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 2011 Nexenta Systems, Inc. All rights reserved.
26 */
27
28 #include <sys/types.h>
29 #include <sys/stream.h>
30 #include <sys/dlpi.h>
31 #include <sys/stropts.h>
32 #include <sys/sysmacros.h>
33 #include <sys/strsubr.h>
34 #include <sys/strlog.h>
35 #include <sys/strsun.h>
36 #include <sys/zone.h>
37 #define _SUN_TPI_VERSION 2
38 #include <sys/tihdr.h>
39 #include <sys/xti_inet.h>
40 #include <sys/ddi.h>
41 #include <sys/suntpi.h>
42 #include <sys/cmn_err.h>
43 #include <sys/debug.h>
44 #include <sys/kobj.h>
45 #include <sys/modctl.h>
46 #include <sys/atomic.h>
47 #include <sys/policy.h>
48 #include <sys/priv.h>
49 #include <sys/taskq.h>
50
51 #include <sys/systm.h>
52 #include <sys/param.h>
53 #include <sys/kmem.h>
54 #include <sys/sdt.h>
55 #include <sys/socket.h>
56 #include <sys/vtrace.h>
57 #include <sys/isa_defs.h>
58 #include <sys/mac.h>
59 #include <net/if.h>
60 #include <net/if_arp.h>
61 #include <net/route.h>
62 #include <sys/sockio.h>
63 #include <netinet/in.h>
64 #include <net/if_dl.h>
65
66 #include <inet/common.h>
67 #include <inet/mi.h>
68 #include <inet/mib2.h>
69 #include <inet/nd.h>
70 #include <inet/arp.h>
71 #include <inet/snmpcom.h>
72 #include <inet/optcom.h>
73 #include <inet/kstatcom.h>
74
75 #include <netinet/igmp_var.h>
76 #include <netinet/ip6.h>
77 #include <netinet/icmp6.h>
78 #include <netinet/sctp.h>
79
80 #include <inet/ip.h>
81 #include <inet/ip_impl.h>
82 #include <inet/ip6.h>
83 #include <inet/ip6_asp.h>
84 #include <inet/tcp.h>
85 #include <inet/tcp_impl.h>
86 #include <inet/ip_multi.h>
87 #include <inet/ip_if.h>
88 #include <inet/ip_ire.h>
89 #include <inet/ip_ftable.h>
90 #include <inet/ip_rts.h>
91 #include <inet/ip_ndp.h>
92 #include <inet/ip_listutils.h>
93 #include <netinet/igmp.h>
94 #include <netinet/ip_mroute.h>
95 #include <inet/ipp_common.h>
96
97 #include <net/pfkeyv2.h>
98 #include <inet/sadb.h>
99 #include <inet/ipsec_impl.h>
100 #include <inet/iptun/iptun_impl.h>
101 #include <inet/ipdrop.h>
102 #include <inet/ip_netinfo.h>
103 #include <inet/ilb_ip.h>
104
105 #include <sys/ethernet.h>
106 #include <net/if_types.h>
107 #include <sys/cpuvar.h>
108
109 #include <ipp/ipp.h>
110 #include <ipp/ipp_impl.h>
111 #include <ipp/ipgpc/ipgpc.h>
112
113 #include <sys/pattr.h>
114 #include <inet/ipclassifier.h>
115 #include <inet/sctp_ip.h>
116 #include <inet/sctp/sctp_impl.h>
117 #include <inet/udp_impl.h>
118 #include <inet/rawip_impl.h>
119 #include <inet/rts_impl.h>
120
121 #include <sys/tsol/label.h>
122 #include <sys/tsol/tnet.h>
123
124 #include <sys/squeue_impl.h>
125 #include <inet/ip_arp.h>
126
127 #include <sys/clock_impl.h> /* For LBOLT_FASTPATH{,64} */
128
129 /*
130 * Values for squeue switch:
131 * IP_SQUEUE_ENTER_NODRAIN: SQ_NODRAIN
132 * IP_SQUEUE_ENTER: SQ_PROCESS
133 * IP_SQUEUE_FILL: SQ_FILL
134 */
135 int ip_squeue_enter = IP_SQUEUE_ENTER; /* Setable in /etc/system */
136
137 int ip_squeue_flag;
138
139 /*
140 * Setable in /etc/system
141 */
142 int ip_poll_normal_ms = 100;
143 int ip_poll_normal_ticks = 0;
144 int ip_modclose_ackwait_ms = 3000;
145
146 /*
147 * It would be nice to have these present only in DEBUG systems, but the
148 * current design of the global symbol checking logic requires them to be
149 * unconditionally present.
150 */
151 uint_t ip_thread_data; /* TSD key for debug support */
152 krwlock_t ip_thread_rwlock;
153 list_t ip_thread_list;
154
155 /*
156 * Structure to represent a linked list of msgblks. Used by ip_snmp_ functions.
157 */
158
159 struct listptr_s {
160 mblk_t *lp_head; /* pointer to the head of the list */
161 mblk_t *lp_tail; /* pointer to the tail of the list */
162 };
163
164 typedef struct listptr_s listptr_t;
165
166 /*
167 * This is used by ip_snmp_get_mib2_ip_route_media and
168 * ip_snmp_get_mib2_ip6_route_media to carry the lists of return data.
169 */
170 typedef struct iproutedata_s {
171 uint_t ird_idx;
172 uint_t ird_flags; /* see below */
173 listptr_t ird_route; /* ipRouteEntryTable */
174 listptr_t ird_netmedia; /* ipNetToMediaEntryTable */
175 listptr_t ird_attrs; /* ipRouteAttributeTable */
176 } iproutedata_t;
177
178 /* Include ire_testhidden and IRE_IF_CLONE routes */
179 #define IRD_REPORT_ALL 0x01
180
181 /*
182 * Cluster specific hooks. These should be NULL when booted as a non-cluster
183 */
184
185 /*
186 * Hook functions to enable cluster networking
187 * On non-clustered systems these vectors must always be NULL.
188 *
189 * Hook function to Check ip specified ip address is a shared ip address
190 * in the cluster
191 *
192 */
193 int (*cl_inet_isclusterwide)(netstackid_t stack_id, uint8_t protocol,
194 sa_family_t addr_family, uint8_t *laddrp, void *args) = NULL;
195
196 /*
197 * Hook function to generate cluster wide ip fragment identifier
198 */
199 uint32_t (*cl_inet_ipident)(netstackid_t stack_id, uint8_t protocol,
200 sa_family_t addr_family, uint8_t *laddrp, uint8_t *faddrp,
201 void *args) = NULL;
202
203 /*
204 * Hook function to generate cluster wide SPI.
205 */
206 void (*cl_inet_getspi)(netstackid_t, uint8_t, uint8_t *, size_t,
207 void *) = NULL;
208
209 /*
210 * Hook function to verify if the SPI is already utlized.
211 */
212
213 int (*cl_inet_checkspi)(netstackid_t, uint8_t, uint32_t, void *) = NULL;
214
215 /*
216 * Hook function to delete the SPI from the cluster wide repository.
217 */
218
219 void (*cl_inet_deletespi)(netstackid_t, uint8_t, uint32_t, void *) = NULL;
220
221 /*
222 * Hook function to inform the cluster when packet received on an IDLE SA
223 */
224
225 void (*cl_inet_idlesa)(netstackid_t, uint8_t, uint32_t, sa_family_t,
226 in6_addr_t, in6_addr_t, void *) = NULL;
227
228 /*
229 * Synchronization notes:
230 *
231 * IP is a fully D_MP STREAMS module/driver. Thus it does not depend on any
232 * MT level protection given by STREAMS. IP uses a combination of its own
233 * internal serialization mechanism and standard Solaris locking techniques.
234 * The internal serialization is per phyint. This is used to serialize
235 * plumbing operations, IPMP operations, most set ioctls, etc.
236 *
237 * Plumbing is a long sequence of operations involving message
238 * exchanges between IP, ARP and device drivers. Many set ioctls are typically
239 * involved in plumbing operations. A natural model is to serialize these
240 * ioctls one per ill. For example plumbing of hme0 and qfe0 can go on in
241 * parallel without any interference. But various set ioctls on hme0 are best
242 * serialized, along with IPMP operations and processing of DLPI control
243 * messages received from drivers on a per phyint basis. This serialization is
244 * provided by the ipsq_t and primitives operating on this. Details can
245 * be found in ip_if.c above the core primitives operating on ipsq_t.
246 *
247 * Lookups of an ipif or ill by a thread return a refheld ipif / ill.
248 * Simiarly lookup of an ire by a thread also returns a refheld ire.
249 * In addition ipif's and ill's referenced by the ire are also indirectly
250 * refheld. Thus no ipif or ill can vanish as long as an ipif is refheld
251 * directly or indirectly. For example an SIOCSLIFADDR ioctl that changes the
252 * address of an ipif has to go through the ipsq_t. This ensures that only
253 * one such exclusive operation proceeds at any time on the ipif. It then
254 * waits for all refcnts
255 * associated with this ipif to come down to zero. The address is changed
256 * only after the ipif has been quiesced. Then the ipif is brought up again.
257 * More details are described above the comment in ip_sioctl_flags.
258 *
259 * Packet processing is based mostly on IREs and are fully multi-threaded
260 * using standard Solaris MT techniques.
261 *
262 * There are explicit locks in IP to handle:
263 * - The ip_g_head list maintained by mi_open_link() and friends.
264 *
265 * - The reassembly data structures (one lock per hash bucket)
266 *
267 * - conn_lock is meant to protect conn_t fields. The fields actually
268 * protected by conn_lock are documented in the conn_t definition.
269 *
270 * - ire_lock to protect some of the fields of the ire, IRE tables
271 * (one lock per hash bucket). Refer to ip_ire.c for details.
272 *
273 * - ndp_g_lock and ncec_lock for protecting NCEs.
274 *
275 * - ill_lock protects fields of the ill and ipif. Details in ip.h
276 *
277 * - ill_g_lock: This is a global reader/writer lock. Protects the following
278 * * The AVL tree based global multi list of all ills.
279 * * The linked list of all ipifs of an ill
280 * * The <ipsq-xop> mapping
281 * * <ill-phyint> association
282 * Insertion/deletion of an ill in the system, insertion/deletion of an ipif
283 * into an ill, changing the <ipsq-xop> mapping of an ill, changing the
284 * <ill-phyint> assoc of an ill will all have to hold the ill_g_lock as
285 * writer for the actual duration of the insertion/deletion/change.
286 *
287 * - ill_lock: This is a per ill mutex.
288 * It protects some members of the ill_t struct; see ip.h for details.
289 * It also protects the <ill-phyint> assoc.
290 * It also protects the list of ipifs hanging off the ill.
291 *
292 * - ipsq_lock: This is a per ipsq_t mutex lock.
293 * This protects some members of the ipsq_t struct; see ip.h for details.
294 * It also protects the <ipsq-ipxop> mapping
295 *
296 * - ipx_lock: This is a per ipxop_t mutex lock.
297 * This protects some members of the ipxop_t struct; see ip.h for details.
298 *
299 * - phyint_lock: This is a per phyint mutex lock. Protects just the
300 * phyint_flags
301 *
302 * - ip_addr_avail_lock: This is used to ensure the uniqueness of IP addresses.
303 * This lock is held in ipif_up_done and the ipif is marked IPIF_UP and the
304 * uniqueness check also done atomically.
305 *
306 * - ill_g_usesrc_lock: This readers/writer lock protects the usesrc
307 * group list linked by ill_usesrc_grp_next. It also protects the
308 * ill_usesrc_ifindex field. It is taken as a writer when a member of the
309 * group is being added or deleted. This lock is taken as a reader when
310 * walking the list/group(eg: to get the number of members in a usesrc group).
311 * Note, it is only necessary to take this lock if the ill_usesrc_grp_next
312 * field is changing state i.e from NULL to non-NULL or vice-versa. For
313 * example, it is not necessary to take this lock in the initial portion
314 * of ip_sioctl_slifusesrc or at all in ip_sioctl_flags since these
315 * operations are executed exclusively and that ensures that the "usesrc
316 * group state" cannot change. The "usesrc group state" change can happen
317 * only in the latter part of ip_sioctl_slifusesrc and in ill_delete.
318 *
319 * Changing <ill-phyint>, <ipsq-xop> assocications:
320 *
321 * To change the <ill-phyint> association, the ill_g_lock must be held
322 * as writer, and the ill_locks of both the v4 and v6 instance of the ill
323 * must be held.
324 *
325 * To change the <ipsq-xop> association, the ill_g_lock must be held as
326 * writer, the ipsq_lock must be held, and one must be writer on the ipsq.
327 * This is only done when ills are added or removed from IPMP groups.
328 *
329 * To add or delete an ipif from the list of ipifs hanging off the ill,
330 * ill_g_lock (writer) and ill_lock must be held and the thread must be
331 * a writer on the associated ipsq.
332 *
333 * To add or delete an ill to the system, the ill_g_lock must be held as
334 * writer and the thread must be a writer on the associated ipsq.
335 *
336 * To add or delete an ilm to an ill, the ill_lock must be held and the thread
337 * must be a writer on the associated ipsq.
338 *
339 * Lock hierarchy
340 *
341 * Some lock hierarchy scenarios are listed below.
342 *
343 * ill_g_lock -> conn_lock -> ill_lock -> ipsq_lock -> ipx_lock
344 * ill_g_lock -> ill_lock(s) -> phyint_lock
345 * ill_g_lock -> ndp_g_lock -> ill_lock -> ncec_lock
346 * ill_g_lock -> ip_addr_avail_lock
347 * conn_lock -> irb_lock -> ill_lock -> ire_lock
348 * ill_g_lock -> ip_g_nd_lock
349 * ill_g_lock -> ips_ipmp_lock -> ill_lock -> nce_lock
350 * ill_g_lock -> ndp_g_lock -> ill_lock -> ncec_lock -> nce_lock
351 * arl_lock -> ill_lock
352 * ips_ire_dep_lock -> irb_lock
353 *
354 * When more than 1 ill lock is needed to be held, all ill lock addresses
355 * are sorted on address and locked starting from highest addressed lock
356 * downward.
357 *
358 * Multicast scenarios
359 * ips_ill_g_lock -> ill_mcast_lock
360 * conn_ilg_lock -> ips_ill_g_lock -> ill_lock
361 * ill_mcast_serializer -> ill_mcast_lock -> ips_ipmp_lock -> ill_lock
362 * ill_mcast_serializer -> ill_mcast_lock -> connf_lock -> conn_lock
363 * ill_mcast_serializer -> ill_mcast_lock -> conn_ilg_lock
364 * ill_mcast_serializer -> ill_mcast_lock -> ips_igmp_timer_lock
365 *
366 * IPsec scenarios
367 *
368 * ipsa_lock -> ill_g_lock -> ill_lock
369 * ill_g_usesrc_lock -> ill_g_lock -> ill_lock
370 *
371 * Trusted Solaris scenarios
372 *
373 * igsa_lock -> gcgrp_rwlock -> gcgrp_lock
374 * igsa_lock -> gcdb_lock
375 * gcgrp_rwlock -> ire_lock
376 * gcgrp_rwlock -> gcdb_lock
377 *
378 * squeue(sq_lock), flow related (ft_lock, fe_lock) locking
379 *
380 * cpu_lock --> ill_lock --> sqset_lock --> sq_lock
381 * sq_lock -> conn_lock -> QLOCK(q)
382 * ill_lock -> ft_lock -> fe_lock
383 *
384 * Routing/forwarding table locking notes:
385 *
386 * Lock acquisition order: Radix tree lock, irb_lock.
387 * Requirements:
388 * i. Walker must not hold any locks during the walker callback.
389 * ii Walker must not see a truncated tree during the walk because of any node
390 * deletion.
391 * iii Existing code assumes ire_bucket is valid if it is non-null and is used
392 * in many places in the code to walk the irb list. Thus even if all the
393 * ires in a bucket have been deleted, we still can't free the radix node
394 * until the ires have actually been inactive'd (freed).
395 *
396 * Tree traversal - Need to hold the global tree lock in read mode.
397 * Before dropping the global tree lock, need to either increment the ire_refcnt
398 * to ensure that the radix node can't be deleted.
399 *
400 * Tree add - Need to hold the global tree lock in write mode to add a
401 * radix node. To prevent the node from being deleted, increment the
402 * irb_refcnt, after the node is added to the tree. The ire itself is
403 * added later while holding the irb_lock, but not the tree lock.
404 *
405 * Tree delete - Need to hold the global tree lock and irb_lock in write mode.
406 * All associated ires must be inactive (i.e. freed), and irb_refcnt
407 * must be zero.
408 *
409 * Walker - Increment irb_refcnt before calling the walker callback. Hold the
410 * global tree lock (read mode) for traversal.
411 *
412 * IRE dependencies - In some cases we hold ips_ire_dep_lock across ire_refrele
413 * hence we will acquire irb_lock while holding ips_ire_dep_lock.
414 *
415 * IPsec notes :
416 *
417 * IP interacts with the IPsec code (AH/ESP) by storing IPsec attributes
418 * in the ip_xmit_attr_t ip_recv_attr_t. For outbound datagrams, the
419 * ip_xmit_attr_t has the
420 * information used by the IPsec code for applying the right level of
421 * protection. The information initialized by IP in the ip_xmit_attr_t
422 * is determined by the per-socket policy or global policy in the system.
423 * For inbound datagrams, the ip_recv_attr_t
424 * starts out with nothing in it. It gets filled
425 * with the right information if it goes through the AH/ESP code, which
426 * happens if the incoming packet is secure. The information initialized
427 * by AH/ESP, is later used by IP (during fanouts to ULP) to see whether
428 * the policy requirements needed by per-socket policy or global policy
429 * is met or not.
430 *
431 * For fully connected sockets i.e dst, src [addr, port] is known,
432 * conn_policy_cached is set indicating that policy has been cached.
433 * conn_in_enforce_policy may or may not be set depending on whether
434 * there is a global policy match or per-socket policy match.
435 * Policy inheriting happpens in ip_policy_set once the destination is known.
436 * Once the right policy is set on the conn_t, policy cannot change for
437 * this socket. This makes life simpler for TCP (UDP ?) where
438 * re-transmissions go out with the same policy. For symmetry, policy
439 * is cached for fully connected UDP sockets also. Thus if policy is cached,
440 * it also implies that policy is latched i.e policy cannot change
441 * on these sockets. As we have the right policy on the conn, we don't
442 * have to lookup global policy for every outbound and inbound datagram
443 * and thus serving as an optimization. Note that a global policy change
444 * does not affect fully connected sockets if they have policy. If fully
445 * connected sockets did not have any policy associated with it, global
446 * policy change may affect them.
447 *
448 * IP Flow control notes:
449 * ---------------------
450 * Non-TCP streams are flow controlled by IP. The way this is accomplished
451 * differs when ILL_CAPAB_DLD_DIRECT is enabled for that IP instance. When
452 * ILL_DIRECT_CAPABLE(ill) is TRUE, IP can do direct function calls into
453 * GLDv3. Otherwise packets are sent down to lower layers using STREAMS
454 * functions.
455 *
456 * Per Tx ring udp flow control:
457 * This is applicable only when ILL_CAPAB_DLD_DIRECT capability is set in
458 * the ill (i.e. ILL_DIRECT_CAPABLE(ill) is true).
459 *
460 * The underlying link can expose multiple Tx rings to the GLDv3 mac layer.
461 * To achieve best performance, outgoing traffic need to be fanned out among
462 * these Tx ring. mac_tx() is called (via str_mdata_fastpath_put()) to send
463 * traffic out of the NIC and it takes a fanout hint. UDP connections pass
464 * the address of connp as fanout hint to mac_tx(). Under flow controlled
465 * condition, mac_tx() returns a non-NULL cookie (ip_mac_tx_cookie_t). This
466 * cookie points to a specific Tx ring that is blocked. The cookie is used to
467 * hash into an idl_tx_list[] entry in idl_tx_list[] array. Each idl_tx_list_t
468 * point to drain_lists (idl_t's). These drain list will store the blocked UDP
469 * connp's. The drain list is not a single list but a configurable number of
470 * lists.
471 *
472 * The diagram below shows idl_tx_list_t's and their drain_lists. ip_stack_t
473 * has an array of idl_tx_list_t. The size of the array is TX_FANOUT_SIZE
474 * which is equal to 128. This array in turn contains a pointer to idl_t[],
475 * the ip drain list. The idl_t[] array size is MIN(max_ncpus, 8). The drain
476 * list will point to the list of connp's that are flow controlled.
477 *
478 * --------------- ------- ------- -------
479 * |->|drain_list[0]|-->|connp|-->|connp|-->|connp|-->
480 * | --------------- ------- ------- -------
481 * | --------------- ------- ------- -------
482 * |->|drain_list[1]|-->|connp|-->|connp|-->|connp|-->
483 * ---------------- | --------------- ------- ------- -------
484 * |idl_tx_list[0]|->| --------------- ------- ------- -------
485 * ---------------- |->|drain_list[2]|-->|connp|-->|connp|-->|connp|-->
486 * | --------------- ------- ------- -------
487 * . . . . .
488 * | --------------- ------- ------- -------
489 * |->|drain_list[n]|-->|connp|-->|connp|-->|connp|-->
490 * --------------- ------- ------- -------
491 * --------------- ------- ------- -------
492 * |->|drain_list[0]|-->|connp|-->|connp|-->|connp|-->
493 * | --------------- ------- ------- -------
494 * | --------------- ------- ------- -------
495 * ---------------- |->|drain_list[1]|-->|connp|-->|connp|-->|connp|-->
496 * |idl_tx_list[1]|->| --------------- ------- ------- -------
497 * ---------------- | . . . .
498 * | --------------- ------- ------- -------
499 * |->|drain_list[n]|-->|connp|-->|connp|-->|connp|-->
500 * --------------- ------- ------- -------
501 * .....
502 * ----------------
503 * |idl_tx_list[n]|-> ...
504 * ----------------
505 *
506 * When mac_tx() returns a cookie, the cookie is hashed into an index into
507 * ips_idl_tx_list[], and conn_drain_insert() is called with the idl_tx_list
508 * to insert the conn onto. conn_drain_insert() asserts flow control for the
509 * sockets via su_txq_full() (non-STREAMS) or QFULL on conn_wq (STREAMS).
510 * Further, conn_blocked is set to indicate that the conn is blocked.
511 *
512 * GLDv3 calls ill_flow_enable() when flow control is relieved. The cookie
513 * passed in the call to ill_flow_enable() identifies the blocked Tx ring and
514 * is again hashed to locate the appropriate idl_tx_list, which is then
515 * drained via conn_walk_drain(). conn_walk_drain() goes through each conn in
516 * the drain list and calls conn_drain_remove() to clear flow control (via
517 * calling su_txq_full() or clearing QFULL), and remove the conn from the
518 * drain list.
519 *
520 * Note that the drain list is not a single list but a (configurable) array of
521 * lists (8 elements by default). Synchronization between drain insertion and
522 * flow control wakeup is handled by using idl_txl->txl_lock, and only
523 * conn_drain_insert() and conn_drain_remove() manipulate the drain list.
524 *
525 * Flow control via STREAMS is used when ILL_DIRECT_CAPABLE() returns FALSE.
526 * On the send side, if the packet cannot be sent down to the driver by IP
527 * (canput() fails), ip_xmit() drops the packet and returns EWOULDBLOCK to the
528 * caller, who may then invoke ixa_check_drain_insert() to insert the conn on
529 * the 0'th drain list. When ip_wsrv() runs on the ill_wq because flow
530 * control has been relieved, the blocked conns in the 0'th drain list are
531 * drained as in the non-STREAMS case.
532 *
533 * In both the STREAMS and non-STREAMS cases, the sockfs upcall to set QFULL
534 * is done when the conn is inserted into the drain list (conn_drain_insert())
535 * and cleared when the conn is removed from the it (conn_drain_remove()).
536 *
537 * IPQOS notes:
538 *
539 * IPQoS Policies are applied to packets using IPPF (IP Policy framework)
540 * and IPQoS modules. IPPF includes hooks in IP at different control points
541 * (callout positions) which direct packets to IPQoS modules for policy
542 * processing. Policies, if present, are global.
543 *
544 * The callout positions are located in the following paths:
545 * o local_in (packets destined for this host)
546 * o local_out (packets orginating from this host )
547 * o fwd_in (packets forwarded by this m/c - inbound)
548 * o fwd_out (packets forwarded by this m/c - outbound)
549 * Hooks at these callout points can be enabled/disabled using the ndd variable
550 * ip_policy_mask (a bit mask with the 4 LSB indicating the callout positions).
551 * By default all the callout positions are enabled.
552 *
553 * Outbound (local_out)
554 * Hooks are placed in ire_send_wire_v4 and ire_send_wire_v6.
555 *
556 * Inbound (local_in)
557 * Hooks are placed in ip_fanout_v4 and ip_fanout_v6.
558 *
559 * Forwarding (in and out)
560 * Hooks are placed in ire_recv_forward_v4/v6.
561 *
562 * IP Policy Framework processing (IPPF processing)
563 * Policy processing for a packet is initiated by ip_process, which ascertains
564 * that the classifier (ipgpc) is loaded and configured, failing which the
565 * packet resumes normal processing in IP. If the clasifier is present, the
566 * packet is acted upon by one or more IPQoS modules (action instances), per
567 * filters configured in ipgpc and resumes normal IP processing thereafter.
568 * An action instance can drop a packet in course of its processing.
569 *
570 * Zones notes:
571 *
572 * The partitioning rules for networking are as follows:
573 * 1) Packets coming from a zone must have a source address belonging to that
574 * zone.
575 * 2) Packets coming from a zone can only be sent on a physical interface on
576 * which the zone has an IP address.
577 * 3) Between two zones on the same machine, packet delivery is only allowed if
578 * there's a matching route for the destination and zone in the forwarding
579 * table.
580 * 4) The TCP and UDP port spaces are per-zone; that is, two processes in
581 * different zones can bind to the same port with the wildcard address
582 * (INADDR_ANY).
583 *
584 * The granularity of interface partitioning is at the logical interface level.
585 * Therefore, every zone has its own IP addresses, and incoming packets can be
586 * attributed to a zone unambiguously. A logical interface is placed into a zone
587 * using the SIOCSLIFZONE ioctl; this sets the ipif_zoneid field in the ipif_t
588 * structure. Rule (1) is implemented by modifying the source address selection
589 * algorithm so that the list of eligible addresses is filtered based on the
590 * sending process zone.
591 *
592 * The Internet Routing Entries (IREs) are either exclusive to a zone or shared
593 * across all zones, depending on their type. Here is the break-up:
594 *
595 * IRE type Shared/exclusive
596 * -------- ----------------
597 * IRE_BROADCAST Exclusive
598 * IRE_DEFAULT (default routes) Shared (*)
599 * IRE_LOCAL Exclusive (x)
600 * IRE_LOOPBACK Exclusive
601 * IRE_PREFIX (net routes) Shared (*)
602 * IRE_IF_NORESOLVER (interface routes) Exclusive
603 * IRE_IF_RESOLVER (interface routes) Exclusive
604 * IRE_IF_CLONE (interface routes) Exclusive
605 * IRE_HOST (host routes) Shared (*)
606 *
607 * (*) A zone can only use a default or off-subnet route if the gateway is
608 * directly reachable from the zone, that is, if the gateway's address matches
609 * one of the zone's logical interfaces.
610 *
611 * (x) IRE_LOCAL are handled a bit differently.
612 * When ip_restrict_interzone_loopback is set (the default),
613 * ire_route_recursive restricts loopback using an IRE_LOCAL
614 * between zone to the case when L2 would have conceptually looped the packet
615 * back, i.e. the loopback which is required since neither Ethernet drivers
616 * nor Ethernet hardware loops them back. This is the case when the normal
617 * routes (ignoring IREs with different zoneids) would send out the packet on
618 * the same ill as the ill with which is IRE_LOCAL is associated.
619 *
620 * Multiple zones can share a common broadcast address; typically all zones
621 * share the 255.255.255.255 address. Incoming as well as locally originated
622 * broadcast packets must be dispatched to all the zones on the broadcast
623 * network. For directed broadcasts (e.g. 10.16.72.255) this is not trivial
624 * since some zones may not be on the 10.16.72/24 network. To handle this, each
625 * zone has its own set of IRE_BROADCAST entries; then, broadcast packets are
626 * sent to every zone that has an IRE_BROADCAST entry for the destination
627 * address on the input ill, see ip_input_broadcast().
628 *
629 * Applications in different zones can join the same multicast group address.
630 * The same logic applies for multicast as for broadcast. ip_input_multicast
631 * dispatches packets to all zones that have members on the physical interface.
632 */
633
634 /*
635 * Squeue Fanout flags:
636 * 0: No fanout.
637 * 1: Fanout across all squeues
638 */
639 boolean_t ip_squeue_fanout = 0;
640
641 /*
642 * Maximum dups allowed per packet.
643 */
644 uint_t ip_max_frag_dups = 10;
645
646 static int ip_open(queue_t *q, dev_t *devp, int flag, int sflag,
647 cred_t *credp, boolean_t isv6);
648 static mblk_t *ip_xmit_attach_llhdr(mblk_t *, nce_t *);
649
650 static boolean_t icmp_inbound_verify_v4(mblk_t *, icmph_t *, ip_recv_attr_t *);
651 static void icmp_inbound_too_big_v4(icmph_t *, ip_recv_attr_t *);
652 static void icmp_inbound_error_fanout_v4(mblk_t *, icmph_t *,
653 ip_recv_attr_t *);
654 static void icmp_options_update(ipha_t *);
655 static void icmp_param_problem(mblk_t *, uint8_t, ip_recv_attr_t *);
656 static void icmp_pkt(mblk_t *, void *, size_t, ip_recv_attr_t *);
657 static mblk_t *icmp_pkt_err_ok(mblk_t *, ip_recv_attr_t *);
658 static void icmp_redirect_v4(mblk_t *mp, ipha_t *, icmph_t *,
659 ip_recv_attr_t *);
660 static void icmp_send_redirect(mblk_t *, ipaddr_t, ip_recv_attr_t *);
661 static void icmp_send_reply_v4(mblk_t *, ipha_t *, icmph_t *,
662 ip_recv_attr_t *);
663
664 mblk_t *ip_dlpi_alloc(size_t, t_uscalar_t);
665 char *ip_dot_addr(ipaddr_t, char *);
666 mblk_t *ip_carve_mp(mblk_t **, ssize_t);
667 int ip_close(queue_t *, int);
668 static char *ip_dot_saddr(uchar_t *, char *);
669 static void ip_lrput(queue_t *, mblk_t *);
670 ipaddr_t ip_net_mask(ipaddr_t);
671 char *ip_nv_lookup(nv_t *, int);
672 void ip_rput(queue_t *, mblk_t *);
673 static void ip_rput_dlpi_writer(ipsq_t *dummy_sq, queue_t *q, mblk_t *mp,
674 void *dummy_arg);
675 int ip_snmp_get(queue_t *, mblk_t *, int, boolean_t);
676 static mblk_t *ip_snmp_get_mib2_ip(queue_t *, mblk_t *,
677 mib2_ipIfStatsEntry_t *, ip_stack_t *, boolean_t);
678 static mblk_t *ip_snmp_get_mib2_ip_traffic_stats(queue_t *, mblk_t *,
679 ip_stack_t *, boolean_t);
680 static mblk_t *ip_snmp_get_mib2_ip6(queue_t *, mblk_t *, ip_stack_t *,
681 boolean_t);
682 static mblk_t *ip_snmp_get_mib2_icmp(queue_t *, mblk_t *, ip_stack_t *ipst);
683 static mblk_t *ip_snmp_get_mib2_icmp6(queue_t *, mblk_t *, ip_stack_t *ipst);
684 static mblk_t *ip_snmp_get_mib2_igmp(queue_t *, mblk_t *, ip_stack_t *ipst);
685 static mblk_t *ip_snmp_get_mib2_multi(queue_t *, mblk_t *, ip_stack_t *ipst);
686 static mblk_t *ip_snmp_get_mib2_ip_addr(queue_t *, mblk_t *,
687 ip_stack_t *ipst, boolean_t);
688 static mblk_t *ip_snmp_get_mib2_ip6_addr(queue_t *, mblk_t *,
689 ip_stack_t *ipst, boolean_t);
690 static mblk_t *ip_snmp_get_mib2_ip_group_src(queue_t *, mblk_t *,
691 ip_stack_t *ipst);
692 static mblk_t *ip_snmp_get_mib2_ip6_group_src(queue_t *, mblk_t *,
693 ip_stack_t *ipst);
694 static mblk_t *ip_snmp_get_mib2_ip_group_mem(queue_t *, mblk_t *,
695 ip_stack_t *ipst);
696 static mblk_t *ip_snmp_get_mib2_ip6_group_mem(queue_t *, mblk_t *,
697 ip_stack_t *ipst);
698 static mblk_t *ip_snmp_get_mib2_virt_multi(queue_t *, mblk_t *,
699 ip_stack_t *ipst);
700 static mblk_t *ip_snmp_get_mib2_multi_rtable(queue_t *, mblk_t *,
701 ip_stack_t *ipst);
702 static mblk_t *ip_snmp_get_mib2_ip_route_media(queue_t *, mblk_t *, int,
703 ip_stack_t *ipst);
704 static mblk_t *ip_snmp_get_mib2_ip6_route_media(queue_t *, mblk_t *, int,
705 ip_stack_t *ipst);
706 static void ip_snmp_get2_v4(ire_t *, iproutedata_t *);
707 static void ip_snmp_get2_v6_route(ire_t *, iproutedata_t *);
708 static int ip_snmp_get2_v4_media(ncec_t *, iproutedata_t *);
709 static int ip_snmp_get2_v6_media(ncec_t *, iproutedata_t *);
710 int ip_snmp_set(queue_t *, int, int, uchar_t *, int);
711
712 static mblk_t *ip_fragment_copyhdr(uchar_t *, int, int, ip_stack_t *,
713 mblk_t *);
714
715 static void conn_drain_init(ip_stack_t *);
716 static void conn_drain_fini(ip_stack_t *);
717 static void conn_drain(conn_t *connp, boolean_t closing);
718
719 static void conn_walk_drain(ip_stack_t *, idl_tx_list_t *);
720 static void conn_walk_sctp(pfv_t, void *, zoneid_t, netstack_t *);
721
722 static void *ip_stack_init(netstackid_t stackid, netstack_t *ns);
723 static void ip_stack_shutdown(netstackid_t stackid, void *arg);
724 static void ip_stack_fini(netstackid_t stackid, void *arg);
725
726 static int ip_multirt_apply_membership(int (*fn)(conn_t *, boolean_t,
727 const in6_addr_t *, ipaddr_t, uint_t, mcast_record_t, const in6_addr_t *),
728 ire_t *, conn_t *, boolean_t, const in6_addr_t *, mcast_record_t,
729 const in6_addr_t *);
730
731 static int ip_squeue_switch(int);
732
733 static void *ip_kstat_init(netstackid_t, ip_stack_t *);
734 static void ip_kstat_fini(netstackid_t, kstat_t *);
735 static int ip_kstat_update(kstat_t *kp, int rw);
736 static void *icmp_kstat_init(netstackid_t);
737 static void icmp_kstat_fini(netstackid_t, kstat_t *);
738 static int icmp_kstat_update(kstat_t *kp, int rw);
739 static void *ip_kstat2_init(netstackid_t, ip_stat_t *);
740 static void ip_kstat2_fini(netstackid_t, kstat_t *);
741
742 static void ipobs_init(ip_stack_t *);
743 static void ipobs_fini(ip_stack_t *);
744
745 static int ip_tp_cpu_update(cpu_setup_t, int, void *);
746
747 ipaddr_t ip_g_all_ones = IP_HOST_MASK;
748
749 static long ip_rput_pullups;
750 int dohwcksum = 1; /* use h/w cksum if supported by the hardware */
751
752 vmem_t *ip_minor_arena_sa; /* for minor nos. from INET_MIN_DEV+2 thru 2^^18-1 */
753 vmem_t *ip_minor_arena_la; /* for minor nos. from 2^^18 thru 2^^32-1 */
754
755 int ip_debug;
756
757 /*
758 * Multirouting/CGTP stuff
759 */
760 int ip_cgtp_filter_rev = CGTP_FILTER_REV; /* CGTP hooks version */
761
762 /*
763 * IP tunables related declarations. Definitions are in ip_tunables.c
764 */
765 extern mod_prop_info_t ip_propinfo_tbl[];
766 extern int ip_propinfo_count;
767
768 /*
769 * Table of IP ioctls encoding the various properties of the ioctl and
770 * indexed based on the last byte of the ioctl command. Occasionally there
771 * is a clash, and there is more than 1 ioctl with the same last byte.
772 * In such a case 1 ioctl is encoded in the ndx table and the remaining
773 * ioctls are encoded in the misc table. An entry in the ndx table is
774 * retrieved by indexing on the last byte of the ioctl command and comparing
775 * the ioctl command with the value in the ndx table. In the event of a
776 * mismatch the misc table is then searched sequentially for the desired
777 * ioctl command.
778 *
779 * Entry: <command> <copyin_size> <flags> <cmd_type> <function> <restart_func>
780 */
781 ip_ioctl_cmd_t ip_ndx_ioctl_table[] = {
782 /* 000 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
783 /* 001 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
784 /* 002 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
785 /* 003 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
786 /* 004 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
787 /* 005 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
788 /* 006 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
789 /* 007 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
790 /* 008 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
791 /* 009 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
792
793 /* 010 */ { SIOCADDRT, sizeof (struct rtentry), IPI_PRIV,
794 MISC_CMD, ip_siocaddrt, NULL },
795 /* 011 */ { SIOCDELRT, sizeof (struct rtentry), IPI_PRIV,
796 MISC_CMD, ip_siocdelrt, NULL },
797
798 /* 012 */ { SIOCSIFADDR, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
799 IF_CMD, ip_sioctl_addr, ip_sioctl_addr_restart },
800 /* 013 */ { SIOCGIFADDR, sizeof (struct ifreq), IPI_GET_CMD,
801 IF_CMD, ip_sioctl_get_addr, NULL },
802
803 /* 014 */ { SIOCSIFDSTADDR, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
804 IF_CMD, ip_sioctl_dstaddr, ip_sioctl_dstaddr_restart },
805 /* 015 */ { SIOCGIFDSTADDR, sizeof (struct ifreq),
806 IPI_GET_CMD, IF_CMD, ip_sioctl_get_dstaddr, NULL },
807
808 /* 016 */ { SIOCSIFFLAGS, sizeof (struct ifreq),
809 IPI_PRIV | IPI_WR,
810 IF_CMD, ip_sioctl_flags, ip_sioctl_flags_restart },
811 /* 017 */ { SIOCGIFFLAGS, sizeof (struct ifreq),
812 IPI_MODOK | IPI_GET_CMD,
813 IF_CMD, ip_sioctl_get_flags, NULL },
814
815 /* 018 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
816 /* 019 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
817
818 /* copyin size cannot be coded for SIOCGIFCONF */
819 /* 020 */ { O_SIOCGIFCONF, 0, IPI_GET_CMD,
820 MISC_CMD, ip_sioctl_get_ifconf, NULL },
821
822 /* 021 */ { SIOCSIFMTU, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
823 IF_CMD, ip_sioctl_mtu, NULL },
824 /* 022 */ { SIOCGIFMTU, sizeof (struct ifreq), IPI_GET_CMD,
825 IF_CMD, ip_sioctl_get_mtu, NULL },
826 /* 023 */ { SIOCGIFBRDADDR, sizeof (struct ifreq),
827 IPI_GET_CMD, IF_CMD, ip_sioctl_get_brdaddr, NULL },
828 /* 024 */ { SIOCSIFBRDADDR, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
829 IF_CMD, ip_sioctl_brdaddr, NULL },
830 /* 025 */ { SIOCGIFNETMASK, sizeof (struct ifreq),
831 IPI_GET_CMD, IF_CMD, ip_sioctl_get_netmask, NULL },
832 /* 026 */ { SIOCSIFNETMASK, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
833 IF_CMD, ip_sioctl_netmask, ip_sioctl_netmask_restart },
834 /* 027 */ { SIOCGIFMETRIC, sizeof (struct ifreq),
835 IPI_GET_CMD, IF_CMD, ip_sioctl_get_metric, NULL },
836 /* 028 */ { SIOCSIFMETRIC, sizeof (struct ifreq), IPI_PRIV,
837 IF_CMD, ip_sioctl_metric, NULL },
838 /* 029 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
839
840 /* See 166-168 below for extended SIOC*XARP ioctls */
841 /* 030 */ { SIOCSARP, sizeof (struct arpreq), IPI_PRIV | IPI_WR,
842 ARP_CMD, ip_sioctl_arp, NULL },
843 /* 031 */ { SIOCGARP, sizeof (struct arpreq), IPI_GET_CMD,
844 ARP_CMD, ip_sioctl_arp, NULL },
845 /* 032 */ { SIOCDARP, sizeof (struct arpreq), IPI_PRIV | IPI_WR,
846 ARP_CMD, ip_sioctl_arp, NULL },
847
848 /* 033 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
849 /* 034 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
850 /* 035 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
851 /* 036 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
852 /* 037 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
853 /* 038 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
854 /* 039 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
855 /* 040 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
856 /* 041 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
857 /* 042 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
858 /* 043 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
859 /* 044 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
860 /* 045 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
861 /* 046 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
862 /* 047 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
863 /* 048 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
864 /* 049 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
865 /* 050 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
866 /* 051 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
867 /* 052 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
868 /* 053 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
869
870 /* 054 */ { IF_UNITSEL, sizeof (int), IPI_PRIV | IPI_WR | IPI_MODOK,
871 MISC_CMD, if_unitsel, if_unitsel_restart },
872
873 /* 055 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
874 /* 056 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
875 /* 057 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
876 /* 058 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
877 /* 059 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
878 /* 060 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
879 /* 061 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
880 /* 062 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
881 /* 063 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
882 /* 064 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
883 /* 065 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
884 /* 066 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
885 /* 067 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
886 /* 068 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
887 /* 069 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
888 /* 070 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
889 /* 071 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
890 /* 072 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
891
892 /* 073 */ { SIOCSIFNAME, sizeof (struct ifreq),
893 IPI_PRIV | IPI_WR | IPI_MODOK,
894 IF_CMD, ip_sioctl_sifname, NULL },
895
896 /* 074 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
897 /* 075 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
898 /* 076 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
899 /* 077 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
900 /* 078 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
901 /* 079 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
902 /* 080 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
903 /* 081 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
904 /* 082 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
905 /* 083 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
906 /* 084 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
907 /* 085 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
908 /* 086 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
909
910 /* 087 */ { SIOCGIFNUM, sizeof (int), IPI_GET_CMD,
911 MISC_CMD, ip_sioctl_get_ifnum, NULL },
912 /* 088 */ { SIOCGIFMUXID, sizeof (struct ifreq), IPI_GET_CMD,
913 IF_CMD, ip_sioctl_get_muxid, NULL },
914 /* 089 */ { SIOCSIFMUXID, sizeof (struct ifreq),
915 IPI_PRIV | IPI_WR, IF_CMD, ip_sioctl_muxid, NULL },
916
917 /* Both if and lif variants share same func */
918 /* 090 */ { SIOCGIFINDEX, sizeof (struct ifreq), IPI_GET_CMD,
919 IF_CMD, ip_sioctl_get_lifindex, NULL },
920 /* Both if and lif variants share same func */
921 /* 091 */ { SIOCSIFINDEX, sizeof (struct ifreq),
922 IPI_PRIV | IPI_WR, IF_CMD, ip_sioctl_slifindex, NULL },
923
924 /* copyin size cannot be coded for SIOCGIFCONF */
925 /* 092 */ { SIOCGIFCONF, 0, IPI_GET_CMD,
926 MISC_CMD, ip_sioctl_get_ifconf, NULL },
927 /* 093 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
928 /* 094 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
929 /* 095 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
930 /* 096 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
931 /* 097 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
932 /* 098 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
933 /* 099 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
934 /* 100 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
935 /* 101 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
936 /* 102 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
937 /* 103 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
938 /* 104 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
939 /* 105 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
940 /* 106 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
941 /* 107 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
942 /* 108 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
943 /* 109 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
944
945 /* 110 */ { SIOCLIFREMOVEIF, sizeof (struct lifreq),
946 IPI_PRIV | IPI_WR, LIF_CMD, ip_sioctl_removeif,
947 ip_sioctl_removeif_restart },
948 /* 111 */ { SIOCLIFADDIF, sizeof (struct lifreq),
949 IPI_GET_CMD | IPI_PRIV | IPI_WR,
950 LIF_CMD, ip_sioctl_addif, NULL },
951 #define SIOCLIFADDR_NDX 112
952 /* 112 */ { SIOCSLIFADDR, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
953 LIF_CMD, ip_sioctl_addr, ip_sioctl_addr_restart },
954 /* 113 */ { SIOCGLIFADDR, sizeof (struct lifreq),
955 IPI_GET_CMD, LIF_CMD, ip_sioctl_get_addr, NULL },
956 /* 114 */ { SIOCSLIFDSTADDR, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
957 LIF_CMD, ip_sioctl_dstaddr, ip_sioctl_dstaddr_restart },
958 /* 115 */ { SIOCGLIFDSTADDR, sizeof (struct lifreq),
959 IPI_GET_CMD, LIF_CMD, ip_sioctl_get_dstaddr, NULL },
960 /* 116 */ { SIOCSLIFFLAGS, sizeof (struct lifreq),
961 IPI_PRIV | IPI_WR,
962 LIF_CMD, ip_sioctl_flags, ip_sioctl_flags_restart },
963 /* 117 */ { SIOCGLIFFLAGS, sizeof (struct lifreq),
964 IPI_GET_CMD | IPI_MODOK,
965 LIF_CMD, ip_sioctl_get_flags, NULL },
966
967 /* 118 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
968 /* 119 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
969
970 /* 120 */ { O_SIOCGLIFCONF, 0, IPI_GET_CMD, MISC_CMD,
971 ip_sioctl_get_lifconf, NULL },
972 /* 121 */ { SIOCSLIFMTU, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
973 LIF_CMD, ip_sioctl_mtu, NULL },
974 /* 122 */ { SIOCGLIFMTU, sizeof (struct lifreq), IPI_GET_CMD,
975 LIF_CMD, ip_sioctl_get_mtu, NULL },
976 /* 123 */ { SIOCGLIFBRDADDR, sizeof (struct lifreq),
977 IPI_GET_CMD, LIF_CMD, ip_sioctl_get_brdaddr, NULL },
978 /* 124 */ { SIOCSLIFBRDADDR, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
979 LIF_CMD, ip_sioctl_brdaddr, NULL },
980 /* 125 */ { SIOCGLIFNETMASK, sizeof (struct lifreq),
981 IPI_GET_CMD, LIF_CMD, ip_sioctl_get_netmask, NULL },
982 /* 126 */ { SIOCSLIFNETMASK, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
983 LIF_CMD, ip_sioctl_netmask, ip_sioctl_netmask_restart },
984 /* 127 */ { SIOCGLIFMETRIC, sizeof (struct lifreq),
985 IPI_GET_CMD, LIF_CMD, ip_sioctl_get_metric, NULL },
986 /* 128 */ { SIOCSLIFMETRIC, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
987 LIF_CMD, ip_sioctl_metric, NULL },
988 /* 129 */ { SIOCSLIFNAME, sizeof (struct lifreq),
989 IPI_PRIV | IPI_WR | IPI_MODOK,
990 LIF_CMD, ip_sioctl_slifname,
991 ip_sioctl_slifname_restart },
992
993 /* 130 */ { SIOCGLIFNUM, sizeof (struct lifnum), IPI_GET_CMD,
994 MISC_CMD, ip_sioctl_get_lifnum, NULL },
995 /* 131 */ { SIOCGLIFMUXID, sizeof (struct lifreq),
996 IPI_GET_CMD, LIF_CMD, ip_sioctl_get_muxid, NULL },
997 /* 132 */ { SIOCSLIFMUXID, sizeof (struct lifreq),
998 IPI_PRIV | IPI_WR, LIF_CMD, ip_sioctl_muxid, NULL },
999 /* 133 */ { SIOCGLIFINDEX, sizeof (struct lifreq),
1000 IPI_GET_CMD, LIF_CMD, ip_sioctl_get_lifindex, 0 },
1001 /* 134 */ { SIOCSLIFINDEX, sizeof (struct lifreq),
1002 IPI_PRIV | IPI_WR, LIF_CMD, ip_sioctl_slifindex, 0 },
1003 /* 135 */ { SIOCSLIFTOKEN, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
1004 LIF_CMD, ip_sioctl_token, NULL },
1005 /* 136 */ { SIOCGLIFTOKEN, sizeof (struct lifreq),
1006 IPI_GET_CMD, LIF_CMD, ip_sioctl_get_token, NULL },
1007 /* 137 */ { SIOCSLIFSUBNET, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
1008 LIF_CMD, ip_sioctl_subnet, ip_sioctl_subnet_restart },
1009 /* 138 */ { SIOCGLIFSUBNET, sizeof (struct lifreq),
1010 IPI_GET_CMD, LIF_CMD, ip_sioctl_get_subnet, NULL },
1011 /* 139 */ { SIOCSLIFLNKINFO, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
1012 LIF_CMD, ip_sioctl_lnkinfo, NULL },
1013
1014 /* 140 */ { SIOCGLIFLNKINFO, sizeof (struct lifreq),
1015 IPI_GET_CMD, LIF_CMD, ip_sioctl_get_lnkinfo, NULL },
1016 /* 141 */ { SIOCLIFDELND, sizeof (struct lifreq), IPI_PRIV,
1017 LIF_CMD, ip_siocdelndp_v6, NULL },
1018 /* 142 */ { SIOCLIFGETND, sizeof (struct lifreq), IPI_GET_CMD,
1019 LIF_CMD, ip_siocqueryndp_v6, NULL },
1020 /* 143 */ { SIOCLIFSETND, sizeof (struct lifreq), IPI_PRIV,
1021 LIF_CMD, ip_siocsetndp_v6, NULL },
1022 /* 144 */ { SIOCTMYADDR, sizeof (struct sioc_addrreq), IPI_GET_CMD,
1023 MISC_CMD, ip_sioctl_tmyaddr, NULL },
1024 /* 145 */ { SIOCTONLINK, sizeof (struct sioc_addrreq), IPI_GET_CMD,
1025 MISC_CMD, ip_sioctl_tonlink, NULL },
1026 /* 146 */ { SIOCTMYSITE, sizeof (struct sioc_addrreq), 0,
1027 MISC_CMD, ip_sioctl_tmysite, NULL },
1028 /* 147 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1029 /* 148 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1030 /* IPSECioctls handled in ip_sioctl_copyin_setup itself */
1031 /* 149 */ { SIOCFIPSECONFIG, 0, IPI_PRIV, MISC_CMD, NULL, NULL },
1032 /* 150 */ { SIOCSIPSECONFIG, 0, IPI_PRIV, MISC_CMD, NULL, NULL },
1033 /* 151 */ { SIOCDIPSECONFIG, 0, IPI_PRIV, MISC_CMD, NULL, NULL },
1034 /* 152 */ { SIOCLIPSECONFIG, 0, IPI_PRIV, MISC_CMD, NULL, NULL },
1035
1036 /* 153 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1037
1038 /* 154 */ { SIOCGLIFBINDING, sizeof (struct lifreq), IPI_GET_CMD,
1039 LIF_CMD, ip_sioctl_get_binding, NULL },
1040 /* 155 */ { SIOCSLIFGROUPNAME, sizeof (struct lifreq),
1041 IPI_PRIV | IPI_WR,
1042 LIF_CMD, ip_sioctl_groupname, ip_sioctl_groupname },
1043 /* 156 */ { SIOCGLIFGROUPNAME, sizeof (struct lifreq),
1044 IPI_GET_CMD, LIF_CMD, ip_sioctl_get_groupname, NULL },
1045 /* 157 */ { SIOCGLIFGROUPINFO, sizeof (lifgroupinfo_t),
1046 IPI_GET_CMD, MISC_CMD, ip_sioctl_groupinfo, NULL },
1047
1048 /* Leave 158-160 unused; used to be SIOC*IFARP ioctls */
1049 /* 158 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1050 /* 159 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1051 /* 160 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1052
1053 /* 161 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1054
1055 /* These are handled in ip_sioctl_copyin_setup itself */
1056 /* 162 */ { SIOCGIP6ADDRPOLICY, 0, IPI_NULL_BCONT,
1057 MISC_CMD, NULL, NULL },
1058 /* 163 */ { SIOCSIP6ADDRPOLICY, 0, IPI_PRIV | IPI_NULL_BCONT,
1059 MISC_CMD, NULL, NULL },
1060 /* 164 */ { SIOCGDSTINFO, 0, IPI_GET_CMD, MISC_CMD, NULL, NULL },
1061
1062 /* 165 */ { SIOCGLIFCONF, 0, IPI_GET_CMD, MISC_CMD,
1063 ip_sioctl_get_lifconf, NULL },
1064
1065 /* 166 */ { SIOCSXARP, sizeof (struct xarpreq), IPI_PRIV | IPI_WR,
1066 XARP_CMD, ip_sioctl_arp, NULL },
1067 /* 167 */ { SIOCGXARP, sizeof (struct xarpreq), IPI_GET_CMD,
1068 XARP_CMD, ip_sioctl_arp, NULL },
1069 /* 168 */ { SIOCDXARP, sizeof (struct xarpreq), IPI_PRIV | IPI_WR,
1070 XARP_CMD, ip_sioctl_arp, NULL },
1071
1072 /* SIOCPOPSOCKFS is not handled by IP */
1073 /* 169 */ { IPI_DONTCARE /* SIOCPOPSOCKFS */, 0, 0, 0, NULL, NULL },
1074
1075 /* 170 */ { SIOCGLIFZONE, sizeof (struct lifreq),
1076 IPI_GET_CMD, LIF_CMD, ip_sioctl_get_lifzone, NULL },
1077 /* 171 */ { SIOCSLIFZONE, sizeof (struct lifreq),
1078 IPI_PRIV | IPI_WR, LIF_CMD, ip_sioctl_slifzone,
1079 ip_sioctl_slifzone_restart },
1080 /* 172-174 are SCTP ioctls and not handled by IP */
1081 /* 172 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1082 /* 173 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1083 /* 174 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1084 /* 175 */ { SIOCGLIFUSESRC, sizeof (struct lifreq),
1085 IPI_GET_CMD, LIF_CMD,
1086 ip_sioctl_get_lifusesrc, 0 },
1087 /* 176 */ { SIOCSLIFUSESRC, sizeof (struct lifreq),
1088 IPI_PRIV | IPI_WR,
1089 LIF_CMD, ip_sioctl_slifusesrc,
1090 NULL },
1091 /* 177 */ { SIOCGLIFSRCOF, 0, IPI_GET_CMD, MISC_CMD,
1092 ip_sioctl_get_lifsrcof, NULL },
1093 /* 178 */ { SIOCGMSFILTER, sizeof (struct group_filter), IPI_GET_CMD,
1094 MSFILT_CMD, ip_sioctl_msfilter, NULL },
1095 /* 179 */ { SIOCSMSFILTER, sizeof (struct group_filter), 0,
1096 MSFILT_CMD, ip_sioctl_msfilter, NULL },
1097 /* 180 */ { SIOCGIPMSFILTER, sizeof (struct ip_msfilter), IPI_GET_CMD,
1098 MSFILT_CMD, ip_sioctl_msfilter, NULL },
1099 /* 181 */ { SIOCSIPMSFILTER, sizeof (struct ip_msfilter), 0,
1100 MSFILT_CMD, ip_sioctl_msfilter, NULL },
1101 /* 182 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1102 /* SIOCSENABLESDP is handled by SDP */
1103 /* 183 */ { IPI_DONTCARE /* SIOCSENABLESDP */, 0, 0, 0, NULL, NULL },
1104 /* 184 */ { IPI_DONTCARE /* SIOCSQPTR */, 0, 0, 0, NULL, NULL },
1105 /* 185 */ { SIOCGIFHWADDR, sizeof (struct ifreq), IPI_GET_CMD,
1106 IF_CMD, ip_sioctl_get_ifhwaddr, NULL },
1107 /* 186 */ { IPI_DONTCARE /* SIOCGSTAMP */, 0, 0, 0, NULL, NULL },
1108 /* 187 */ { SIOCILB, 0, IPI_PRIV | IPI_GET_CMD, MISC_CMD,
1109 ip_sioctl_ilb_cmd, NULL },
1110 /* 188 */ { SIOCGETPROP, 0, IPI_GET_CMD, 0, NULL, NULL },
1111 /* 189 */ { SIOCSETPROP, 0, IPI_PRIV | IPI_WR, 0, NULL, NULL},
1112 /* 190 */ { SIOCGLIFDADSTATE, sizeof (struct lifreq),
1113 IPI_GET_CMD, LIF_CMD, ip_sioctl_get_dadstate, NULL },
1114 /* 191 */ { SIOCSLIFPREFIX, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
1115 LIF_CMD, ip_sioctl_prefix, ip_sioctl_prefix_restart },
1116 /* 192 */ { SIOCGLIFHWADDR, sizeof (struct lifreq), IPI_GET_CMD,
1117 LIF_CMD, ip_sioctl_get_lifhwaddr, NULL }
1118 };
1119
1120 int ip_ndx_ioctl_count = sizeof (ip_ndx_ioctl_table) / sizeof (ip_ioctl_cmd_t);
1121
1122 ip_ioctl_cmd_t ip_misc_ioctl_table[] = {
1123 { I_LINK, 0, IPI_PRIV | IPI_WR, 0, NULL, NULL },
1124 { I_UNLINK, 0, IPI_PRIV | IPI_WR, 0, NULL, NULL },
1125 { I_PLINK, 0, IPI_PRIV | IPI_WR, 0, NULL, NULL },
1126 { I_PUNLINK, 0, IPI_PRIV | IPI_WR, 0, NULL, NULL },
1127 { ND_GET, 0, 0, 0, NULL, NULL },
1128 { ND_SET, 0, IPI_PRIV | IPI_WR, 0, NULL, NULL },
1129 { IP_IOCTL, 0, 0, 0, NULL, NULL },
1130 { SIOCGETVIFCNT, sizeof (struct sioc_vif_req), IPI_GET_CMD,
1131 MISC_CMD, mrt_ioctl},
1132 { SIOCGETSGCNT, sizeof (struct sioc_sg_req), IPI_GET_CMD,
1133 MISC_CMD, mrt_ioctl},
1134 { SIOCGETLSGCNT, sizeof (struct sioc_lsg_req), IPI_GET_CMD,
1135 MISC_CMD, mrt_ioctl}
1136 };
1137
1138 int ip_misc_ioctl_count =
1139 sizeof (ip_misc_ioctl_table) / sizeof (ip_ioctl_cmd_t);
1140
1141 int conn_drain_nthreads; /* Number of drainers reqd. */
1142 /* Settable in /etc/system */
1143 /* Defined in ip_ire.c */
1144 extern uint32_t ip_ire_max_bucket_cnt, ip6_ire_max_bucket_cnt;
1145 extern uint32_t ip_ire_min_bucket_cnt, ip6_ire_min_bucket_cnt;
1146 extern uint32_t ip_ire_mem_ratio, ip_ire_cpu_ratio;
1147
1148 static nv_t ire_nv_arr[] = {
1149 { IRE_BROADCAST, "BROADCAST" },
1150 { IRE_LOCAL, "LOCAL" },
1151 { IRE_LOOPBACK, "LOOPBACK" },
1152 { IRE_DEFAULT, "DEFAULT" },
1153 { IRE_PREFIX, "PREFIX" },
1154 { IRE_IF_NORESOLVER, "IF_NORESOL" },
1155 { IRE_IF_RESOLVER, "IF_RESOLV" },
1156 { IRE_IF_CLONE, "IF_CLONE" },
1157 { IRE_HOST, "HOST" },
1158 { IRE_MULTICAST, "MULTICAST" },
1159 { IRE_NOROUTE, "NOROUTE" },
1160 { 0 }
1161 };
1162
1163 nv_t *ire_nv_tbl = ire_nv_arr;
1164
1165 /* Simple ICMP IP Header Template */
1166 static ipha_t icmp_ipha = {
1167 IP_SIMPLE_HDR_VERSION, 0, 0, 0, 0, 0, IPPROTO_ICMP
1168 };
1169
1170 struct module_info ip_mod_info = {
1171 IP_MOD_ID, IP_MOD_NAME, IP_MOD_MINPSZ, IP_MOD_MAXPSZ, IP_MOD_HIWAT,
1172 IP_MOD_LOWAT
1173 };
1174
1175 /*
1176 * Duplicate static symbols within a module confuses mdb; so we avoid the
1177 * problem by making the symbols here distinct from those in udp.c.
1178 */
1179
1180 /*
1181 * Entry points for IP as a device and as a module.
1182 * We have separate open functions for the /dev/ip and /dev/ip6 devices.
1183 */
1184 static struct qinit iprinitv4 = {
1185 (pfi_t)ip_rput, NULL, ip_openv4, ip_close, NULL,
1186 &ip_mod_info
1187 };
1188
1189 struct qinit iprinitv6 = {
1190 (pfi_t)ip_rput_v6, NULL, ip_openv6, ip_close, NULL,
1191 &ip_mod_info
1192 };
1193
1194 static struct qinit ipwinit = {
1195 (pfi_t)ip_wput_nondata, (pfi_t)ip_wsrv, NULL, NULL, NULL,
1196 &ip_mod_info
1197 };
1198
1199 static struct qinit iplrinit = {
1200 (pfi_t)ip_lrput, NULL, ip_openv4, ip_close, NULL,
1201 &ip_mod_info
1202 };
1203
1204 static struct qinit iplwinit = {
1205 (pfi_t)ip_lwput, NULL, NULL, NULL, NULL,
1206 &ip_mod_info
1207 };
1208
1209 /* For AF_INET aka /dev/ip */
1210 struct streamtab ipinfov4 = {
1211 &iprinitv4, &ipwinit, &iplrinit, &iplwinit
1212 };
1213
1214 /* For AF_INET6 aka /dev/ip6 */
1215 struct streamtab ipinfov6 = {
1216 &iprinitv6, &ipwinit, &iplrinit, &iplwinit
1217 };
1218
1219 #ifdef DEBUG
1220 boolean_t skip_sctp_cksum = B_FALSE;
1221 #endif
1222
1223 /*
1224 * Generate an ICMP fragmentation needed message.
1225 * When called from ip_output side a minimal ip_recv_attr_t needs to be
1226 * constructed by the caller.
1227 */
1228 void
1229 icmp_frag_needed(mblk_t *mp, int mtu, ip_recv_attr_t *ira)
1230 {
1231 icmph_t icmph;
1232 ip_stack_t *ipst = ira->ira_ill->ill_ipst;
1233
1234 mp = icmp_pkt_err_ok(mp, ira);
1235 if (mp == NULL)
1236 return;
1237
1238 bzero(&icmph, sizeof (icmph_t));
1239 icmph.icmph_type = ICMP_DEST_UNREACHABLE;
1240 icmph.icmph_code = ICMP_FRAGMENTATION_NEEDED;
1241 icmph.icmph_du_mtu = htons((uint16_t)mtu);
1242 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutFragNeeded);
1243 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDestUnreachs);
1244
1245 icmp_pkt(mp, &icmph, sizeof (icmph_t), ira);
1246 }
1247
1248 /*
1249 * icmp_inbound_v4 deals with ICMP messages that are handled by IP.
1250 * If the ICMP message is consumed by IP, i.e., it should not be delivered
1251 * to any IPPROTO_ICMP raw sockets, then it returns NULL.
1252 * Likewise, if the ICMP error is misformed (too short, etc), then it
1253 * returns NULL. The caller uses this to determine whether or not to send
1254 * to raw sockets.
1255 *
1256 * All error messages are passed to the matching transport stream.
1257 *
1258 * The following cases are handled by icmp_inbound:
1259 * 1) It needs to send a reply back and possibly delivering it
1260 * to the "interested" upper clients.
1261 * 2) Return the mblk so that the caller can pass it to the RAW socket clients.
1262 * 3) It needs to change some values in IP only.
1263 * 4) It needs to change some values in IP and upper layers e.g TCP
1264 * by delivering an error to the upper layers.
1265 *
1266 * We handle the above three cases in the context of IPsec in the
1267 * following way :
1268 *
1269 * 1) Send the reply back in the same way as the request came in.
1270 * If it came in encrypted, it goes out encrypted. If it came in
1271 * clear, it goes out in clear. Thus, this will prevent chosen
1272 * plain text attack.
1273 * 2) The client may or may not expect things to come in secure.
1274 * If it comes in secure, the policy constraints are checked
1275 * before delivering it to the upper layers. If it comes in
1276 * clear, ipsec_inbound_accept_clear will decide whether to
1277 * accept this in clear or not. In both the cases, if the returned
1278 * message (IP header + 8 bytes) that caused the icmp message has
1279 * AH/ESP headers, it is sent up to AH/ESP for validation before
1280 * sending up. If there are only 8 bytes of returned message, then
1281 * upper client will not be notified.
1282 * 3) Check with global policy to see whether it matches the constaints.
1283 * But this will be done only if icmp_accept_messages_in_clear is
1284 * zero.
1285 * 4) If we need to change both in IP and ULP, then the decision taken
1286 * while affecting the values in IP and while delivering up to TCP
1287 * should be the same.
1288 *
1289 * There are two cases.
1290 *
1291 * a) If we reject data at the IP layer (ipsec_check_global_policy()
1292 * failed), we will not deliver it to the ULP, even though they
1293 * are *willing* to accept in *clear*. This is fine as our global
1294 * disposition to icmp messages asks us reject the datagram.
1295 *
1296 * b) If we accept data at the IP layer (ipsec_check_global_policy()
1297 * succeeded or icmp_accept_messages_in_clear is 1), and not able
1298 * to deliver it to ULP (policy failed), it can lead to
1299 * consistency problems. The cases known at this time are
1300 * ICMP_DESTINATION_UNREACHABLE messages with following code
1301 * values :
1302 *
1303 * - ICMP_FRAGMENTATION_NEEDED : IP adapts to the new value
1304 * and Upper layer rejects. Then the communication will
1305 * come to a stop. This is solved by making similar decisions
1306 * at both levels. Currently, when we are unable to deliver
1307 * to the Upper Layer (due to policy failures) while IP has
1308 * adjusted dce_pmtu, the next outbound datagram would
1309 * generate a local ICMP_FRAGMENTATION_NEEDED message - which
1310 * will be with the right level of protection. Thus the right
1311 * value will be communicated even if we are not able to
1312 * communicate when we get from the wire initially. But this
1313 * assumes there would be at least one outbound datagram after
1314 * IP has adjusted its dce_pmtu value. To make things
1315 * simpler, we accept in clear after the validation of
1316 * AH/ESP headers.
1317 *
1318 * - Other ICMP ERRORS : We may not be able to deliver it to the
1319 * upper layer depending on the level of protection the upper
1320 * layer expects and the disposition in ipsec_inbound_accept_clear().
1321 * ipsec_inbound_accept_clear() decides whether a given ICMP error
1322 * should be accepted in clear when the Upper layer expects secure.
1323 * Thus the communication may get aborted by some bad ICMP
1324 * packets.
1325 */
1326 mblk_t *
1327 icmp_inbound_v4(mblk_t *mp, ip_recv_attr_t *ira)
1328 {
1329 icmph_t *icmph;
1330 ipha_t *ipha; /* Outer header */
1331 int ip_hdr_length; /* Outer header length */
1332 boolean_t interested;
1333 ipif_t *ipif;
1334 uint32_t ts;
1335 uint32_t *tsp;
1336 timestruc_t now;
1337 ill_t *ill = ira->ira_ill;
1338 ip_stack_t *ipst = ill->ill_ipst;
1339 zoneid_t zoneid = ira->ira_zoneid;
1340 int len_needed;
1341 mblk_t *mp_ret = NULL;
1342
1343 ipha = (ipha_t *)mp->b_rptr;
1344
1345 BUMP_MIB(&ipst->ips_icmp_mib, icmpInMsgs);
1346
1347 ip_hdr_length = ira->ira_ip_hdr_length;
1348 if ((mp->b_wptr - mp->b_rptr) < (ip_hdr_length + ICMPH_SIZE)) {
1349 if (ira->ira_pktlen < (ip_hdr_length + ICMPH_SIZE)) {
1350 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInTruncatedPkts);
1351 ip_drop_input("ipIfStatsInTruncatedPkts", mp, ill);
1352 freemsg(mp);
1353 return (NULL);
1354 }
1355 /* Last chance to get real. */
1356 ipha = ip_pullup(mp, ip_hdr_length + ICMPH_SIZE, ira);
1357 if (ipha == NULL) {
1358 BUMP_MIB(&ipst->ips_icmp_mib, icmpInErrors);
1359 freemsg(mp);
1360 return (NULL);
1361 }
1362 }
1363
1364 /* The IP header will always be a multiple of four bytes */
1365 icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1366 ip2dbg(("icmp_inbound_v4: type %d code %d\n", icmph->icmph_type,
1367 icmph->icmph_code));
1368
1369 /*
1370 * We will set "interested" to "true" if we should pass a copy to
1371 * the transport or if we handle the packet locally.
1372 */
1373 interested = B_FALSE;
1374 switch (icmph->icmph_type) {
1375 case ICMP_ECHO_REPLY:
1376 BUMP_MIB(&ipst->ips_icmp_mib, icmpInEchoReps);
1377 break;
1378 case ICMP_DEST_UNREACHABLE:
1379 if (icmph->icmph_code == ICMP_FRAGMENTATION_NEEDED)
1380 BUMP_MIB(&ipst->ips_icmp_mib, icmpInFragNeeded);
1381 interested = B_TRUE; /* Pass up to transport */
1382 BUMP_MIB(&ipst->ips_icmp_mib, icmpInDestUnreachs);
1383 break;
1384 case ICMP_SOURCE_QUENCH:
1385 interested = B_TRUE; /* Pass up to transport */
1386 BUMP_MIB(&ipst->ips_icmp_mib, icmpInSrcQuenchs);
1387 break;
1388 case ICMP_REDIRECT:
1389 if (!ipst->ips_ip_ignore_redirect)
1390 interested = B_TRUE;
1391 BUMP_MIB(&ipst->ips_icmp_mib, icmpInRedirects);
1392 break;
1393 case ICMP_ECHO_REQUEST:
1394 /*
1395 * Whether to respond to echo requests that come in as IP
1396 * broadcasts or as IP multicast is subject to debate
1397 * (what isn't?). We aim to please, you pick it.
