 /*
12854                  * The only PROTO messages we expect are SNMP-related.
12855                  */
12856                 switch (((union T_primitives *)mp->b_rptr)->type) {
12857                 case T_SVR4_OPTMGMT_REQ:
12858                         ip2dbg(("ip_wput_nondata: T_SVR4_OPTMGMT_REQ "
12859                             "flags %x\n",
12860                             ((struct T_optmgmt_req *)mp->b_rptr)->MGMT_flags));
12861 
12862                         if (connp == NULL) {
12863                                 proto_str = "T_SVR4_OPTMGMT_REQ";
12864                                 goto protonak;
12865                         }
12866 
12867                         /*
12868                          * All Solaris components should pass a db_credp
12869                          * for this TPI message, hence we ASSERT.
12870                          * But in case there is some other M_PROTO that looks
12871                          * like a TPI message sent by some other kernel
12872                          * component, we check and return an error.
12873                          */
12874                         cr = msg_getcred(mp, NULL);
12875                         ASSERT(cr != NULL);
12876                         if (cr == NULL) {
12877                                 mp = mi_tpi_err_ack_alloc(mp, TSYSERR, EINVAL);
12878                                 if (mp != NULL)
12879                                         qreply(q, mp);
12880                                 return (0);
12881                         }
12882 
12883                         if (!snmpcom_req(q, mp, ip_snmp_set, ip_snmp_get, cr)) {
12884                                 proto_str = "Bad SNMPCOM request?";
12885                                 goto protonak;
12886                         }
12887                         return (0);
12888                 default:
12889                         ip1dbg(("ip_wput_nondata: dropping M_PROTO prim %u\n",
12890                             (int)*(uint_t *)mp->b_rptr));
12891                         freemsg(mp);
12892                         return (0);
12893                 }
12894         default:
12895                 break;
12896         }
12897         if (q->q_next) {
12898                 putnext(q, mp);
12899         } else
12900                 freemsg(mp);
12901         return (0);
12902 
12903 nak:
12904         iocp->ioc_error = EINVAL;
12905         mp->b_datap->db_type = M_IOCNAK;
12906         iocp->ioc_count = 0;
12907         qreply(q, mp);
12908         return (0);
12909 
12910 protonak:
12911         cmn_err(CE_NOTE, "IP doesn't process %s as a module", proto_str);
12912         if ((mp = mi_tpi_err_ack_alloc(mp, TPROTO, EINVAL)) != NULL)
12913                 qreply(q, mp);
12914         return (0);
12915 }
12916 
12917 /*
12918  * Process IP options in an outbound packet.  Verify that the nexthop in a
12919  * strict source route is onlink.
12920  * Returns non-zero if something fails in which case an ICMP error has been
12921  * sent and mp freed.
12922  *
12923  * Assumes the ULP has called ip_massage_options to move nexthop into ipha_dst.
12924  */
12925 int
12926 ip_output_options(mblk_t *mp, ipha_t *ipha, ip_xmit_attr_t *ixa, ill_t *ill)
12927 {
12928         ipoptp_t        opts;
12929         uchar_t         *opt;
12930         uint8_t         optval;
12931         uint8_t         optlen;
12932         ipaddr_t        dst;
12933         intptr_t        code = 0;
12934         ire_t           *ire;
12935         ip_stack_t      *ipst = ixa->ixa_ipst;
12936         ip_recv_attr_t  iras;
12937 
12938         ip2dbg(("ip_output_options\n"));
12939 
12940         dst = ipha->ipha_dst;
12941         for (optval = ipoptp_first(&opts, ipha);
12942             optval != IPOPT_EOL;
12943             optval = ipoptp_next(&opts)) {
12944                 opt = opts.ipoptp_cur;
12945                 optlen = opts.ipoptp_len;
12946                 ip2dbg(("ip_output_options: opt %d, len %d\n",
12947                     optval, optlen));
12948                 switch (optval) {
12949                         uint32_t off;
12950                 case IPOPT_SSRR:
12951                 case IPOPT_LSRR:
12952                         if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
12953                                 ip1dbg((
12954                                     "ip_output_options: bad option offset\n"));
12955                                 code = (char *)&opt[IPOPT_OLEN] -
12956                                     (char *)ipha;
12957                                 goto param_prob;
12958                         }
12959                         off = opt[IPOPT_OFFSET];
12960                         ip1dbg(("ip_output_options: next hop 0x%x\n",
12961                             ntohl(dst)));
12962                         /*
12963                          * For strict: verify that dst is directly
12964                          * reachable.
12965                          */
12966                         if (optval == IPOPT_SSRR) {
12967                                 ire = ire_ftable_lookup_v4(dst, 0, 0,
12968                                     IRE_INTERFACE, NULL, ALL_ZONES,
12969                                     ixa->ixa_tsl,
12970                                     MATCH_IRE_TYPE | MATCH_IRE_SECATTR, 0, ipst,
12971                                     NULL);
12972                                 if (ire == NULL) {
12973                                         ip1dbg(("ip_output_options: SSRR not"
12974                                             " directly reachable: 0x%x\n",
12975                                             ntohl(dst)));
12976                                         goto bad_src_route;
12977                                 }
12978                                 ire_refrele(ire);
12979                         }
12980                         break;
12981                 case IPOPT_RR:
12982                         if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
12983                                 ip1dbg((
12984                                     "ip_output_options: bad option offset\n"));
12985                                 code = (char *)&opt[IPOPT_OLEN] -
12986                                     (char *)ipha;
12987                                 goto param_prob;
12988                         }
12989                         break;
12990                 case IPOPT_TS:
12991                         /*
12992                          * Verify that length >=5 and that there is either
12993                          * room for another timestamp or that the overflow
12994                          * counter is not maxed out.
12995                          */
12996                         code = (char *)&opt[IPOPT_OLEN] - (char *)ipha;
12997                         if (optlen < IPOPT_MINLEN_IT) {
12998                                 goto param_prob;
12999                         }
13000                         if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
13001                                 ip1dbg((
13002                                     "ip_output_options: bad option offset\n"));
13003                                 code = (char *)&opt[IPOPT_OFFSET] -
13004                                     (char *)ipha;
13005                                 goto param_prob;
13006                         }
13007                         switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
13008                         case IPOPT_TS_TSONLY:
13009                                 off = IPOPT_TS_TIMELEN;
13010                                 break;
13011                         case IPOPT_TS_TSANDADDR:
13012                         case IPOPT_TS_PRESPEC:
13013                         case IPOPT_TS_PRESPEC_RFC791:
13014                                 off = IP_ADDR_LEN + IPOPT_TS_TIMELEN;
13015                                 break;
13016                         default:
13017                                 code = (char *)&opt[IPOPT_POS_OV_FLG] -
13018                                     (char *)ipha;
13019                                 goto param_prob;
13020                         }
13021                         if (opt[IPOPT_OFFSET] - 1 + off > optlen &&
13022                             (opt[IPOPT_POS_OV_FLG] & 0xF0) == 0xF0) {
13023                                 /*
13024                                  * No room and the overflow counter is 15
13025                                  * already.
13026                                  */
13027                                 goto param_prob;
13028                         }
13029                         break;
13030                 }
13031         }
13032 
13033         if ((opts.ipoptp_flags & IPOPTP_ERROR) == 0)
13034                 return (0);
13035 
13036         ip1dbg(("ip_output_options: error processing IP options."));
13037         code = (char *)&opt[IPOPT_OFFSET] - (char *)ipha;
13038 
13039 param_prob:
13040         bzero(&iras, sizeof (iras));
13041         iras.ira_ill = iras.ira_rill = ill;
13042         iras.ira_ruifindex = ill->ill_phyint->phyint_ifindex;
13043         iras.ira_rifindex = iras.ira_ruifindex;
13044         iras.ira_flags = IRAF_IS_IPV4;
13045 
13046         ip_drop_output("ip_output_options", mp, ill);
13047         icmp_param_problem(mp, (uint8_t)code, &iras);
13048         ASSERT(!(iras.ira_flags & IRAF_IPSEC_SECURE));
13049         return (-1);
13050 
13051 bad_src_route:
13052         bzero(&iras, sizeof (iras));
13053         iras.ira_ill = iras.ira_rill = ill;
13054         iras.ira_ruifindex = ill->ill_phyint->phyint_ifindex;
13055         iras.ira_rifindex = iras.ira_ruifindex;
13056         iras.ira_flags = IRAF_IS_IPV4;
13057 
13058         ip_drop_input("ICMP_SOURCE_ROUTE_FAILED", mp, ill);
13059         icmp_unreachable(mp, ICMP_SOURCE_ROUTE_FAILED, &iras);
13060         ASSERT(!(iras.ira_flags & IRAF_IPSEC_SECURE));
13061         return (-1);
13062 }
13063 
13064 /*
13065  * The maximum value of conn_drain_list_cnt is CONN_MAXDRAINCNT.
13066  * conn_drain_list_cnt can be changed by setting conn_drain_nthreads
13067  * thru /etc/system.
13068  */
13069 #define CONN_MAXDRAINCNT        64
13070 
13071 static void
13072 conn_drain_init(ip_stack_t *ipst)
13073 {
13074         int i, j;
13075         idl_tx_list_t *itl_tx;
13076 
13077         ipst->ips_conn_drain_list_cnt = conn_drain_nthreads;
13078 
13079         if ((ipst->ips_conn_drain_list_cnt == 0) ||
13080             (ipst->ips_conn_drain_list_cnt > CONN_MAXDRAINCNT)) {
13081                 /*
13082                  * Default value of the number of drainers is the
13083                  * number of cpus, subject to maximum of 8 drainers.
13084                  */
13085                 if (boot_max_ncpus != -1)
13086                         ipst->ips_conn_drain_list_cnt = MIN(boot_max_ncpus, 8);
13087                 else
13088                         ipst->ips_conn_drain_list_cnt = MIN(max_ncpus, 8);
13089         }
13090 
13091         ipst->ips_idl_tx_list =
13092             kmem_zalloc(TX_FANOUT_SIZE * sizeof (idl_tx_list_t), KM_SLEEP);
13093         for (i = 0; i < TX_FANOUT_SIZE; i++) {
13094                 itl_tx =  &ipst->ips_idl_tx_list[i];
13095                 itl_tx->txl_drain_list =
13096                     kmem_zalloc(ipst->ips_conn_drain_list_cnt *
13097                     sizeof (idl_t), KM_SLEEP);
13098                 mutex_init(&itl_tx->txl_lock, NULL, MUTEX_DEFAULT, NULL);
13099                 for (j = 0; j < ipst->ips_conn_drain_list_cnt; j++) {
13100                         mutex_init(&itl_tx->txl_drain_list[j].idl_lock, NULL,
13101                             MUTEX_DEFAULT, NULL);
13102                         itl_tx->txl_drain_list[j].idl_itl = itl_tx;
13103                 }
13104         }
13105 }
13106 
13107 static void
13108 conn_drain_fini(ip_stack_t *ipst)
13109 {
13110         int i;
13111         idl_tx_list_t *itl_tx;
13112 
13113         for (i = 0; i < TX_FANOUT_SIZE; i++) {
13114                 itl_tx =  &ipst->ips_idl_tx_list[i];
13115                 kmem_free(itl_tx->txl_drain_list,
13116                     ipst->ips_conn_drain_list_cnt * sizeof (idl_t));
13117         }
13118         kmem_free(ipst->ips_idl_tx_list,
13119             TX_FANOUT_SIZE * sizeof (idl_tx_list_t));
13120         ipst->ips_idl_tx_list = NULL;
13121 }
13122 
13123 /*
13124  * Flow control has blocked us from proceeding.  Insert the given conn in one
13125  * of the conn drain lists.  When flow control is unblocked, either ip_wsrv()
13126  * (STREAMS) or ill_flow_enable() (direct) will be called back, which in turn
13127  * will call conn_walk_drain().  See the flow control notes at the top of this
13128  * file for more details.
13129  */
13130 void
13131 conn_drain_insert(conn_t *connp, idl_tx_list_t *tx_list)
13132 {
13133         idl_t   *idl = tx_list->txl_drain_list;
13134         uint_t  index;
13135         ip_stack_t      *ipst = connp->conn_netstack->netstack_ip;
13136 
13137         mutex_enter(&connp->conn_lock);
13138         if (connp->conn_state_flags & CONN_CLOSING) {
13139                 /*
13140                  * The conn is closing as a result of which CONN_CLOSING
13141                  * is set. Return.
13142                  */
13143                 mutex_exit(&connp->conn_lock);
13144                 return;
13145         } else if (connp->conn_idl == NULL) {
13146                 /*
13147                  * Assign the next drain list round robin. We dont' use
13148                  * a lock, and thus it may not be strictly round robin.
13149                  * Atomicity of load/stores is enough to make sure that
13150                  * conn_drain_list_index is always within bounds.
13151                  */
13152                 index = tx_list->txl_drain_index;
13153                 ASSERT(index < ipst->ips_conn_drain_list_cnt);
13154                 connp->conn_idl = &tx_list->txl_drain_list[index];
13155                 index++;
13156                 if (index == ipst->ips_conn_drain_list_cnt)
13157                         index = 0;
13158                 tx_list->txl_drain_index = index;
13159         } else {
13160                 ASSERT(connp->conn_idl->idl_itl == tx_list);
13161         }
13162         mutex_exit(&connp->conn_lock);
13163 
13164         idl = connp->conn_idl;
13165         mutex_enter(&idl->idl_lock);
13166         if ((connp->conn_drain_prev != NULL) ||
13167             (connp->conn_state_flags & CONN_CLOSING)) {
13168                 /*
13169                  * The conn is either already in the drain list or closing.