1398 * Default is do it.
1399 */
1400 if (ira->ira_flags & IRAF_MULTICAST) {
1401 /* multicast: respond based on tunable */
1402 interested = ipst->ips_ip_g_resp_to_echo_mcast;
1403 } else if (ira->ira_flags & IRAF_BROADCAST) {
1404 /* broadcast: respond based on tunable */
1405 interested = ipst->ips_ip_g_resp_to_echo_bcast;
1406 } else {
1407 /* unicast: always respond */
1408 interested = B_TRUE;
1409 }
1410 BUMP_MIB(&ipst->ips_icmp_mib, icmpInEchos);
1411 if (!interested) {
1412 /* We never pass these to RAW sockets */
1413 freemsg(mp);
1414 return (NULL);
1415 }
1416
1417 /* Check db_ref to make sure we can modify the packet. */
1418 if (mp->b_datap->db_ref > 1) {
1419 mblk_t *mp1;
1420
1421 mp1 = copymsg(mp);
1422 freemsg(mp);
1423 if (!mp1) {
1424 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
1425 return (NULL);
1426 }
1427 mp = mp1;
1428 ipha = (ipha_t *)mp->b_rptr;
1429 icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1430 }
1431 icmph->icmph_type = ICMP_ECHO_REPLY;
1432 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutEchoReps);
1433 icmp_send_reply_v4(mp, ipha, icmph, ira);
1434 return (NULL);
1435
1436 case ICMP_ROUTER_ADVERTISEMENT:
1437 case ICMP_ROUTER_SOLICITATION:
1438 break;
1439 case ICMP_TIME_EXCEEDED:
1440 interested = B_TRUE; /* Pass up to transport */
1441 BUMP_MIB(&ipst->ips_icmp_mib, icmpInTimeExcds);
1442 break;
1443 case ICMP_PARAM_PROBLEM:
1444 interested = B_TRUE; /* Pass up to transport */
1445 BUMP_MIB(&ipst->ips_icmp_mib, icmpInParmProbs);
1446 break;
1447 case ICMP_TIME_STAMP_REQUEST:
1448 /* Response to Time Stamp Requests is local policy. */
1449 if (ipst->ips_ip_g_resp_to_timestamp) {
1450 if (ira->ira_flags & IRAF_MULTIBROADCAST)
1451 interested =
1452 ipst->ips_ip_g_resp_to_timestamp_bcast;
1453 else
1454 interested = B_TRUE;
1455 }
1456 if (!interested) {
1457 /* We never pass these to RAW sockets */
1458 freemsg(mp);
1459 return (NULL);
1460 }
1461
1462 /* Make sure we have enough of the packet */
1463 len_needed = ip_hdr_length + ICMPH_SIZE +
1464 3 * sizeof (uint32_t);
1465
1466 if (mp->b_wptr - mp->b_rptr < len_needed) {
1467 ipha = ip_pullup(mp, len_needed, ira);
1468 if (ipha == NULL) {
1469 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1470 ip_drop_input("ipIfStatsInDiscards - ip_pullup",
1471 mp, ill);
1472 freemsg(mp);
1473 return (NULL);
1474 }
1475 /* Refresh following the pullup. */
1476 icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1477 }
1478 BUMP_MIB(&ipst->ips_icmp_mib, icmpInTimestamps);
1479 /* Check db_ref to make sure we can modify the packet. */
1480 if (mp->b_datap->db_ref > 1) {
1481 mblk_t *mp1;
1482
1483 mp1 = copymsg(mp);
1484 freemsg(mp);
1485 if (!mp1) {
1486 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
1487 return (NULL);
1488 }
1489 mp = mp1;
1490 ipha = (ipha_t *)mp->b_rptr;
1491 icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1492 }
1493 icmph->icmph_type = ICMP_TIME_STAMP_REPLY;
1494 tsp = (uint32_t *)&icmph[1];
1495 tsp++; /* Skip past 'originate time' */
1496 /* Compute # of milliseconds since midnight */
1497 gethrestime(&now);
1498 ts = (now.tv_sec % (24 * 60 * 60)) * 1000 +
1499 now.tv_nsec / (NANOSEC / MILLISEC);
1500 *tsp++ = htonl(ts); /* Lay in 'receive time' */
1501 *tsp++ = htonl(ts); /* Lay in 'send time' */
1502 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutTimestampReps);
1503 icmp_send_reply_v4(mp, ipha, icmph, ira);
1504 return (NULL);
1505
1506 case ICMP_TIME_STAMP_REPLY:
1507 BUMP_MIB(&ipst->ips_icmp_mib, icmpInTimestampReps);
1508 break;
1509 case ICMP_INFO_REQUEST:
1510 /* Per RFC 1122 3.2.2.7, ignore this. */
1511 case ICMP_INFO_REPLY:
1512 break;
1513 case ICMP_ADDRESS_MASK_REQUEST:
1514 if (ira->ira_flags & IRAF_MULTIBROADCAST) {
1515 interested =
1516 ipst->ips_ip_respond_to_address_mask_broadcast;
1517 } else {
1518 interested = B_TRUE;
1519 }
1520 if (!interested) {
1521 /* We never pass these to RAW sockets */
1522 freemsg(mp);
1523 return (NULL);
1524 }
1525 len_needed = ip_hdr_length + ICMPH_SIZE + IP_ADDR_LEN;
1526 if (mp->b_wptr - mp->b_rptr < len_needed) {
1527 ipha = ip_pullup(mp, len_needed, ira);
1528 if (ipha == NULL) {
1529 BUMP_MIB(ill->ill_ip_mib,
1530 ipIfStatsInTruncatedPkts);
1531 ip_drop_input("ipIfStatsInTruncatedPkts", mp,
1532 ill);
1533 freemsg(mp);
1534 return (NULL);
1535 }
1536 /* Refresh following the pullup. */
1537 icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1538 }
1539 BUMP_MIB(&ipst->ips_icmp_mib, icmpInAddrMasks);
1540 /* Check db_ref to make sure we can modify the packet. */
1541 if (mp->b_datap->db_ref > 1) {
1542 mblk_t *mp1;
1543
1544 mp1 = copymsg(mp);
1545 freemsg(mp);
1546 if (!mp1) {
1547 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
1548 return (NULL);
1549 }
1550 mp = mp1;
1551 ipha = (ipha_t *)mp->b_rptr;
1552 icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1553 }
1554 /*
1555 * Need the ipif with the mask be the same as the source
1556 * address of the mask reply. For unicast we have a specific
1557 * ipif. For multicast/broadcast we only handle onlink
1558 * senders, and use the source address to pick an ipif.
1559 */
1560 ipif = ipif_lookup_addr(ipha->ipha_dst, ill, zoneid, ipst);
1561 if (ipif == NULL) {
1562 /* Broadcast or multicast */
1563 ipif = ipif_lookup_remote(ill, ipha->ipha_src, zoneid);
1564 if (ipif == NULL) {
1565 freemsg(mp);
1566 return (NULL);
1567 }
1568 }
1569 icmph->icmph_type = ICMP_ADDRESS_MASK_REPLY;
1570 bcopy(&ipif->ipif_net_mask, &icmph[1], IP_ADDR_LEN);
1571 ipif_refrele(ipif);
1572 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutAddrMaskReps);
1573 icmp_send_reply_v4(mp, ipha, icmph, ira);
1574 return (NULL);
1575
1576 case ICMP_ADDRESS_MASK_REPLY:
1577 BUMP_MIB(&ipst->ips_icmp_mib, icmpInAddrMaskReps);
1578 break;
1579 default:
1580 interested = B_TRUE; /* Pass up to transport */
1581 BUMP_MIB(&ipst->ips_icmp_mib, icmpInUnknowns);
1582 break;
1583 }
1584 /*
1585 * See if there is an ICMP client to avoid an extra copymsg/freemsg
1586 * if there isn't one.
1587 */
1588 if (ipst->ips_ipcl_proto_fanout_v4[IPPROTO_ICMP].connf_head != NULL) {
1589 /* If there is an ICMP client and we want one too, copy it. */
1590
1591 if (!interested) {
1592 /* Caller will deliver to RAW sockets */
1593 return (mp);
1594 }
1595 mp_ret = copymsg(mp);
1596 if (mp_ret == NULL) {
1597 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1598 ip_drop_input("ipIfStatsInDiscards - copymsg", mp, ill);
1599 }
1600 } else if (!interested) {
1601 /* Neither we nor raw sockets are interested. Drop packet now */
1602 freemsg(mp);
1603 return (NULL);
1604 }
1605
1606 /*
1607 * ICMP error or redirect packet. Make sure we have enough of
1608 * the header and that db_ref == 1 since we might end up modifying
1609 * the packet.
1610 */
1611 if (mp->b_cont != NULL) {
1612 if (ip_pullup(mp, -1, ira) == NULL) {
1613 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1614 ip_drop_input("ipIfStatsInDiscards - ip_pullup",
1615 mp, ill);
1616 freemsg(mp);
1617 return (mp_ret);
1618 }
1619 }
1620
1621 if (mp->b_datap->db_ref > 1) {
1622 mblk_t *mp1;
1623
1624 mp1 = copymsg(mp);
1625 if (mp1 == NULL) {
1626 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1627 ip_drop_input("ipIfStatsInDiscards - copymsg", mp, ill);
1628 freemsg(mp);
1629 return (mp_ret);
1630 }
1631 freemsg(mp);
1632 mp = mp1;
1633 }
1634
1635 /*
1636 * In case mp has changed, verify the message before any further
1637 * processes.
1638 */
1639 ipha = (ipha_t *)mp->b_rptr;
1640 icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1641 if (!icmp_inbound_verify_v4(mp, icmph, ira)) {
1642 freemsg(mp);
1643 return (mp_ret);
1644 }
1645
1646 switch (icmph->icmph_type) {
1647 case ICMP_REDIRECT:
1648 icmp_redirect_v4(mp, ipha, icmph, ira);
1649 break;
1650 case ICMP_DEST_UNREACHABLE:
1651 if (icmph->icmph_code == ICMP_FRAGMENTATION_NEEDED) {
1652 /* Update DCE and adjust MTU is icmp header if needed */
1653 icmp_inbound_too_big_v4(icmph, ira);
1654 }
1655 /* FALLTHRU */
1656 default:
1657 icmp_inbound_error_fanout_v4(mp, icmph, ira);
1658 break;
1659 }
1660 return (mp_ret);
1661 }
1662
1663 /*
1664 * Send an ICMP echo, timestamp or address mask reply.
1665 * The caller has already updated the payload part of the packet.
1666 * We handle the ICMP checksum, IP source address selection and feed
1667 * the packet into ip_output_simple.
1668 */
1669 static void
1670 icmp_send_reply_v4(mblk_t *mp, ipha_t *ipha, icmph_t *icmph,
1671 ip_recv_attr_t *ira)
1672 {
1673 uint_t ip_hdr_length = ira->ira_ip_hdr_length;
1674 ill_t *ill = ira->ira_ill;
1675 ip_stack_t *ipst = ill->ill_ipst;
1676 ip_xmit_attr_t ixas;
1677
1678 /* Send out an ICMP packet */
1679 icmph->icmph_checksum = 0;
1680 icmph->icmph_checksum = IP_CSUM(mp, ip_hdr_length, 0);
1681 /* Reset time to live. */
1682 ipha->ipha_ttl = ipst->ips_ip_def_ttl;
1683 {
1684 /* Swap source and destination addresses */
1685 ipaddr_t tmp;
1686
1687 tmp = ipha->ipha_src;
1688 ipha->ipha_src = ipha->ipha_dst;
1689 ipha->ipha_dst = tmp;
1690 }
1691 ipha->ipha_ident = 0;
1692 if (!IS_SIMPLE_IPH(ipha))
1693 icmp_options_update(ipha);
1694
1695 bzero(&ixas, sizeof (ixas));
1696 ixas.ixa_flags = IXAF_BASIC_SIMPLE_V4;
1697 ixas.ixa_zoneid = ira->ira_zoneid;
1698 ixas.ixa_cred = kcred;
1699 ixas.ixa_cpid = NOPID;
1700 ixas.ixa_tsl = ira->ira_tsl; /* Behave as a multi-level responder */
1701 ixas.ixa_ifindex = 0;
1702 ixas.ixa_ipst = ipst;
1703 ixas.ixa_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1704
1705 if (!(ira->ira_flags & IRAF_IPSEC_SECURE)) {
1706 /*
1707 * This packet should go out the same way as it
1708 * came in i.e in clear, independent of the IPsec policy
1709 * for transmitting packets.
1710 */
1711 ixas.ixa_flags |= IXAF_NO_IPSEC;
1712 } else {
1713 if (!ipsec_in_to_out(ira, &ixas, mp, ipha, NULL)) {
1714 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1715 /* Note: mp already consumed and ip_drop_packet done */
1716 return;
1717 }
1718 }
1719 if (ira->ira_flags & IRAF_MULTIBROADCAST) {
1720 /*
1721 * Not one or our addresses (IRE_LOCALs), thus we let
1722 * ip_output_simple pick the source.
1723 */
1724 ipha->ipha_src = INADDR_ANY;
1725 ixas.ixa_flags |= IXAF_SET_SOURCE;
1726 }
1727 /* Should we send with DF and use dce_pmtu? */
1728 if (ipst->ips_ipv4_icmp_return_pmtu) {
1729 ixas.ixa_flags |= IXAF_PMTU_DISCOVERY;
1730 ipha->ipha_fragment_offset_and_flags |= IPH_DF_HTONS;
1731 }
1732
1733 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutMsgs);
1734
1735 (void) ip_output_simple(mp, &ixas);
1736 ixa_cleanup(&ixas);
1737 }
1738
1739 /*
1740 * Verify the ICMP messages for either for ICMP error or redirect packet.
1741 * The caller should have fully pulled up the message. If it's a redirect
1742 * packet, only basic checks on IP header will be done; otherwise, verify
1743 * the packet by looking at the included ULP header.
1744 *
1745 * Called before icmp_inbound_error_fanout_v4 is called.
1746 */
1747 static boolean_t
1748 icmp_inbound_verify_v4(mblk_t *mp, icmph_t *icmph, ip_recv_attr_t *ira)
1749 {
1750 ill_t *ill = ira->ira_ill;
1751 int hdr_length;
1752 ip_stack_t *ipst = ira->ira_ill->ill_ipst;
1753 conn_t *connp;
1754 ipha_t *ipha; /* Inner IP header */
1755
1756 ipha = (ipha_t *)&icmph[1];
1757 if ((uchar_t *)ipha + IP_SIMPLE_HDR_LENGTH > mp->b_wptr)
1758 goto truncated;
1759
1760 hdr_length = IPH_HDR_LENGTH(ipha);
1761
1762 if ((IPH_HDR_VERSION(ipha) != IPV4_VERSION))
1763 goto discard_pkt;
1764
1765 if (hdr_length < sizeof (ipha_t))
1766 goto truncated;
1767
1768 if ((uchar_t *)ipha + hdr_length > mp->b_wptr)
1769 goto truncated;
1770
1771 /*
1772 * Stop here for ICMP_REDIRECT.
1773 */
1774 if (icmph->icmph_type == ICMP_REDIRECT)
1775 return (B_TRUE);
1776
1777 /*
1778 * ICMP errors only.
1779 */
1780 switch (ipha->ipha_protocol) {
1781 case IPPROTO_UDP:
1782 /*
1783 * Verify we have at least ICMP_MIN_TP_HDR_LEN bytes of
1784 * transport header.
1785 */
1786 if ((uchar_t *)ipha + hdr_length + ICMP_MIN_TP_HDR_LEN >
1787 mp->b_wptr)
1788 goto truncated;
1789 break;
1790 case IPPROTO_TCP: {
1791 tcpha_t *tcpha;
1792
1793 /*
1794 * Verify we have at least ICMP_MIN_TP_HDR_LEN bytes of
1795 * transport header.
1796 */
1797 if ((uchar_t *)ipha + hdr_length + ICMP_MIN_TP_HDR_LEN >
1798 mp->b_wptr)
1799 goto truncated;
1800
1801 tcpha = (tcpha_t *)((uchar_t *)ipha + hdr_length);
1802 connp = ipcl_tcp_lookup_reversed_ipv4(ipha, tcpha, TCPS_LISTEN,
1803 ipst);
1804 if (connp == NULL)
1805 goto discard_pkt;
1806
1807 if ((connp->conn_verifyicmp != NULL) &&
1808 !connp->conn_verifyicmp(connp, tcpha, icmph, NULL, ira)) {
1809 CONN_DEC_REF(connp);
1810 goto discard_pkt;
1811 }
1812 CONN_DEC_REF(connp);
1813 break;
1814 }
1815 case IPPROTO_SCTP:
1816 /*
1817 * Verify we have at least ICMP_MIN_TP_HDR_LEN bytes of
1818 * transport header.
1819 */
1820 if ((uchar_t *)ipha + hdr_length + ICMP_MIN_TP_HDR_LEN >
1821 mp->b_wptr)
1822 goto truncated;
1823 break;
1824 case IPPROTO_ESP:
1825 case IPPROTO_AH:
1826 break;
1827 case IPPROTO_ENCAP:
1828 if ((uchar_t *)ipha + hdr_length + sizeof (ipha_t) >
1829 mp->b_wptr)
1830 goto truncated;
1831 break;
1832 default:
1833 break;
1834 }
1835
1836 return (B_TRUE);
1837
1838 discard_pkt:
1839 /* Bogus ICMP error. */
1840 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1841 return (B_FALSE);
1842
1843 truncated:
1844 /* We pulled up everthing already. Must be truncated */
1845 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInTruncatedPkts);
1846 ip_drop_input("ipIfStatsInTruncatedPkts", mp, ill);
1847 return (B_FALSE);
1848 }
1849
1850 /* Table from RFC 1191 */
1851 static int icmp_frag_size_table[] =
1852 { 32000, 17914, 8166, 4352, 2002, 1496, 1006, 508, 296, 68 };
1853
1854 /*
1855 * Process received ICMP Packet too big.
1856 * Just handles the DCE create/update, including using the above table of
1857 * PMTU guesses. The caller is responsible for validating the packet before
1858 * passing it in and also to fanout the ICMP error to any matching transport
1859 * conns. Assumes the message has been fully pulled up and verified.
1860 *
1861 * Before getting here, the caller has called icmp_inbound_verify_v4()
1862 * that should have verified with ULP to prevent undoing the changes we're
1863 * going to make to DCE. For example, TCP might have verified that the packet
1864 * which generated error is in the send window.
1865 *
1866 * In some cases modified this MTU in the ICMP header packet; the caller
1867 * should pass to the matching ULP after this returns.
1868 */
1869 static void
1870 icmp_inbound_too_big_v4(icmph_t *icmph, ip_recv_attr_t *ira)
1871 {
1872 dce_t *dce;
1873 int old_mtu;
1874 int mtu, orig_mtu;
1875 ipaddr_t dst;
1876 boolean_t disable_pmtud;
1877 ill_t *ill = ira->ira_ill;
1878 ip_stack_t *ipst = ill->ill_ipst;
1879 uint_t hdr_length;
1880 ipha_t *ipha;
1881
1882 /* Caller already pulled up everything. */
1883 ipha = (ipha_t *)&icmph[1];
1884 ASSERT(icmph->icmph_type == ICMP_DEST_UNREACHABLE &&
1885 icmph->icmph_code == ICMP_FRAGMENTATION_NEEDED);
1886 ASSERT(ill != NULL);
1887
1888 hdr_length = IPH_HDR_LENGTH(ipha);
1889
1890 /*
1891 * We handle path MTU for source routed packets since the DCE
1892 * is looked up using the final destination.
1893 */
1894 dst = ip_get_dst(ipha);
1895
1896 dce = dce_lookup_and_add_v4(dst, ipst);
1897 if (dce == NULL) {
1898 /* Couldn't add a unique one - ENOMEM */
1899 ip1dbg(("icmp_inbound_too_big_v4: no dce for 0x%x\n",
1900 ntohl(dst)));
1901 return;
1902 }
1903
1904 /* Check for MTU discovery advice as described in RFC 1191 */
1905 mtu = ntohs(icmph->icmph_du_mtu);
1906 orig_mtu = mtu;
1907 disable_pmtud = B_FALSE;
1908
1909 mutex_enter(&dce->dce_lock);
1910 if (dce->dce_flags & DCEF_PMTU)
1911 old_mtu = dce->dce_pmtu;
1912 else
1913 old_mtu = ill->ill_mtu;
1914
1915 if (icmph->icmph_du_zero != 0 || mtu < ipst->ips_ip_pmtu_min) {
1916 uint32_t length;
1917 int i;
1918
1919 /*
1920 * Use the table from RFC 1191 to figure out
1921 * the next "plateau" based on the length in
1922 * the original IP packet.
1923 */
1924 length = ntohs(ipha->ipha_length);
1925 DTRACE_PROBE2(ip4__pmtu__guess, dce_t *, dce,
1926 uint32_t, length);
1927 if (old_mtu <= length &&
1928 old_mtu >= length - hdr_length) {
1929 /*
1930 * Handle broken BSD 4.2 systems that
1931 * return the wrong ipha_length in ICMP
1932 * errors.
1933 */
1934 ip1dbg(("Wrong mtu: sent %d, dce %d\n",
1935 length, old_mtu));
1936 length -= hdr_length;
1937 }
1938 for (i = 0; i < A_CNT(icmp_frag_size_table); i++) {
1939 if (length > icmp_frag_size_table[i])
1940 break;
1941 }
1942 if (i == A_CNT(icmp_frag_size_table)) {
1943 /* Smaller than IP_MIN_MTU! */
1944 ip1dbg(("Too big for packet size %d\n",
1945 length));
1946 disable_pmtud = B_TRUE;
1947 mtu = ipst->ips_ip_pmtu_min;
1948 } else {
1949 mtu = icmp_frag_size_table[i];
1950 ip1dbg(("Calculated mtu %d, packet size %d, "
1951 "before %d\n", mtu, length, old_mtu));
1952 if (mtu < ipst->ips_ip_pmtu_min) {
1953 mtu = ipst->ips_ip_pmtu_min;
1954 disable_pmtud = B_TRUE;
1955 }
1956 }
1957 }
1958 if (disable_pmtud)
1959 dce->dce_flags |= DCEF_TOO_SMALL_PMTU;
1960 else
1961 dce->dce_flags &= ~DCEF_TOO_SMALL_PMTU;
1962
1963 dce->dce_pmtu = MIN(old_mtu, mtu);
1964 /* Prepare to send the new max frag size for the ULP. */
1965 icmph->icmph_du_zero = 0;
1966 icmph->icmph_du_mtu = htons((uint16_t)dce->dce_pmtu);
1967 DTRACE_PROBE4(ip4__pmtu__change, icmph_t *, icmph, dce_t *,
1968 dce, int, orig_mtu, int, mtu);
1969
1970 /* We now have a PMTU for sure */
1971 dce->dce_flags |= DCEF_PMTU;
1972 dce->dce_last_change_time = TICK_TO_SEC(ddi_get_lbolt64());
1973 mutex_exit(&dce->dce_lock);
1974 /*
1975 * After dropping the lock the new value is visible to everyone.
1976 * Then we bump the generation number so any cached values reinspect
1977 * the dce_t.
1978 */
1979 dce_increment_generation(dce);
1980 dce_refrele(dce);
1981 }
1982
1983 /*
1984 * If the packet in error is Self-Encapsulated, icmp_inbound_error_fanout_v4
1985 * calls this function.
1986 */
1987 static mblk_t *
1988 icmp_inbound_self_encap_error_v4(mblk_t *mp, ipha_t *ipha, ipha_t *in_ipha)
1989 {
1990 int length;
1991
1992 ASSERT(mp->b_datap->db_type == M_DATA);
1993
1994 /* icmp_inbound_v4 has already pulled up the whole error packet */
1995 ASSERT(mp->b_cont == NULL);
1996
1997 /*
1998 * The length that we want to overlay is the inner header
1999 * and what follows it.
2000 */
2001 length = msgdsize(mp) - ((uchar_t *)in_ipha - mp->b_rptr);
2002
2003 /*
2004 * Overlay the inner header and whatever follows it over the
2005 * outer header.
2006 */
2007 bcopy((uchar_t *)in_ipha, (uchar_t *)ipha, length);
2008
2009 /* Adjust for what we removed */
2010 mp->b_wptr -= (uchar_t *)in_ipha - (uchar_t *)ipha;
2011 return (mp);
2012 }
2013
2014 /*
2015 * Try to pass the ICMP message upstream in case the ULP cares.
2016 *
2017 * If the packet that caused the ICMP error is secure, we send
2018 * it to AH/ESP to make sure that the attached packet has a
2019 * valid association. ipha in the code below points to the
2020 * IP header of the packet that caused the error.
2021 *
2022 * For IPsec cases, we let the next-layer-up (which has access to
2023 * cached policy on the conn_t, or can query the SPD directly)
2024 * subtract out any IPsec overhead if they must. We therefore make no
2025 * adjustments here for IPsec overhead.
2026 *
2027 * IFN could have been generated locally or by some router.
2028 *
2029 * LOCAL : ire_send_wire (before calling ipsec_out_process) can call
2030 * icmp_frag_needed/icmp_pkt2big_v6 to generated a local IFN.
2031 * This happens because IP adjusted its value of MTU on an
2032 * earlier IFN message and could not tell the upper layer,
2033 * the new adjusted value of MTU e.g. Packet was encrypted
2034 * or there was not enough information to fanout to upper
2035 * layers. Thus on the next outbound datagram, ire_send_wire
2036 * generates the IFN, where IPsec processing has *not* been
2037 * done.
2038 *
2039 * Note that we retain ixa_fragsize across IPsec thus once
2040 * we have picking ixa_fragsize and entered ipsec_out_process we do
2041 * no change the fragsize even if the path MTU changes before
2042 * we reach ip_output_post_ipsec.
2043 *
2044 * In the local case, IRAF_LOOPBACK will be set indicating
2045 * that IFN was generated locally.
2046 *
2047 * ROUTER : IFN could be secure or non-secure.
2048 *
2049 * * SECURE : We use the IPSEC_IN to fanout to AH/ESP if the
2050 * packet in error has AH/ESP headers to validate the AH/ESP
2051 * headers. AH/ESP will verify whether there is a valid SA or
2052 * not and send it back. We will fanout again if we have more
2053 * data in the packet.
2054 *
2055 * If the packet in error does not have AH/ESP, we handle it
2056 * like any other case.
2057 *
2058 * * NON_SECURE : If the packet in error has AH/ESP headers, we send it
2059 * up to AH/ESP for validation. AH/ESP will verify whether there is a
2060 * valid SA or not and send it back. We will fanout again if
2061 * we have more data in the packet.
2062 *
2063 * If the packet in error does not have AH/ESP, we handle it
2064 * like any other case.
2065 *
2066 * The caller must have called icmp_inbound_verify_v4.
2067 */
2068 static void
2069 icmp_inbound_error_fanout_v4(mblk_t *mp, icmph_t *icmph, ip_recv_attr_t *ira)
2070 {
2071 uint16_t *up; /* Pointer to ports in ULP header */
2072 uint32_t ports; /* reversed ports for fanout */
2073 ipha_t ripha; /* With reversed addresses */
2074 ipha_t *ipha; /* Inner IP header */
2075 uint_t hdr_length; /* Inner IP header length */
2076 tcpha_t *tcpha;
2077 conn_t *connp;
2078 ill_t *ill = ira->ira_ill;
2079 ip_stack_t *ipst = ill->ill_ipst;
2080 ipsec_stack_t *ipss = ipst->ips_netstack->netstack_ipsec;
2081 ill_t *rill = ira->ira_rill;
2082
2083 /* Caller already pulled up everything. */
2084 ipha = (ipha_t *)&icmph[1];
2085 ASSERT((uchar_t *)&ipha[1] <= mp->b_wptr);
2086 ASSERT(mp->b_cont == NULL);
2087
2088 hdr_length = IPH_HDR_LENGTH(ipha);
2089 ira->ira_protocol = ipha->ipha_protocol;
2090
2091 /*
2092 * We need a separate IP header with the source and destination
2093 * addresses reversed to do fanout/classification because the ipha in
2094 * the ICMP error is in the form we sent it out.
2095 */
2096 ripha.ipha_src = ipha->ipha_dst;
2097 ripha.ipha_dst = ipha->ipha_src;
2098 ripha.ipha_protocol = ipha->ipha_protocol;
2099 ripha.ipha_version_and_hdr_length = ipha->ipha_version_and_hdr_length;
2100
2101 ip2dbg(("icmp_inbound_error_v4: proto %d %x to %x: %d/%d\n",
2102 ripha.ipha_protocol, ntohl(ipha->ipha_src),
2103 ntohl(ipha->ipha_dst),
2104 icmph->icmph_type, icmph->icmph_code));
2105
2106 switch (ipha->ipha_protocol) {
2107 case IPPROTO_UDP:
2108 up = (uint16_t *)((uchar_t *)ipha + hdr_length);
2109
2110 /* Attempt to find a client stream based on port. */
2111 ip2dbg(("icmp_inbound_error_v4: UDP ports %d to %d\n",
2112 ntohs(up[0]), ntohs(up[1])));
2113
2114 /* Note that we send error to all matches. */
2115 ira->ira_flags |= IRAF_ICMP_ERROR;
2116 ip_fanout_udp_multi_v4(mp, &ripha, up[0], up[1], ira);
2117 ira->ira_flags &= ~IRAF_ICMP_ERROR;
2118 return;
2119
2120 case IPPROTO_TCP:
2121 /*
2122 * Find a TCP client stream for this packet.
2123 * Note that we do a reverse lookup since the header is
2124 * in the form we sent it out.
2125 */
2126 tcpha = (tcpha_t *)((uchar_t *)ipha + hdr_length);
2127 connp = ipcl_tcp_lookup_reversed_ipv4(ipha, tcpha, TCPS_LISTEN,
2128 ipst);
2129 if (connp == NULL)
2130 goto discard_pkt;
2131
2132 if (CONN_INBOUND_POLICY_PRESENT(connp, ipss) ||
2133 (ira->ira_flags & IRAF_IPSEC_SECURE)) {
2134 mp = ipsec_check_inbound_policy(mp, connp,
2135 ipha, NULL, ira);
2136 if (mp == NULL) {
2137 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
2138 /* Note that mp is NULL */
2139 ip_drop_input("ipIfStatsInDiscards", mp, ill);
2140 CONN_DEC_REF(connp);
2141 return;
2142 }
2143 }
2144
2145 ira->ira_flags |= IRAF_ICMP_ERROR;
2146 ira->ira_ill = ira->ira_rill = NULL;
2147 if (IPCL_IS_TCP(connp)) {
2148 SQUEUE_ENTER_ONE(connp->conn_sqp, mp,
2149 connp->conn_recvicmp, connp, ira, SQ_FILL,
2150 SQTAG_TCP_INPUT_ICMP_ERR);
2151 } else {
2152 /* Not TCP; must be SOCK_RAW, IPPROTO_TCP */
2153 (connp->conn_recv)(connp, mp, NULL, ira);
2154 CONN_DEC_REF(connp);
2155 }
2156 ira->ira_ill = ill;
2157 ira->ira_rill = rill;
2158 ira->ira_flags &= ~IRAF_ICMP_ERROR;
2159 return;
2160
2161 case IPPROTO_SCTP:
2162 up = (uint16_t *)((uchar_t *)ipha + hdr_length);
2163 /* Find a SCTP client stream for this packet. */
2164 ((uint16_t *)&ports)[0] = up[1];
2165 ((uint16_t *)&ports)[1] = up[0];
2166
2167 ira->ira_flags |= IRAF_ICMP_ERROR;
2168 ip_fanout_sctp(mp, &ripha, NULL, ports, ira);
2169 ira->ira_flags &= ~IRAF_ICMP_ERROR;
2170 return;
2171
2172 case IPPROTO_ESP:
2173 case IPPROTO_AH:
2174 if (!ipsec_loaded(ipss)) {
2175 ip_proto_not_sup(mp, ira);
2176 return;
2177 }
2178
2179 if (ipha->ipha_protocol == IPPROTO_ESP)
2180 mp = ipsecesp_icmp_error(mp, ira);
2181 else
2182 mp = ipsecah_icmp_error(mp, ira);
2183 if (mp == NULL)
2184 return;
2185
2186 /* Just in case ipsec didn't preserve the NULL b_cont */
2187 if (mp->b_cont != NULL) {
2188 if (!pullupmsg(mp, -1))
2189 goto discard_pkt;
2190 }
2191
2192 /*
2193 * Note that ira_pktlen and ira_ip_hdr_length are no longer
2194 * correct, but we don't use them any more here.
2195 *
2196 * If succesful, the mp has been modified to not include
2197 * the ESP/AH header so we can fanout to the ULP's icmp
2198 * error handler.
2199 */
2200 if (mp->b_wptr - mp->b_rptr < IP_SIMPLE_HDR_LENGTH)
2201 goto truncated;
2202
2203 /* Verify the modified message before any further processes. */
2204 ipha = (ipha_t *)mp->b_rptr;
2205 hdr_length = IPH_HDR_LENGTH(ipha);
2206 icmph = (icmph_t *)&mp->b_rptr[hdr_length];
2207 if (!icmp_inbound_verify_v4(mp, icmph, ira)) {
2208 freemsg(mp);
2209 return;
2210 }
2211
2212 icmp_inbound_error_fanout_v4(mp, icmph, ira);
2213 return;
2214
2215 case IPPROTO_ENCAP: {
2216 /* Look for self-encapsulated packets that caused an error */
2217 ipha_t *in_ipha;
2218
2219 /*
2220 * Caller has verified that length has to be
2221 * at least the size of IP header.
2222 */
2223 ASSERT(hdr_length >= sizeof (ipha_t));
2224 /*
2225 * Check the sanity of the inner IP header like
2226 * we did for the outer header.
2227 */
2228 in_ipha = (ipha_t *)((uchar_t *)ipha + hdr_length);
2229 if ((IPH_HDR_VERSION(in_ipha) != IPV4_VERSION)) {
2230 goto discard_pkt;
2231 }
2232 if (IPH_HDR_LENGTH(in_ipha) < sizeof (ipha_t)) {
2233 goto discard_pkt;
2234 }
2235 /* Check for Self-encapsulated tunnels */
2236 if (in_ipha->ipha_src == ipha->ipha_src &&
2237 in_ipha->ipha_dst == ipha->ipha_dst) {
2238
2239 mp = icmp_inbound_self_encap_error_v4(mp, ipha,
2240 in_ipha);
2241 if (mp == NULL)
2242 goto discard_pkt;
2243
2244 /*
2245 * Just in case self_encap didn't preserve the NULL
2246 * b_cont
2247 */
2248 if (mp->b_cont != NULL) {
2249 if (!pullupmsg(mp, -1))
2250 goto discard_pkt;
2251 }
2252 /*
2253 * Note that ira_pktlen and ira_ip_hdr_length are no
2254 * longer correct, but we don't use them any more here.
2255 */
2256 if (mp->b_wptr - mp->b_rptr < IP_SIMPLE_HDR_LENGTH)
2257 goto truncated;
2258
2259 /*
2260 * Verify the modified message before any further
2261 * processes.
2262 */
2263 ipha = (ipha_t *)mp->b_rptr;
2264 hdr_length = IPH_HDR_LENGTH(ipha);
2265 icmph = (icmph_t *)&mp->b_rptr[hdr_length];
2266 if (!icmp_inbound_verify_v4(mp, icmph, ira)) {
2267 freemsg(mp);
2268 return;
2269 }
2270
2271 /*
2272 * The packet in error is self-encapsualted.
2273 * And we are finding it further encapsulated
2274 * which we could not have possibly generated.
2275 */
2276 if (ipha->ipha_protocol == IPPROTO_ENCAP) {
2277 goto discard_pkt;
2278 }
2279 icmp_inbound_error_fanout_v4(mp, icmph, ira);
2280 return;
2281 }
2282 /* No self-encapsulated */
2283 /* FALLTHRU */
2284 }
2285 case IPPROTO_IPV6:
2286 if ((connp = ipcl_iptun_classify_v4(&ripha.ipha_src,
2287 &ripha.ipha_dst, ipst)) != NULL) {
2288 ira->ira_flags |= IRAF_ICMP_ERROR;
2289 connp->conn_recvicmp(connp, mp, NULL, ira);
2290 CONN_DEC_REF(connp);
2291 ira->ira_flags &= ~IRAF_ICMP_ERROR;
2292 return;
2293 }
2294 /*
2295 * No IP tunnel is interested, fallthrough and see
2296 * if a raw socket will want it.
2297 */
2298 /* FALLTHRU */
2299 default:
2300 ira->ira_flags |= IRAF_ICMP_ERROR;
2301 ip_fanout_proto_v4(mp, &ripha, ira);
2302 ira->ira_flags &= ~IRAF_ICMP_ERROR;
2303 return;
2304 }
2305 /* NOTREACHED */
2306 discard_pkt:
2307 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
2308 ip1dbg(("icmp_inbound_error_fanout_v4: drop pkt\n"));
2309 ip_drop_input("ipIfStatsInDiscards", mp, ill);
2310 freemsg(mp);
2311 return;
2312
2313 truncated:
2314 /* We pulled up everthing already. Must be truncated */
2315 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInTruncatedPkts);
2316 ip_drop_input("ipIfStatsInTruncatedPkts", mp, ill);
2317 freemsg(mp);
2318 }
2319
2320 /*
2321 * Common IP options parser.
2322 *
2323 * Setup routine: fill in *optp with options-parsing state, then
2324 * tail-call ipoptp_next to return the first option.
2325 */
2326 uint8_t
2327 ipoptp_first(ipoptp_t *optp, ipha_t *ipha)
2328 {
2329 uint32_t totallen; /* total length of all options */
2330
2331 totallen = ipha->ipha_version_and_hdr_length -
2332 (uint8_t)((IP_VERSION << 4) + IP_SIMPLE_HDR_LENGTH_IN_WORDS);
2333 totallen <<= 2;
2334 optp->ipoptp_next = (uint8_t *)(&ipha[1]);
2335 optp->ipoptp_end = optp->ipoptp_next + totallen;
2336 optp->ipoptp_flags = 0;
2337 return (ipoptp_next(optp));
2338 }
2339
2340 /* Like above but without an ipha_t */
2341 uint8_t
2342 ipoptp_first2(ipoptp_t *optp, uint32_t totallen, uint8_t *opt)
2343 {
2344 optp->ipoptp_next = opt;
2345 optp->ipoptp_end = optp->ipoptp_next + totallen;
2346 optp->ipoptp_flags = 0;
2347 return (ipoptp_next(optp));
2348 }
2349
2350 /*
2351 * Common IP options parser: extract next option.
2352 */
2353 uint8_t
2354 ipoptp_next(ipoptp_t *optp)
2355 {
2356 uint8_t *end = optp->ipoptp_end;
2357 uint8_t *cur = optp->ipoptp_next;
2358 uint8_t opt, len, pointer;
2359
2360 /*
2361 * If cur > end already, then the ipoptp_end or ipoptp_next pointer
2362 * has been corrupted.
2363 */
2364 ASSERT(cur <= end);
2365
2366 if (cur == end)
2367 return (IPOPT_EOL);
2368
2369 opt = cur[IPOPT_OPTVAL];
2370
2371 /*
2372 * Skip any NOP options.
2373 */
2374 while (opt == IPOPT_NOP) {
2375 cur++;
2376 if (cur == end)
2377 return (IPOPT_EOL);
2378 opt = cur[IPOPT_OPTVAL];
2379 }
2380
2381 if (opt == IPOPT_EOL)
2382 return (IPOPT_EOL);
2383
2384 /*
2385 * Option requiring a length.
2386 */
2387 if ((cur + 1) >= end) {
2388 optp->ipoptp_flags |= IPOPTP_ERROR;
2389 return (IPOPT_EOL);
2390 }
2391 len = cur[IPOPT_OLEN];
2392 if (len < 2) {
2393 optp->ipoptp_flags |= IPOPTP_ERROR;
2394 return (IPOPT_EOL);
2395 }
2396 optp->ipoptp_cur = cur;
2397 optp->ipoptp_len = len;
2398 optp->ipoptp_next = cur + len;
2399 if (cur + len > end) {
2400 optp->ipoptp_flags |= IPOPTP_ERROR;
2401 return (IPOPT_EOL);
2402 }
2403
2404 /*
2405 * For the options which require a pointer field, make sure
2406 * its there, and make sure it points to either something
2407 * inside this option, or the end of the option.
2408 */
2409 switch (opt) {
2410 case IPOPT_RR:
2411 case IPOPT_TS:
2412 case IPOPT_LSRR:
2413 case IPOPT_SSRR:
2414 if (len <= IPOPT_OFFSET) {
2415 optp->ipoptp_flags |= IPOPTP_ERROR;
2416 return (opt);
2417 }
2418 pointer = cur[IPOPT_OFFSET];
2419 if (pointer - 1 > len) {
2420 optp->ipoptp_flags |= IPOPTP_ERROR;
2421 return (opt);
2422 }
2423 break;
2424 }
2425
2426 /*
2427 * Sanity check the pointer field based on the type of the
2428 * option.
2429 */
2430 switch (opt) {
2431 case IPOPT_RR:
2432 case IPOPT_SSRR:
2433 case IPOPT_LSRR:
2434 if (pointer < IPOPT_MINOFF_SR)
2435 optp->ipoptp_flags |= IPOPTP_ERROR;
2436 break;
2437 case IPOPT_TS:
2438 if (pointer < IPOPT_MINOFF_IT)
2439 optp->ipoptp_flags |= IPOPTP_ERROR;
2440 /*
2441 * Note that the Internet Timestamp option also
2442 * contains two four bit fields (the Overflow field,
2443 * and the Flag field), which follow the pointer
2444 * field. We don't need to check that these fields
2445 * fall within the length of the option because this
2446 * was implicitely done above. We've checked that the
2447 * pointer value is at least IPOPT_MINOFF_IT, and that
2448 * it falls within the option. Since IPOPT_MINOFF_IT >
2449 * IPOPT_POS_OV_FLG, we don't need the explicit check.
2450 */
2451 ASSERT(len > IPOPT_POS_OV_FLG);
2452 break;
2453 }
2454
2455 return (opt);
2456 }
2457
2458 /*
2459 * Use the outgoing IP header to create an IP_OPTIONS option the way
2460 * it was passed down from the application.
2461 *
2462 * This is compatible with BSD in that it returns
2463 * the reverse source route with the final destination
2464 * as the last entry. The first 4 bytes of the option
2465 * will contain the final destination.
2466 */
2467 int
2468 ip_opt_get_user(conn_t *connp, uchar_t *buf)
2469 {
2470 ipoptp_t opts;
2471 uchar_t *opt;
2472 uint8_t optval;
2473 uint8_t optlen;
2474 uint32_t len = 0;
2475 uchar_t *buf1 = buf;
2476 uint32_t totallen;
2477 ipaddr_t dst;
2478 ip_pkt_t *ipp = &connp->conn_xmit_ipp;
2479
2480 if (!(ipp->ipp_fields & IPPF_IPV4_OPTIONS))
2481 return (0);
2482
2483 totallen = ipp->ipp_ipv4_options_len;
2484 if (totallen & 0x3)
2485 return (0);
2486
2487 buf += IP_ADDR_LEN; /* Leave room for final destination */
2488 len += IP_ADDR_LEN;
2489 bzero(buf1, IP_ADDR_LEN);
2490
2491 dst = connp->conn_faddr_v4;
2492
2493 for (optval = ipoptp_first2(&opts, totallen, ipp->ipp_ipv4_options);
2494 optval != IPOPT_EOL;
2495 optval = ipoptp_next(&opts)) {
2496 int off;
2497
2498 opt = opts.ipoptp_cur;
2499 if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
2500 break;
2501 }
2502 optlen = opts.ipoptp_len;
2503
2504 switch (optval) {
2505 case IPOPT_SSRR:
2506 case IPOPT_LSRR:
2507
2508 /*
2509 * Insert destination as the first entry in the source
2510 * route and move down the entries on step.
2511 * The last entry gets placed at buf1.
2512 */
2513 buf[IPOPT_OPTVAL] = optval;
2514 buf[IPOPT_OLEN] = optlen;
2515 buf[IPOPT_OFFSET] = optlen;
2516
2517 off = optlen - IP_ADDR_LEN;
2518 if (off < 0) {
2519 /* No entries in source route */
2520 break;
2521 }
2522 /* Last entry in source route if not already set */
2523 if (dst == INADDR_ANY)
2524 bcopy(opt + off, buf1, IP_ADDR_LEN);
2525 off -= IP_ADDR_LEN;
2526
2527 while (off > 0) {
2528 bcopy(opt + off,
2529 buf + off + IP_ADDR_LEN,
2530 IP_ADDR_LEN);
2531 off -= IP_ADDR_LEN;
2532 }
2533 /* ipha_dst into first slot */
2534 bcopy(&dst, buf + off + IP_ADDR_LEN,
2535 IP_ADDR_LEN);
2536 buf += optlen;
2537 len += optlen;
2538 break;
2539
2540 default:
2541 bcopy(opt, buf, optlen);
2542 buf += optlen;
2543 len += optlen;
2544 break;
2545 }
2546 }
2547 done:
2548 /* Pad the resulting options */
2549 while (len & 0x3) {
2550 *buf++ = IPOPT_EOL;
2551 len++;
2552 }
2553 return (len);
2554 }
2555
2556 /*
2557 * Update any record route or timestamp options to include this host.
2558 * Reverse any source route option.
2559 * This routine assumes that the options are well formed i.e. that they
2560 * have already been checked.
2561 */
2562 static void
2563 icmp_options_update(ipha_t *ipha)
2564 {
2565 ipoptp_t opts;
2566 uchar_t *opt;
2567 uint8_t optval;
2568 ipaddr_t src; /* Our local address */
2569 ipaddr_t dst;
2570
2571 ip2dbg(("icmp_options_update\n"));
2572 src = ipha->ipha_src;
2573 dst = ipha->ipha_dst;
2574
2575 for (optval = ipoptp_first(&opts, ipha);
2576 optval != IPOPT_EOL;
2577 optval = ipoptp_next(&opts)) {
2578 ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
2579 opt = opts.ipoptp_cur;
2580 ip2dbg(("icmp_options_update: opt %d, len %d\n",
2581 optval, opts.ipoptp_len));
2582 switch (optval) {
2583 int off1, off2;
2584 case IPOPT_SSRR:
2585 case IPOPT_LSRR:
2586 /*
2587 * Reverse the source route. The first entry
2588 * should be the next to last one in the current
2589 * source route (the last entry is our address).
2590 * The last entry should be the final destination.
2591 */
2592 off1 = IPOPT_MINOFF_SR - 1;
2593 off2 = opt[IPOPT_OFFSET] - IP_ADDR_LEN - 1;
2594 if (off2 < 0) {
2595 /* No entries in source route */
2596 ip1dbg((
2597 "icmp_options_update: bad src route\n"));
2598 break;
2599 }
2600 bcopy((char *)opt + off2, &dst, IP_ADDR_LEN);
2601 bcopy(&ipha->ipha_dst, (char *)opt + off2, IP_ADDR_LEN);
2602 bcopy(&dst, &ipha->ipha_dst, IP_ADDR_LEN);
2603 off2 -= IP_ADDR_LEN;
2604
2605 while (off1 < off2) {
2606 bcopy((char *)opt + off1, &src, IP_ADDR_LEN);
2607 bcopy((char *)opt + off2, (char *)opt + off1,
2608 IP_ADDR_LEN);
2609 bcopy(&src, (char *)opt + off2, IP_ADDR_LEN);
2610 off1 += IP_ADDR_LEN;
2611 off2 -= IP_ADDR_LEN;
2612 }
2613 opt[IPOPT_OFFSET] = IPOPT_MINOFF_SR;
2614 break;
2615 }
2616 }
2617 }
2618
2619 /*
2620 * Process received ICMP Redirect messages.
2621 * Assumes the caller has verified that the headers are in the pulled up mblk.
2622 * Consumes mp.
2623 */
2624 static void
2625 icmp_redirect_v4(mblk_t *mp, ipha_t *ipha, icmph_t *icmph, ip_recv_attr_t *ira)
2626 {
2627 ire_t *ire, *nire;
2628 ire_t *prev_ire;
2629 ipaddr_t src, dst, gateway;
2630 ip_stack_t *ipst = ira->ira_ill->ill_ipst;
2631 ipha_t *inner_ipha; /* Inner IP header */
2632
2633 /* Caller already pulled up everything. */
2634 inner_ipha = (ipha_t *)&icmph[1];
2635 src = ipha->ipha_src;
2636 dst = inner_ipha->ipha_dst;
2637 gateway = icmph->icmph_rd_gateway;
2638 /* Make sure the new gateway is reachable somehow. */
2639 ire = ire_ftable_lookup_v4(gateway, 0, 0, IRE_ONLINK, NULL,
2640 ALL_ZONES, NULL, MATCH_IRE_TYPE, 0, ipst, NULL);
2641 /*
2642 * Make sure we had a route for the dest in question and that
2643 * that route was pointing to the old gateway (the source of the
2644 * redirect packet.)
2645 * We do longest match and then compare ire_gateway_addr below.
2646 */
2647 prev_ire = ire_ftable_lookup_v4(dst, 0, 0, 0, NULL, ALL_ZONES,
2648 NULL, MATCH_IRE_DSTONLY, 0, ipst, NULL);
2649 /*
2650 * Check that
2651 * the redirect was not from ourselves
2652 * the new gateway and the old gateway are directly reachable
2653 */
2654 if (prev_ire == NULL || ire == NULL ||
2655 (prev_ire->ire_type & (IRE_LOCAL|IRE_LOOPBACK)) ||
2656 (prev_ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE)) ||
2657 !(ire->ire_type & IRE_IF_ALL) ||
2658 prev_ire->ire_gateway_addr != src) {
2659 BUMP_MIB(&ipst->ips_icmp_mib, icmpInBadRedirects);
2660 ip_drop_input("icmpInBadRedirects - ire", mp, ira->ira_ill);
2661 freemsg(mp);
2662 if (ire != NULL)
2663 ire_refrele(ire);
2664 if (prev_ire != NULL)
2665 ire_refrele(prev_ire);
2666 return;
2667 }
2668
2669 ire_refrele(prev_ire);
2670 ire_refrele(ire);
2671
2672 /*
2673 * TODO: more precise handling for cases 0, 2, 3, the latter two
2674 * require TOS routing
2675 */
2676 switch (icmph->icmph_code) {
2677 case 0:
2678 case 1:
2679 /* TODO: TOS specificity for cases 2 and 3 */
2680 case 2:
2681 case 3:
2682 break;
2683 default:
2684 BUMP_MIB(&ipst->ips_icmp_mib, icmpInBadRedirects);
2685 ip_drop_input("icmpInBadRedirects - code", mp, ira->ira_ill);
2686 freemsg(mp);
2687 return;
2688 }
2689 /*
2690 * Create a Route Association. This will allow us to remember that
2691 * someone we believe told us to use the particular gateway.
2692 */
2693 ire = ire_create(
2694 (uchar_t *)&dst, /* dest addr */
2695 (uchar_t *)&ip_g_all_ones, /* mask */
2696 (uchar_t *)&gateway, /* gateway addr */
2697 IRE_HOST,
2698 NULL, /* ill */
2699 ALL_ZONES,
2700 (RTF_DYNAMIC | RTF_GATEWAY | RTF_HOST),
2701 NULL, /* tsol_gc_t */
2702 ipst);
2703
2704 if (ire == NULL) {
2705 freemsg(mp);
2706 return;
2707 }
2708 nire = ire_add(ire);
2709 /* Check if it was a duplicate entry */
2710 if (nire != NULL && nire != ire) {
2711 ASSERT(nire->ire_identical_ref > 1);
2712 ire_delete(nire);
2713 ire_refrele(nire);
2714 nire = NULL;
2715 }
2716 ire = nire;
2717 if (ire != NULL) {
2718 ire_refrele(ire); /* Held in ire_add */
2719
2720 /* tell routing sockets that we received a redirect */
2721 ip_rts_change(RTM_REDIRECT, dst, gateway, IP_HOST_MASK, 0, src,
2722 (RTF_DYNAMIC | RTF_GATEWAY | RTF_HOST), 0,
2723 (RTA_DST | RTA_GATEWAY | RTA_NETMASK | RTA_AUTHOR), ipst);
2724 }
2725
2726 /*
2727 * Delete any existing IRE_HOST type redirect ires for this destination.
2728 * This together with the added IRE has the effect of
2729 * modifying an existing redirect.
2730 */
2731 prev_ire = ire_ftable_lookup_v4(dst, 0, src, IRE_HOST, NULL,
2732 ALL_ZONES, NULL, (MATCH_IRE_GW | MATCH_IRE_TYPE), 0, ipst, NULL);
2733 if (prev_ire != NULL) {
2734 if (prev_ire ->ire_flags & RTF_DYNAMIC)
2735 ire_delete(prev_ire);
2736 ire_refrele(prev_ire);
2737 }
2738
2739 freemsg(mp);
2740 }
2741
2742 /*
2743 * Generate an ICMP parameter problem message.
2744 * When called from ip_output side a minimal ip_recv_attr_t needs to be
2745 * constructed by the caller.
2746 */
2747 static void
2748 icmp_param_problem(mblk_t *mp, uint8_t ptr, ip_recv_attr_t *ira)
2749 {
2750 icmph_t icmph;
2751 ip_stack_t *ipst = ira->ira_ill->ill_ipst;
2752
2753 mp = icmp_pkt_err_ok(mp, ira);
2754 if (mp == NULL)
2755 return;
2756
2757 bzero(&icmph, sizeof (icmph_t));
2758 icmph.icmph_type = ICMP_PARAM_PROBLEM;
2759 icmph.icmph_pp_ptr = ptr;
2760 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutParmProbs);
2761 icmp_pkt(mp, &icmph, sizeof (icmph_t), ira);
2762 }
2763
2764 /*
2765 * Build and ship an IPv4 ICMP message using the packet data in mp, and
2766 * the ICMP header pointed to by "stuff". (May be called as writer.)
2767 * Note: assumes that icmp_pkt_err_ok has been called to verify that
2768 * an icmp error packet can be sent.
2769 * Assigns an appropriate source address to the packet. If ipha_dst is
2770 * one of our addresses use it for source. Otherwise let ip_output_simple
2771 * pick the source address.
2772 */
2773 static void
2774 icmp_pkt(mblk_t *mp, void *stuff, size_t len, ip_recv_attr_t *ira)
2775 {
2776 ipaddr_t dst;
2777 icmph_t *icmph;
2778 ipha_t *ipha;
2779 uint_t len_needed;
2780 size_t msg_len;
2781 mblk_t *mp1;
2782 ipaddr_t src;
2783 ire_t *ire;
2784 ip_xmit_attr_t ixas;
2785 ip_stack_t *ipst = ira->ira_ill->ill_ipst;
2786
2787 ipha = (ipha_t *)mp->b_rptr;
2788
2789 bzero(&ixas, sizeof (ixas));
2790 ixas.ixa_flags = IXAF_BASIC_SIMPLE_V4;
2791 ixas.ixa_zoneid = ira->ira_zoneid;
2792 ixas.ixa_ifindex = 0;
2793 ixas.ixa_ipst = ipst;
2794 ixas.ixa_cred = kcred;
2795 ixas.ixa_cpid = NOPID;
2796 ixas.ixa_tsl = ira->ira_tsl; /* Behave as a multi-level responder */
2797 ixas.ixa_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
2798
2799 if (ira->ira_flags & IRAF_IPSEC_SECURE) {
2800 /*
2801 * Apply IPsec based on how IPsec was applied to
2802 * the packet that had the error.
2803 *
2804 * If it was an outbound packet that caused the ICMP
2805 * error, then the caller will have setup the IRA
2806 * appropriately.
2807 */
2808 if (!ipsec_in_to_out(ira, &ixas, mp, ipha, NULL)) {
2809 BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsOutDiscards);
2810 /* Note: mp already consumed and ip_drop_packet done */
2811 return;
2812 }
2813 } else {
2814 /*
2815 * This is in clear. The icmp message we are building
2816 * here should go out in clear, independent of our policy.
2817 */
2818 ixas.ixa_flags |= IXAF_NO_IPSEC;
2819 }
2820
2821 /* Remember our eventual destination */
2822 dst = ipha->ipha_src;
2823
2824 /*
2825 * If the packet was for one of our unicast addresses, make
2826 * sure we respond with that as the source. Otherwise
2827 * have ip_output_simple pick the source address.
2828 */
2829 ire = ire_ftable_lookup_v4(ipha->ipha_dst, 0, 0,
2830 (IRE_LOCAL|IRE_LOOPBACK), NULL, ira->ira_zoneid, NULL,
2831 MATCH_IRE_TYPE|MATCH_IRE_ZONEONLY, 0, ipst, NULL);
2832 if (ire != NULL) {
2833 ire_refrele(ire);
2834 src = ipha->ipha_dst;
2835 } else {
2836 src = INADDR_ANY;
2837 ixas.ixa_flags |= IXAF_SET_SOURCE;
2838 }
2839
2840 /*
2841 * Check if we can send back more then 8 bytes in addition to
2842 * the IP header. We try to send 64 bytes of data and the internal
2843 * header in the special cases of ipv4 encapsulated ipv4 or ipv6.
2844 */
2845 len_needed = IPH_HDR_LENGTH(ipha);
2846 if (ipha->ipha_protocol == IPPROTO_ENCAP ||
2847 ipha->ipha_protocol == IPPROTO_IPV6) {
2848 if (!pullupmsg(mp, -1)) {
2849 BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsOutDiscards);
2850 ip_drop_output("ipIfStatsOutDiscards", mp, NULL);
2851 freemsg(mp);
2852 return;
2853 }
2854 ipha = (ipha_t *)mp->b_rptr;
2855
2856 if (ipha->ipha_protocol == IPPROTO_ENCAP) {
2857 len_needed += IPH_HDR_LENGTH(((uchar_t *)ipha +
2858 len_needed));
2859 } else {
2860 ip6_t *ip6h = (ip6_t *)((uchar_t *)ipha + len_needed);
2861
2862 ASSERT(ipha->ipha_protocol == IPPROTO_IPV6);
2863 len_needed += ip_hdr_length_v6(mp, ip6h);
2864 }
2865 }
2866 len_needed += ipst->ips_ip_icmp_return;
2867 msg_len = msgdsize(mp);
2868 if (msg_len > len_needed) {
2869 (void) adjmsg(mp, len_needed - msg_len);
2870 msg_len = len_needed;
2871 }
2872 mp1 = allocb(sizeof (icmp_ipha) + len, BPRI_MED);
2873 if (mp1 == NULL) {
2874 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutErrors);
2875 freemsg(mp);
2876 return;
2877 }
2878 mp1->b_cont = mp;
2879 mp = mp1;
2880
2881 /*
2882 * Set IXAF_TRUSTED_ICMP so we can let the ICMP messages this
2883 * node generates be accepted in peace by all on-host destinations.
2884 * If we do NOT assume that all on-host destinations trust
2885 * self-generated ICMP messages, then rework here, ip6.c, and spd.c.
2886 * (Look for IXAF_TRUSTED_ICMP).
2887 */
2888 ixas.ixa_flags |= IXAF_TRUSTED_ICMP;
2889
2890 ipha = (ipha_t *)mp->b_rptr;
2891 mp1->b_wptr = (uchar_t *)ipha + (sizeof (icmp_ipha) + len);
2892 *ipha = icmp_ipha;
2893 ipha->ipha_src = src;
2894 ipha->ipha_dst = dst;
2895 ipha->ipha_ttl = ipst->ips_ip_def_ttl;
2896 msg_len += sizeof (icmp_ipha) + len;
2897 if (msg_len > IP_MAXPACKET) {
2898 (void) adjmsg(mp, IP_MAXPACKET - msg_len);
2899 msg_len = IP_MAXPACKET;
2900 }
2901 ipha->ipha_length = htons((uint16_t)msg_len);
2902 icmph = (icmph_t *)&ipha[1];
2903 bcopy(stuff, icmph, len);
2904 icmph->icmph_checksum = 0;
2905 icmph->icmph_checksum = IP_CSUM(mp, (int32_t)sizeof (ipha_t), 0);
2906 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutMsgs);
2907
2908 (void) ip_output_simple(mp, &ixas);
2909 ixa_cleanup(&ixas);
2910 }
2911
2912 /*
2913 * Determine if an ICMP error packet can be sent given the rate limit.
2914 * The limit consists of an average frequency (icmp_pkt_err_interval measured
2915 * in milliseconds) and a burst size. Burst size number of packets can
2916 * be sent arbitrarely closely spaced.
2917 * The state is tracked using two variables to implement an approximate
2918 * token bucket filter:
2919 * icmp_pkt_err_last - lbolt value when the last burst started
2920 * icmp_pkt_err_sent - number of packets sent in current burst
2921 */
2922 boolean_t
2923 icmp_err_rate_limit(ip_stack_t *ipst)
2924 {
2925 clock_t now = TICK_TO_MSEC(ddi_get_lbolt());
2926 uint_t refilled; /* Number of packets refilled in tbf since last */
2927 /* Guard against changes by loading into local variable */
2928 uint_t err_interval = ipst->ips_ip_icmp_err_interval;
2929
2930 if (err_interval == 0)
2931 return (B_FALSE);
2932
2933 if (ipst->ips_icmp_pkt_err_last > now) {
2934 /* 100HZ lbolt in ms for 32bit arch wraps every 49.7 days */
2935 ipst->ips_icmp_pkt_err_last = 0;
2936 ipst->ips_icmp_pkt_err_sent = 0;
2937 }
2938 /*
2939 * If we are in a burst update the token bucket filter.
2940 * Update the "last" time to be close to "now" but make sure
2941 * we don't loose precision.
2942 */
2943 if (ipst->ips_icmp_pkt_err_sent != 0) {
2944 refilled = (now - ipst->ips_icmp_pkt_err_last)/err_interval;
2945 if (refilled > ipst->ips_icmp_pkt_err_sent) {
2946 ipst->ips_icmp_pkt_err_sent = 0;
2947 } else {
2948 ipst->ips_icmp_pkt_err_sent -= refilled;
2949 ipst->ips_icmp_pkt_err_last += refilled * err_interval;
2950 }
2951 }
2952 if (ipst->ips_icmp_pkt_err_sent == 0) {
2953 /* Start of new burst */
2954 ipst->ips_icmp_pkt_err_last = now;
2955 }
2956 if (ipst->ips_icmp_pkt_err_sent < ipst->ips_ip_icmp_err_burst) {
2957 ipst->ips_icmp_pkt_err_sent++;
2958 ip1dbg(("icmp_err_rate_limit: %d sent in burst\n",
2959 ipst->ips_icmp_pkt_err_sent));
2960 return (B_FALSE);
2961 }
2962 ip1dbg(("icmp_err_rate_limit: dropped\n"));
2963 return (B_TRUE);
2964 }
2965
2966 /*
2967 * Check if it is ok to send an IPv4 ICMP error packet in
2968 * response to the IPv4 packet in mp.
2969 * Free the message and return null if no
2970 * ICMP error packet should be sent.
2971 */
2972 static mblk_t *
2973 icmp_pkt_err_ok(mblk_t *mp, ip_recv_attr_t *ira)
2974 {
2975 ip_stack_t *ipst = ira->ira_ill->ill_ipst;
2976 icmph_t *icmph;
2977 ipha_t *ipha;
2978 uint_t len_needed;
2979
2980 if (!mp)
2981 return (NULL);
2982 ipha = (ipha_t *)mp->b_rptr;
2983 if (ip_csum_hdr(ipha)) {
2984 BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsInCksumErrs);
2985 ip_drop_input("ipIfStatsInCksumErrs", mp, NULL);
2986 freemsg(mp);
2987 return (NULL);
2988 }
2989 if (ip_type_v4(ipha->ipha_dst, ipst) == IRE_BROADCAST ||
2990 ip_type_v4(ipha->ipha_src, ipst) == IRE_BROADCAST ||
2991 CLASSD(ipha->ipha_dst) ||
2992 CLASSD(ipha->ipha_src) ||
2993 (ntohs(ipha->ipha_fragment_offset_and_flags) & IPH_OFFSET)) {
2994 /* Note: only errors to the fragment with offset 0 */
2995 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
2996 freemsg(mp);
2997 return (NULL);
2998 }
2999 if (ipha->ipha_protocol == IPPROTO_ICMP) {
3000 /*
3001 * Check the ICMP type. RFC 1122 sez: don't send ICMP
3002 * errors in response to any ICMP errors.
3003 */
3004 len_needed = IPH_HDR_LENGTH(ipha) + ICMPH_SIZE;
3005 if (mp->b_wptr - mp->b_rptr < len_needed) {
3006 if (!pullupmsg(mp, len_needed)) {
3007 BUMP_MIB(&ipst->ips_icmp_mib, icmpInErrors);
3008 freemsg(mp);
3009 return (NULL);
3010 }
3011 ipha = (ipha_t *)mp->b_rptr;
3012 }
3013 icmph = (icmph_t *)
3014 (&((char *)ipha)[IPH_HDR_LENGTH(ipha)]);
3015 switch (icmph->icmph_type) {
3016 case ICMP_DEST_UNREACHABLE:
3017 case ICMP_SOURCE_QUENCH:
3018 case ICMP_TIME_EXCEEDED:
3019 case ICMP_PARAM_PROBLEM:
3020 case ICMP_REDIRECT:
3021 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
3022 freemsg(mp);
3023 return (NULL);
3024 default:
3025 break;
3026 }
3027 }
3028 /*
3029 * If this is a labeled system, then check to see if we're allowed to
3030 * send a response to this particular sender. If not, then just drop.
3031 */
3032 if (is_system_labeled() && !tsol_can_reply_error(mp, ira)) {
3033 ip2dbg(("icmp_pkt_err_ok: can't respond to packet\n"));
3034 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
3035 freemsg(mp);
3036 return (NULL);
3037 }
3038 if (icmp_err_rate_limit(ipst)) {
3039 /*
3040 * Only send ICMP error packets every so often.
3041 * This should be done on a per port/source basis,
3042 * but for now this will suffice.
3043 */
3044 freemsg(mp);
3045 return (NULL);
3046 }
3047 return (mp);
3048 }
3049
3050 /*
3051 * Called when a packet was sent out the same link that it arrived on.
3052 * Check if it is ok to send a redirect and then send it.
3053 */
3054 void
3055 ip_send_potential_redirect_v4(mblk_t *mp, ipha_t *ipha, ire_t *ire,
3056 ip_recv_attr_t *ira)
3057 {
3058 ip_stack_t *ipst = ira->ira_ill->ill_ipst;
3059 ipaddr_t src, nhop;
3060 mblk_t *mp1;
3061 ire_t *nhop_ire;
3062
3063 /*
3064 * Check the source address to see if it originated
3065 * on the same logical subnet it is going back out on.
3066 * If so, we should be able to send it a redirect.
3067 * Avoid sending a redirect if the destination
3068 * is directly connected (i.e., we matched an IRE_ONLINK),
3069 * or if the packet was source routed out this interface.
3070 *
3071 * We avoid sending a redirect if the
3072 * destination is directly connected
3073 * because it is possible that multiple
3074 * IP subnets may have been configured on
3075 * the link, and the source may not
3076 * be on the same subnet as ip destination,
3077 * even though they are on the same
3078 * physical link.
3079 */
3080 if ((ire->ire_type & IRE_ONLINK) ||
3081 ip_source_routed(ipha, ipst))
3082 return;
3083
3084 nhop_ire = ire_nexthop(ire);
3085 if (nhop_ire == NULL)
3086 return;
3087
3088 nhop = nhop_ire->ire_addr;
3089
3090 if (nhop_ire->ire_type & IRE_IF_CLONE) {
3091 ire_t *ire2;
3092
3093 /* Follow ire_dep_parent to find non-clone IRE_INTERFACE */
3094 mutex_enter(&nhop_ire->ire_lock);
3095 ire2 = nhop_ire->ire_dep_parent;
3096 if (ire2 != NULL)
3097 ire_refhold(ire2);
3098 mutex_exit(&nhop_ire->ire_lock);
3099 ire_refrele(nhop_ire);
3100 nhop_ire = ire2;
3101 }
3102 if (nhop_ire == NULL)
3103 return;
3104
3105 ASSERT(!(nhop_ire->ire_type & IRE_IF_CLONE));
3106
3107 src = ipha->ipha_src;
3108
3109 /*
3110 * We look at the interface ire for the nexthop,
3111 * to see if ipha_src is in the same subnet
3112 * as the nexthop.
3113 */
3114 if ((src & nhop_ire->ire_mask) == (nhop & nhop_ire->ire_mask)) {
3115 /*
3116 * The source is directly connected.
3117 */
3118 mp1 = copymsg(mp);
3119 if (mp1 != NULL) {
3120 icmp_send_redirect(mp1, nhop, ira);
3121 }
3122 }
3123 ire_refrele(nhop_ire);
3124 }
3125
3126 /*
3127 * Generate an ICMP redirect message.
3128 */
3129 static void
3130 icmp_send_redirect(mblk_t *mp, ipaddr_t gateway, ip_recv_attr_t *ira)
3131 {
3132 icmph_t icmph;
3133 ip_stack_t *ipst = ira->ira_ill->ill_ipst;
3134
3135 mp = icmp_pkt_err_ok(mp, ira);
3136 if (mp == NULL)
3137 return;
3138
3139 bzero(&icmph, sizeof (icmph_t));
3140 icmph.icmph_type = ICMP_REDIRECT;
3141 icmph.icmph_code = 1;
3142 icmph.icmph_rd_gateway = gateway;
3143 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutRedirects);
3144 icmp_pkt(mp, &icmph, sizeof (icmph_t), ira);
3145 }
3146
3147 /*
3148 * Generate an ICMP time exceeded message.