13170                  * (We needed to check for CONN_CLOSING again since close can
13171                  * sneak in between dropping conn_lock and acquiring idl_lock.)
13172                  */
13173                 mutex_exit(&idl->idl_lock);
13174                 return;
13175         }
13176 
13177         /*
13178          * The conn is not in the drain list. Insert it at the
13179          * tail of the drain list. The drain list is circular
13180          * and doubly linked. idl_conn points to the 1st element
13181          * in the list.
13182          */
13183         if (idl->idl_conn == NULL) {
13184                 idl->idl_conn = connp;
13185                 connp->conn_drain_next = connp;
13186                 connp->conn_drain_prev = connp;
13187         } else {
13188                 conn_t *head = idl->idl_conn;
13189 
13190                 connp->conn_drain_next = head;
13191                 connp->conn_drain_prev = head->conn_drain_prev;
13192                 head->conn_drain_prev->conn_drain_next = connp;
13193                 head->conn_drain_prev = connp;
13194         }
13195         /*
13196          * For non streams based sockets assert flow control.
13197          */
13198         conn_setqfull(connp, NULL);
13199         mutex_exit(&idl->idl_lock);
13200 }
13201 
13202 static void
13203 conn_drain_remove(conn_t *connp)
13204 {
13205         idl_t *idl = connp->conn_idl;
13206 
13207         if (idl != NULL) {
13208                 /*
13209                  * Remove ourself from the drain list.
13210                  */
13211                 if (connp->conn_drain_next == connp) {
13212                         /* Singleton in the list */
13213                         ASSERT(connp->conn_drain_prev == connp);
13214                         idl->idl_conn = NULL;
13215                 } else {
13216                         connp->conn_drain_prev->conn_drain_next =
13217                             connp->conn_drain_next;
13218                         connp->conn_drain_next->conn_drain_prev =
13219                             connp->conn_drain_prev;
13220                         if (idl->idl_conn == connp)
13221                                 idl->idl_conn = connp->conn_drain_next;
13222                 }
13223 
13224                 /*
13225                  * NOTE: because conn_idl is associated with a specific drain
13226                  * list which in turn is tied to the index the TX ring
13227                  * (txl_cookie) hashes to, and because the TX ring can change
13228                  * over the lifetime of the conn_t, we must clear conn_idl so
13229                  * a subsequent conn_drain_insert() will set conn_idl again
13230                  * based on the latest txl_cookie.
13231                  */
13232                 connp->conn_idl = NULL;
13233         }
13234         connp->conn_drain_next = NULL;
13235         connp->conn_drain_prev = NULL;
13236 
13237         conn_clrqfull(connp, NULL);
13238         /*
13239          * For streams based sockets open up flow control.
13240          */
13241         if (!IPCL_IS_NONSTR(connp))
13242                 enableok(connp->conn_wq);
13243 }
13244 
13245 /*
13246  * This conn is closing, and we are called from ip_close. OR
13247  * this conn is draining because flow-control on the ill has been relieved.
13248  *
13249  * We must also need to remove conn's on this idl from the list, and also
13250  * inform the sockfs upcalls about the change in flow-control.
13251  */
13252 static void
13253 conn_drain(conn_t *connp, boolean_t closing)
13254 {
13255         idl_t *idl;
13256         conn_t *next_connp;
13257 
13258         /*
13259          * connp->conn_idl is stable at this point, and no lock is needed
13260          * to check it. If we are called from ip_close, close has already
13261          * set CONN_CLOSING, thus freezing the value of conn_idl, and
13262          * called us only because conn_idl is non-null. If we are called thru
13263          * service, conn_idl could be null, but it cannot change because
13264          * service is single-threaded per queue, and there cannot be another
13265          * instance of service trying to call conn_drain_insert on this conn
13266          * now.
13267          */
13268         ASSERT(!closing || connp == NULL || connp->conn_idl != NULL);
13269 
13270         /*
13271          * If the conn doesn't exist or is not on a drain list, bail.
13272          */
13273         if (connp == NULL || connp->conn_idl == NULL ||
13274             connp->conn_drain_prev == NULL) {
13275                 return;
13276         }
13277 
13278         idl = connp->conn_idl;
13279         ASSERT(MUTEX_HELD(&idl->idl_lock));
13280 
13281         if (!closing) {
13282                 next_connp = connp->conn_drain_next;
13283                 while (next_connp != connp) {
13284                         conn_t *delconnp = next_connp;
13285 
13286                         next_connp = next_connp->conn_drain_next;
13287                         conn_drain_remove(delconnp);
13288                 }
13289                 ASSERT(connp->conn_drain_next == idl->idl_conn);
13290         }
13291         conn_drain_remove(connp);
13292 }
13293 
13294 /*
13295  * Write service routine. Shared perimeter entry point.
13296  * The device queue's messages has fallen below the low water mark and STREAMS
13297  * has backenabled the ill_wq. Send sockfs notification about flow-control on
13298  * each waiting conn.
13299  */
13300 int
13301 ip_wsrv(queue_t *q)
13302 {
13303         ill_t   *ill;
13304 
13305         ill = (ill_t *)q->q_ptr;
13306         if (ill->ill_state_flags == 0) {
13307                 ip_stack_t *ipst = ill->ill_ipst;
13308 
13309                 /*
13310                  * The device flow control has opened up.
13311                  * Walk through conn drain lists and qenable the
13312                  * first conn in each list. This makes sense only
13313                  * if the stream is fully plumbed and setup.
13314                  * Hence the ill_state_flags check above.
13315                  */
13316                 ip1dbg(("ip_wsrv: walking\n"));
13317                 conn_walk_drain(ipst, &ipst->ips_idl_tx_list[0]);
13318                 enableok(ill->ill_wq);
13319         }
13320         return (0);
13321 }
13322 
13323 /*
13324  * Callback to disable flow control in IP.
13325  *
13326  * This is a mac client callback added when the DLD_CAPAB_DIRECT capability
13327  * is enabled.
13328  *
13329  * When MAC_TX() is not able to send any more packets, dld sets its queue
13330  * to QFULL and enable the STREAMS flow control. Later, when the underlying
13331  * driver is able to continue to send packets, it calls mac_tx_(ring_)update()
13332  * function and wakes up corresponding mac worker threads, which in turn
13333  * calls this callback function, and disables flow control.
13334  */
13335 void
13336 ill_flow_enable(void *arg, ip_mac_tx_cookie_t cookie)
13337 {
13338         ill_t *ill = (ill_t *)arg;
13339         ip_stack_t *ipst = ill->ill_ipst;
13340         idl_tx_list_t *idl_txl;
13341 
13342         idl_txl = &ipst->ips_idl_tx_list[IDLHASHINDEX(cookie)];
13343         mutex_enter(&idl_txl->txl_lock);
13344         /* add code to to set a flag to indicate idl_txl is enabled */
13345         conn_walk_drain(ipst, idl_txl);
13346         mutex_exit(&idl_txl->txl_lock);
13347 }
13348 
13349 /*
13350  * Flow control has been relieved and STREAMS has backenabled us; drain
13351  * all the conn lists on `tx_list'.
13352  */
13353 static void
13354 conn_walk_drain(ip_stack_t *ipst, idl_tx_list_t *tx_list)
13355 {
13356         int i;
13357         idl_t *idl;
13358 
13359         IP_STAT(ipst, ip_conn_walk_drain);
13360 
13361         for (i = 0; i < ipst->ips_conn_drain_list_cnt; i++) {
13362                 idl = &tx_list->txl_drain_list[i];
13363                 mutex_enter(&idl->idl_lock);
13364                 conn_drain(idl->idl_conn, B_FALSE);
13365                 mutex_exit(&idl->idl_lock);
13366         }
13367 }
13368 
13369 /*
13370  * Determine if the ill and multicast aspects of that packets
13371  * "matches" the conn.
13372  */
13373 boolean_t
13374 conn_wantpacket(conn_t *connp, ip_recv_attr_t *ira, ipha_t *ipha)
13375 {
13376         ill_t           *ill = ira->ira_rill;
13377         zoneid_t        zoneid = ira->ira_zoneid;
13378         uint_t          in_ifindex;
13379         ipaddr_t        dst, src;
13380 
13381         dst = ipha->ipha_dst;
13382         src = ipha->ipha_src;
13383 
13384         /*
13385          * conn_incoming_ifindex is set by IP_BOUND_IF which limits
13386          * unicast, broadcast and multicast reception to
13387          * conn_incoming_ifindex.
13388          * conn_wantpacket is called for unicast, broadcast and
13389          * multicast packets.
13390          */
13391         in_ifindex = connp->conn_incoming_ifindex;
13392 
13393         /* mpathd can bind to the under IPMP interface, which we allow */
13394         if (in_ifindex != 0 && in_ifindex != ill->ill_phyint->phyint_ifindex) {
13395                 if (!IS_UNDER_IPMP(ill))
13396                         return (B_FALSE);
13397 
13398                 if (in_ifindex != ipmp_ill_get_ipmp_ifindex(ill))
13399                         return (B_FALSE);
13400         }
13401 
13402         if (!IPCL_ZONE_MATCH(connp, zoneid))
13403                 return (B_FALSE);
13404 
13405         if (!(ira->ira_flags & IRAF_MULTICAST))
13406                 return (B_TRUE);
13407 
13408         if (connp->conn_multi_router) {
13409                 /* multicast packet and multicast router socket: send up */
13410                 return (B_TRUE);
13411         }
13412 
13413         if (ipha->ipha_protocol == IPPROTO_PIM ||
13414             ipha->ipha_protocol == IPPROTO_RSVP)
13415                 return (B_TRUE);
13416 
13417         return (conn_hasmembers_ill_withsrc_v4(connp, dst, src, ira->ira_ill));
13418 }
13419 
13420 void
13421 conn_setqfull(conn_t *connp, boolean_t *flow_stopped)
13422 {
13423         if (IPCL_IS_NONSTR(connp)) {
13424                 (*connp->conn_upcalls->su_txq_full)
13425                     (connp->conn_upper_handle, B_TRUE);
13426                 if (flow_stopped != NULL)
13427                         *flow_stopped = B_TRUE;
13428         } else {
13429                 queue_t *q = connp->conn_wq;
13430 
13431                 ASSERT(q != NULL);
13432                 if (!(q->q_flag & QFULL)) {
13433                         mutex_enter(QLOCK(q));
13434                         if (!(q->q_flag & QFULL)) {
13435                                 /* still need to set QFULL */
13436                                 q->q_flag |= QFULL;
13437                                 /* set flow_stopped to true under QLOCK */
13438                                 if (flow_stopped != NULL)
13439                                         *flow_stopped = B_TRUE;
13440                                 mutex_exit(QLOCK(q));
13441                         } else {
13442                                 /* flow_stopped is left unchanged */
13443                                 mutex_exit(QLOCK(q));
13444                         }
13445                 }
13446         }
13447 }
13448 
13449 void
13450 conn_clrqfull(conn_t *connp, boolean_t *flow_stopped)
13451 {
13452         if (IPCL_IS_NONSTR(connp)) {
13453                 (*connp->conn_upcalls->su_txq_full)
13454                     (connp->conn_upper_handle, B_FALSE);
13455                 if (flow_stopped != NULL)
13456                         *flow_stopped = B_FALSE;
13457         } else {
13458                 queue_t *q = connp->conn_wq;
13459 
13460                 ASSERT(q != NULL);
13461                 if (q->q_flag & QFULL) {
13462                         mutex_enter(QLOCK(q));
13463                         if (q->q_flag & QFULL) {
13464                                 q->q_flag &= ~QFULL;
13465                                 /* set flow_stopped to false under QLOCK */
13466                                 if (flow_stopped != NULL)
13467                                         *flow_stopped = B_FALSE;
13468                                 mutex_exit(QLOCK(q));
13469                                 if (q->q_flag & QWANTW)
13470                                         qbackenable(q, 0);
13471                         } else {
13472                                 /* flow_stopped is left unchanged */
13473                                 mutex_exit(QLOCK(q));
13474                         }
13475                 }
13476         }
13477 
13478         mutex_enter(&connp->conn_lock);
13479         connp->conn_blocked = B_FALSE;
13480         mutex_exit(&connp->conn_lock);
13481 }
13482 
13483 /*
13484  * Return the length in bytes of the IPv4 headers (base header, label, and
13485  * other IP options) that will be needed based on the
13486  * ip_pkt_t structure passed by the caller.
13487  *
13488  * The returned length does not include the length of the upper level
13489  * protocol (ULP) header.
13490  * The caller needs to check that the length doesn't exceed the max for IPv4.
13491  */
13492 int
13493 ip_total_hdrs_len_v4(const ip_pkt_t *ipp)
13494 {
13495         int len;
13496 
13497         len = IP_SIMPLE_HDR_LENGTH;
13498         if (ipp->ipp_fields & IPPF_LABEL_V4) {
13499                 ASSERT(ipp->ipp_label_len_v4 != 0);
13500                 /* We need to round up here */
13501                 len += (ipp->ipp_label_len_v4 + 3) & ~3;
13502         }
13503 
13504         if (ipp->ipp_fields & IPPF_IPV4_OPTIONS) {
13505                 ASSERT(ipp->ipp_ipv4_options_len != 0);
13506                 ASSERT((ipp->ipp_ipv4_options_len & 3) == 0);
13507                 len += ipp->ipp_ipv4_options_len;
13508         }
13509         return (len);
13510 }
13511 
13512 /*
13513  * All-purpose routine to build an IPv4 header with options based
13514  * on the abstract ip_pkt_t.
13515  *
13516  * The caller has to set the source and destination address as well as
13517  * ipha_length. The caller has to massage any source route and compensate
13518  * for the ULP pseudo-header checksum due to the source route.
13519  */
13520 void
13521 ip_build_hdrs_v4(uchar_t *buf, uint_t buf_len, const ip_pkt_t *ipp,
13522     uint8_t protocol)
13523 {
13524         ipha_t  *ipha = (ipha_t *)buf;
13525         uint8_t *cp;
13526 
13527         /* Initialize IPv4 header */
13528         ipha->ipha_type_of_service = ipp->ipp_type_of_service;
13529         ipha->ipha_length = 0;       /* Caller will set later */
13530         ipha->ipha_ident = 0;
13531         ipha->ipha_fragment_offset_and_flags = 0;
13532         ipha->ipha_ttl = ipp->ipp_unicast_hops;
13533         ipha->ipha_protocol = protocol;
13534         ipha->ipha_hdr_checksum = 0;
13535 
13536         if ((ipp->ipp_fields & IPPF_ADDR) &&
13537             IN6_IS_ADDR_V4MAPPED(&ipp->ipp_addr))
13538                 ipha->ipha_src = ipp->ipp_addr_v4;
13539 
13540         cp = (uint8_t *)&ipha[1];
13541         if (ipp->ipp_fields & IPPF_LABEL_V4) {
13542                 ASSERT(ipp->ipp_label_len_v4 != 0);
13543                 bcopy(ipp->ipp_label_v4, cp, ipp->ipp_label_len_v4);
13544                 cp += ipp->ipp_label_len_v4;
13545                 /* We need to round up here */
13546                 while ((uintptr_t)cp & 0x3) {
13547                         *cp++ = IPOPT_NOP;
13548                 }
13549         }
13550 
13551         if (ipp->ipp_fields & IPPF_IPV4_OPTIONS) {
13552                 ASSERT(ipp->ipp_ipv4_options_len != 0);
13553                 ASSERT((ipp->ipp_ipv4_options_len & 3) == 0);
13554                 bcopy(ipp->ipp_ipv4_options, cp, ipp->ipp_ipv4_options_len);
13555                 cp += ipp->ipp_ipv4_options_len;
13556         }
13557         ipha->ipha_version_and_hdr_length =
13558             (uint8_t)((IP_VERSION << 4) + buf_len / 4);
13559 
13560         ASSERT((int)(cp - buf) == buf_len);
13561 }
13562 
13563 /* Allocate the private structure */
13564 static int
13565 ip_priv_alloc(void **bufp)
13566 {
13567         void    *buf;
13568 
13569         if ((buf = kmem_alloc(sizeof (ip_priv_t), KM_NOSLEEP)) == NULL)
13570                 return (ENOMEM);
13571 
13572         *bufp = buf;
13573         return (0);
13574 }
13575 
13576 /* Function to delete the private structure */
13577 void
13578 ip_priv_free(void *buf)
13579 {
13580         ASSERT(buf != NULL);
13581         kmem_free(buf, sizeof (ip_priv_t));
13582 }
13583 
13584 /*
13585  * The entry point for IPPF processing.
13586  * If the classifier (IPGPC_CLASSIFY) is not loaded and configured, the
13587  * routine just returns.
13588  *
13589  * When called, ip_process generates an ipp_packet_t structure
13590  * which holds the state information for this packet and invokes the
13591  * the classifier (via ipp_packet_process). The classification, depending on
13592  * configured filters, results in a list of actions for this packet. Invoking
13593  * an action may cause the packet to be dropped, in which case we return NULL.
13594  * proc indicates the callout position for
13595  * this packet and ill is the interface this packet arrived on or will leave
13596  * on (inbound and outbound resp.).
13597  *
13598  * We do the processing on the rill (mapped to the upper if ipmp), but MIB
13599  * on the ill corrsponding to the destination IP address.
13600  */
13601 mblk_t *
13602 ip_process(ip_proc_t proc, mblk_t *mp, ill_t *rill, ill_t *ill)
13603 {
13604         ip_priv_t       *priv;
13605         ipp_action_id_t aid;
13606         int             rc = 0;
13607         ipp_packet_t    *pp;
13608 
13609         /* If the classifier is not loaded, return  */
13610         if ((aid = ipp_action_lookup(IPGPC_CLASSIFY)) == IPP_ACTION_INVAL) {
13611                 return (mp);
13612         }
13613 
13614         ASSERT(mp != NULL);
13615 
13616         /* Allocate the packet structure */
13617         rc = ipp_packet_alloc(&pp, "ip", aid);
13618         if (rc != 0)
13619                 goto drop;
13620 
13621         /* Allocate the private structure */
13622         rc = ip_priv_alloc((void **)&priv);
13623         if (rc != 0) {
13624                 ipp_packet_free(pp);
13625                 goto drop;
13626         }
13627         priv->proc = proc;
13628         priv->ill_index = ill_get_upper_ifindex(rill);
13629 
13630         ipp_packet_set_private(pp, priv, ip_priv_free);
13631         ipp_packet_set_data(pp, mp);
13632 
13633         /* Invoke the classifier */
13634         rc = ipp_packet_process(&pp);
13635         if (pp != NULL) {
13636                 mp = ipp_packet_get_data(pp);
13637                 ipp_packet_free(pp);
13638                 if (rc != 0)
13639                         goto drop;
13640                 return (mp);
13641         } else {
13642                 /* No mp to trace in ip_drop_input/ip_drop_output  */
13643                 mp = NULL;
13644         }
13645 drop:
13646         if (proc == IPP_LOCAL_IN || proc == IPP_FWD_IN) {
13647                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
13648                 ip_drop_input("ip_process", mp, ill);
13649         } else {
13650                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
13651                 ip_drop_output("ip_process", mp, ill);
13652         }
13653         freemsg(mp);
13654         return (NULL);
13655 }
13656 
13657 /*
13658  * Propagate a multicast group membership operation (add/drop) on
13659  * all the interfaces crossed by the related multirt routes.
13660  * The call is considered successful if the operation succeeds
13661  * on at least one interface.
13662  *
13663  * This assumes that a set of IRE_HOST/RTF_MULTIRT has been created for the
13664  * multicast addresses with the ire argument being the first one.
13665  * We walk the bucket to find all the of those.
13666  *
13667  * Common to IPv4 and IPv6.
13668  */
13669 static int
13670 ip_multirt_apply_membership(int (*fn)(conn_t *, boolean_t,
13671     const in6_addr_t *, ipaddr_t, uint_t, mcast_record_t, const in6_addr_t *),
13672     ire_t *ire, conn_t *connp, boolean_t checkonly, const in6_addr_t *v6group,
13673     mcast_record_t fmode, const in6_addr_t *v6src)
13674 {
13675         ire_t           *ire_gw;
13676         irb_t           *irb;
13677         int             ifindex;
13678         int             error = 0;
13679         int             result;
13680         ip_stack_t      *ipst = ire->ire_ipst;
13681         ipaddr_t        group;
13682         boolean_t       isv6;
13683         int             match_flags;
13684 
13685         if (IN6_IS_ADDR_V4MAPPED(v6group)) {
13686                 IN6_V4MAPPED_TO_IPADDR(v6group, group);
13687                 isv6 = B_FALSE;
13688         } else {
13689                 isv6 = B_TRUE;
13690         }
13691 
13692         irb = ire->ire_bucket;
13693         ASSERT(irb != NULL);
13694 
13695         result = 0;
13696         irb_refhold(irb);
13697         for (; ire != NULL; ire = ire->ire_next) {
13698                 if ((ire->ire_flags & RTF_MULTIRT) == 0)
13699                         continue;
13700 
13701                 /* We handle -ifp routes by matching on the ill if set */
13702                 match_flags = MATCH_IRE_TYPE;
13703                 if (ire->ire_ill != NULL)
13704                         match_flags |= MATCH_IRE_ILL;
13705 
13706                 if (isv6) {
13707                         if (!IN6_ARE_ADDR_EQUAL(&ire->ire_addr_v6, v6group))
13708                                 continue;
13709 
13710                         ire_gw = ire_ftable_lookup_v6(&ire->ire_gateway_addr_v6,
13711                             0, 0, IRE_INTERFACE, ire->ire_ill, ALL_ZONES, NULL,
13712                             match_flags, 0, ipst, NULL);
13713                 } else {
13714                         if (ire->ire_addr != group)
13715                                 continue;
13716 
13717                         ire_gw = ire_ftable_lookup_v4(ire->ire_gateway_addr,
13718                             0, 0, IRE_INTERFACE, ire->ire_ill, ALL_ZONES, NULL,
13719                             match_flags, 0, ipst, NULL);
13720                 }
13721                 /* No interface route exists for the gateway; skip this ire. */
13722                 if (ire_gw == NULL)
13723                         continue;
13724                 if (ire_gw->ire_flags & (RTF_REJECT|RTF_BLACKHOLE)) {
13725                         ire_refrele(ire_gw);
13726                         continue;
13727                 }
13728                 ASSERT(ire_gw->ire_ill != NULL);     /* IRE_INTERFACE */
13729                 ifindex = ire_gw->ire_ill->ill_phyint->phyint_ifindex;
13730 
13731                 /*
13732                  * The operation is considered a success if
13733                  * it succeeds at least once on any one interface.
13734                  */
13735                 error = fn(connp, checkonly, v6group, INADDR_ANY, ifindex,
13736                     fmode, v6src);
13737                 if (error == 0)
13738                         result = CGTP_MCAST_SUCCESS;
13739 
13740                 ire_refrele(ire_gw);
13741         }
13742         irb_refrele(irb);
13743         /*
13744          * Consider the call as successful if we succeeded on at least
13745          * one interface. Otherwise, return the last encountered error.
13746          */
13747         return (result == CGTP_MCAST_SUCCESS ? 0 : error);
13748 }
13749 
13750 /*
13751  * Return the expected CGTP hooks version number.
13752  */
13753 int
13754 ip_cgtp_filter_supported(void)
13755 {
13756         return (ip_cgtp_filter_rev);
13757 }
13758 
13759 /*
13760  * CGTP hooks can be registered by invoking this function.
13761  * Checks that the version number matches.
13762  */
13763 int
13764 ip_cgtp_filter_register(netstackid_t stackid, cgtp_filter_ops_t *ops)
13765 {
13766         netstack_t *ns;
13767         ip_stack_t *ipst;
13768 
13769         if (ops->cfo_filter_rev != CGTP_FILTER_REV)
13770                 return (ENOTSUP);
13771 
13772         ns = netstack_find_by_stackid(stackid);
13773         if (ns == NULL)
13774                 return (EINVAL);
13775         ipst = ns->netstack_ip;
13776         ASSERT(ipst != NULL);
13777 
13778         if (ipst->ips_ip_cgtp_filter_ops != NULL) {
13779                 netstack_rele(ns);
13780                 return (EALREADY);
13781         }
13782 
13783         ipst->ips_ip_cgtp_filter_ops = ops;
13784 
13785         ill_set_inputfn_all(ipst);
13786 
13787         netstack_rele(ns);
13788         return (0);
13789 }
13790 
13791 /*
13792  * CGTP hooks can be unregistered by invoking this function.
13793  * Returns ENXIO if there was no registration.
13794  * Returns EBUSY if the ndd variable has not been turned off.
13795  */
13796 int
13797 ip_cgtp_filter_unregister(netstackid_t stackid)
13798 {
13799         netstack_t *ns;
13800         ip_stack_t *ipst;
13801 
13802         ns = netstack_find_by_stackid(stackid);
13803         if (ns == NULL)
13804                 return (EINVAL);
13805         ipst = ns->netstack_ip;
13806         ASSERT(ipst != NULL);
13807 
13808         if (ipst->ips_ip_cgtp_filter) {
13809                 netstack_rele(ns);
13810                 return (EBUSY);
13811         }
13812 
13813         if (ipst->ips_ip_cgtp_filter_ops == NULL) {
13814                 netstack_rele(ns);
13815                 return (ENXIO);
13816         }
13817         ipst->ips_ip_cgtp_filter_ops = NULL;
13818 
13819         ill_set_inputfn_all(ipst);
13820 
13821         netstack_rele(ns);
13822         return (0);
13823 }
13824 
13825 /*
13826  * Check whether there is a CGTP filter registration.
13827  * Returns non-zero if there is a registration, otherwise returns zero.
13828  * Note: returns zero if bad stackid.