3149 */
3150 void
3151 icmp_time_exceeded(mblk_t *mp, uint8_t code, ip_recv_attr_t *ira)
3152 {
3153 icmph_t icmph;
3154 ip_stack_t *ipst = ira->ira_ill->ill_ipst;
3155
3156 mp = icmp_pkt_err_ok(mp, ira);
3157 if (mp == NULL)
3158 return;
3159
3160 bzero(&icmph, sizeof (icmph_t));
3161 icmph.icmph_type = ICMP_TIME_EXCEEDED;
3162 icmph.icmph_code = code;
3163 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutTimeExcds);
3164 icmp_pkt(mp, &icmph, sizeof (icmph_t), ira);
3165 }
3166
3167 /*
3168 * Generate an ICMP unreachable message.
3169 * When called from ip_output side a minimal ip_recv_attr_t needs to be
3170 * constructed by the caller.
3171 */
3172 void
3173 icmp_unreachable(mblk_t *mp, uint8_t code, ip_recv_attr_t *ira)
3174 {
3175 icmph_t icmph;
3176 ip_stack_t *ipst = ira->ira_ill->ill_ipst;
3177
3178 mp = icmp_pkt_err_ok(mp, ira);
3179 if (mp == NULL)
3180 return;
3181
3182 bzero(&icmph, sizeof (icmph_t));
3183 icmph.icmph_type = ICMP_DEST_UNREACHABLE;
3184 icmph.icmph_code = code;
3185 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDestUnreachs);
3186 icmp_pkt(mp, &icmph, sizeof (icmph_t), ira);
3187 }
3188
3189 /*
3190 * Latch in the IPsec state for a stream based the policy in the listener
3191 * and the actions in the ip_recv_attr_t.
3192 * Called directly from TCP and SCTP.
3193 */
3194 boolean_t
3195 ip_ipsec_policy_inherit(conn_t *connp, conn_t *lconnp, ip_recv_attr_t *ira)
3196 {
3197 ASSERT(lconnp->conn_policy != NULL);
3198 ASSERT(connp->conn_policy == NULL);
3199
3200 IPPH_REFHOLD(lconnp->conn_policy);
3201 connp->conn_policy = lconnp->conn_policy;
3202
3203 if (ira->ira_ipsec_action != NULL) {
3204 if (connp->conn_latch == NULL) {
3205 connp->conn_latch = iplatch_create();
3206 if (connp->conn_latch == NULL)
3207 return (B_FALSE);
3208 }
3209 ipsec_latch_inbound(connp, ira);
3210 }
3211 return (B_TRUE);
3212 }
3213
3214 /*
3215 * Verify whether or not the IP address is a valid local address.
3216 * Could be a unicast, including one for a down interface.
3217 * If allow_mcbc then a multicast or broadcast address is also
3218 * acceptable.
3219 *
3220 * In the case of a broadcast/multicast address, however, the
3221 * upper protocol is expected to reset the src address
3222 * to zero when we return IPVL_MCAST/IPVL_BCAST so that
3223 * no packets are emitted with broadcast/multicast address as
3224 * source address (that violates hosts requirements RFC 1122)
3225 * The addresses valid for bind are:
3226 * (1) - INADDR_ANY (0)
3227 * (2) - IP address of an UP interface
3228 * (3) - IP address of a DOWN interface
3229 * (4) - valid local IP broadcast addresses. In this case
3230 * the conn will only receive packets destined to
3231 * the specified broadcast address.
3232 * (5) - a multicast address. In this case
3233 * the conn will only receive packets destined to
3234 * the specified multicast address. Note: the
3235 * application still has to issue an
3236 * IP_ADD_MEMBERSHIP socket option.
3237 *
3238 * In all the above cases, the bound address must be valid in the current zone.
3239 * When the address is loopback, multicast or broadcast, there might be many
3240 * matching IREs so bind has to look up based on the zone.
3241 */
3242 ip_laddr_t
3243 ip_laddr_verify_v4(ipaddr_t src_addr, zoneid_t zoneid,
3244 ip_stack_t *ipst, boolean_t allow_mcbc)
3245 {
3246 ire_t *src_ire;
3247
3248 ASSERT(src_addr != INADDR_ANY);
3249
3250 src_ire = ire_ftable_lookup_v4(src_addr, 0, 0, 0,
3251 NULL, zoneid, NULL, MATCH_IRE_ZONEONLY, 0, ipst, NULL);
3252
3253 /*
3254 * If an address other than in6addr_any is requested,
3255 * we verify that it is a valid address for bind
3256 * Note: Following code is in if-else-if form for
3257 * readability compared to a condition check.
3258 */
3259 if (src_ire != NULL && (src_ire->ire_type & (IRE_LOCAL|IRE_LOOPBACK))) {
3260 /*
3261 * (2) Bind to address of local UP interface
3262 */
3263 ire_refrele(src_ire);
3264 return (IPVL_UNICAST_UP);
3265 } else if (src_ire != NULL && src_ire->ire_type & IRE_BROADCAST) {
3266 /*
3267 * (4) Bind to broadcast address
3268 */
3269 ire_refrele(src_ire);
3270 if (allow_mcbc)
3271 return (IPVL_BCAST);
3272 else
3273 return (IPVL_BAD);
3274 } else if (CLASSD(src_addr)) {
3275 /* (5) bind to multicast address. */
3276 if (src_ire != NULL)
3277 ire_refrele(src_ire);
3278
3279 if (allow_mcbc)
3280 return (IPVL_MCAST);
3281 else
3282 return (IPVL_BAD);
3283 } else {
3284 ipif_t *ipif;
3285
3286 /*
3287 * (3) Bind to address of local DOWN interface?
3288 * (ipif_lookup_addr() looks up all interfaces
3289 * but we do not get here for UP interfaces
3290 * - case (2) above)
3291 */
3292 if (src_ire != NULL)
3293 ire_refrele(src_ire);
3294
3295 ipif = ipif_lookup_addr(src_addr, NULL, zoneid, ipst);
3296 if (ipif == NULL)
3297 return (IPVL_BAD);
3298
3299 /* Not a useful source? */
3300 if (ipif->ipif_flags & (IPIF_NOLOCAL | IPIF_ANYCAST)) {
3301 ipif_refrele(ipif);
3302 return (IPVL_BAD);
3303 }
3304 ipif_refrele(ipif);
3305 return (IPVL_UNICAST_DOWN);
3306 }
3307 }
3308
3309 /*
3310 * Insert in the bind fanout for IPv4 and IPv6.
3311 * The caller should already have used ip_laddr_verify_v*() before calling
3312 * this.
3313 */
3314 int
3315 ip_laddr_fanout_insert(conn_t *connp)
3316 {
3317 int error;
3318
3319 /*
3320 * Allow setting new policies. For example, disconnects result
3321 * in us being called. As we would have set conn_policy_cached
3322 * to B_TRUE before, we should set it to B_FALSE, so that policy
3323 * can change after the disconnect.
3324 */
3325 connp->conn_policy_cached = B_FALSE;
3326
3327 error = ipcl_bind_insert(connp);
3328 if (error != 0) {
3329 if (connp->conn_anon_port) {
3330 (void) tsol_mlp_anon(crgetzone(connp->conn_cred),
3331 connp->conn_mlp_type, connp->conn_proto,
3332 ntohs(connp->conn_lport), B_FALSE);
3333 }
3334 connp->conn_mlp_type = mlptSingle;
3335 }
3336 return (error);
3337 }
3338
3339 /*
3340 * Verify that both the source and destination addresses are valid. If
3341 * IPDF_VERIFY_DST is not set, then the destination address may be unreachable,
3342 * i.e. have no route to it. Protocols like TCP want to verify destination
3343 * reachability, while tunnels do not.
3344 *
3345 * Determine the route, the interface, and (optionally) the source address
3346 * to use to reach a given destination.
3347 * Note that we allow connect to broadcast and multicast addresses when
3348 * IPDF_ALLOW_MCBC is set.
3349 * first_hop and dst_addr are normally the same, but if source routing
3350 * they will differ; in that case the first_hop is what we'll use for the
3351 * routing lookup but the dce and label checks will be done on dst_addr,
3352 *
3353 * If uinfo is set, then we fill in the best available information
3354 * we have for the destination. This is based on (in priority order) any
3355 * metrics and path MTU stored in a dce_t, route metrics, and finally the
3356 * ill_mtu/ill_mc_mtu.
3357 *
3358 * Tsol note: If we have a source route then dst_addr != firsthop. But we
3359 * always do the label check on dst_addr.
3360 */
3361 int
3362 ip_set_destination_v4(ipaddr_t *src_addrp, ipaddr_t dst_addr, ipaddr_t firsthop,
3363 ip_xmit_attr_t *ixa, iulp_t *uinfo, uint32_t flags, uint_t mac_mode)
3364 {
3365 ire_t *ire = NULL;
3366 int error = 0;
3367 ipaddr_t setsrc; /* RTF_SETSRC */
3368 zoneid_t zoneid = ixa->ixa_zoneid; /* Honors SO_ALLZONES */
3369 ip_stack_t *ipst = ixa->ixa_ipst;
3370 dce_t *dce;
3371 uint_t pmtu;
3372 uint_t generation;
3373 nce_t *nce;
3374 ill_t *ill = NULL;
3375 boolean_t multirt = B_FALSE;
3376
3377 ASSERT(ixa->ixa_flags & IXAF_IS_IPV4);
3378
3379 /*
3380 * We never send to zero; the ULPs map it to the loopback address.
3381 * We can't allow it since we use zero to mean unitialized in some
3382 * places.
3383 */
3384 ASSERT(dst_addr != INADDR_ANY);
3385
3386 if (is_system_labeled()) {
3387 ts_label_t *tsl = NULL;
3388
3389 error = tsol_check_dest(ixa->ixa_tsl, &dst_addr, IPV4_VERSION,
3390 mac_mode, (flags & IPDF_ZONE_IS_GLOBAL) != 0, &tsl);
3391 if (error != 0)
3392 return (error);
3393 if (tsl != NULL) {
3394 /* Update the label */
3395 ip_xmit_attr_replace_tsl(ixa, tsl);
3396 }
3397 }
3398
3399 setsrc = INADDR_ANY;
3400 /*
3401 * Select a route; For IPMP interfaces, we would only select
3402 * a "hidden" route (i.e., going through a specific under_ill)
3403 * if ixa_ifindex has been specified.
3404 */
3405 ire = ip_select_route_v4(firsthop, *src_addrp, ixa,
3406 &generation, &setsrc, &error, &multirt);
3407 ASSERT(ire != NULL); /* IRE_NOROUTE if none found */
3408 if (error != 0)
3409 goto bad_addr;
3410
3411 /*
3412 * ire can't be a broadcast or multicast unless IPDF_ALLOW_MCBC is set.
3413 * If IPDF_VERIFY_DST is set, the destination must be reachable;
3414 * Otherwise the destination needn't be reachable.
3415 *
3416 * If we match on a reject or black hole, then we've got a
3417 * local failure. May as well fail out the connect() attempt,
3418 * since it's never going to succeed.
3419 */
3420 if (ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE)) {
3421 /*
3422 * If we're verifying destination reachability, we always want
3423 * to complain here.
3424 *
3425 * If we're not verifying destination reachability but the
3426 * destination has a route, we still want to fail on the
3427 * temporary address and broadcast address tests.
3428 *
3429 * In both cases do we let the code continue so some reasonable
3430 * information is returned to the caller. That enables the
3431 * caller to use (and even cache) the IRE. conn_ip_ouput will
3432 * use the generation mismatch path to check for the unreachable
3433 * case thereby avoiding any specific check in the main path.
3434 */
3435 ASSERT(generation == IRE_GENERATION_VERIFY);
3436 if (flags & IPDF_VERIFY_DST) {
3437 /*
3438 * Set errno but continue to set up ixa_ire to be
3439 * the RTF_REJECT|RTF_BLACKHOLE IRE.
3440 * That allows callers to use ip_output to get an
3441 * ICMP error back.
3442 */
3443 if (!(ire->ire_type & IRE_HOST))
3444 error = ENETUNREACH;
3445 else
3446 error = EHOSTUNREACH;
3447 }
3448 }
3449
3450 if ((ire->ire_type & (IRE_BROADCAST|IRE_MULTICAST)) &&
3451 !(flags & IPDF_ALLOW_MCBC)) {
3452 ire_refrele(ire);
3453 ire = ire_reject(ipst, B_FALSE);
3454 generation = IRE_GENERATION_VERIFY;
3455 error = ENETUNREACH;
3456 }
3457
3458 /* Cache things */
3459 if (ixa->ixa_ire != NULL)
3460 ire_refrele_notr(ixa->ixa_ire);
3461 #ifdef DEBUG
3462 ire_refhold_notr(ire);
3463 ire_refrele(ire);
3464 #endif
3465 ixa->ixa_ire = ire;
3466 ixa->ixa_ire_generation = generation;
3467
3468 /*
3469 * Ensure that ixa_dce is always set any time that ixa_ire is set,
3470 * since some callers will send a packet to conn_ip_output() even if
3471 * there's an error.
3472 */
3473 if (flags & IPDF_UNIQUE_DCE) {
3474 /* Fallback to the default dce if allocation fails */
3475 dce = dce_lookup_and_add_v4(dst_addr, ipst);
3476 if (dce != NULL)
3477 generation = dce->dce_generation;
3478 else
3479 dce = dce_lookup_v4(dst_addr, ipst, &generation);
3480 } else {
3481 dce = dce_lookup_v4(dst_addr, ipst, &generation);
3482 }
3483 ASSERT(dce != NULL);
3484 if (ixa->ixa_dce != NULL)
3485 dce_refrele_notr(ixa->ixa_dce);
3486 #ifdef DEBUG
3487 dce_refhold_notr(dce);
3488 dce_refrele(dce);
3489 #endif
3490 ixa->ixa_dce = dce;
3491 ixa->ixa_dce_generation = generation;
3492
3493 /*
3494 * For multicast with multirt we have a flag passed back from
3495 * ire_lookup_multi_ill_v4 since we don't have an IRE for each
3496 * possible multicast address.
3497 * We also need a flag for multicast since we can't check
3498 * whether RTF_MULTIRT is set in ixa_ire for multicast.
3499 */
3500 if (multirt) {
3501 ixa->ixa_postfragfn = ip_postfrag_multirt_v4;
3502 ixa->ixa_flags |= IXAF_MULTIRT_MULTICAST;
3503 } else {
3504 ixa->ixa_postfragfn = ire->ire_postfragfn;
3505 ixa->ixa_flags &= ~IXAF_MULTIRT_MULTICAST;
3506 }
3507 if (!(ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE))) {
3508 /* Get an nce to cache. */
3509 nce = ire_to_nce(ire, firsthop, NULL);
3510 if (nce == NULL) {
3511 /* Allocation failure? */
3512 ixa->ixa_ire_generation = IRE_GENERATION_VERIFY;
3513 } else {
3514 if (ixa->ixa_nce != NULL)
3515 nce_refrele(ixa->ixa_nce);
3516 ixa->ixa_nce = nce;
3517 }
3518 }
3519
3520 /*
3521 * If the source address is a loopback address, the
3522 * destination had best be local or multicast.
3523 * If we are sending to an IRE_LOCAL using a loopback source then
3524 * it had better be the same zoneid.
3525 */
3526 if (*src_addrp == htonl(INADDR_LOOPBACK)) {
3527 if ((ire->ire_type & IRE_LOCAL) && ire->ire_zoneid != zoneid) {
3528 ire = NULL; /* Stored in ixa_ire */
3529 error = EADDRNOTAVAIL;
3530 goto bad_addr;
3531 }
3532 if (!(ire->ire_type & (IRE_LOOPBACK|IRE_LOCAL|IRE_MULTICAST))) {
3533 ire = NULL; /* Stored in ixa_ire */
3534 error = EADDRNOTAVAIL;
3535 goto bad_addr;
3536 }
3537 }
3538 if (ire->ire_type & IRE_BROADCAST) {
3539 /*
3540 * If the ULP didn't have a specified source, then we
3541 * make sure we reselect the source when sending
3542 * broadcasts out different interfaces.
3543 */
3544 if (flags & IPDF_SELECT_SRC)
3545 ixa->ixa_flags |= IXAF_SET_SOURCE;
3546 else
3547 ixa->ixa_flags &= ~IXAF_SET_SOURCE;
3548 }
3549
3550 /*
3551 * Does the caller want us to pick a source address?
3552 */
3553 if (flags & IPDF_SELECT_SRC) {
3554 ipaddr_t src_addr;
3555
3556 /*
3557 * We use use ire_nexthop_ill to avoid the under ipmp
3558 * interface for source address selection. Note that for ipmp
3559 * probe packets, ixa_ifindex would have been specified, and
3560 * the ip_select_route() invocation would have picked an ire
3561 * will ire_ill pointing at an under interface.
3562 */
3563 ill = ire_nexthop_ill(ire);
3564
3565 /* If unreachable we have no ill but need some source */
3566 if (ill == NULL) {
3567 src_addr = htonl(INADDR_LOOPBACK);
3568 /* Make sure we look for a better source address */
3569 generation = SRC_GENERATION_VERIFY;
3570 } else {
3571 error = ip_select_source_v4(ill, setsrc, dst_addr,
3572 ixa->ixa_multicast_ifaddr, zoneid,
3573 ipst, &src_addr, &generation, NULL);
3574 if (error != 0) {
3575 ire = NULL; /* Stored in ixa_ire */
3576 goto bad_addr;
3577 }
3578 }
3579
3580 /*
3581 * We allow the source address to to down.
3582 * However, we check that we don't use the loopback address
3583 * as a source when sending out on the wire.
3584 */
3585 if ((src_addr == htonl(INADDR_LOOPBACK)) &&
3586 !(ire->ire_type & (IRE_LOCAL|IRE_LOOPBACK|IRE_MULTICAST)) &&
3587 !(ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE))) {
3588 ire = NULL; /* Stored in ixa_ire */
3589 error = EADDRNOTAVAIL;
3590 goto bad_addr;
3591 }
3592
3593 *src_addrp = src_addr;
3594 ixa->ixa_src_generation = generation;
3595 }
3596
3597 /*
3598 * Make sure we don't leave an unreachable ixa_nce in place
3599 * since ip_select_route is used when we unplumb i.e., remove
3600 * references on ixa_ire, ixa_nce, and ixa_dce.
3601 */
3602 nce = ixa->ixa_nce;
3603 if (nce != NULL && nce->nce_is_condemned) {
3604 nce_refrele(nce);
3605 ixa->ixa_nce = NULL;
3606 ixa->ixa_ire_generation = IRE_GENERATION_VERIFY;
3607 }
3608
3609 /*
3610 * The caller has set IXAF_PMTU_DISCOVERY if path MTU is desired.
3611 * However, we can't do it for IPv4 multicast or broadcast.
3612 */
3613 if (ire->ire_type & (IRE_BROADCAST|IRE_MULTICAST))
3614 ixa->ixa_flags &= ~IXAF_PMTU_DISCOVERY;
3615
3616 /*
3617 * Set initial value for fragmentation limit. Either conn_ip_output
3618 * or ULP might updates it when there are routing changes.
3619 * Handles a NULL ixa_ire->ire_ill or a NULL ixa_nce for RTF_REJECT.
3620 */
3621 pmtu = ip_get_pmtu(ixa);
3622 ixa->ixa_fragsize = pmtu;
3623 /* Make sure ixa_fragsize and ixa_pmtu remain identical */
3624 if (ixa->ixa_flags & IXAF_VERIFY_PMTU)
3625 ixa->ixa_pmtu = pmtu;
3626
3627 /*
3628 * Extract information useful for some transports.
3629 * First we look for DCE metrics. Then we take what we have in
3630 * the metrics in the route, where the offlink is used if we have
3631 * one.
3632 */
3633 if (uinfo != NULL) {
3634 bzero(uinfo, sizeof (*uinfo));
3635
3636 if (dce->dce_flags & DCEF_UINFO)
3637 *uinfo = dce->dce_uinfo;
3638
3639 rts_merge_metrics(uinfo, &ire->ire_metrics);
3640
3641 /* Allow ire_metrics to decrease the path MTU from above */
3642 if (uinfo->iulp_mtu == 0 || uinfo->iulp_mtu > pmtu)
3643 uinfo->iulp_mtu = pmtu;
3644
3645 uinfo->iulp_localnet = (ire->ire_type & IRE_ONLINK) != 0;
3646 uinfo->iulp_loopback = (ire->ire_type & IRE_LOOPBACK) != 0;
3647 uinfo->iulp_local = (ire->ire_type & IRE_LOCAL) != 0;
3648 }
3649
3650 if (ill != NULL)
3651 ill_refrele(ill);
3652
3653 return (error);
3654
3655 bad_addr:
3656 if (ire != NULL)
3657 ire_refrele(ire);
3658
3659 if (ill != NULL)
3660 ill_refrele(ill);
3661
3662 /*
3663 * Make sure we don't leave an unreachable ixa_nce in place
3664 * since ip_select_route is used when we unplumb i.e., remove
3665 * references on ixa_ire, ixa_nce, and ixa_dce.
3666 */
3667 nce = ixa->ixa_nce;
3668 if (nce != NULL && nce->nce_is_condemned) {
3669 nce_refrele(nce);
3670 ixa->ixa_nce = NULL;
3671 ixa->ixa_ire_generation = IRE_GENERATION_VERIFY;
3672 }
3673
3674 return (error);
3675 }
3676
3677
3678 /*
3679 * Get the base MTU for the case when path MTU discovery is not used.
3680 * Takes the MTU of the IRE into account.
3681 */
3682 uint_t
3683 ip_get_base_mtu(ill_t *ill, ire_t *ire)
3684 {
3685 uint_t mtu;
3686 uint_t iremtu = ire->ire_metrics.iulp_mtu;
3687
3688 if (ire->ire_type & (IRE_MULTICAST|IRE_BROADCAST))
3689 mtu = ill->ill_mc_mtu;
3690 else
3691 mtu = ill->ill_mtu;
3692
3693 if (iremtu != 0 && iremtu < mtu)
3694 mtu = iremtu;
3695
3696 return (mtu);
3697 }
3698
3699 /*
3700 * Get the PMTU for the attributes. Handles both IPv4 and IPv6.
3701 * Assumes that ixa_ire, dce, and nce have already been set up.
3702 *
3703 * The caller has set IXAF_PMTU_DISCOVERY if path MTU discovery is desired.
3704 * We avoid path MTU discovery if it is disabled with ndd.
3705 * Furtermore, if the path MTU is too small, then we don't set DF for IPv4.
3706 *
3707 * NOTE: We also used to turn it off for source routed packets. That
3708 * is no longer required since the dce is per final destination.
3709 */
3710 uint_t
3711 ip_get_pmtu(ip_xmit_attr_t *ixa)
3712 {
3713 ip_stack_t *ipst = ixa->ixa_ipst;
3714 dce_t *dce;
3715 nce_t *nce;
3716 ire_t *ire;
3717 uint_t pmtu;
3718
3719 ire = ixa->ixa_ire;
3720 dce = ixa->ixa_dce;
3721 nce = ixa->ixa_nce;
3722
3723 /*
3724 * If path MTU discovery has been turned off by ndd, then we ignore
3725 * any dce_pmtu and for IPv4 we will not set DF.
3726 */
3727 if (!ipst->ips_ip_path_mtu_discovery)
3728 ixa->ixa_flags &= ~IXAF_PMTU_DISCOVERY;
3729
3730 pmtu = IP_MAXPACKET;
3731 /*
3732 * Decide whether whether IPv4 sets DF
3733 * For IPv6 "no DF" means to use the 1280 mtu
3734 */
3735 if (ixa->ixa_flags & IXAF_PMTU_DISCOVERY) {
3736 ixa->ixa_flags |= IXAF_PMTU_IPV4_DF;
3737 } else {
3738 ixa->ixa_flags &= ~IXAF_PMTU_IPV4_DF;
3739 if (!(ixa->ixa_flags & IXAF_IS_IPV4))
3740 pmtu = IPV6_MIN_MTU;
3741 }
3742
3743 /* Check if the PMTU is to old before we use it */
3744 if ((dce->dce_flags & DCEF_PMTU) &&
3745 TICK_TO_SEC(ddi_get_lbolt64()) - dce->dce_last_change_time >
3746 ipst->ips_ip_pathmtu_interval) {
3747 /*
3748 * Older than 20 minutes. Drop the path MTU information.
3749 */
3750 mutex_enter(&dce->dce_lock);
3751 dce->dce_flags &= ~(DCEF_PMTU|DCEF_TOO_SMALL_PMTU);
3752 dce->dce_last_change_time = TICK_TO_SEC(ddi_get_lbolt64());
3753 mutex_exit(&dce->dce_lock);
3754 dce_increment_generation(dce);
3755 }
3756
3757 /* The metrics on the route can lower the path MTU */
3758 if (ire->ire_metrics.iulp_mtu != 0 &&
3759 ire->ire_metrics.iulp_mtu < pmtu)
3760 pmtu = ire->ire_metrics.iulp_mtu;
3761
3762 /*
3763 * If the path MTU is smaller than some minimum, we still use dce_pmtu
3764 * above (would be 576 for IPv4 and 1280 for IPv6), but we clear
3765 * IXAF_PMTU_IPV4_DF so that we avoid setting DF for IPv4.
3766 */
3767 if (ixa->ixa_flags & IXAF_PMTU_DISCOVERY) {
3768 if (dce->dce_flags & DCEF_PMTU) {
3769 if (dce->dce_pmtu < pmtu)
3770 pmtu = dce->dce_pmtu;
3771
3772 if (dce->dce_flags & DCEF_TOO_SMALL_PMTU) {
3773 ixa->ixa_flags |= IXAF_PMTU_TOO_SMALL;
3774 ixa->ixa_flags &= ~IXAF_PMTU_IPV4_DF;
3775 } else {
3776 ixa->ixa_flags &= ~IXAF_PMTU_TOO_SMALL;
3777 ixa->ixa_flags |= IXAF_PMTU_IPV4_DF;
3778 }
3779 } else {
3780 ixa->ixa_flags &= ~IXAF_PMTU_TOO_SMALL;
3781 ixa->ixa_flags |= IXAF_PMTU_IPV4_DF;
3782 }
3783 }
3784
3785 /*
3786 * If we have an IRE_LOCAL we use the loopback mtu instead of
3787 * the ill for going out the wire i.e., IRE_LOCAL gets the same
3788 * mtu as IRE_LOOPBACK.
3789 */
3790 if (ire->ire_type & (IRE_LOCAL|IRE_LOOPBACK)) {
3791 uint_t loopback_mtu;
3792
3793 loopback_mtu = (ire->ire_ipversion == IPV6_VERSION) ?
3794 ip_loopback_mtu_v6plus : ip_loopback_mtuplus;
3795
3796 if (loopback_mtu < pmtu)
3797 pmtu = loopback_mtu;
3798 } else if (nce != NULL) {
3799 /*
3800 * Make sure we don't exceed the interface MTU.
3801 * In the case of RTF_REJECT or RTF_BLACKHOLE we might not have
3802 * an ill. We'd use the above IP_MAXPACKET in that case just
3803 * to tell the transport something larger than zero.
3804 */
3805 if (ire->ire_type & (IRE_MULTICAST|IRE_BROADCAST)) {
3806 if (nce->nce_common->ncec_ill->ill_mc_mtu < pmtu)
3807 pmtu = nce->nce_common->ncec_ill->ill_mc_mtu;
3808 if (nce->nce_common->ncec_ill != nce->nce_ill &&
3809 nce->nce_ill->ill_mc_mtu < pmtu) {
3810 /*
3811 * for interfaces in an IPMP group, the mtu of
3812 * the nce_ill (under_ill) could be different
3813 * from the mtu of the ncec_ill, so we take the
3814 * min of the two.
3815 */
3816 pmtu = nce->nce_ill->ill_mc_mtu;
3817 }
3818 } else {
3819 if (nce->nce_common->ncec_ill->ill_mtu < pmtu)
3820 pmtu = nce->nce_common->ncec_ill->ill_mtu;
3821 if (nce->nce_common->ncec_ill != nce->nce_ill &&
3822 nce->nce_ill->ill_mtu < pmtu) {
3823 /*
3824 * for interfaces in an IPMP group, the mtu of
3825 * the nce_ill (under_ill) could be different
3826 * from the mtu of the ncec_ill, so we take the
3827 * min of the two.
3828 */
3829 pmtu = nce->nce_ill->ill_mtu;
3830 }
3831 }
3832 }
3833
3834 /*
3835 * Handle the IPV6_USE_MIN_MTU socket option or ancillary data.
3836 * Only applies to IPv6.
3837 */
3838 if (!(ixa->ixa_flags & IXAF_IS_IPV4)) {
3839 if (ixa->ixa_flags & IXAF_USE_MIN_MTU) {
3840 switch (ixa->ixa_use_min_mtu) {
3841 case IPV6_USE_MIN_MTU_MULTICAST:
3842 if (ire->ire_type & IRE_MULTICAST)
3843 pmtu = IPV6_MIN_MTU;
3844 break;
3845 case IPV6_USE_MIN_MTU_ALWAYS:
3846 pmtu = IPV6_MIN_MTU;
3847 break;
3848 case IPV6_USE_MIN_MTU_NEVER:
3849 break;
3850 }
3851 } else {
3852 /* Default is IPV6_USE_MIN_MTU_MULTICAST */
3853 if (ire->ire_type & IRE_MULTICAST)
3854 pmtu = IPV6_MIN_MTU;
3855 }
3856 }
3857
3858 /*
3859 * After receiving an ICMPv6 "packet too big" message with a
3860 * MTU < 1280, and for multirouted IPv6 packets, the IP layer
3861 * will insert a 8-byte fragment header in every packet. We compensate
3862 * for those cases by returning a smaller path MTU to the ULP.
3863 *
3864 * In the case of CGTP then ip_output will add a fragment header.
3865 * Make sure there is room for it by telling a smaller number
3866 * to the transport.
3867 *
3868 * When IXAF_IPV6_ADDR_FRAGHDR we subtract the frag hdr here
3869 * so the ULPs consistently see a iulp_pmtu and ip_get_pmtu()
3870 * which is the size of the packets it can send.
3871 */
3872 if (!(ixa->ixa_flags & IXAF_IS_IPV4)) {
3873 if ((dce->dce_flags & DCEF_TOO_SMALL_PMTU) ||
3874 (ire->ire_flags & RTF_MULTIRT) ||
3875 (ixa->ixa_flags & IXAF_MULTIRT_MULTICAST)) {
3876 pmtu -= sizeof (ip6_frag_t);
3877 ixa->ixa_flags |= IXAF_IPV6_ADD_FRAGHDR;
3878 }
3879 }
3880
3881 return (pmtu);
3882 }
3883
3884 /*
3885 * Carve "len" bytes out of an mblk chain, consuming any we empty, and duping
3886 * the final piece where we don't. Return a pointer to the first mblk in the
3887 * result, and update the pointer to the next mblk to chew on. If anything
3888 * goes wrong (i.e., dupb fails), we waste everything in sight and return a
3889 * NULL pointer.
3890 */
3891 mblk_t *
3892 ip_carve_mp(mblk_t **mpp, ssize_t len)
3893 {
3894 mblk_t *mp0;
3895 mblk_t *mp1;
3896 mblk_t *mp2;
3897
3898 if (!len || !mpp || !(mp0 = *mpp))
3899 return (NULL);
3900 /* If we aren't going to consume the first mblk, we need a dup. */
3901 if (mp0->b_wptr - mp0->b_rptr > len) {
3902 mp1 = dupb(mp0);
3903 if (mp1) {
3904 /* Partition the data between the two mblks. */
3905 mp1->b_wptr = mp1->b_rptr + len;
3906 mp0->b_rptr = mp1->b_wptr;
3907 /*
3908 * after adjustments if mblk not consumed is now
3909 * unaligned, try to align it. If this fails free
3910 * all messages and let upper layer recover.
3911 */
3912 if (!OK_32PTR(mp0->b_rptr)) {
3913 if (!pullupmsg(mp0, -1)) {
3914 freemsg(mp0);
3915 freemsg(mp1);
3916 *mpp = NULL;
3917 return (NULL);
3918 }
3919 }
3920 }
3921 return (mp1);
3922 }
3923 /* Eat through as many mblks as we need to get len bytes. */
3924 len -= mp0->b_wptr - mp0->b_rptr;
3925 for (mp2 = mp1 = mp0; (mp2 = mp2->b_cont) != 0 && len; mp1 = mp2) {
3926 if (mp2->b_wptr - mp2->b_rptr > len) {
3927 /*
3928 * We won't consume the entire last mblk. Like
3929 * above, dup and partition it.
3930 */
3931 mp1->b_cont = dupb(mp2);
3932 mp1 = mp1->b_cont;
3933 if (!mp1) {
3934 /*
3935 * Trouble. Rather than go to a lot of
3936 * trouble to clean up, we free the messages.
3937 * This won't be any worse than losing it on
3938 * the wire.
3939 */
3940 freemsg(mp0);
3941 freemsg(mp2);
3942 *mpp = NULL;
3943 return (NULL);
3944 }
3945 mp1->b_wptr = mp1->b_rptr + len;
3946 mp2->b_rptr = mp1->b_wptr;
3947 /*
3948 * after adjustments if mblk not consumed is now
3949 * unaligned, try to align it. If this fails free
3950 * all messages and let upper layer recover.
3951 */
3952 if (!OK_32PTR(mp2->b_rptr)) {
3953 if (!pullupmsg(mp2, -1)) {
3954 freemsg(mp0);
3955 freemsg(mp2);
3956 *mpp = NULL;
3957 return (NULL);
3958 }
3959 }
3960 *mpp = mp2;
3961 return (mp0);
3962 }
3963 /* Decrement len by the amount we just got. */
3964 len -= mp2->b_wptr - mp2->b_rptr;
3965 }
3966 /*
3967 * len should be reduced to zero now. If not our caller has
3968 * screwed up.
3969 */
3970 if (len) {
3971 /* Shouldn't happen! */
3972 freemsg(mp0);
3973 *mpp = NULL;
3974 return (NULL);
3975 }
3976 /*
3977 * We consumed up to exactly the end of an mblk. Detach the part
3978 * we are returning from the rest of the chain.
3979 */
3980 mp1->b_cont = NULL;
3981 *mpp = mp2;
3982 return (mp0);
3983 }
3984
3985 /* The ill stream is being unplumbed. Called from ip_close */
3986 int
3987 ip_modclose(ill_t *ill)
3988 {
3989 boolean_t success;
3990 ipsq_t *ipsq;
3991 ipif_t *ipif;
3992 queue_t *q = ill->ill_rq;
3993 ip_stack_t *ipst = ill->ill_ipst;
3994 int i;
3995 arl_ill_common_t *ai = ill->ill_common;
3996
3997 /*
3998 * The punlink prior to this may have initiated a capability
3999 * negotiation. But ipsq_enter will block until that finishes or
4000 * times out.
4001 */
4002 success = ipsq_enter(ill, B_FALSE, NEW_OP);
4003
4004 /*
4005 * Open/close/push/pop is guaranteed to be single threaded
4006 * per stream by STREAMS. FS guarantees that all references
4007 * from top are gone before close is called. So there can't
4008 * be another close thread that has set CONDEMNED on this ill.
4009 * and cause ipsq_enter to return failure.
4010 */
4011 ASSERT(success);
4012 ipsq = ill->ill_phyint->phyint_ipsq;
4013
4014 /*
4015 * Mark it condemned. No new reference will be made to this ill.
4016 * Lookup functions will return an error. Threads that try to
4017 * increment the refcnt must check for ILL_CAN_LOOKUP. This ensures
4018 * that the refcnt will drop down to zero.
4019 */
4020 mutex_enter(&ill->ill_lock);
4021 ill->ill_state_flags |= ILL_CONDEMNED;
4022 for (ipif = ill->ill_ipif; ipif != NULL;
4023 ipif = ipif->ipif_next) {
4024 ipif->ipif_state_flags |= IPIF_CONDEMNED;
4025 }
4026 /*
4027 * Wake up anybody waiting to enter the ipsq. ipsq_enter
4028 * returns error if ILL_CONDEMNED is set
4029 */
4030 cv_broadcast(&ill->ill_cv);
4031 mutex_exit(&ill->ill_lock);
4032
4033 /*
4034 * Send all the deferred DLPI messages downstream which came in
4035 * during the small window right before ipsq_enter(). We do this
4036 * without waiting for the ACKs because all the ACKs for M_PROTO
4037 * messages are ignored in ip_rput() when ILL_CONDEMNED is set.
4038 */
4039 ill_dlpi_send_deferred(ill);
4040
4041 /*
4042 * Shut down fragmentation reassembly.
4043 * ill_frag_timer won't start a timer again.
4044 * Now cancel any existing timer
4045 */
4046 (void) untimeout(ill->ill_frag_timer_id);
4047 (void) ill_frag_timeout(ill, 0);
4048
4049 /*
4050 * Call ill_delete to bring down the ipifs, ilms and ill on
4051 * this ill. Then wait for the refcnts to drop to zero.
4052 * ill_is_freeable checks whether the ill is really quiescent.
4053 * Then make sure that threads that are waiting to enter the
4054 * ipsq have seen the error returned by ipsq_enter and have
4055 * gone away. Then we call ill_delete_tail which does the
4056 * DL_UNBIND_REQ with the driver and then qprocsoff.
4057 */
4058 ill_delete(ill);
4059 mutex_enter(&ill->ill_lock);
4060 while (!ill_is_freeable(ill))
4061 cv_wait(&ill->ill_cv, &ill->ill_lock);
4062
4063 while (ill->ill_waiters)
4064 cv_wait(&ill->ill_cv, &ill->ill_lock);
4065
4066 mutex_exit(&ill->ill_lock);
4067
4068 /*
4069 * ill_delete_tail drops reference on ill_ipst, but we need to keep
4070 * it held until the end of the function since the cleanup
4071 * below needs to be able to use the ip_stack_t.
4072 */
4073 netstack_hold(ipst->ips_netstack);
4074
4075 /* qprocsoff is done via ill_delete_tail */
4076 ill_delete_tail(ill);
4077 /*
4078 * synchronously wait for arp stream to unbind. After this, we
4079 * cannot get any data packets up from the driver.
4080 */
4081 arp_unbind_complete(ill);
4082 ASSERT(ill->ill_ipst == NULL);
4083
4084 /*
4085 * Walk through all conns and qenable those that have queued data.
4086 * Close synchronization needs this to
4087 * be done to ensure that all upper layers blocked
4088 * due to flow control to the closing device
4089 * get unblocked.
4090 */
4091 ip1dbg(("ip_wsrv: walking\n"));
4092 for (i = 0; i < TX_FANOUT_SIZE; i++) {
4093 conn_walk_drain(ipst, &ipst->ips_idl_tx_list[i]);
4094 }
4095
4096 /*
4097 * ai can be null if this is an IPv6 ill, or if the IPv4
4098 * stream is being torn down before ARP was plumbed (e.g.,
4099 * /sbin/ifconfig plumbing a stream twice, and encountering
4100 * an error
4101 */
4102 if (ai != NULL) {
4103 ASSERT(!ill->ill_isv6);
4104 mutex_enter(&ai->ai_lock);
4105 ai->ai_ill = NULL;
4106 if (ai->ai_arl == NULL) {
4107 mutex_destroy(&ai->ai_lock);
4108 kmem_free(ai, sizeof (*ai));
4109 } else {
4110 cv_signal(&ai->ai_ill_unplumb_done);
4111 mutex_exit(&ai->ai_lock);
4112 }
4113 }
4114
4115 mutex_enter(&ipst->ips_ip_mi_lock);
4116 mi_close_unlink(&ipst->ips_ip_g_head, (IDP)ill);
4117 mutex_exit(&ipst->ips_ip_mi_lock);
4118
4119 /*
4120 * credp could be null if the open didn't succeed and ip_modopen
4121 * itself calls ip_close.
4122 */
4123 if (ill->ill_credp != NULL)
4124 crfree(ill->ill_credp);
4125
4126 mutex_destroy(&ill->ill_saved_ire_lock);
4127 mutex_destroy(&ill->ill_lock);
4128 rw_destroy(&ill->ill_mcast_lock);
4129 mutex_destroy(&ill->ill_mcast_serializer);
4130 list_destroy(&ill->ill_nce);
4131
4132 /*
4133 * Now we are done with the module close pieces that
4134 * need the netstack_t.
4135 */
4136 netstack_rele(ipst->ips_netstack);
4137
4138 mi_close_free((IDP)ill);
4139 q->q_ptr = WR(q)->q_ptr = NULL;
4140
4141 ipsq_exit(ipsq);
4142
4143 return (0);
4144 }
4145
4146 /*
4147 * This is called as part of close() for IP, UDP, ICMP, and RTS
4148 * in order to quiesce the conn.
4149 */
4150 void
4151 ip_quiesce_conn(conn_t *connp)
4152 {
4153 boolean_t drain_cleanup_reqd = B_FALSE;
4154 boolean_t conn_ioctl_cleanup_reqd = B_FALSE;
4155 boolean_t ilg_cleanup_reqd = B_FALSE;
4156 ip_stack_t *ipst;
4157
4158 ASSERT(!IPCL_IS_TCP(connp));
4159 ipst = connp->conn_netstack->netstack_ip;
4160
4161 /*
4162 * Mark the conn as closing, and this conn must not be
4163 * inserted in future into any list. Eg. conn_drain_insert(),
4164 * won't insert this conn into the conn_drain_list.
4165 *
4166 * conn_idl, and conn_ilg cannot get set henceforth.
4167 */
4168 mutex_enter(&connp->conn_lock);
4169 ASSERT(!(connp->conn_state_flags & CONN_QUIESCED));
4170 connp->conn_state_flags |= CONN_CLOSING;
4171 if (connp->conn_idl != NULL)
4172 drain_cleanup_reqd = B_TRUE;
4173 if (connp->conn_oper_pending_ill != NULL)
4174 conn_ioctl_cleanup_reqd = B_TRUE;
4175 if (connp->conn_dhcpinit_ill != NULL) {
4176 ASSERT(connp->conn_dhcpinit_ill->ill_dhcpinit != 0);
4177 atomic_dec_32(&connp->conn_dhcpinit_ill->ill_dhcpinit);
4178 ill_set_inputfn(connp->conn_dhcpinit_ill);
4179 connp->conn_dhcpinit_ill = NULL;
4180 }
4181 if (connp->conn_ilg != NULL)
4182 ilg_cleanup_reqd = B_TRUE;
4183 mutex_exit(&connp->conn_lock);
4184
4185 if (conn_ioctl_cleanup_reqd)
4186 conn_ioctl_cleanup(connp);
4187
4188 if (is_system_labeled() && connp->conn_anon_port) {
4189 (void) tsol_mlp_anon(crgetzone(connp->conn_cred),
4190 connp->conn_mlp_type, connp->conn_proto,
4191 ntohs(connp->conn_lport), B_FALSE);
4192 connp->conn_anon_port = 0;
4193 }
4194 connp->conn_mlp_type = mlptSingle;
4195
4196 /*
4197 * Remove this conn from any fanout list it is on.
4198 * and then wait for any threads currently operating
4199 * on this endpoint to finish
4200 */
4201 ipcl_hash_remove(connp);
4202
4203 /*
4204 * Remove this conn from the drain list, and do any other cleanup that
4205 * may be required. (TCP conns are never flow controlled, and
4206 * conn_idl will be NULL.)
4207 */
4208 if (drain_cleanup_reqd && connp->conn_idl != NULL) {
4209 idl_t *idl = connp->conn_idl;
4210
4211 mutex_enter(&idl->idl_lock);
4212 conn_drain(connp, B_TRUE);
4213 mutex_exit(&idl->idl_lock);
4214 }
4215
4216 if (connp == ipst->ips_ip_g_mrouter)
4217 (void) ip_mrouter_done(ipst);
4218
4219 if (ilg_cleanup_reqd)
4220 ilg_delete_all(connp);
4221
4222 /*
4223 * Now conn refcnt can increase only thru CONN_INC_REF_LOCKED.
4224 * callers from write side can't be there now because close
4225 * is in progress. The only other caller is ipcl_walk
4226 * which checks for the condemned flag.
4227 */
4228 mutex_enter(&connp->conn_lock);
4229 connp->conn_state_flags |= CONN_CONDEMNED;
4230 while (connp->conn_ref != 1)
4231 cv_wait(&connp->conn_cv, &connp->conn_lock);
4232 connp->conn_state_flags |= CONN_QUIESCED;
4233 mutex_exit(&connp->conn_lock);
4234 }
4235
4236 /* ARGSUSED */
4237 int
4238 ip_close(queue_t *q, int flags)
4239 {
4240 conn_t *connp;
4241
4242 /*
4243 * Call the appropriate delete routine depending on whether this is
4244 * a module or device.
4245 */
4246 if (WR(q)->q_next != NULL) {
4247 /* This is a module close */
4248 return (ip_modclose((ill_t *)q->q_ptr));
4249 }
4250
4251 connp = q->q_ptr;
4252 ip_quiesce_conn(connp);
4253
4254 qprocsoff(q);
4255
4256 /*
4257 * Now we are truly single threaded on this stream, and can
4258 * delete the things hanging off the connp, and finally the connp.
4259 * We removed this connp from the fanout list, it cannot be
4260 * accessed thru the fanouts, and we already waited for the
4261 * conn_ref to drop to 0. We are already in close, so
4262 * there cannot be any other thread from the top. qprocsoff
4263 * has completed, and service has completed or won't run in
4264 * future.
4265 */
4266 ASSERT(connp->conn_ref == 1);
4267
4268 inet_minor_free(connp->conn_minor_arena, connp->conn_dev);
4269
4270 connp->conn_ref--;
4271 ipcl_conn_destroy(connp);
4272
4273 q->q_ptr = WR(q)->q_ptr = NULL;
4274 return (0);
4275 }
4276
4277 /*
4278 * Wapper around putnext() so that ip_rts_request can merely use
4279 * conn_recv.
4280 */
4281 /*ARGSUSED2*/
4282 static void
4283 ip_conn_input(void *arg1, mblk_t *mp, void *arg2, ip_recv_attr_t *ira)
4284 {
4285 conn_t *connp = (conn_t *)arg1;
4286
4287 putnext(connp->conn_rq, mp);
4288 }
4289
4290 /* Dummy in case ICMP error delivery is attempted to a /dev/ip instance */
4291 /* ARGSUSED */
4292 static void
4293 ip_conn_input_icmp(void *arg1, mblk_t *mp, void *arg2, ip_recv_attr_t *ira)
4294 {
4295 freemsg(mp);
4296 }
4297
4298 /*
4299 * Called when the module is about to be unloaded
4300 */
4301 void
4302 ip_ddi_destroy(void)
4303 {
4304 /* This needs to be called before destroying any transports. */
4305 mutex_enter(&cpu_lock);
4306 unregister_cpu_setup_func(ip_tp_cpu_update, NULL);
4307 mutex_exit(&cpu_lock);
4308
4309 tnet_fini();
4310
4311 icmp_ddi_g_destroy();
4312 rts_ddi_g_destroy();
4313 udp_ddi_g_destroy();
4314 sctp_ddi_g_destroy();
4315 tcp_ddi_g_destroy();
4316 ilb_ddi_g_destroy();
4317 dce_g_destroy();
4318 ipsec_policy_g_destroy();
4319 ipcl_g_destroy();
4320 ip_net_g_destroy();
4321 ip_ire_g_fini();
4322 inet_minor_destroy(ip_minor_arena_sa);
4323 #if defined(_LP64)
4324 inet_minor_destroy(ip_minor_arena_la);
4325 #endif
4326
4327 #ifdef DEBUG
4328 list_destroy(&ip_thread_list);
4329 rw_destroy(&ip_thread_rwlock);
4330 tsd_destroy(&ip_thread_data);
4331 #endif
4332
4333 netstack_unregister(NS_IP);
4334 }
4335
4336 /*
4337 * First step in cleanup.
4338 */
4339 /* ARGSUSED */
4340 static void
4341 ip_stack_shutdown(netstackid_t stackid, void *arg)
4342 {
4343 ip_stack_t *ipst = (ip_stack_t *)arg;
4344
4345 #ifdef NS_DEBUG
4346 printf("ip_stack_shutdown(%p, stack %d)\n", (void *)ipst, stackid);
4347 #endif
4348
4349 /*
4350 * Perform cleanup for special interfaces (loopback and IPMP).
4351 */
4352 ip_interface_cleanup(ipst);
4353
4354 /*
4355 * The *_hook_shutdown()s start the process of notifying any
4356 * consumers that things are going away.... nothing is destroyed.
4357 */
4358 ipv4_hook_shutdown(ipst);
4359 ipv6_hook_shutdown(ipst);
4360 arp_hook_shutdown(ipst);
4361
4362 mutex_enter(&ipst->ips_capab_taskq_lock);
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 /*
4369 * Free the IP stack instance.
4370 */
4371 static void
4372 ip_stack_fini(netstackid_t stackid, void *arg)
4373 {
4374 ip_stack_t *ipst = (ip_stack_t *)arg;
4375 int ret;
4376
4377 #ifdef NS_DEBUG
4378 printf("ip_stack_fini(%p, stack %d)\n", (void *)ipst, stackid);
4379 #endif
4380 /*
4381 * At this point, all of the notifications that the events and
4382 * protocols are going away have been run, meaning that we can
4383 * now set about starting to clean things up.
4384 */
4385 ipobs_fini(ipst);
4386 ipv4_hook_destroy(ipst);
4387 ipv6_hook_destroy(ipst);
4388 arp_hook_destroy(ipst);
4389 ip_net_destroy(ipst);
4390
4391 ipmp_destroy(ipst);
4392
4393 ip_kstat_fini(stackid, ipst->ips_ip_mibkp);
4394 ipst->ips_ip_mibkp = NULL;
4395 icmp_kstat_fini(stackid, ipst->ips_icmp_mibkp);
4396 ipst->ips_icmp_mibkp = NULL;
4397 ip_kstat2_fini(stackid, ipst->ips_ip_kstat);
4398 ipst->ips_ip_kstat = NULL;
4399 bzero(&ipst->ips_ip_statistics, sizeof (ipst->ips_ip_statistics));
4400 ip6_kstat_fini(stackid, ipst->ips_ip6_kstat);
4401 ipst->ips_ip6_kstat = NULL;
4402 bzero(&ipst->ips_ip6_statistics, sizeof (ipst->ips_ip6_statistics));
4403
4404 kmem_free(ipst->ips_propinfo_tbl,
4405 ip_propinfo_count * sizeof (mod_prop_info_t));
4406 ipst->ips_propinfo_tbl = NULL;
4407
4408 dce_stack_destroy(ipst);
4409 ip_mrouter_stack_destroy(ipst);
4410
4411 ret = untimeout(ipst->ips_igmp_timeout_id);
4412 if (ret == -1) {
4413 ASSERT(ipst->ips_igmp_timeout_id == 0);
4414 } else {
4415 ASSERT(ipst->ips_igmp_timeout_id != 0);
4416 ipst->ips_igmp_timeout_id = 0;
4417 }
4418 ret = untimeout(ipst->ips_igmp_slowtimeout_id);
4419 if (ret == -1) {
4420 ASSERT(ipst->ips_igmp_slowtimeout_id == 0);
4421 } else {
4422 ASSERT(ipst->ips_igmp_slowtimeout_id != 0);
4423 ipst->ips_igmp_slowtimeout_id = 0;
4424 }
4425 ret = untimeout(ipst->ips_mld_timeout_id);
4426 if (ret == -1) {
4427 ASSERT(ipst->ips_mld_timeout_id == 0);
4428 } else {
4429 ASSERT(ipst->ips_mld_timeout_id != 0);
4430 ipst->ips_mld_timeout_id = 0;
4431 }
4432 ret = untimeout(ipst->ips_mld_slowtimeout_id);
4433 if (ret == -1) {
4434 ASSERT(ipst->ips_mld_slowtimeout_id == 0);
4435 } else {
4436 ASSERT(ipst->ips_mld_slowtimeout_id != 0);
4437 ipst->ips_mld_slowtimeout_id = 0;
4438 }
4439
4440 ip_ire_fini(ipst);
4441 ip6_asp_free(ipst);
4442 conn_drain_fini(ipst);
4443 ipcl_destroy(ipst);
4444
4445 mutex_destroy(&ipst->ips_ndp4->ndp_g_lock);
4446 mutex_destroy(&ipst->ips_ndp6->ndp_g_lock);
4447 kmem_free(ipst->ips_ndp4, sizeof (ndp_g_t));
4448 ipst->ips_ndp4 = NULL;
4449 kmem_free(ipst->ips_ndp6, sizeof (ndp_g_t));
4450 ipst->ips_ndp6 = NULL;
4451
4452 if (ipst->ips_loopback_ksp != NULL) {
4453 kstat_delete_netstack(ipst->ips_loopback_ksp, stackid);
4454 ipst->ips_loopback_ksp = NULL;
4455 }
4456
4457 mutex_destroy(&ipst->ips_capab_taskq_lock);
4458 cv_destroy(&ipst->ips_capab_taskq_cv);
4459
4460 rw_destroy(&ipst->ips_srcid_lock);
4461
4462 mutex_destroy(&ipst->ips_ip_mi_lock);
4463 rw_destroy(&ipst->ips_ill_g_usesrc_lock);
4464
4465 mutex_destroy(&ipst->ips_igmp_timer_lock);
4466 mutex_destroy(&ipst->ips_mld_timer_lock);
4467 mutex_destroy(&ipst->ips_igmp_slowtimeout_lock);
4468 mutex_destroy(&ipst->ips_mld_slowtimeout_lock);
4469 mutex_destroy(&ipst->ips_ip_addr_avail_lock);
4470 rw_destroy(&ipst->ips_ill_g_lock);
4471
4472 kmem_free(ipst->ips_phyint_g_list, sizeof (phyint_list_t));
4473 ipst->ips_phyint_g_list = NULL;
4474 kmem_free(ipst->ips_ill_g_heads, sizeof (ill_g_head_t) * MAX_G_HEADS);
4475 ipst->ips_ill_g_heads = NULL;
4476
4477 ldi_ident_release(ipst->ips_ldi_ident);
4478 kmem_free(ipst, sizeof (*ipst));
4479 }
4480
4481 /*
4482 * This function is called from the TSD destructor, and is used to debug
4483 * reference count issues in IP. See block comment in <inet/ip_if.h> for
4484 * details.
4485 */
4486 static void
4487 ip_thread_exit(void *phash)
4488 {
4489 th_hash_t *thh = phash;
4490
4491 rw_enter(&ip_thread_rwlock, RW_WRITER);
4492 list_remove(&ip_thread_list, thh);
4493 rw_exit(&ip_thread_rwlock);
4494 mod_hash_destroy_hash(thh->thh_hash);
4495 kmem_free(thh, sizeof (*thh));
4496 }
4497
4498 /*
4499 * Called when the IP kernel module is loaded into the kernel
4500 */
4501 void
4502 ip_ddi_init(void)
4503 {
4504 ip_squeue_flag = ip_squeue_switch(ip_squeue_enter);
4505
4506 /*
4507 * For IP and TCP the minor numbers should start from 2 since we have 4
4508 * initial devices: ip, ip6, tcp, tcp6.
4509 */
4510 /*
4511 * If this is a 64-bit kernel, then create two separate arenas -
4512 * one for TLIs in the range of INET_MIN_DEV+2 through 2^^18-1, and the
4513 * other for socket apps in the range 2^^18 through 2^^32-1.
4514 */
4515 ip_minor_arena_la = NULL;
4516 ip_minor_arena_sa = NULL;
4517 #if defined(_LP64)
4518 if ((ip_minor_arena_sa = inet_minor_create("ip_minor_arena_sa",
4519 INET_MIN_DEV + 2, MAXMIN32, KM_SLEEP)) == NULL) {
4520 cmn_err(CE_PANIC,
4521 "ip_ddi_init: ip_minor_arena_sa creation failed\n");
4522 }
4523 if ((ip_minor_arena_la = inet_minor_create("ip_minor_arena_la",
4524 MAXMIN32 + 1, MAXMIN64, KM_SLEEP)) == NULL) {
4525 cmn_err(CE_PANIC,
4526 "ip_ddi_init: ip_minor_arena_la creation failed\n");
4527 }
4528 #else
4529 if ((ip_minor_arena_sa = inet_minor_create("ip_minor_arena_sa",
4530 INET_MIN_DEV + 2, MAXMIN, KM_SLEEP)) == NULL) {
4531 cmn_err(CE_PANIC,
4532 "ip_ddi_init: ip_minor_arena_sa creation failed\n");
4533 }
4534 #endif
4535 ip_poll_normal_ticks = MSEC_TO_TICK_ROUNDUP(ip_poll_normal_ms);
4536
4537 ipcl_g_init();
4538 ip_ire_g_init();
4539 ip_net_g_init();
4540
4541 #ifdef DEBUG
4542 tsd_create(&ip_thread_data, ip_thread_exit);
4543 rw_init(&ip_thread_rwlock, NULL, RW_DEFAULT, NULL);
4544 list_create(&ip_thread_list, sizeof (th_hash_t),
4545 offsetof(th_hash_t, thh_link));
4546 #endif
4547 ipsec_policy_g_init();
4548 tcp_ddi_g_init();
4549 sctp_ddi_g_init();
4550 dce_g_init();
4551
4552 /*
4553 * We want to be informed each time a stack is created or
4554 * destroyed in the kernel, so we can maintain the
4555 * set of udp_stack_t's.
4556 */
4557 netstack_register(NS_IP, ip_stack_init, ip_stack_shutdown,
4558 ip_stack_fini);
4559
4560 tnet_init();
4561
4562 udp_ddi_g_init();
4563 rts_ddi_g_init();
4564 icmp_ddi_g_init();
4565 ilb_ddi_g_init();
4566
4567 /* This needs to be called after all transports are initialized. */
4568 mutex_enter(&cpu_lock);
4569 register_cpu_setup_func(ip_tp_cpu_update, NULL);
4570 mutex_exit(&cpu_lock);
4571 }
4572
4573 /*
4574 * Initialize the IP stack instance.
4575 */
4576 static void *
4577 ip_stack_init(netstackid_t stackid, netstack_t *ns)
4578 {
4579 ip_stack_t *ipst;
4580 size_t arrsz;
4581 major_t major;
4582
4583 #ifdef NS_DEBUG
4584 printf("ip_stack_init(stack %d)\n", stackid);
4585 #endif
4586
4587 ipst = (ip_stack_t *)kmem_zalloc(sizeof (*ipst), KM_SLEEP);
4588 ipst->ips_netstack = ns;
4589
4590 ipst->ips_ill_g_heads = kmem_zalloc(sizeof (ill_g_head_t) * MAX_G_HEADS,
4591 KM_SLEEP);
4592 ipst->ips_phyint_g_list = kmem_zalloc(sizeof (phyint_list_t),
4593 KM_SLEEP);
4594 ipst->ips_ndp4 = kmem_zalloc(sizeof (ndp_g_t), KM_SLEEP);
4595 ipst->ips_ndp6 = kmem_zalloc(sizeof (ndp_g_t), KM_SLEEP);
4596 mutex_init(&ipst->ips_ndp4->ndp_g_lock, NULL, MUTEX_DEFAULT, NULL);
4597 mutex_init(&ipst->ips_ndp6->ndp_g_lock, NULL, MUTEX_DEFAULT, NULL);
4598
4599 mutex_init(&ipst->ips_igmp_timer_lock, NULL, MUTEX_DEFAULT, NULL);
4600 ipst->ips_igmp_deferred_next = INFINITY;
4601 mutex_init(&ipst->ips_mld_timer_lock, NULL, MUTEX_DEFAULT, NULL);
4602 ipst->ips_mld_deferred_next = INFINITY;
4603 mutex_init(&ipst->ips_igmp_slowtimeout_lock, NULL, MUTEX_DEFAULT, NULL);
4604 mutex_init(&ipst->ips_mld_slowtimeout_lock, NULL, MUTEX_DEFAULT, NULL);
4605 mutex_init(&ipst->ips_ip_mi_lock, NULL, MUTEX_DEFAULT, NULL);
4606 mutex_init(&ipst->ips_ip_addr_avail_lock, NULL, MUTEX_DEFAULT, NULL);
4607 rw_init(&ipst->ips_ill_g_lock, NULL, RW_DEFAULT, NULL);
4608 rw_init(&ipst->ips_ill_g_usesrc_lock, NULL, RW_DEFAULT, NULL);
4609
4610 ipcl_init(ipst);
4611 ip_ire_init(ipst);
4612 ip6_asp_init(ipst);
4613 ipif_init(ipst);
4614 conn_drain_init(ipst);
4615 ip_mrouter_stack_init(ipst);
4616 dce_stack_init(ipst);
4617
4618 ipst->ips_ip_multirt_log_interval = 1000;
4619
4620 ipst->ips_ill_index = 1;
4621
4622 ipst->ips_saved_ip_forwarding = -1;
4623 ipst->ips_reg_vif_num = ALL_VIFS; /* Index to Register vif */
4624
4625 arrsz = ip_propinfo_count * sizeof (mod_prop_info_t);
4626 ipst->ips_propinfo_tbl = (mod_prop_info_t *)kmem_alloc(arrsz, KM_SLEEP);
4627 bcopy(ip_propinfo_tbl, ipst->ips_propinfo_tbl, arrsz);
4628
4629 ipst->ips_ip_mibkp = ip_kstat_init(stackid, ipst);
4630 ipst->ips_icmp_mibkp = icmp_kstat_init(stackid);
4631 ipst->ips_ip_kstat = ip_kstat2_init(stackid, &ipst->ips_ip_statistics);
4632 ipst->ips_ip6_kstat =
4633 ip6_kstat_init(stackid, &ipst->ips_ip6_statistics);
4634
4635 ipst->ips_ip_src_id = 1;
4636 rw_init(&ipst->ips_srcid_lock, NULL, RW_DEFAULT, NULL);
4637
4638 ipst->ips_src_generation = SRC_GENERATION_INITIAL;
4639
4640 ip_net_init(ipst, ns);
4641 ipv4_hook_init(ipst);
4642 ipv6_hook_init(ipst);
4643 arp_hook_init(ipst);
4644 ipmp_init(ipst);
4645 ipobs_init(ipst);
4646
4647 /*
4648 * Create the taskq dispatcher thread and initialize related stuff.
4649 */
4650 ipst->ips_capab_taskq_thread = thread_create(NULL, 0,
4651 ill_taskq_dispatch, ipst, 0, &p0, TS_RUN, minclsyspri);
4652 mutex_init(&ipst->ips_capab_taskq_lock, NULL, MUTEX_DEFAULT, NULL);
4653 cv_init(&ipst->ips_capab_taskq_cv, NULL, CV_DEFAULT, NULL);
4654
4655 major = mod_name_to_major(INET_NAME);
4656 (void) ldi_ident_from_major(major, &ipst->ips_ldi_ident);
4657 return (ipst);
4658 }
4659
4660 /*
4661 * Allocate and initialize a DLPI template of the specified length. (May be
4662 * called as writer.)
4663 */
4664 mblk_t *
4665 ip_dlpi_alloc(size_t len, t_uscalar_t prim)
4666 {
4667 mblk_t *mp;
4668
4669 mp = allocb(len, BPRI_MED);
4670 if (!mp)
4671 return (NULL);
4672
4673 /*
4674 * DLPIv2 says that DL_INFO_REQ and DL_TOKEN_REQ (the latter
4675 * of which we don't seem to use) are sent with M_PCPROTO, and
4676 * that other DLPI are M_PROTO.
4677 */
4678 if (prim == DL_INFO_REQ) {
4679 mp->b_datap->db_type = M_PCPROTO;
4680 } else {
4681 mp->b_datap->db_type = M_PROTO;
4682 }
4683
4684 mp->b_wptr = mp->b_rptr + len;
4685 bzero(mp->b_rptr, len);
4686 ((dl_unitdata_req_t *)mp->b_rptr)->dl_primitive = prim;
4687 return (mp);
4688 }
4689
4690 /*
4691 * Allocate and initialize a DLPI notification. (May be called as writer.)
4692 */
4693 mblk_t *
4694 ip_dlnotify_alloc(uint_t notification, uint_t data)
4695 {
4696 dl_notify_ind_t *notifyp;
4697 mblk_t *mp;
4698
4699 if ((mp = ip_dlpi_alloc(DL_NOTIFY_IND_SIZE, DL_NOTIFY_IND)) == NULL)
4700 return (NULL);
4701
4702 notifyp = (dl_notify_ind_t *)mp->b_rptr;
4703 notifyp->dl_notification = notification;
4704 notifyp->dl_data = data;
4705 return (mp);
4706 }
4707
4708 mblk_t *
4709 ip_dlnotify_alloc2(uint_t notification, uint_t data1, uint_t data2)
4710 {
4711 dl_notify_ind_t *notifyp;
4712 mblk_t *mp;
4713
4714 if ((mp = ip_dlpi_alloc(DL_NOTIFY_IND_SIZE, DL_NOTIFY_IND)) == NULL)
4715 return (NULL);
4716
4717 notifyp = (dl_notify_ind_t *)mp->b_rptr;
4718 notifyp->dl_notification = notification;
4719 notifyp->dl_data1 = data1;
4720 notifyp->dl_data2 = data2;
4721 return (mp);
4722 }
4723
4724 /*
4725 * Debug formatting routine. Returns a character string representation of the
4726 * addr in buf, of the form xxx.xxx.xxx.xxx. This routine takes the address
4727 * in the form of a ipaddr_t and calls ip_dot_saddr with a pointer.
4728 *
4729 * Once the ndd table-printing interfaces are removed, this can be changed to
4730 * standard dotted-decimal form.
4731 */
4732 char *
4733 ip_dot_addr(ipaddr_t addr, char *buf)
4734 {
4735 uint8_t *ap = (uint8_t *)&addr;
4736
4737 (void) mi_sprintf(buf, "%03d.%03d.%03d.%03d",
4738 ap[0] & 0xFF, ap[1] & 0xFF, ap[2] & 0xFF, ap[3] & 0xFF);
4739 return (buf);
4740 }
4741
4742 /*
4743 * Write the given MAC address as a printable string in the usual colon-
4744 * separated format.
4745 */
4746 const char *
4747 mac_colon_addr(const uint8_t *addr, size_t alen, char *buf, size_t buflen)
4748 {
4749 char *bp;
4750
4751 if (alen == 0 || buflen < 4)
4752 return ("?");
4753 bp = buf;
4754 for (;;) {
4755 /*
4756 * If there are more MAC address bytes available, but we won't
4757 * have any room to print them, then add "..." to the string
4758 * instead. See below for the 'magic number' explanation.
4759 */
4760 if ((alen == 2 && buflen < 6) || (alen > 2 && buflen < 7)) {
4761 (void) strcpy(bp, "...");
4762 break;
4763 }
4764 (void) sprintf(bp, "%02x", *addr++);
4765 bp += 2;
4766 if (--alen == 0)
4767 break;
4768 *bp++ = ':';
4769 buflen -= 3;
4770 /*
4771 * At this point, based on the first 'if' statement above,
4772 * either alen == 1 and buflen >= 3, or alen > 1 and
4773 * buflen >= 4. The first case leaves room for the final "xx"
4774 * number and trailing NUL byte. The second leaves room for at
4775 * least "...". Thus the apparently 'magic' numbers chosen for
4776 * that statement.
4777 */
4778 }
4779 return (buf);
4780 }
4781
4782 /*
4783 * Called when it is conceptually a ULP that would sent the packet
4784 * e.g., port unreachable and protocol unreachable. Check that the packet
4785 * would have passed the IPsec global policy before sending the error.
4786 *
4787 * Send an ICMP error after patching up the packet appropriately.
4788 * Uses ip_drop_input and bumps the appropriate MIB.
4789 */
4790 void
4791 ip_fanout_send_icmp_v4(mblk_t *mp, uint_t icmp_type, uint_t icmp_code,
4792 ip_recv_attr_t *ira)
4793 {
4794 ipha_t *ipha;
4795 boolean_t secure;
4796 ill_t *ill = ira->ira_ill;
4797 ip_stack_t *ipst = ill->ill_ipst;
4798 netstack_t *ns = ipst->ips_netstack;
4799 ipsec_stack_t *ipss = ns->netstack_ipsec;
4800
4801 secure = ira->ira_flags & IRAF_IPSEC_SECURE;
4802
4803 /*
4804 * We are generating an icmp error for some inbound packet.
4805 * Called from all ip_fanout_(udp, tcp, proto) functions.
4806 * Before we generate an error, check with global policy
4807 * to see whether this is allowed to enter the system. As
4808 * there is no "conn", we are checking with global policy.
4809 */
4810 ipha = (ipha_t *)mp->b_rptr;
4811 if (secure || ipss->ipsec_inbound_v4_policy_present) {
4812 mp = ipsec_check_global_policy(mp, NULL, ipha, NULL, ira, ns);
4813 if (mp == NULL)
4814 return;
4815 }
4816
4817 /* We never send errors for protocols that we do implement */
4818 if (ira->ira_protocol == IPPROTO_ICMP ||
4819 ira->ira_protocol == IPPROTO_IGMP) {
4820 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
4821 ip_drop_input("ip_fanout_send_icmp_v4", mp, ill);
4822 freemsg(mp);
4823 return;
4824 }
4825 /*
4826 * Have to correct checksum since
4827 * the packet might have been
4828 * fragmented and the reassembly code in ip_rput
4829 * does not restore the IP checksum.
4830 */
4831 ipha->ipha_hdr_checksum = 0;
4832 ipha->ipha_hdr_checksum = ip_csum_hdr(ipha);
4833
4834 switch (icmp_type) {
4835 case ICMP_DEST_UNREACHABLE:
4836 switch (icmp_code) {
4837 case ICMP_PROTOCOL_UNREACHABLE:
4838 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInUnknownProtos);
4839 ip_drop_input("ipIfStatsInUnknownProtos", mp, ill);
4840 break;
4841 case ICMP_PORT_UNREACHABLE:
4842 BUMP_MIB(ill->ill_ip_mib, udpIfStatsNoPorts);
4843 ip_drop_input("ipIfStatsNoPorts", mp, ill);
4844 break;
4845 }
4846
4847 icmp_unreachable(mp, icmp_code, ira);
4848 break;
4849 default:
4850 #ifdef DEBUG
4851 panic("ip_fanout_send_icmp_v4: wrong type");
4852 /*NOTREACHED*/
4853 #else
4854 freemsg(mp);
4855 break;
4856 #endif
4857 }
4858 }
4859
4860 /*
4861 * Used to send an ICMP error message when a packet is received for
4862 * a protocol that is not supported. The mblk passed as argument
4863 * is consumed by this function.