13829  */
13830 int
13831 ip_cgtp_filter_is_registered(netstackid_t stackid)
13832 {
13833         netstack_t *ns;
13834         ip_stack_t *ipst;
13835         int ret;
13836 
13837         ns = netstack_find_by_stackid(stackid);
13838         if (ns == NULL)
13839                 return (0);
13840         ipst = ns->netstack_ip;
13841         ASSERT(ipst != NULL);
13842 
13843         if (ipst->ips_ip_cgtp_filter_ops != NULL)
13844                 ret = 1;
13845         else
13846                 ret = 0;
13847 
13848         netstack_rele(ns);
13849         return (ret);
13850 }
13851 
13852 static int
13853 ip_squeue_switch(int val)
13854 {
13855         int rval;
13856 
13857         switch (val) {
13858         case IP_SQUEUE_ENTER_NODRAIN:
13859                 rval = SQ_NODRAIN;
13860                 break;
13861         case IP_SQUEUE_ENTER:
13862                 rval = SQ_PROCESS;
13863                 break;
13864         case IP_SQUEUE_FILL:
13865         default:
13866                 rval = SQ_FILL;
13867                 break;
13868         }
13869         return (rval);
13870 }
13871 
13872 static void *
13873 ip_kstat2_init(netstackid_t stackid, ip_stat_t *ip_statisticsp)
13874 {
13875         kstat_t *ksp;
13876 
13877         ip_stat_t template = {
13878                 { "ip_udp_fannorm",             KSTAT_DATA_UINT64 },
13879                 { "ip_udp_fanmb",               KSTAT_DATA_UINT64 },
13880                 { "ip_recv_pullup",             KSTAT_DATA_UINT64 },
13881                 { "ip_db_ref",                  KSTAT_DATA_UINT64 },
13882                 { "ip_notaligned",              KSTAT_DATA_UINT64 },
13883                 { "ip_multimblk",               KSTAT_DATA_UINT64 },
13884                 { "ip_opt",                     KSTAT_DATA_UINT64 },
13885                 { "ipsec_proto_ahesp",          KSTAT_DATA_UINT64 },
13886                 { "ip_conn_flputbq",            KSTAT_DATA_UINT64 },
13887                 { "ip_conn_walk_drain",         KSTAT_DATA_UINT64 },
13888                 { "ip_out_sw_cksum",            KSTAT_DATA_UINT64 },
13889                 { "ip_out_sw_cksum_bytes",      KSTAT_DATA_UINT64 },
13890                 { "ip_in_sw_cksum",             KSTAT_DATA_UINT64 },
13891                 { "ip_ire_reclaim_calls",       KSTAT_DATA_UINT64 },
13892                 { "ip_ire_reclaim_deleted",     KSTAT_DATA_UINT64 },
13893                 { "ip_nce_reclaim_calls",       KSTAT_DATA_UINT64 },
13894                 { "ip_nce_reclaim_deleted",     KSTAT_DATA_UINT64 },
13895                 { "ip_nce_mcast_reclaim_calls", KSTAT_DATA_UINT64 },
13896                 { "ip_nce_mcast_reclaim_deleted",       KSTAT_DATA_UINT64 },
13897                 { "ip_nce_mcast_reclaim_tqfail",        KSTAT_DATA_UINT64 },
13898                 { "ip_dce_reclaim_calls",       KSTAT_DATA_UINT64 },
13899                 { "ip_dce_reclaim_deleted",     KSTAT_DATA_UINT64 },
13900                 { "ip_tcp_in_full_hw_cksum_err",        KSTAT_DATA_UINT64 },
13901                 { "ip_tcp_in_part_hw_cksum_err",        KSTAT_DATA_UINT64 },
13902                 { "ip_tcp_in_sw_cksum_err",             KSTAT_DATA_UINT64 },
13903                 { "ip_udp_in_full_hw_cksum_err",        KSTAT_DATA_UINT64 },
13904                 { "ip_udp_in_part_hw_cksum_err",        KSTAT_DATA_UINT64 },
13905                 { "ip_udp_in_sw_cksum_err",     KSTAT_DATA_UINT64 },
13906                 { "conn_in_recvdstaddr",        KSTAT_DATA_UINT64 },
13907                 { "conn_in_recvopts",           KSTAT_DATA_UINT64 },
13908                 { "conn_in_recvif",             KSTAT_DATA_UINT64 },
13909                 { "conn_in_recvslla",           KSTAT_DATA_UINT64 },
13910                 { "conn_in_recvucred",          KSTAT_DATA_UINT64 },
13911                 { "conn_in_recvttl",            KSTAT_DATA_UINT64 },
13912                 { "conn_in_recvhopopts",        KSTAT_DATA_UINT64 },
13913                 { "conn_in_recvhoplimit",       KSTAT_DATA_UINT64 },
13914                 { "conn_in_recvdstopts",        KSTAT_DATA_UINT64 },
13915                 { "conn_in_recvrthdrdstopts",   KSTAT_DATA_UINT64 },
13916                 { "conn_in_recvrthdr",          KSTAT_DATA_UINT64 },
13917                 { "conn_in_recvpktinfo",        KSTAT_DATA_UINT64 },
13918                 { "conn_in_recvtclass",         KSTAT_DATA_UINT64 },
13919                 { "conn_in_timestamp",          KSTAT_DATA_UINT64 },
13920         };
13921 
13922         ksp = kstat_create_netstack("ip", 0, "ipstat", "net",
13923             KSTAT_TYPE_NAMED, sizeof (template) / sizeof (kstat_named_t),
13924             KSTAT_FLAG_VIRTUAL, stackid);
13925 
13926         if (ksp == NULL)
13927                 return (NULL);
13928 
13929         bcopy(&template, ip_statisticsp, sizeof (template));
13930         ksp->ks_data = (void *)ip_statisticsp;
13931         ksp->ks_private = (void *)(uintptr_t)stackid;
13932 
13933         kstat_install(ksp);
13934         return (ksp);
13935 }
13936 
13937 static void
13938 ip_kstat2_fini(netstackid_t stackid, kstat_t *ksp)
13939 {
13940         if (ksp != NULL) {
13941                 ASSERT(stackid == (netstackid_t)(uintptr_t)ksp->ks_private);
13942                 kstat_delete_netstack(ksp, stackid);
13943         }
13944 }
13945 
13946 static void *
13947 ip_kstat_init(netstackid_t stackid, ip_stack_t *ipst)
13948 {
13949         kstat_t *ksp;
13950 
13951         ip_named_kstat_t template = {
13952                 { "forwarding",         KSTAT_DATA_UINT32, 0 },
13953                 { "defaultTTL",         KSTAT_DATA_UINT32, 0 },
13954                 { "inReceives",         KSTAT_DATA_UINT64, 0 },
13955                 { "inHdrErrors",        KSTAT_DATA_UINT32, 0 },
13956                 { "inAddrErrors",       KSTAT_DATA_UINT32, 0 },
13957                 { "forwDatagrams",      KSTAT_DATA_UINT64, 0 },
13958                 { "inUnknownProtos",    KSTAT_DATA_UINT32, 0 },
13959                 { "inDiscards",         KSTAT_DATA_UINT32, 0 },
13960                 { "inDelivers",         KSTAT_DATA_UINT64, 0 },
13961                 { "outRequests",        KSTAT_DATA_UINT64, 0 },
13962                 { "outDiscards",        KSTAT_DATA_UINT32, 0 },
13963                 { "outNoRoutes",        KSTAT_DATA_UINT32, 0 },
13964                 { "reasmTimeout",       KSTAT_DATA_UINT32, 0 },
13965                 { "reasmReqds",         KSTAT_DATA_UINT32, 0 },
13966                 { "reasmOKs",           KSTAT_DATA_UINT32, 0 },
13967                 { "reasmFails",         KSTAT_DATA_UINT32, 0 },
13968                 { "fragOKs",            KSTAT_DATA_UINT32, 0 },
13969                 { "fragFails",          KSTAT_DATA_UINT32, 0 },
13970                 { "fragCreates",        KSTAT_DATA_UINT32, 0 },
13971                 { "addrEntrySize",      KSTAT_DATA_INT32, 0 },
13972                 { "routeEntrySize",     KSTAT_DATA_INT32, 0 },
13973                 { "netToMediaEntrySize",        KSTAT_DATA_INT32, 0 },
13974                 { "routingDiscards",    KSTAT_DATA_UINT32, 0 },
13975                 { "inErrs",             KSTAT_DATA_UINT32, 0 },
13976                 { "noPorts",            KSTAT_DATA_UINT32, 0 },
13977                 { "inCksumErrs",        KSTAT_DATA_UINT32, 0 },
13978                 { "reasmDuplicates",    KSTAT_DATA_UINT32, 0 },
13979                 { "reasmPartDups",      KSTAT_DATA_UINT32, 0 },
13980                 { "forwProhibits",      KSTAT_DATA_UINT32, 0 },
13981                 { "udpInCksumErrs",     KSTAT_DATA_UINT32, 0 },
13982                 { "udpInOverflows",     KSTAT_DATA_UINT32, 0 },
13983                 { "rawipInOverflows",   KSTAT_DATA_UINT32, 0 },
13984                 { "ipsecInSucceeded",   KSTAT_DATA_UINT32, 0 },
13985                 { "ipsecInFailed",      KSTAT_DATA_INT32, 0 },
13986                 { "memberEntrySize",    KSTAT_DATA_INT32, 0 },
13987                 { "inIPv6",             KSTAT_DATA_UINT32, 0 },
13988                 { "outIPv6",            KSTAT_DATA_UINT32, 0 },
13989                 { "outSwitchIPv6",      KSTAT_DATA_UINT32, 0 },
13990         };
13991 
13992         ksp = kstat_create_netstack("ip", 0, "ip", "mib2", KSTAT_TYPE_NAMED,
13993             NUM_OF_FIELDS(ip_named_kstat_t), 0, stackid);
13994         if (ksp == NULL || ksp->ks_data == NULL)
13995                 return (NULL);
13996 
13997         template.forwarding.value.ui32 = WE_ARE_FORWARDING(ipst) ? 1:2;
13998         template.defaultTTL.value.ui32 = (uint32_t)ipst->ips_ip_def_ttl;
13999         template.reasmTimeout.value.ui32 = ipst->ips_ip_reassembly_timeout;
14000         template.addrEntrySize.value.i32 = sizeof (mib2_ipAddrEntry_t);
14001         template.routeEntrySize.value.i32 = sizeof (mib2_ipRouteEntry_t);
14002 
14003         template.netToMediaEntrySize.value.i32 =
14004             sizeof (mib2_ipNetToMediaEntry_t);
14005 
14006         template.memberEntrySize.value.i32 = sizeof (ipv6_member_t);
14007 
14008         bcopy(&template, ksp->ks_data, sizeof (template));
14009         ksp->ks_update = ip_kstat_update;
14010         ksp->ks_private = (void *)(uintptr_t)stackid;
14011 
14012         kstat_install(ksp);
14013         return (ksp);
14014 }
14015 
14016 static void
14017 ip_kstat_fini(netstackid_t stackid, kstat_t *ksp)
14018 {
14019         if (ksp != NULL) {
14020                 ASSERT(stackid == (netstackid_t)(uintptr_t)ksp->ks_private);
14021                 kstat_delete_netstack(ksp, stackid);
14022         }
14023 }
14024 
14025 static int
14026 ip_kstat_update(kstat_t *kp, int rw)
14027 {
14028         ip_named_kstat_t *ipkp;
14029         mib2_ipIfStatsEntry_t ipmib;
14030         ill_walk_context_t ctx;
14031         ill_t *ill;
14032         netstackid_t    stackid = (zoneid_t)(uintptr_t)kp->ks_private;
14033         netstack_t      *ns;
14034         ip_stack_t      *ipst;
14035 
14036         if (kp->ks_data == NULL)
14037                 return (EIO);
14038 
14039         if (rw == KSTAT_WRITE)
14040                 return (EACCES);
14041 
14042         ns = netstack_find_by_stackid(stackid);
14043         if (ns == NULL)
14044                 return (-1);
14045         ipst = ns->netstack_ip;
14046         if (ipst == NULL) {
14047                 netstack_rele(ns);
14048                 return (-1);
14049         }
14050         ipkp = (ip_named_kstat_t *)kp->ks_data;
14051 
14052         bcopy(&ipst->ips_ip_mib, &ipmib, sizeof (ipmib));
14053         rw_enter(&ipst->ips_ill_g_lock, RW_READER);
14054         ill = ILL_START_WALK_V4(&ctx, ipst);
14055         for (; ill != NULL; ill = ill_next(&ctx, ill))
14056                 ip_mib2_add_ip_stats(&ipmib, ill->ill_ip_mib);
14057         rw_exit(&ipst->ips_ill_g_lock);
14058 
14059         ipkp->forwarding.value.ui32 =                ipmib.ipIfStatsForwarding;
14060         ipkp->defaultTTL.value.ui32 =                ipmib.ipIfStatsDefaultTTL;
14061         ipkp->inReceives.value.ui64 =                ipmib.ipIfStatsHCInReceives;
14062         ipkp->inHdrErrors.value.ui32 =               ipmib.ipIfStatsInHdrErrors;
14063         ipkp->inAddrErrors.value.ui32 =              ipmib.ipIfStatsInAddrErrors;
14064         ipkp->forwDatagrams.value.ui64 = ipmib.ipIfStatsHCOutForwDatagrams;
14065         ipkp->inUnknownProtos.value.ui32 =   ipmib.ipIfStatsInUnknownProtos;
14066         ipkp->inDiscards.value.ui32 =                ipmib.ipIfStatsInDiscards;
14067         ipkp->inDelivers.value.ui64 =                ipmib.ipIfStatsHCInDelivers;
14068         ipkp->outRequests.value.ui64 =               ipmib.ipIfStatsHCOutRequests;
14069         ipkp->outDiscards.value.ui32 =               ipmib.ipIfStatsOutDiscards;
14070         ipkp->outNoRoutes.value.ui32 =               ipmib.ipIfStatsOutNoRoutes;
14071         ipkp->reasmTimeout.value.ui32 =              ipst->ips_ip_reassembly_timeout;
14072         ipkp->reasmReqds.value.ui32 =                ipmib.ipIfStatsReasmReqds;
14073         ipkp->reasmOKs.value.ui32 =          ipmib.ipIfStatsReasmOKs;
14074         ipkp->reasmFails.value.ui32 =                ipmib.ipIfStatsReasmFails;
14075         ipkp->fragOKs.value.ui32 =           ipmib.ipIfStatsOutFragOKs;
14076         ipkp->fragFails.value.ui32 =         ipmib.ipIfStatsOutFragFails;
14077         ipkp->fragCreates.value.ui32 =               ipmib.ipIfStatsOutFragCreates;
14078 
14079         ipkp->routingDiscards.value.ui32 =   0;
14080         ipkp->inErrs.value.ui32 =            ipmib.tcpIfStatsInErrs;
14081         ipkp->noPorts.value.ui32 =           ipmib.udpIfStatsNoPorts;
14082         ipkp->inCksumErrs.value.ui32 =               ipmib.ipIfStatsInCksumErrs;
14083         ipkp->reasmDuplicates.value.ui32 =   ipmib.ipIfStatsReasmDuplicates;
14084         ipkp->reasmPartDups.value.ui32 =     ipmib.ipIfStatsReasmPartDups;
14085         ipkp->forwProhibits.value.ui32 =     ipmib.ipIfStatsForwProhibits;
14086         ipkp->udpInCksumErrs.value.ui32 =    ipmib.udpIfStatsInCksumErrs;
14087         ipkp->udpInOverflows.value.ui32 =    ipmib.udpIfStatsInOverflows;
14088         ipkp->rawipInOverflows.value.ui32 =  ipmib.rawipIfStatsInOverflows;
14089         ipkp->ipsecInSucceeded.value.ui32 =  ipmib.ipsecIfStatsInSucceeded;
14090         ipkp->ipsecInFailed.value.i32 =              ipmib.ipsecIfStatsInFailed;
14091 
14092         ipkp->inIPv6.value.ui32 =    ipmib.ipIfStatsInWrongIPVersion;
14093         ipkp->outIPv6.value.ui32 =   ipmib.ipIfStatsOutWrongIPVersion;
14094         ipkp->outSwitchIPv6.value.ui32 = ipmib.ipIfStatsOutSwitchIPVersion;
14095 
14096         netstack_rele(ns);
14097 
14098         return (0);
14099 }
14100 
14101 static void *
14102 icmp_kstat_init(netstackid_t stackid)
14103 {
14104         kstat_t *ksp;
14105 
14106         icmp_named_kstat_t template = {
14107                 { "inMsgs",             KSTAT_DATA_UINT32 },
14108                 { "inErrors",           KSTAT_DATA_UINT32 },
14109                 { "inDestUnreachs",     KSTAT_DATA_UINT32 },
14110                 { "inTimeExcds",        KSTAT_DATA_UINT32 },
14111                 { "inParmProbs",        KSTAT_DATA_UINT32 },
14112                 { "inSrcQuenchs",       KSTAT_DATA_UINT32 },
14113                 { "inRedirects",        KSTAT_DATA_UINT32 },
14114                 { "inEchos",            KSTAT_DATA_UINT32 },
14115                 { "inEchoReps",         KSTAT_DATA_UINT32 },
14116                 { "inTimestamps",       KSTAT_DATA_UINT32 },
14117                 { "inTimestampReps",    KSTAT_DATA_UINT32 },
14118                 { "inAddrMasks",        KSTAT_DATA_UINT32 },
14119                 { "inAddrMaskReps",     KSTAT_DATA_UINT32 },
14120                 { "outMsgs",            KSTAT_DATA_UINT32 },
14121                 { "outErrors",          KSTAT_DATA_UINT32 },
14122                 { "outDestUnreachs",    KSTAT_DATA_UINT32 },
14123                 { "outTimeExcds",       KSTAT_DATA_UINT32 },
14124                 { "outParmProbs",       KSTAT_DATA_UINT32 },
14125                 { "outSrcQuenchs",      KSTAT_DATA_UINT32 },
14126                 { "outRedirects",       KSTAT_DATA_UINT32 },
14127                 { "outEchos",           KSTAT_DATA_UINT32 },
14128                 { "outEchoReps",        KSTAT_DATA_UINT32 },
14129                 { "outTimestamps",      KSTAT_DATA_UINT32 },
14130                 { "outTimestampReps",   KSTAT_DATA_UINT32 },
14131                 { "outAddrMasks",       KSTAT_DATA_UINT32 },
14132                 { "outAddrMaskReps",    KSTAT_DATA_UINT32 },
14133                 { "inChksumErrs",       KSTAT_DATA_UINT32 },
14134                 { "inUnknowns",         KSTAT_DATA_UINT32 },
14135                 { "inFragNeeded",       KSTAT_DATA_UINT32 },
14136                 { "outFragNeeded",      KSTAT_DATA_UINT32 },
14137                 { "outDrops",           KSTAT_DATA_UINT32 },
14138                 { "inOverFlows",        KSTAT_DATA_UINT32 },
14139                 { "inBadRedirects",     KSTAT_DATA_UINT32 },
14140         };
14141 
14142         ksp = kstat_create_netstack("ip", 0, "icmp", "mib2", KSTAT_TYPE_NAMED,
14143             NUM_OF_FIELDS(icmp_named_kstat_t), 0, stackid);
14144         if (ksp == NULL || ksp->ks_data == NULL)
14145                 return (NULL);
14146 
14147         bcopy(&template, ksp->ks_data, sizeof (template));
14148 
14149         ksp->ks_update = icmp_kstat_update;
14150         ksp->ks_private = (void *)(uintptr_t)stackid;
14151 
14152         kstat_install(ksp);
14153         return (ksp);
14154 }
14155 
14156 static void
14157 icmp_kstat_fini(netstackid_t stackid, kstat_t *ksp)
14158 {
14159         if (ksp != NULL) {
14160                 ASSERT(stackid == (netstackid_t)(uintptr_t)ksp->ks_private);
14161                 kstat_delete_netstack(ksp, stackid);
14162         }
14163 }
14164 
14165 static int
14166 icmp_kstat_update(kstat_t *kp, int rw)
14167 {
14168         icmp_named_kstat_t *icmpkp;
14169         netstackid_t    stackid = (zoneid_t)(uintptr_t)kp->ks_private;
14170         netstack_t      *ns;
14171         ip_stack_t      *ipst;
14172 
14173         if (kp->ks_data == NULL)
14174                 return (EIO);
14175 
14176         if (rw == KSTAT_WRITE)
14177                 return (EACCES);
14178 
14179         ns = netstack_find_by_stackid(stackid);
14180         if (ns == NULL)
14181                 return (-1);
14182         ipst = ns->netstack_ip;
14183         if (ipst == NULL) {
14184                 netstack_rele(ns);
14185                 return (-1);
14186         }
14187         icmpkp = (icmp_named_kstat_t *)kp->ks_data;
14188 
14189         icmpkp->inMsgs.value.ui32 =      ipst->ips_icmp_mib.icmpInMsgs;
14190         icmpkp->inErrors.value.ui32 =            ipst->ips_icmp_mib.icmpInErrors;
14191         icmpkp->inDestUnreachs.value.ui32 =
14192             ipst->ips_icmp_mib.icmpInDestUnreachs;
14193         icmpkp->inTimeExcds.value.ui32 =    ipst->ips_icmp_mib.icmpInTimeExcds;
14194         icmpkp->inParmProbs.value.ui32 =    ipst->ips_icmp_mib.icmpInParmProbs;
14195         icmpkp->inSrcQuenchs.value.ui32 =   ipst->ips_icmp_mib.icmpInSrcQuenchs;
14196         icmpkp->inRedirects.value.ui32 =    ipst->ips_icmp_mib.icmpInRedirects;
14197         icmpkp->inEchos.value.ui32 =     ipst->ips_icmp_mib.icmpInEchos;
14198         icmpkp->inEchoReps.value.ui32 =          ipst->ips_icmp_mib.icmpInEchoReps;
14199         icmpkp->inTimestamps.value.ui32 =   ipst->ips_icmp_mib.icmpInTimestamps;
14200         icmpkp->inTimestampReps.value.ui32 =
14201             ipst->ips_icmp_mib.icmpInTimestampReps;
14202         icmpkp->inAddrMasks.value.ui32 =    ipst->ips_icmp_mib.icmpInAddrMasks;
14203         icmpkp->inAddrMaskReps.value.ui32 =
14204             ipst->ips_icmp_mib.icmpInAddrMaskReps;
14205         icmpkp->outMsgs.value.ui32 =     ipst->ips_icmp_mib.icmpOutMsgs;
14206         icmpkp->outErrors.value.ui32 =           ipst->ips_icmp_mib.icmpOutErrors;
14207         icmpkp->outDestUnreachs.value.ui32 =
14208             ipst->ips_icmp_mib.icmpOutDestUnreachs;
14209         icmpkp->outTimeExcds.value.ui32 =   ipst->ips_icmp_mib.icmpOutTimeExcds;
14210         icmpkp->outParmProbs.value.ui32 =   ipst->ips_icmp_mib.icmpOutParmProbs;
14211         icmpkp->outSrcQuenchs.value.ui32 =
14212             ipst->ips_icmp_mib.icmpOutSrcQuenchs;
14213         icmpkp->outRedirects.value.ui32 =   ipst->ips_icmp_mib.icmpOutRedirects;
14214         icmpkp->outEchos.value.ui32 =            ipst->ips_icmp_mib.icmpOutEchos;
14215         icmpkp->outEchoReps.value.ui32 =    ipst->ips_icmp_mib.icmpOutEchoReps;
14216         icmpkp->outTimestamps.value.ui32 =
14217             ipst->ips_icmp_mib.icmpOutTimestamps;
14218         icmpkp->outTimestampReps.value.ui32 =
14219             ipst->ips_icmp_mib.icmpOutTimestampReps;
14220         icmpkp->outAddrMasks.value.ui32 =
14221             ipst->ips_icmp_mib.icmpOutAddrMasks;
14222         icmpkp->outAddrMaskReps.value.ui32 =
14223             ipst->ips_icmp_mib.icmpOutAddrMaskReps;
14224         icmpkp->inCksumErrs.value.ui32 =    ipst->ips_icmp_mib.icmpInCksumErrs;
14225         icmpkp->inUnknowns.value.ui32 =          ipst->ips_icmp_mib.icmpInUnknowns;
14226         icmpkp->inFragNeeded.value.ui32 =   ipst->ips_icmp_mib.icmpInFragNeeded;
14227         icmpkp->outFragNeeded.value.ui32 =
14228             ipst->ips_icmp_mib.icmpOutFragNeeded;
14229         icmpkp->outDrops.value.ui32 =            ipst->ips_icmp_mib.icmpOutDrops;
14230         icmpkp->inOverflows.value.ui32 =    ipst->ips_icmp_mib.icmpInOverflows;
14231         icmpkp->inBadRedirects.value.ui32 =
14232             ipst->ips_icmp_mib.icmpInBadRedirects;
14233 
14234         netstack_rele(ns);
14235         return (0);
14236 }
14237 
14238 /*
14239  * This is the fanout function for raw socket opened for SCTP.  Note
14240  * that it is called after SCTP checks that there is no socket which
14241  * wants a packet.  Then before SCTP handles this out of the blue packet,
14242  * this function is called to see if there is any raw socket for SCTP.
14243  * If there is and it is bound to the correct address, the packet will
14244  * be sent to that socket.  Note that only one raw socket can be bound to
14245  * a port.  This is assured in ipcl_sctp_hash_insert();
14246  */
14247 void
14248 ip_fanout_sctp_raw(mblk_t *mp, ipha_t *ipha, ip6_t *ip6h, uint32_t ports,
14249     ip_recv_attr_t *ira)
14250 {
14251         conn_t          *connp;
14252         queue_t         *rq;
14253         boolean_t       secure;
14254         ill_t           *ill = ira->ira_ill;
14255         ip_stack_t      *ipst = ill->ill_ipst;
14256         ipsec_stack_t   *ipss = ipst->ips_netstack->netstack_ipsec;
14257         sctp_stack_t    *sctps = ipst->ips_netstack->netstack_sctp;
14258         iaflags_t       iraflags = ira->ira_flags;
14259         ill_t           *rill = ira->ira_rill;
14260 
14261         secure = iraflags & IRAF_IPSEC_SECURE;
14262 
14263         connp = ipcl_classify_raw(mp, IPPROTO_SCTP, ports, ipha, ip6h,
14264             ira, ipst);
14265         if (connp == NULL) {
14266                 /*
14267                  * Although raw sctp is not summed, OOB chunks must be.
14268                  * Drop the packet here if the sctp checksum failed.
14269                  */
14270                 if (iraflags & IRAF_SCTP_CSUM_ERR) {
14271                         SCTPS_BUMP_MIB(sctps, sctpChecksumError);
14272                         freemsg(mp);
14273                         return;
14274                 }
14275                 ira->ira_ill = ira->ira_rill = NULL;
14276                 sctp_ootb_input(mp, ira, ipst);
14277                 ira->ira_ill = ill;
14278                 ira->ira_rill = rill;
14279                 return;
14280         }
14281         rq = connp->conn_rq;
14282         if (IPCL_IS_NONSTR(connp) ? connp->conn_flow_cntrld : !canputnext(rq)) {
14283                 CONN_DEC_REF(connp);
14284                 BUMP_MIB(ill->ill_ip_mib, rawipIfStatsInOverflows);
14285                 freemsg(mp);
14286                 return;
14287         }
14288         if (((iraflags & IRAF_IS_IPV4) ?
14289             CONN_INBOUND_POLICY_PRESENT(connp, ipss) :
14290             CONN_INBOUND_POLICY_PRESENT_V6(connp, ipss)) ||
14291             secure) {
14292                 mp = ipsec_check_inbound_policy(mp, connp, ipha,
14293                     ip6h, ira);
14294                 if (mp == NULL) {
14295                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
14296                         /* Note that mp is NULL */
14297                         ip_drop_input("ipIfStatsInDiscards", mp, ill);
14298                         CONN_DEC_REF(connp);
14299                         return;
14300                 }
14301         }
14302 
14303         if (iraflags & IRAF_ICMP_ERROR) {
14304                 (connp->conn_recvicmp)(connp, mp, NULL, ira);
14305         } else {
14306                 ill_t *rill = ira->ira_rill;
14307 
14308                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCInDelivers);
14309                 /* This is the SOCK_RAW, IPPROTO_SCTP case. */
14310                 ira->ira_ill = ira->ira_rill = NULL;
14311                 (connp->conn_recv)(connp, mp, NULL, ira);
14312                 ira->ira_ill = ill;
14313                 ira->ira_rill = rill;
14314         }
14315         CONN_DEC_REF(connp);
14316 }
14317 
14318 /*
14319  * Free a packet that has the link-layer dl_unitdata_req_t or fast-path
14320  * header before the ip payload.