4864 */
4865 void
4866 ip_proto_not_sup(mblk_t *mp, ip_recv_attr_t *ira)
4867 {
4868 ipha_t *ipha;
4869
4870 ipha = (ipha_t *)mp->b_rptr;
4871 if (ira->ira_flags & IRAF_IS_IPV4) {
4872 ASSERT(IPH_HDR_VERSION(ipha) == IP_VERSION);
4873 ip_fanout_send_icmp_v4(mp, ICMP_DEST_UNREACHABLE,
4874 ICMP_PROTOCOL_UNREACHABLE, ira);
4875 } else {
4876 ASSERT(IPH_HDR_VERSION(ipha) == IPV6_VERSION);
4877 ip_fanout_send_icmp_v6(mp, ICMP6_PARAM_PROB,
4878 ICMP6_PARAMPROB_NEXTHEADER, ira);
4879 }
4880 }
4881
4882 /*
4883 * Deliver a rawip packet to the given conn, possibly applying ipsec policy.
4884 * Handles IPv4 and IPv6.
4885 * We are responsible for disposing of mp, such as by freemsg() or putnext()
4886 * Caller is responsible for dropping references to the conn.
4887 */
4888 void
4889 ip_fanout_proto_conn(conn_t *connp, mblk_t *mp, ipha_t *ipha, ip6_t *ip6h,
4890 ip_recv_attr_t *ira)
4891 {
4892 ill_t *ill = ira->ira_ill;
4893 ip_stack_t *ipst = ill->ill_ipst;
4894 ipsec_stack_t *ipss = ipst->ips_netstack->netstack_ipsec;
4895 boolean_t secure;
4896 uint_t protocol = ira->ira_protocol;
4897 iaflags_t iraflags = ira->ira_flags;
4898 queue_t *rq;
4899
4900 secure = iraflags & IRAF_IPSEC_SECURE;
4901
4902 rq = connp->conn_rq;
4903 if (IPCL_IS_NONSTR(connp) ? connp->conn_flow_cntrld : !canputnext(rq)) {
4904 switch (protocol) {
4905 case IPPROTO_ICMPV6:
4906 BUMP_MIB(ill->ill_icmp6_mib, ipv6IfIcmpInOverflows);
4907 break;
4908 case IPPROTO_ICMP:
4909 BUMP_MIB(&ipst->ips_icmp_mib, icmpInOverflows);
4910 break;
4911 default:
4912 BUMP_MIB(ill->ill_ip_mib, rawipIfStatsInOverflows);
4913 break;
4914 }
4915 freemsg(mp);
4916 return;
4917 }
4918
4919 ASSERT(!(IPCL_IS_IPTUN(connp)));
4920
4921 if (((iraflags & IRAF_IS_IPV4) ?
4922 CONN_INBOUND_POLICY_PRESENT(connp, ipss) :
4923 CONN_INBOUND_POLICY_PRESENT_V6(connp, ipss)) ||
4924 secure) {
4925 mp = ipsec_check_inbound_policy(mp, connp, ipha,
4926 ip6h, ira);
4927 if (mp == NULL) {
4928 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
4929 /* Note that mp is NULL */
4930 ip_drop_input("ipIfStatsInDiscards", mp, ill);
4931 return;
4932 }
4933 }
4934
4935 if (iraflags & IRAF_ICMP_ERROR) {
4936 (connp->conn_recvicmp)(connp, mp, NULL, ira);
4937 } else {
4938 ill_t *rill = ira->ira_rill;
4939
4940 BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCInDelivers);
4941 ira->ira_ill = ira->ira_rill = NULL;
4942 /* Send it upstream */
4943 (connp->conn_recv)(connp, mp, NULL, ira);
4944 ira->ira_ill = ill;
4945 ira->ira_rill = rill;
4946 }
4947 }
4948
4949 /*
4950 * Handle protocols with which IP is less intimate. There
4951 * can be more than one stream bound to a particular
4952 * protocol. When this is the case, normally each one gets a copy
4953 * of any incoming packets.
4954 *
4955 * IPsec NOTE :
4956 *
4957 * Don't allow a secure packet going up a non-secure connection.
4958 * We don't allow this because
4959 *
4960 * 1) Reply might go out in clear which will be dropped at
4961 * the sending side.
4962 * 2) If the reply goes out in clear it will give the
4963 * adversary enough information for getting the key in
4964 * most of the cases.
4965 *
4966 * Moreover getting a secure packet when we expect clear
4967 * implies that SA's were added without checking for
4968 * policy on both ends. This should not happen once ISAKMP
4969 * is used to negotiate SAs as SAs will be added only after
4970 * verifying the policy.
4971 *
4972 * Zones notes:
4973 * Earlier in ip_input on a system with multiple shared-IP zones we
4974 * duplicate the multicast and broadcast packets and send them up
4975 * with each explicit zoneid that exists on that ill.
4976 * This means that here we can match the zoneid with SO_ALLZONES being special.
4977 */
4978 void
4979 ip_fanout_proto_v4(mblk_t *mp, ipha_t *ipha, ip_recv_attr_t *ira)
4980 {
4981 mblk_t *mp1;
4982 ipaddr_t laddr;
4983 conn_t *connp, *first_connp, *next_connp;
4984 connf_t *connfp;
4985 ill_t *ill = ira->ira_ill;
4986 ip_stack_t *ipst = ill->ill_ipst;
4987
4988 laddr = ipha->ipha_dst;
4989
4990 connfp = &ipst->ips_ipcl_proto_fanout_v4[ira->ira_protocol];
4991 mutex_enter(&connfp->connf_lock);
4992 connp = connfp->connf_head;
4993 for (connp = connfp->connf_head; connp != NULL;
4994 connp = connp->conn_next) {
4995 /* Note: IPCL_PROTO_MATCH includes conn_wantpacket */
4996 if (IPCL_PROTO_MATCH(connp, ira, ipha) &&
4997 (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
4998 tsol_receive_local(mp, &laddr, IPV4_VERSION, ira, connp))) {
4999 break;
5000 }
5001 }
5002
5003 if (connp == NULL) {
5004 /*
5005 * No one bound to these addresses. Is
5006 * there a client that wants all
5007 * unclaimed datagrams?
5008 */
5009 mutex_exit(&connfp->connf_lock);
5010 ip_fanout_send_icmp_v4(mp, ICMP_DEST_UNREACHABLE,
5011 ICMP_PROTOCOL_UNREACHABLE, ira);
5012 return;
5013 }
5014
5015 ASSERT(IPCL_IS_NONSTR(connp) || connp->conn_rq != NULL);
5016
5017 CONN_INC_REF(connp);
5018 first_connp = connp;
5019 connp = connp->conn_next;
5020
5021 for (;;) {
5022 while (connp != NULL) {
5023 /* Note: IPCL_PROTO_MATCH includes conn_wantpacket */
5024 if (IPCL_PROTO_MATCH(connp, ira, ipha) &&
5025 (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5026 tsol_receive_local(mp, &laddr, IPV4_VERSION,
5027 ira, connp)))
5028 break;
5029 connp = connp->conn_next;
5030 }
5031
5032 if (connp == NULL) {
5033 /* No more interested clients */
5034 connp = first_connp;
5035 break;
5036 }
5037 if (((mp1 = dupmsg(mp)) == NULL) &&
5038 ((mp1 = copymsg(mp)) == NULL)) {
5039 /* Memory allocation failed */
5040 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
5041 ip_drop_input("ipIfStatsInDiscards", mp, ill);
5042 connp = first_connp;
5043 break;
5044 }
5045
5046 CONN_INC_REF(connp);
5047 mutex_exit(&connfp->connf_lock);
5048
5049 ip_fanout_proto_conn(connp, mp1, (ipha_t *)mp1->b_rptr, NULL,
5050 ira);
5051
5052 mutex_enter(&connfp->connf_lock);
5053 /* Follow the next pointer before releasing the conn. */
5054 next_connp = connp->conn_next;
5055 CONN_DEC_REF(connp);
5056 connp = next_connp;
5057 }
5058
5059 /* Last one. Send it upstream. */
5060 mutex_exit(&connfp->connf_lock);
5061
5062 ip_fanout_proto_conn(connp, mp, ipha, NULL, ira);
5063
5064 CONN_DEC_REF(connp);
5065 }
5066
5067 /*
5068 * If we have a IPsec NAT-Traversal packet, strip the zero-SPI or
5069 * pass it along to ESP if the SPI is non-zero. Returns the mblk if the mblk
5070 * is not consumed.
5071 *
5072 * One of three things can happen, all of which affect the passed-in mblk:
5073 *
5074 * 1.) The packet is stock UDP and gets its zero-SPI stripped. Return mblk..
5075 *
5076 * 2.) The packet is ESP-in-UDP, gets transformed into an equivalent
5077 * ESP packet, and is passed along to ESP for consumption. Return NULL.
5078 *
5079 * 3.) The packet is an ESP-in-UDP Keepalive. Drop it and return NULL.
5080 */
5081 mblk_t *
5082 zero_spi_check(mblk_t *mp, ip_recv_attr_t *ira)
5083 {
5084 int shift, plen, iph_len;
5085 ipha_t *ipha;
5086 udpha_t *udpha;
5087 uint32_t *spi;
5088 uint32_t esp_ports;
5089 uint8_t *orptr;
5090 ip_stack_t *ipst = ira->ira_ill->ill_ipst;
5091 ipsec_stack_t *ipss = ipst->ips_netstack->netstack_ipsec;
5092
5093 ipha = (ipha_t *)mp->b_rptr;
5094 iph_len = ira->ira_ip_hdr_length;
5095 plen = ira->ira_pktlen;
5096
5097 if (plen - iph_len - sizeof (udpha_t) < sizeof (uint32_t)) {
5098 /*
5099 * Most likely a keepalive for the benefit of an intervening
5100 * NAT. These aren't for us, per se, so drop it.
5101 *
5102 * RFC 3947/8 doesn't say for sure what to do for 2-3
5103 * byte packets (keepalives are 1-byte), but we'll drop them
5104 * also.
5105 */
5106 ip_drop_packet(mp, B_TRUE, ira->ira_ill,
5107 DROPPER(ipss, ipds_esp_nat_t_ka), &ipss->ipsec_dropper);
5108 return (NULL);
5109 }
5110
5111 if (MBLKL(mp) < iph_len + sizeof (udpha_t) + sizeof (*spi)) {
5112 /* might as well pull it all up - it might be ESP. */
5113 if (!pullupmsg(mp, -1)) {
5114 ip_drop_packet(mp, B_TRUE, ira->ira_ill,
5115 DROPPER(ipss, ipds_esp_nomem),
5116 &ipss->ipsec_dropper);
5117 return (NULL);
5118 }
5119
5120 ipha = (ipha_t *)mp->b_rptr;
5121 }
5122 spi = (uint32_t *)(mp->b_rptr + iph_len + sizeof (udpha_t));
5123 if (*spi == 0) {
5124 /* UDP packet - remove 0-spi. */
5125 shift = sizeof (uint32_t);
5126 } else {
5127 /* ESP-in-UDP packet - reduce to ESP. */
5128 ipha->ipha_protocol = IPPROTO_ESP;
5129 shift = sizeof (udpha_t);
5130 }
5131
5132 /* Fix IP header */
5133 ira->ira_pktlen = (plen - shift);
5134 ipha->ipha_length = htons(ira->ira_pktlen);
5135 ipha->ipha_hdr_checksum = 0;
5136
5137 orptr = mp->b_rptr;
5138 mp->b_rptr += shift;
5139
5140 udpha = (udpha_t *)(orptr + iph_len);
5141 if (*spi == 0) {
5142 ASSERT((uint8_t *)ipha == orptr);
5143 udpha->uha_length = htons(plen - shift - iph_len);
5144 iph_len += sizeof (udpha_t); /* For the call to ovbcopy(). */
5145 esp_ports = 0;
5146 } else {
5147 esp_ports = *((uint32_t *)udpha);
5148 ASSERT(esp_ports != 0);
5149 }
5150 ovbcopy(orptr, orptr + shift, iph_len);
5151 if (esp_ports != 0) /* Punt up for ESP processing. */ {
5152 ipha = (ipha_t *)(orptr + shift);
5153
5154 ira->ira_flags |= IRAF_ESP_UDP_PORTS;
5155 ira->ira_esp_udp_ports = esp_ports;
5156 ip_fanout_v4(mp, ipha, ira);
5157 return (NULL);
5158 }
5159 return (mp);
5160 }
5161
5162 /*
5163 * Deliver a udp packet to the given conn, possibly applying ipsec policy.
5164 * Handles IPv4 and IPv6.
5165 * We are responsible for disposing of mp, such as by freemsg() or putnext()
5166 * Caller is responsible for dropping references to the conn.
5167 */
5168 void
5169 ip_fanout_udp_conn(conn_t *connp, mblk_t *mp, ipha_t *ipha, ip6_t *ip6h,
5170 ip_recv_attr_t *ira)
5171 {
5172 ill_t *ill = ira->ira_ill;
5173 ip_stack_t *ipst = ill->ill_ipst;
5174 ipsec_stack_t *ipss = ipst->ips_netstack->netstack_ipsec;
5175 boolean_t secure;
5176 iaflags_t iraflags = ira->ira_flags;
5177
5178 secure = iraflags & IRAF_IPSEC_SECURE;
5179
5180 if (IPCL_IS_NONSTR(connp) ? connp->conn_flow_cntrld :
5181 !canputnext(connp->conn_rq)) {
5182 BUMP_MIB(ill->ill_ip_mib, udpIfStatsInOverflows);
5183 freemsg(mp);
5184 return;
5185 }
5186
5187 if (((iraflags & IRAF_IS_IPV4) ?
5188 CONN_INBOUND_POLICY_PRESENT(connp, ipss) :
5189 CONN_INBOUND_POLICY_PRESENT_V6(connp, ipss)) ||
5190 secure) {
5191 mp = ipsec_check_inbound_policy(mp, connp, ipha,
5192 ip6h, ira);
5193 if (mp == NULL) {
5194 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
5195 /* Note that mp is NULL */
5196 ip_drop_input("ipIfStatsInDiscards", mp, ill);
5197 return;
5198 }
5199 }
5200
5201 /*
5202 * Since this code is not used for UDP unicast we don't need a NAT_T
5203 * check. Only ip_fanout_v4 has that check.
5204 */
5205 if (ira->ira_flags & IRAF_ICMP_ERROR) {
5206 (connp->conn_recvicmp)(connp, mp, NULL, ira);
5207 } else {
5208 ill_t *rill = ira->ira_rill;
5209
5210 BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCInDelivers);
5211 ira->ira_ill = ira->ira_rill = NULL;
5212 /* Send it upstream */
5213 (connp->conn_recv)(connp, mp, NULL, ira);
5214 ira->ira_ill = ill;
5215 ira->ira_rill = rill;
5216 }
5217 }
5218
5219 /*
5220 * Fanout for UDP packets that are multicast or broadcast, and ICMP errors.
5221 * (Unicast fanout is handled in ip_input_v4.)
5222 *
5223 * If SO_REUSEADDR is set all multicast and broadcast packets
5224 * will be delivered to all conns bound to the same port.
5225 *
5226 * If there is at least one matching AF_INET receiver, then we will
5227 * ignore any AF_INET6 receivers.
5228 * In the special case where an AF_INET socket binds to 0.0.0.0/<port> and an
5229 * AF_INET6 socket binds to ::/<port>, only the AF_INET socket receives the IPv4
5230 * packets.
5231 *
5232 * Zones notes:
5233 * Earlier in ip_input on a system with multiple shared-IP zones we
5234 * duplicate the multicast and broadcast packets and send them up
5235 * with each explicit zoneid that exists on that ill.
5236 * This means that here we can match the zoneid with SO_ALLZONES being special.
5237 */
5238 void
5239 ip_fanout_udp_multi_v4(mblk_t *mp, ipha_t *ipha, uint16_t lport, uint16_t fport,
5240 ip_recv_attr_t *ira)
5241 {
5242 ipaddr_t laddr;
5243 in6_addr_t v6faddr;
5244 conn_t *connp;
5245 connf_t *connfp;
5246 ipaddr_t faddr;
5247 ill_t *ill = ira->ira_ill;
5248 ip_stack_t *ipst = ill->ill_ipst;
5249
5250 ASSERT(ira->ira_flags & (IRAF_MULTIBROADCAST|IRAF_ICMP_ERROR));
5251
5252 laddr = ipha->ipha_dst;
5253 faddr = ipha->ipha_src;
5254
5255 connfp = &ipst->ips_ipcl_udp_fanout[IPCL_UDP_HASH(lport, ipst)];
5256 mutex_enter(&connfp->connf_lock);
5257 connp = connfp->connf_head;
5258
5259 /*
5260 * If SO_REUSEADDR has been set on the first we send the
5261 * packet to all clients that have joined the group and
5262 * match the port.
5263 */
5264 while (connp != NULL) {
5265 if ((IPCL_UDP_MATCH(connp, lport, laddr, fport, faddr)) &&
5266 conn_wantpacket(connp, ira, ipha) &&
5267 (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5268 tsol_receive_local(mp, &laddr, IPV4_VERSION, ira, connp)))
5269 break;
5270 connp = connp->conn_next;
5271 }
5272
5273 if (connp == NULL)
5274 goto notfound;
5275
5276 CONN_INC_REF(connp);
5277
5278 if (connp->conn_reuseaddr) {
5279 conn_t *first_connp = connp;
5280 conn_t *next_connp;
5281 mblk_t *mp1;
5282
5283 connp = connp->conn_next;
5284 for (;;) {
5285 while (connp != NULL) {
5286 if (IPCL_UDP_MATCH(connp, lport, laddr,
5287 fport, faddr) &&
5288 conn_wantpacket(connp, ira, ipha) &&
5289 (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5290 tsol_receive_local(mp, &laddr, IPV4_VERSION,
5291 ira, connp)))
5292 break;
5293 connp = connp->conn_next;
5294 }
5295 if (connp == NULL) {
5296 /* No more interested clients */
5297 connp = first_connp;
5298 break;
5299 }
5300 if (((mp1 = dupmsg(mp)) == NULL) &&
5301 ((mp1 = copymsg(mp)) == NULL)) {
5302 /* Memory allocation failed */
5303 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
5304 ip_drop_input("ipIfStatsInDiscards", mp, ill);
5305 connp = first_connp;
5306 break;
5307 }
5308 CONN_INC_REF(connp);
5309 mutex_exit(&connfp->connf_lock);
5310
5311 IP_STAT(ipst, ip_udp_fanmb);
5312 ip_fanout_udp_conn(connp, mp1, (ipha_t *)mp1->b_rptr,
5313 NULL, ira);
5314 mutex_enter(&connfp->connf_lock);
5315 /* Follow the next pointer before releasing the conn */
5316 next_connp = connp->conn_next;
5317 CONN_DEC_REF(connp);
5318 connp = next_connp;
5319 }
5320 }
5321
5322 /* Last one. Send it upstream. */
5323 mutex_exit(&connfp->connf_lock);
5324 IP_STAT(ipst, ip_udp_fanmb);
5325 ip_fanout_udp_conn(connp, mp, ipha, NULL, ira);
5326 CONN_DEC_REF(connp);
5327 return;
5328
5329 notfound:
5330 mutex_exit(&connfp->connf_lock);
5331 /*
5332 * IPv6 endpoints bound to multicast IPv4-mapped addresses
5333 * have already been matched above, since they live in the IPv4
5334 * fanout tables. This implies we only need to
5335 * check for IPv6 in6addr_any endpoints here.
5336 * Thus we compare using ipv6_all_zeros instead of the destination
5337 * address, except for the multicast group membership lookup which
5338 * uses the IPv4 destination.
5339 */
5340 IN6_IPADDR_TO_V4MAPPED(ipha->ipha_src, &v6faddr);
5341 connfp = &ipst->ips_ipcl_udp_fanout[IPCL_UDP_HASH(lport, ipst)];
5342 mutex_enter(&connfp->connf_lock);
5343 connp = connfp->connf_head;
5344 /*
5345 * IPv4 multicast packet being delivered to an AF_INET6
5346 * in6addr_any endpoint.
5347 * Need to check conn_wantpacket(). Note that we use conn_wantpacket()
5348 * and not conn_wantpacket_v6() since any multicast membership is
5349 * for an IPv4-mapped multicast address.
5350 */
5351 while (connp != NULL) {
5352 if (IPCL_UDP_MATCH_V6(connp, lport, ipv6_all_zeros,
5353 fport, v6faddr) &&
5354 conn_wantpacket(connp, ira, ipha) &&
5355 (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5356 tsol_receive_local(mp, &laddr, IPV4_VERSION, ira, connp)))
5357 break;
5358 connp = connp->conn_next;
5359 }
5360
5361 if (connp == NULL) {
5362 /*
5363 * No one bound to this port. Is
5364 * there a client that wants all
5365 * unclaimed datagrams?
5366 */
5367 mutex_exit(&connfp->connf_lock);
5368
5369 if (ipst->ips_ipcl_proto_fanout_v4[IPPROTO_UDP].connf_head !=
5370 NULL) {
5371 ASSERT(ira->ira_protocol == IPPROTO_UDP);
5372 ip_fanout_proto_v4(mp, ipha, ira);
5373 } else {
5374 /*
5375 * We used to attempt to send an icmp error here, but
5376 * since this is known to be a multicast packet
5377 * and we don't send icmp errors in response to
5378 * multicast, just drop the packet and give up sooner.
5379 */
5380 BUMP_MIB(ill->ill_ip_mib, udpIfStatsNoPorts);
5381 freemsg(mp);
5382 }
5383 return;
5384 }
5385 ASSERT(IPCL_IS_NONSTR(connp) || connp->conn_rq != NULL);
5386
5387 /*
5388 * If SO_REUSEADDR has been set on the first we send the
5389 * packet to all clients that have joined the group and
5390 * match the port.
5391 */
5392 if (connp->conn_reuseaddr) {
5393 conn_t *first_connp = connp;
5394 conn_t *next_connp;
5395 mblk_t *mp1;
5396
5397 CONN_INC_REF(connp);
5398 connp = connp->conn_next;
5399 for (;;) {
5400 while (connp != NULL) {
5401 if (IPCL_UDP_MATCH_V6(connp, lport,
5402 ipv6_all_zeros, fport, v6faddr) &&
5403 conn_wantpacket(connp, ira, ipha) &&
5404 (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5405 tsol_receive_local(mp, &laddr, IPV4_VERSION,
5406 ira, connp)))
5407 break;
5408 connp = connp->conn_next;
5409 }
5410 if (connp == NULL) {
5411 /* No more interested clients */
5412 connp = first_connp;
5413 break;
5414 }
5415 if (((mp1 = dupmsg(mp)) == NULL) &&
5416 ((mp1 = copymsg(mp)) == NULL)) {
5417 /* Memory allocation failed */
5418 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
5419 ip_drop_input("ipIfStatsInDiscards", mp, ill);
5420 connp = first_connp;
5421 break;
5422 }
5423 CONN_INC_REF(connp);
5424 mutex_exit(&connfp->connf_lock);
5425
5426 IP_STAT(ipst, ip_udp_fanmb);
5427 ip_fanout_udp_conn(connp, mp1, (ipha_t *)mp1->b_rptr,
5428 NULL, ira);
5429 mutex_enter(&connfp->connf_lock);
5430 /* Follow the next pointer before releasing the conn */
5431 next_connp = connp->conn_next;
5432 CONN_DEC_REF(connp);
5433 connp = next_connp;
5434 }
5435 }
5436
5437 /* Last one. Send it upstream. */
5438 mutex_exit(&connfp->connf_lock);
5439 IP_STAT(ipst, ip_udp_fanmb);
5440 ip_fanout_udp_conn(connp, mp, ipha, NULL, ira);
5441 CONN_DEC_REF(connp);
5442 }
5443
5444 /*
5445 * Split an incoming packet's IPv4 options into the label and the other options.
5446 * If 'allocate' is set it does memory allocation for the ip_pkt_t, including
5447 * clearing out any leftover label or options.
5448 * Otherwise it just makes ipp point into the packet.
5449 *
5450 * Returns zero if ok; ENOMEM if the buffer couldn't be allocated.
5451 */
5452 int
5453 ip_find_hdr_v4(ipha_t *ipha, ip_pkt_t *ipp, boolean_t allocate)
5454 {
5455 uchar_t *opt;
5456 uint32_t totallen;
5457 uint32_t optval;
5458 uint32_t optlen;
5459
5460 ipp->ipp_fields |= IPPF_HOPLIMIT | IPPF_TCLASS | IPPF_ADDR;
5461 ipp->ipp_hoplimit = ipha->ipha_ttl;
5462 ipp->ipp_type_of_service = ipha->ipha_type_of_service;
5463 IN6_IPADDR_TO_V4MAPPED(ipha->ipha_dst, &ipp->ipp_addr);
5464
5465 /*
5466 * Get length (in 4 byte octets) of IP header options.
5467 */
5468 totallen = ipha->ipha_version_and_hdr_length -
5469 (uint8_t)((IP_VERSION << 4) + IP_SIMPLE_HDR_LENGTH_IN_WORDS);
5470
5471 if (totallen == 0) {
5472 if (!allocate)
5473 return (0);
5474
5475 /* Clear out anything from a previous packet */
5476 if (ipp->ipp_fields & IPPF_IPV4_OPTIONS) {
5477 kmem_free(ipp->ipp_ipv4_options,
5478 ipp->ipp_ipv4_options_len);
5479 ipp->ipp_ipv4_options = NULL;
5480 ipp->ipp_ipv4_options_len = 0;
5481 ipp->ipp_fields &= ~IPPF_IPV4_OPTIONS;
5482 }
5483 if (ipp->ipp_fields & IPPF_LABEL_V4) {
5484 kmem_free(ipp->ipp_label_v4, ipp->ipp_label_len_v4);
5485 ipp->ipp_label_v4 = NULL;
5486 ipp->ipp_label_len_v4 = 0;
5487 ipp->ipp_fields &= ~IPPF_LABEL_V4;
5488 }
5489 return (0);
5490 }
5491
5492 totallen <<= 2;
5493 opt = (uchar_t *)&ipha[1];
5494 if (!is_system_labeled()) {
5495
5496 copyall:
5497 if (!allocate) {
5498 if (totallen != 0) {
5499 ipp->ipp_ipv4_options = opt;
5500 ipp->ipp_ipv4_options_len = totallen;
5501 ipp->ipp_fields |= IPPF_IPV4_OPTIONS;
5502 }
5503 return (0);
5504 }
5505 /* Just copy all of options */
5506 if (ipp->ipp_fields & IPPF_IPV4_OPTIONS) {
5507 if (totallen == ipp->ipp_ipv4_options_len) {
5508 bcopy(opt, ipp->ipp_ipv4_options, totallen);
5509 return (0);
5510 }
5511 kmem_free(ipp->ipp_ipv4_options,
5512 ipp->ipp_ipv4_options_len);
5513 ipp->ipp_ipv4_options = NULL;
5514 ipp->ipp_ipv4_options_len = 0;
5515 ipp->ipp_fields &= ~IPPF_IPV4_OPTIONS;
5516 }
5517 if (totallen == 0)
5518 return (0);
5519
5520 ipp->ipp_ipv4_options = kmem_alloc(totallen, KM_NOSLEEP);
5521 if (ipp->ipp_ipv4_options == NULL)
5522 return (ENOMEM);
5523 ipp->ipp_ipv4_options_len = totallen;
5524 ipp->ipp_fields |= IPPF_IPV4_OPTIONS;
5525 bcopy(opt, ipp->ipp_ipv4_options, totallen);
5526 return (0);
5527 }
5528
5529 if (allocate && (ipp->ipp_fields & IPPF_LABEL_V4)) {
5530 kmem_free(ipp->ipp_label_v4, ipp->ipp_label_len_v4);
5531 ipp->ipp_label_v4 = NULL;
5532 ipp->ipp_label_len_v4 = 0;
5533 ipp->ipp_fields &= ~IPPF_LABEL_V4;
5534 }
5535
5536 /*
5537 * Search for CIPSO option.
5538 * We assume CIPSO is first in options if it is present.
5539 * If it isn't, then ipp_opt_ipv4_options will not include the options
5540 * prior to the CIPSO option.
5541 */
5542 while (totallen != 0) {
5543 switch (optval = opt[IPOPT_OPTVAL]) {
5544 case IPOPT_EOL:
5545 return (0);
5546 case IPOPT_NOP:
5547 optlen = 1;
5548 break;
5549 default:
5550 if (totallen <= IPOPT_OLEN)
5551 return (EINVAL);
5552 optlen = opt[IPOPT_OLEN];
5553 if (optlen < 2)
5554 return (EINVAL);
5555 }
5556 if (optlen > totallen)
5557 return (EINVAL);
5558
5559 switch (optval) {
5560 case IPOPT_COMSEC:
5561 if (!allocate) {
5562 ipp->ipp_label_v4 = opt;
5563 ipp->ipp_label_len_v4 = optlen;
5564 ipp->ipp_fields |= IPPF_LABEL_V4;
5565 } else {
5566 ipp->ipp_label_v4 = kmem_alloc(optlen,
5567 KM_NOSLEEP);
5568 if (ipp->ipp_label_v4 == NULL)
5569 return (ENOMEM);
5570 ipp->ipp_label_len_v4 = optlen;
5571 ipp->ipp_fields |= IPPF_LABEL_V4;
5572 bcopy(opt, ipp->ipp_label_v4, optlen);
5573 }
5574 totallen -= optlen;
5575 opt += optlen;
5576
5577 /* Skip padding bytes until we get to a multiple of 4 */
5578 while ((totallen & 3) != 0 && opt[0] == IPOPT_NOP) {
5579 totallen--;
5580 opt++;
5581 }
5582 /* Remaining as ipp_ipv4_options */
5583 goto copyall;
5584 }
5585 totallen -= optlen;
5586 opt += optlen;
5587 }
5588 /* No CIPSO found; return everything as ipp_ipv4_options */
5589 totallen = ipha->ipha_version_and_hdr_length -
5590 (uint8_t)((IP_VERSION << 4) + IP_SIMPLE_HDR_LENGTH_IN_WORDS);
5591 totallen <<= 2;
5592 opt = (uchar_t *)&ipha[1];
5593 goto copyall;
5594 }
5595
5596 /*
5597 * Efficient versions of lookup for an IRE when we only
5598 * match the address.
5599 * For RTF_REJECT or BLACKHOLE we return IRE_NOROUTE.
5600 * Does not handle multicast addresses.
5601 */
5602 uint_t
5603 ip_type_v4(ipaddr_t addr, ip_stack_t *ipst)
5604 {
5605 ire_t *ire;
5606 uint_t result;
5607
5608 ire = ire_ftable_lookup_simple_v4(addr, 0, ipst, NULL);
5609 ASSERT(ire != NULL);
5610 if (ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE))
5611 result = IRE_NOROUTE;
5612 else
5613 result = ire->ire_type;
5614 ire_refrele(ire);
5615 return (result);
5616 }
5617
5618 /*
5619 * Efficient versions of lookup for an IRE when we only
5620 * match the address.
5621 * For RTF_REJECT or BLACKHOLE we return IRE_NOROUTE.
5622 * Does not handle multicast addresses.
5623 */
5624 uint_t
5625 ip_type_v6(const in6_addr_t *addr, ip_stack_t *ipst)
5626 {
5627 ire_t *ire;
5628 uint_t result;
5629
5630 ire = ire_ftable_lookup_simple_v6(addr, 0, ipst, NULL);
5631 ASSERT(ire != NULL);
5632 if (ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE))
5633 result = IRE_NOROUTE;
5634 else
5635 result = ire->ire_type;
5636 ire_refrele(ire);
5637 return (result);
5638 }
5639
5640 /*
5641 * Nobody should be sending
5642 * packets up this stream
5643 */
5644 static void
5645 ip_lrput(queue_t *q, mblk_t *mp)
5646 {
5647 switch (mp->b_datap->db_type) {
5648 case M_FLUSH:
5649 /* Turn around */
5650 if (*mp->b_rptr & FLUSHW) {
5651 *mp->b_rptr &= ~FLUSHR;
5652 qreply(q, mp);
5653 return;
5654 }
5655 break;
5656 }
5657 freemsg(mp);
5658 }
5659
5660 /* Nobody should be sending packets down this stream */
5661 /* ARGSUSED */
5662 void
5663 ip_lwput(queue_t *q, mblk_t *mp)
5664 {
5665 freemsg(mp);
5666 }
5667
5668 /*
5669 * Move the first hop in any source route to ipha_dst and remove that part of
5670 * the source route. Called by other protocols. Errors in option formatting
5671 * are ignored - will be handled by ip_output_options. Return the final
5672 * destination (either ipha_dst or the last entry in a source route.)
5673 */
5674 ipaddr_t
5675 ip_massage_options(ipha_t *ipha, netstack_t *ns)
5676 {
5677 ipoptp_t opts;
5678 uchar_t *opt;
5679 uint8_t optval;
5680 uint8_t optlen;
5681 ipaddr_t dst;
5682 int i;
5683 ip_stack_t *ipst = ns->netstack_ip;
5684
5685 ip2dbg(("ip_massage_options\n"));
5686 dst = ipha->ipha_dst;
5687 for (optval = ipoptp_first(&opts, ipha);
5688 optval != IPOPT_EOL;
5689 optval = ipoptp_next(&opts)) {
5690 opt = opts.ipoptp_cur;
5691 switch (optval) {
5692 uint8_t off;
5693 case IPOPT_SSRR:
5694 case IPOPT_LSRR:
5695 if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
5696 ip1dbg(("ip_massage_options: bad src route\n"));
5697 break;
5698 }
5699 optlen = opts.ipoptp_len;
5700 off = opt[IPOPT_OFFSET];
5701 off--;
5702 redo_srr:
5703 if (optlen < IP_ADDR_LEN ||
5704 off > optlen - IP_ADDR_LEN) {
5705 /* End of source route */
5706 ip1dbg(("ip_massage_options: end of SR\n"));
5707 break;
5708 }
5709 bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
5710 ip1dbg(("ip_massage_options: next hop 0x%x\n",
5711 ntohl(dst)));
5712 /*
5713 * Check if our address is present more than
5714 * once as consecutive hops in source route.
5715 * XXX verify per-interface ip_forwarding
5716 * for source route?
5717 */
5718 if (ip_type_v4(dst, ipst) == IRE_LOCAL) {
5719 off += IP_ADDR_LEN;
5720 goto redo_srr;
5721 }
5722 if (dst == htonl(INADDR_LOOPBACK)) {
5723 ip1dbg(("ip_massage_options: loopback addr in "
5724 "source route!\n"));
5725 break;
5726 }
5727 /*
5728 * Update ipha_dst to be the first hop and remove the
5729 * first hop from the source route (by overwriting
5730 * part of the option with NOP options).
5731 */
5732 ipha->ipha_dst = dst;
5733 /* Put the last entry in dst */
5734 off = ((optlen - IP_ADDR_LEN - 3) & ~(IP_ADDR_LEN-1)) +
5735 3;
5736 bcopy(&opt[off], &dst, IP_ADDR_LEN);
5737
5738 ip1dbg(("ip_massage_options: last hop 0x%x\n",
5739 ntohl(dst)));
5740 /* Move down and overwrite */
5741 opt[IP_ADDR_LEN] = opt[0];
5742 opt[IP_ADDR_LEN+1] = opt[IPOPT_OLEN] - IP_ADDR_LEN;
5743 opt[IP_ADDR_LEN+2] = opt[IPOPT_OFFSET];
5744 for (i = 0; i < IP_ADDR_LEN; i++)
5745 opt[i] = IPOPT_NOP;
5746 break;
5747 }
5748 }
5749 return (dst);
5750 }
5751
5752 /*
5753 * Return the network mask
5754 * associated with the specified address.
5755 */
5756 ipaddr_t
5757 ip_net_mask(ipaddr_t addr)
5758 {
5759 uchar_t *up = (uchar_t *)&addr;
5760 ipaddr_t mask = 0;
5761 uchar_t *maskp = (uchar_t *)&mask;
5762
5763 #if defined(__i386) || defined(__amd64)
5764 #define TOTALLY_BRAIN_DAMAGED_C_COMPILER
5765 #endif
5766 #ifdef TOTALLY_BRAIN_DAMAGED_C_COMPILER
5767 maskp[0] = maskp[1] = maskp[2] = maskp[3] = 0;
5768 #endif
5769 if (CLASSD(addr)) {
5770 maskp[0] = 0xF0;
5771 return (mask);
5772 }
5773
5774 /* We assume Class E default netmask to be 32 */
5775 if (CLASSE(addr))
5776 return (0xffffffffU);
5777
5778 if (addr == 0)
5779 return (0);
5780 maskp[0] = 0xFF;
5781 if ((up[0] & 0x80) == 0)
5782 return (mask);
5783
5784 maskp[1] = 0xFF;
5785 if ((up[0] & 0xC0) == 0x80)
5786 return (mask);
5787
5788 maskp[2] = 0xFF;
5789 if ((up[0] & 0xE0) == 0xC0)
5790 return (mask);
5791
5792 /* Otherwise return no mask */
5793 return ((ipaddr_t)0);
5794 }
5795
5796 /* Name/Value Table Lookup Routine */
5797 char *
5798 ip_nv_lookup(nv_t *nv, int value)
5799 {
5800 if (!nv)
5801 return (NULL);
5802 for (; nv->nv_name; nv++) {
5803 if (nv->nv_value == value)
5804 return (nv->nv_name);
5805 }
5806 return ("unknown");
5807 }
5808
5809 static int
5810 ip_wait_for_info_ack(ill_t *ill)
5811 {
5812 int err;
5813
5814 mutex_enter(&ill->ill_lock);
5815 while (ill->ill_state_flags & ILL_LL_SUBNET_PENDING) {
5816 /*
5817 * Return value of 0 indicates a pending signal.
5818 */
5819 err = cv_wait_sig(&ill->ill_cv, &ill->ill_lock);
5820 if (err == 0) {
5821 mutex_exit(&ill->ill_lock);
5822 return (EINTR);
5823 }
5824 }
5825 mutex_exit(&ill->ill_lock);
5826 /*
5827 * ip_rput_other could have set an error in ill_error on
5828 * receipt of M_ERROR.
5829 */
5830 return (ill->ill_error);
5831 }
5832
5833 /*
5834 * This is a module open, i.e. this is a control stream for access
5835 * to a DLPI device. We allocate an ill_t as the instance data in
5836 * this case.
5837 */
5838 static int
5839 ip_modopen(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
5840 {
5841 ill_t *ill;
5842 int err;
5843 zoneid_t zoneid;
5844 netstack_t *ns;
5845 ip_stack_t *ipst;
5846
5847 /*
5848 * Prevent unprivileged processes from pushing IP so that
5849 * they can't send raw IP.
5850 */
5851 if (secpolicy_net_rawaccess(credp) != 0)
5852 return (EPERM);
5853
5854 ns = netstack_find_by_cred(credp);
5855 ASSERT(ns != NULL);
5856 ipst = ns->netstack_ip;
5857 ASSERT(ipst != NULL);
5858
5859 /*
5860 * For exclusive stacks we set the zoneid to zero
5861 * to make IP operate as if in the global zone.
5862 */
5863 if (ipst->ips_netstack->netstack_stackid != GLOBAL_NETSTACKID)
5864 zoneid = GLOBAL_ZONEID;
5865 else
5866 zoneid = crgetzoneid(credp);
5867
5868 ill = (ill_t *)mi_open_alloc_sleep(sizeof (ill_t));
5869 q->q_ptr = WR(q)->q_ptr = ill;
5870 ill->ill_ipst = ipst;
5871 ill->ill_zoneid = zoneid;
5872
5873 /*
5874 * ill_init initializes the ill fields and then sends down
5875 * down a DL_INFO_REQ after calling qprocson.
5876 */
5877 err = ill_init(q, ill);
5878
5879 if (err != 0) {
5880 mi_free(ill);
5881 netstack_rele(ipst->ips_netstack);
5882 q->q_ptr = NULL;
5883 WR(q)->q_ptr = NULL;
5884 return (err);
5885 }
5886
5887 /*
5888 * Wait for the DL_INFO_ACK if a DL_INFO_REQ was sent.
5889 *
5890 * ill_init initializes the ipsq marking this thread as
5891 * writer
5892 */
5893 ipsq_exit(ill->ill_phyint->phyint_ipsq);
5894 err = ip_wait_for_info_ack(ill);
5895 if (err == 0)
5896 ill->ill_credp = credp;
5897 else
5898 goto fail;
5899
5900 crhold(credp);
5901
5902 mutex_enter(&ipst->ips_ip_mi_lock);
5903 err = mi_open_link(&ipst->ips_ip_g_head, (IDP)q->q_ptr, devp, flag,
5904 sflag, credp);
5905 mutex_exit(&ipst->ips_ip_mi_lock);
5906 fail:
5907 if (err) {
5908 (void) ip_close(q, 0);
5909 return (err);
5910 }
5911 return (0);
5912 }
5913
5914 /* For /dev/ip aka AF_INET open */
5915 int
5916 ip_openv4(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
5917 {
5918 return (ip_open(q, devp, flag, sflag, credp, B_FALSE));
5919 }
5920
5921 /* For /dev/ip6 aka AF_INET6 open */
5922 int
5923 ip_openv6(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
5924 {
5925 return (ip_open(q, devp, flag, sflag, credp, B_TRUE));
5926 }
5927
5928 /* IP open routine. */
5929 int
5930 ip_open(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp,
5931 boolean_t isv6)
5932 {
5933 conn_t *connp;
5934 major_t maj;
5935 zoneid_t zoneid;
5936 netstack_t *ns;
5937 ip_stack_t *ipst;
5938
5939 /* Allow reopen. */
5940 if (q->q_ptr != NULL)
5941 return (0);
5942
5943 if (sflag & MODOPEN) {
5944 /* This is a module open */
5945 return (ip_modopen(q, devp, flag, sflag, credp));
5946 }
5947
5948 if ((flag & ~(FKLYR)) == IP_HELPER_STR) {
5949 /*
5950 * Non streams based socket looking for a stream
5951 * to access IP
5952 */
5953 return (ip_helper_stream_setup(q, devp, flag, sflag,
5954 credp, isv6));
5955 }
5956
5957 ns = netstack_find_by_cred(credp);
5958 ASSERT(ns != NULL);
5959 ipst = ns->netstack_ip;
5960 ASSERT(ipst != NULL);
5961
5962 /*
5963 * For exclusive stacks we set the zoneid to zero
5964 * to make IP operate as if in the global zone.
5965 */
5966 if (ipst->ips_netstack->netstack_stackid != GLOBAL_NETSTACKID)
5967 zoneid = GLOBAL_ZONEID;
5968 else
5969 zoneid = crgetzoneid(credp);
5970
5971 /*
5972 * We are opening as a device. This is an IP client stream, and we
5973 * allocate an conn_t as the instance data.
5974 */
5975 connp = ipcl_conn_create(IPCL_IPCCONN, KM_SLEEP, ipst->ips_netstack);
5976
5977 /*
5978 * ipcl_conn_create did a netstack_hold. Undo the hold that was
5979 * done by netstack_find_by_cred()
5980 */
5981 netstack_rele(ipst->ips_netstack);
5982
5983 connp->conn_ixa->ixa_flags |= IXAF_MULTICAST_LOOP | IXAF_SET_ULP_CKSUM;
5984 /* conn_allzones can not be set this early, hence no IPCL_ZONEID */
5985 connp->conn_ixa->ixa_zoneid = zoneid;
5986 connp->conn_zoneid = zoneid;
5987
5988 connp->conn_rq = q;
5989 q->q_ptr = WR(q)->q_ptr = connp;
5990
5991 /* Minor tells us which /dev entry was opened */
5992 if (isv6) {
5993 connp->conn_family = AF_INET6;
5994 connp->conn_ipversion = IPV6_VERSION;
5995 connp->conn_ixa->ixa_flags &= ~IXAF_IS_IPV4;
5996 connp->conn_ixa->ixa_src_preferences = IPV6_PREFER_SRC_DEFAULT;
5997 } else {
5998 connp->conn_family = AF_INET;
5999 connp->conn_ipversion = IPV4_VERSION;
6000 connp->conn_ixa->ixa_flags |= IXAF_IS_IPV4;
6001 }
6002
6003 if ((ip_minor_arena_la != NULL) && (flag & SO_SOCKSTR) &&
6004 ((connp->conn_dev = inet_minor_alloc(ip_minor_arena_la)) != 0)) {
6005 connp->conn_minor_arena = ip_minor_arena_la;
6006 } else {
6007 /*
6008 * Either minor numbers in the large arena were exhausted
6009 * or a non socket application is doing the open.
6010 * Try to allocate from the small arena.
6011 */
6012 if ((connp->conn_dev =
6013 inet_minor_alloc(ip_minor_arena_sa)) == 0) {
6014 /* CONN_DEC_REF takes care of netstack_rele() */
6015 q->q_ptr = WR(q)->q_ptr = NULL;
6016 CONN_DEC_REF(connp);
6017 return (EBUSY);
6018 }
6019 connp->conn_minor_arena = ip_minor_arena_sa;
6020 }
6021
6022 maj = getemajor(*devp);
6023 *devp = makedevice(maj, (minor_t)connp->conn_dev);
6024
6025 /*
6026 * connp->conn_cred is crfree()ed in ipcl_conn_destroy()
6027 */
6028 connp->conn_cred = credp;
6029 connp->conn_cpid = curproc->p_pid;
6030 /* Cache things in ixa without an extra refhold */
6031 ASSERT(!(connp->conn_ixa->ixa_free_flags & IXA_FREE_CRED));
6032 connp->conn_ixa->ixa_cred = connp->conn_cred;
6033 connp->conn_ixa->ixa_cpid = connp->conn_cpid;
6034 if (is_system_labeled())
6035 connp->conn_ixa->ixa_tsl = crgetlabel(connp->conn_cred);
6036
6037 /*
6038 * Handle IP_IOC_RTS_REQUEST and other ioctls which use conn_recv
6039 */
6040 connp->conn_recv = ip_conn_input;
6041 connp->conn_recvicmp = ip_conn_input_icmp;
6042
6043 crhold(connp->conn_cred);
6044
6045 /*
6046 * If the caller has the process-wide flag set, then default to MAC
6047 * exempt mode. This allows read-down to unlabeled hosts.
6048 */
6049 if (getpflags(NET_MAC_AWARE, credp) != 0)
6050 connp->conn_mac_mode = CONN_MAC_AWARE;
6051
6052 connp->conn_zone_is_global = (crgetzoneid(credp) == GLOBAL_ZONEID);
6053
6054 connp->conn_rq = q;
6055 connp->conn_wq = WR(q);
6056
6057 /* Non-zero default values */
6058 connp->conn_ixa->ixa_flags |= IXAF_MULTICAST_LOOP;
6059
6060 /*
6061 * Make the conn globally visible to walkers
6062 */
6063 ASSERT(connp->conn_ref == 1);
6064 mutex_enter(&connp->conn_lock);
6065 connp->conn_state_flags &= ~CONN_INCIPIENT;
6066 mutex_exit(&connp->conn_lock);
6067
6068 qprocson(q);
6069
6070 return (0);
6071 }
6072
6073 /*
6074 * Set IPsec policy from an ipsec_req_t. If the req is not "zero" and valid,
6075 * all of them are copied to the conn_t. If the req is "zero", the policy is
6076 * zeroed out. A "zero" policy has zero ipsr_{ah,req,self_encap}_req
6077 * fields.
6078 * We keep only the latest setting of the policy and thus policy setting
6079 * is not incremental/cumulative.
6080 *
6081 * Requests to set policies with multiple alternative actions will
6082 * go through a different API.
6083 */
6084 int
6085 ipsec_set_req(cred_t *cr, conn_t *connp, ipsec_req_t *req)
6086 {
6087 uint_t ah_req = 0;
6088 uint_t esp_req = 0;
6089 uint_t se_req = 0;
6090 ipsec_act_t *actp = NULL;
6091 uint_t nact;
6092 ipsec_policy_head_t *ph;
6093 boolean_t is_pol_reset, is_pol_inserted = B_FALSE;
6094 int error = 0;
6095 netstack_t *ns = connp->conn_netstack;
6096 ip_stack_t *ipst = ns->netstack_ip;
6097 ipsec_stack_t *ipss = ns->netstack_ipsec;
6098
6099 #define REQ_MASK (IPSEC_PREF_REQUIRED|IPSEC_PREF_NEVER)
6100
6101 /*
6102 * The IP_SEC_OPT option does not allow variable length parameters,
6103 * hence a request cannot be NULL.
6104 */
6105 if (req == NULL)
6106 return (EINVAL);
6107
6108 ah_req = req->ipsr_ah_req;
6109 esp_req = req->ipsr_esp_req;
6110 se_req = req->ipsr_self_encap_req;
6111
6112 /* Don't allow setting self-encap without one or more of AH/ESP. */
6113 if (se_req != 0 && esp_req == 0 && ah_req == 0)
6114 return (EINVAL);
6115
6116 /*
6117 * Are we dealing with a request to reset the policy (i.e.
6118 * zero requests).
6119 */
6120 is_pol_reset = ((ah_req & REQ_MASK) == 0 &&
6121 (esp_req & REQ_MASK) == 0 &&
6122 (se_req & REQ_MASK) == 0);
6123
6124 if (!is_pol_reset) {
6125 /*
6126 * If we couldn't load IPsec, fail with "protocol
6127 * not supported".
6128 * IPsec may not have been loaded for a request with zero
6129 * policies, so we don't fail in this case.
6130 */
6131 mutex_enter(&ipss->ipsec_loader_lock);
6132 if (ipss->ipsec_loader_state != IPSEC_LOADER_SUCCEEDED) {
6133 mutex_exit(&ipss->ipsec_loader_lock);
6134 return (EPROTONOSUPPORT);
6135 }
6136 mutex_exit(&ipss->ipsec_loader_lock);
6137
6138 /*
6139 * Test for valid requests. Invalid algorithms
6140 * need to be tested by IPsec code because new
6141 * algorithms can be added dynamically.
6142 */
6143 if ((ah_req & ~(REQ_MASK|IPSEC_PREF_UNIQUE)) != 0 ||
6144 (esp_req & ~(REQ_MASK|IPSEC_PREF_UNIQUE)) != 0 ||
6145 (se_req & ~(REQ_MASK|IPSEC_PREF_UNIQUE)) != 0) {
6146 return (EINVAL);
6147 }
6148
6149 /*
6150 * Only privileged users can issue these
6151 * requests.
6152 */
6153 if (((ah_req & IPSEC_PREF_NEVER) ||
6154 (esp_req & IPSEC_PREF_NEVER) ||
6155 (se_req & IPSEC_PREF_NEVER)) &&
6156 secpolicy_ip_config(cr, B_FALSE) != 0) {
6157 return (EPERM);
6158 }
6159
6160 /*
6161 * The IPSEC_PREF_REQUIRED and IPSEC_PREF_NEVER
6162 * are mutually exclusive.
6163 */
6164 if (((ah_req & REQ_MASK) == REQ_MASK) ||
6165 ((esp_req & REQ_MASK) == REQ_MASK) ||
6166 ((se_req & REQ_MASK) == REQ_MASK)) {
6167 /* Both of them are set */
6168 return (EINVAL);
6169 }
6170 }
6171
6172 ASSERT(MUTEX_HELD(&connp->conn_lock));
6173
6174 /*
6175 * If we have already cached policies in conn_connect(), don't
6176 * let them change now. We cache policies for connections
6177 * whose src,dst [addr, port] is known.
6178 */
6179 if (connp->conn_policy_cached) {
6180 return (EINVAL);
6181 }
6182
6183 /*
6184 * We have a zero policies, reset the connection policy if already
6185 * set. This will cause the connection to inherit the
6186 * global policy, if any.
6187 */
6188 if (is_pol_reset) {
6189 if (connp->conn_policy != NULL) {
6190 IPPH_REFRELE(connp->conn_policy, ipst->ips_netstack);
6191 connp->conn_policy = NULL;
6192 }
6193 connp->conn_in_enforce_policy = B_FALSE;
6194 connp->conn_out_enforce_policy = B_FALSE;
6195 return (0);
6196 }
6197
6198 ph = connp->conn_policy = ipsec_polhead_split(connp->conn_policy,
6199 ipst->ips_netstack);
6200 if (ph == NULL)
6201 goto enomem;
6202
6203 ipsec_actvec_from_req(req, &actp, &nact, ipst->ips_netstack);
6204 if (actp == NULL)
6205 goto enomem;
6206
6207 /*
6208 * Always insert IPv4 policy entries, since they can also apply to
6209 * ipv6 sockets being used in ipv4-compat mode.
6210 */
6211 if (!ipsec_polhead_insert(ph, actp, nact, IPSEC_AF_V4,
6212 IPSEC_TYPE_INBOUND, ns))
6213 goto enomem;
6214 is_pol_inserted = B_TRUE;
6215 if (!ipsec_polhead_insert(ph, actp, nact, IPSEC_AF_V4,
6216 IPSEC_TYPE_OUTBOUND, ns))
6217 goto enomem;
6218
6219 /*
6220 * We're looking at a v6 socket, also insert the v6-specific
6221 * entries.
6222 */
6223 if (connp->conn_family == AF_INET6) {
6224 if (!ipsec_polhead_insert(ph, actp, nact, IPSEC_AF_V6,
6225 IPSEC_TYPE_INBOUND, ns))
6226 goto enomem;
6227 if (!ipsec_polhead_insert(ph, actp, nact, IPSEC_AF_V6,
6228 IPSEC_TYPE_OUTBOUND, ns))
6229 goto enomem;
6230 }
6231
6232 ipsec_actvec_free(actp, nact);
6233
6234 /*
6235 * If the requests need security, set enforce_policy.
6236 * If the requests are IPSEC_PREF_NEVER, one should
6237 * still set conn_out_enforce_policy so that ip_set_destination
6238 * marks the ip_xmit_attr_t appropriatly. This is needed so that
6239 * for connections that we don't cache policy in at connect time,
6240 * if global policy matches in ip_output_attach_policy, we
6241 * don't wrongly inherit global policy. Similarly, we need
6242 * to set conn_in_enforce_policy also so that we don't verify
6243 * policy wrongly.
6244 */
6245 if ((ah_req & REQ_MASK) != 0 ||
6246 (esp_req & REQ_MASK) != 0 ||
6247 (se_req & REQ_MASK) != 0) {
6248 connp->conn_in_enforce_policy = B_TRUE;
6249 connp->conn_out_enforce_policy = B_TRUE;
6250 }
6251
6252 return (error);
6253 #undef REQ_MASK
6254
6255 /*
6256 * Common memory-allocation-failure exit path.
6257 */
6258 enomem:
6259 if (actp != NULL)
6260 ipsec_actvec_free(actp, nact);
6261 if (is_pol_inserted)
6262 ipsec_polhead_flush(ph, ns);
6263 return (ENOMEM);
6264 }
6265
6266 /*
6267 * Set socket options for joining and leaving multicast groups.
6268 * Common to IPv4 and IPv6; inet6 indicates the type of socket.
6269 * The caller has already check that the option name is consistent with
6270 * the address family of the socket.
6271 */
6272 int
6273 ip_opt_set_multicast_group(conn_t *connp, t_scalar_t name,
6274 uchar_t *invalp, boolean_t inet6, boolean_t checkonly)
6275 {
6276 int *i1 = (int *)invalp;
6277 int error = 0;
6278 ip_stack_t *ipst = connp->conn_netstack->netstack_ip;
6279 struct ip_mreq *v4_mreqp;
6280 struct ipv6_mreq *v6_mreqp;
6281 struct group_req *greqp;
6282 ire_t *ire;
6283 boolean_t done = B_FALSE;
6284 ipaddr_t ifaddr;
6285 in6_addr_t v6group;
6286 uint_t ifindex;
6287 boolean_t mcast_opt = B_TRUE;
6288 mcast_record_t fmode;
6289 int (*optfn)(conn_t *, boolean_t, const in6_addr_t *,
6290 ipaddr_t, uint_t, mcast_record_t, const in6_addr_t *);
6291
6292 switch (name) {
6293 case IP_ADD_MEMBERSHIP:
6294 case IPV6_JOIN_GROUP:
6295 mcast_opt = B_FALSE;
6296 /* FALLTHRU */
6297 case MCAST_JOIN_GROUP:
6298 fmode = MODE_IS_EXCLUDE;
6299 optfn = ip_opt_add_group;
6300 break;
6301
6302 case IP_DROP_MEMBERSHIP:
6303 case IPV6_LEAVE_GROUP:
6304 mcast_opt = B_FALSE;
6305 /* FALLTHRU */
6306 case MCAST_LEAVE_GROUP:
6307 fmode = MODE_IS_INCLUDE;
6308 optfn = ip_opt_delete_group;
6309 break;
6310 default:
6311 ASSERT(0);
6312 }
6313
6314 if (mcast_opt) {
6315 struct sockaddr_in *sin;
6316 struct sockaddr_in6 *sin6;
6317
6318 greqp = (struct group_req *)i1;
6319 if (greqp->gr_group.ss_family == AF_INET) {
6320 sin = (struct sockaddr_in *)&(greqp->gr_group);
6321 IN6_INADDR_TO_V4MAPPED(&sin->sin_addr, &v6group);
6322 } else {
6323 if (!inet6)
6324 return (EINVAL); /* Not on INET socket */
6325
6326 sin6 = (struct sockaddr_in6 *)&(greqp->gr_group);
6327 v6group = sin6->sin6_addr;
6328 }
6329 ifaddr = INADDR_ANY;
6330 ifindex = greqp->gr_interface;
6331 } else if (inet6) {
6332 v6_mreqp = (struct ipv6_mreq *)i1;
6333 v6group = v6_mreqp->ipv6mr_multiaddr;
6334 ifaddr = INADDR_ANY;
6335 ifindex = v6_mreqp->ipv6mr_interface;
6336 } else {
6337 v4_mreqp = (struct ip_mreq *)i1;
6338 IN6_INADDR_TO_V4MAPPED(&v4_mreqp->imr_multiaddr, &v6group);
6339 ifaddr = (ipaddr_t)v4_mreqp->imr_interface.s_addr;
6340 ifindex = 0;
6341 }
6342
6343 /*
6344 * In the multirouting case, we need to replicate
6345 * the request on all interfaces that will take part
6346 * in replication. We do so because multirouting is
6347 * reflective, thus we will probably receive multi-
6348 * casts on those interfaces.
6349 * The ip_multirt_apply_membership() succeeds if
6350 * the operation succeeds on at least one interface.
6351 */
6352 if (IN6_IS_ADDR_V4MAPPED(&v6group)) {
6353 ipaddr_t group;
6354
6355 IN6_V4MAPPED_TO_IPADDR(&v6group, group);
6356
6357 ire = ire_ftable_lookup_v4(group, IP_HOST_MASK, 0,
6358 IRE_HOST | IRE_INTERFACE, NULL, ALL_ZONES, NULL,
6359 MATCH_IRE_MASK | MATCH_IRE_TYPE, 0, ipst, NULL);
6360 } else {
6361 ire = ire_ftable_lookup_v6(&v6group, &ipv6_all_ones, 0,
6362 IRE_HOST | IRE_INTERFACE, NULL, ALL_ZONES, NULL,
6363 MATCH_IRE_MASK | MATCH_IRE_TYPE, 0, ipst, NULL);
6364 }
6365 if (ire != NULL) {
6366 if (ire->ire_flags & RTF_MULTIRT) {
6367 error = ip_multirt_apply_membership(optfn, ire, connp,
6368 checkonly, &v6group, fmode, &ipv6_all_zeros);
6369 done = B_TRUE;
6370 }
6371 ire_refrele(ire);
6372 }
6373
6374 if (!done) {
6375 error = optfn(connp, checkonly, &v6group, ifaddr, ifindex,
6376 fmode, &ipv6_all_zeros);
6377 }
6378 return (error);
6379 }
6380
6381 /*
6382 * Set socket options for joining and leaving multicast groups
6383 * for specific sources.
6384 * Common to IPv4 and IPv6; inet6 indicates the type of socket.
6385 * The caller has already check that the option name is consistent with
6386 * the address family of the socket.
6387 */
6388 int
6389 ip_opt_set_multicast_sources(conn_t *connp, t_scalar_t name,
6390 uchar_t *invalp, boolean_t inet6, boolean_t checkonly)
6391 {
6392 int *i1 = (int *)invalp;
6393 int error = 0;
6394 ip_stack_t *ipst = connp->conn_netstack->netstack_ip;
6395 struct ip_mreq_source *imreqp;
6396 struct group_source_req *gsreqp;
6397 in6_addr_t v6group, v6src;
6398 uint32_t ifindex;
6399 ipaddr_t ifaddr;
6400 boolean_t mcast_opt = B_TRUE;
6401 mcast_record_t fmode;
6402 ire_t *ire;
6403 boolean_t done = B_FALSE;
6404 int (*optfn)(conn_t *, boolean_t, const in6_addr_t *,
6405 ipaddr_t, uint_t, mcast_record_t, const in6_addr_t *);
6406
6407 switch (name) {
6408 case IP_BLOCK_SOURCE:
6409 mcast_opt = B_FALSE;
6410 /* FALLTHRU */
6411 case MCAST_BLOCK_SOURCE:
6412 fmode = MODE_IS_EXCLUDE;
6413 optfn = ip_opt_add_group;
6414 break;
6415
6416 case IP_UNBLOCK_SOURCE:
6417 mcast_opt = B_FALSE;
6418 /* FALLTHRU */
6419 case MCAST_UNBLOCK_SOURCE:
6420 fmode = MODE_IS_EXCLUDE;
6421 optfn = ip_opt_delete_group;
6422 break;
6423
6424 case IP_ADD_SOURCE_MEMBERSHIP:
6425 mcast_opt = B_FALSE;
6426 /* FALLTHRU */
6427 case MCAST_JOIN_SOURCE_GROUP:
6428 fmode = MODE_IS_INCLUDE;
6429 optfn = ip_opt_add_group;
6430 break;
6431
6432 case IP_DROP_SOURCE_MEMBERSHIP:
6433 mcast_opt = B_FALSE;
6434 /* FALLTHRU */
6435 case MCAST_LEAVE_SOURCE_GROUP:
6436 fmode = MODE_IS_INCLUDE;
6437 optfn = ip_opt_delete_group;
6438 break;
6439 default:
6440 ASSERT(0);
6441 }
6442
6443 if (mcast_opt) {
6444 gsreqp = (struct group_source_req *)i1;
6445 ifindex = gsreqp->gsr_interface;
6446 if (gsreqp->gsr_group.ss_family == AF_INET) {
6447 struct sockaddr_in *s;
6448 s = (struct sockaddr_in *)&gsreqp->gsr_group;
6449 IN6_INADDR_TO_V4MAPPED(&s->sin_addr, &v6group);
6450 s = (struct sockaddr_in *)&gsreqp->gsr_source;
6451 IN6_INADDR_TO_V4MAPPED(&s->sin_addr, &v6src);
6452 } else {
6453 struct sockaddr_in6 *s6;
6454
6455 if (!inet6)
6456 return (EINVAL); /* Not on INET socket */
6457
6458 s6 = (struct sockaddr_in6 *)&gsreqp->gsr_group;
6459 v6group = s6->sin6_addr;
6460 s6 = (struct sockaddr_in6 *)&gsreqp->gsr_source;
6461 v6src = s6->sin6_addr;
6462 }
6463 ifaddr = INADDR_ANY;
6464 } else {
6465 imreqp = (struct ip_mreq_source *)i1;
6466 IN6_INADDR_TO_V4MAPPED(&imreqp->imr_multiaddr, &v6group);
6467 IN6_INADDR_TO_V4MAPPED(&imreqp->imr_sourceaddr, &v6src);
6468 ifaddr = (ipaddr_t)imreqp->imr_interface.s_addr;
6469 ifindex = 0;
6470 }
6471
6472 /*
6473 * Handle src being mapped INADDR_ANY by changing it to unspecified.
6474 */
6475 if (IN6_IS_ADDR_V4MAPPED_ANY(&v6src))
6476 v6src = ipv6_all_zeros;
6477
6478 /*
6479 * In the multirouting case, we need to replicate
6480 * the request as noted in the mcast cases above.
6481 */
6482 if (IN6_IS_ADDR_V4MAPPED(&v6group)) {
6483 ipaddr_t group;
6484
6485 IN6_V4MAPPED_TO_IPADDR(&v6group, group);
6486
6487 ire = ire_ftable_lookup_v4(group, IP_HOST_MASK, 0,
6488 IRE_HOST | IRE_INTERFACE, NULL, ALL_ZONES, NULL,
6489 MATCH_IRE_MASK | MATCH_IRE_TYPE, 0, ipst, NULL);
6490 } else {
6491 ire = ire_ftable_lookup_v6(&v6group, &ipv6_all_ones, 0,
6492 IRE_HOST | IRE_INTERFACE, NULL, ALL_ZONES, NULL,
6493 MATCH_IRE_MASK | MATCH_IRE_TYPE, 0, ipst, NULL);
6494 }
6495 if (ire != NULL) {
6496 if (ire->ire_flags & RTF_MULTIRT) {
6497 error = ip_multirt_apply_membership(optfn, ire, connp,
6498 checkonly, &v6group, fmode, &v6src);
6499 done = B_TRUE;
6500 }
6501 ire_refrele(ire);
6502 }
6503 if (!done) {
6504 error = optfn(connp, checkonly, &v6group, ifaddr, ifindex,
6505 fmode, &v6src);
6506 }
6507 return (error);
6508 }
6509
6510 /*
6511 * Given a destination address and a pointer to where to put the information
6512 * this routine fills in the mtuinfo.
6513 * The socket must be connected.
6514 * For sctp conn_faddr is the primary address.
6515 */
6516 int
6517 ip_fill_mtuinfo(conn_t *connp, ip_xmit_attr_t *ixa, struct ip6_mtuinfo *mtuinfo)
6518 {
6519 uint32_t pmtu = IP_MAXPACKET;
6520 uint_t scopeid;
6521
6522 if (IN6_IS_ADDR_UNSPECIFIED(&connp->conn_faddr_v6))
6523 return (-1);
6524
6525 /* In case we never sent or called ip_set_destination_v4/v6 */
6526 if (ixa->ixa_ire != NULL)
6527 pmtu = ip_get_pmtu(ixa);
6528
6529 if (ixa->ixa_flags & IXAF_SCOPEID_SET)
6530 scopeid = ixa->ixa_scopeid;
6531 else
6532 scopeid = 0;
6533
6534 bzero(mtuinfo, sizeof (*mtuinfo));
6535 mtuinfo->ip6m_addr.sin6_family = AF_INET6;
6536 mtuinfo->ip6m_addr.sin6_port = connp->conn_fport;
6537 mtuinfo->ip6m_addr.sin6_addr = connp->conn_faddr_v6;
6538 mtuinfo->ip6m_addr.sin6_scope_id = scopeid;
6539 mtuinfo->ip6m_mtu = pmtu;
6540
6541 return (sizeof (struct ip6_mtuinfo));
6542 }
6543
6544 /*
6545 * When the src multihoming is changed from weak to [strong, preferred]
6546 * ip_ire_rebind_walker is called to walk the list of all ire_t entries
6547 * and identify routes that were created by user-applications in the
6548 * unbound state (i.e., without RTA_IFP), and for which an ire_ill is not
6549 * currently defined. These routes are then 'rebound', i.e., their ire_ill
6550 * is selected by finding an interface route for the gateway.
6551 */
6552 /* ARGSUSED */
6553 void
6554 ip_ire_rebind_walker(ire_t *ire, void *notused)
6555 {
6556 if (!ire->ire_unbound || ire->ire_ill != NULL)
6557 return;
6558 ire_rebind(ire);
6559 ire_delete(ire);
6560 }
6561
6562 /*
6563 * When the src multihoming is changed from [strong, preferred] to weak,
6564 * ip_ire_unbind_walker is called to walk the list of all ire_t entries, and
6565 * set any entries that were created by user-applications in the unbound state
6566 * (i.e., without RTA_IFP) back to having a NULL ire_ill.
6567 */
6568 /* ARGSUSED */
6569 void
6570 ip_ire_unbind_walker(ire_t *ire, void *notused)
6571 {
6572 ire_t *new_ire;
6573
6574 if (!ire->ire_unbound || ire->ire_ill == NULL)
6575 return;
6576 if (ire->ire_ipversion == IPV6_VERSION) {
6577 new_ire = ire_create_v6(&ire->ire_addr_v6, &ire->ire_mask_v6,
6578 &ire->ire_gateway_addr_v6, ire->ire_type, NULL,
6579 ire->ire_zoneid, ire->ire_flags, NULL, ire->ire_ipst);
6580 } else {
6581 new_ire = ire_create((uchar_t *)&ire->ire_addr,
6582 (uchar_t *)&ire->ire_mask,
6583 (uchar_t *)&ire->ire_gateway_addr, ire->ire_type, NULL,
6584 ire->ire_zoneid, ire->ire_flags, NULL, ire->ire_ipst);
6585 }
6586 if (new_ire == NULL)
6587 return;
6588 new_ire->ire_unbound = B_TRUE;
6589 /*
6590 * The bound ire must first be deleted so that we don't return
6591 * the existing one on the attempt to add the unbound new_ire.
6592 */
6593 ire_delete(ire);
6594 new_ire = ire_add(new_ire);
6595 if (new_ire != NULL)
6596 ire_refrele(new_ire);
6597 }
6598
6599 /*
6600 * When the settings of ip*_strict_src_multihoming tunables are changed,
6601 * all cached routes need to be recomputed. This recomputation needs to be
6602 * done when going from weaker to stronger modes so that the cached ire
6603 * for the connection does not violate the current ip*_strict_src_multihoming
6604 * setting. It also needs to be done when going from stronger to weaker modes,
6605 * so that we fall back to matching on the longest-matching-route (as opposed
6606 * to a shorter match that may have been selected in the strong mode
6607 * to satisfy src_multihoming settings).
6608 *
6609 * The cached ixa_ire entires for all conn_t entries are marked as
6610 * "verify" so that they will be recomputed for the next packet.
6611 */
6612 void
6613 conn_ire_revalidate(conn_t *connp, void *arg)
6614 {
6615 boolean_t isv6 = (boolean_t)arg;
6616
6617 if ((isv6 && connp->conn_ipversion != IPV6_VERSION) ||
6618 (!isv6 && connp->conn_ipversion != IPV4_VERSION))
6619 return;
6620 connp->conn_ixa->ixa_ire_generation = IRE_GENERATION_VERIFY;
6621 }
6622
6623 /*
6624 * Handles both IPv4 and IPv6 reassembly - doing the out-of-order cases,
6625 * When an ipf is passed here for the first time, if
6626 * we already have in-order fragments on the queue, we convert from the fast-
6627 * path reassembly scheme to the hard-case scheme. From then on, additional
6628 * fragments are reassembled here. We keep track of the start and end offsets
6629 * of each piece, and the number of holes in the chain. When the hole count
6630 * goes to zero, we are done!
6631 *
6632 * The ipf_count will be updated to account for any mblk(s) added (pointed to
6633 * by mp) or subtracted (freeb()ed dups), upon return the caller must update
6634 * ipfb_count and ill_frag_count by the difference of ipf_count before and
6635 * after the call to ip_reassemble().
6636 */
6637 int
6638 ip_reassemble(mblk_t *mp, ipf_t *ipf, uint_t start, boolean_t more, ill_t *ill,
6639 size_t msg_len)
6640 {
6641 uint_t end;
6642 mblk_t *next_mp;
6643 mblk_t *mp1;
6644 uint_t offset;
6645 boolean_t incr_dups = B_TRUE;
6646 boolean_t offset_zero_seen = B_FALSE;
6647 boolean_t pkt_boundary_checked = B_FALSE;
6648
6649 /* If start == 0 then ipf_nf_hdr_len has to be set. */
6650 ASSERT(start != 0 || ipf->ipf_nf_hdr_len != 0);
6651
6652 /* Add in byte count */
6653 ipf->ipf_count += msg_len;
6654 if (ipf->ipf_end) {
6655 /*
6656 * We were part way through in-order reassembly, but now there
6657 * is a hole. We walk through messages already queued, and
6658 * mark them for hard case reassembly. We know that up till
6659 * now they were in order starting from offset zero.
6660 */
6661 offset = 0;
6662 for (mp1 = ipf->ipf_mp->b_cont; mp1; mp1 = mp1->b_cont) {
6663 IP_REASS_SET_START(mp1, offset);
6664 if (offset == 0) {
6665 ASSERT(ipf->ipf_nf_hdr_len != 0);
6666 offset = -ipf->ipf_nf_hdr_len;
6667 }
6668 offset += mp1->b_wptr - mp1->b_rptr;
6669 IP_REASS_SET_END(mp1, offset);
6670 }
6671 /* One hole at the end. */
6672 ipf->ipf_hole_cnt = 1;
6673 /* Brand it as a hard case, forever. */
6674 ipf->ipf_end = 0;
6675 }
6676 /* Walk through all the new pieces. */
6677 do {
6678 end = start + (mp->b_wptr - mp->b_rptr);
6679 /*
6680 * If start is 0, decrease 'end' only for the first mblk of
6681 * the fragment. Otherwise 'end' can get wrong value in the
6682 * second pass of the loop if first mblk is exactly the
6683 * size of ipf_nf_hdr_len.
6684 */
6685 if (start == 0 && !offset_zero_seen) {
6686 /* First segment */
6687 ASSERT(ipf->ipf_nf_hdr_len != 0);
6688 end -= ipf->ipf_nf_hdr_len;
6689 offset_zero_seen = B_TRUE;
6690 }
6691 next_mp = mp->b_cont;
6692 /*
6693 * We are checking to see if there is any interesing data
6694 * to process. If there isn't and the mblk isn't the
6695 * one which carries the unfragmentable header then we
6696 * drop it. It's possible to have just the unfragmentable
6697 * header come through without any data. That needs to be
6698 * saved.
6699 *
6700 * If the assert at the top of this function holds then the
6701 * term "ipf->ipf_nf_hdr_len != 0" isn't needed. This code
6702 * is infrequently traveled enough that the test is left in
6703 * to protect against future code changes which break that
6704 * invariant.
6705 */
6706 if (start == end && start != 0 && ipf->ipf_nf_hdr_len != 0) {
6707 /* Empty. Blast it. */
6708 IP_REASS_SET_START(mp, 0);
6709 IP_REASS_SET_END(mp, 0);
6710 /*
6711 * If the ipf points to the mblk we are about to free,
6712 * update ipf to point to the next mblk (or NULL
6713 * if none).
6714 */
6715 if (ipf->ipf_mp->b_cont == mp)
6716 ipf->ipf_mp->b_cont = next_mp;
6717 freeb(mp);
6718 continue;
6719 }
6720 mp->b_cont = NULL;
6721 IP_REASS_SET_START(mp, start);
6722 IP_REASS_SET_END(mp, end);
6723 if (!ipf->ipf_tail_mp) {
6724 ipf->ipf_tail_mp = mp;
6725 ipf->ipf_mp->b_cont = mp;
6726 if (start == 0 || !more) {
6727 ipf->ipf_hole_cnt = 1;
6728 /*
6729 * if the first fragment comes in more than one
6730 * mblk, this loop will be executed for each
6731 * mblk. Need to adjust hole count so exiting
6732 * this routine will leave hole count at 1.
6733 */
6734 if (next_mp)
6735 ipf->ipf_hole_cnt++;
6736 } else
6737 ipf->ipf_hole_cnt = 2;
6738 continue;
6739 } else if (ipf->ipf_last_frag_seen && !more &&
6740 !pkt_boundary_checked) {
6741 /*
6742 * We check datagram boundary only if this fragment
6743 * claims to be the last fragment and we have seen a
6744 * last fragment in the past too. We do this only
6745 * once for a given fragment.
6746 *
6747 * start cannot be 0 here as fragments with start=0
6748 * and MF=0 gets handled as a complete packet. These
6749 * fragments should not reach here.
6750 */
6751
6752 if (start + msgdsize(mp) !=
6753 IP_REASS_END(ipf->ipf_tail_mp)) {
6754 /*
6755 * We have two fragments both of which claim
6756 * to be the last fragment but gives conflicting
6757 * information about the whole datagram size.
6758 * Something fishy is going on. Drop the
6759 * fragment and free up the reassembly list.
6760 */
6761 return (IP_REASS_FAILED);
6762 }
6763
6764 /*
6765 * We shouldn't come to this code block again for this
6766 * particular fragment.
6767 */
6768 pkt_boundary_checked = B_TRUE;
6769 }
6770
6771 /* New stuff at or beyond tail? */
6772 offset = IP_REASS_END(ipf->ipf_tail_mp);
6773 if (start >= offset) {
6774 if (ipf->ipf_last_frag_seen) {
6775 /* current fragment is beyond last fragment */
6776 return (IP_REASS_FAILED);
6777 }
6778 /* Link it on end. */
6779 ipf->ipf_tail_mp->b_cont = mp;
6780 ipf->ipf_tail_mp = mp;
6781 if (more) {
6782 if (start != offset)
6783 ipf->ipf_hole_cnt++;
6784 } else if (start == offset && next_mp == NULL)
6785 ipf->ipf_hole_cnt--;
6786 continue;
6787 }
6788 mp1 = ipf->ipf_mp->b_cont;
6789 offset = IP_REASS_START(mp1);
6790 /* New stuff at the front? */
6791 if (start < offset) {
6792 if (start == 0) {
6793 if (end >= offset) {
6794 /* Nailed the hole at the begining. */
6795 ipf->ipf_hole_cnt--;
6796 }
6797 } else if (end < offset) {
6798 /*
6799 * A hole, stuff, and a hole where there used
6800 * to be just a hole.
6801 */
6802 ipf->ipf_hole_cnt++;
6803 }
6804 mp->b_cont = mp1;
6805 /* Check for overlap. */
6806 while (end > offset) {
6807 if (end < IP_REASS_END(mp1)) {
6808 mp->b_wptr -= end - offset;
6809 IP_REASS_SET_END(mp, offset);
6810 BUMP_MIB(ill->ill_ip_mib,
6811 ipIfStatsReasmPartDups);
6812 break;
6813 }
6814 /* Did we cover another hole? */
6815 if ((mp1->b_cont &&
6816 IP_REASS_END(mp1) !=
6817 IP_REASS_START(mp1->b_cont) &&
6818 end >= IP_REASS_START(mp1->b_cont)) ||
6819 (!ipf->ipf_last_frag_seen && !more)) {
6820 ipf->ipf_hole_cnt--;
6821 }
6822 /* Clip out mp1. */
6823 if ((mp->b_cont = mp1->b_cont) == NULL) {
6824 /*
6825 * After clipping out mp1, this guy
6826 * is now hanging off the end.
6827 */
6828 ipf->ipf_tail_mp = mp;
6829 }
6830 IP_REASS_SET_START(mp1, 0);
6831 IP_REASS_SET_END(mp1, 0);
6832 /* Subtract byte count */
6833 ipf->ipf_count -= mp1->b_datap->db_lim -
6834 mp1->b_datap->db_base;
6835 freeb(mp1);
6836 BUMP_MIB(ill->ill_ip_mib,
6837 ipIfStatsReasmPartDups);
6838 mp1 = mp->b_cont;
6839 if (!mp1)
6840 break;
6841 offset = IP_REASS_START(mp1);
6842 }
6843 ipf->ipf_mp->b_cont = mp;
6844 continue;
6845 }
6846 /*
6847 * The new piece starts somewhere between the start of the head
6848 * and before the end of the tail.
6849 */
6850 for (; mp1; mp1 = mp1->b_cont) {
6851 offset = IP_REASS_END(mp1);
6852 if (start < offset) {
6853 if (end <= offset) {
6854 /* Nothing new. */
6855 IP_REASS_SET_START(mp, 0);
6856 IP_REASS_SET_END(mp, 0);
6857 /* Subtract byte count */
6858 ipf->ipf_count -= mp->b_datap->db_lim -
6859 mp->b_datap->db_base;
6860 if (incr_dups) {
6861 ipf->ipf_num_dups++;
6862 incr_dups = B_FALSE;
6863 }
6864 freeb(mp);
6865 BUMP_MIB(ill->ill_ip_mib,
6866 ipIfStatsReasmDuplicates);
6867 break;
6868 }
6869 /*
6870 * Trim redundant stuff off beginning of new
6871 * piece.
6872 */
6873 IP_REASS_SET_START(mp, offset);
6874 mp->b_rptr += offset - start;
6875 BUMP_MIB(ill->ill_ip_mib,
6876 ipIfStatsReasmPartDups);
6877 start = offset;
6878 if (!mp1->b_cont) {
6879 /*
6880 * After trimming, this guy is now
6881 * hanging off the end.
6882 */
6883 mp1->b_cont = mp;
6884 ipf->ipf_tail_mp = mp;
6885 if (!more) {
6886 ipf->ipf_hole_cnt--;
6887 }
6888 break;
6889 }
6890 }
6891 if (start >= IP_REASS_START(mp1->b_cont))
6892 continue;
6893 /* Fill a hole */
6894 if (start > offset)
6895 ipf->ipf_hole_cnt++;
6896 mp->b_cont = mp1->b_cont;
6897 mp1->b_cont = mp;
6898 mp1 = mp->b_cont;
6899 offset = IP_REASS_START(mp1);
6900 if (end >= offset) {
6901 ipf->ipf_hole_cnt--;
6902 /* Check for overlap. */
6903 while (end > offset) {
6904 if (end < IP_REASS_END(mp1)) {
6905 mp->b_wptr -= end - offset;
6906 IP_REASS_SET_END(mp, offset);
6907 /*
6908 * TODO we might bump
6909 * this up twice if there is
6910 * overlap at both ends.
6911 */
6912 BUMP_MIB(ill->ill_ip_mib,
6913 ipIfStatsReasmPartDups);
6914 break;
6915 }
6916 /* Did we cover another hole? */
6917 if ((mp1->b_cont &&
6918 IP_REASS_END(mp1)
6919 != IP_REASS_START(mp1->b_cont) &&
6920 end >=
6921 IP_REASS_START(mp1->b_cont)) ||
6922 (!ipf->ipf_last_frag_seen &&
6923 !more)) {
6924 ipf->ipf_hole_cnt--;
6925 }
6926 /* Clip out mp1. */
6927 if ((mp->b_cont = mp1->b_cont) ==
6928 NULL) {
6929 /*
6930 * After clipping out mp1,
6931 * this guy is now hanging
6932 * off the end.
6933 */
6934 ipf->ipf_tail_mp = mp;
6935 }
6936 IP_REASS_SET_START(mp1, 0);
6937 IP_REASS_SET_END(mp1, 0);
6938 /* Subtract byte count */
6939 ipf->ipf_count -=
6940 mp1->b_datap->db_lim -
6941 mp1->b_datap->db_base;
6942 freeb(mp1);
6943 BUMP_MIB(ill->ill_ip_mib,
6944 ipIfStatsReasmPartDups);
6945 mp1 = mp->b_cont;
6946 if (!mp1)
6947 break;
6948 offset = IP_REASS_START(mp1);
6949 }
6950 }
6951 break;
6952 }
6953 } while (start = end, mp = next_mp);
6954
6955 /* Fragment just processed could be the last one. Remember this fact */
6956 if (!more)
6957 ipf->ipf_last_frag_seen = B_TRUE;
6958
6959 /* Still got holes? */
6960 if (ipf->ipf_hole_cnt)
6961 return (IP_REASS_PARTIAL);
6962 /* Clean up overloaded fields to avoid upstream disasters. */
6963 for (mp1 = ipf->ipf_mp->b_cont; mp1; mp1 = mp1->b_cont) {
6964 IP_REASS_SET_START(mp1, 0);
6965 IP_REASS_SET_END(mp1, 0);
6966 }
6967 return (IP_REASS_COMPLETE);
6968 }
6969
6970 /*
6971 * Fragmentation reassembly. Each ILL has a hash table for
6972 * queuing packets undergoing reassembly for all IPIFs
6973 * associated with the ILL. The hash is based on the packet
6974 * IP ident field. The ILL frag hash table was allocated
6975 * as a timer block at the time the ILL was created. Whenever
6976 * there is anything on the reassembly queue, the timer will
6977 * be running. Returns the reassembled packet if reassembly completes.
6978 */
6979 mblk_t *
6980 ip_input_fragment(mblk_t *mp, ipha_t *ipha, ip_recv_attr_t *ira)
6981 {
6982 uint32_t frag_offset_flags;
6983 mblk_t *t_mp;
6984 ipaddr_t dst;
6985 uint8_t proto = ipha->ipha_protocol;
6986 uint32_t sum_val;
6987 uint16_t sum_flags;
6988 ipf_t *ipf;
6989 ipf_t **ipfp;
6990 ipfb_t *ipfb;
6991 uint16_t ident;
6992 uint32_t offset;
6993 ipaddr_t src;
6994 uint_t hdr_length;
6995 uint32_t end;
6996 mblk_t *mp1;
6997 mblk_t *tail_mp;
6998 size_t count;
6999 size_t msg_len;
7000 uint8_t ecn_info = 0;
7001 uint32_t packet_size;
7002 boolean_t pruned = B_FALSE;
7003 ill_t *ill = ira->ira_ill;
7004 ip_stack_t *ipst = ill->ill_ipst;
7005
7006 /*
7007 * Drop the fragmented as early as possible, if
7008 * we don't have resource(s) to re-assemble.
7009 */
7010 if (ipst->ips_ip_reass_queue_bytes == 0) {
7011 freemsg(mp);
7012 return (NULL);
7013 }
7014
7015 /* Check for fragmentation offset; return if there's none */
7016 if ((frag_offset_flags = ntohs(ipha->ipha_fragment_offset_and_flags) &
7017 (IPH_MF | IPH_OFFSET)) == 0)
7018 return (mp);
7019
7020 /*
7021 * We utilize hardware computed checksum info only for UDP since
7022 * IP fragmentation is a normal occurrence for the protocol. In
7023 * addition, checksum offload support for IP fragments carrying
7024 * UDP payload is commonly implemented across network adapters.
7025 */
7026 ASSERT(ira->ira_rill != NULL);
7027 if (proto == IPPROTO_UDP && dohwcksum &&
7028 ILL_HCKSUM_CAPABLE(ira->ira_rill) &&
7029 (DB_CKSUMFLAGS(mp) & (HCK_FULLCKSUM | HCK_PARTIALCKSUM))) {
7030 mblk_t *mp1 = mp->b_cont;
7031 int32_t len;
7032
7033 /* Record checksum information from the packet */
7034 sum_val = (uint32_t)DB_CKSUM16(mp);
7035 sum_flags = DB_CKSUMFLAGS(mp);
7036
7037 /* IP payload offset from beginning of mblk */
7038 offset = ((uchar_t *)ipha + IPH_HDR_LENGTH(ipha)) - mp->b_rptr;
7039
7040 if ((sum_flags & HCK_PARTIALCKSUM) &&
7041 (mp1 == NULL || mp1->b_cont == NULL) &&
7042 offset >= DB_CKSUMSTART(mp) &&
7043 ((len = offset - DB_CKSUMSTART(mp)) & 1) == 0) {
7044 uint32_t adj;
7045 /*
7046 * Partial checksum has been calculated by hardware
7047 * and attached to the packet; in addition, any
7048 * prepended extraneous data is even byte aligned.
7049 * If any such data exists, we adjust the checksum;
7050 * this would also handle any postpended data.
7051 */
7052 IP_ADJCKSUM_PARTIAL(mp->b_rptr + DB_CKSUMSTART(mp),
7053 mp, mp1, len, adj);
7054
7055 /* One's complement subtract extraneous checksum */
7056 if (adj >= sum_val)
7057 sum_val = ~(adj - sum_val) & 0xFFFF;
7058 else
7059 sum_val -= adj;
7060 }
7061 } else {
7062 sum_val = 0;
7063 sum_flags = 0;
7064 }
7065
7066 /* Clear hardware checksumming flag */
7067 DB_CKSUMFLAGS(mp) = 0;
7068
7069 ident = ipha->ipha_ident;
7070 offset = (frag_offset_flags << 3) & 0xFFFF;
7071 src = ipha->ipha_src;
7072 dst = ipha->ipha_dst;
7073 hdr_length = IPH_HDR_LENGTH(ipha);
7074 end = ntohs(ipha->ipha_length) - hdr_length;
7075
7076 /* If end == 0 then we have a packet with no data, so just free it */
7077 if (end == 0) {
7078 freemsg(mp);
7079 return (NULL);
7080 }
7081
7082 /* Record the ECN field info. */
7083 ecn_info = (ipha->ipha_type_of_service & 0x3);
7084 if (offset != 0) {
7085 /*
7086 * If this isn't the first piece, strip the header, and
7087 * add the offset to the end value.
7088 */
7089 mp->b_rptr += hdr_length;
7090 end += offset;
7091 }
7092
7093 /* Handle vnic loopback of fragments */
7094 if (mp->b_datap->db_ref > 2)
7095 msg_len = 0;
7096 else
7097 msg_len = MBLKSIZE(mp);
7098
7099 tail_mp = mp;
7100 while (tail_mp->b_cont != NULL) {
7101 tail_mp = tail_mp->b_cont;
7102 if (tail_mp->b_datap->db_ref <= 2)
7103 msg_len += MBLKSIZE(tail_mp);
7104 }
7105
7106 /* If the reassembly list for this ILL will get too big, prune it */
7107 if ((msg_len + sizeof (*ipf) + ill->ill_frag_count) >=
7108 ipst->ips_ip_reass_queue_bytes) {
7109 DTRACE_PROBE3(ip_reass_queue_bytes, uint_t, msg_len,
7110 uint_t, ill->ill_frag_count,
7111 uint_t, ipst->ips_ip_reass_queue_bytes);
7112 ill_frag_prune(ill,
7113 (ipst->ips_ip_reass_queue_bytes < msg_len) ? 0 :
7114 (ipst->ips_ip_reass_queue_bytes - msg_len));
7115 pruned = B_TRUE;
7116 }
7117
7118 ipfb = &ill->ill_frag_hash_tbl[ILL_FRAG_HASH(src, ident)];
7119 mutex_enter(&ipfb->ipfb_lock);
7120
7121 ipfp = &ipfb->ipfb_ipf;
7122 /* Try to find an existing fragment queue for this packet. */
7123 for (;;) {
7124 ipf = ipfp[0];
7125 if (ipf != NULL) {
7126 /*
7127 * It has to match on ident and src/dst address.
7128 */
7129 if (ipf->ipf_ident == ident &&
7130 ipf->ipf_src == src &&
7131 ipf->ipf_dst == dst &&
7132 ipf->ipf_protocol == proto) {
7133 /*
7134 * If we have received too many
7135 * duplicate fragments for this packet
7136 * free it.
7137 */
7138 if (ipf->ipf_num_dups > ip_max_frag_dups) {
7139 ill_frag_free_pkts(ill, ipfb, ipf, 1);
7140 freemsg(mp);
7141 mutex_exit(&ipfb->ipfb_lock);
7142 return (NULL);
7143 }
7144 /* Found it. */
7145 break;
7146 }
7147 ipfp = &ipf->ipf_hash_next;
7148 continue;
7149 }
7150
7151 /*
7152 * If we pruned the list, do we want to store this new
7153 * fragment?. We apply an optimization here based on the
7154 * fact that most fragments will be received in order.
7155 * So if the offset of this incoming fragment is zero,
7156 * it is the first fragment of a new packet. We will
7157 * keep it. Otherwise drop the fragment, as we have
7158 * probably pruned the packet already (since the
7159 * packet cannot be found).
7160 */
7161 if (pruned && offset != 0) {
7162 mutex_exit(&ipfb->ipfb_lock);
7163 freemsg(mp);
7164 return (NULL);
7165 }
7166
7167 if (ipfb->ipfb_frag_pkts >= MAX_FRAG_PKTS(ipst)) {
7168 /*
7169 * Too many fragmented packets in this hash
7170 * bucket. Free the oldest.
7171 */
7172 ill_frag_free_pkts(ill, ipfb, ipfb->ipfb_ipf, 1);
7173 }
7174
7175 /* New guy. Allocate a frag message. */
7176 mp1 = allocb(sizeof (*ipf), BPRI_MED);
7177 if (mp1 == NULL) {
7178 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
7179 ip_drop_input("ipIfStatsInDiscards", mp, ill);
7180 freemsg(mp);
7181 reass_done:
7182 mutex_exit(&ipfb->ipfb_lock);
7183 return (NULL);
7184 }
7185
7186 BUMP_MIB(ill->ill_ip_mib, ipIfStatsReasmReqds);
7187 mp1->b_cont = mp;
7188
7189 /* Initialize the fragment header. */
7190 ipf = (ipf_t *)mp1->b_rptr;
7191 ipf->ipf_mp = mp1;
7192 ipf->ipf_ptphn = ipfp;
7193 ipfp[0] = ipf;
7194 ipf->ipf_hash_next = NULL;
7195 ipf->ipf_ident = ident;
7196 ipf->ipf_protocol = proto;
7197 ipf->ipf_src = src;
7198 ipf->ipf_dst = dst;
7199 ipf->ipf_nf_hdr_len = 0;
7200 /* Record reassembly start time. */
7201 ipf->ipf_timestamp = gethrestime_sec();
7202 /* Record ipf generation and account for frag header */
7203 ipf->ipf_gen = ill->ill_ipf_gen++;
7204 ipf->ipf_count = MBLKSIZE(mp1);
7205 ipf->ipf_last_frag_seen = B_FALSE;
7206 ipf->ipf_ecn = ecn_info;
7207 ipf->ipf_num_dups = 0;
7208 ipfb->ipfb_frag_pkts++;
7209 ipf->ipf_checksum = 0;
7210 ipf->ipf_checksum_flags = 0;
7211
7212 /* Store checksum value in fragment header */
7213 if (sum_flags != 0) {
7214 sum_val = (sum_val & 0xFFFF) + (sum_val >> 16);
7215 sum_val = (sum_val & 0xFFFF) + (sum_val >> 16);
7216 ipf->ipf_checksum = sum_val;
7217 ipf->ipf_checksum_flags = sum_flags;
7218 }
7219
7220 /*
7221 * We handle reassembly two ways. In the easy case,
7222 * where all the fragments show up in order, we do
7223 * minimal bookkeeping, and just clip new pieces on
7224 * the end. If we ever see a hole, then we go off
7225 * to ip_reassemble which has to mark the pieces and
7226 * keep track of the number of holes, etc. Obviously,
7227 * the point of having both mechanisms is so we can
7228 * handle the easy case as efficiently as possible.
7229 */
7230 if (offset == 0) {
7231 /* Easy case, in-order reassembly so far. */
7232 ipf->ipf_count += msg_len;
7233 ipf->ipf_tail_mp = tail_mp;
7234 /*
7235 * Keep track of next expected offset in
7236 * ipf_end.
7237 */
7238 ipf->ipf_end = end;
7239 ipf->ipf_nf_hdr_len = hdr_length;
7240 } else {
7241 /* Hard case, hole at the beginning. */
7242 ipf->ipf_tail_mp = NULL;
7243 /*
7244 * ipf_end == 0 means that we have given up
7245 * on easy reassembly.
7246 */
7247 ipf->ipf_end = 0;
7248
7249 /* Forget checksum offload from now on */
7250 ipf->ipf_checksum_flags = 0;
7251
7252 /*
7253 * ipf_hole_cnt is set by ip_reassemble.
7254 * ipf_count is updated by ip_reassemble.
7255 * No need to check for return value here
7256 * as we don't expect reassembly to complete
7257 * or fail for the first fragment itself.
7258 */
7259 (void) ip_reassemble(mp, ipf,
7260 (frag_offset_flags & IPH_OFFSET) << 3,
7261 (frag_offset_flags & IPH_MF), ill, msg_len);
7262 }
7263 /* Update per ipfb and ill byte counts */
7264 ipfb->ipfb_count += ipf->ipf_count;
7265 ASSERT(ipfb->ipfb_count > 0); /* Wraparound */
7266 atomic_add_32(&ill->ill_frag_count, ipf->ipf_count);
7267 /* If the frag timer wasn't already going, start it. */
7268 mutex_enter(&ill->ill_lock);
7269 ill_frag_timer_start(ill);
7270 mutex_exit(&ill->ill_lock);
7271 goto reass_done;
7272 }
7273
7274 /*
7275 * If the packet's flag has changed (it could be coming up
7276 * from an interface different than the previous, therefore
7277 * possibly different checksum capability), then forget about
7278 * any stored checksum states. Otherwise add the value to
7279 * the existing one stored in the fragment header.
7280 */
7281 if (sum_flags != 0 && sum_flags == ipf->ipf_checksum_flags) {
7282 sum_val += ipf->ipf_checksum;
7283 sum_val = (sum_val & 0xFFFF) + (sum_val >> 16);
7284 sum_val = (sum_val & 0xFFFF) + (sum_val >> 16);
7285 ipf->ipf_checksum = sum_val;
7286 } else if (ipf->ipf_checksum_flags != 0) {
7287 /* Forget checksum offload from now on */
7288 ipf->ipf_checksum_flags = 0;
7289 }
7290
7291 /*
7292 * We have a new piece of a datagram which is already being
7293 * reassembled. Update the ECN info if all IP fragments
7294 * are ECN capable. If there is one which is not, clear
7295 * all the info. If there is at least one which has CE
7296 * code point, IP needs to report that up to transport.
7297 */
7298 if (ecn_info != IPH_ECN_NECT && ipf->ipf_ecn != IPH_ECN_NECT) {
7299 if (ecn_info == IPH_ECN_CE)
7300 ipf->ipf_ecn = IPH_ECN_CE;
7301 } else {
7302 ipf->ipf_ecn = IPH_ECN_NECT;
7303 }
7304 if (offset && ipf->ipf_end == offset) {
7305 /* The new fragment fits at the end */
7306 ipf->ipf_tail_mp->b_cont = mp;
7307 /* Update the byte count */
7308 ipf->ipf_count += msg_len;
7309 /* Update per ipfb and ill byte counts */
7310 ipfb->ipfb_count += msg_len;
7311 ASSERT(ipfb->ipfb_count > 0); /* Wraparound */
7312 atomic_add_32(&ill->ill_frag_count, msg_len);
7313 if (frag_offset_flags & IPH_MF) {
7314 /* More to come. */
7315 ipf->ipf_end = end;
7316 ipf->ipf_tail_mp = tail_mp;
7317 goto reass_done;
7318 }
7319 } else {
7320 /* Go do the hard cases. */
7321 int ret;
7322
7323 if (offset == 0)
7324 ipf->ipf_nf_hdr_len = hdr_length;
7325
7326 /* Save current byte count */
7327 count = ipf->ipf_count;
7328 ret = ip_reassemble(mp, ipf,
7329 (frag_offset_flags & IPH_OFFSET) << 3,
7330 (frag_offset_flags & IPH_MF), ill, msg_len);
7331 /* Count of bytes added and subtracted (freeb()ed) */
7332 count = ipf->ipf_count - count;
7333 if (count) {
7334 /* Update per ipfb and ill byte counts */
7335 ipfb->ipfb_count += count;
7336 ASSERT(ipfb->ipfb_count > 0); /* Wraparound */
7337 atomic_add_32(&ill->ill_frag_count, count);
7338 }
7339 if (ret == IP_REASS_PARTIAL) {
7340 goto reass_done;
7341 } else if (ret == IP_REASS_FAILED) {
7342 /* Reassembly failed. Free up all resources */
7343 ill_frag_free_pkts(ill, ipfb, ipf, 1);
7344 for (t_mp = mp; t_mp != NULL; t_mp = t_mp->b_cont) {
7345 IP_REASS_SET_START(t_mp, 0);
7346 IP_REASS_SET_END(t_mp, 0);
7347 }
7348 freemsg(mp);
7349 goto reass_done;
7350 }
7351 /* We will reach here iff 'ret' is IP_REASS_COMPLETE */
7352 }
7353 /*
7354 * We have completed reassembly. Unhook the frag header from
7355 * the reassembly list.
7356 *
7357 * Before we free the frag header, record the ECN info
7358 * to report back to the transport.
7359 */
7360 ecn_info = ipf->ipf_ecn;
7361 BUMP_MIB(ill->ill_ip_mib, ipIfStatsReasmOKs);
7362 ipfp = ipf->ipf_ptphn;
7363
7364 /* We need to supply these to caller */
7365 if ((sum_flags = ipf->ipf_checksum_flags) != 0)
7366 sum_val = ipf->ipf_checksum;
7367 else
7368 sum_val = 0;
7369
7370 mp1 = ipf->ipf_mp;
7371 count = ipf->ipf_count;
7372 ipf = ipf->ipf_hash_next;
7373 if (ipf != NULL)
7374 ipf->ipf_ptphn = ipfp;
7375 ipfp[0] = ipf;
7376 atomic_add_32(&ill->ill_frag_count, -count);
7377 ASSERT(ipfb->ipfb_count >= count);
7378 ipfb->ipfb_count -= count;
7379 ipfb->ipfb_frag_pkts--;
7380 mutex_exit(&ipfb->ipfb_lock);
7381 /* Ditch the frag header. */
7382 mp = mp1->b_cont;
7383
7384 freeb(mp1);
7385
7386 /* Restore original IP length in header. */
7387 packet_size = (uint32_t)msgdsize(mp);
7388 if (packet_size > IP_MAXPACKET) {
7389 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7390 ip_drop_input("Reassembled packet too large", mp, ill);
7391 freemsg(mp);
7392 return (NULL);
7393 }
7394
7395 if (DB_REF(mp) > 1) {
7396 mblk_t *mp2 = copymsg(mp);
7397
7398 if (mp2 == NULL) {
7399 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
7400 ip_drop_input("ipIfStatsInDiscards", mp, ill);
7401 freemsg(mp);
7402 return (NULL);
7403 }
7404 freemsg(mp);
7405 mp = mp2;
7406 }
7407 ipha = (ipha_t *)mp->b_rptr;
7408
7409 ipha->ipha_length = htons((uint16_t)packet_size);
7410 /* We're now complete, zip the frag state */
7411 ipha->ipha_fragment_offset_and_flags = 0;
7412 /* Record the ECN info. */
7413 ipha->ipha_type_of_service &= 0xFC;
7414 ipha->ipha_type_of_service |= ecn_info;
7415
7416 /* Update the receive attributes */
7417 ira->ira_pktlen = packet_size;
7418 ira->ira_ip_hdr_length = IPH_HDR_LENGTH(ipha);
7419
7420 /* Reassembly is successful; set checksum information in packet */
7421 DB_CKSUM16(mp) = (uint16_t)sum_val;
7422 DB_CKSUMFLAGS(mp) = sum_flags;
7423 DB_CKSUMSTART(mp) = ira->ira_ip_hdr_length;
7424
7425 return (mp);
7426 }
7427
7428 /*
7429 * Pullup function that should be used for IP input in order to
7430 * ensure we do not loose the L2 source address; we need the l2 source
7431 * address for IP_RECVSLLA and for ndp_input.
7432 *
7433 * We return either NULL or b_rptr.
7434 */
7435 void *
7436 ip_pullup(mblk_t *mp, ssize_t len, ip_recv_attr_t *ira)
7437 {
7438 ill_t *ill = ira->ira_ill;
7439
7440 if (ip_rput_pullups++ == 0) {
7441 (void) mi_strlog(ill->ill_rq, 1, SL_ERROR|SL_TRACE,
7442 "ip_pullup: %s forced us to "
7443 " pullup pkt, hdr len %ld, hdr addr %p",
7444 ill->ill_name, len, (void *)mp->b_rptr);
7445 }
7446 if (!(ira->ira_flags & IRAF_L2SRC_SET))
7447 ip_setl2src(mp, ira, ira->ira_rill);
7448 ASSERT(ira->ira_flags & IRAF_L2SRC_SET);
7449 if (!pullupmsg(mp, len))
7450 return (NULL);
7451 else
7452 return (mp->b_rptr);
7453 }
7454
7455 /*
7456 * Make sure ira_l2src has an address. If we don't have one fill with zeros.
7457 * When called from the ULP ira_rill will be NULL hence the caller has to
7458 * pass in the ill.
7459 */
7460 /* ARGSUSED */
7461 void
7462 ip_setl2src(mblk_t *mp, ip_recv_attr_t *ira, ill_t *ill)
7463 {
7464 const uchar_t *addr;
7465 int alen;
7466
7467 if (ira->ira_flags & IRAF_L2SRC_SET)
7468 return;
7469
7470 ASSERT(ill != NULL);
7471 alen = ill->ill_phys_addr_length;
7472 ASSERT(alen <= sizeof (ira->ira_l2src));
7473 if (ira->ira_mhip != NULL &&
7474 (addr = ira->ira_mhip->mhi_saddr) != NULL) {
7475 bcopy(addr, ira->ira_l2src, alen);
7476 } else if ((ira->ira_flags & IRAF_L2SRC_LOOPBACK) &&
7477 (addr = ill->ill_phys_addr) != NULL) {
7478 bcopy(addr, ira->ira_l2src, alen);
7479 } else {
7480 bzero(ira->ira_l2src, alen);
7481 }
7482 ira->ira_flags |= IRAF_L2SRC_SET;
7483 }
7484
7485 /*
7486 * check ip header length and align it.
7487 */
7488 mblk_t *
7489 ip_check_and_align_header(mblk_t *mp, uint_t min_size, ip_recv_attr_t *ira)
7490 {
7491 ill_t *ill = ira->ira_ill;
7492 ssize_t len;
7493
7494 len = MBLKL(mp);
7495
7496 if (!OK_32PTR(mp->b_rptr))
7497 IP_STAT(ill->ill_ipst, ip_notaligned);
7498 else
7499 IP_STAT(ill->ill_ipst, ip_recv_pullup);
7500
7501 /* Guard against bogus device drivers */
7502 if (len < 0) {
7503 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7504 ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7505 freemsg(mp);
7506 return (NULL);
7507 }
7508
7509 if (len == 0) {
7510 /* GLD sometimes sends up mblk with b_rptr == b_wptr! */
7511 mblk_t *mp1 = mp->b_cont;
7512
7513 if (!(ira->ira_flags & IRAF_L2SRC_SET))
7514 ip_setl2src(mp, ira, ira->ira_rill);
7515 ASSERT(ira->ira_flags & IRAF_L2SRC_SET);
7516
7517 freeb(mp);
7518 mp = mp1;
7519 if (mp == NULL)
7520 return (NULL);
7521
7522 if (OK_32PTR(mp->b_rptr) && MBLKL(mp) >= min_size)
7523 return (mp);
7524 }
7525 if (ip_pullup(mp, min_size, ira) == NULL) {
7526 if (msgdsize(mp) < min_size) {
7527 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7528 ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7529 } else {
7530 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
7531 ip_drop_input("ipIfStatsInDiscards", mp, ill);
7532 }
7533 freemsg(mp);
7534 return (NULL);
7535 }
7536 return (mp);
7537 }
7538
7539 /*
7540 * Common code for IPv4 and IPv6 to check and pullup multi-mblks
7541 */
7542 mblk_t *
7543 ip_check_length(mblk_t *mp, uchar_t *rptr, ssize_t len, uint_t pkt_len,
7544 uint_t min_size, ip_recv_attr_t *ira)
7545 {
7546 ill_t *ill = ira->ira_ill;
7547
7548 /*
7549 * Make sure we have data length consistent
7550 * with the IP header.
7551 */
7552 if (mp->b_cont == NULL) {
7553 /* pkt_len is based on ipha_len, not the mblk length */
7554 if (pkt_len < min_size) {
7555 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7556 ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7557 freemsg(mp);
7558 return (NULL);
7559 }
7560 if (len < 0) {
7561 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInTruncatedPkts);
7562 ip_drop_input("ipIfStatsInTruncatedPkts", mp, ill);
7563 freemsg(mp);
7564 return (NULL);
7565 }
7566 /* Drop any pad */
7567 mp->b_wptr = rptr + pkt_len;
7568 } else if ((len += msgdsize(mp->b_cont)) != 0) {
7569 ASSERT(pkt_len >= min_size);
7570 if (pkt_len < min_size) {
7571 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7572 ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7573 freemsg(mp);
7574 return (NULL);
7575 }
7576 if (len < 0) {
7577 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInTruncatedPkts);
7578 ip_drop_input("ipIfStatsInTruncatedPkts", mp, ill);
7579 freemsg(mp);
7580 return (NULL);
7581 }
7582 /* Drop any pad */
7583 (void) adjmsg(mp, -len);
7584 /*
7585 * adjmsg may have freed an mblk from the chain, hence
7586 * invalidate any hw checksum here. This will force IP to
7587 * calculate the checksum in sw, but only for this packet.
7588 */
7589 DB_CKSUMFLAGS(mp) = 0;
7590 IP_STAT(ill->ill_ipst, ip_multimblk);
7591 }
7592 return (mp);
7593 }
7594
7595 /*
7596 * Check that the IPv4 opt_len is consistent with the packet and pullup
7597 * the options.
7598 */
7599 mblk_t *
7600 ip_check_optlen(mblk_t *mp, ipha_t *ipha, uint_t opt_len, uint_t pkt_len,
7601 ip_recv_attr_t *ira)
7602 {
7603 ill_t *ill = ira->ira_ill;
7604 ssize_t len;
7605
7606 /* Assume no IPv6 packets arrive over the IPv4 queue */
7607 if (IPH_HDR_VERSION(ipha) != IPV4_VERSION) {
7608 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7609 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInWrongIPVersion);
7610 ip_drop_input("IPvN packet on IPv4 ill", mp, ill);
7611 freemsg(mp);
7612 return (NULL);
7613 }
7614
7615 if (opt_len > (15 - IP_SIMPLE_HDR_LENGTH_IN_WORDS)) {
7616 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7617 ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7618 freemsg(mp);
7619 return (NULL);
7620 }
7621 /*
7622 * Recompute complete header length and make sure we
7623 * have access to all of it.
7624 */
7625 len = ((size_t)opt_len + IP_SIMPLE_HDR_LENGTH_IN_WORDS) << 2;
7626 if (len > (mp->b_wptr - mp->b_rptr)) {
7627 if (len > pkt_len) {
7628 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7629 ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7630 freemsg(mp);
7631 return (NULL);
7632 }
7633 if (ip_pullup(mp, len, ira) == NULL) {
7634 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
7635 ip_drop_input("ipIfStatsInDiscards", mp, ill);
7636 freemsg(mp);
7637 return (NULL);
7638 }
7639 }
7640 return (mp);
7641 }
7642
7643 /*
7644 * Returns a new ire, or the same ire, or NULL.
7645 * If a different IRE is returned, then it is held; the caller
7646 * needs to release it.
7647 * In no case is there any hold/release on the ire argument.
7648 */
7649 ire_t *
7650 ip_check_multihome(void *addr, ire_t *ire, ill_t *ill)
7651 {
7652 ire_t *new_ire;
7653 ill_t *ire_ill;
7654 uint_t ifindex;
7655 ip_stack_t *ipst = ill->ill_ipst;
7656 boolean_t strict_check = B_FALSE;
7657
7658 /*
7659 * IPMP common case: if IRE and ILL are in the same group, there's no
7660 * issue (e.g. packet received on an underlying interface matched an
7661 * IRE_LOCAL on its associated group interface).
7662 */
7663 ASSERT(ire->ire_ill != NULL);
7664 if (IS_IN_SAME_ILLGRP(ill, ire->ire_ill))
7665 return (ire);
7666
7667 /*
7668 * Do another ire lookup here, using the ingress ill, to see if the
7669 * interface is in a usesrc group.
7670 * As long as the ills belong to the same group, we don't consider
7671 * them to be arriving on the wrong interface. Thus, if the switch
7672 * is doing inbound load spreading, we won't drop packets when the
7673 * ip*_strict_dst_multihoming switch is on.
7674 * We also need to check for IPIF_UNNUMBERED point2point interfaces
7675 * where the local address may not be unique. In this case we were
7676 * at the mercy of the initial ire lookup and the IRE_LOCAL it
7677 * actually returned. The new lookup, which is more specific, should
7678 * only find the IRE_LOCAL associated with the ingress ill if one
7679 * exists.
7680 */
7681 if (ire->ire_ipversion == IPV4_VERSION) {
7682 if (ipst->ips_ip_strict_dst_multihoming)
7683 strict_check = B_TRUE;
7684 new_ire = ire_ftable_lookup_v4(*((ipaddr_t *)addr), 0, 0,
7685 IRE_LOCAL, ill, ALL_ZONES, NULL,
7686 (MATCH_IRE_TYPE|MATCH_IRE_ILL), 0, ipst, NULL);
7687 } else {
7688 ASSERT(!IN6_IS_ADDR_MULTICAST((in6_addr_t *)addr));
7689 if (ipst->ips_ipv6_strict_dst_multihoming)
7690 strict_check = B_TRUE;
7691 new_ire = ire_ftable_lookup_v6((in6_addr_t *)addr, NULL, NULL,
7692 IRE_LOCAL, ill, ALL_ZONES, NULL,
7693 (MATCH_IRE_TYPE|MATCH_IRE_ILL), 0, ipst, NULL);
7694 }
7695 /*
7696 * If the same ire that was returned in ip_input() is found then this
7697 * is an indication that usesrc groups are in use. The packet
7698 * arrived on a different ill in the group than the one associated with
7699 * the destination address. If a different ire was found then the same
7700 * IP address must be hosted on multiple ills. This is possible with
7701 * unnumbered point2point interfaces. We switch to use this new ire in
7702 * order to have accurate interface statistics.
7703 */
7704 if (new_ire != NULL) {
7705 /* Note: held in one case but not the other? Caller handles */
7706 if (new_ire != ire)
7707 return (new_ire);
7708 /* Unchanged */
7709 ire_refrele(new_ire);
7710 return (ire);
7711 }
7712
7713 /*
7714 * Chase pointers once and store locally.
7715 */
7716 ASSERT(ire->ire_ill != NULL);
7717 ire_ill = ire->ire_ill;
7718 ifindex = ill->ill_usesrc_ifindex;
7719
7720 /*
7721 * Check if it's a legal address on the 'usesrc' interface.
7722 * For IPMP data addresses the IRE_LOCAL is the upper, hence we
7723 * can just check phyint_ifindex.
7724 */
7725 if (ifindex != 0 && ifindex == ire_ill->ill_phyint->phyint_ifindex) {
7726 return (ire);
7727 }
7728
7729 /*
7730 * If the ip*_strict_dst_multihoming switch is on then we can
7731 * only accept this packet if the interface is marked as routing.
7732 */
7733 if (!(strict_check))
7734 return (ire);
7735
7736 if ((ill->ill_flags & ire->ire_ill->ill_flags & ILLF_ROUTER) != 0) {
7737 return (ire);
7738 }
7739 return (NULL);
7740 }
7741
7742 /*
7743 * This function is used to construct a mac_header_info_s from a
7744 * DL_UNITDATA_IND message.
7745 * The address fields in the mhi structure points into the message,
7746 * thus the caller can't use those fields after freeing the message.
7747 *
7748 * We determine whether the packet received is a non-unicast packet
7749 * and in doing so, determine whether or not it is broadcast vs multicast.
7750 * For it to be a broadcast packet, we must have the appropriate mblk_t
7751 * hanging off the ill_t. If this is either not present or doesn't match
7752 * the destination mac address in the DL_UNITDATA_IND, the packet is deemed
7753 * to be multicast. Thus NICs that have no broadcast address (or no
7754 * capability for one, such as point to point links) cannot return as
7755 * the packet being broadcast.
7756 */
7757 void
7758 ip_dlur_to_mhi(ill_t *ill, mblk_t *mb, struct mac_header_info_s *mhip)
7759 {
7760 dl_unitdata_ind_t *ind = (dl_unitdata_ind_t *)mb->b_rptr;
7761 mblk_t *bmp;
7762 uint_t extra_offset;
7763
7764 bzero(mhip, sizeof (struct mac_header_info_s));
7765
7766 mhip->mhi_dsttype = MAC_ADDRTYPE_UNICAST;
7767
7768 if (ill->ill_sap_length < 0)
7769 extra_offset = 0;
7770 else
7771 extra_offset = ill->ill_sap_length;
7772
7773 mhip->mhi_daddr = (uchar_t *)ind + ind->dl_dest_addr_offset +
7774 extra_offset;
7775 mhip->mhi_saddr = (uchar_t *)ind + ind->dl_src_addr_offset +
7776 extra_offset;
7777
7778 if (!ind->dl_group_address)
7779 return;
7780
7781 /* Multicast or broadcast */
7782 mhip->mhi_dsttype = MAC_ADDRTYPE_MULTICAST;
7783
7784 if (ind->dl_dest_addr_offset > sizeof (*ind) &&
7785 ind->dl_dest_addr_offset + ind->dl_dest_addr_length < MBLKL(mb) &&
7786 (bmp = ill->ill_bcast_mp) != NULL) {
7787 dl_unitdata_req_t *dlur;
7788 uint8_t *bphys_addr;
7789
7790 dlur = (dl_unitdata_req_t *)bmp->b_rptr;
7791 bphys_addr = (uchar_t *)dlur + dlur->dl_dest_addr_offset +
7792 extra_offset;
7793
7794 if (bcmp(mhip->mhi_daddr, bphys_addr,
7795 ind->dl_dest_addr_length) == 0)
7796 mhip->mhi_dsttype = MAC_ADDRTYPE_BROADCAST;
7797 }
7798 }
7799
7800 /*
7801 * This function is used to construct a mac_header_info_s from a
7802 * M_DATA fastpath message from a DLPI driver.
7803 * The address fields in the mhi structure points into the message,
7804 * thus the caller can't use those fields after freeing the message.
7805 *
7806 * We determine whether the packet received is a non-unicast packet
7807 * and in doing so, determine whether or not it is broadcast vs multicast.
7808 * For it to be a broadcast packet, we must have the appropriate mblk_t
7809 * hanging off the ill_t. If this is either not present or doesn't match
7810 * the destination mac address in the DL_UNITDATA_IND, the packet is deemed
7811 * to be multicast. Thus NICs that have no broadcast address (or no
7812 * capability for one, such as point to point links) cannot return as
7813 * the packet being broadcast.
7814 */
7815 void
7816 ip_mdata_to_mhi(ill_t *ill, mblk_t *mp, struct mac_header_info_s *mhip)
7817 {
7818 mblk_t *bmp;
7819 struct ether_header *pether;
7820
7821 bzero(mhip, sizeof (struct mac_header_info_s));
7822
7823 mhip->mhi_dsttype = MAC_ADDRTYPE_UNICAST;
7824
7825 pether = (struct ether_header *)((char *)mp->b_rptr
7826 - sizeof (struct ether_header));
7827
7828 /*
7829 * Make sure the interface is an ethernet type, since we don't
7830 * know the header format for anything but Ethernet. Also make
7831 * sure we are pointing correctly above db_base.
7832 */
7833 if (ill->ill_type != IFT_ETHER)
7834 return;
7835
7836 /*
7837 * PPP (including PPPoE) interfaces lie about being an ethernet, but
7838 * have other clues to indicate we should just bail now. We exploit
7839 * the 0 broadcast address length.
7840 *
7841 * XXX KEBE ASKS --> what about IP multicast packets?
7842 */
7843 if (ill->ill_bcast_addr_length == 0)
7844 return;
7845
7846 retry:
7847 if ((uchar_t *)pether < mp->b_datap->db_base)
7848 return;
7849
7850 /* Is there a VLAN tag? */
7851 if (ill->ill_isv6) {
7852 if (pether->ether_type != htons(ETHERTYPE_IPV6)) {
7853 pether = (struct ether_header *)((char *)pether - 4);
7854 goto retry;
7855 }
7856 } else {
7857 if (pether->ether_type != htons(ETHERTYPE_IP)) {
7858 pether = (struct ether_header *)((char *)pether - 4);
7859 goto retry;
7860 }
7861 }
7862 mhip->mhi_daddr = (uchar_t *)&pether->ether_dhost;
7863 mhip->mhi_saddr = (uchar_t *)&pether->ether_shost;
7864
7865 if (!(mhip->mhi_daddr[0] & 0x01))
7866 return;
7867
7868 /* Multicast or broadcast */
7869 mhip->mhi_dsttype = MAC_ADDRTYPE_MULTICAST;
7870
7871 if ((bmp = ill->ill_bcast_mp) != NULL) {
7872 dl_unitdata_req_t *dlur;
7873 uint8_t *bphys_addr;
7874 uint_t addrlen;
7875
7876 dlur = (dl_unitdata_req_t *)bmp->b_rptr;
7877 addrlen = dlur->dl_dest_addr_length;
7878 if (ill->ill_sap_length < 0) {
7879 bphys_addr = (uchar_t *)dlur +
7880 dlur->dl_dest_addr_offset;
7881 addrlen += ill->ill_sap_length;
7882 } else {
7883 bphys_addr = (uchar_t *)dlur +
7884 dlur->dl_dest_addr_offset +
7885 ill->ill_sap_length;
7886 addrlen -= ill->ill_sap_length;
7887 }
7888 if (bcmp(mhip->mhi_daddr, bphys_addr, addrlen) == 0)
7889 mhip->mhi_dsttype = MAC_ADDRTYPE_BROADCAST;
7890 }
7891 }
7892
7893 /*
7894 * Handle anything but M_DATA messages
7895 * We see the DL_UNITDATA_IND which are part
7896 * of the data path, and also the other messages from the driver.
7897 */
7898 void
7899 ip_rput_notdata(ill_t *ill, mblk_t *mp)
7900 {
7901 mblk_t *first_mp;
7902 struct iocblk *iocp;
7903 struct mac_header_info_s mhi;
7904
7905 switch (DB_TYPE(mp)) {
7906 case M_PROTO:
7907 case M_PCPROTO: {
7908 if (((dl_unitdata_ind_t *)mp->b_rptr)->dl_primitive !=
7909 DL_UNITDATA_IND) {
7910 /* Go handle anything other than data elsewhere. */
7911 ip_rput_dlpi(ill, mp);
7912 return;
7913 }
7914
7915 first_mp = mp;
7916 mp = first_mp->b_cont;
7917 first_mp->b_cont = NULL;
7918
7919 if (mp == NULL) {
7920 freeb(first_mp);
7921 return;
7922 }
7923 ip_dlur_to_mhi(ill, first_mp, &mhi);
7924 if (ill->ill_isv6)
7925 ip_input_v6(ill, NULL, mp, &mhi);
7926 else
7927 ip_input(ill, NULL, mp, &mhi);
7928
7929 /* Ditch the DLPI header. */
7930 freeb(first_mp);
7931 return;
7932 }
7933 case M_IOCACK:
7934 iocp = (struct iocblk *)mp->b_rptr;
7935 switch (iocp->ioc_cmd) {
7936 case DL_IOC_HDR_INFO:
7937 ill_fastpath_ack(ill, mp);
7938 return;
7939 default:
7940 putnext(ill->ill_rq, mp);
7941 return;
7942 }
7943 /* FALLTHRU */
7944 case M_ERROR:
7945 case M_HANGUP:
7946 mutex_enter(&ill->ill_lock);
7947 if (ill->ill_state_flags & ILL_CONDEMNED) {
7948 mutex_exit(&ill->ill_lock);
7949 freemsg(mp);
7950 return;
7951 }
7952 ill_refhold_locked(ill);
7953 mutex_exit(&ill->ill_lock);
7954 qwriter_ip(ill, ill->ill_rq, mp, ip_rput_other, CUR_OP,
7955 B_FALSE);
7956 return;
7957 case M_CTL:
7958 putnext(ill->ill_rq, mp);
7959 return;
7960 case M_IOCNAK:
7961 ip1dbg(("got iocnak "));
7962 iocp = (struct iocblk *)mp->b_rptr;
7963 switch (iocp->ioc_cmd) {
7964 case DL_IOC_HDR_INFO:
7965 ip_rput_other(NULL, ill->ill_rq, mp, NULL);
7966 return;
7967 default:
7968 break;
7969 }
7970 /* FALLTHRU */
7971 default:
7972 putnext(ill->ill_rq, mp);
7973 return;
7974 }
7975 }
7976
7977 /* Read side put procedure. Packets coming from the wire arrive here. */
7978 void
7979 ip_rput(queue_t *q, mblk_t *mp)
7980 {
7981 ill_t *ill;
7982 union DL_primitives *dl;
7983
7984 ill = (ill_t *)q->q_ptr;
7985
7986 if (ill->ill_state_flags & (ILL_CONDEMNED | ILL_LL_SUBNET_PENDING)) {
7987 /*
7988 * If things are opening or closing, only accept high-priority
7989 * DLPI messages. (On open ill->ill_ipif has not yet been
7990 * created; on close, things hanging off the ill may have been
7991 * freed already.)
7992 */
7993 dl = (union DL_primitives *)mp->b_rptr;
7994 if (DB_TYPE(mp) != M_PCPROTO ||
7995 dl->dl_primitive == DL_UNITDATA_IND) {
7996 inet_freemsg(mp);
7997 return;
7998 }
7999 }
8000 if (DB_TYPE(mp) == M_DATA) {
8001 struct mac_header_info_s mhi;
8002
8003 ip_mdata_to_mhi(ill, mp, &mhi);
8004 ip_input(ill, NULL, mp, &mhi);
8005 } else {
8006 ip_rput_notdata(ill, mp);
8007 }
8008 }
8009
8010 /*
8011 * Move the information to a copy.
8012 */
8013 mblk_t *
8014 ip_fix_dbref(mblk_t *mp, ip_recv_attr_t *ira)
8015 {
8016 mblk_t *mp1;
8017 ill_t *ill = ira->ira_ill;
8018 ip_stack_t *ipst = ill->ill_ipst;
8019
8020 IP_STAT(ipst, ip_db_ref);
8021
8022 /* Make sure we have ira_l2src before we loose the original mblk */
8023 if (!(ira->ira_flags & IRAF_L2SRC_SET))
8024 ip_setl2src(mp, ira, ira->ira_rill);
8025
8026 mp1 = copymsg(mp);
8027 if (mp1 == NULL) {
8028 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
8029 ip_drop_input("ipIfStatsInDiscards", mp, ill);
8030 freemsg(mp);
8031 return (NULL);
8032 }
8033 /* preserve the hardware checksum flags and data, if present */
8034 if (DB_CKSUMFLAGS(mp) != 0) {
8035 DB_CKSUMFLAGS(mp1) = DB_CKSUMFLAGS(mp);
8036 DB_CKSUMSTART(mp1) = DB_CKSUMSTART(mp);
8037 DB_CKSUMSTUFF(mp1) = DB_CKSUMSTUFF(mp);
8038 DB_CKSUMEND(mp1) = DB_CKSUMEND(mp);
8039 DB_CKSUM16(mp1) = DB_CKSUM16(mp);
8040 }
8041 freemsg(mp);
8042 return (mp1);
8043 }
8044
8045 static void
8046 ip_dlpi_error(ill_t *ill, t_uscalar_t prim, t_uscalar_t dl_err,
8047 t_uscalar_t err)
8048 {
8049 if (dl_err == DL_SYSERR) {
8050 (void) mi_strlog(ill->ill_rq, 1, SL_CONSOLE|SL_ERROR|SL_TRACE,
8051 "%s: %s failed: DL_SYSERR (errno %u)\n",
8052 ill->ill_name, dl_primstr(prim), err);
8053 return;
8054 }
8055
8056 (void) mi_strlog(ill->ill_rq, 1, SL_CONSOLE|SL_ERROR|SL_TRACE,
8057 "%s: %s failed: %s\n", ill->ill_name, dl_primstr(prim),
8058 dl_errstr(dl_err));
8059 }
8060
8061 /*
8062 * ip_rput_dlpi is called by ip_rput to handle all DLPI messages other
8063 * than DL_UNITDATA_IND messages. If we need to process this message
8064 * exclusively, we call qwriter_ip, in which case we also need to call
8065 * ill_refhold before that, since qwriter_ip does an ill_refrele.
8066 */
8067 void
8068 ip_rput_dlpi(ill_t *ill, mblk_t *mp)
8069 {
8070 dl_ok_ack_t *dloa = (dl_ok_ack_t *)mp->b_rptr;
8071 dl_error_ack_t *dlea = (dl_error_ack_t *)dloa;
8072 queue_t *q = ill->ill_rq;
8073 t_uscalar_t prim = dloa->dl_primitive;
8074 t_uscalar_t reqprim = DL_PRIM_INVAL;
8075
8076 DTRACE_PROBE3(ill__dlpi, char *, "ip_rput_dlpi",
8077 char *, dl_primstr(prim), ill_t *, ill);
8078 ip1dbg(("ip_rput_dlpi"));
8079
8080 /*
8081 * If we received an ACK but didn't send a request for it, then it
8082 * can't be part of any pending operation; discard up-front.
8083 */
8084 switch (prim) {
8085 case DL_ERROR_ACK:
8086 reqprim = dlea->dl_error_primitive;
8087 ip2dbg(("ip_rput_dlpi(%s): DL_ERROR_ACK for %s (0x%x): %s "
8088 "(0x%x), unix %u\n", ill->ill_name, dl_primstr(reqprim),
8089 reqprim, dl_errstr(dlea->dl_errno), dlea->dl_errno,
8090 dlea->dl_unix_errno));
8091 break;
8092 case DL_OK_ACK:
8093 reqprim = dloa->dl_correct_primitive;
8094 break;
8095 case DL_INFO_ACK:
8096 reqprim = DL_INFO_REQ;
8097 break;
8098 case DL_BIND_ACK:
8099 reqprim = DL_BIND_REQ;
8100 break;
8101 case DL_PHYS_ADDR_ACK:
8102 reqprim = DL_PHYS_ADDR_REQ;
8103 break;
8104 case DL_NOTIFY_ACK:
8105 reqprim = DL_NOTIFY_REQ;
8106 break;
8107 case DL_CAPABILITY_ACK:
8108 reqprim = DL_CAPABILITY_REQ;
8109 break;
8110 }
8111
8112 if (prim != DL_NOTIFY_IND) {
8113 if (reqprim == DL_PRIM_INVAL ||
8114 !ill_dlpi_pending(ill, reqprim)) {
8115 /* Not a DLPI message we support or expected */
8116 freemsg(mp);
8117 return;
8118 }
8119 ip1dbg(("ip_rput: received %s for %s\n", dl_primstr(prim),
8120 dl_primstr(reqprim)));
8121 }
8122
8123 switch (reqprim) {
8124 case DL_UNBIND_REQ:
8125 /*
8126 * NOTE: we mark the unbind as complete even if we got a
8127 * DL_ERROR_ACK, since there's not much else we can do.
8128 */
8129 mutex_enter(&ill->ill_lock);
8130 ill->ill_state_flags &= ~ILL_DL_UNBIND_IN_PROGRESS;
8131 cv_signal(&ill->ill_cv);
8132 mutex_exit(&ill->ill_lock);
8133 break;
8134
8135 case DL_ENABMULTI_REQ:
8136 if (prim == DL_OK_ACK) {
8137 if (ill->ill_dlpi_multicast_state == IDS_INPROGRESS)
8138 ill->ill_dlpi_multicast_state = IDS_OK;
8139 }
8140 break;
8141 }
8142
8143 /*
8144 * The message is one we're waiting for (or DL_NOTIFY_IND), but we
8145 * need to become writer to continue to process it. Because an
8146 * exclusive operation doesn't complete until replies to all queued
8147 * DLPI messages have been received, we know we're in the middle of an
8148 * exclusive operation and pass CUR_OP (except for DL_NOTIFY_IND).
8149 *
8150 * As required by qwriter_ip(), we refhold the ill; it will refrele.
8151 * Since this is on the ill stream we unconditionally bump up the
8152 * refcount without doing ILL_CAN_LOOKUP().
8153 */
8154 ill_refhold(ill);
8155 if (prim == DL_NOTIFY_IND)
8156 qwriter_ip(ill, q, mp, ip_rput_dlpi_writer, NEW_OP, B_FALSE);
8157 else
8158 qwriter_ip(ill, q, mp, ip_rput_dlpi_writer, CUR_OP, B_FALSE);
8159 }
8160
8161 /*
8162 * Handling of DLPI messages that require exclusive access to the ipsq.
8163 *
8164 * Need to do ipsq_pending_mp_get on ioctl completion, which could
8165 * happen here. (along with mi_copy_done)
8166 */
8167 /* ARGSUSED */
8168 static void
8169 ip_rput_dlpi_writer(ipsq_t *ipsq, queue_t *q, mblk_t *mp, void *dummy_arg)
8170 {
8171 dl_ok_ack_t *dloa = (dl_ok_ack_t *)mp->b_rptr;
8172 dl_error_ack_t *dlea = (dl_error_ack_t *)dloa;
8173 int err = 0;
8174 ill_t *ill = (ill_t *)q->q_ptr;
8175 ipif_t *ipif = NULL;
8176 mblk_t *mp1 = NULL;
8177 conn_t *connp = NULL;
8178 t_uscalar_t paddrreq;
8179 mblk_t *mp_hw;
8180 boolean_t success;
8181 boolean_t ioctl_aborted = B_FALSE;
8182 boolean_t log = B_TRUE;
8183
8184 DTRACE_PROBE3(ill__dlpi, char *, "ip_rput_dlpi_writer",
8185 char *, dl_primstr(dloa->dl_primitive), ill_t *, ill);
8186
8187 ip1dbg(("ip_rput_dlpi_writer .."));
8188 ASSERT(ipsq->ipsq_xop == ill->ill_phyint->phyint_ipsq->ipsq_xop);
8189 ASSERT(IAM_WRITER_ILL(ill));
8190
8191 ipif = ipsq->ipsq_xop->ipx_pending_ipif;
8192 /*
8193 * The current ioctl could have been aborted by the user and a new
8194 * ioctl to bring up another ill could have started. We could still
8195 * get a response from the driver later.
8196 */
8197 if (ipif != NULL && ipif->ipif_ill != ill)
8198 ioctl_aborted = B_TRUE;
8199
8200 switch (dloa->dl_primitive) {
8201 case DL_ERROR_ACK:
8202 ip1dbg(("ip_rput_dlpi_writer: got DL_ERROR_ACK for %s\n",
8203 dl_primstr(dlea->dl_error_primitive)));
8204
8205 DTRACE_PROBE3(ill__dlpi, char *, "ip_rput_dlpi_writer error",
8206 char *, dl_primstr(dlea->dl_error_primitive),
8207 ill_t *, ill);
8208
8209 switch (dlea->dl_error_primitive) {
8210 case DL_DISABMULTI_REQ:
8211 ill_dlpi_done(ill, dlea->dl_error_primitive);
8212 break;
8213 case DL_PROMISCON_REQ:
8214 case DL_PROMISCOFF_REQ:
8215 case DL_UNBIND_REQ:
8216 case DL_ATTACH_REQ:
8217 case DL_INFO_REQ:
8218 ill_dlpi_done(ill, dlea->dl_error_primitive);
8219 break;
8220 case DL_NOTIFY_REQ:
8221 ill_dlpi_done(ill, DL_NOTIFY_REQ);
8222 log = B_FALSE;
8223 break;
8224 case DL_PHYS_ADDR_REQ:
8225 /*
8226 * For IPv6 only, there are two additional
8227 * phys_addr_req's sent to the driver to get the
8228 * IPv6 token and lla. This allows IP to acquire
8229 * the hardware address format for a given interface
8230 * without having built in knowledge of the hardware
8231 * address. ill_phys_addr_pend keeps track of the last
8232 * DL_PAR sent so we know which response we are
8233 * dealing with. ill_dlpi_done will update
8234 * ill_phys_addr_pend when it sends the next req.
8235 * We don't complete the IOCTL until all three DL_PARs
8236 * have been attempted, so set *_len to 0 and break.
8237 */
8238 paddrreq = ill->ill_phys_addr_pend;
8239 ill_dlpi_done(ill, DL_PHYS_ADDR_REQ);
8240 if (paddrreq == DL_IPV6_TOKEN) {
8241 ill->ill_token_length = 0;
8242 log = B_FALSE;
8243 break;
8244 } else if (paddrreq == DL_IPV6_LINK_LAYER_ADDR) {
8245 ill->ill_nd_lla_len = 0;
8246 log = B_FALSE;
8247 break;
8248 }
8249 /*
8250 * Something went wrong with the DL_PHYS_ADDR_REQ.
8251 * We presumably have an IOCTL hanging out waiting
8252 * for completion. Find it and complete the IOCTL
8253 * with the error noted.
8254 * However, ill_dl_phys was called on an ill queue
8255 * (from SIOCSLIFNAME), thus conn_pending_ill is not
8256 * set. But the ioctl is known to be pending on ill_wq.
8257 */
8258 if (!ill->ill_ifname_pending)
8259 break;
8260 ill->ill_ifname_pending = 0;
8261 if (!ioctl_aborted)
8262 mp1 = ipsq_pending_mp_get(ipsq, &connp);
8263 if (mp1 != NULL) {
8264 /*
8265 * This operation (SIOCSLIFNAME) must have
8266 * happened on the ill. Assert there is no conn
8267 */
8268 ASSERT(connp == NULL);
8269 q = ill->ill_wq;
8270 }
8271 break;
8272 case DL_BIND_REQ:
8273 ill_dlpi_done(ill, DL_BIND_REQ);
8274 if (ill->ill_ifname_pending)
8275 break;
8276 mutex_enter(&ill->ill_lock);
8277 ill->ill_state_flags &= ~ILL_DOWN_IN_PROGRESS;
8278 mutex_exit(&ill->ill_lock);
8279 /*
8280 * Something went wrong with the bind. We presumably
8281 * have an IOCTL hanging out waiting for completion.
8282 * Find it, take down the interface that was coming
8283 * up, and complete the IOCTL with the error noted.
8284 */
8285 if (!ioctl_aborted)
8286 mp1 = ipsq_pending_mp_get(ipsq, &connp);
8287 if (mp1 != NULL) {
8288 /*
8289 * This might be a result of a DL_NOTE_REPLUMB
8290 * notification. In that case, connp is NULL.
8291 */
8292 if (connp != NULL)
8293 q = CONNP_TO_WQ(connp);
8294
8295 (void) ipif_down(ipif, NULL, NULL);
8296 /* error is set below the switch */
8297 }
8298 break;
8299 case DL_ENABMULTI_REQ:
8300 ill_dlpi_done(ill, DL_ENABMULTI_REQ);
8301
8302 if (ill->ill_dlpi_multicast_state == IDS_INPROGRESS)
8303 ill->ill_dlpi_multicast_state = IDS_FAILED;
8304 if (ill->ill_dlpi_multicast_state == IDS_FAILED) {
8305
8306 printf("ip: joining multicasts failed (%d)"
8307 " on %s - will use link layer "
8308 "broadcasts for multicast\n",
8309 dlea->dl_errno, ill->ill_name);
8310
8311 /*
8312 * Set up for multi_bcast; We are the
8313 * writer, so ok to access ill->ill_ipif
8314 * without any lock.
8315 */
8316 mutex_enter(&ill->ill_phyint->phyint_lock);
8317 ill->ill_phyint->phyint_flags |=
8318 PHYI_MULTI_BCAST;
8319 mutex_exit(&ill->ill_phyint->phyint_lock);
8320
8321 }
8322 freemsg(mp); /* Don't want to pass this up */
8323 return;
8324 case DL_CAPABILITY_REQ:
8325 ip1dbg(("ip_rput_dlpi_writer: got DL_ERROR_ACK for "
8326 "DL_CAPABILITY REQ\n"));
8327 if (ill->ill_dlpi_capab_state == IDCS_PROBE_SENT)
8328 ill->ill_dlpi_capab_state = IDCS_FAILED;
8329 ill_capability_done(ill);
8330 freemsg(mp);
8331 return;
8332 }
8333 /*
8334 * Note the error for IOCTL completion (mp1 is set when
8335 * ready to complete ioctl). If ill_ifname_pending_err is
8336 * set, an error occured during plumbing (ill_ifname_pending),
8337 * so we want to report that error.
8338 *
8339 * NOTE: there are two addtional DL_PHYS_ADDR_REQ's
8340 * (DL_IPV6_TOKEN and DL_IPV6_LINK_LAYER_ADDR) that are
8341 * expected to get errack'd if the driver doesn't support
8342 * these flags (e.g. ethernet). log will be set to B_FALSE
8343 * if these error conditions are encountered.
8344 */
8345 if (mp1 != NULL) {
8346 if (ill->ill_ifname_pending_err != 0) {
8347 err = ill->ill_ifname_pending_err;
8348 ill->ill_ifname_pending_err = 0;
8349 } else {
8350 err = dlea->dl_unix_errno ?
8351 dlea->dl_unix_errno : ENXIO;
8352 }
8353 /*
8354 * If we're plumbing an interface and an error hasn't already
8355 * been saved, set ill_ifname_pending_err to the error passed
8356 * up. Ignore the error if log is B_FALSE (see comment above).