14321  */
14322 static void
14323 ip_xmit_flowctl_drop(ill_t *ill, mblk_t *mp, boolean_t is_fp_mp, int fp_mp_len)
14324 {
14325         int len = (mp->b_wptr - mp->b_rptr);
14326         mblk_t *ip_mp;
14327 
14328         BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
14329         if (is_fp_mp || len != fp_mp_len) {
14330                 if (len > fp_mp_len) {
14331                         /*
14332                          * fastpath header and ip header in the first mblk
14333                          */
14334                         mp->b_rptr += fp_mp_len;
14335                 } else {
14336                         /*
14337                          * ip_xmit_attach_llhdr had to prepend an mblk to
14338                          * attach the fastpath header before ip header.
14339                          */
14340                         ip_mp = mp->b_cont;
14341                         freeb(mp);
14342                         mp = ip_mp;
14343                         mp->b_rptr += (fp_mp_len - len);
14344                 }
14345         } else {
14346                 ip_mp = mp->b_cont;
14347                 freeb(mp);
14348                 mp = ip_mp;
14349         }
14350         ip_drop_output("ipIfStatsOutDiscards - flow ctl", mp, ill);
14351         freemsg(mp);
14352 }
14353 
14354 /*
14355  * Normal post fragmentation function.
14356  *
14357  * Send a packet using the passed in nce. This handles both IPv4 and IPv6
14358  * using the same state machine.
14359  *
14360  * We return an error on failure. In particular we return EWOULDBLOCK
14361  * when the driver flow controls. In that case this ensures that ip_wsrv runs
14362  * (currently by canputnext failure resulting in backenabling from GLD.)
14363  * This allows the callers of conn_ip_output() to use EWOULDBLOCK as an
14364  * indication that they can flow control until ip_wsrv() tells then to restart.
14365  *
14366  * If the nce passed by caller is incomplete, this function
14367  * queues the packet and if necessary, sends ARP request and bails.
14368  * If the Neighbor Cache passed is fully resolved, we simply prepend
14369  * the link-layer header to the packet, do ipsec hw acceleration
14370  * work if necessary, and send the packet out on the wire.
14371  */
14372 /* ARGSUSED6 */
14373 int
14374 ip_xmit(mblk_t *mp, nce_t *nce, iaflags_t ixaflags, uint_t pkt_len,
14375     uint32_t xmit_hint, zoneid_t szone, zoneid_t nolzid, uintptr_t *ixacookie)
14376 {
14377         queue_t         *wq;
14378         ill_t           *ill = nce->nce_ill;
14379         ip_stack_t      *ipst = ill->ill_ipst;
14380         uint64_t        delta;
14381         boolean_t       isv6 = ill->ill_isv6;
14382         boolean_t       fp_mp;
14383         ncec_t          *ncec = nce->nce_common;
14384         int64_t         now = LBOLT_FASTPATH64;
14385         boolean_t       is_probe;
14386 
14387         DTRACE_PROBE1(ip__xmit, nce_t *, nce);
14388 
14389         ASSERT(mp != NULL);
14390         ASSERT(mp->b_datap->db_type == M_DATA);
14391         ASSERT(pkt_len == msgdsize(mp));
14392 
14393         /*
14394          * If we have already been here and are coming back after ARP/ND.
14395          * the IXAF_NO_TRACE flag is set. We skip FW_HOOKS, DTRACE and ipobs
14396          * in that case since they have seen the packet when it came here
14397          * the first time.
14398          */
14399         if (ixaflags & IXAF_NO_TRACE)
14400                 goto sendit;
14401 
14402         if (ixaflags & IXAF_IS_IPV4) {
14403                 ipha_t *ipha = (ipha_t *)mp->b_rptr;
14404 
14405                 ASSERT(!isv6);
14406                 ASSERT(pkt_len == ntohs(((ipha_t *)mp->b_rptr)->ipha_length));
14407                 if (HOOKS4_INTERESTED_PHYSICAL_OUT(ipst) &&
14408                     !(ixaflags & IXAF_NO_PFHOOK)) {
14409                         int     error;
14410 
14411                         FW_HOOKS(ipst->ips_ip4_physical_out_event,
14412                             ipst->ips_ipv4firewall_physical_out,
14413                             NULL, ill, ipha, mp, mp, 0, ipst, error);
14414                         DTRACE_PROBE1(ip4__physical__out__end,
14415                             mblk_t *, mp);
14416                         if (mp == NULL)
14417                                 return (error);
14418 
14419                         /* The length could have changed */
14420                         pkt_len = msgdsize(mp);
14421                 }
14422                 if (ipst->ips_ip4_observe.he_interested) {
14423                         /*
14424                          * Note that for TX the zoneid is the sending
14425                          * zone, whether or not MLP is in play.
14426                          * Since the szone argument is the IP zoneid (i.e.,
14427                          * zero for exclusive-IP zones) and ipobs wants
14428                          * the system zoneid, we map it here.
14429                          */
14430                         szone = IP_REAL_ZONEID(szone, ipst);
14431 
14432                         /*
14433                          * On the outbound path the destination zone will be
14434                          * unknown as we're sending this packet out on the
14435                          * wire.
14436                          */
14437                         ipobs_hook(mp, IPOBS_HOOK_OUTBOUND, szone, ALL_ZONES,
14438                             ill, ipst);
14439                 }
14440                 DTRACE_IP7(send, mblk_t *, mp,  conn_t *, NULL,
14441                     void_ip_t *, ipha,  __dtrace_ipsr_ill_t *, ill,
14442                     ipha_t *, ipha, ip6_t *, NULL, int, 0);
14443         } else {
14444                 ip6_t *ip6h = (ip6_t *)mp->b_rptr;
14445 
14446                 ASSERT(isv6);
14447                 ASSERT(pkt_len ==
14448                     ntohs(((ip6_t *)mp->b_rptr)->ip6_plen) + IPV6_HDR_LEN);
14449                 if (HOOKS6_INTERESTED_PHYSICAL_OUT(ipst) &&
14450                     !(ixaflags & IXAF_NO_PFHOOK)) {
14451                         int     error;
14452 
14453                         FW_HOOKS6(ipst->ips_ip6_physical_out_event,
14454                             ipst->ips_ipv6firewall_physical_out,
14455                             NULL, ill, ip6h, mp, mp, 0, ipst, error);
14456                         DTRACE_PROBE1(ip6__physical__out__end,
14457                             mblk_t *, mp);
14458                         if (mp == NULL)
14459                                 return (error);
14460 
14461                         /* The length could have changed */
14462                         pkt_len = msgdsize(mp);
14463                 }
14464                 if (ipst->ips_ip6_observe.he_interested) {
14465                         /* See above */
14466                         szone = IP_REAL_ZONEID(szone, ipst);
14467 
14468                         ipobs_hook(mp, IPOBS_HOOK_OUTBOUND, szone, ALL_ZONES,
14469                             ill, ipst);
14470                 }
14471                 DTRACE_IP7(send, mblk_t *, mp,  conn_t *, NULL,
14472                     void_ip_t *, ip6h,  __dtrace_ipsr_ill_t *, ill,
14473                     ipha_t *, NULL, ip6_t *, ip6h, int, 0);
14474         }
14475 
14476 sendit:
14477         /*
14478          * We check the state without a lock because the state can never
14479          * move "backwards" to initial or incomplete.
14480          */
14481         switch (ncec->ncec_state) {
14482         case ND_REACHABLE:
14483         case ND_STALE:
14484         case ND_DELAY:
14485         case ND_PROBE:
14486                 mp = ip_xmit_attach_llhdr(mp, nce);
14487                 if (mp == NULL) {
14488                         /*
14489                          * ip_xmit_attach_llhdr has increased
14490                          * ipIfStatsOutDiscards and called ip_drop_output()
14491                          */
14492                         return (ENOBUFS);
14493                 }
14494                 /*
14495                  * check if nce_fastpath completed and we tagged on a
14496                  * copy of nce_fp_mp in ip_xmit_attach_llhdr().
14497                  */
14498                 fp_mp = (mp->b_datap->db_type == M_DATA);
14499 
14500                 if (fp_mp &&
14501                     (ill->ill_capabilities & ILL_CAPAB_DLD_DIRECT)) {
14502                         ill_dld_direct_t *idd;
14503 
14504                         idd = &ill->ill_dld_capab->idc_direct;
14505                         /*
14506                          * Send the packet directly to DLD, where it
14507                          * may be queued depending on the availability
14508                          * of transmit resources at the media layer.
14509                          * Return value should be taken into
14510                          * account and flow control the TCP.
14511                          */
14512                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCOutTransmits);
14513                         UPDATE_MIB(ill->ill_ip_mib, ipIfStatsHCOutOctets,
14514                             pkt_len);
14515 
14516                         if (ixaflags & IXAF_NO_DEV_FLOW_CTL) {
14517                                 (void) idd->idd_tx_df(idd->idd_tx_dh, mp,
14518                                     (uintptr_t)xmit_hint, IP_DROP_ON_NO_DESC);
14519                         } else {
14520                                 uintptr_t cookie;
14521 
14522                                 if ((cookie = idd->idd_tx_df(idd->idd_tx_dh,
14523                                     mp, (uintptr_t)xmit_hint, 0)) != 0) {
14524                                         if (ixacookie != NULL)
14525                                                 *ixacookie = cookie;
14526                                         return (EWOULDBLOCK);
14527                                 }
14528                         }
14529                 } else {
14530                         wq = ill->ill_wq;
14531 
14532                         if (!(ixaflags & IXAF_NO_DEV_FLOW_CTL) &&
14533                             !canputnext(wq)) {
14534                                 if (ixacookie != NULL)
14535                                         *ixacookie = 0;
14536                                 ip_xmit_flowctl_drop(ill, mp, fp_mp,
14537                                     nce->nce_fp_mp != NULL ?
14538                                     MBLKL(nce->nce_fp_mp) : 0);
14539                                 return (EWOULDBLOCK);
14540                         }
14541                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCOutTransmits);
14542                         UPDATE_MIB(ill->ill_ip_mib, ipIfStatsHCOutOctets,
14543                             pkt_len);
14544                         putnext(wq, mp);
14545                 }
14546 
14547                 /*
14548                  * The rest of this function implements Neighbor Unreachability
14549                  * detection. Determine if the ncec is eligible for NUD.
14550                  */
14551                 if (ncec->ncec_flags & NCE_F_NONUD)
14552                         return (0);
14553 
14554                 ASSERT(ncec->ncec_state != ND_INCOMPLETE);
14555 
14556                 /*
14557                  * Check for upper layer advice
14558                  */
14559                 if (ixaflags & IXAF_REACH_CONF) {
14560                         timeout_id_t tid;
14561 
14562                         /*
14563                          * It should be o.k. to check the state without
14564                          * a lock here, at most we lose an advice.
14565                          */
14566                         ncec->ncec_last = TICK_TO_MSEC(now);
14567                         if (ncec->ncec_state != ND_REACHABLE) {
14568                                 mutex_enter(&ncec->ncec_lock);
14569                                 ncec->ncec_state = ND_REACHABLE;
14570                                 tid = ncec->ncec_timeout_id;
14571                                 ncec->ncec_timeout_id = 0;
14572                                 mutex_exit(&ncec->ncec_lock);
14573                                 (void) untimeout(tid);
14574                                 if (ip_debug > 2) {
14575                                         /* ip1dbg */
14576                                         pr_addr_dbg("ip_xmit: state"
14577                                             " for %s changed to"
14578                                             " REACHABLE\n", AF_INET6,
14579                                             &ncec->ncec_addr);
14580                                 }
14581                         }
14582                         return (0);
14583                 }
14584 
14585                 delta =  TICK_TO_MSEC(now) - ncec->ncec_last;
14586                 ip1dbg(("ip_xmit: delta = %" PRId64
14587                     " ill_reachable_time = %d \n", delta,
14588                     ill->ill_reachable_time));
14589                 if (delta > (uint64_t)ill->ill_reachable_time) {
14590                         mutex_enter(&ncec->ncec_lock);
14591                         switch (ncec->ncec_state) {
14592                         case ND_REACHABLE:
14593                                 ASSERT((ncec->ncec_flags & NCE_F_NONUD) == 0);
14594                                 /* FALLTHROUGH */
14595                         case ND_STALE:
14596                                 /*
14597                                  * ND_REACHABLE is identical to
14598                                  * ND_STALE in this specific case. If
14599                                  * reachable time has expired for this
14600                                  * neighbor (delta is greater than
14601                                  * reachable time), conceptually, the
14602                                  * neighbor cache is no longer in
14603                                  * REACHABLE state, but already in
14604                                  * STALE state.  So the correct
14605                                  * transition here is to ND_DELAY.
14606                                  */
14607                                 ncec->ncec_state = ND_DELAY;
14608                                 mutex_exit(&ncec->ncec_lock);
14609                                 nce_restart_timer(ncec,
14610                                     ipst->ips_delay_first_probe_time);
14611                                 if (ip_debug > 3) {
14612                                         /* ip2dbg */
14613                                         pr_addr_dbg("ip_xmit: state"
14614                                             " for %s changed to"
14615                                             " DELAY\n", AF_INET6,
14616                                             &ncec->ncec_addr);
14617                                 }
14618                                 break;
14619                         case ND_DELAY:
14620                         case ND_PROBE:
14621                                 mutex_exit(&ncec->ncec_lock);
14622                                 /* Timers have already started */
14623                                 break;
14624                         case ND_UNREACHABLE:
14625                                 /*
14626                                  * nce_timer has detected that this ncec
14627                                  * is unreachable and initiated deleting
14628                                  * this ncec.