8357 */
8358 } else if (log && ill->ill_ifname_pending &&
8359 ill->ill_ifname_pending_err == 0) {
8360 ill->ill_ifname_pending_err = dlea->dl_unix_errno ?
8361 dlea->dl_unix_errno : ENXIO;
8362 }
8363
8364 if (log)
8365 ip_dlpi_error(ill, dlea->dl_error_primitive,
8366 dlea->dl_errno, dlea->dl_unix_errno);
8367 break;
8368 case DL_CAPABILITY_ACK:
8369 ill_capability_ack(ill, mp);
8370 /*
8371 * The message has been handed off to ill_capability_ack
8372 * and must not be freed below
8373 */
8374 mp = NULL;
8375 break;
8376
8377 case DL_INFO_ACK:
8378 /* Call a routine to handle this one. */
8379 ill_dlpi_done(ill, DL_INFO_REQ);
8380 ip_ll_subnet_defaults(ill, mp);
8381 ASSERT(!MUTEX_HELD(&ill->ill_phyint->phyint_ipsq->ipsq_lock));
8382 return;
8383 case DL_BIND_ACK:
8384 /*
8385 * We should have an IOCTL waiting on this unless
8386 * sent by ill_dl_phys, in which case just return
8387 */
8388 ill_dlpi_done(ill, DL_BIND_REQ);
8389
8390 if (ill->ill_ifname_pending) {
8391 DTRACE_PROBE2(ip__rput__dlpi__ifname__pending,
8392 ill_t *, ill, mblk_t *, mp);
8393 break;
8394 }
8395 mutex_enter(&ill->ill_lock);
8396 ill->ill_dl_up = 1;
8397 ill->ill_state_flags &= ~ILL_DOWN_IN_PROGRESS;
8398 mutex_exit(&ill->ill_lock);
8399
8400 if (!ioctl_aborted)
8401 mp1 = ipsq_pending_mp_get(ipsq, &connp);
8402 if (mp1 == NULL) {
8403 DTRACE_PROBE1(ip__rput__dlpi__no__mblk, ill_t *, ill);
8404 break;
8405 }
8406 /*
8407 * mp1 was added by ill_dl_up(). if that is a result of
8408 * a DL_NOTE_REPLUMB notification, connp could be NULL.
8409 */
8410 if (connp != NULL)
8411 q = CONNP_TO_WQ(connp);
8412 /*
8413 * We are exclusive. So nothing can change even after
8414 * we get the pending mp.
8415 */
8416 ip1dbg(("ip_rput_dlpi: bind_ack %s\n", ill->ill_name));
8417 DTRACE_PROBE1(ip__rput__dlpi__bind__ack, ill_t *, ill);
8418 ill_nic_event_dispatch(ill, 0, NE_UP, NULL, 0);
8419
8420 /*
8421 * Now bring up the resolver; when that is complete, we'll
8422 * create IREs. Note that we intentionally mirror what
8423 * ipif_up() would have done, because we got here by way of
8424 * ill_dl_up(), which stopped ipif_up()'s processing.
8425 */
8426 if (ill->ill_isv6) {
8427 /*
8428 * v6 interfaces.
8429 * Unlike ARP which has to do another bind
8430 * and attach, once we get here we are
8431 * done with NDP
8432 */
8433 (void) ipif_resolver_up(ipif, Res_act_initial);
8434 if ((err = ipif_ndp_up(ipif, B_TRUE)) == 0)
8435 err = ipif_up_done_v6(ipif);
8436 } else if (ill->ill_net_type == IRE_IF_RESOLVER) {
8437 /*
8438 * ARP and other v4 external resolvers.
8439 * Leave the pending mblk intact so that
8440 * the ioctl completes in ip_rput().
8441 */
8442 if (connp != NULL)
8443 mutex_enter(&connp->conn_lock);
8444 mutex_enter(&ill->ill_lock);
8445 success = ipsq_pending_mp_add(connp, ipif, q, mp1, 0);
8446 mutex_exit(&ill->ill_lock);
8447 if (connp != NULL)
8448 mutex_exit(&connp->conn_lock);
8449 if (success) {
8450 err = ipif_resolver_up(ipif, Res_act_initial);
8451 if (err == EINPROGRESS) {
8452 freemsg(mp);
8453 return;
8454 }
8455 mp1 = ipsq_pending_mp_get(ipsq, &connp);
8456 } else {
8457 /* The conn has started closing */
8458 err = EINTR;
8459 }
8460 } else {
8461 /*
8462 * This one is complete. Reply to pending ioctl.
8463 */
8464 (void) ipif_resolver_up(ipif, Res_act_initial);
8465 err = ipif_up_done(ipif);
8466 }
8467
8468 if ((err == 0) && (ill->ill_up_ipifs)) {
8469 err = ill_up_ipifs(ill, q, mp1);
8470 if (err == EINPROGRESS) {
8471 freemsg(mp);
8472 return;
8473 }
8474 }
8475
8476 /*
8477 * If we have a moved ipif to bring up, and everything has
8478 * succeeded to this point, bring it up on the IPMP ill.
8479 * Otherwise, leave it down -- the admin can try to bring it
8480 * up by hand if need be.
8481 */
8482 if (ill->ill_move_ipif != NULL) {
8483 if (err != 0) {
8484 ill->ill_move_ipif = NULL;
8485 } else {
8486 ipif = ill->ill_move_ipif;
8487 ill->ill_move_ipif = NULL;
8488 err = ipif_up(ipif, q, mp1);
8489 if (err == EINPROGRESS) {
8490 freemsg(mp);
8491 return;
8492 }
8493 }
8494 }
8495 break;
8496
8497 case DL_NOTIFY_IND: {
8498 dl_notify_ind_t *notify = (dl_notify_ind_t *)mp->b_rptr;
8499 uint_t orig_mtu, orig_mc_mtu;
8500
8501 switch (notify->dl_notification) {
8502 case DL_NOTE_PHYS_ADDR:
8503 err = ill_set_phys_addr(ill, mp);
8504 break;
8505
8506 case DL_NOTE_REPLUMB:
8507 /*
8508 * Directly return after calling ill_replumb().
8509 * Note that we should not free mp as it is reused
8510 * in the ill_replumb() function.
8511 */
8512 err = ill_replumb(ill, mp);
8513 return;
8514
8515 case DL_NOTE_FASTPATH_FLUSH:
8516 nce_flush(ill, B_FALSE);
8517 break;
8518
8519 case DL_NOTE_SDU_SIZE:
8520 case DL_NOTE_SDU_SIZE2:
8521 /*
8522 * The dce and fragmentation code can cope with
8523 * this changing while packets are being sent.
8524 * When packets are sent ip_output will discover
8525 * a change.
8526 *
8527 * Change the MTU size of the interface.
8528 */
8529 mutex_enter(&ill->ill_lock);
8530 orig_mtu = ill->ill_mtu;
8531 orig_mc_mtu = ill->ill_mc_mtu;
8532 switch (notify->dl_notification) {
8533 case DL_NOTE_SDU_SIZE:
8534 ill->ill_current_frag =
8535 (uint_t)notify->dl_data;
8536 ill->ill_mc_mtu = (uint_t)notify->dl_data;
8537 break;
8538 case DL_NOTE_SDU_SIZE2:
8539 ill->ill_current_frag =
8540 (uint_t)notify->dl_data1;
8541 ill->ill_mc_mtu = (uint_t)notify->dl_data2;
8542 break;
8543 }
8544 if (ill->ill_current_frag > ill->ill_max_frag)
8545 ill->ill_max_frag = ill->ill_current_frag;
8546
8547 if (!(ill->ill_flags & ILLF_FIXEDMTU)) {
8548 ill->ill_mtu = ill->ill_current_frag;
8549
8550 /*
8551 * If ill_user_mtu was set (via
8552 * SIOCSLIFLNKINFO), clamp ill_mtu at it.
8553 */
8554 if (ill->ill_user_mtu != 0 &&
8555 ill->ill_user_mtu < ill->ill_mtu)
8556 ill->ill_mtu = ill->ill_user_mtu;
8557
8558 if (ill->ill_user_mtu != 0 &&
8559 ill->ill_user_mtu < ill->ill_mc_mtu)
8560 ill->ill_mc_mtu = ill->ill_user_mtu;
8561
8562 if (ill->ill_isv6) {
8563 if (ill->ill_mtu < IPV6_MIN_MTU)
8564 ill->ill_mtu = IPV6_MIN_MTU;
8565 if (ill->ill_mc_mtu < IPV6_MIN_MTU)
8566 ill->ill_mc_mtu = IPV6_MIN_MTU;
8567 } else {
8568 if (ill->ill_mtu < IP_MIN_MTU)
8569 ill->ill_mtu = IP_MIN_MTU;
8570 if (ill->ill_mc_mtu < IP_MIN_MTU)
8571 ill->ill_mc_mtu = IP_MIN_MTU;
8572 }
8573 } else if (ill->ill_mc_mtu > ill->ill_mtu) {
8574 ill->ill_mc_mtu = ill->ill_mtu;
8575 }
8576
8577 mutex_exit(&ill->ill_lock);
8578 /*
8579 * Make sure all dce_generation checks find out
8580 * that ill_mtu/ill_mc_mtu has changed.
8581 */
8582 if (orig_mtu != ill->ill_mtu ||
8583 orig_mc_mtu != ill->ill_mc_mtu) {
8584 dce_increment_all_generations(ill->ill_isv6,
8585 ill->ill_ipst);
8586 }
8587
8588 /*
8589 * Refresh IPMP meta-interface MTU if necessary.
8590 */
8591 if (IS_UNDER_IPMP(ill))
8592 ipmp_illgrp_refresh_mtu(ill->ill_grp);
8593 break;
8594
8595 case DL_NOTE_LINK_UP:
8596 case DL_NOTE_LINK_DOWN: {
8597 /*
8598 * We are writer. ill / phyint / ipsq assocs stable.
8599 * The RUNNING flag reflects the state of the link.
8600 */
8601 phyint_t *phyint = ill->ill_phyint;
8602 uint64_t new_phyint_flags;
8603 boolean_t changed = B_FALSE;
8604 boolean_t went_up;
8605
8606 went_up = notify->dl_notification == DL_NOTE_LINK_UP;
8607 mutex_enter(&phyint->phyint_lock);
8608
8609 new_phyint_flags = went_up ?
8610 phyint->phyint_flags | PHYI_RUNNING :
8611 phyint->phyint_flags & ~PHYI_RUNNING;
8612
8613 if (IS_IPMP(ill)) {
8614 new_phyint_flags = went_up ?
8615 new_phyint_flags & ~PHYI_FAILED :
8616 new_phyint_flags | PHYI_FAILED;
8617 }
8618
8619 if (new_phyint_flags != phyint->phyint_flags) {
8620 phyint->phyint_flags = new_phyint_flags;
8621 changed = B_TRUE;
8622 }
8623 mutex_exit(&phyint->phyint_lock);
8624 /*
8625 * ill_restart_dad handles the DAD restart and routing
8626 * socket notification logic.
8627 */
8628 if (changed) {
8629 ill_restart_dad(phyint->phyint_illv4, went_up);
8630 ill_restart_dad(phyint->phyint_illv6, went_up);
8631 }
8632 break;
8633 }
8634 case DL_NOTE_PROMISC_ON_PHYS: {
8635 phyint_t *phyint = ill->ill_phyint;
8636
8637 mutex_enter(&phyint->phyint_lock);
8638 phyint->phyint_flags |= PHYI_PROMISC;
8639 mutex_exit(&phyint->phyint_lock);
8640 break;
8641 }
8642 case DL_NOTE_PROMISC_OFF_PHYS: {
8643 phyint_t *phyint = ill->ill_phyint;
8644
8645 mutex_enter(&phyint->phyint_lock);
8646 phyint->phyint_flags &= ~PHYI_PROMISC;
8647 mutex_exit(&phyint->phyint_lock);
8648 break;
8649 }
8650 case DL_NOTE_CAPAB_RENEG:
8651 /*
8652 * Something changed on the driver side.
8653 * It wants us to renegotiate the capabilities
8654 * on this ill. One possible cause is the aggregation
8655 * interface under us where a port got added or
8656 * went away.
8657 *
8658 * If the capability negotiation is already done
8659 * or is in progress, reset the capabilities and
8660 * mark the ill's ill_capab_reneg to be B_TRUE,
8661 * so that when the ack comes back, we can start
8662 * the renegotiation process.
8663 *
8664 * Note that if ill_capab_reneg is already B_TRUE
8665 * (ill_dlpi_capab_state is IDS_UNKNOWN in this case),
8666 * the capability resetting request has been sent
8667 * and the renegotiation has not been started yet;
8668 * nothing needs to be done in this case.
8669 */
8670 ipsq_current_start(ipsq, ill->ill_ipif, 0);
8671 ill_capability_reset(ill, B_TRUE);
8672 ipsq_current_finish(ipsq);
8673 break;
8674
8675 case DL_NOTE_ALLOWED_IPS:
8676 ill_set_allowed_ips(ill, mp);
8677 break;
8678 default:
8679 ip0dbg(("ip_rput_dlpi_writer: unknown notification "
8680 "type 0x%x for DL_NOTIFY_IND\n",
8681 notify->dl_notification));
8682 break;
8683 }
8684
8685 /*
8686 * As this is an asynchronous operation, we
8687 * should not call ill_dlpi_done
8688 */
8689 break;
8690 }
8691 case DL_NOTIFY_ACK: {
8692 dl_notify_ack_t *noteack = (dl_notify_ack_t *)mp->b_rptr;
8693
8694 if (noteack->dl_notifications & DL_NOTE_LINK_UP)
8695 ill->ill_note_link = 1;
8696 ill_dlpi_done(ill, DL_NOTIFY_REQ);
8697 break;
8698 }
8699 case DL_PHYS_ADDR_ACK: {
8700 /*
8701 * As part of plumbing the interface via SIOCSLIFNAME,
8702 * ill_dl_phys() will queue a series of DL_PHYS_ADDR_REQs,
8703 * whose answers we receive here. As each answer is received,
8704 * we call ill_dlpi_done() to dispatch the next request as
8705 * we're processing the current one. Once all answers have
8706 * been received, we use ipsq_pending_mp_get() to dequeue the
8707 * outstanding IOCTL and reply to it. (Because ill_dl_phys()
8708 * is invoked from an ill queue, conn_oper_pending_ill is not
8709 * available, but we know the ioctl is pending on ill_wq.)
8710 */
8711 uint_t paddrlen, paddroff;
8712 uint8_t *addr;
8713
8714 paddrreq = ill->ill_phys_addr_pend;
8715 paddrlen = ((dl_phys_addr_ack_t *)mp->b_rptr)->dl_addr_length;
8716 paddroff = ((dl_phys_addr_ack_t *)mp->b_rptr)->dl_addr_offset;
8717 addr = mp->b_rptr + paddroff;
8718
8719 ill_dlpi_done(ill, DL_PHYS_ADDR_REQ);
8720 if (paddrreq == DL_IPV6_TOKEN) {
8721 /*
8722 * bcopy to low-order bits of ill_token
8723 *
8724 * XXX Temporary hack - currently, all known tokens
8725 * are 64 bits, so I'll cheat for the moment.
8726 */
8727 bcopy(addr, &ill->ill_token.s6_addr32[2], paddrlen);
8728 ill->ill_token_length = paddrlen;
8729 break;
8730 } else if (paddrreq == DL_IPV6_LINK_LAYER_ADDR) {
8731 ASSERT(ill->ill_nd_lla_mp == NULL);
8732 ill_set_ndmp(ill, mp, paddroff, paddrlen);
8733 mp = NULL;
8734 break;
8735 } else if (paddrreq == DL_CURR_DEST_ADDR) {
8736 ASSERT(ill->ill_dest_addr_mp == NULL);
8737 ill->ill_dest_addr_mp = mp;
8738 ill->ill_dest_addr = addr;
8739 mp = NULL;
8740 if (ill->ill_isv6) {
8741 ill_setdesttoken(ill);
8742 ipif_setdestlinklocal(ill->ill_ipif);
8743 }
8744 break;
8745 }
8746
8747 ASSERT(paddrreq == DL_CURR_PHYS_ADDR);
8748 ASSERT(ill->ill_phys_addr_mp == NULL);
8749 if (!ill->ill_ifname_pending)
8750 break;
8751 ill->ill_ifname_pending = 0;
8752 if (!ioctl_aborted)
8753 mp1 = ipsq_pending_mp_get(ipsq, &connp);
8754 if (mp1 != NULL) {
8755 ASSERT(connp == NULL);
8756 q = ill->ill_wq;
8757 }
8758 /*
8759 * If any error acks received during the plumbing sequence,
8760 * ill_ifname_pending_err will be set. Break out and send up
8761 * the error to the pending ioctl.
8762 */
8763 if (ill->ill_ifname_pending_err != 0) {
8764 err = ill->ill_ifname_pending_err;
8765 ill->ill_ifname_pending_err = 0;
8766 break;
8767 }
8768
8769 ill->ill_phys_addr_mp = mp;
8770 ill->ill_phys_addr = (paddrlen == 0 ? NULL : addr);
8771 mp = NULL;
8772
8773 /*
8774 * If paddrlen or ill_phys_addr_length is zero, the DLPI
8775 * provider doesn't support physical addresses. We check both
8776 * paddrlen and ill_phys_addr_length because sppp (PPP) does
8777 * not have physical addresses, but historically adversises a
8778 * physical address length of 0 in its DL_INFO_ACK, but 6 in
8779 * its DL_PHYS_ADDR_ACK.
8780 */
8781 if (paddrlen == 0 || ill->ill_phys_addr_length == 0) {
8782 ill->ill_phys_addr = NULL;
8783 } else if (paddrlen != ill->ill_phys_addr_length) {
8784 ip0dbg(("DL_PHYS_ADDR_ACK: got addrlen %d, expected %d",
8785 paddrlen, ill->ill_phys_addr_length));
8786 err = EINVAL;
8787 break;
8788 }
8789
8790 if (ill->ill_nd_lla_mp == NULL) {
8791 if ((mp_hw = copyb(ill->ill_phys_addr_mp)) == NULL) {
8792 err = ENOMEM;
8793 break;
8794 }
8795 ill_set_ndmp(ill, mp_hw, paddroff, paddrlen);
8796 }
8797
8798 if (ill->ill_isv6) {
8799 ill_setdefaulttoken(ill);
8800 ipif_setlinklocal(ill->ill_ipif);
8801 }
8802 break;
8803 }
8804 case DL_OK_ACK:
8805 ip2dbg(("DL_OK_ACK %s (0x%x)\n",
8806 dl_primstr((int)dloa->dl_correct_primitive),
8807 dloa->dl_correct_primitive));
8808 DTRACE_PROBE3(ill__dlpi, char *, "ip_rput_dlpi_writer ok",
8809 char *, dl_primstr(dloa->dl_correct_primitive),
8810 ill_t *, ill);
8811
8812 switch (dloa->dl_correct_primitive) {
8813 case DL_ENABMULTI_REQ:
8814 case DL_DISABMULTI_REQ:
8815 ill_dlpi_done(ill, dloa->dl_correct_primitive);
8816 break;
8817 case DL_PROMISCON_REQ:
8818 case DL_PROMISCOFF_REQ:
8819 case DL_UNBIND_REQ:
8820 case DL_ATTACH_REQ:
8821 ill_dlpi_done(ill, dloa->dl_correct_primitive);
8822 break;
8823 }
8824 break;
8825 default:
8826 break;
8827 }
8828
8829 freemsg(mp);
8830 if (mp1 == NULL)
8831 return;
8832
8833 /*
8834 * The operation must complete without EINPROGRESS since
8835 * ipsq_pending_mp_get() has removed the mblk (mp1). Otherwise,
8836 * the operation will be stuck forever inside the IPSQ.
8837 */
8838 ASSERT(err != EINPROGRESS);
8839
8840 DTRACE_PROBE4(ipif__ioctl, char *, "ip_rput_dlpi_writer finish",
8841 int, ipsq->ipsq_xop->ipx_current_ioctl, ill_t *, ill,
8842 ipif_t *, NULL);
8843
8844 switch (ipsq->ipsq_xop->ipx_current_ioctl) {
8845 case 0:
8846 ipsq_current_finish(ipsq);
8847 break;
8848
8849 case SIOCSLIFNAME:
8850 case IF_UNITSEL: {
8851 ill_t *ill_other = ILL_OTHER(ill);
8852
8853 /*
8854 * If SIOCSLIFNAME or IF_UNITSEL is about to succeed, and the
8855 * ill has a peer which is in an IPMP group, then place ill
8856 * into the same group. One catch: although ifconfig plumbs
8857 * the appropriate IPMP meta-interface prior to plumbing this
8858 * ill, it is possible for multiple ifconfig applications to
8859 * race (or for another application to adjust plumbing), in
8860 * which case the IPMP meta-interface we need will be missing.
8861 * If so, kick the phyint out of the group.
8862 */
8863 if (err == 0 && ill_other != NULL && IS_UNDER_IPMP(ill_other)) {
8864 ipmp_grp_t *grp = ill->ill_phyint->phyint_grp;
8865 ipmp_illgrp_t *illg;
8866
8867 illg = ill->ill_isv6 ? grp->gr_v6 : grp->gr_v4;
8868 if (illg == NULL)
8869 ipmp_phyint_leave_grp(ill->ill_phyint);
8870 else
8871 ipmp_ill_join_illgrp(ill, illg);
8872 }
8873
8874 if (ipsq->ipsq_xop->ipx_current_ioctl == IF_UNITSEL)
8875 ip_ioctl_finish(q, mp1, err, NO_COPYOUT, ipsq);
8876 else
8877 ip_ioctl_finish(q, mp1, err, COPYOUT, ipsq);
8878 break;
8879 }
8880 case SIOCLIFADDIF:
8881 ip_ioctl_finish(q, mp1, err, COPYOUT, ipsq);
8882 break;
8883
8884 default:
8885 ip_ioctl_finish(q, mp1, err, NO_COPYOUT, ipsq);
8886 break;
8887 }
8888 }
8889
8890 /*
8891 * ip_rput_other is called by ip_rput to handle messages modifying the global
8892 * state in IP. If 'ipsq' is non-NULL, caller is writer on it.
8893 */
8894 /* ARGSUSED */
8895 void
8896 ip_rput_other(ipsq_t *ipsq, queue_t *q, mblk_t *mp, void *dummy_arg)
8897 {
8898 ill_t *ill = q->q_ptr;
8899 struct iocblk *iocp;
8900
8901 ip1dbg(("ip_rput_other "));
8902 if (ipsq != NULL) {
8903 ASSERT(IAM_WRITER_IPSQ(ipsq));
8904 ASSERT(ipsq->ipsq_xop ==
8905 ill->ill_phyint->phyint_ipsq->ipsq_xop);
8906 }
8907
8908 switch (mp->b_datap->db_type) {
8909 case M_ERROR:
8910 case M_HANGUP:
8911 /*
8912 * The device has a problem. We force the ILL down. It can
8913 * be brought up again manually using SIOCSIFFLAGS (via
8914 * ifconfig or equivalent).
8915 */
8916 ASSERT(ipsq != NULL);
8917 if (mp->b_rptr < mp->b_wptr)
8918 ill->ill_error = (int)(*mp->b_rptr & 0xFF);
8919 if (ill->ill_error == 0)
8920 ill->ill_error = ENXIO;
8921 if (!ill_down_start(q, mp))
8922 return;
8923 ipif_all_down_tail(ipsq, q, mp, NULL);
8924 break;
8925 case M_IOCNAK: {
8926 iocp = (struct iocblk *)mp->b_rptr;
8927
8928 ASSERT(iocp->ioc_cmd == DL_IOC_HDR_INFO);
8929 /*
8930 * If this was the first attempt, turn off the fastpath
8931 * probing.
8932 */
8933 mutex_enter(&ill->ill_lock);
8934 if (ill->ill_dlpi_fastpath_state == IDS_INPROGRESS) {
8935 ill->ill_dlpi_fastpath_state = IDS_FAILED;
8936 mutex_exit(&ill->ill_lock);
8937 /*
8938 * don't flush the nce_t entries: we use them
8939 * as an index to the ncec itself.
8940 */
8941 ip1dbg(("ip_rput: DLPI fastpath off on interface %s\n",
8942 ill->ill_name));
8943 } else {
8944 mutex_exit(&ill->ill_lock);
8945 }
8946 freemsg(mp);
8947 break;
8948 }
8949 default:
8950 ASSERT(0);
8951 break;
8952 }
8953 }
8954
8955 /*
8956 * Update any source route, record route or timestamp options
8957 * When it fails it has consumed the message and BUMPed the MIB.
8958 */
8959 boolean_t
8960 ip_forward_options(mblk_t *mp, ipha_t *ipha, ill_t *dst_ill,
8961 ip_recv_attr_t *ira)
8962 {
8963 ipoptp_t opts;
8964 uchar_t *opt;
8965 uint8_t optval;
8966 uint8_t optlen;
8967 ipaddr_t dst;
8968 ipaddr_t ifaddr;
8969 uint32_t ts;
8970 timestruc_t now;
8971 ip_stack_t *ipst = ira->ira_ill->ill_ipst;
8972
8973 ip2dbg(("ip_forward_options\n"));
8974 dst = ipha->ipha_dst;
8975 for (optval = ipoptp_first(&opts, ipha);
8976 optval != IPOPT_EOL;
8977 optval = ipoptp_next(&opts)) {
8978 ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
8979 opt = opts.ipoptp_cur;
8980 optlen = opts.ipoptp_len;
8981 ip2dbg(("ip_forward_options: opt %d, len %d\n",
8982 optval, opts.ipoptp_len));
8983 switch (optval) {
8984 uint32_t off;
8985 case IPOPT_SSRR:
8986 case IPOPT_LSRR:
8987 /* Check if adminstratively disabled */
8988 if (!ipst->ips_ip_forward_src_routed) {
8989 BUMP_MIB(dst_ill->ill_ip_mib,
8990 ipIfStatsForwProhibits);
8991 ip_drop_input("ICMP_SOURCE_ROUTE_FAILED",
8992 mp, dst_ill);
8993 icmp_unreachable(mp, ICMP_SOURCE_ROUTE_FAILED,
8994 ira);
8995 return (B_FALSE);
8996 }
8997 if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
8998 /*
8999 * Must be partial since ip_input_options
9000 * checked for strict.
9001 */
9002 break;
9003 }
9004 off = opt[IPOPT_OFFSET];
9005 off--;
9006 redo_srr:
9007 if (optlen < IP_ADDR_LEN ||
9008 off > optlen - IP_ADDR_LEN) {
9009 /* End of source route */
9010 ip1dbg((
9011 "ip_forward_options: end of SR\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((char *)opt + off, &dst, IP_ADDR_LEN);
9023 bcopy(&ifaddr, (char *)opt + off, IP_ADDR_LEN);
9024 ip1dbg(("ip_forward_options: next hop 0x%x\n",
9025 ntohl(dst)));
9026
9027 /*
9028 * Check if our address is present more than
9029 * once as consecutive hops in source route.
9030 */
9031 if (ip_type_v4(dst, ipst) == IRE_LOCAL) {
9032 off += IP_ADDR_LEN;
9033 opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9034 goto redo_srr;
9035 }
9036 ipha->ipha_dst = dst;
9037 opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9038 break;
9039 case IPOPT_RR:
9040 off = opt[IPOPT_OFFSET];
9041 off--;
9042 if (optlen < IP_ADDR_LEN ||
9043 off > optlen - IP_ADDR_LEN) {
9044 /* No more room - ignore */
9045 ip1dbg((
9046 "ip_forward_options: end of RR\n"));
9047 break;
9048 }
9049 /* Pick a reasonable address on the outbound if */
9050 ASSERT(dst_ill != NULL);
9051 if (ip_select_source_v4(dst_ill, INADDR_ANY, dst,
9052 INADDR_ANY, ALL_ZONES, ipst, &ifaddr, NULL,
9053 NULL) != 0) {
9054 /* No source! Shouldn't happen */
9055 ifaddr = INADDR_ANY;
9056 }
9057 bcopy(&ifaddr, (char *)opt + off, IP_ADDR_LEN);
9058 opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9059 break;
9060 case IPOPT_TS:
9061 /* Insert timestamp if there is room */
9062 switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
9063 case IPOPT_TS_TSONLY:
9064 off = IPOPT_TS_TIMELEN;
9065 break;
9066 case IPOPT_TS_PRESPEC:
9067 case IPOPT_TS_PRESPEC_RFC791:
9068 /* Verify that the address matched */
9069 off = opt[IPOPT_OFFSET] - 1;
9070 bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
9071 if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
9072 /* Not for us */
9073 break;
9074 }
9075 /* FALLTHRU */
9076 case IPOPT_TS_TSANDADDR:
9077 off = IP_ADDR_LEN + IPOPT_TS_TIMELEN;
9078 break;
9079 default:
9080 /*
9081 * ip_*put_options should have already
9082 * dropped this packet.
9083 */
9084 cmn_err(CE_PANIC, "ip_forward_options: "
9085 "unknown IT - bug in ip_input_options?\n");
9086 return (B_TRUE); /* Keep "lint" happy */
9087 }
9088 if (opt[IPOPT_OFFSET] - 1 + off > optlen) {
9089 /* Increase overflow counter */
9090 off = (opt[IPOPT_POS_OV_FLG] >> 4) + 1;
9091 opt[IPOPT_POS_OV_FLG] =
9092 (uint8_t)((opt[IPOPT_POS_OV_FLG] & 0x0F) |
9093 (off << 4));
9094 break;
9095 }
9096 off = opt[IPOPT_OFFSET] - 1;
9097 switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
9098 case IPOPT_TS_PRESPEC:
9099 case IPOPT_TS_PRESPEC_RFC791:
9100 case IPOPT_TS_TSANDADDR:
9101 /* Pick a reasonable addr on the outbound if */
9102 ASSERT(dst_ill != NULL);
9103 if (ip_select_source_v4(dst_ill, INADDR_ANY,
9104 dst, INADDR_ANY, ALL_ZONES, ipst, &ifaddr,
9105 NULL, NULL) != 0) {
9106 /* No source! Shouldn't happen */
9107 ifaddr = INADDR_ANY;
9108 }
9109 bcopy(&ifaddr, (char *)opt + off, IP_ADDR_LEN);
9110 opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9111 /* FALLTHRU */
9112 case IPOPT_TS_TSONLY:
9113 off = opt[IPOPT_OFFSET] - 1;
9114 /* Compute # of milliseconds since midnight */
9115 gethrestime(&now);
9116 ts = (now.tv_sec % (24 * 60 * 60)) * 1000 +
9117 now.tv_nsec / (NANOSEC / MILLISEC);
9118 bcopy(&ts, (char *)opt + off, IPOPT_TS_TIMELEN);
9119 opt[IPOPT_OFFSET] += IPOPT_TS_TIMELEN;
9120 break;
9121 }
9122 break;
9123 }
9124 }
9125 return (B_TRUE);
9126 }
9127
9128 /*
9129 * Call ill_frag_timeout to do garbage collection. ill_frag_timeout
9130 * returns 'true' if there are still fragments left on the queue, in
9131 * which case we restart the timer.
9132 */
9133 void
9134 ill_frag_timer(void *arg)
9135 {
9136 ill_t *ill = (ill_t *)arg;
9137 boolean_t frag_pending;
9138 ip_stack_t *ipst = ill->ill_ipst;
9139 time_t timeout;
9140
9141 mutex_enter(&ill->ill_lock);
9142 ASSERT(!ill->ill_fragtimer_executing);
9143 if (ill->ill_state_flags & ILL_CONDEMNED) {
9144 ill->ill_frag_timer_id = 0;
9145 mutex_exit(&ill->ill_lock);
9146 return;
9147 }
9148 ill->ill_fragtimer_executing = 1;
9149 mutex_exit(&ill->ill_lock);
9150
9151 timeout = (ill->ill_isv6 ? ipst->ips_ipv6_reassembly_timeout :
9152 ipst->ips_ip_reassembly_timeout);
9153
9154 frag_pending = ill_frag_timeout(ill, timeout);
9155
9156 /*
9157 * Restart the timer, if we have fragments pending or if someone
9158 * wanted us to be scheduled again.
9159 */
9160 mutex_enter(&ill->ill_lock);
9161 ill->ill_fragtimer_executing = 0;
9162 ill->ill_frag_timer_id = 0;
9163 if (frag_pending || ill->ill_fragtimer_needrestart)
9164 ill_frag_timer_start(ill);
9165 mutex_exit(&ill->ill_lock);
9166 }
9167
9168 void
9169 ill_frag_timer_start(ill_t *ill)
9170 {
9171 ip_stack_t *ipst = ill->ill_ipst;
9172 clock_t timeo_ms;
9173
9174 ASSERT(MUTEX_HELD(&ill->ill_lock));
9175
9176 /* If the ill is closing or opening don't proceed */
9177 if (ill->ill_state_flags & ILL_CONDEMNED)
9178 return;
9179
9180 if (ill->ill_fragtimer_executing) {
9181 /*
9182 * ill_frag_timer is currently executing. Just record the
9183 * the fact that we want the timer to be restarted.
9184 * ill_frag_timer will post a timeout before it returns,
9185 * ensuring it will be called again.
9186 */
9187 ill->ill_fragtimer_needrestart = 1;
9188 return;
9189 }
9190
9191 if (ill->ill_frag_timer_id == 0) {
9192 timeo_ms = (ill->ill_isv6 ? ipst->ips_ipv6_reassembly_timeout :
9193 ipst->ips_ip_reassembly_timeout) * SECONDS;
9194
9195 /*
9196 * The timer is neither running nor is the timeout handler
9197 * executing. Post a timeout so that ill_frag_timer will be
9198 * called
9199 */
9200 ill->ill_frag_timer_id = timeout(ill_frag_timer, ill,
9201 MSEC_TO_TICK(timeo_ms >> 1));
9202 ill->ill_fragtimer_needrestart = 0;
9203 }
9204 }
9205
9206 /*
9207 * Update any source route, record route or timestamp options.
9208 * Check that we are at end of strict source route.
9209 * The options have already been checked for sanity in ip_input_options().
9210 */
9211 boolean_t
9212 ip_input_local_options(mblk_t *mp, ipha_t *ipha, ip_recv_attr_t *ira)
9213 {
9214 ipoptp_t opts;
9215 uchar_t *opt;
9216 uint8_t optval;
9217 uint8_t optlen;
9218 ipaddr_t dst;
9219 ipaddr_t ifaddr;
9220 uint32_t ts;
9221 timestruc_t now;
9222 ill_t *ill = ira->ira_ill;
9223 ip_stack_t *ipst = ill->ill_ipst;
9224
9225 ip2dbg(("ip_input_local_options\n"));
9226
9227 for (optval = ipoptp_first(&opts, ipha);
9228 optval != IPOPT_EOL;
9229 optval = ipoptp_next(&opts)) {
9230 ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
9231 opt = opts.ipoptp_cur;
9232 optlen = opts.ipoptp_len;
9233 ip2dbg(("ip_input_local_options: opt %d, len %d\n",
9234 optval, optlen));
9235 switch (optval) {
9236 uint32_t off;
9237 case IPOPT_SSRR:
9238 case IPOPT_LSRR:
9239 off = opt[IPOPT_OFFSET];
9240 off--;
9241 if (optlen < IP_ADDR_LEN ||
9242 off > optlen - IP_ADDR_LEN) {
9243 /* End of source route */
9244 ip1dbg(("ip_input_local_options: end of SR\n"));
9245 break;
9246 }
9247 /*
9248 * This will only happen if two consecutive entries
9249 * in the source route contains our address or if
9250 * it is a packet with a loose source route which
9251 * reaches us before consuming the whole source route
9252 */
9253 ip1dbg(("ip_input_local_options: not end of SR\n"));
9254 if (optval == IPOPT_SSRR) {
9255 goto bad_src_route;
9256 }
9257 /*
9258 * Hack: instead of dropping the packet truncate the
9259 * source route to what has been used by filling the
9260 * rest with IPOPT_NOP.
9261 */
9262 opt[IPOPT_OLEN] = (uint8_t)off;
9263 while (off < optlen) {
9264 opt[off++] = IPOPT_NOP;
9265 }
9266 break;
9267 case IPOPT_RR:
9268 off = opt[IPOPT_OFFSET];
9269 off--;
9270 if (optlen < IP_ADDR_LEN ||
9271 off > optlen - IP_ADDR_LEN) {
9272 /* No more room - ignore */
9273 ip1dbg((
9274 "ip_input_local_options: end of RR\n"));
9275 break;
9276 }
9277 /* Pick a reasonable address on the outbound if */
9278 if (ip_select_source_v4(ill, INADDR_ANY, ipha->ipha_dst,
9279 INADDR_ANY, ALL_ZONES, ipst, &ifaddr, NULL,
9280 NULL) != 0) {
9281 /* No source! Shouldn't happen */
9282 ifaddr = INADDR_ANY;
9283 }
9284 bcopy(&ifaddr, (char *)opt + off, IP_ADDR_LEN);
9285 opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9286 break;
9287 case IPOPT_TS:
9288 /* Insert timestamp if there is romm */
9289 switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
9290 case IPOPT_TS_TSONLY:
9291 off = IPOPT_TS_TIMELEN;
9292 break;
9293 case IPOPT_TS_PRESPEC:
9294 case IPOPT_TS_PRESPEC_RFC791:
9295 /* Verify that the address matched */
9296 off = opt[IPOPT_OFFSET] - 1;
9297 bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
9298 if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
9299 /* Not for us */
9300 break;
9301 }
9302 /* FALLTHRU */
9303 case IPOPT_TS_TSANDADDR:
9304 off = IP_ADDR_LEN + IPOPT_TS_TIMELEN;
9305 break;
9306 default:
9307 /*
9308 * ip_*put_options should have already
9309 * dropped this packet.
9310 */
9311 cmn_err(CE_PANIC, "ip_input_local_options: "
9312 "unknown IT - bug in ip_input_options?\n");
9313 return (B_TRUE); /* Keep "lint" happy */
9314 }
9315 if (opt[IPOPT_OFFSET] - 1 + off > optlen) {
9316 /* Increase overflow counter */
9317 off = (opt[IPOPT_POS_OV_FLG] >> 4) + 1;
9318 opt[IPOPT_POS_OV_FLG] =
9319 (uint8_t)((opt[IPOPT_POS_OV_FLG] & 0x0F) |
9320 (off << 4));
9321 break;
9322 }
9323 off = opt[IPOPT_OFFSET] - 1;
9324 switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
9325 case IPOPT_TS_PRESPEC:
9326 case IPOPT_TS_PRESPEC_RFC791:
9327 case IPOPT_TS_TSANDADDR:
9328 /* Pick a reasonable addr on the outbound if */
9329 if (ip_select_source_v4(ill, INADDR_ANY,
9330 ipha->ipha_dst, INADDR_ANY, ALL_ZONES, ipst,
9331 &ifaddr, NULL, NULL) != 0) {
9332 /* No source! Shouldn't happen */
9333 ifaddr = INADDR_ANY;
9334 }
9335 bcopy(&ifaddr, (char *)opt + off, IP_ADDR_LEN);
9336 opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9337 /* FALLTHRU */
9338 case IPOPT_TS_TSONLY:
9339 off = opt[IPOPT_OFFSET] - 1;
9340 /* Compute # of milliseconds since midnight */
9341 gethrestime(&now);
9342 ts = (now.tv_sec % (24 * 60 * 60)) * 1000 +
9343 now.tv_nsec / (NANOSEC / MILLISEC);
9344 bcopy(&ts, (char *)opt + off, IPOPT_TS_TIMELEN);
9345 opt[IPOPT_OFFSET] += IPOPT_TS_TIMELEN;
9346 break;
9347 }
9348 break;
9349 }
9350 }
9351 return (B_TRUE);
9352
9353 bad_src_route:
9354 /* make sure we clear any indication of a hardware checksum */
9355 DB_CKSUMFLAGS(mp) = 0;
9356 ip_drop_input("ICMP_SOURCE_ROUTE_FAILED", mp, ill);
9357 icmp_unreachable(mp, ICMP_SOURCE_ROUTE_FAILED, ira);
9358 return (B_FALSE);
9359
9360 }
9361
9362 /*
9363 * Process IP options in an inbound packet. Always returns the nexthop.
9364 * Normally this is the passed in nexthop, but if there is an option
9365 * that effects the nexthop (such as a source route) that will be returned.
9366 * Sets *errorp if there is an error, in which case an ICMP error has been sent
9367 * and mp freed.
9368 */
9369 ipaddr_t
9370 ip_input_options(ipha_t *ipha, ipaddr_t dst, mblk_t *mp,
9371 ip_recv_attr_t *ira, int *errorp)
9372 {
9373 ip_stack_t *ipst = ira->ira_ill->ill_ipst;
9374 ipoptp_t opts;
9375 uchar_t *opt;
9376 uint8_t optval;
9377 uint8_t optlen;
9378 intptr_t code = 0;
9379 ire_t *ire;
9380
9381 ip2dbg(("ip_input_options\n"));
9382 *errorp = 0;
9383 for (optval = ipoptp_first(&opts, ipha);
9384 optval != IPOPT_EOL;
9385 optval = ipoptp_next(&opts)) {
9386 opt = opts.ipoptp_cur;
9387 optlen = opts.ipoptp_len;
9388 ip2dbg(("ip_input_options: opt %d, len %d\n",
9389 optval, optlen));
9390 /*
9391 * Note: we need to verify the checksum before we
9392 * modify anything thus this routine only extracts the next
9393 * hop dst from any source route.
9394 */
9395 switch (optval) {
9396 uint32_t off;
9397 case IPOPT_SSRR:
9398 case IPOPT_LSRR:
9399 if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
9400 if (optval == IPOPT_SSRR) {
9401 ip1dbg(("ip_input_options: not next"
9402 " strict source route 0x%x\n",
9403 ntohl(dst)));
9404 code = (char *)&ipha->ipha_dst -
9405 (char *)ipha;
9406 goto param_prob; /* RouterReq's */
9407 }
9408 ip2dbg(("ip_input_options: "
9409 "not next source route 0x%x\n",
9410 ntohl(dst)));
9411 break;
9412 }
9413
9414 if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
9415 ip1dbg((
9416 "ip_input_options: bad option offset\n"));
9417 code = (char *)&opt[IPOPT_OLEN] -
9418 (char *)ipha;
9419 goto param_prob;
9420 }
9421 off = opt[IPOPT_OFFSET];
9422 off--;
9423 redo_srr:
9424 if (optlen < IP_ADDR_LEN ||
9425 off > optlen - IP_ADDR_LEN) {
9426 /* End of source route */
9427 ip1dbg(("ip_input_options: end of SR\n"));
9428 break;
9429 }
9430 bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
9431 ip1dbg(("ip_input_options: next hop 0x%x\n",
9432 ntohl(dst)));
9433
9434 /*
9435 * Check if our address is present more than
9436 * once as consecutive hops in source route.
9437 * XXX verify per-interface ip_forwarding
9438 * for source route?
9439 */
9440 if (ip_type_v4(dst, ipst) == IRE_LOCAL) {
9441 off += IP_ADDR_LEN;
9442 goto redo_srr;
9443 }
9444
9445 if (dst == htonl(INADDR_LOOPBACK)) {
9446 ip1dbg(("ip_input_options: loopback addr in "
9447 "source route!\n"));
9448 goto bad_src_route;
9449 }
9450 /*
9451 * For strict: verify that dst is directly
9452 * reachable.
9453 */
9454 if (optval == IPOPT_SSRR) {
9455 ire = ire_ftable_lookup_v4(dst, 0, 0,
9456 IRE_INTERFACE, NULL, ALL_ZONES,
9457 ira->ira_tsl,
9458 MATCH_IRE_TYPE | MATCH_IRE_SECATTR, 0, ipst,
9459 NULL);
9460 if (ire == NULL) {
9461 ip1dbg(("ip_input_options: SSRR not "
9462 "directly reachable: 0x%x\n",
9463 ntohl(dst)));
9464 goto bad_src_route;
9465 }
9466 ire_refrele(ire);
9467 }
9468 /*
9469 * Defer update of the offset and the record route
9470 * until the packet is forwarded.
9471 */
9472 break;
9473 case IPOPT_RR:
9474 if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
9475 ip1dbg((
9476 "ip_input_options: bad option offset\n"));
9477 code = (char *)&opt[IPOPT_OLEN] -
9478 (char *)ipha;
9479 goto param_prob;
9480 }
9481 break;
9482 case IPOPT_TS:
9483 /*
9484 * Verify that length >= 5 and that there is either
9485 * room for another timestamp or that the overflow
9486 * counter is not maxed out.
9487 */
9488 code = (char *)&opt[IPOPT_OLEN] - (char *)ipha;
9489 if (optlen < IPOPT_MINLEN_IT) {
9490 goto param_prob;
9491 }
9492 if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
9493 ip1dbg((
9494 "ip_input_options: bad option offset\n"));
9495 code = (char *)&opt[IPOPT_OFFSET] -
9496 (char *)ipha;
9497 goto param_prob;
9498 }
9499 switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
9500 case IPOPT_TS_TSONLY:
9501 off = IPOPT_TS_TIMELEN;
9502 break;
9503 case IPOPT_TS_TSANDADDR:
9504 case IPOPT_TS_PRESPEC:
9505 case IPOPT_TS_PRESPEC_RFC791:
9506 off = IP_ADDR_LEN + IPOPT_TS_TIMELEN;
9507 break;
9508 default:
9509 code = (char *)&opt[IPOPT_POS_OV_FLG] -
9510 (char *)ipha;
9511 goto param_prob;
9512 }
9513 if (opt[IPOPT_OFFSET] - 1 + off > optlen &&
9514 (opt[IPOPT_POS_OV_FLG] & 0xF0) == 0xF0) {
9515 /*
9516 * No room and the overflow counter is 15
9517 * already.
9518 */
9519 goto param_prob;
9520 }
9521 break;
9522 }
9523 }
9524
9525 if ((opts.ipoptp_flags & IPOPTP_ERROR) == 0) {
9526 return (dst);
9527 }
9528
9529 ip1dbg(("ip_input_options: error processing IP options."));
9530 code = (char *)&opt[IPOPT_OFFSET] - (char *)ipha;
9531
9532 param_prob:
9533 /* make sure we clear any indication of a hardware checksum */
9534 DB_CKSUMFLAGS(mp) = 0;
9535 ip_drop_input("ICMP_PARAM_PROBLEM", mp, ira->ira_ill);
9536 icmp_param_problem(mp, (uint8_t)code, ira);
9537 *errorp = -1;
9538 return (dst);
9539
9540 bad_src_route:
9541 /* make sure we clear any indication of a hardware checksum */
9542 DB_CKSUMFLAGS(mp) = 0;
9543 ip_drop_input("ICMP_SOURCE_ROUTE_FAILED", mp, ira->ira_ill);
9544 icmp_unreachable(mp, ICMP_SOURCE_ROUTE_FAILED, ira);
9545 *errorp = -1;
9546 return (dst);
9547 }
9548
9549 /*
9550 * IP & ICMP info in >=14 msg's ...
9551 * - ip fixed part (mib2_ip_t)
9552 * - icmp fixed part (mib2_icmp_t)
9553 * - ipAddrEntryTable (ip 20) all IPv4 ipifs
9554 * - ipRouteEntryTable (ip 21) all IPv4 IREs
9555 * - ipNetToMediaEntryTable (ip 22) all IPv4 Neighbor Cache entries
9556 * - ipRouteAttributeTable (ip 102) labeled routes
9557 * - ip multicast membership (ip_member_t)
9558 * - ip multicast source filtering (ip_grpsrc_t)
9559 * - igmp fixed part (struct igmpstat)
9560 * - multicast routing stats (struct mrtstat)
9561 * - multicast routing vifs (array of struct vifctl)
9562 * - multicast routing routes (array of struct mfcctl)
9563 * - ip6 fixed part (mib2_ipv6IfStatsEntry_t)
9564 * One per ill plus one generic
9565 * - icmp6 fixed part (mib2_ipv6IfIcmpEntry_t)
9566 * One per ill plus one generic
9567 * - ipv6RouteEntry all IPv6 IREs
9568 * - ipv6RouteAttributeTable (ip6 102) labeled routes
9569 * - ipv6NetToMediaEntry all IPv6 Neighbor Cache entries
9570 * - ipv6AddrEntry all IPv6 ipifs
9571 * - ipv6 multicast membership (ipv6_member_t)
9572 * - ipv6 multicast source filtering (ipv6_grpsrc_t)
9573 *
9574 * NOTE: original mpctl is copied for msg's 2..N, since its ctl part is
9575 * already filled in by the caller.
9576 * If legacy_req is true then MIB structures needs to be truncated to their
9577 * legacy sizes before being returned.
9578 * Return value of 0 indicates that no messages were sent and caller
9579 * should free mpctl.
9580 */
9581 int
9582 ip_snmp_get(queue_t *q, mblk_t *mpctl, int level, boolean_t legacy_req)
9583 {
9584 ip_stack_t *ipst;
9585 sctp_stack_t *sctps;
9586
9587 if (q->q_next != NULL) {
9588 ipst = ILLQ_TO_IPST(q);
9589 } else {
9590 ipst = CONNQ_TO_IPST(q);
9591 }
9592 ASSERT(ipst != NULL);
9593 sctps = ipst->ips_netstack->netstack_sctp;
9594
9595 if (mpctl == NULL || mpctl->b_cont == NULL) {
9596 return (0);
9597 }
9598
9599 /*
9600 * For the purposes of the (broken) packet shell use
9601 * of the level we make sure MIB2_TCP/MIB2_UDP can be used
9602 * to make TCP and UDP appear first in the list of mib items.
9603 * TBD: We could expand this and use it in netstat so that
9604 * the kernel doesn't have to produce large tables (connections,
9605 * routes, etc) when netstat only wants the statistics or a particular
9606 * table.
9607 */
9608 if (!(level == MIB2_TCP || level == MIB2_UDP)) {
9609 if ((mpctl = icmp_snmp_get(q, mpctl)) == NULL) {
9610 return (1);
9611 }
9612 }
9613
9614 if (level != MIB2_TCP) {
9615 if ((mpctl = udp_snmp_get(q, mpctl, legacy_req)) == NULL) {
9616 return (1);
9617 }
9618 }
9619
9620 if (level != MIB2_UDP) {
9621 if ((mpctl = tcp_snmp_get(q, mpctl, legacy_req)) == NULL) {
9622 return (1);
9623 }
9624 }
9625
9626 if ((mpctl = ip_snmp_get_mib2_ip_traffic_stats(q, mpctl,
9627 ipst, legacy_req)) == NULL) {
9628 return (1);
9629 }
9630
9631 if ((mpctl = ip_snmp_get_mib2_ip6(q, mpctl, ipst,
9632 legacy_req)) == NULL) {
9633 return (1);
9634 }
9635
9636 if ((mpctl = ip_snmp_get_mib2_icmp(q, mpctl, ipst)) == NULL) {
9637 return (1);
9638 }
9639
9640 if ((mpctl = ip_snmp_get_mib2_icmp6(q, mpctl, ipst)) == NULL) {
9641 return (1);
9642 }
9643
9644 if ((mpctl = ip_snmp_get_mib2_igmp(q, mpctl, ipst)) == NULL) {
9645 return (1);
9646 }
9647
9648 if ((mpctl = ip_snmp_get_mib2_multi(q, mpctl, ipst)) == NULL) {
9649 return (1);
9650 }
9651
9652 if ((mpctl = ip_snmp_get_mib2_ip_addr(q, mpctl, ipst,
9653 legacy_req)) == NULL) {
9654 return (1);
9655 }
9656
9657 if ((mpctl = ip_snmp_get_mib2_ip6_addr(q, mpctl, ipst,
9658 legacy_req)) == NULL) {
9659 return (1);
9660 }
9661
9662 if ((mpctl = ip_snmp_get_mib2_ip_group_mem(q, mpctl, ipst)) == NULL) {
9663 return (1);
9664 }
9665
9666 if ((mpctl = ip_snmp_get_mib2_ip6_group_mem(q, mpctl, ipst)) == NULL) {
9667 return (1);
9668 }
9669
9670 if ((mpctl = ip_snmp_get_mib2_ip_group_src(q, mpctl, ipst)) == NULL) {
9671 return (1);
9672 }
9673
9674 if ((mpctl = ip_snmp_get_mib2_ip6_group_src(q, mpctl, ipst)) == NULL) {
9675 return (1);
9676 }
9677
9678 if ((mpctl = ip_snmp_get_mib2_virt_multi(q, mpctl, ipst)) == NULL) {
9679 return (1);
9680 }
9681
9682 if ((mpctl = ip_snmp_get_mib2_multi_rtable(q, mpctl, ipst)) == NULL) {
9683 return (1);
9684 }
9685
9686 mpctl = ip_snmp_get_mib2_ip_route_media(q, mpctl, level, ipst);
9687 if (mpctl == NULL)
9688 return (1);
9689
9690 mpctl = ip_snmp_get_mib2_ip6_route_media(q, mpctl, level, ipst);
9691 if (mpctl == NULL)
9692 return (1);
9693
9694 if ((mpctl = sctp_snmp_get_mib2(q, mpctl, sctps)) == NULL) {
9695 return (1);
9696 }
9697 if ((mpctl = ip_snmp_get_mib2_ip_dce(q, mpctl, ipst)) == NULL) {
9698 return (1);
9699 }
9700 freemsg(mpctl);
9701 return (1);
9702 }
9703
9704 /* Get global (legacy) IPv4 statistics */
9705 static mblk_t *
9706 ip_snmp_get_mib2_ip(queue_t *q, mblk_t *mpctl, mib2_ipIfStatsEntry_t *ipmib,
9707 ip_stack_t *ipst, boolean_t legacy_req)
9708 {
9709 mib2_ip_t old_ip_mib;
9710 struct opthdr *optp;
9711 mblk_t *mp2ctl;
9712 mib2_ipAddrEntry_t mae;
9713
9714 /*
9715 * make a copy of the original message
9716 */
9717 mp2ctl = copymsg(mpctl);
9718
9719 /* fixed length IP structure... */
9720 optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
9721 optp->level = MIB2_IP;
9722 optp->name = 0;
9723 SET_MIB(old_ip_mib.ipForwarding,
9724 (WE_ARE_FORWARDING(ipst) ? 1 : 2));
9725 SET_MIB(old_ip_mib.ipDefaultTTL,
9726 (uint32_t)ipst->ips_ip_def_ttl);
9727 SET_MIB(old_ip_mib.ipReasmTimeout,
9728 ipst->ips_ip_reassembly_timeout);
9729 SET_MIB(old_ip_mib.ipAddrEntrySize,
9730 (legacy_req) ? LEGACY_MIB_SIZE(&mae, mib2_ipAddrEntry_t) :
9731 sizeof (mib2_ipAddrEntry_t));
9732 SET_MIB(old_ip_mib.ipRouteEntrySize,
9733 sizeof (mib2_ipRouteEntry_t));
9734 SET_MIB(old_ip_mib.ipNetToMediaEntrySize,
9735 sizeof (mib2_ipNetToMediaEntry_t));
9736 SET_MIB(old_ip_mib.ipMemberEntrySize, sizeof (ip_member_t));
9737 SET_MIB(old_ip_mib.ipGroupSourceEntrySize, sizeof (ip_grpsrc_t));
9738 SET_MIB(old_ip_mib.ipRouteAttributeSize,
9739 sizeof (mib2_ipAttributeEntry_t));
9740 SET_MIB(old_ip_mib.transportMLPSize, sizeof (mib2_transportMLPEntry_t));
9741 SET_MIB(old_ip_mib.ipDestEntrySize, sizeof (dest_cache_entry_t));
9742
9743 /*
9744 * Grab the statistics from the new IP MIB
9745 */
9746 SET_MIB(old_ip_mib.ipInReceives,
9747 (uint32_t)ipmib->ipIfStatsHCInReceives);
9748 SET_MIB(old_ip_mib.ipInHdrErrors, ipmib->ipIfStatsInHdrErrors);
9749 SET_MIB(old_ip_mib.ipInAddrErrors, ipmib->ipIfStatsInAddrErrors);
9750 SET_MIB(old_ip_mib.ipForwDatagrams,
9751 (uint32_t)ipmib->ipIfStatsHCOutForwDatagrams);
9752 SET_MIB(old_ip_mib.ipInUnknownProtos,
9753 ipmib->ipIfStatsInUnknownProtos);
9754 SET_MIB(old_ip_mib.ipInDiscards, ipmib->ipIfStatsInDiscards);
9755 SET_MIB(old_ip_mib.ipInDelivers,
9756 (uint32_t)ipmib->ipIfStatsHCInDelivers);
9757 SET_MIB(old_ip_mib.ipOutRequests,
9758 (uint32_t)ipmib->ipIfStatsHCOutRequests);
9759 SET_MIB(old_ip_mib.ipOutDiscards, ipmib->ipIfStatsOutDiscards);
9760 SET_MIB(old_ip_mib.ipOutNoRoutes, ipmib->ipIfStatsOutNoRoutes);
9761 SET_MIB(old_ip_mib.ipReasmReqds, ipmib->ipIfStatsReasmReqds);
9762 SET_MIB(old_ip_mib.ipReasmOKs, ipmib->ipIfStatsReasmOKs);
9763 SET_MIB(old_ip_mib.ipReasmFails, ipmib->ipIfStatsReasmFails);
9764 SET_MIB(old_ip_mib.ipFragOKs, ipmib->ipIfStatsOutFragOKs);
9765 SET_MIB(old_ip_mib.ipFragFails, ipmib->ipIfStatsOutFragFails);
9766 SET_MIB(old_ip_mib.ipFragCreates, ipmib->ipIfStatsOutFragCreates);
9767
9768 /* ipRoutingDiscards is not being used */
9769 SET_MIB(old_ip_mib.ipRoutingDiscards, 0);
9770 SET_MIB(old_ip_mib.tcpInErrs, ipmib->tcpIfStatsInErrs);
9771 SET_MIB(old_ip_mib.udpNoPorts, ipmib->udpIfStatsNoPorts);
9772 SET_MIB(old_ip_mib.ipInCksumErrs, ipmib->ipIfStatsInCksumErrs);
9773 SET_MIB(old_ip_mib.ipReasmDuplicates,
9774 ipmib->ipIfStatsReasmDuplicates);
9775 SET_MIB(old_ip_mib.ipReasmPartDups, ipmib->ipIfStatsReasmPartDups);
9776 SET_MIB(old_ip_mib.ipForwProhibits, ipmib->ipIfStatsForwProhibits);
9777 SET_MIB(old_ip_mib.udpInCksumErrs, ipmib->udpIfStatsInCksumErrs);
9778 SET_MIB(old_ip_mib.udpInOverflows, ipmib->udpIfStatsInOverflows);
9779 SET_MIB(old_ip_mib.rawipInOverflows,
9780 ipmib->rawipIfStatsInOverflows);
9781
9782 SET_MIB(old_ip_mib.ipsecInSucceeded, ipmib->ipsecIfStatsInSucceeded);
9783 SET_MIB(old_ip_mib.ipsecInFailed, ipmib->ipsecIfStatsInFailed);
9784 SET_MIB(old_ip_mib.ipInIPv6, ipmib->ipIfStatsInWrongIPVersion);
9785 SET_MIB(old_ip_mib.ipOutIPv6, ipmib->ipIfStatsOutWrongIPVersion);
9786 SET_MIB(old_ip_mib.ipOutSwitchIPv6,
9787 ipmib->ipIfStatsOutSwitchIPVersion);
9788
9789 if (!snmp_append_data(mpctl->b_cont, (char *)&old_ip_mib,
9790 (int)sizeof (old_ip_mib))) {
9791 ip1dbg(("ip_snmp_get_mib2_ip: failed to allocate %u bytes\n",
9792 (uint_t)sizeof (old_ip_mib)));
9793 }
9794
9795 optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
9796 ip3dbg(("ip_snmp_get_mib2_ip: level %d, name %d, len %d\n",
9797 (int)optp->level, (int)optp->name, (int)optp->len));
9798 qreply(q, mpctl);
9799 return (mp2ctl);
9800 }
9801
9802 /* Per interface IPv4 statistics */
9803 static mblk_t *
9804 ip_snmp_get_mib2_ip_traffic_stats(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst,
9805 boolean_t legacy_req)
9806 {
9807 struct opthdr *optp;
9808 mblk_t *mp2ctl;
9809 ill_t *ill;
9810 ill_walk_context_t ctx;
9811 mblk_t *mp_tail = NULL;
9812 mib2_ipIfStatsEntry_t global_ip_mib;
9813 mib2_ipAddrEntry_t mae;
9814
9815 /*
9816 * Make a copy of the original message
9817 */
9818 mp2ctl = copymsg(mpctl);
9819
9820 optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
9821 optp->level = MIB2_IP;
9822 optp->name = MIB2_IP_TRAFFIC_STATS;
9823 /* Include "unknown interface" ip_mib */
9824 ipst->ips_ip_mib.ipIfStatsIPVersion = MIB2_INETADDRESSTYPE_ipv4;
9825 ipst->ips_ip_mib.ipIfStatsIfIndex =
9826 MIB2_UNKNOWN_INTERFACE; /* Flag to netstat */
9827 SET_MIB(ipst->ips_ip_mib.ipIfStatsForwarding,
9828 (ipst->ips_ip_forwarding ? 1 : 2));
9829 SET_MIB(ipst->ips_ip_mib.ipIfStatsDefaultTTL,
9830 (uint32_t)ipst->ips_ip_def_ttl);
9831 SET_MIB(ipst->ips_ip_mib.ipIfStatsEntrySize,
9832 sizeof (mib2_ipIfStatsEntry_t));
9833 SET_MIB(ipst->ips_ip_mib.ipIfStatsAddrEntrySize,
9834 sizeof (mib2_ipAddrEntry_t));
9835 SET_MIB(ipst->ips_ip_mib.ipIfStatsRouteEntrySize,
9836 sizeof (mib2_ipRouteEntry_t));
9837 SET_MIB(ipst->ips_ip_mib.ipIfStatsNetToMediaEntrySize,
9838 sizeof (mib2_ipNetToMediaEntry_t));
9839 SET_MIB(ipst->ips_ip_mib.ipIfStatsMemberEntrySize,
9840 sizeof (ip_member_t));
9841 SET_MIB(ipst->ips_ip_mib.ipIfStatsGroupSourceEntrySize,
9842 sizeof (ip_grpsrc_t));
9843
9844 bcopy(&ipst->ips_ip_mib, &global_ip_mib, sizeof (global_ip_mib));
9845
9846 if (legacy_req) {
9847 SET_MIB(global_ip_mib.ipIfStatsAddrEntrySize,
9848 LEGACY_MIB_SIZE(&mae, mib2_ipAddrEntry_t));
9849 }
9850
9851 if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
9852 (char *)&global_ip_mib, (int)sizeof (global_ip_mib))) {
9853 ip1dbg(("ip_snmp_get_mib2_ip_traffic_stats: "
9854 "failed to allocate %u bytes\n",
9855 (uint_t)sizeof (global_ip_mib)));
9856 }
9857
9858 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
9859 ill = ILL_START_WALK_V4(&ctx, ipst);
9860 for (; ill != NULL; ill = ill_next(&ctx, ill)) {
9861 ill->ill_ip_mib->ipIfStatsIfIndex =
9862 ill->ill_phyint->phyint_ifindex;
9863 SET_MIB(ill->ill_ip_mib->ipIfStatsForwarding,
9864 (ipst->ips_ip_forwarding ? 1 : 2));
9865 SET_MIB(ill->ill_ip_mib->ipIfStatsDefaultTTL,
9866 (uint32_t)ipst->ips_ip_def_ttl);
9867
9868 ip_mib2_add_ip_stats(&global_ip_mib, ill->ill_ip_mib);
9869 if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
9870 (char *)ill->ill_ip_mib,
9871 (int)sizeof (*ill->ill_ip_mib))) {
9872 ip1dbg(("ip_snmp_get_mib2_ip_traffic_stats: "
9873 "failed to allocate %u bytes\n",
9874 (uint_t)sizeof (*ill->ill_ip_mib)));
9875 }
9876 }
9877 rw_exit(&ipst->ips_ill_g_lock);
9878
9879 optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
9880 ip3dbg(("ip_snmp_get_mib2_ip_traffic_stats: "
9881 "level %d, name %d, len %d\n",
9882 (int)optp->level, (int)optp->name, (int)optp->len));
9883 qreply(q, mpctl);
9884
9885 if (mp2ctl == NULL)
9886 return (NULL);
9887
9888 return (ip_snmp_get_mib2_ip(q, mp2ctl, &global_ip_mib, ipst,
9889 legacy_req));
9890 }
9891
9892 /* Global IPv4 ICMP statistics */
9893 static mblk_t *
9894 ip_snmp_get_mib2_icmp(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
9895 {
9896 struct opthdr *optp;
9897 mblk_t *mp2ctl;
9898
9899 /*
9900 * Make a copy of the original message
9901 */
9902 mp2ctl = copymsg(mpctl);
9903
9904 optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
9905 optp->level = MIB2_ICMP;
9906 optp->name = 0;
9907 if (!snmp_append_data(mpctl->b_cont, (char *)&ipst->ips_icmp_mib,
9908 (int)sizeof (ipst->ips_icmp_mib))) {
9909 ip1dbg(("ip_snmp_get_mib2_icmp: failed to allocate %u bytes\n",
9910 (uint_t)sizeof (ipst->ips_icmp_mib)));
9911 }
9912 optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
9913 ip3dbg(("ip_snmp_get_mib2_icmp: level %d, name %d, len %d\n",
9914 (int)optp->level, (int)optp->name, (int)optp->len));
9915 qreply(q, mpctl);
9916 return (mp2ctl);
9917 }
9918
9919 /* Global IPv4 IGMP statistics */
9920 static mblk_t *
9921 ip_snmp_get_mib2_igmp(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
9922 {
9923 struct opthdr *optp;
9924 mblk_t *mp2ctl;
9925
9926 /*
9927 * make a copy of the original message
9928 */
9929 mp2ctl = copymsg(mpctl);
9930
9931 optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
9932 optp->level = EXPER_IGMP;
9933 optp->name = 0;
9934 if (!snmp_append_data(mpctl->b_cont, (char *)&ipst->ips_igmpstat,
9935 (int)sizeof (ipst->ips_igmpstat))) {
9936 ip1dbg(("ip_snmp_get_mib2_igmp: failed to allocate %u bytes\n",
9937 (uint_t)sizeof (ipst->ips_igmpstat)));
9938 }
9939 optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
9940 ip3dbg(("ip_snmp_get_mib2_igmp: level %d, name %d, len %d\n",
9941 (int)optp->level, (int)optp->name, (int)optp->len));
9942 qreply(q, mpctl);
9943 return (mp2ctl);
9944 }
9945
9946 /* Global IPv4 Multicast Routing statistics */
9947 static mblk_t *
9948 ip_snmp_get_mib2_multi(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
9949 {
9950 struct opthdr *optp;
9951 mblk_t *mp2ctl;
9952
9953 /*
9954 * make a copy of the original message
9955 */
9956 mp2ctl = copymsg(mpctl);
9957
9958 optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
9959 optp->level = EXPER_DVMRP;
9960 optp->name = 0;
9961 if (!ip_mroute_stats(mpctl->b_cont, ipst)) {
9962 ip0dbg(("ip_mroute_stats: failed\n"));
9963 }
9964 optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
9965 ip3dbg(("ip_snmp_get_mib2_multi: level %d, name %d, len %d\n",
9966 (int)optp->level, (int)optp->name, (int)optp->len));
9967 qreply(q, mpctl);
9968 return (mp2ctl);
9969 }
9970
9971 /* IPv4 address information */
9972 static mblk_t *
9973 ip_snmp_get_mib2_ip_addr(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst,
9974 boolean_t legacy_req)
9975 {
9976 struct opthdr *optp;
9977 mblk_t *mp2ctl;
9978 mblk_t *mp_tail = NULL;
9979 ill_t *ill;
9980 ipif_t *ipif;
9981 uint_t bitval;
9982 mib2_ipAddrEntry_t mae;
9983 size_t mae_size;
9984 zoneid_t zoneid;
9985 ill_walk_context_t ctx;
9986
9987 /*
9988 * make a copy of the original message
9989 */
9990 mp2ctl = copymsg(mpctl);
9991
9992 mae_size = (legacy_req) ? LEGACY_MIB_SIZE(&mae, mib2_ipAddrEntry_t) :
9993 sizeof (mib2_ipAddrEntry_t);
9994
9995 /* ipAddrEntryTable */
9996
9997 optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
9998 optp->level = MIB2_IP;
9999 optp->name = MIB2_IP_ADDR;
10000 zoneid = Q_TO_CONN(q)->conn_zoneid;
10001
10002 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10003 ill = ILL_START_WALK_V4(&ctx, ipst);
10004 for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10005 for (ipif = ill->ill_ipif; ipif != NULL;
10006 ipif = ipif->ipif_next) {
10007 if (ipif->ipif_zoneid != zoneid &&
10008 ipif->ipif_zoneid != ALL_ZONES)
10009 continue;
10010 /* Sum of count from dead IRE_LO* and our current */
10011 mae.ipAdEntInfo.ae_ibcnt = ipif->ipif_ib_pkt_count;
10012 if (ipif->ipif_ire_local != NULL) {
10013 mae.ipAdEntInfo.ae_ibcnt +=
10014 ipif->ipif_ire_local->ire_ib_pkt_count;
10015 }
10016 mae.ipAdEntInfo.ae_obcnt = 0;
10017 mae.ipAdEntInfo.ae_focnt = 0;
10018
10019 ipif_get_name(ipif, mae.ipAdEntIfIndex.o_bytes,
10020 OCTET_LENGTH);
10021 mae.ipAdEntIfIndex.o_length =
10022 mi_strlen(mae.ipAdEntIfIndex.o_bytes);
10023 mae.ipAdEntAddr = ipif->ipif_lcl_addr;
10024 mae.ipAdEntNetMask = ipif->ipif_net_mask;
10025 mae.ipAdEntInfo.ae_subnet = ipif->ipif_subnet;
10026 mae.ipAdEntInfo.ae_subnet_len =
10027 ip_mask_to_plen(ipif->ipif_net_mask);
10028 mae.ipAdEntInfo.ae_src_addr = ipif->ipif_lcl_addr;
10029 for (bitval = 1;
10030 bitval &&
10031 !(bitval & ipif->ipif_brd_addr);
10032 bitval <<= 1)
10033 noop;
10034 mae.ipAdEntBcastAddr = bitval;
10035 mae.ipAdEntReasmMaxSize = IP_MAXPACKET;
10036 mae.ipAdEntInfo.ae_mtu = ipif->ipif_ill->ill_mtu;
10037 mae.ipAdEntInfo.ae_metric = ipif->ipif_ill->ill_metric;
10038 mae.ipAdEntInfo.ae_broadcast_addr =
10039 ipif->ipif_brd_addr;
10040 mae.ipAdEntInfo.ae_pp_dst_addr =
10041 ipif->ipif_pp_dst_addr;
10042 mae.ipAdEntInfo.ae_flags = ipif->ipif_flags |
10043 ill->ill_flags | ill->ill_phyint->phyint_flags;
10044 mae.ipAdEntRetransmitTime =
10045 ill->ill_reachable_retrans_time;
10046
10047 if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10048 (char *)&mae, (int)mae_size)) {
10049 ip1dbg(("ip_snmp_get_mib2_ip_addr: failed to "
10050 "allocate %u bytes\n", (uint_t)mae_size));
10051 }
10052 }
10053 }
10054 rw_exit(&ipst->ips_ill_g_lock);
10055
10056 optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10057 ip3dbg(("ip_snmp_get_mib2_ip_addr: level %d, name %d, len %d\n",
10058 (int)optp->level, (int)optp->name, (int)optp->len));
10059 qreply(q, mpctl);
10060 return (mp2ctl);
10061 }
10062
10063 /* IPv6 address information */
10064 static mblk_t *
10065 ip_snmp_get_mib2_ip6_addr(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst,
10066 boolean_t legacy_req)
10067 {
10068 struct opthdr *optp;
10069 mblk_t *mp2ctl;
10070 mblk_t *mp_tail = NULL;
10071 ill_t *ill;
10072 ipif_t *ipif;
10073 mib2_ipv6AddrEntry_t mae6;
10074 size_t mae6_size;
10075 zoneid_t zoneid;
10076 ill_walk_context_t ctx;
10077
10078 /*
10079 * make a copy of the original message
10080 */
10081 mp2ctl = copymsg(mpctl);
10082
10083 mae6_size = (legacy_req) ?
10084 LEGACY_MIB_SIZE(&mae6, mib2_ipv6AddrEntry_t) :
10085 sizeof (mib2_ipv6AddrEntry_t);
10086
10087 /* ipv6AddrEntryTable */
10088
10089 optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10090 optp->level = MIB2_IP6;
10091 optp->name = MIB2_IP6_ADDR;
10092 zoneid = Q_TO_CONN(q)->conn_zoneid;
10093
10094 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10095 ill = ILL_START_WALK_V6(&ctx, ipst);
10096 for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10097 for (ipif = ill->ill_ipif; ipif != NULL;
10098 ipif = ipif->ipif_next) {
10099 if (ipif->ipif_zoneid != zoneid &&
10100 ipif->ipif_zoneid != ALL_ZONES)
10101 continue;
10102 /* Sum of count from dead IRE_LO* and our current */
10103 mae6.ipv6AddrInfo.ae_ibcnt = ipif->ipif_ib_pkt_count;
10104 if (ipif->ipif_ire_local != NULL) {
10105 mae6.ipv6AddrInfo.ae_ibcnt +=
10106 ipif->ipif_ire_local->ire_ib_pkt_count;
10107 }
10108 mae6.ipv6AddrInfo.ae_obcnt = 0;
10109 mae6.ipv6AddrInfo.ae_focnt = 0;
10110
10111 ipif_get_name(ipif, mae6.ipv6AddrIfIndex.o_bytes,
10112 OCTET_LENGTH);
10113 mae6.ipv6AddrIfIndex.o_length =
10114 mi_strlen(mae6.ipv6AddrIfIndex.o_bytes);
10115 mae6.ipv6AddrAddress = ipif->ipif_v6lcl_addr;
10116 mae6.ipv6AddrPfxLength =
10117 ip_mask_to_plen_v6(&ipif->ipif_v6net_mask);
10118 mae6.ipv6AddrInfo.ae_subnet = ipif->ipif_v6subnet;
10119 mae6.ipv6AddrInfo.ae_subnet_len =
10120 mae6.ipv6AddrPfxLength;
10121 mae6.ipv6AddrInfo.ae_src_addr = ipif->ipif_v6lcl_addr;
10122
10123 /* Type: stateless(1), stateful(2), unknown(3) */
10124 if (ipif->ipif_flags & IPIF_ADDRCONF)
10125 mae6.ipv6AddrType = 1;
10126 else
10127 mae6.ipv6AddrType = 2;
10128 /* Anycast: true(1), false(2) */
10129 if (ipif->ipif_flags & IPIF_ANYCAST)
10130 mae6.ipv6AddrAnycastFlag = 1;
10131 else
10132 mae6.ipv6AddrAnycastFlag = 2;
10133
10134 /*
10135 * Address status: preferred(1), deprecated(2),
10136 * invalid(3), inaccessible(4), unknown(5)
10137 */
10138 if (ipif->ipif_flags & IPIF_NOLOCAL)
10139 mae6.ipv6AddrStatus = 3;
10140 else if (ipif->ipif_flags & IPIF_DEPRECATED)
10141 mae6.ipv6AddrStatus = 2;
10142 else
10143 mae6.ipv6AddrStatus = 1;
10144 mae6.ipv6AddrInfo.ae_mtu = ipif->ipif_ill->ill_mtu;
10145 mae6.ipv6AddrInfo.ae_metric =
10146 ipif->ipif_ill->ill_metric;
10147 mae6.ipv6AddrInfo.ae_pp_dst_addr =
10148 ipif->ipif_v6pp_dst_addr;
10149 mae6.ipv6AddrInfo.ae_flags = ipif->ipif_flags |
10150 ill->ill_flags | ill->ill_phyint->phyint_flags;
10151 mae6.ipv6AddrReasmMaxSize = IP_MAXPACKET;
10152 mae6.ipv6AddrIdentifier = ill->ill_token;
10153 mae6.ipv6AddrIdentifierLen = ill->ill_token_length;
10154 mae6.ipv6AddrReachableTime = ill->ill_reachable_time;
10155 mae6.ipv6AddrRetransmitTime =
10156 ill->ill_reachable_retrans_time;
10157 if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10158 (char *)&mae6, (int)mae6_size)) {
10159 ip1dbg(("ip_snmp_get_mib2_ip6_addr: failed to "
10160 "allocate %u bytes\n",
10161 (uint_t)mae6_size));
10162 }
10163 }
10164 }
10165 rw_exit(&ipst->ips_ill_g_lock);
10166
10167 optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10168 ip3dbg(("ip_snmp_get_mib2_ip6_addr: level %d, name %d, len %d\n",
10169 (int)optp->level, (int)optp->name, (int)optp->len));
10170 qreply(q, mpctl);
10171 return (mp2ctl);
10172 }
10173
10174 /* IPv4 multicast group membership. */
10175 static mblk_t *
10176 ip_snmp_get_mib2_ip_group_mem(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10177 {
10178 struct opthdr *optp;
10179 mblk_t *mp2ctl;
10180 ill_t *ill;
10181 ipif_t *ipif;
10182 ilm_t *ilm;
10183 ip_member_t ipm;
10184 mblk_t *mp_tail = NULL;
10185 ill_walk_context_t ctx;
10186 zoneid_t zoneid;
10187
10188 /*
10189 * make a copy of the original message
10190 */
10191 mp2ctl = copymsg(mpctl);
10192 zoneid = Q_TO_CONN(q)->conn_zoneid;
10193
10194 /* ipGroupMember table */
10195 optp = (struct opthdr *)&mpctl->b_rptr[
10196 sizeof (struct T_optmgmt_ack)];
10197 optp->level = MIB2_IP;
10198 optp->name = EXPER_IP_GROUP_MEMBERSHIP;
10199
10200 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10201 ill = ILL_START_WALK_V4(&ctx, ipst);
10202 for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10203 /* Make sure the ill isn't going away. */
10204 if (!ill_check_and_refhold(ill))
10205 continue;
10206 rw_exit(&ipst->ips_ill_g_lock);
10207 rw_enter(&ill->ill_mcast_lock, RW_READER);
10208 for (ilm = ill->ill_ilm; ilm; ilm = ilm->ilm_next) {
10209 if (ilm->ilm_zoneid != zoneid &&
10210 ilm->ilm_zoneid != ALL_ZONES)
10211 continue;
10212
10213 /* Is there an ipif for ilm_ifaddr? */
10214 for (ipif = ill->ill_ipif; ipif != NULL;
10215 ipif = ipif->ipif_next) {
10216 if (!IPIF_IS_CONDEMNED(ipif) &&
10217 ipif->ipif_lcl_addr == ilm->ilm_ifaddr &&
10218 ilm->ilm_ifaddr != INADDR_ANY)
10219 break;
10220 }
10221 if (ipif != NULL) {
10222 ipif_get_name(ipif,
10223 ipm.ipGroupMemberIfIndex.o_bytes,
10224 OCTET_LENGTH);
10225 } else {
10226 ill_get_name(ill,
10227 ipm.ipGroupMemberIfIndex.o_bytes,
10228 OCTET_LENGTH);
10229 }
10230 ipm.ipGroupMemberIfIndex.o_length =
10231 mi_strlen(ipm.ipGroupMemberIfIndex.o_bytes);
10232
10233 ipm.ipGroupMemberAddress = ilm->ilm_addr;
10234 ipm.ipGroupMemberRefCnt = ilm->ilm_refcnt;
10235 ipm.ipGroupMemberFilterMode = ilm->ilm_fmode;
10236 if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10237 (char *)&ipm, (int)sizeof (ipm))) {
10238 ip1dbg(("ip_snmp_get_mib2_ip_group: "
10239 "failed to allocate %u bytes\n",
10240 (uint_t)sizeof (ipm)));
10241 }
10242 }
10243 rw_exit(&ill->ill_mcast_lock);
10244 ill_refrele(ill);
10245 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10246 }
10247 rw_exit(&ipst->ips_ill_g_lock);
10248 optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10249 ip3dbg(("ip_snmp_get: level %d, name %d, len %d\n",
10250 (int)optp->level, (int)optp->name, (int)optp->len));
10251 qreply(q, mpctl);
10252 return (mp2ctl);
10253 }
10254
10255 /* IPv6 multicast group membership. */
10256 static mblk_t *
10257 ip_snmp_get_mib2_ip6_group_mem(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10258 {
10259 struct opthdr *optp;
10260 mblk_t *mp2ctl;
10261 ill_t *ill;
10262 ilm_t *ilm;
10263 ipv6_member_t ipm6;
10264 mblk_t *mp_tail = NULL;
10265 ill_walk_context_t ctx;
10266 zoneid_t zoneid;
10267
10268 /*
10269 * make a copy of the original message
10270 */
10271 mp2ctl = copymsg(mpctl);
10272 zoneid = Q_TO_CONN(q)->conn_zoneid;
10273
10274 /* ip6GroupMember table */
10275 optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10276 optp->level = MIB2_IP6;
10277 optp->name = EXPER_IP6_GROUP_MEMBERSHIP;
10278
10279 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10280 ill = ILL_START_WALK_V6(&ctx, ipst);
10281 for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10282 /* Make sure the ill isn't going away. */
10283 if (!ill_check_and_refhold(ill))
10284 continue;
10285 rw_exit(&ipst->ips_ill_g_lock);
10286 /*
10287 * Normally we don't have any members on under IPMP interfaces.
10288 * We report them as a debugging aid.
10289 */
10290 rw_enter(&ill->ill_mcast_lock, RW_READER);
10291 ipm6.ipv6GroupMemberIfIndex = ill->ill_phyint->phyint_ifindex;
10292 for (ilm = ill->ill_ilm; ilm; ilm = ilm->ilm_next) {
10293 if (ilm->ilm_zoneid != zoneid &&
10294 ilm->ilm_zoneid != ALL_ZONES)
10295 continue; /* not this zone */
10296 ipm6.ipv6GroupMemberAddress = ilm->ilm_v6addr;
10297 ipm6.ipv6GroupMemberRefCnt = ilm->ilm_refcnt;
10298 ipm6.ipv6GroupMemberFilterMode = ilm->ilm_fmode;
10299 if (!snmp_append_data2(mpctl->b_cont,
10300 &mp_tail,
10301 (char *)&ipm6, (int)sizeof (ipm6))) {
10302 ip1dbg(("ip_snmp_get_mib2_ip6_group: "
10303 "failed to allocate %u bytes\n",
10304 (uint_t)sizeof (ipm6)));
10305 }
10306 }
10307 rw_exit(&ill->ill_mcast_lock);
10308 ill_refrele(ill);
10309 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10310 }
10311 rw_exit(&ipst->ips_ill_g_lock);
10312
10313 optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10314 ip3dbg(("ip_snmp_get: level %d, name %d, len %d\n",
10315 (int)optp->level, (int)optp->name, (int)optp->len));
10316 qreply(q, mpctl);
10317 return (mp2ctl);
10318 }
10319
10320 /* IP multicast filtered sources */
10321 static mblk_t *
10322 ip_snmp_get_mib2_ip_group_src(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10323 {
10324 struct opthdr *optp;
10325 mblk_t *mp2ctl;
10326 ill_t *ill;
10327 ipif_t *ipif;
10328 ilm_t *ilm;
10329 ip_grpsrc_t ips;
10330 mblk_t *mp_tail = NULL;
10331 ill_walk_context_t ctx;
10332 zoneid_t zoneid;
10333 int i;
10334 slist_t *sl;
10335
10336 /*
10337 * make a copy of the original message
10338 */
10339 mp2ctl = copymsg(mpctl);
10340 zoneid = Q_TO_CONN(q)->conn_zoneid;
10341
10342 /* ipGroupSource table */
10343 optp = (struct opthdr *)&mpctl->b_rptr[
10344 sizeof (struct T_optmgmt_ack)];
10345 optp->level = MIB2_IP;
10346 optp->name = EXPER_IP_GROUP_SOURCES;
10347
10348 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10349 ill = ILL_START_WALK_V4(&ctx, ipst);
10350 for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10351 /* Make sure the ill isn't going away. */
10352 if (!ill_check_and_refhold(ill))
10353 continue;
10354 rw_exit(&ipst->ips_ill_g_lock);
10355 rw_enter(&ill->ill_mcast_lock, RW_READER);
10356 for (ilm = ill->ill_ilm; ilm; ilm = ilm->ilm_next) {
10357 sl = ilm->ilm_filter;
10358 if (ilm->ilm_zoneid != zoneid &&
10359 ilm->ilm_zoneid != ALL_ZONES)
10360 continue;
10361 if (SLIST_IS_EMPTY(sl))
10362 continue;
10363
10364 /* Is there an ipif for ilm_ifaddr? */
10365 for (ipif = ill->ill_ipif; ipif != NULL;
10366 ipif = ipif->ipif_next) {
10367 if (!IPIF_IS_CONDEMNED(ipif) &&
10368 ipif->ipif_lcl_addr == ilm->ilm_ifaddr &&
10369 ilm->ilm_ifaddr != INADDR_ANY)
10370 break;
10371 }
10372 if (ipif != NULL) {
10373 ipif_get_name(ipif,
10374 ips.ipGroupSourceIfIndex.o_bytes,
10375 OCTET_LENGTH);
10376 } else {
10377 ill_get_name(ill,
10378 ips.ipGroupSourceIfIndex.o_bytes,
10379 OCTET_LENGTH);
10380 }
10381 ips.ipGroupSourceIfIndex.o_length =
10382 mi_strlen(ips.ipGroupSourceIfIndex.o_bytes);
10383
10384 ips.ipGroupSourceGroup = ilm->ilm_addr;
10385 for (i = 0; i < sl->sl_numsrc; i++) {
10386 if (!IN6_IS_ADDR_V4MAPPED(&sl->sl_addr[i]))
10387 continue;
10388 IN6_V4MAPPED_TO_IPADDR(&sl->sl_addr[i],
10389 ips.ipGroupSourceAddress);
10390 if (snmp_append_data2(mpctl->b_cont, &mp_tail,
10391 (char *)&ips, (int)sizeof (ips)) == 0) {
10392 ip1dbg(("ip_snmp_get_mib2_ip_group_src:"
10393 " failed to allocate %u bytes\n",
10394 (uint_t)sizeof (ips)));
10395 }
10396 }
10397 }
10398 rw_exit(&ill->ill_mcast_lock);
10399 ill_refrele(ill);
10400 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10401 }
10402 rw_exit(&ipst->ips_ill_g_lock);
10403 optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10404 ip3dbg(("ip_snmp_get: level %d, name %d, len %d\n",
10405 (int)optp->level, (int)optp->name, (int)optp->len));
10406 qreply(q, mpctl);
10407 return (mp2ctl);
10408 }
10409
10410 /* IPv6 multicast filtered sources. */
10411 static mblk_t *
10412 ip_snmp_get_mib2_ip6_group_src(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10413 {
10414 struct opthdr *optp;
10415 mblk_t *mp2ctl;
10416 ill_t *ill;
10417 ilm_t *ilm;
10418 ipv6_grpsrc_t ips6;
10419 mblk_t *mp_tail = NULL;
10420 ill_walk_context_t ctx;
10421 zoneid_t zoneid;
10422 int i;
10423 slist_t *sl;
10424
10425 /*
10426 * make a copy of the original message
10427 */
10428 mp2ctl = copymsg(mpctl);
10429 zoneid = Q_TO_CONN(q)->conn_zoneid;
10430
10431 /* ip6GroupMember table */
10432 optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10433 optp->level = MIB2_IP6;
10434 optp->name = EXPER_IP6_GROUP_SOURCES;
10435
10436 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10437 ill = ILL_START_WALK_V6(&ctx, ipst);
10438 for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10439 /* Make sure the ill isn't going away. */
10440 if (!ill_check_and_refhold(ill))
10441 continue;
10442 rw_exit(&ipst->ips_ill_g_lock);
10443 /*
10444 * Normally we don't have any members on under IPMP interfaces.
10445 * We report them as a debugging aid.
10446 */
10447 rw_enter(&ill->ill_mcast_lock, RW_READER);
10448 ips6.ipv6GroupSourceIfIndex = ill->ill_phyint->phyint_ifindex;
10449 for (ilm = ill->ill_ilm; ilm; ilm = ilm->ilm_next) {
10450 sl = ilm->ilm_filter;
10451 if (ilm->ilm_zoneid != zoneid &&
10452 ilm->ilm_zoneid != ALL_ZONES)
10453 continue;
10454 if (SLIST_IS_EMPTY(sl))
10455 continue;
10456 ips6.ipv6GroupSourceGroup = ilm->ilm_v6addr;
10457 for (i = 0; i < sl->sl_numsrc; i++) {
10458 ips6.ipv6GroupSourceAddress = sl->sl_addr[i];
10459 if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10460 (char *)&ips6, (int)sizeof (ips6))) {
10461 ip1dbg(("ip_snmp_get_mib2_ip6_"
10462 "group_src: failed to allocate "
10463 "%u bytes\n",
10464 (uint_t)sizeof (ips6)));
10465 }
10466 }
10467 }
10468 rw_exit(&ill->ill_mcast_lock);
10469 ill_refrele(ill);
10470 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10471 }
10472 rw_exit(&ipst->ips_ill_g_lock);
10473
10474 optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10475 ip3dbg(("ip_snmp_get: level %d, name %d, len %d\n",
10476 (int)optp->level, (int)optp->name, (int)optp->len));
10477 qreply(q, mpctl);
10478 return (mp2ctl);
10479 }
10480
10481 /* Multicast routing virtual interface table. */
10482 static mblk_t *
10483 ip_snmp_get_mib2_virt_multi(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10484 {
10485 struct opthdr *optp;
10486 mblk_t *mp2ctl;
10487
10488 /*
10489 * make a copy of the original message
10490 */
10491 mp2ctl = copymsg(mpctl);
10492
10493 optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10494 optp->level = EXPER_DVMRP;
10495 optp->name = EXPER_DVMRP_VIF;
10496 if (!ip_mroute_vif(mpctl->b_cont, ipst)) {
10497 ip0dbg(("ip_mroute_vif: failed\n"));
10498 }
10499 optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10500 ip3dbg(("ip_snmp_get_mib2_virt_multi: level %d, name %d, len %d\n",
10501 (int)optp->level, (int)optp->name, (int)optp->len));
10502 qreply(q, mpctl);
10503 return (mp2ctl);
10504 }
10505
10506 /* Multicast routing table. */
10507 static mblk_t *
10508 ip_snmp_get_mib2_multi_rtable(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10509 {
10510 struct opthdr *optp;
10511 mblk_t *mp2ctl;
10512
10513 /*
10514 * make a copy of the original message
10515 */
10516 mp2ctl = copymsg(mpctl);
10517
10518 optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10519 optp->level = EXPER_DVMRP;
10520 optp->name = EXPER_DVMRP_MRT;
10521 if (!ip_mroute_mrt(mpctl->b_cont, ipst)) {
10522 ip0dbg(("ip_mroute_mrt: failed\n"));
10523 }
10524 optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10525 ip3dbg(("ip_snmp_get_mib2_multi_rtable: level %d, name %d, len %d\n",
10526 (int)optp->level, (int)optp->name, (int)optp->len));
10527 qreply(q, mpctl);
10528 return (mp2ctl);
10529 }
10530
10531 /*
10532 * Return ipRouteEntryTable, ipNetToMediaEntryTable, and ipRouteAttributeTable
10533 * in one IRE walk.
10534 */
10535 static mblk_t *
10536 ip_snmp_get_mib2_ip_route_media(queue_t *q, mblk_t *mpctl, int level,
10537 ip_stack_t *ipst)
10538 {
10539 struct opthdr *optp;
10540 mblk_t *mp2ctl; /* Returned */
10541 mblk_t *mp3ctl; /* nettomedia */
10542 mblk_t *mp4ctl; /* routeattrs */
10543 iproutedata_t ird;
10544 zoneid_t zoneid;
10545
10546 /*
10547 * make copies of the original message
10548 * - mp2ctl is returned unchanged to the caller for his use
10549 * - mpctl is sent upstream as ipRouteEntryTable
10550 * - mp3ctl is sent upstream as ipNetToMediaEntryTable
10551 * - mp4ctl is sent upstream as ipRouteAttributeTable
10552 */
10553 mp2ctl = copymsg(mpctl);
10554 mp3ctl = copymsg(mpctl);
10555 mp4ctl = copymsg(mpctl);
10556 if (mp3ctl == NULL || mp4ctl == NULL) {
10557 freemsg(mp4ctl);
10558 freemsg(mp3ctl);
10559 freemsg(mp2ctl);
10560 freemsg(mpctl);
10561 return (NULL);
10562 }
10563
10564 bzero(&ird, sizeof (ird));
10565
10566 ird.ird_route.lp_head = mpctl->b_cont;
10567 ird.ird_netmedia.lp_head = mp3ctl->b_cont;
10568 ird.ird_attrs.lp_head = mp4ctl->b_cont;
10569 /*
10570 * If the level has been set the special EXPER_IP_AND_ALL_IRES value,
10571 * then also include ire_testhidden IREs and IRE_IF_CLONE. This is
10572 * intended a temporary solution until a proper MIB API is provided
10573 * that provides complete filtering/caller-opt-in.
10574 */
10575 if (level == EXPER_IP_AND_ALL_IRES)
10576 ird.ird_flags |= IRD_REPORT_ALL;
10577
10578 zoneid = Q_TO_CONN(q)->conn_zoneid;
10579 ire_walk_v4(ip_snmp_get2_v4, &ird, zoneid, ipst);
10580
10581 /* ipRouteEntryTable in mpctl */
10582 optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10583 optp->level = MIB2_IP;
10584 optp->name = MIB2_IP_ROUTE;
10585 optp->len = msgdsize(ird.ird_route.lp_head);
10586 ip3dbg(("ip_snmp_get_mib2_ip_route_media: level %d, name %d, len %d\n",
10587 (int)optp->level, (int)optp->name, (int)optp->len));
10588 qreply(q, mpctl);
10589
10590 /* ipNetToMediaEntryTable in mp3ctl */
10591 ncec_walk(NULL, ip_snmp_get2_v4_media, &ird, ipst);
10592
10593 optp = (struct opthdr *)&mp3ctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10594 optp->level = MIB2_IP;
10595 optp->name = MIB2_IP_MEDIA;
10596 optp->len = msgdsize(ird.ird_netmedia.lp_head);
10597 ip3dbg(("ip_snmp_get_mib2_ip_route_media: level %d, name %d, len %d\n",
10598 (int)optp->level, (int)optp->name, (int)optp->len));
10599 qreply(q, mp3ctl);
10600
10601 /* ipRouteAttributeTable in mp4ctl */
10602 optp = (struct opthdr *)&mp4ctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10603 optp->level = MIB2_IP;
10604 optp->name = EXPER_IP_RTATTR;
10605 optp->len = msgdsize(ird.ird_attrs.lp_head);
10606 ip3dbg(("ip_snmp_get_mib2_ip_route_media: level %d, name %d, len %d\n",
10607 (int)optp->level, (int)optp->name, (int)optp->len));
10608 if (optp->len == 0)
10609 freemsg(mp4ctl);
10610 else
10611 qreply(q, mp4ctl);
10612
10613 return (mp2ctl);
10614 }
10615
10616 /*
10617 * Return ipv6RouteEntryTable and ipv6RouteAttributeTable in one IRE walk, and
10618 * ipv6NetToMediaEntryTable in an NDP walk.
10619 */
10620 static mblk_t *
10621 ip_snmp_get_mib2_ip6_route_media(queue_t *q, mblk_t *mpctl, int level,
10622 ip_stack_t *ipst)
10623 {
10624 struct opthdr *optp;
10625 mblk_t *mp2ctl; /* Returned */
10626 mblk_t *mp3ctl; /* nettomedia */
10627 mblk_t *mp4ctl; /* routeattrs */
10628 iproutedata_t ird;
10629 zoneid_t zoneid;
10630
10631 /*
10632 * make copies of the original message
10633 * - mp2ctl is returned unchanged to the caller for his use
10634 * - mpctl is sent upstream as ipv6RouteEntryTable
10635 * - mp3ctl is sent upstream as ipv6NetToMediaEntryTable
10636 * - mp4ctl is sent upstream as ipv6RouteAttributeTable
10637 */
10638 mp2ctl = copymsg(mpctl);
10639 mp3ctl = copymsg(mpctl);
10640 mp4ctl = copymsg(mpctl);
10641 if (mp3ctl == NULL || mp4ctl == NULL) {
10642 freemsg(mp4ctl);
10643 freemsg(mp3ctl);
10644 freemsg(mp2ctl);
10645 freemsg(mpctl);
10646 return (NULL);
10647 }
10648
10649 bzero(&ird, sizeof (ird));
10650
10651 ird.ird_route.lp_head = mpctl->b_cont;
10652 ird.ird_netmedia.lp_head = mp3ctl->b_cont;
10653 ird.ird_attrs.lp_head = mp4ctl->b_cont;
10654 /*
10655 * If the level has been set the special EXPER_IP_AND_ALL_IRES value,
10656 * then also include ire_testhidden IREs and IRE_IF_CLONE. This is
10657 * intended a temporary solution until a proper MIB API is provided
10658 * that provides complete filtering/caller-opt-in.
10659 */
10660 if (level == EXPER_IP_AND_ALL_IRES)
10661 ird.ird_flags |= IRD_REPORT_ALL;
10662
10663 zoneid = Q_TO_CONN(q)->conn_zoneid;
10664 ire_walk_v6(ip_snmp_get2_v6_route, &ird, zoneid, ipst);
10665
10666 optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10667 optp->level = MIB2_IP6;
10668 optp->name = MIB2_IP6_ROUTE;
10669 optp->len = msgdsize(ird.ird_route.lp_head);
10670 ip3dbg(("ip_snmp_get_mib2_ip6_route_media: level %d, name %d, len %d\n",
10671 (int)optp->level, (int)optp->name, (int)optp->len));
10672 qreply(q, mpctl);
10673
10674 /* ipv6NetToMediaEntryTable in mp3ctl */
10675 ncec_walk(NULL, ip_snmp_get2_v6_media, &ird, ipst);
10676
10677 optp = (struct opthdr *)&mp3ctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10678 optp->level = MIB2_IP6;
10679 optp->name = MIB2_IP6_MEDIA;
10680 optp->len = msgdsize(ird.ird_netmedia.lp_head);
10681 ip3dbg(("ip_snmp_get_mib2_ip6_route_media: level %d, name %d, len %d\n",
10682 (int)optp->level, (int)optp->name, (int)optp->len));
10683 qreply(q, mp3ctl);
10684
10685 /* ipv6RouteAttributeTable in mp4ctl */
10686 optp = (struct opthdr *)&mp4ctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10687 optp->level = MIB2_IP6;
10688 optp->name = EXPER_IP_RTATTR;
10689 optp->len = msgdsize(ird.ird_attrs.lp_head);
10690 ip3dbg(("ip_snmp_get_mib2_ip6_route_media: level %d, name %d, len %d\n",
10691 (int)optp->level, (int)optp->name, (int)optp->len));
10692 if (optp->len == 0)
10693 freemsg(mp4ctl);
10694 else
10695 qreply(q, mp4ctl);
10696
10697 return (mp2ctl);
10698 }
10699
10700 /*
10701 * IPv6 mib: One per ill
10702 */
10703 static mblk_t *
10704 ip_snmp_get_mib2_ip6(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst,
10705 boolean_t legacy_req)
10706 {
10707 struct opthdr *optp;
10708 mblk_t *mp2ctl;
10709 ill_t *ill;
10710 ill_walk_context_t ctx;
10711 mblk_t *mp_tail = NULL;
10712 mib2_ipv6AddrEntry_t mae6;
10713 mib2_ipIfStatsEntry_t *ise;
10714 size_t ise_size, iae_size;
10715
10716 /*
10717 * Make a copy of the original message
10718 */
10719 mp2ctl = copymsg(mpctl);
10720
10721 /* fixed length IPv6 structure ... */
10722
10723 if (legacy_req) {
10724 ise_size = LEGACY_MIB_SIZE(&ipst->ips_ip6_mib,
10725 mib2_ipIfStatsEntry_t);
10726 iae_size = LEGACY_MIB_SIZE(&mae6, mib2_ipv6AddrEntry_t);
10727 } else {
10728 ise_size = sizeof (mib2_ipIfStatsEntry_t);
10729 iae_size = sizeof (mib2_ipv6AddrEntry_t);
10730 }
10731
10732 optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10733 optp->level = MIB2_IP6;
10734 optp->name = 0;
10735 /* Include "unknown interface" ip6_mib */
10736 ipst->ips_ip6_mib.ipIfStatsIPVersion = MIB2_INETADDRESSTYPE_ipv6;
10737 ipst->ips_ip6_mib.ipIfStatsIfIndex =
10738 MIB2_UNKNOWN_INTERFACE; /* Flag to netstat */
10739 SET_MIB(ipst->ips_ip6_mib.ipIfStatsForwarding,
10740 ipst->ips_ipv6_forwarding ? 1 : 2);
10741 SET_MIB(ipst->ips_ip6_mib.ipIfStatsDefaultHopLimit,
10742 ipst->ips_ipv6_def_hops);
10743 SET_MIB(ipst->ips_ip6_mib.ipIfStatsEntrySize,
10744 sizeof (mib2_ipIfStatsEntry_t));
10745 SET_MIB(ipst->ips_ip6_mib.ipIfStatsAddrEntrySize,
10746 sizeof (mib2_ipv6AddrEntry_t));
10747 SET_MIB(ipst->ips_ip6_mib.ipIfStatsRouteEntrySize,
10748 sizeof (mib2_ipv6RouteEntry_t));
10749 SET_MIB(ipst->ips_ip6_mib.ipIfStatsNetToMediaEntrySize,
10750 sizeof (mib2_ipv6NetToMediaEntry_t));
10751 SET_MIB(ipst->ips_ip6_mib.ipIfStatsMemberEntrySize,
10752 sizeof (ipv6_member_t));
10753 SET_MIB(ipst->ips_ip6_mib.ipIfStatsGroupSourceEntrySize,
10754 sizeof (ipv6_grpsrc_t));
10755
10756 /*
10757 * Synchronize 64- and 32-bit counters
10758 */
10759 SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsInReceives,
10760 ipIfStatsHCInReceives);
10761 SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsInDelivers,
10762 ipIfStatsHCInDelivers);
10763 SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsOutRequests,
10764 ipIfStatsHCOutRequests);
10765 SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsOutForwDatagrams,
10766 ipIfStatsHCOutForwDatagrams);
10767 SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsOutMcastPkts,
10768 ipIfStatsHCOutMcastPkts);
10769 SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsInMcastPkts,
10770 ipIfStatsHCInMcastPkts);
10771
10772 if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10773 (char *)&ipst->ips_ip6_mib, (int)ise_size)) {
10774 ip1dbg(("ip_snmp_get_mib2_ip6: failed to allocate %u bytes\n",
10775 (uint_t)ise_size));
10776 } else if (legacy_req) {
10777 /* Adjust the EntrySize fields for legacy requests. */
10778 ise =
10779 (mib2_ipIfStatsEntry_t *)(mp_tail->b_wptr - (int)ise_size);
10780 SET_MIB(ise->ipIfStatsEntrySize, ise_size);
10781 SET_MIB(ise->ipIfStatsAddrEntrySize, iae_size);
10782 }
10783
10784 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10785 ill = ILL_START_WALK_V6(&ctx, ipst);
10786 for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10787 ill->ill_ip_mib->ipIfStatsIfIndex =
10788 ill->ill_phyint->phyint_ifindex;
10789 SET_MIB(ill->ill_ip_mib->ipIfStatsForwarding,
10790 ipst->ips_ipv6_forwarding ? 1 : 2);
10791 SET_MIB(ill->ill_ip_mib->ipIfStatsDefaultHopLimit,
10792 ill->ill_max_hops);
10793
10794 /*
10795 * Synchronize 64- and 32-bit counters
10796 */
10797 SYNC32_MIB(ill->ill_ip_mib, ipIfStatsInReceives,
10798 ipIfStatsHCInReceives);
10799 SYNC32_MIB(ill->ill_ip_mib, ipIfStatsInDelivers,
10800 ipIfStatsHCInDelivers);
10801 SYNC32_MIB(ill->ill_ip_mib, ipIfStatsOutRequests,
10802 ipIfStatsHCOutRequests);
10803 SYNC32_MIB(ill->ill_ip_mib, ipIfStatsOutForwDatagrams,
10804 ipIfStatsHCOutForwDatagrams);
10805 SYNC32_MIB(ill->ill_ip_mib, ipIfStatsOutMcastPkts,
10806 ipIfStatsHCOutMcastPkts);
10807 SYNC32_MIB(ill->ill_ip_mib, ipIfStatsInMcastPkts,
10808 ipIfStatsHCInMcastPkts);
10809
10810 if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10811 (char *)ill->ill_ip_mib, (int)ise_size)) {
10812 ip1dbg(("ip_snmp_get_mib2_ip6: failed to allocate "
10813 "%u bytes\n", (uint_t)ise_size));
10814 } else if (legacy_req) {
10815 /* Adjust the EntrySize fields for legacy requests. */
10816 ise = (mib2_ipIfStatsEntry_t *)(mp_tail->b_wptr -
10817 (int)ise_size);
10818 SET_MIB(ise->ipIfStatsEntrySize, ise_size);
10819 SET_MIB(ise->ipIfStatsAddrEntrySize, iae_size);
10820 }
10821 }
10822 rw_exit(&ipst->ips_ill_g_lock);
10823
10824 optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10825 ip3dbg(("ip_snmp_get_mib2_ip6: level %d, name %d, len %d\n",
10826 (int)optp->level, (int)optp->name, (int)optp->len));
10827 qreply(q, mpctl);
10828 return (mp2ctl);
10829 }
10830
10831 /*
10832 * ICMPv6 mib: One per ill
10833 */
10834 static mblk_t *
10835 ip_snmp_get_mib2_icmp6(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10836 {
10837 struct opthdr *optp;
10838 mblk_t *mp2ctl;
10839 ill_t *ill;
10840 ill_walk_context_t ctx;
10841 mblk_t *mp_tail = NULL;
10842 /*
10843 * Make a copy of the original message
10844 */
10845 mp2ctl = copymsg(mpctl);
10846
10847 /* fixed length ICMPv6 structure ... */
10848
10849 optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10850 optp->level = MIB2_ICMP6;
10851 optp->name = 0;
10852 /* Include "unknown interface" icmp6_mib */
10853 ipst->ips_icmp6_mib.ipv6IfIcmpIfIndex =
10854 MIB2_UNKNOWN_INTERFACE; /* netstat flag */
10855 ipst->ips_icmp6_mib.ipv6IfIcmpEntrySize =
10856 sizeof (mib2_ipv6IfIcmpEntry_t);
10857 if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10858 (char *)&ipst->ips_icmp6_mib,
10859 (int)sizeof (ipst->ips_icmp6_mib))) {
10860 ip1dbg(("ip_snmp_get_mib2_icmp6: failed to allocate %u bytes\n",
10861 (uint_t)sizeof (ipst->ips_icmp6_mib)));
10862 }
10863
10864 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10865 ill = ILL_START_WALK_V6(&ctx, ipst);
10866 for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10867 ill->ill_icmp6_mib->ipv6IfIcmpIfIndex =
10868 ill->ill_phyint->phyint_ifindex;
10869 if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10870 (char *)ill->ill_icmp6_mib,
10871 (int)sizeof (*ill->ill_icmp6_mib))) {
10872 ip1dbg(("ip_snmp_get_mib2_icmp6: failed to allocate "
10873 "%u bytes\n",
10874 (uint_t)sizeof (*ill->ill_icmp6_mib)));
10875 }
10876 }
10877 rw_exit(&ipst->ips_ill_g_lock);
10878
10879 optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10880 ip3dbg(("ip_snmp_get_mib2_icmp6: level %d, name %d, len %d\n",
10881 (int)optp->level, (int)optp->name, (int)optp->len));
10882 qreply(q, mpctl);
10883 return (mp2ctl);
10884 }
10885
10886 /*
10887 * ire_walk routine to create both ipRouteEntryTable and
10888 * ipRouteAttributeTable in one IRE walk
10889 */
10890 static void
10891 ip_snmp_get2_v4(ire_t *ire, iproutedata_t *ird)
10892 {
10893 ill_t *ill;
10894 mib2_ipRouteEntry_t *re;
10895 mib2_ipAttributeEntry_t iaes;
10896 tsol_ire_gw_secattr_t *attrp;
10897 tsol_gc_t *gc = NULL;
10898 tsol_gcgrp_t *gcgrp = NULL;
10899 ip_stack_t *ipst = ire->ire_ipst;
10900
10901 ASSERT(ire->ire_ipversion == IPV4_VERSION);
10902
10903 if (!(ird->ird_flags & IRD_REPORT_ALL)) {
10904 if (ire->ire_testhidden)
10905 return;
10906 if (ire->ire_type & IRE_IF_CLONE)
10907 return;
10908 }
10909
10910 if ((re = kmem_zalloc(sizeof (*re), KM_NOSLEEP)) == NULL)
10911 return;
10912
10913 if ((attrp = ire->ire_gw_secattr) != NULL) {
10914 mutex_enter(&attrp->igsa_lock);
10915 if ((gc = attrp->igsa_gc) != NULL) {
10916 gcgrp = gc->gc_grp;
10917 ASSERT(gcgrp != NULL);
10918 rw_enter(&gcgrp->gcgrp_rwlock, RW_READER);
10919 }
10920 mutex_exit(&attrp->igsa_lock);
10921 }
10922 /*
10923 * Return all IRE types for route table... let caller pick and choose
10924 */
10925 re->ipRouteDest = ire->ire_addr;
10926 ill = ire->ire_ill;
10927 re->ipRouteIfIndex.o_length = 0;
10928 if (ill != NULL) {
10929 ill_get_name(ill, re->ipRouteIfIndex.o_bytes, OCTET_LENGTH);
10930 re->ipRouteIfIndex.o_length =
10931 mi_strlen(re->ipRouteIfIndex.o_bytes);
10932 }
10933 re->ipRouteMetric1 = -1;
10934 re->ipRouteMetric2 = -1;
10935 re->ipRouteMetric3 = -1;
10936 re->ipRouteMetric4 = -1;
10937
10938 re->ipRouteNextHop = ire->ire_gateway_addr;
10939 /* indirect(4), direct(3), or invalid(2) */
10940 if (ire->ire_flags & (RTF_REJECT | RTF_BLACKHOLE))
10941 re->ipRouteType = 2;
10942 else if (ire->ire_type & IRE_ONLINK)
10943 re->ipRouteType = 3;
10944 else
10945 re->ipRouteType = 4;
10946
10947 re->ipRouteProto = -1;
10948 re->ipRouteAge = gethrestime_sec() - ire->ire_create_time;
10949 re->ipRouteMask = ire->ire_mask;
10950 re->ipRouteMetric5 = -1;
10951 re->ipRouteInfo.re_max_frag = ire->ire_metrics.iulp_mtu;
10952 if (ire->ire_ill != NULL && re->ipRouteInfo.re_max_frag == 0)
10953 re->ipRouteInfo.re_max_frag = ire->ire_ill->ill_mtu;
10954
10955 re->ipRouteInfo.re_frag_flag = 0;
10956 re->ipRouteInfo.re_rtt = 0;
10957 re->ipRouteInfo.re_src_addr = 0;
10958 re->ipRouteInfo.re_ref = ire->ire_refcnt;
10959 re->ipRouteInfo.re_obpkt = ire->ire_ob_pkt_count;
10960 re->ipRouteInfo.re_ibpkt = ire->ire_ib_pkt_count;
10961 re->ipRouteInfo.re_flags = ire->ire_flags;
10962
10963 /* Add the IRE_IF_CLONE's counters to their parent IRE_INTERFACE */
10964 if (ire->ire_type & IRE_INTERFACE) {
10965 ire_t *child;
10966
10967 rw_enter(&ipst->ips_ire_dep_lock, RW_READER);
10968 child = ire->ire_dep_children;
10969 while (child != NULL) {
10970 re->ipRouteInfo.re_obpkt += child->ire_ob_pkt_count;
10971 re->ipRouteInfo.re_ibpkt += child->ire_ib_pkt_count;
10972 child = child->ire_dep_sib_next;
10973 }
10974 rw_exit(&ipst->ips_ire_dep_lock);
10975 }
10976
10977 if (ire->ire_flags & RTF_DYNAMIC) {
10978 re->ipRouteInfo.re_ire_type = IRE_HOST_REDIRECT;
10979 } else {
10980 re->ipRouteInfo.re_ire_type = ire->ire_type;
10981 }
10982
10983 if (!snmp_append_data2(ird->ird_route.lp_head, &ird->ird_route.lp_tail,
10984 (char *)re, (int)sizeof (*re))) {
10985 ip1dbg(("ip_snmp_get2_v4: failed to allocate %u bytes\n",
10986 (uint_t)sizeof (*re)));
10987 }
10988
10989 if (gc != NULL) {
10990 iaes.iae_routeidx = ird->ird_idx;
10991 iaes.iae_doi = gc->gc_db->gcdb_doi;
10992 iaes.iae_slrange = gc->gc_db->gcdb_slrange;
10993
10994 if (!snmp_append_data2(ird->ird_attrs.lp_head,
10995 &ird->ird_attrs.lp_tail, (char *)&iaes, sizeof (iaes))) {
10996 ip1dbg(("ip_snmp_get2_v4: failed to allocate %u "
10997 "bytes\n", (uint_t)sizeof (iaes)));
10998 }
10999 }
11000
11001 /* bump route index for next pass */
11002 ird->ird_idx++;
11003
11004 kmem_free(re, sizeof (*re));
11005 if (gcgrp != NULL)
11006 rw_exit(&gcgrp->gcgrp_rwlock);
11007 }
11008
11009 /*
11010 * ire_walk routine to create ipv6RouteEntryTable and ipRouteEntryTable.
11011 */
11012 static void
11013 ip_snmp_get2_v6_route(ire_t *ire, iproutedata_t *ird)
11014 {
11015 ill_t *ill;
11016 mib2_ipv6RouteEntry_t *re;
11017 mib2_ipAttributeEntry_t iaes;
11018 tsol_ire_gw_secattr_t *attrp;
11019 tsol_gc_t *gc = NULL;
11020 tsol_gcgrp_t *gcgrp = NULL;
11021 ip_stack_t *ipst = ire->ire_ipst;
11022
11023 ASSERT(ire->ire_ipversion == IPV6_VERSION);
11024
11025 if (!(ird->ird_flags & IRD_REPORT_ALL)) {
11026 if (ire->ire_testhidden)
11027 return;
11028 if (ire->ire_type & IRE_IF_CLONE)
11029 return;
11030 }
11031
11032 if ((re = kmem_zalloc(sizeof (*re), KM_NOSLEEP)) == NULL)
11033 return;
11034
11035 if ((attrp = ire->ire_gw_secattr) != NULL) {
11036 mutex_enter(&attrp->igsa_lock);
11037 if ((gc = attrp->igsa_gc) != NULL) {
11038 gcgrp = gc->gc_grp;
11039 ASSERT(gcgrp != NULL);
11040 rw_enter(&gcgrp->gcgrp_rwlock, RW_READER);
11041 }
11042 mutex_exit(&attrp->igsa_lock);
11043 }
11044 /*
11045 * Return all IRE types for route table... let caller pick and choose
11046 */
11047 re->ipv6RouteDest = ire->ire_addr_v6;
11048 re->ipv6RoutePfxLength = ip_mask_to_plen_v6(&ire->ire_mask_v6);
11049 re->ipv6RouteIndex = 0; /* Unique when multiple with same dest/plen */
11050 re->ipv6RouteIfIndex.o_length = 0;
11051 ill = ire->ire_ill;
11052 if (ill != NULL) {
11053 ill_get_name(ill, re->ipv6RouteIfIndex.o_bytes, OCTET_LENGTH);
11054 re->ipv6RouteIfIndex.o_length =
11055 mi_strlen(re->ipv6RouteIfIndex.o_bytes);
11056 }
11057
11058 ASSERT(!(ire->ire_type & IRE_BROADCAST));
11059
11060 mutex_enter(&ire->ire_lock);
11061 re->ipv6RouteNextHop = ire->ire_gateway_addr_v6;
11062 mutex_exit(&ire->ire_lock);
11063
11064 /* remote(4), local(3), or discard(2) */
11065 if (ire->ire_flags & (RTF_REJECT | RTF_BLACKHOLE))
11066 re->ipv6RouteType = 2;
11067 else if (ire->ire_type & IRE_ONLINK)
11068 re->ipv6RouteType = 3;
11069 else
11070 re->ipv6RouteType = 4;
11071
11072 re->ipv6RouteProtocol = -1;
11073 re->ipv6RoutePolicy = 0;
11074 re->ipv6RouteAge = gethrestime_sec() - ire->ire_create_time;
11075 re->ipv6RouteNextHopRDI = 0;
11076 re->ipv6RouteWeight = 0;
11077 re->ipv6RouteMetric = 0;
11078 re->ipv6RouteInfo.re_max_frag = ire->ire_metrics.iulp_mtu;
11079 if (ire->ire_ill != NULL && re->ipv6RouteInfo.re_max_frag == 0)
11080 re->ipv6RouteInfo.re_max_frag = ire->ire_ill->ill_mtu;
11081
11082 re->ipv6RouteInfo.re_frag_flag = 0;
11083 re->ipv6RouteInfo.re_rtt = 0;
11084 re->ipv6RouteInfo.re_src_addr = ipv6_all_zeros;
11085 re->ipv6RouteInfo.re_obpkt = ire->ire_ob_pkt_count;
11086 re->ipv6RouteInfo.re_ibpkt = ire->ire_ib_pkt_count;
11087 re->ipv6RouteInfo.re_ref = ire->ire_refcnt;
11088 re->ipv6RouteInfo.re_flags = ire->ire_flags;
11089
11090 /* Add the IRE_IF_CLONE's counters to their parent IRE_INTERFACE */
11091 if (ire->ire_type & IRE_INTERFACE) {
11092 ire_t *child;
11093
11094 rw_enter(&ipst->ips_ire_dep_lock, RW_READER);
11095 child = ire->ire_dep_children;
11096 while (child != NULL) {
11097 re->ipv6RouteInfo.re_obpkt += child->ire_ob_pkt_count;
11098 re->ipv6RouteInfo.re_ibpkt += child->ire_ib_pkt_count;
11099 child = child->ire_dep_sib_next;
11100 }
11101 rw_exit(&ipst->ips_ire_dep_lock);
11102 }
11103 if (ire->ire_flags & RTF_DYNAMIC) {
11104 re->ipv6RouteInfo.re_ire_type = IRE_HOST_REDIRECT;
11105 } else {
11106 re->ipv6RouteInfo.re_ire_type = ire->ire_type;
11107 }
11108
11109 if (!snmp_append_data2(ird->ird_route.lp_head, &ird->ird_route.lp_tail,
11110 (char *)re, (int)sizeof (*re))) {
11111 ip1dbg(("ip_snmp_get2_v6: failed to allocate %u bytes\n",
11112 (uint_t)sizeof (*re)));
11113 }
11114
11115 if (gc != NULL) {
11116 iaes.iae_routeidx = ird->ird_idx;
11117 iaes.iae_doi = gc->gc_db->gcdb_doi;
11118 iaes.iae_slrange = gc->gc_db->gcdb_slrange;
11119
11120 if (!snmp_append_data2(ird->ird_attrs.lp_head,
11121 &ird->ird_attrs.lp_tail, (char *)&iaes, sizeof (iaes))) {
11122 ip1dbg(("ip_snmp_get2_v6: failed to allocate %u "
11123 "bytes\n", (uint_t)sizeof (iaes)));
11124 }
11125 }
11126
11127 /* bump route index for next pass */
11128 ird->ird_idx++;
11129
11130 kmem_free(re, sizeof (*re));
11131 if (gcgrp != NULL)
11132 rw_exit(&gcgrp->gcgrp_rwlock);
11133 }
11134
11135 /*
11136 * ncec_walk routine to create ipv6NetToMediaEntryTable
11137 */
11138 static int
11139 ip_snmp_get2_v6_media(ncec_t *ncec, iproutedata_t *ird)
11140 {
11141 ill_t *ill;
11142 mib2_ipv6NetToMediaEntry_t ntme;
11143
11144 ill = ncec->ncec_ill;
11145 /* skip arpce entries, and loopback ncec entries */
11146 if (ill->ill_isv6 == B_FALSE || ill->ill_net_type == IRE_LOOPBACK)
11147 return (0);
11148 /*
11149 * Neighbor cache entry attached to IRE with on-link
11150 * destination.
11151 * We report all IPMP groups on ncec_ill which is normally the upper.
11152 */
11153 ntme.ipv6NetToMediaIfIndex = ill->ill_phyint->phyint_ifindex;
11154 ntme.ipv6NetToMediaNetAddress = ncec->ncec_addr;
11155 ntme.ipv6NetToMediaPhysAddress.o_length = ill->ill_phys_addr_length;
11156 if (ncec->ncec_lladdr != NULL) {
11157 bcopy(ncec->ncec_lladdr, ntme.ipv6NetToMediaPhysAddress.o_bytes,
11158 ntme.ipv6NetToMediaPhysAddress.o_length);
11159 }
11160 /*
11161 * Note: Returns ND_* states. Should be:
11162 * reachable(1), stale(2), delay(3), probe(4),
11163 * invalid(5), unknown(6)
11164 */
11165 ntme.ipv6NetToMediaState = ncec->ncec_state;
11166 ntme.ipv6NetToMediaLastUpdated = 0;
11167
11168 /* other(1), dynamic(2), static(3), local(4) */
11169 if (NCE_MYADDR(ncec)) {
11170 ntme.ipv6NetToMediaType = 4;
11171 } else if (ncec->ncec_flags & NCE_F_PUBLISH) {
11172 ntme.ipv6NetToMediaType = 1; /* proxy */
11173 } else if (ncec->ncec_flags & NCE_F_STATIC) {
11174 ntme.ipv6NetToMediaType = 3;
11175 } else if (ncec->ncec_flags & (NCE_F_MCAST|NCE_F_BCAST)) {
11176 ntme.ipv6NetToMediaType = 1;
11177 } else {
11178 ntme.ipv6NetToMediaType = 2;
11179 }
11180
11181 if (!snmp_append_data2(ird->ird_netmedia.lp_head,
11182 &ird->ird_netmedia.lp_tail, (char *)&ntme, sizeof (ntme))) {
11183 ip1dbg(("ip_snmp_get2_v6_media: failed to allocate %u bytes\n",
11184 (uint_t)sizeof (ntme)));
11185 }
11186 return (0);
11187 }
11188
11189 int
11190 nce2ace(ncec_t *ncec)
11191 {
11192 int flags = 0;
11193
11194 if (NCE_ISREACHABLE(ncec))
11195 flags |= ACE_F_RESOLVED;
11196 if (ncec->ncec_flags & NCE_F_AUTHORITY)
11197 flags |= ACE_F_AUTHORITY;
11198 if (ncec->ncec_flags & NCE_F_PUBLISH)
11199 flags |= ACE_F_PUBLISH;
11200 if ((ncec->ncec_flags & NCE_F_NONUD) != 0)
11201 flags |= ACE_F_PERMANENT;
11202 if (NCE_MYADDR(ncec))
11203 flags |= (ACE_F_MYADDR | ACE_F_AUTHORITY);
11204 if (ncec->ncec_flags & NCE_F_UNVERIFIED)
11205 flags |= ACE_F_UNVERIFIED;
11206 if (ncec->ncec_flags & NCE_F_AUTHORITY)
11207 flags |= ACE_F_AUTHORITY;
11208 if (ncec->ncec_flags & NCE_F_DELAYED)
11209 flags |= ACE_F_DELAYED;
11210 return (flags);
11211 }
11212
11213 /*
11214 * ncec_walk routine to create ipNetToMediaEntryTable
11215 */
11216 static int
11217 ip_snmp_get2_v4_media(ncec_t *ncec, iproutedata_t *ird)
11218 {
11219 ill_t *ill;
11220 mib2_ipNetToMediaEntry_t ntme;
11221 const char *name = "unknown";
11222 ipaddr_t ncec_addr;
11223
11224 ill = ncec->ncec_ill;
11225 if (ill->ill_isv6 || (ncec->ncec_flags & NCE_F_BCAST) ||
11226 ill->ill_net_type == IRE_LOOPBACK)
11227 return (0);
11228
11229 /* We report all IPMP groups on ncec_ill which is normally the upper. */
11230 name = ill->ill_name;
11231 /* Based on RFC 4293: other(1), inval(2), dyn(3), stat(4) */
11232 if (NCE_MYADDR(ncec)) {
11233 ntme.ipNetToMediaType = 4;
11234 } else if (ncec->ncec_flags & (NCE_F_MCAST|NCE_F_BCAST|NCE_F_PUBLISH)) {
11235 ntme.ipNetToMediaType = 1;
11236 } else {
11237 ntme.ipNetToMediaType = 3;
11238 }
11239 ntme.ipNetToMediaIfIndex.o_length = MIN(OCTET_LENGTH, strlen(name));
11240 bcopy(name, ntme.ipNetToMediaIfIndex.o_bytes,
11241 ntme.ipNetToMediaIfIndex.o_length);
11242
11243 IN6_V4MAPPED_TO_IPADDR(&ncec->ncec_addr, ncec_addr);
11244 bcopy(&ncec_addr, &ntme.ipNetToMediaNetAddress, sizeof (ncec_addr));
11245
11246 ntme.ipNetToMediaInfo.ntm_mask.o_length = sizeof (ipaddr_t);
11247 ncec_addr = INADDR_BROADCAST;
11248 bcopy(&ncec_addr, ntme.ipNetToMediaInfo.ntm_mask.o_bytes,
11249 sizeof (ncec_addr));
11250 /*
11251 * map all the flags to the ACE counterpart.
11252 */
11253 ntme.ipNetToMediaInfo.ntm_flags = nce2ace(ncec);
11254
11255 ntme.ipNetToMediaPhysAddress.o_length =
11256 MIN(OCTET_LENGTH, ill->ill_phys_addr_length);
11257
11258 if (!NCE_ISREACHABLE(ncec))
11259 ntme.ipNetToMediaPhysAddress.o_length = 0;
11260 else {
11261 if (ncec->ncec_lladdr != NULL) {
11262 bcopy(ncec->ncec_lladdr,
11263 ntme.ipNetToMediaPhysAddress.o_bytes,
11264 ntme.ipNetToMediaPhysAddress.o_length);
11265 }
11266 }
11267
11268 if (!snmp_append_data2(ird->ird_netmedia.lp_head,
11269 &ird->ird_netmedia.lp_tail, (char *)&ntme, sizeof (ntme))) {
11270 ip1dbg(("ip_snmp_get2_v4_media: failed to allocate %u bytes\n",
11271 (uint_t)sizeof (ntme)));
11272 }
11273 return (0);
11274 }
11275
11276 /*
11277 * return (0) if invalid set request, 1 otherwise, including non-tcp requests
11278 */
11279 /* ARGSUSED */
11280 int
11281 ip_snmp_set(queue_t *q, int level, int name, uchar_t *ptr, int len)
11282 {
11283 switch (level) {
11284 case MIB2_IP:
11285 case MIB2_ICMP:
11286 switch (name) {
11287 default:
11288 break;
11289 }
11290 return (1);
11291 default:
11292 return (1);
11293 }
11294 }
11295
11296 /*
11297 * When there exists both a 64- and 32-bit counter of a particular type
11298 * (i.e., InReceives), only the 64-bit counters are added.
11299 */
11300 void
11301 ip_mib2_add_ip_stats(mib2_ipIfStatsEntry_t *o1, mib2_ipIfStatsEntry_t *o2)
11302 {
11303 UPDATE_MIB(o1, ipIfStatsInHdrErrors, o2->ipIfStatsInHdrErrors);
11304 UPDATE_MIB(o1, ipIfStatsInTooBigErrors, o2->ipIfStatsInTooBigErrors);
11305 UPDATE_MIB(o1, ipIfStatsInNoRoutes, o2->ipIfStatsInNoRoutes);
11306 UPDATE_MIB(o1, ipIfStatsInAddrErrors, o2->ipIfStatsInAddrErrors);
11307 UPDATE_MIB(o1, ipIfStatsInUnknownProtos, o2->ipIfStatsInUnknownProtos);
11308 UPDATE_MIB(o1, ipIfStatsInTruncatedPkts, o2->ipIfStatsInTruncatedPkts);
11309 UPDATE_MIB(o1, ipIfStatsInDiscards, o2->ipIfStatsInDiscards);
11310 UPDATE_MIB(o1, ipIfStatsOutDiscards, o2->ipIfStatsOutDiscards);
11311 UPDATE_MIB(o1, ipIfStatsOutFragOKs, o2->ipIfStatsOutFragOKs);
11312 UPDATE_MIB(o1, ipIfStatsOutFragFails, o2->ipIfStatsOutFragFails);
11313 UPDATE_MIB(o1, ipIfStatsOutFragCreates, o2->ipIfStatsOutFragCreates);
11314 UPDATE_MIB(o1, ipIfStatsReasmReqds, o2->ipIfStatsReasmReqds);
11315 UPDATE_MIB(o1, ipIfStatsReasmOKs, o2->ipIfStatsReasmOKs);
11316 UPDATE_MIB(o1, ipIfStatsReasmFails, o2->ipIfStatsReasmFails);
11317 UPDATE_MIB(o1, ipIfStatsOutNoRoutes, o2->ipIfStatsOutNoRoutes);
11318 UPDATE_MIB(o1, ipIfStatsReasmDuplicates, o2->ipIfStatsReasmDuplicates);
11319 UPDATE_MIB(o1, ipIfStatsReasmPartDups, o2->ipIfStatsReasmPartDups);
11320 UPDATE_MIB(o1, ipIfStatsForwProhibits, o2->ipIfStatsForwProhibits);
11321 UPDATE_MIB(o1, udpInCksumErrs, o2->udpInCksumErrs);
11322 UPDATE_MIB(o1, udpInOverflows, o2->udpInOverflows);
11323 UPDATE_MIB(o1, rawipInOverflows, o2->rawipInOverflows);
11324 UPDATE_MIB(o1, ipIfStatsInWrongIPVersion,
11325 o2->ipIfStatsInWrongIPVersion);
11326 UPDATE_MIB(o1, ipIfStatsOutWrongIPVersion,
11327 o2->ipIfStatsInWrongIPVersion);
11328 UPDATE_MIB(o1, ipIfStatsOutSwitchIPVersion,
11329 o2->ipIfStatsOutSwitchIPVersion);
11330 UPDATE_MIB(o1, ipIfStatsHCInReceives, o2->ipIfStatsHCInReceives);
11331 UPDATE_MIB(o1, ipIfStatsHCInOctets, o2->ipIfStatsHCInOctets);
11332 UPDATE_MIB(o1, ipIfStatsHCInForwDatagrams,
11333 o2->ipIfStatsHCInForwDatagrams);
11334 UPDATE_MIB(o1, ipIfStatsHCInDelivers, o2->ipIfStatsHCInDelivers);
11335 UPDATE_MIB(o1, ipIfStatsHCOutRequests, o2->ipIfStatsHCOutRequests);
11336 UPDATE_MIB(o1, ipIfStatsHCOutForwDatagrams,
11337 o2->ipIfStatsHCOutForwDatagrams);
11338 UPDATE_MIB(o1, ipIfStatsOutFragReqds, o2->ipIfStatsOutFragReqds);
11339 UPDATE_MIB(o1, ipIfStatsHCOutTransmits, o2->ipIfStatsHCOutTransmits);
11340 UPDATE_MIB(o1, ipIfStatsHCOutOctets, o2->ipIfStatsHCOutOctets);
11341 UPDATE_MIB(o1, ipIfStatsHCInMcastPkts, o2->ipIfStatsHCInMcastPkts);
11342 UPDATE_MIB(o1, ipIfStatsHCInMcastOctets, o2->ipIfStatsHCInMcastOctets);
11343 UPDATE_MIB(o1, ipIfStatsHCOutMcastPkts, o2->ipIfStatsHCOutMcastPkts);
11344 UPDATE_MIB(o1, ipIfStatsHCOutMcastOctets,
11345 o2->ipIfStatsHCOutMcastOctets);
11346 UPDATE_MIB(o1, ipIfStatsHCInBcastPkts, o2->ipIfStatsHCInBcastPkts);
11347 UPDATE_MIB(o1, ipIfStatsHCOutBcastPkts, o2->ipIfStatsHCOutBcastPkts);
11348 UPDATE_MIB(o1, ipsecInSucceeded, o2->ipsecInSucceeded);
11349 UPDATE_MIB(o1, ipsecInFailed, o2->ipsecInFailed);
11350 UPDATE_MIB(o1, ipInCksumErrs, o2->ipInCksumErrs);
11351 UPDATE_MIB(o1, tcpInErrs, o2->tcpInErrs);
11352 UPDATE_MIB(o1, udpNoPorts, o2->udpNoPorts);
11353 }
11354
11355 void
11356 ip_mib2_add_icmp6_stats(mib2_ipv6IfIcmpEntry_t *o1, mib2_ipv6IfIcmpEntry_t *o2)
11357 {
11358 UPDATE_MIB(o1, ipv6IfIcmpInMsgs, o2->ipv6IfIcmpInMsgs);
11359 UPDATE_MIB(o1, ipv6IfIcmpInErrors, o2->ipv6IfIcmpInErrors);
11360 UPDATE_MIB(o1, ipv6IfIcmpInDestUnreachs, o2->ipv6IfIcmpInDestUnreachs);
11361 UPDATE_MIB(o1, ipv6IfIcmpInAdminProhibs, o2->ipv6IfIcmpInAdminProhibs);
11362 UPDATE_MIB(o1, ipv6IfIcmpInTimeExcds, o2->ipv6IfIcmpInTimeExcds);
11363 UPDATE_MIB(o1, ipv6IfIcmpInParmProblems, o2->ipv6IfIcmpInParmProblems);
11364 UPDATE_MIB(o1, ipv6IfIcmpInPktTooBigs, o2->ipv6IfIcmpInPktTooBigs);
11365 UPDATE_MIB(o1, ipv6IfIcmpInEchos, o2->ipv6IfIcmpInEchos);
11366 UPDATE_MIB(o1, ipv6IfIcmpInEchoReplies, o2->ipv6IfIcmpInEchoReplies);
11367 UPDATE_MIB(o1, ipv6IfIcmpInRouterSolicits,
11368 o2->ipv6IfIcmpInRouterSolicits);
11369 UPDATE_MIB(o1, ipv6IfIcmpInRouterAdvertisements,
11370 o2->ipv6IfIcmpInRouterAdvertisements);
11371 UPDATE_MIB(o1, ipv6IfIcmpInNeighborSolicits,
11372 o2->ipv6IfIcmpInNeighborSolicits);
11373 UPDATE_MIB(o1, ipv6IfIcmpInNeighborAdvertisements,
11374 o2->ipv6IfIcmpInNeighborAdvertisements);
11375 UPDATE_MIB(o1, ipv6IfIcmpInRedirects, o2->ipv6IfIcmpInRedirects);
11376 UPDATE_MIB(o1, ipv6IfIcmpInGroupMembQueries,
11377 o2->ipv6IfIcmpInGroupMembQueries);
11378 UPDATE_MIB(o1, ipv6IfIcmpInGroupMembResponses,
11379 o2->ipv6IfIcmpInGroupMembResponses);
11380 UPDATE_MIB(o1, ipv6IfIcmpInGroupMembReductions,
11381 o2->ipv6IfIcmpInGroupMembReductions);
11382 UPDATE_MIB(o1, ipv6IfIcmpOutMsgs, o2->ipv6IfIcmpOutMsgs);
11383 UPDATE_MIB(o1, ipv6IfIcmpOutErrors, o2->ipv6IfIcmpOutErrors);
11384 UPDATE_MIB(o1, ipv6IfIcmpOutDestUnreachs,
11385 o2->ipv6IfIcmpOutDestUnreachs);
11386 UPDATE_MIB(o1, ipv6IfIcmpOutAdminProhibs,
11387 o2->ipv6IfIcmpOutAdminProhibs);
11388 UPDATE_MIB(o1, ipv6IfIcmpOutTimeExcds, o2->ipv6IfIcmpOutTimeExcds);
11389 UPDATE_MIB(o1, ipv6IfIcmpOutParmProblems,
11390 o2->ipv6IfIcmpOutParmProblems);
11391 UPDATE_MIB(o1, ipv6IfIcmpOutPktTooBigs, o2->ipv6IfIcmpOutPktTooBigs);
11392 UPDATE_MIB(o1, ipv6IfIcmpOutEchos, o2->ipv6IfIcmpOutEchos);
11393 UPDATE_MIB(o1, ipv6IfIcmpOutEchoReplies, o2->ipv6IfIcmpOutEchoReplies);
11394 UPDATE_MIB(o1, ipv6IfIcmpOutRouterSolicits,
11395 o2->ipv6IfIcmpOutRouterSolicits);
11396 UPDATE_MIB(o1, ipv6IfIcmpOutRouterAdvertisements,
11397 o2->ipv6IfIcmpOutRouterAdvertisements);
11398 UPDATE_MIB(o1, ipv6IfIcmpOutNeighborSolicits,
11399 o2->ipv6IfIcmpOutNeighborSolicits);
11400 UPDATE_MIB(o1, ipv6IfIcmpOutNeighborAdvertisements,
11401 o2->ipv6IfIcmpOutNeighborAdvertisements);
11402 UPDATE_MIB(o1, ipv6IfIcmpOutRedirects, o2->ipv6IfIcmpOutRedirects);
11403 UPDATE_MIB(o1, ipv6IfIcmpOutGroupMembQueries,
11404 o2->ipv6IfIcmpOutGroupMembQueries);
11405 UPDATE_MIB(o1, ipv6IfIcmpOutGroupMembResponses,
11406 o2->ipv6IfIcmpOutGroupMembResponses);
11407 UPDATE_MIB(o1, ipv6IfIcmpOutGroupMembReductions,
11408 o2->ipv6IfIcmpOutGroupMembReductions);
11409 UPDATE_MIB(o1, ipv6IfIcmpInOverflows, o2->ipv6IfIcmpInOverflows);
11410 UPDATE_MIB(o1, ipv6IfIcmpBadHoplimit, o2->ipv6IfIcmpBadHoplimit);
11411 UPDATE_MIB(o1, ipv6IfIcmpInBadNeighborAdvertisements,
11412 o2->ipv6IfIcmpInBadNeighborAdvertisements);
11413 UPDATE_MIB(o1, ipv6IfIcmpInBadNeighborSolicitations,
11414 o2->ipv6IfIcmpInBadNeighborSolicitations);
11415 UPDATE_MIB(o1, ipv6IfIcmpInBadRedirects, o2->ipv6IfIcmpInBadRedirects);
11416 UPDATE_MIB(o1, ipv6IfIcmpInGroupMembTotal,
11417 o2->ipv6IfIcmpInGroupMembTotal);
11418 UPDATE_MIB(o1, ipv6IfIcmpInGroupMembBadQueries,
11419 o2->ipv6IfIcmpInGroupMembBadQueries);
11420 UPDATE_MIB(o1, ipv6IfIcmpInGroupMembBadReports,
11421 o2->ipv6IfIcmpInGroupMembBadReports);
11422 UPDATE_MIB(o1, ipv6IfIcmpInGroupMembOurReports,
11423 o2->ipv6IfIcmpInGroupMembOurReports);
11424 }
11425
11426 /*
11427 * Called before the options are updated to check if this packet will
11428 * be source routed from here.
11429 * This routine assumes that the options are well formed i.e. that they
11430 * have already been checked.