14629                                  * This is a harmless race where we found the
14630                                  * ncec before it was deleted and have
14631                                  * just sent out a packet using this
14632                                  * unreachable ncec.
14633                                  */
14634                                 mutex_exit(&ncec->ncec_lock);
14635                                 break;
14636                         default:
14637                                 ASSERT(0);
14638                                 mutex_exit(&ncec->ncec_lock);
14639                         }
14640                 }
14641                 return (0);
14642 
14643         case ND_INCOMPLETE:
14644                 /*
14645                  * the state could have changed since we didn't hold the lock.
14646                  * Re-verify state under lock.
14647                  */
14648                 is_probe = ipmp_packet_is_probe(mp, nce->nce_ill);
14649                 mutex_enter(&ncec->ncec_lock);
14650                 if (NCE_ISREACHABLE(ncec)) {
14651                         mutex_exit(&ncec->ncec_lock);
14652                         goto sendit;
14653                 }
14654                 /* queue the packet */
14655                 nce_queue_mp(ncec, mp, is_probe);
14656                 mutex_exit(&ncec->ncec_lock);
14657                 DTRACE_PROBE2(ip__xmit__incomplete,
14658                     (ncec_t *), ncec, (mblk_t *), mp);
14659                 return (0);
14660 
14661         case ND_INITIAL:
14662                 /*
14663                  * State could have changed since we didn't hold the lock, so
14664                  * re-verify state.
14665                  */
14666                 is_probe = ipmp_packet_is_probe(mp, nce->nce_ill);
14667                 mutex_enter(&ncec->ncec_lock);
14668                 if (NCE_ISREACHABLE(ncec))  {
14669                         mutex_exit(&ncec->ncec_lock);
14670                         goto sendit;
14671                 }
14672                 nce_queue_mp(ncec, mp, is_probe);
14673                 if (ncec->ncec_state == ND_INITIAL) {
14674                         ncec->ncec_state = ND_INCOMPLETE;
14675                         mutex_exit(&ncec->ncec_lock);
14676                         /*
14677                          * figure out the source we want to use
14678                          * and resolve it.
14679                          */
14680                         ip_ndp_resolve(ncec);
14681                 } else  {
14682                         mutex_exit(&ncec->ncec_lock);
14683                 }
14684                 return (0);
14685 
14686         case ND_UNREACHABLE:
14687                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
14688                 ip_drop_output("ipIfStatsOutDiscards - ND_UNREACHABLE",
14689                     mp, ill);
14690                 freemsg(mp);
14691                 return (0);
14692 
14693         default:
14694                 ASSERT(0);
14695                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
14696                 ip_drop_output("ipIfStatsOutDiscards - ND_other",
14697                     mp, ill);
14698                 freemsg(mp);
14699                 return (ENETUNREACH);
14700         }
14701 }
14702 
14703 /*
14704  * Return B_TRUE if the buffers differ in length or content.
14705  * This is used for comparing extension header buffers.
14706  * Note that an extension header would be declared different
14707  * even if all that changed was the next header value in that header i.e.
14708  * what really changed is the next extension header.
14709  */
14710 boolean_t
14711 ip_cmpbuf(const void *abuf, uint_t alen, boolean_t b_valid, const void *bbuf,
14712     uint_t blen)
14713 {
14714         if (!b_valid)
14715                 blen = 0;
14716 
14717         if (alen != blen)
14718                 return (B_TRUE);
14719         if (alen == 0)
14720                 return (B_FALSE);       /* Both zero length */
14721         return (bcmp(abuf, bbuf, alen));
14722 }
14723 
14724 /*
14725  * Preallocate memory for ip_savebuf(). Returns B_TRUE if ok.
14726  * Return B_FALSE if memory allocation fails - don't change any state!
14727  */
14728 boolean_t
14729 ip_allocbuf(void **dstp, uint_t *dstlenp, boolean_t src_valid,
14730     const void *src, uint_t srclen)
14731 {
14732         void *dst;
14733 
14734         if (!src_valid)
14735                 srclen = 0;
14736 
14737         ASSERT(*dstlenp == 0);
14738         if (src != NULL && srclen != 0) {
14739                 dst = mi_alloc(srclen, BPRI_MED);
14740                 if (dst == NULL)
14741                         return (B_FALSE);
14742         } else {
14743                 dst = NULL;
14744         }
14745         if (*dstp != NULL)
14746                 mi_free(*dstp);
14747         *dstp = dst;
14748         *dstlenp = dst == NULL ? 0 : srclen;
14749         return (B_TRUE);
14750 }
14751 
14752 /*
14753  * Replace what is in *dst, *dstlen with the source.
14754  * Assumes ip_allocbuf has already been called.
14755  */
14756 void
14757 ip_savebuf(void **dstp, uint_t *dstlenp, boolean_t src_valid,
14758     const void *src, uint_t srclen)
14759 {
14760         if (!src_valid)
14761                 srclen = 0;
14762 
14763         ASSERT(*dstlenp == srclen);
14764         if (src != NULL && srclen != 0)
14765                 bcopy(src, *dstp, srclen);
14766 }
14767 
14768 /*
14769  * Free the storage pointed to by the members of an ip_pkt_t.
14770  */
14771 void
14772 ip_pkt_free(ip_pkt_t *ipp)
14773 {
14774         uint_t  fields = ipp->ipp_fields;
14775 
14776         if (fields & IPPF_HOPOPTS) {
14777                 kmem_free(ipp->ipp_hopopts, ipp->ipp_hopoptslen);
14778                 ipp->ipp_hopopts = NULL;
14779                 ipp->ipp_hopoptslen = 0;
14780         }
14781         if (fields & IPPF_RTHDRDSTOPTS) {
14782                 kmem_free(ipp->ipp_rthdrdstopts, ipp->ipp_rthdrdstoptslen);
14783                 ipp->ipp_rthdrdstopts = NULL;
14784                 ipp->ipp_rthdrdstoptslen = 0;
14785         }
14786         if (fields & IPPF_DSTOPTS) {
14787                 kmem_free(ipp->ipp_dstopts, ipp->ipp_dstoptslen);
14788                 ipp->ipp_dstopts = NULL;
14789                 ipp->ipp_dstoptslen = 0;
14790         }
14791         if (fields & IPPF_RTHDR) {
14792                 kmem_free(ipp->ipp_rthdr, ipp->ipp_rthdrlen);
14793                 ipp->ipp_rthdr = NULL;
14794                 ipp->ipp_rthdrlen = 0;
14795         }
14796         if (fields & IPPF_IPV4_OPTIONS) {
14797                 kmem_free(ipp->ipp_ipv4_options, ipp->ipp_ipv4_options_len);
14798                 ipp->ipp_ipv4_options = NULL;
14799                 ipp->ipp_ipv4_options_len = 0;
14800         }
14801         if (fields & IPPF_LABEL_V4) {
14802                 kmem_free(ipp->ipp_label_v4, ipp->ipp_label_len_v4);
14803                 ipp->ipp_label_v4 = NULL;
14804                 ipp->ipp_label_len_v4 = 0;
14805         }
14806         if (fields & IPPF_LABEL_V6) {
14807                 kmem_free(ipp->ipp_label_v6, ipp->ipp_label_len_v6);
14808                 ipp->ipp_label_v6 = NULL;
14809                 ipp->ipp_label_len_v6 = 0;
14810         }
14811         ipp->ipp_fields &= ~(IPPF_HOPOPTS | IPPF_RTHDRDSTOPTS | IPPF_DSTOPTS |
14812             IPPF_RTHDR | IPPF_IPV4_OPTIONS | IPPF_LABEL_V4 | IPPF_LABEL_V6);
14813 }
14814 
14815 /*
14816  * Copy from src to dst and allocate as needed.
14817  * Returns zero or ENOMEM.
14818  *
14819  * The caller must initialize dst to zero.
14820  */
14821 int
14822 ip_pkt_copy(ip_pkt_t *src, ip_pkt_t *dst, int kmflag)
14823 {
14824         uint_t  fields = src->ipp_fields;
14825 
14826         /* Start with fields that don't require memory allocation */
14827         dst->ipp_fields = fields &
14828             ~(IPPF_HOPOPTS | IPPF_RTHDRDSTOPTS | IPPF_DSTOPTS |
14829             IPPF_RTHDR | IPPF_IPV4_OPTIONS | IPPF_LABEL_V4 | IPPF_LABEL_V6);
14830 
14831         dst->ipp_addr = src->ipp_addr;
14832         dst->ipp_unicast_hops = src->ipp_unicast_hops;
14833         dst->ipp_hoplimit = src->ipp_hoplimit;
14834         dst->ipp_tclass = src->ipp_tclass;
14835         dst->ipp_type_of_service = src->ipp_type_of_service;
14836 
14837         if (!(fields & (IPPF_HOPOPTS | IPPF_RTHDRDSTOPTS | IPPF_DSTOPTS |
14838             IPPF_RTHDR | IPPF_IPV4_OPTIONS | IPPF_LABEL_V4 | IPPF_LABEL_V6)))
14839                 return (0);
14840 
14841         if (fields & IPPF_HOPOPTS) {
14842                 dst->ipp_hopopts = kmem_alloc(src->ipp_hopoptslen, kmflag);
14843                 if (dst->ipp_hopopts == NULL) {
14844                         ip_pkt_free(dst);
14845                         return (ENOMEM);
14846                 }
14847                 dst->ipp_fields |= IPPF_HOPOPTS;
14848                 bcopy(src->ipp_hopopts, dst->ipp_hopopts,
14849                     src->ipp_hopoptslen);
14850                 dst->ipp_hopoptslen = src->ipp_hopoptslen;
14851         }
14852         if (fields & IPPF_RTHDRDSTOPTS) {
14853                 dst->ipp_rthdrdstopts = kmem_alloc(src->ipp_rthdrdstoptslen,
14854                     kmflag);
14855                 if (dst->ipp_rthdrdstopts == NULL) {
14856                         ip_pkt_free(dst);
14857                         return (ENOMEM);
14858                 }
14859                 dst->ipp_fields |= IPPF_RTHDRDSTOPTS;
14860                 bcopy(src->ipp_rthdrdstopts, dst->ipp_rthdrdstopts,
14861                     src->ipp_rthdrdstoptslen);
14862                 dst->ipp_rthdrdstoptslen = src->ipp_rthdrdstoptslen;
14863         }
14864         if (fields & IPPF_DSTOPTS) {
14865                 dst->ipp_dstopts = kmem_alloc(src->ipp_dstoptslen, kmflag);
14866                 if (dst->ipp_dstopts == NULL) {
14867                         ip_pkt_free(dst);
14868                         return (ENOMEM);
14869                 }
14870                 dst->ipp_fields |= IPPF_DSTOPTS;
14871                 bcopy(src->ipp_dstopts, dst->ipp_dstopts,
14872                     src->ipp_dstoptslen);
14873                 dst->ipp_dstoptslen = src->ipp_dstoptslen;
14874         }
14875         if (fields & IPPF_RTHDR) {
14876                 dst->ipp_rthdr = kmem_alloc(src->ipp_rthdrlen, kmflag);
14877                 if (dst->ipp_rthdr == NULL) {
14878                         ip_pkt_free(dst);
14879                         return (ENOMEM);
14880                 }
14881                 dst->ipp_fields |= IPPF_RTHDR;
14882                 bcopy(src->ipp_rthdr, dst->ipp_rthdr,
14883                     src->ipp_rthdrlen);
14884                 dst->ipp_rthdrlen = src->ipp_rthdrlen;
14885         }
14886         if (fields & IPPF_IPV4_OPTIONS) {
14887                 dst->ipp_ipv4_options = kmem_alloc(src->ipp_ipv4_options_len,
14888                     kmflag);
14889                 if (dst->ipp_ipv4_options == NULL) {
14890                         ip_pkt_free(dst);
14891                         return (ENOMEM);
14892                 }
14893                 dst->ipp_fields |= IPPF_IPV4_OPTIONS;
14894                 bcopy(src->ipp_ipv4_options, dst->ipp_ipv4_options,
14895                     src->ipp_ipv4_options_len);
14896                 dst->ipp_ipv4_options_len = src->ipp_ipv4_options_len;
14897         }
14898         if (fields & IPPF_LABEL_V4) {
14899                 dst->ipp_label_v4 = kmem_alloc(src->ipp_label_len_v4, kmflag);
14900                 if (dst->ipp_label_v4 == NULL) {
14901                         ip_pkt_free(dst);
14902                         return (ENOMEM);
14903                 }
14904                 dst->ipp_fields |= IPPF_LABEL_V4;
14905                 bcopy(src->ipp_label_v4, dst->ipp_label_v4,
14906                     src->ipp_label_len_v4);
14907                 dst->ipp_label_len_v4 = src->ipp_label_len_v4;
14908         }
14909         if (fields & IPPF_LABEL_V6) {
14910                 dst->ipp_label_v6 = kmem_alloc(src->ipp_label_len_v6, kmflag);
14911                 if (dst->ipp_label_v6 == NULL) {
14912                         ip_pkt_free(dst);
14913                         return (ENOMEM);
14914                 }
14915                 dst->ipp_fields |= IPPF_LABEL_V6;
14916                 bcopy(src->ipp_label_v6, dst->ipp_label_v6,
14917                     src->ipp_label_len_v6);
14918                 dst->ipp_label_len_v6 = src->ipp_label_len_v6;
14919         }
14920         if (fields & IPPF_FRAGHDR) {
14921                 dst->ipp_fraghdr = kmem_alloc(src->ipp_fraghdrlen, kmflag);
14922                 if (dst->ipp_fraghdr == NULL) {
14923                         ip_pkt_free(dst);
14924                         return (ENOMEM);
14925                 }
14926                 dst->ipp_fields |= IPPF_FRAGHDR;
14927                 bcopy(src->ipp_fraghdr, dst->ipp_fraghdr,
14928                     src->ipp_fraghdrlen);
14929                 dst->ipp_fraghdrlen = src->ipp_fraghdrlen;
14930         }
14931         return (0);
14932 }
14933 
14934 /*
14935  * Returns INADDR_ANY if no source route
14936  */
14937 ipaddr_t
14938 ip_pkt_source_route_v4(const ip_pkt_t *ipp)
14939 {
14940         ipaddr_t        nexthop = INADDR_ANY;
14941         ipoptp_t        opts;
14942         uchar_t         *opt;
14943         uint8_t         optval;
14944         uint8_t         optlen;
14945         uint32_t        totallen;
14946 
14947         if (!(ipp->ipp_fields & IPPF_IPV4_OPTIONS))
14948                 return (INADDR_ANY);
14949 
14950         totallen = ipp->ipp_ipv4_options_len;
14951         if (totallen & 0x3)
14952                 return (INADDR_ANY);
14953 
14954         for (optval = ipoptp_first2(&opts, totallen, ipp->ipp_ipv4_options);
14955             optval != IPOPT_EOL;
14956             optval = ipoptp_next(&opts)) {
14957                 opt = opts.ipoptp_cur;
14958                 switch (optval) {
14959                         uint8_t off;
14960                 case IPOPT_SSRR:
14961                 case IPOPT_LSRR:
14962                         if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
14963                                 break;
14964                         }
14965                         optlen = opts.ipoptp_len;
14966                         off = opt[IPOPT_OFFSET];
14967                         off--;
14968                         if (optlen < IP_ADDR_LEN ||
14969                             off > optlen - IP_ADDR_LEN) {
14970                                 /* End of source route */
14971                                 break;
14972                         }
14973                         bcopy((char *)opt + off, &nexthop, IP_ADDR_LEN);
14974                         if (nexthop == htonl(INADDR_LOOPBACK)) {
14975                                 /* Ignore */
14976                                 nexthop = INADDR_ANY;
14977                                 break;
14978                         }
14979                         break;
14980                 }
14981         }
14982         return (nexthop);
14983 }
14984 
14985 /*
14986  * Reverse a source route.
14987  */
14988 void
14989 ip_pkt_source_route_reverse_v4(ip_pkt_t *ipp)
14990 {
14991         ipaddr_t        tmp;
14992         ipoptp_t        opts;
14993         uchar_t         *opt;
14994         uint8_t         optval;
14995         uint32_t        totallen;
14996 
14997         if (!(ipp->ipp_fields & IPPF_IPV4_OPTIONS))
14998                 return;
14999 
15000         totallen = ipp->ipp_ipv4_options_len;
15001         if (totallen & 0x3)
15002                 return;
15003 
15004         for (optval = ipoptp_first2(&opts, totallen, ipp->ipp_ipv4_options);
15005             optval != IPOPT_EOL;
15006             optval = ipoptp_next(&opts)) {
15007                 uint8_t off1, off2;
15008 
15009                 opt = opts.ipoptp_cur;
15010                 switch (optval) {
15011                 case IPOPT_SSRR:
15012                 case IPOPT_LSRR:
15013                         if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
15014                                 break;
15015                         }
15016                         off1 = IPOPT_MINOFF_SR - 1;
15017                         off2 = opt[IPOPT_OFFSET] - IP_ADDR_LEN - 1;
15018                         while (off2 > off1) {
15019                                 bcopy(opt + off2, &tmp, IP_ADDR_LEN);
15020                                 bcopy(opt + off1, opt + off2, IP_ADDR_LEN);
15021                                 bcopy(&tmp, opt + off2, IP_ADDR_LEN);
15022                                 off2 -= IP_ADDR_LEN;
15023                                 off1 += IP_ADDR_LEN;
15024                         }
15025                         opt[IPOPT_OFFSET] = IPOPT_MINOFF_SR;
15026                         break;
15027                 }
15028         }
15029 }
15030 
15031 /*
15032  * Returns NULL if no routing header
15033  */
15034 in6_addr_t *
15035 ip_pkt_source_route_v6(const ip_pkt_t *ipp)
15036 {
15037         in6_addr_t      *nexthop = NULL;
15038         ip6_rthdr0_t    *rthdr;
15039 
15040         if (!(ipp->ipp_fields & IPPF_RTHDR))
15041                 return (NULL);
15042 
15043         rthdr = (ip6_rthdr0_t *)ipp->ipp_rthdr;
15044         if (rthdr->ip6r0_segleft == 0)
15045                 return (NULL);
15046 
15047         nexthop = (in6_addr_t *)((char *)rthdr + sizeof (*rthdr));
15048         return (nexthop);
15049 }
15050 
15051 zoneid_t
15052 ip_get_zoneid_v4(ipaddr_t addr, mblk_t *mp, ip_recv_attr_t *ira,
15053     zoneid_t lookup_zoneid)
15054 {
15055         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
15056         ire_t           *ire;
15057         int             ire_flags = MATCH_IRE_TYPE;
15058         zoneid_t        zoneid = ALL_ZONES;
15059 
15060         if (is_system_labeled() && !tsol_can_accept_raw(mp, ira, B_FALSE))
15061                 return (ALL_ZONES);
15062 
15063         if (lookup_zoneid != ALL_ZONES)
15064                 ire_flags |= MATCH_IRE_ZONEONLY;
15065         ire = ire_ftable_lookup_v4(addr, 0, 0, IRE_LOCAL | IRE_LOOPBACK,
15066             NULL, lookup_zoneid, NULL, ire_flags, 0, ipst, NULL);
15067         if (ire != NULL) {
15068                 zoneid = IP_REAL_ZONEID(ire->ire_zoneid, ipst);
15069                 ire_refrele(ire);
15070         }
15071         return (zoneid);
15072 }
15073 
15074 zoneid_t
15075 ip_get_zoneid_v6(in6_addr_t *addr, mblk_t *mp, const ill_t *ill,
15076     ip_recv_attr_t *ira, zoneid_t lookup_zoneid)
15077 {
15078         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
15079         ire_t           *ire;
15080         int             ire_flags = MATCH_IRE_TYPE;
15081         zoneid_t        zoneid = ALL_ZONES;
15082 
15083         if (is_system_labeled() && !tsol_can_accept_raw(mp, ira, B_FALSE))
15084                 return (ALL_ZONES);
15085 
15086         if (IN6_IS_ADDR_LINKLOCAL(addr))
15087                 ire_flags |= MATCH_IRE_ILL;
15088 
15089         if (lookup_zoneid != ALL_ZONES)
15090                 ire_flags |= MATCH_IRE_ZONEONLY;
15091         ire = ire_ftable_lookup_v6(addr, NULL, NULL, IRE_LOCAL | IRE_LOOPBACK,
15092             ill, lookup_zoneid, NULL, ire_flags, 0, ipst, NULL);
15093         if (ire != NULL) {
15094                 zoneid = IP_REAL_ZONEID(ire->ire_zoneid, ipst);
15095                 ire_refrele(ire);
15096         }
15097         return (zoneid);
15098 }
15099 
15100 /*
15101  * IP obserability hook support functions.
15102  */
15103 static void
15104 ipobs_init(ip_stack_t *ipst)
15105 {
15106         netid_t id;
15107 
15108         id = net_getnetidbynetstackid(ipst->ips_netstack->netstack_stackid);
15109 
15110         ipst->ips_ip4_observe_pr = net_protocol_lookup(id, NHF_INET);
15111         VERIFY(ipst->ips_ip4_observe_pr != NULL);
15112 
15113         ipst->ips_ip6_observe_pr = net_protocol_lookup(id, NHF_INET6);
15114         VERIFY(ipst->ips_ip6_observe_pr != NULL);
15115 }
15116 
15117 static void
15118 ipobs_fini(ip_stack_t *ipst)
15119 {
15120 
15121         VERIFY(net_protocol_release(ipst->ips_ip4_observe_pr) == 0);
15122         VERIFY(net_protocol_release(ipst->ips_ip6_observe_pr) == 0);
15123 }
15124 
15125 /*
15126  * hook_pkt_observe_t is composed in network byte order so that the
15127  * entire mblk_t chain handed into hook_run can be used as-is.
15128  * The caveat is that use of the fields, such as the zone fields,
15129  * requires conversion into host byte order first.
15130  */
15131 void
15132 ipobs_hook(mblk_t *mp, int htype, zoneid_t zsrc, zoneid_t zdst,
15133     const ill_t *ill, ip_stack_t *ipst)
15134 {
15135         hook_pkt_observe_t *hdr;
15136         uint64_t grifindex;
15137         mblk_t *imp;
15138 
15139         imp = allocb(sizeof (*hdr), BPRI_HI);
15140         if (imp == NULL)
15141                 return;
15142 
15143         hdr = (hook_pkt_observe_t *)imp->b_rptr;
15144         /*
15145          * b_wptr is set to make the apparent size of the data in the mblk_t
15146          * to exclude the pointers at the end of hook_pkt_observer_t.
15147          */
15148         imp->b_wptr = imp->b_rptr + sizeof (dl_ipnetinfo_t);
15149         imp->b_cont = mp;
15150 
15151         ASSERT(DB_TYPE(mp) == M_DATA);
15152 
15153         if (IS_UNDER_IPMP(ill))
15154                 grifindex = ipmp_ill_get_ipmp_ifindex(ill);
15155         else
15156                 grifindex = 0;
15157 
15158         hdr->hpo_version = 1;
15159         hdr->hpo_htype = htons(htype);
15160         hdr->hpo_pktlen = htonl((ulong_t)msgdsize(mp));
15161         hdr->hpo_ifindex = htonl(ill->ill_phyint->phyint_ifindex);
15162         hdr->hpo_grifindex = htonl(grifindex);
15163         hdr->hpo_zsrc = htonl(zsrc);
15164         hdr->hpo_zdst = htonl(zdst);
15165         hdr->hpo_pkt = imp;
15166         hdr->hpo_ctx = ipst->ips_netstack;
15167 
15168         if (ill->ill_isv6) {
15169                 hdr->hpo_family = AF_INET6;
15170                 (void) hook_run(ipst->ips_ipv6_net_data->netd_hooks,
15171                     ipst->ips_ipv6observing, (hook_data_t)hdr);
15172         } else {
15173                 hdr->hpo_family = AF_INET;
15174                 (void) hook_run(ipst->ips_ipv4_net_data->netd_hooks,
15175                     ipst->ips_ipv4observing, (hook_data_t)hdr);
15176         }
15177 
15178         imp->b_cont = NULL;
15179         freemsg(imp);
15180 }
15181 
15182 /*
15183  * Utility routine that checks if `v4srcp' is a valid address on underlying
15184  * interface `ill'.  If `ipifp' is non-NULL, it's set to a held ipif
15185  * associated with `v4srcp' on success.  NOTE: if this is not called from
15186  * inside the IPSQ (ill_g_lock is not held), `ill' may be removed from the
15187  * group during or after this lookup.
15188  */
15189 boolean_t
15190 ipif_lookup_testaddr_v4(ill_t *ill, const in_addr_t *v4srcp, ipif_t **ipifp)
15191 {
15192         ipif_t *ipif;
15193 
15194         ipif = ipif_lookup_addr_exact(*v4srcp, ill, ill->ill_ipst);
15195         if (ipif != NULL) {
15196                 if (ipifp != NULL)
15197                         *ipifp = ipif;
15198                 else
15199                         ipif_refrele(ipif);
15200                 return (B_TRUE);
15201         }
15202 
15203         ip1dbg(("ipif_lookup_testaddr_v4: cannot find ipif for src %x\n",
15204             *v4srcp));
15205         return (B_FALSE);
15206 }
15207 
15208 /*
15209  * Transport protocol call back function for CPU state change.
15210  */
15211 /* ARGSUSED */
15212 static int
15213 ip_tp_cpu_update(cpu_setup_t what, int id, void *arg)
15214 {
15215         processorid_t cpu_seqid;
15216         netstack_handle_t nh;
15217         netstack_t *ns;
15218 
15219         ASSERT(MUTEX_HELD(&cpu_lock));
15220 
15221         switch (what) {
15222         case CPU_CONFIG:
15223         case CPU_ON:
15224         case CPU_INIT:
15225         case CPU_CPUPART_IN:
15226                 cpu_seqid = cpu[id]->cpu_seqid;
15227                 netstack_next_init(&nh);
15228                 while ((ns = netstack_next(&nh)) != NULL) {
15229                         tcp_stack_cpu_add(ns->netstack_tcp, cpu_seqid);
15230                         sctp_stack_cpu_add(ns->netstack_sctp, cpu_seqid);
15231                         udp_stack_cpu_add(ns->netstack_udp, cpu_seqid);
15232                         netstack_rele(ns);
15233                 }
15234                 netstack_next_fini(&nh);
15235                 break;
15236         case CPU_UNCONFIG:
15237         case CPU_OFF:
15238         case CPU_CPUPART_OUT:
15239                 /*
15240                  * Nothing to do.  We don't remove the per CPU stats from
15241                  * the IP stack even when the CPU goes offline.
15242                  */
15243                 break;
15244         default:
15245                 break;
15246         }
15247         return (0);
15248 }