11431 */
11432 boolean_t
11433 ip_source_routed(ipha_t *ipha, ip_stack_t *ipst)
11434 {
11435 ipoptp_t opts;
11436 uchar_t *opt;
11437 uint8_t optval;
11438 uint8_t optlen;
11439 ipaddr_t dst;
11440
11441 if (IS_SIMPLE_IPH(ipha)) {
11442 ip2dbg(("not source routed\n"));
11443 return (B_FALSE);
11444 }
11445 dst = ipha->ipha_dst;
11446 for (optval = ipoptp_first(&opts, ipha);
11447 optval != IPOPT_EOL;
11448 optval = ipoptp_next(&opts)) {
11449 ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
11450 opt = opts.ipoptp_cur;
11451 optlen = opts.ipoptp_len;
11452 ip2dbg(("ip_source_routed: opt %d, len %d\n",
11453 optval, optlen));
11454 switch (optval) {
11455 uint32_t off;
11456 case IPOPT_SSRR:
11457 case IPOPT_LSRR:
11458 /*
11459 * If dst is one of our addresses and there are some
11460 * entries left in the source route return (true).
11461 */
11462 if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
11463 ip2dbg(("ip_source_routed: not next"
11464 " source route 0x%x\n",
11465 ntohl(dst)));
11466 return (B_FALSE);
11467 }
11468 off = opt[IPOPT_OFFSET];
11469 off--;
11470 if (optlen < IP_ADDR_LEN ||
11471 off > optlen - IP_ADDR_LEN) {
11472 /* End of source route */
11473 ip1dbg(("ip_source_routed: end of SR\n"));
11474 return (B_FALSE);
11475 }
11476 return (B_TRUE);
11477 }
11478 }
11479 ip2dbg(("not source routed\n"));
11480 return (B_FALSE);
11481 }
11482
11483 /*
11484 * ip_unbind is called by the transports to remove a conn from
11485 * the fanout table.
11486 */
11487 void
11488 ip_unbind(conn_t *connp)
11489 {
11490
11491 ASSERT(!MUTEX_HELD(&connp->conn_lock));
11492
11493 if (is_system_labeled() && connp->conn_anon_port) {
11494 (void) tsol_mlp_anon(crgetzone(connp->conn_cred),
11495 connp->conn_mlp_type, connp->conn_proto,
11496 ntohs(connp->conn_lport), B_FALSE);
11497 connp->conn_anon_port = 0;
11498 }
11499 connp->conn_mlp_type = mlptSingle;
11500
11501 ipcl_hash_remove(connp);
11502 }
11503
11504 /*
11505 * Used for deciding the MSS size for the upper layer. Thus
11506 * we need to check the outbound policy values in the conn.
11507 */
11508 int
11509 conn_ipsec_length(conn_t *connp)
11510 {
11511 ipsec_latch_t *ipl;
11512
11513 ipl = connp->conn_latch;
11514 if (ipl == NULL)
11515 return (0);
11516
11517 if (connp->conn_ixa->ixa_ipsec_policy == NULL)
11518 return (0);
11519
11520 return (connp->conn_ixa->ixa_ipsec_policy->ipsp_act->ipa_ovhd);
11521 }
11522
11523 /*
11524 * Returns an estimate of the IPsec headers size. This is used if
11525 * we don't want to call into IPsec to get the exact size.
11526 */
11527 int
11528 ipsec_out_extra_length(ip_xmit_attr_t *ixa)
11529 {
11530 ipsec_action_t *a;
11531
11532 if (!(ixa->ixa_flags & IXAF_IPSEC_SECURE))
11533 return (0);
11534
11535 a = ixa->ixa_ipsec_action;
11536 if (a == NULL) {
11537 ASSERT(ixa->ixa_ipsec_policy != NULL);
11538 a = ixa->ixa_ipsec_policy->ipsp_act;
11539 }
11540 ASSERT(a != NULL);
11541
11542 return (a->ipa_ovhd);
11543 }
11544
11545 /*
11546 * If there are any source route options, return the true final
11547 * destination. Otherwise, return the destination.
11548 */
11549 ipaddr_t
11550 ip_get_dst(ipha_t *ipha)
11551 {
11552 ipoptp_t opts;
11553 uchar_t *opt;
11554 uint8_t optval;
11555 uint8_t optlen;
11556 ipaddr_t dst;
11557 uint32_t off;
11558
11559 dst = ipha->ipha_dst;
11560
11561 if (IS_SIMPLE_IPH(ipha))
11562 return (dst);
11563
11564 for (optval = ipoptp_first(&opts, ipha);
11565 optval != IPOPT_EOL;
11566 optval = ipoptp_next(&opts)) {
11567 opt = opts.ipoptp_cur;
11568 optlen = opts.ipoptp_len;
11569 ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
11570 switch (optval) {
11571 case IPOPT_SSRR:
11572 case IPOPT_LSRR:
11573 off = opt[IPOPT_OFFSET];
11574 /*
11575 * If one of the conditions is true, it means
11576 * end of options and dst already has the right
11577 * value.
11578 */
11579 if (!(optlen < IP_ADDR_LEN || off > optlen - 3)) {
11580 off = optlen - IP_ADDR_LEN;
11581 bcopy(&opt[off], &dst, IP_ADDR_LEN);
11582 }
11583 return (dst);
11584 default:
11585 break;
11586 }
11587 }
11588
11589 return (dst);
11590 }
11591
11592 /*
11593 * Outbound IP fragmentation routine.
11594 * Assumes the caller has checked whether or not fragmentation should
11595 * be allowed. Here we copy the DF bit from the header to all the generated
11596 * fragments.
11597 */
11598 int
11599 ip_fragment_v4(mblk_t *mp_orig, nce_t *nce, iaflags_t ixaflags,
11600 uint_t pkt_len, uint32_t max_frag, uint32_t xmit_hint, zoneid_t szone,
11601 zoneid_t nolzid, pfirepostfrag_t postfragfn, uintptr_t *ixa_cookie)
11602 {
11603 int i1;
11604 int hdr_len;
11605 mblk_t *hdr_mp;
11606 ipha_t *ipha;
11607 int ip_data_end;
11608 int len;
11609 mblk_t *mp = mp_orig;
11610 int offset;
11611 ill_t *ill = nce->nce_ill;
11612 ip_stack_t *ipst = ill->ill_ipst;
11613 mblk_t *carve_mp;
11614 uint32_t frag_flag;
11615 uint_t priority = mp->b_band;
11616 int error = 0;
11617
11618 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragReqds);
11619
11620 if (pkt_len != msgdsize(mp)) {
11621 ip0dbg(("Packet length mismatch: %d, %ld\n",
11622 pkt_len, msgdsize(mp)));
11623 freemsg(mp);
11624 return (EINVAL);
11625 }
11626
11627 if (max_frag == 0) {
11628 ip1dbg(("ip_fragment_v4: max_frag is zero. Dropping packet\n"));
11629 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11630 ip_drop_output("FragFails: zero max_frag", mp, ill);
11631 freemsg(mp);
11632 return (EINVAL);
11633 }
11634
11635 ASSERT(MBLKL(mp) >= sizeof (ipha_t));
11636 ipha = (ipha_t *)mp->b_rptr;
11637 ASSERT(ntohs(ipha->ipha_length) == pkt_len);
11638 frag_flag = ntohs(ipha->ipha_fragment_offset_and_flags) & IPH_DF;
11639
11640 /*
11641 * Establish the starting offset. May not be zero if we are fragging
11642 * a fragment that is being forwarded.
11643 */
11644 offset = ntohs(ipha->ipha_fragment_offset_and_flags) & IPH_OFFSET;
11645
11646 /* TODO why is this test needed? */
11647 if (((max_frag - ntohs(ipha->ipha_length)) & ~7) < 8) {
11648 /* TODO: notify ulp somehow */
11649 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11650 ip_drop_output("FragFails: bad starting offset", mp, ill);
11651 freemsg(mp);
11652 return (EINVAL);
11653 }
11654
11655 hdr_len = IPH_HDR_LENGTH(ipha);
11656 ipha->ipha_hdr_checksum = 0;
11657
11658 /*
11659 * Establish the number of bytes maximum per frag, after putting
11660 * in the header.
11661 */
11662 len = (max_frag - hdr_len) & ~7;
11663
11664 /* Get a copy of the header for the trailing frags */
11665 hdr_mp = ip_fragment_copyhdr((uchar_t *)ipha, hdr_len, offset, ipst,
11666 mp);
11667 if (hdr_mp == NULL) {
11668 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11669 ip_drop_output("FragFails: no hdr_mp", mp, ill);
11670 freemsg(mp);
11671 return (ENOBUFS);
11672 }
11673
11674 /* Store the starting offset, with the MoreFrags flag. */
11675 i1 = offset | IPH_MF | frag_flag;
11676 ipha->ipha_fragment_offset_and_flags = htons((uint16_t)i1);
11677
11678 /* Establish the ending byte offset, based on the starting offset. */
11679 offset <<= 3;
11680 ip_data_end = offset + ntohs(ipha->ipha_length) - hdr_len;
11681
11682 /* Store the length of the first fragment in the IP header. */
11683 i1 = len + hdr_len;
11684 ASSERT(i1 <= IP_MAXPACKET);
11685 ipha->ipha_length = htons((uint16_t)i1);
11686
11687 /*
11688 * Compute the IP header checksum for the first frag. We have to
11689 * watch out that we stop at the end of the header.
11690 */
11691 ipha->ipha_hdr_checksum = ip_csum_hdr(ipha);
11692
11693 /*
11694 * Now carve off the first frag. Note that this will include the
11695 * original IP header.
11696 */
11697 if (!(mp = ip_carve_mp(&mp_orig, i1))) {
11698 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11699 ip_drop_output("FragFails: could not carve mp", mp_orig, ill);
11700 freeb(hdr_mp);
11701 freemsg(mp_orig);
11702 return (ENOBUFS);
11703 }
11704
11705 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragCreates);
11706
11707 error = postfragfn(mp, nce, ixaflags, i1, xmit_hint, szone, nolzid,
11708 ixa_cookie);
11709 if (error != 0 && error != EWOULDBLOCK) {
11710 /* No point in sending the other fragments */
11711 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11712 ip_drop_output("FragFails: postfragfn failed", mp_orig, ill);
11713 freeb(hdr_mp);
11714 freemsg(mp_orig);
11715 return (error);
11716 }
11717
11718 /* No need to redo state machine in loop */
11719 ixaflags &= ~IXAF_REACH_CONF;
11720
11721 /* Advance the offset to the second frag starting point. */
11722 offset += len;
11723 /*
11724 * Update hdr_len from the copied header - there might be less options
11725 * in the later fragments.
11726 */
11727 hdr_len = IPH_HDR_LENGTH(hdr_mp->b_rptr);
11728 /* Loop until done. */
11729 for (;;) {
11730 uint16_t offset_and_flags;
11731 uint16_t ip_len;
11732
11733 if (ip_data_end - offset > len) {
11734 /*
11735 * Carve off the appropriate amount from the original
11736 * datagram.
11737 */
11738 if (!(carve_mp = ip_carve_mp(&mp_orig, len))) {
11739 mp = NULL;
11740 break;
11741 }
11742 /*
11743 * More frags after this one. Get another copy
11744 * of the header.
11745 */
11746 if (carve_mp->b_datap->db_ref == 1 &&
11747 hdr_mp->b_wptr - hdr_mp->b_rptr <
11748 carve_mp->b_rptr - carve_mp->b_datap->db_base) {
11749 /* Inline IP header */
11750 carve_mp->b_rptr -= hdr_mp->b_wptr -
11751 hdr_mp->b_rptr;
11752 bcopy(hdr_mp->b_rptr, carve_mp->b_rptr,
11753 hdr_mp->b_wptr - hdr_mp->b_rptr);
11754 mp = carve_mp;
11755 } else {
11756 if (!(mp = copyb(hdr_mp))) {
11757 freemsg(carve_mp);
11758 break;
11759 }
11760 /* Get priority marking, if any. */
11761 mp->b_band = priority;
11762 mp->b_cont = carve_mp;
11763 }
11764 ipha = (ipha_t *)mp->b_rptr;
11765 offset_and_flags = IPH_MF;
11766 } else {
11767 /*
11768 * Last frag. Consume the header. Set len to
11769 * the length of this last piece.
11770 */
11771 len = ip_data_end - offset;
11772
11773 /*
11774 * Carve off the appropriate amount from the original
11775 * datagram.
11776 */
11777 if (!(carve_mp = ip_carve_mp(&mp_orig, len))) {
11778 mp = NULL;
11779 break;
11780 }
11781 if (carve_mp->b_datap->db_ref == 1 &&
11782 hdr_mp->b_wptr - hdr_mp->b_rptr <
11783 carve_mp->b_rptr - carve_mp->b_datap->db_base) {
11784 /* Inline IP header */
11785 carve_mp->b_rptr -= hdr_mp->b_wptr -
11786 hdr_mp->b_rptr;
11787 bcopy(hdr_mp->b_rptr, carve_mp->b_rptr,
11788 hdr_mp->b_wptr - hdr_mp->b_rptr);
11789 mp = carve_mp;
11790 freeb(hdr_mp);
11791 hdr_mp = mp;
11792 } else {
11793 mp = hdr_mp;
11794 /* Get priority marking, if any. */
11795 mp->b_band = priority;
11796 mp->b_cont = carve_mp;
11797 }
11798 ipha = (ipha_t *)mp->b_rptr;
11799 /* A frag of a frag might have IPH_MF non-zero */
11800 offset_and_flags =
11801 ntohs(ipha->ipha_fragment_offset_and_flags) &
11802 IPH_MF;
11803 }
11804 offset_and_flags |= (uint16_t)(offset >> 3);
11805 offset_and_flags |= (uint16_t)frag_flag;
11806 /* Store the offset and flags in the IP header. */
11807 ipha->ipha_fragment_offset_and_flags = htons(offset_and_flags);
11808
11809 /* Store the length in the IP header. */
11810 ip_len = (uint16_t)(len + hdr_len);
11811 ipha->ipha_length = htons(ip_len);
11812
11813 /*
11814 * Set the IP header checksum. Note that mp is just
11815 * the header, so this is easy to pass to ip_csum.
11816 */
11817 ipha->ipha_hdr_checksum = ip_csum_hdr(ipha);
11818
11819 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragCreates);
11820
11821 error = postfragfn(mp, nce, ixaflags, ip_len, xmit_hint, szone,
11822 nolzid, ixa_cookie);
11823 /* All done if we just consumed the hdr_mp. */
11824 if (mp == hdr_mp) {
11825 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragOKs);
11826 return (error);
11827 }
11828 if (error != 0 && error != EWOULDBLOCK) {
11829 DTRACE_PROBE2(ip__xmit__frag__fail, ill_t *, ill,
11830 mblk_t *, hdr_mp);
11831 /* No point in sending the other fragments */
11832 break;
11833 }
11834
11835 /* Otherwise, advance and loop. */
11836 offset += len;
11837 }
11838 /* Clean up following allocation failure. */
11839 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11840 ip_drop_output("FragFails: loop ended", NULL, ill);
11841 if (mp != hdr_mp)
11842 freeb(hdr_mp);
11843 if (mp != mp_orig)
11844 freemsg(mp_orig);
11845 return (error);
11846 }
11847
11848 /*
11849 * Copy the header plus those options which have the copy bit set
11850 */
11851 static mblk_t *
11852 ip_fragment_copyhdr(uchar_t *rptr, int hdr_len, int offset, ip_stack_t *ipst,
11853 mblk_t *src)
11854 {
11855 mblk_t *mp;
11856 uchar_t *up;
11857
11858 /*
11859 * Quick check if we need to look for options without the copy bit
11860 * set
11861 */
11862 mp = allocb_tmpl(ipst->ips_ip_wroff_extra + hdr_len, src);
11863 if (!mp)
11864 return (mp);
11865 mp->b_rptr += ipst->ips_ip_wroff_extra;
11866 if (hdr_len == IP_SIMPLE_HDR_LENGTH || offset != 0) {
11867 bcopy(rptr, mp->b_rptr, hdr_len);
11868 mp->b_wptr += hdr_len + ipst->ips_ip_wroff_extra;
11869 return (mp);
11870 }
11871 up = mp->b_rptr;
11872 bcopy(rptr, up, IP_SIMPLE_HDR_LENGTH);
11873 up += IP_SIMPLE_HDR_LENGTH;
11874 rptr += IP_SIMPLE_HDR_LENGTH;
11875 hdr_len -= IP_SIMPLE_HDR_LENGTH;
11876 while (hdr_len > 0) {
11877 uint32_t optval;
11878 uint32_t optlen;
11879
11880 optval = *rptr;
11881 if (optval == IPOPT_EOL)
11882 break;
11883 if (optval == IPOPT_NOP)
11884 optlen = 1;
11885 else
11886 optlen = rptr[1];
11887 if (optval & IPOPT_COPY) {
11888 bcopy(rptr, up, optlen);
11889 up += optlen;
11890 }
11891 rptr += optlen;
11892 hdr_len -= optlen;
11893 }
11894 /*
11895 * Make sure that we drop an even number of words by filling
11896 * with EOL to the next word boundary.
11897 */
11898 for (hdr_len = up - (mp->b_rptr + IP_SIMPLE_HDR_LENGTH);
11899 hdr_len & 0x3; hdr_len++)
11900 *up++ = IPOPT_EOL;
11901 mp->b_wptr = up;
11902 /* Update header length */
11903 mp->b_rptr[0] = (uint8_t)((IP_VERSION << 4) | ((up - mp->b_rptr) >> 2));
11904 return (mp);
11905 }
11906
11907 /*
11908 * Update any source route, record route, or timestamp options when
11909 * sending a packet back to ourselves.
11910 * Check that we are at end of strict source route.
11911 * The options have been sanity checked by ip_output_options().
11912 */
11913 void
11914 ip_output_local_options(ipha_t *ipha, ip_stack_t *ipst)
11915 {
11916 ipoptp_t opts;
11917 uchar_t *opt;
11918 uint8_t optval;
11919 uint8_t optlen;
11920 ipaddr_t dst;
11921 uint32_t ts;
11922 timestruc_t now;
11923
11924 for (optval = ipoptp_first(&opts, ipha);
11925 optval != IPOPT_EOL;
11926 optval = ipoptp_next(&opts)) {
11927 opt = opts.ipoptp_cur;
11928 optlen = opts.ipoptp_len;
11929 ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
11930 switch (optval) {
11931 uint32_t off;
11932 case IPOPT_SSRR:
11933 case IPOPT_LSRR:
11934 off = opt[IPOPT_OFFSET];
11935 off--;
11936 if (optlen < IP_ADDR_LEN ||
11937 off > optlen - IP_ADDR_LEN) {
11938 /* End of source route */
11939 break;
11940 }
11941 /*
11942 * This will only happen if two consecutive entries
11943 * in the source route contains our address or if
11944 * it is a packet with a loose source route which
11945 * reaches us before consuming the whole source route
11946 */
11947
11948 if (optval == IPOPT_SSRR) {
11949 return;
11950 }
11951 /*
11952 * Hack: instead of dropping the packet truncate the
11953 * source route to what has been used by filling the
11954 * rest with IPOPT_NOP.
11955 */
11956 opt[IPOPT_OLEN] = (uint8_t)off;
11957 while (off < optlen) {
11958 opt[off++] = IPOPT_NOP;
11959 }
11960 break;
11961 case IPOPT_RR:
11962 off = opt[IPOPT_OFFSET];
11963 off--;
11964 if (optlen < IP_ADDR_LEN ||
11965 off > optlen - IP_ADDR_LEN) {
11966 /* No more room - ignore */
11967 ip1dbg((
11968 "ip_output_local_options: end of RR\n"));
11969 break;
11970 }
11971 dst = htonl(INADDR_LOOPBACK);
11972 bcopy(&dst, (char *)opt + off, IP_ADDR_LEN);
11973 opt[IPOPT_OFFSET] += IP_ADDR_LEN;
11974 break;
11975 case IPOPT_TS:
11976 /* Insert timestamp if there is romm */
11977 switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
11978 case IPOPT_TS_TSONLY:
11979 off = IPOPT_TS_TIMELEN;
11980 break;
11981 case IPOPT_TS_PRESPEC:
11982 case IPOPT_TS_PRESPEC_RFC791:
11983 /* Verify that the address matched */
11984 off = opt[IPOPT_OFFSET] - 1;
11985 bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
11986 if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
11987 /* Not for us */
11988 break;
11989 }
11990 /* FALLTHRU */
11991 case IPOPT_TS_TSANDADDR:
11992 off = IP_ADDR_LEN + IPOPT_TS_TIMELEN;
11993 break;
11994 default:
11995 /*
11996 * ip_*put_options should have already
11997 * dropped this packet.
11998 */
11999 cmn_err(CE_PANIC, "ip_output_local_options: "
12000 "unknown IT - bug in ip_output_options?\n");
12001 return; /* Keep "lint" happy */
12002 }
12003 if (opt[IPOPT_OFFSET] - 1 + off > optlen) {
12004 /* Increase overflow counter */
12005 off = (opt[IPOPT_POS_OV_FLG] >> 4) + 1;
12006 opt[IPOPT_POS_OV_FLG] = (uint8_t)
12007 (opt[IPOPT_POS_OV_FLG] & 0x0F) |
12008 (off << 4);
12009 break;
12010 }
12011 off = opt[IPOPT_OFFSET] - 1;
12012 switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
12013 case IPOPT_TS_PRESPEC:
12014 case IPOPT_TS_PRESPEC_RFC791:
12015 case IPOPT_TS_TSANDADDR:
12016 dst = htonl(INADDR_LOOPBACK);
12017 bcopy(&dst, (char *)opt + off, IP_ADDR_LEN);
12018 opt[IPOPT_OFFSET] += IP_ADDR_LEN;
12019 /* FALLTHRU */
12020 case IPOPT_TS_TSONLY:
12021 off = opt[IPOPT_OFFSET] - 1;
12022 /* Compute # of milliseconds since midnight */
12023 gethrestime(&now);
12024 ts = (now.tv_sec % (24 * 60 * 60)) * 1000 +
12025 now.tv_nsec / (NANOSEC / MILLISEC);
12026 bcopy(&ts, (char *)opt + off, IPOPT_TS_TIMELEN);
12027 opt[IPOPT_OFFSET] += IPOPT_TS_TIMELEN;
12028 break;
12029 }
12030 break;
12031 }
12032 }
12033 }
12034
12035 /*
12036 * Prepend an M_DATA fastpath header, and if none present prepend a
12037 * DL_UNITDATA_REQ. Frees the mblk on failure.
12038 *
12039 * nce_dlur_mp and nce_fp_mp can not disappear once they have been set.
12040 * If there is a change to them, the nce will be deleted (condemned) and
12041 * a new nce_t will be created when packets are sent. Thus we need no locks
12042 * to access those fields.
12043 *
12044 * We preserve b_band to support IPQoS. If a DL_UNITDATA_REQ is prepended
12045 * we place b_band in dl_priority.dl_max.
12046 */
12047 static mblk_t *
12048 ip_xmit_attach_llhdr(mblk_t *mp, nce_t *nce)
12049 {
12050 uint_t hlen;
12051 mblk_t *mp1;
12052 uint_t priority;
12053 uchar_t *rptr;
12054
12055 rptr = mp->b_rptr;
12056
12057 ASSERT(DB_TYPE(mp) == M_DATA);
12058 priority = mp->b_band;
12059
12060 ASSERT(nce != NULL);
12061 if ((mp1 = nce->nce_fp_mp) != NULL) {
12062 hlen = MBLKL(mp1);
12063 /*
12064 * Check if we have enough room to prepend fastpath
12065 * header
12066 */
12067 if (hlen != 0 && (rptr - mp->b_datap->db_base) >= hlen) {
12068 rptr -= hlen;
12069 bcopy(mp1->b_rptr, rptr, hlen);
12070 /*
12071 * Set the b_rptr to the start of the link layer
12072 * header
12073 */
12074 mp->b_rptr = rptr;
12075 return (mp);
12076 }
12077 mp1 = copyb(mp1);
12078 if (mp1 == NULL) {
12079 ill_t *ill = nce->nce_ill;
12080
12081 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
12082 ip_drop_output("ipIfStatsOutDiscards", mp, ill);
12083 freemsg(mp);
12084 return (NULL);
12085 }
12086 mp1->b_band = priority;
12087 mp1->b_cont = mp;
12088 DB_CKSUMSTART(mp1) = DB_CKSUMSTART(mp);
12089 DB_CKSUMSTUFF(mp1) = DB_CKSUMSTUFF(mp);
12090 DB_CKSUMEND(mp1) = DB_CKSUMEND(mp);
12091 DB_CKSUMFLAGS(mp1) = DB_CKSUMFLAGS(mp);
12092 DB_LSOMSS(mp1) = DB_LSOMSS(mp);
12093 DTRACE_PROBE1(ip__xmit__copyb, (mblk_t *), mp1);
12094 /*
12095 * XXX disable ICK_VALID and compute checksum
12096 * here; can happen if nce_fp_mp changes and
12097 * it can't be copied now due to insufficient
12098 * space. (unlikely, fp mp can change, but it
12099 * does not increase in length)
12100 */
12101 return (mp1);
12102 }
12103 mp1 = copyb(nce->nce_dlur_mp);
12104
12105 if (mp1 == NULL) {
12106 ill_t *ill = nce->nce_ill;
12107
12108 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
12109 ip_drop_output("ipIfStatsOutDiscards", mp, ill);
12110 freemsg(mp);
12111 return (NULL);
12112 }
12113 mp1->b_cont = mp;
12114 if (priority != 0) {
12115 mp1->b_band = priority;
12116 ((dl_unitdata_req_t *)(mp1->b_rptr))->dl_priority.dl_max =
12117 priority;
12118 }
12119 return (mp1);
12120 #undef rptr
12121 }
12122
12123 /*
12124 * Finish the outbound IPsec processing. This function is called from
12125 * ipsec_out_process() if the IPsec packet was processed
12126 * synchronously, or from {ah,esp}_kcf_callback_outbound() if it was processed
12127 * asynchronously.
12128 *
12129 * This is common to IPv4 and IPv6.
12130 */
12131 int
12132 ip_output_post_ipsec(mblk_t *mp, ip_xmit_attr_t *ixa)
12133 {
12134 iaflags_t ixaflags = ixa->ixa_flags;
12135 uint_t pktlen;
12136
12137
12138 /* AH/ESP don't update ixa_pktlen when they modify the packet */
12139 if (ixaflags & IXAF_IS_IPV4) {
12140 ipha_t *ipha = (ipha_t *)mp->b_rptr;
12141
12142 ASSERT(IPH_HDR_VERSION(ipha) == IPV4_VERSION);
12143 pktlen = ntohs(ipha->ipha_length);
12144 } else {
12145 ip6_t *ip6h = (ip6_t *)mp->b_rptr;
12146
12147 ASSERT(IPH_HDR_VERSION(mp->b_rptr) == IPV6_VERSION);
12148 pktlen = ntohs(ip6h->ip6_plen) + IPV6_HDR_LEN;
12149 }
12150
12151 /*
12152 * We release any hard reference on the SAs here to make
12153 * sure the SAs can be garbage collected. ipsr_sa has a soft reference
12154 * on the SAs.
12155 * If in the future we want the hard latching of the SAs in the
12156 * ip_xmit_attr_t then we should remove this.
12157 */
12158 if (ixa->ixa_ipsec_esp_sa != NULL) {
12159 IPSA_REFRELE(ixa->ixa_ipsec_esp_sa);
12160 ixa->ixa_ipsec_esp_sa = NULL;
12161 }
12162 if (ixa->ixa_ipsec_ah_sa != NULL) {
12163 IPSA_REFRELE(ixa->ixa_ipsec_ah_sa);
12164 ixa->ixa_ipsec_ah_sa = NULL;
12165 }
12166
12167 /* Do we need to fragment? */
12168 if ((ixa->ixa_flags & IXAF_IPV6_ADD_FRAGHDR) ||
12169 pktlen > ixa->ixa_fragsize) {
12170 if (ixaflags & IXAF_IS_IPV4) {
12171 ASSERT(!(ixa->ixa_flags & IXAF_IPV6_ADD_FRAGHDR));
12172 /*
12173 * We check for the DF case in ipsec_out_process
12174 * hence this only handles the non-DF case.
12175 */
12176 return (ip_fragment_v4(mp, ixa->ixa_nce, ixa->ixa_flags,
12177 pktlen, ixa->ixa_fragsize,
12178 ixa->ixa_xmit_hint, ixa->ixa_zoneid,
12179 ixa->ixa_no_loop_zoneid, ixa->ixa_postfragfn,
12180 &ixa->ixa_cookie));
12181 } else {
12182 mp = ip_fraghdr_add_v6(mp, ixa->ixa_ident, ixa);
12183 if (mp == NULL) {
12184 /* MIB and ip_drop_output already done */
12185 return (ENOMEM);
12186 }
12187 pktlen += sizeof (ip6_frag_t);
12188 if (pktlen > ixa->ixa_fragsize) {
12189 return (ip_fragment_v6(mp, ixa->ixa_nce,
12190 ixa->ixa_flags, pktlen,
12191 ixa->ixa_fragsize, ixa->ixa_xmit_hint,
12192 ixa->ixa_zoneid, ixa->ixa_no_loop_zoneid,
12193 ixa->ixa_postfragfn, &ixa->ixa_cookie));
12194 }
12195 }
12196 }
12197 return ((ixa->ixa_postfragfn)(mp, ixa->ixa_nce, ixa->ixa_flags,
12198 pktlen, ixa->ixa_xmit_hint, ixa->ixa_zoneid,
12199 ixa->ixa_no_loop_zoneid, NULL));
12200 }
12201
12202 /*
12203 * Finish the inbound IPsec processing. This function is called from
12204 * ipsec_out_process() if the IPsec packet was processed
12205 * synchronously, or from {ah,esp}_kcf_callback_outbound() if it was processed
12206 * asynchronously.
12207 *
12208 * This is common to IPv4 and IPv6.
12209 */
12210 void
12211 ip_input_post_ipsec(mblk_t *mp, ip_recv_attr_t *ira)
12212 {
12213 iaflags_t iraflags = ira->ira_flags;
12214
12215 /* Length might have changed */
12216 if (iraflags & IRAF_IS_IPV4) {
12217 ipha_t *ipha = (ipha_t *)mp->b_rptr;
12218
12219 ASSERT(IPH_HDR_VERSION(ipha) == IPV4_VERSION);
12220 ira->ira_pktlen = ntohs(ipha->ipha_length);
12221 ira->ira_ip_hdr_length = IPH_HDR_LENGTH(ipha);
12222 ira->ira_protocol = ipha->ipha_protocol;
12223
12224 ip_fanout_v4(mp, ipha, ira);
12225 } else {
12226 ip6_t *ip6h = (ip6_t *)mp->b_rptr;
12227 uint8_t *nexthdrp;
12228
12229 ASSERT(IPH_HDR_VERSION(mp->b_rptr) == IPV6_VERSION);
12230 ira->ira_pktlen = ntohs(ip6h->ip6_plen) + IPV6_HDR_LEN;
12231 if (!ip_hdr_length_nexthdr_v6(mp, ip6h, &ira->ira_ip_hdr_length,
12232 &nexthdrp)) {
12233 /* Malformed packet */
12234 BUMP_MIB(ira->ira_ill->ill_ip_mib, ipIfStatsInDiscards);
12235 ip_drop_input("ipIfStatsInDiscards", mp, ira->ira_ill);
12236 freemsg(mp);
12237 return;
12238 }
12239 ira->ira_protocol = *nexthdrp;
12240 ip_fanout_v6(mp, ip6h, ira);
12241 }
12242 }
12243
12244 /*
12245 * Select which AH & ESP SA's to use (if any) for the outbound packet.
12246 *
12247 * If this function returns B_TRUE, the requested SA's have been filled
12248 * into the ixa_ipsec_*_sa pointers.
12249 *
12250 * If the function returns B_FALSE, the packet has been "consumed", most
12251 * likely by an ACQUIRE sent up via PF_KEY to a key management daemon.
12252 *
12253 * The SA references created by the protocol-specific "select"
12254 * function will be released in ip_output_post_ipsec.
12255 */
12256 static boolean_t
12257 ipsec_out_select_sa(mblk_t *mp, ip_xmit_attr_t *ixa)
12258 {
12259 boolean_t need_ah_acquire = B_FALSE, need_esp_acquire = B_FALSE;
12260 ipsec_policy_t *pp;
12261 ipsec_action_t *ap;
12262
12263 ASSERT(ixa->ixa_flags & IXAF_IPSEC_SECURE);
12264 ASSERT((ixa->ixa_ipsec_policy != NULL) ||
12265 (ixa->ixa_ipsec_action != NULL));
12266
12267 ap = ixa->ixa_ipsec_action;
12268 if (ap == NULL) {
12269 pp = ixa->ixa_ipsec_policy;
12270 ASSERT(pp != NULL);
12271 ap = pp->ipsp_act;
12272 ASSERT(ap != NULL);
12273 }
12274
12275 /*
12276 * We have an action. now, let's select SA's.
12277 * A side effect of setting ixa_ipsec_*_sa is that it will
12278 * be cached in the conn_t.
12279 */
12280 if (ap->ipa_want_esp) {
12281 if (ixa->ixa_ipsec_esp_sa == NULL) {
12282 need_esp_acquire = !ipsec_outbound_sa(mp, ixa,
12283 IPPROTO_ESP);
12284 }
12285 ASSERT(need_esp_acquire || ixa->ixa_ipsec_esp_sa != NULL);
12286 }
12287
12288 if (ap->ipa_want_ah) {
12289 if (ixa->ixa_ipsec_ah_sa == NULL) {
12290 need_ah_acquire = !ipsec_outbound_sa(mp, ixa,
12291 IPPROTO_AH);
12292 }
12293 ASSERT(need_ah_acquire || ixa->ixa_ipsec_ah_sa != NULL);
12294 /*
12295 * The ESP and AH processing order needs to be preserved
12296 * when both protocols are required (ESP should be applied
12297 * before AH for an outbound packet). Force an ESP ACQUIRE
12298 * when both ESP and AH are required, and an AH ACQUIRE
12299 * is needed.
12300 */
12301 if (ap->ipa_want_esp && need_ah_acquire)
12302 need_esp_acquire = B_TRUE;
12303 }
12304
12305 /*
12306 * Send an ACQUIRE (extended, regular, or both) if we need one.
12307 * Release SAs that got referenced, but will not be used until we
12308 * acquire _all_ of the SAs we need.
12309 */
12310 if (need_ah_acquire || need_esp_acquire) {
12311 if (ixa->ixa_ipsec_ah_sa != NULL) {
12312 IPSA_REFRELE(ixa->ixa_ipsec_ah_sa);
12313 ixa->ixa_ipsec_ah_sa = NULL;
12314 }
12315 if (ixa->ixa_ipsec_esp_sa != NULL) {
12316 IPSA_REFRELE(ixa->ixa_ipsec_esp_sa);
12317 ixa->ixa_ipsec_esp_sa = NULL;
12318 }
12319
12320 sadb_acquire(mp, ixa, need_ah_acquire, need_esp_acquire);
12321 return (B_FALSE);
12322 }
12323
12324 return (B_TRUE);
12325 }
12326
12327 /*
12328 * Handle IPsec output processing.
12329 * This function is only entered once for a given packet.
12330 * We try to do things synchronously, but if we need to have user-level
12331 * set up SAs, or ESP or AH uses asynchronous kEF, then the operation
12332 * will be completed
12333 * - when the SAs are added in esp_add_sa_finish/ah_add_sa_finish
12334 * - when asynchronous ESP is done it will do AH
12335 *
12336 * In all cases we come back in ip_output_post_ipsec() to fragment and
12337 * send out the packet.
12338 */
12339 int
12340 ipsec_out_process(mblk_t *mp, ip_xmit_attr_t *ixa)
12341 {
12342 ill_t *ill = ixa->ixa_nce->nce_ill;
12343 ip_stack_t *ipst = ixa->ixa_ipst;
12344 ipsec_stack_t *ipss;
12345 ipsec_policy_t *pp;
12346 ipsec_action_t *ap;
12347
12348 ASSERT(ixa->ixa_flags & IXAF_IPSEC_SECURE);
12349
12350 ASSERT((ixa->ixa_ipsec_policy != NULL) ||
12351 (ixa->ixa_ipsec_action != NULL));
12352
12353 ipss = ipst->ips_netstack->netstack_ipsec;
12354 if (!ipsec_loaded(ipss)) {
12355 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
12356 ip_drop_packet(mp, B_TRUE, ill,
12357 DROPPER(ipss, ipds_ip_ipsec_not_loaded),
12358 &ipss->ipsec_dropper);
12359 return (ENOTSUP);
12360 }
12361
12362 ap = ixa->ixa_ipsec_action;
12363 if (ap == NULL) {
12364 pp = ixa->ixa_ipsec_policy;
12365 ASSERT(pp != NULL);
12366 ap = pp->ipsp_act;
12367 ASSERT(ap != NULL);
12368 }
12369
12370 /* Handle explicit drop action and bypass. */
12371 switch (ap->ipa_act.ipa_type) {
12372 case IPSEC_ACT_DISCARD:
12373 case IPSEC_ACT_REJECT:
12374 ip_drop_packet(mp, B_FALSE, ill,
12375 DROPPER(ipss, ipds_spd_explicit), &ipss->ipsec_spd_dropper);
12376 return (EHOSTUNREACH); /* IPsec policy failure */
12377 case IPSEC_ACT_BYPASS:
12378 return (ip_output_post_ipsec(mp, ixa));
12379 }
12380
12381 /*
12382 * The order of processing is first insert a IP header if needed.
12383 * Then insert the ESP header and then the AH header.
12384 */
12385 if ((ixa->ixa_flags & IXAF_IS_IPV4) && ap->ipa_want_se) {
12386 /*
12387 * First get the outer IP header before sending
12388 * it to ESP.
12389 */
12390 ipha_t *oipha, *iipha;
12391 mblk_t *outer_mp, *inner_mp;
12392
12393 if ((outer_mp = allocb(sizeof (ipha_t), BPRI_HI)) == NULL) {
12394 (void) mi_strlog(ill->ill_rq, 0,
12395 SL_ERROR|SL_TRACE|SL_CONSOLE,
12396 "ipsec_out_process: "
12397 "Self-Encapsulation failed: Out of memory\n");
12398 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
12399 ip_drop_output("ipIfStatsOutDiscards", mp, ill);
12400 freemsg(mp);
12401 return (ENOBUFS);
12402 }
12403 inner_mp = mp;
12404 ASSERT(inner_mp->b_datap->db_type == M_DATA);
12405 oipha = (ipha_t *)outer_mp->b_rptr;
12406 iipha = (ipha_t *)inner_mp->b_rptr;
12407 *oipha = *iipha;
12408 outer_mp->b_wptr += sizeof (ipha_t);
12409 oipha->ipha_length = htons(ntohs(iipha->ipha_length) +
12410 sizeof (ipha_t));
12411 oipha->ipha_protocol = IPPROTO_ENCAP;
12412 oipha->ipha_version_and_hdr_length =
12413 IP_SIMPLE_HDR_VERSION;
12414 oipha->ipha_hdr_checksum = 0;
12415 oipha->ipha_hdr_checksum = ip_csum_hdr(oipha);
12416 outer_mp->b_cont = inner_mp;
12417 mp = outer_mp;
12418
12419 ixa->ixa_flags |= IXAF_IPSEC_TUNNEL;
12420 }
12421
12422 /* If we need to wait for a SA then we can't return any errno */
12423 if (((ap->ipa_want_ah && (ixa->ixa_ipsec_ah_sa == NULL)) ||
12424 (ap->ipa_want_esp && (ixa->ixa_ipsec_esp_sa == NULL))) &&
12425 !ipsec_out_select_sa(mp, ixa))
12426 return (0);
12427
12428 /*
12429 * By now, we know what SA's to use. Toss over to ESP & AH
12430 * to do the heavy lifting.
12431 */
12432 if (ap->ipa_want_esp) {
12433 ASSERT(ixa->ixa_ipsec_esp_sa != NULL);
12434
12435 mp = ixa->ixa_ipsec_esp_sa->ipsa_output_func(mp, ixa);
12436 if (mp == NULL) {
12437 /*
12438 * Either it failed or is pending. In the former case
12439 * ipIfStatsInDiscards was increased.
12440 */
12441 return (0);
12442 }
12443 }
12444
12445 if (ap->ipa_want_ah) {
12446 ASSERT(ixa->ixa_ipsec_ah_sa != NULL);
12447
12448 mp = ixa->ixa_ipsec_ah_sa->ipsa_output_func(mp, ixa);
12449 if (mp == NULL) {
12450 /*
12451 * Either it failed or is pending. In the former case
12452 * ipIfStatsInDiscards was increased.
12453 */
12454 return (0);
12455 }
12456 }
12457 /*
12458 * We are done with IPsec processing. Send it over
12459 * the wire.
12460 */
12461 return (ip_output_post_ipsec(mp, ixa));
12462 }
12463
12464 /*
12465 * ioctls that go through a down/up sequence may need to wait for the down
12466 * to complete. This involves waiting for the ire and ipif refcnts to go down
12467 * to zero. Subsequently the ioctl is restarted from ipif_ill_refrele_tail.
12468 */
12469 /* ARGSUSED */
12470 void
12471 ip_reprocess_ioctl(ipsq_t *ipsq, queue_t *q, mblk_t *mp, void *dummy_arg)
12472 {
12473 struct iocblk *iocp;
12474 mblk_t *mp1;
12475 ip_ioctl_cmd_t *ipip;
12476 int err;
12477 sin_t *sin;
12478 struct lifreq *lifr;
12479 struct ifreq *ifr;
12480
12481 iocp = (struct iocblk *)mp->b_rptr;
12482 ASSERT(ipsq != NULL);
12483 /* Existence of mp1 verified in ip_wput_nondata */
12484 mp1 = mp->b_cont->b_cont;
12485 ipip = ip_sioctl_lookup(iocp->ioc_cmd);
12486 if (ipip->ipi_cmd == SIOCSLIFNAME || ipip->ipi_cmd == IF_UNITSEL) {
12487 /*
12488 * Special case where ipx_current_ipif is not set:
12489 * ill_phyint_reinit merged the v4 and v6 into a single ipsq.
12490 * We are here as were not able to complete the operation in
12491 * ipif_set_values because we could not become exclusive on
12492 * the new ipsq.
12493 */
12494 ill_t *ill = q->q_ptr;
12495 ipsq_current_start(ipsq, ill->ill_ipif, ipip->ipi_cmd);
12496 }
12497 ASSERT(ipsq->ipsq_xop->ipx_current_ipif != NULL);
12498
12499 if (ipip->ipi_cmd_type == IF_CMD) {
12500 /* This a old style SIOC[GS]IF* command */
12501 ifr = (struct ifreq *)mp1->b_rptr;
12502 sin = (sin_t *)&ifr->ifr_addr;
12503 } else if (ipip->ipi_cmd_type == LIF_CMD) {
12504 /* This a new style SIOC[GS]LIF* command */
12505 lifr = (struct lifreq *)mp1->b_rptr;
12506 sin = (sin_t *)&lifr->lifr_addr;
12507 } else {
12508 sin = NULL;
12509 }
12510
12511 err = (*ipip->ipi_func_restart)(ipsq->ipsq_xop->ipx_current_ipif, sin,
12512 q, mp, ipip, mp1->b_rptr);
12513
12514 DTRACE_PROBE4(ipif__ioctl, char *, "ip_reprocess_ioctl finish",
12515 int, ipip->ipi_cmd,
12516 ill_t *, ipsq->ipsq_xop->ipx_current_ipif->ipif_ill,
12517 ipif_t *, ipsq->ipsq_xop->ipx_current_ipif);
12518
12519 ip_ioctl_finish(q, mp, err, IPI2MODE(ipip), ipsq);
12520 }
12521
12522 /*
12523 * ioctl processing
12524 *
12525 * ioctl processing starts with ip_sioctl_copyin_setup(), which looks up
12526 * the ioctl command in the ioctl tables, determines the copyin data size
12527 * from the ipi_copyin_size field, and does an mi_copyin() of that size.
12528 *
12529 * ioctl processing then continues when the M_IOCDATA makes its way down to
12530 * ip_wput_nondata(). The ioctl is looked up again in the ioctl table, its
12531 * associated 'conn' is refheld till the end of the ioctl and the general
12532 * ioctl processing function ip_process_ioctl() is called to extract the
12533 * arguments and process the ioctl. To simplify extraction, ioctl commands
12534 * are "typed" based on the arguments they take (e.g., LIF_CMD which takes a
12535 * `struct lifreq'), and a common extract function (e.g., ip_extract_lifreq())
12536 * is used to extract the ioctl's arguments.
12537 *
12538 * ip_process_ioctl determines if the ioctl needs to be serialized, and if
12539 * so goes thru the serialization primitive ipsq_try_enter. Then the
12540 * appropriate function to handle the ioctl is called based on the entry in
12541 * the ioctl table. ioctl completion is encapsulated in ip_ioctl_finish
12542 * which also refreleases the 'conn' that was refheld at the start of the
12543 * ioctl. Finally ipsq_exit is called if needed to exit the ipsq.
12544 *
12545 * Many exclusive ioctls go thru an internal down up sequence as part of
12546 * the operation. For example an attempt to change the IP address of an
12547 * ipif entails ipif_down, set address, ipif_up. Bringing down the interface
12548 * does all the cleanup such as deleting all ires that use this address.
12549 * Then we need to wait till all references to the interface go away.
12550 */
12551 void
12552 ip_process_ioctl(ipsq_t *ipsq, queue_t *q, mblk_t *mp, void *arg)
12553 {
12554 struct iocblk *iocp = (struct iocblk *)mp->b_rptr;
12555 ip_ioctl_cmd_t *ipip = arg;
12556 ip_extract_func_t *extract_funcp;
12557 cmd_info_t ci;
12558 int err;
12559 boolean_t entered_ipsq = B_FALSE;
12560
12561 ip3dbg(("ip_process_ioctl: ioctl %X\n", iocp->ioc_cmd));
12562
12563 if (ipip == NULL)
12564 ipip = ip_sioctl_lookup(iocp->ioc_cmd);
12565
12566 /*
12567 * SIOCLIFADDIF needs to go thru a special path since the
12568 * ill may not exist yet. This happens in the case of lo0
12569 * which is created using this ioctl.
12570 */
12571 if (ipip->ipi_cmd == SIOCLIFADDIF) {
12572 err = ip_sioctl_addif(NULL, NULL, q, mp, NULL, NULL);
12573 DTRACE_PROBE4(ipif__ioctl, char *, "ip_process_ioctl finish",
12574 int, ipip->ipi_cmd, ill_t *, NULL, ipif_t *, NULL);
12575 ip_ioctl_finish(q, mp, err, IPI2MODE(ipip), NULL);
12576 return;
12577 }
12578
12579 ci.ci_ipif = NULL;
12580 switch (ipip->ipi_cmd_type) {
12581 case MISC_CMD:
12582 case MSFILT_CMD:
12583 /*
12584 * All MISC_CMD ioctls come in here -- e.g. SIOCGLIFCONF.
12585 */
12586 if (ipip->ipi_cmd == IF_UNITSEL) {
12587 /* ioctl comes down the ill */
12588 ci.ci_ipif = ((ill_t *)q->q_ptr)->ill_ipif;
12589 ipif_refhold(ci.ci_ipif);
12590 }
12591 err = 0;
12592 ci.ci_sin = NULL;
12593 ci.ci_sin6 = NULL;
12594 ci.ci_lifr = NULL;
12595 extract_funcp = NULL;
12596 break;
12597
12598 case IF_CMD:
12599 case LIF_CMD:
12600 extract_funcp = ip_extract_lifreq;
12601 break;
12602
12603 case ARP_CMD:
12604 case XARP_CMD:
12605 extract_funcp = ip_extract_arpreq;
12606 break;
12607
12608 default:
12609 ASSERT(0);
12610 }
12611
12612 if (extract_funcp != NULL) {
12613 err = (*extract_funcp)(q, mp, ipip, &ci);
12614 if (err != 0) {
12615 DTRACE_PROBE4(ipif__ioctl,
12616 char *, "ip_process_ioctl finish err",
12617 int, ipip->ipi_cmd, ill_t *, NULL, ipif_t *, NULL);
12618 ip_ioctl_finish(q, mp, err, IPI2MODE(ipip), NULL);
12619 return;
12620 }
12621
12622 /*
12623 * All of the extraction functions return a refheld ipif.
12624 */
12625 ASSERT(ci.ci_ipif != NULL);
12626 }
12627
12628 if (!(ipip->ipi_flags & IPI_WR)) {
12629 /*
12630 * A return value of EINPROGRESS means the ioctl is
12631 * either queued and waiting for some reason or has
12632 * already completed.
12633 */
12634 err = (*ipip->ipi_func)(ci.ci_ipif, ci.ci_sin, q, mp, ipip,
12635 ci.ci_lifr);
12636 if (ci.ci_ipif != NULL) {
12637 DTRACE_PROBE4(ipif__ioctl,
12638 char *, "ip_process_ioctl finish RD",
12639 int, ipip->ipi_cmd, ill_t *, ci.ci_ipif->ipif_ill,
12640 ipif_t *, ci.ci_ipif);
12641 ipif_refrele(ci.ci_ipif);
12642 } else {
12643 DTRACE_PROBE4(ipif__ioctl,
12644 char *, "ip_process_ioctl finish RD",
12645 int, ipip->ipi_cmd, ill_t *, NULL, ipif_t *, NULL);
12646 }
12647 ip_ioctl_finish(q, mp, err, IPI2MODE(ipip), NULL);
12648 return;
12649 }
12650
12651 ASSERT(ci.ci_ipif != NULL);
12652
12653 /*
12654 * If ipsq is non-NULL, we are already being called exclusively
12655 */
12656 ASSERT(ipsq == NULL || IAM_WRITER_IPSQ(ipsq));
12657 if (ipsq == NULL) {
12658 ipsq = ipsq_try_enter(ci.ci_ipif, NULL, q, mp, ip_process_ioctl,
12659 NEW_OP, B_TRUE);
12660 if (ipsq == NULL) {
12661 ipif_refrele(ci.ci_ipif);
12662 return;
12663 }
12664 entered_ipsq = B_TRUE;
12665 }
12666 /*
12667 * Release the ipif so that ipif_down and friends that wait for
12668 * references to go away are not misled about the current ipif_refcnt
12669 * values. We are writer so we can access the ipif even after releasing
12670 * the ipif.
12671 */
12672 ipif_refrele(ci.ci_ipif);
12673
12674 ipsq_current_start(ipsq, ci.ci_ipif, ipip->ipi_cmd);
12675
12676 /*
12677 * A return value of EINPROGRESS means the ioctl is
12678 * either queued and waiting for some reason or has
12679 * already completed.
12680 */
12681 err = (*ipip->ipi_func)(ci.ci_ipif, ci.ci_sin, q, mp, ipip, ci.ci_lifr);
12682
12683 DTRACE_PROBE4(ipif__ioctl, char *, "ip_process_ioctl finish WR",
12684 int, ipip->ipi_cmd,
12685 ill_t *, ci.ci_ipif == NULL ? NULL : ci.ci_ipif->ipif_ill,
12686 ipif_t *, ci.ci_ipif);
12687 ip_ioctl_finish(q, mp, err, IPI2MODE(ipip), ipsq);
12688
12689 if (entered_ipsq)
12690 ipsq_exit(ipsq);
12691 }
12692
12693 /*
12694 * Complete the ioctl. Typically ioctls use the mi package and need to
12695 * do mi_copyout/mi_copy_done.
12696 */
12697 void
12698 ip_ioctl_finish(queue_t *q, mblk_t *mp, int err, int mode, ipsq_t *ipsq)
12699 {
12700 conn_t *connp = NULL;
12701
12702 if (err == EINPROGRESS)
12703 return;
12704
12705 if (CONN_Q(q)) {
12706 connp = Q_TO_CONN(q);
12707 ASSERT(connp->conn_ref >= 2);
12708 }
12709
12710 switch (mode) {
12711 case COPYOUT:
12712 if (err == 0)
12713 mi_copyout(q, mp);
12714 else
12715 mi_copy_done(q, mp, err);
12716 break;
12717
12718 case NO_COPYOUT:
12719 mi_copy_done(q, mp, err);
12720 break;
12721
12722 default:
12723 ASSERT(mode == CONN_CLOSE); /* aborted through CONN_CLOSE */
12724 break;
12725 }
12726
12727 /*
12728 * The conn refhold and ioctlref placed on the conn at the start of the
12729 * ioctl are released here.
12730 */
12731 if (connp != NULL) {
12732 CONN_DEC_IOCTLREF(connp);
12733 CONN_OPER_PENDING_DONE(connp);
12734 }
12735
12736 if (ipsq != NULL)
12737 ipsq_current_finish(ipsq);
12738 }
12739
12740 /* Handles all non data messages */
12741 void
12742 ip_wput_nondata(queue_t *q, mblk_t *mp)
12743 {
12744 mblk_t *mp1;
12745 struct iocblk *iocp;
12746 ip_ioctl_cmd_t *ipip;
12747 conn_t *connp;
12748 cred_t *cr;
12749 char *proto_str;
12750
12751 if (CONN_Q(q))
12752 connp = Q_TO_CONN(q);
12753 else
12754 connp = NULL;
12755
12756 switch (DB_TYPE(mp)) {
12757 case M_IOCTL:
12758 /*
12759 * IOCTL processing begins in ip_sioctl_copyin_setup which
12760 * will arrange to copy in associated control structures.
12761 */
12762 ip_sioctl_copyin_setup(q, mp);
12763 return;
12764 case M_IOCDATA:
12765 /*
12766 * Ensure that this is associated with one of our trans-
12767 * parent ioctls. If it's not ours, discard it if we're
12768 * running as a driver, or pass it on if we're a module.
12769 */
12770 iocp = (struct iocblk *)mp->b_rptr;
12771 ipip = ip_sioctl_lookup(iocp->ioc_cmd);
12772 if (ipip == NULL) {
12773 if (q->q_next == NULL) {
12774 goto nak;
12775 } else {
12776 putnext(q, mp);
12777 }
12778 return;
12779 }
12780 if ((q->q_next != NULL) && !(ipip->ipi_flags & IPI_MODOK)) {
12781 /*
12782 * The ioctl is one we recognise, but is not consumed
12783 * by IP as a module and we are a module, so we drop
12784 */
12785 goto nak;
12786 }
12787
12788 /* IOCTL continuation following copyin or copyout. */
12789 if (mi_copy_state(q, mp, NULL) == -1) {
12790 /*
12791 * The copy operation failed. mi_copy_state already
12792 * cleaned up, so we're out of here.
12793 */
12794 return;
12795 }
12796 /*
12797 * If we just completed a copy in, we become writer and
12798 * continue processing in ip_sioctl_copyin_done. If it
12799 * was a copy out, we call mi_copyout again. If there is
12800 * nothing more to copy out, it will complete the IOCTL.
12801 */
12802 if (MI_COPY_DIRECTION(mp) == MI_COPY_IN) {
12803 if (!(mp1 = mp->b_cont) || !(mp1 = mp1->b_cont)) {
12804 mi_copy_done(q, mp, EPROTO);
12805 return;
12806 }
12807 /*
12808 * Check for cases that need more copying. A return
12809 * value of 0 means a second copyin has been started,
12810 * so we return; a return value of 1 means no more
12811 * copying is needed, so we continue.
12812 */
12813 if (ipip->ipi_cmd_type == MSFILT_CMD &&
12814 MI_COPY_COUNT(mp) == 1) {
12815 if (ip_copyin_msfilter(q, mp) == 0)
12816 return;
12817 }
12818 /*
12819 * Refhold the conn, till the ioctl completes. This is
12820 * needed in case the ioctl ends up in the pending mp
12821 * list. Every mp in the ipx_pending_mp list must have
12822 * a refhold on the conn to resume processing. The
12823 * refhold is released when the ioctl completes
12824 * (whether normally or abnormally). An ioctlref is also
12825 * placed on the conn to prevent TCP from removing the
12826 * queue needed to send the ioctl reply back.
12827 * In all cases ip_ioctl_finish is called to finish
12828 * the ioctl and release the refholds.
12829 */
12830 if (connp != NULL) {
12831 /* This is not a reentry */
12832 CONN_INC_REF(connp);
12833 CONN_INC_IOCTLREF(connp);
12834 } else {
12835 if (!(ipip->ipi_flags & IPI_MODOK)) {
12836 mi_copy_done(q, mp, EINVAL);
12837 return;
12838 }
12839 }
12840
12841 ip_process_ioctl(NULL, q, mp, ipip);
12842
12843 } else {
12844 mi_copyout(q, mp);
12845 }
12846 return;
12847
12848 case M_IOCNAK:
12849 /*
12850 * The only way we could get here is if a resolver didn't like
12851 * an IOCTL we sent it. This shouldn't happen.
12852 */
12853 (void) mi_strlog(q, 1, SL_ERROR|SL_TRACE,
12854 "ip_wput_nondata: unexpected M_IOCNAK, ioc_cmd 0x%x",
12855 ((struct iocblk *)mp->b_rptr)->ioc_cmd);
12856 freemsg(mp);
12857 return;
12858 case M_IOCACK:
12859 /* /dev/ip shouldn't see this */
12860 goto nak;
12861 case M_FLUSH:
12862 if (*mp->b_rptr & FLUSHW)
12863 flushq(q, FLUSHALL);
12864 if (q->q_next) {
12865 putnext(q, mp);
12866 return;
12867 }
12868 if (*mp->b_rptr & FLUSHR) {
12869 *mp->b_rptr &= ~FLUSHW;
12870 qreply(q, mp);
12871 return;
12872 }
12873 freemsg(mp);
12874 return;
12875 case M_CTL:
12876 break;
12877 case M_PROTO:
12878 case M_PCPROTO:
12879 /*
12880 * The only PROTO messages we expect are SNMP-related.
12881 */
12882 switch (((union T_primitives *)mp->b_rptr)->type) {
12883 case T_SVR4_OPTMGMT_REQ:
12884 ip2dbg(("ip_wput_nondata: T_SVR4_OPTMGMT_REQ "
12885 "flags %x\n",
12886 ((struct T_optmgmt_req *)mp->b_rptr)->MGMT_flags));
12887
12888 if (connp == NULL) {
12889 proto_str = "T_SVR4_OPTMGMT_REQ";
12890 goto protonak;
12891 }
12892
12893 /*
12894 * All Solaris components should pass a db_credp
12895 * for this TPI message, hence we ASSERT.
12896 * But in case there is some other M_PROTO that looks
12897 * like a TPI message sent by some other kernel
12898 * component, we check and return an error.
12899 */
12900 cr = msg_getcred(mp, NULL);
12901 ASSERT(cr != NULL);
12902 if (cr == NULL) {
12903 mp = mi_tpi_err_ack_alloc(mp, TSYSERR, EINVAL);
12904 if (mp != NULL)
12905 qreply(q, mp);
12906 return;
12907 }
12908
12909 if (!snmpcom_req(q, mp, ip_snmp_set, ip_snmp_get, cr)) {
12910 proto_str = "Bad SNMPCOM request?";
12911 goto protonak;
12912 }
12913 return;
12914 default:
12915 ip1dbg(("ip_wput_nondata: dropping M_PROTO prim %u\n",
12916 (int)*(uint_t *)mp->b_rptr));
12917 freemsg(mp);
12918 return;
12919 }
12920 default:
12921 break;
12922 }
12923 if (q->q_next) {
12924 putnext(q, mp);
12925 } else
12926 freemsg(mp);
12927 return;
12928
12929 nak:
12930 iocp->ioc_error = EINVAL;
12931 mp->b_datap->db_type = M_IOCNAK;
12932 iocp->ioc_count = 0;
12933 qreply(q, mp);
12934 return;
12935
12936 protonak:
12937 cmn_err(CE_NOTE, "IP doesn't process %s as a module", proto_str);
12938 if ((mp = mi_tpi_err_ack_alloc(mp, TPROTO, EINVAL)) != NULL)
12939 qreply(q, mp);
12940 }
12941
12942 /*
12943 * Process IP options in an outbound packet. Verify that the nexthop in a
12944 * strict source route is onlink.
12945 * Returns non-zero if something fails in which case an ICMP error has been
12946 * sent and mp freed.
12947 *
12948 * Assumes the ULP has called ip_massage_options to move nexthop into ipha_dst.
12949 */
12950 int
12951 ip_output_options(mblk_t *mp, ipha_t *ipha, ip_xmit_attr_t *ixa, ill_t *ill)
12952 {
12953 ipoptp_t opts;
12954 uchar_t *opt;
12955 uint8_t optval;
12956 uint8_t optlen;
12957 ipaddr_t dst;
12958 intptr_t code = 0;
12959 ire_t *ire;
12960 ip_stack_t *ipst = ixa->ixa_ipst;
12961 ip_recv_attr_t iras;
12962
12963 ip2dbg(("ip_output_options\n"));
12964
12965 dst = ipha->ipha_dst;
12966 for (optval = ipoptp_first(&opts, ipha);
12967 optval != IPOPT_EOL;
12968 optval = ipoptp_next(&opts)) {
12969 opt = opts.ipoptp_cur;
12970 optlen = opts.ipoptp_len;
12971 ip2dbg(("ip_output_options: opt %d, len %d\n",
12972 optval, optlen));
12973 switch (optval) {
12974 uint32_t off;
12975 case IPOPT_SSRR:
12976 case IPOPT_LSRR:
12977 if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
12978 ip1dbg((
12979 "ip_output_options: bad option offset\n"));
12980 code = (char *)&opt[IPOPT_OLEN] -
12981 (char *)ipha;
12982 goto param_prob;
12983 }
12984 off = opt[IPOPT_OFFSET];
12985 ip1dbg(("ip_output_options: next hop 0x%x\n",
12986 ntohl(dst)));
12987 /*
12988 * For strict: verify that dst is directly
12989 * reachable.
12990 */
12991 if (optval == IPOPT_SSRR) {
12992 ire = ire_ftable_lookup_v4(dst, 0, 0,
12993 IRE_INTERFACE, NULL, ALL_ZONES,
12994 ixa->ixa_tsl,
12995 MATCH_IRE_TYPE | MATCH_IRE_SECATTR, 0, ipst,
12996 NULL);
12997 if (ire == NULL) {
12998 ip1dbg(("ip_output_options: SSRR not"
12999 " directly reachable: 0x%x\n",
13000 ntohl(dst)));
13001 goto bad_src_route;
13002 }
13003 ire_refrele(ire);
13004 }
13005 break;
13006 case IPOPT_RR:
13007 if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
13008 ip1dbg((
13009 "ip_output_options: bad option offset\n"));
13010 code = (char *)&opt[IPOPT_OLEN] -
13011 (char *)ipha;
13012 goto param_prob;
13013 }
13014 break;
13015 case IPOPT_TS:
13016 /*
13017 * Verify that length >=5 and that there is either
13018 * room for another timestamp or that the overflow
13019 * counter is not maxed out.
13020 */
13021 code = (char *)&opt[IPOPT_OLEN] - (char *)ipha;
13022 if (optlen < IPOPT_MINLEN_IT) {
13023 goto param_prob;
13024 }
13025 if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
13026 ip1dbg((
13027 "ip_output_options: bad option offset\n"));
13028 code = (char *)&opt[IPOPT_OFFSET] -
13029 (char *)ipha;
13030 goto param_prob;
13031 }
13032 switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
13033 case IPOPT_TS_TSONLY:
13034 off = IPOPT_TS_TIMELEN;
13035 break;
13036 case IPOPT_TS_TSANDADDR:
13037 case IPOPT_TS_PRESPEC:
13038 case IPOPT_TS_PRESPEC_RFC791:
13039 off = IP_ADDR_LEN + IPOPT_TS_TIMELEN;
13040 break;
13041 default:
13042 code = (char *)&opt[IPOPT_POS_OV_FLG] -
13043 (char *)ipha;
13044 goto param_prob;
13045 }
13046 if (opt[IPOPT_OFFSET] - 1 + off > optlen &&
13047 (opt[IPOPT_POS_OV_FLG] & 0xF0) == 0xF0) {
13048 /*
13049 * No room and the overflow counter is 15
13050 * already.
13051 */
13052 goto param_prob;
13053 }
13054 break;
13055 }
13056 }
13057
13058 if ((opts.ipoptp_flags & IPOPTP_ERROR) == 0)
13059 return (0);
13060
13061 ip1dbg(("ip_output_options: error processing IP options."));
13062 code = (char *)&opt[IPOPT_OFFSET] - (char *)ipha;
13063
13064 param_prob:
13065 bzero(&iras, sizeof (iras));
13066 iras.ira_ill = iras.ira_rill = ill;
13067 iras.ira_ruifindex = ill->ill_phyint->phyint_ifindex;
13068 iras.ira_rifindex = iras.ira_ruifindex;
13069 iras.ira_flags = IRAF_IS_IPV4;
13070
13071 ip_drop_output("ip_output_options", mp, ill);
13072 icmp_param_problem(mp, (uint8_t)code, &iras);
13073 ASSERT(!(iras.ira_flags & IRAF_IPSEC_SECURE));
13074 return (-1);
13075
13076 bad_src_route:
13077 bzero(&iras, sizeof (iras));
13078 iras.ira_ill = iras.ira_rill = ill;
13079 iras.ira_ruifindex = ill->ill_phyint->phyint_ifindex;
13080 iras.ira_rifindex = iras.ira_ruifindex;
13081 iras.ira_flags = IRAF_IS_IPV4;
13082
13083 ip_drop_input("ICMP_SOURCE_ROUTE_FAILED", mp, ill);
13084 icmp_unreachable(mp, ICMP_SOURCE_ROUTE_FAILED, &iras);
13085 ASSERT(!(iras.ira_flags & IRAF_IPSEC_SECURE));
13086 return (-1);
13087 }
13088
13089 /*
13090 * The maximum value of conn_drain_list_cnt is CONN_MAXDRAINCNT.
13091 * conn_drain_list_cnt can be changed by setting conn_drain_nthreads
13092 * thru /etc/system.
13093 */
13094 #define CONN_MAXDRAINCNT 64
13095
13096 static void
13097 conn_drain_init(ip_stack_t *ipst)
13098 {
13099 int i, j;
13100 idl_tx_list_t *itl_tx;
13101
13102 ipst->ips_conn_drain_list_cnt = conn_drain_nthreads;
13103
13104 if ((ipst->ips_conn_drain_list_cnt == 0) ||
13105 (ipst->ips_conn_drain_list_cnt > CONN_MAXDRAINCNT)) {
13106 /*
13107 * Default value of the number of drainers is the
13108 * number of cpus, subject to maximum of 8 drainers.
13109 */
13110 if (boot_max_ncpus != -1)
13111 ipst->ips_conn_drain_list_cnt = MIN(boot_max_ncpus, 8);
13112 else
13113 ipst->ips_conn_drain_list_cnt = MIN(max_ncpus, 8);
13114 }
13115
13116 ipst->ips_idl_tx_list =
13117 kmem_zalloc(TX_FANOUT_SIZE * sizeof (idl_tx_list_t), KM_SLEEP);
13118 for (i = 0; i < TX_FANOUT_SIZE; i++) {
13119 itl_tx = &ipst->ips_idl_tx_list[i];
13120 itl_tx->txl_drain_list =
13121 kmem_zalloc(ipst->ips_conn_drain_list_cnt *
13122 sizeof (idl_t), KM_SLEEP);
13123 mutex_init(&itl_tx->txl_lock, NULL, MUTEX_DEFAULT, NULL);
13124 for (j = 0; j < ipst->ips_conn_drain_list_cnt; j++) {
13125 mutex_init(&itl_tx->txl_drain_list[j].idl_lock, NULL,
13126 MUTEX_DEFAULT, NULL);
13127 itl_tx->txl_drain_list[j].idl_itl = itl_tx;
13128 }
13129 }
13130 }
13131
13132 static void
13133 conn_drain_fini(ip_stack_t *ipst)
13134 {
13135 int i;
13136 idl_tx_list_t *itl_tx;
13137
13138 for (i = 0; i < TX_FANOUT_SIZE; i++) {
13139 itl_tx = &ipst->ips_idl_tx_list[i];
13140 kmem_free(itl_tx->txl_drain_list,
13141 ipst->ips_conn_drain_list_cnt * sizeof (idl_t));
13142 }
13143 kmem_free(ipst->ips_idl_tx_list,
13144 TX_FANOUT_SIZE * sizeof (idl_tx_list_t));
13145 ipst->ips_idl_tx_list = NULL;
13146 }
13147
13148 /*
13149 * Flow control has blocked us from proceeding. Insert the given conn in one
13150 * of the conn drain lists. When flow control is unblocked, either ip_wsrv()
13151 * (STREAMS) or ill_flow_enable() (direct) will be called back, which in turn
13152 * will call conn_walk_drain(). See the flow control notes at the top of this
13153 * file for more details.
13154 */
13155 void
13156 conn_drain_insert(conn_t *connp, idl_tx_list_t *tx_list)
13157 {
13158 idl_t *idl = tx_list->txl_drain_list;
13159 uint_t index;
13160 ip_stack_t *ipst = connp->conn_netstack->netstack_ip;
13161
13162 mutex_enter(&connp->conn_lock);
13163 if (connp->conn_state_flags & CONN_CLOSING) {
13164 /*
13165 * The conn is closing as a result of which CONN_CLOSING
13166 * is set. Return.
13167 */
13168 mutex_exit(&connp->conn_lock);
13169 return;
13170 } else if (connp->conn_idl == NULL) {
13171 /*
13172 * Assign the next drain list round robin. We dont' use
13173 * a lock, and thus it may not be strictly round robin.
13174 * Atomicity of load/stores is enough to make sure that
13175 * conn_drain_list_index is always within bounds.
13176 */
13177 index = tx_list->txl_drain_index;
13178 ASSERT(index < ipst->ips_conn_drain_list_cnt);
13179 connp->conn_idl = &tx_list->txl_drain_list[index];
13180 index++;
13181 if (index == ipst->ips_conn_drain_list_cnt)
13182 index = 0;
13183 tx_list->txl_drain_index = index;
13184 } else {
13185 ASSERT(connp->conn_idl->idl_itl == tx_list);
13186 }
13187 mutex_exit(&connp->conn_lock);
13188
13189 idl = connp->conn_idl;
13190 mutex_enter(&idl->idl_lock);
13191 if ((connp->conn_drain_prev != NULL) ||
13192 (connp->conn_state_flags & CONN_CLOSING)) {
13193 /*
13194 * The conn is either already in the drain list or closing.
13195 * (We needed to check for CONN_CLOSING again since close can
13196 * sneak in between dropping conn_lock and acquiring idl_lock.)
13197 */
13198 mutex_exit(&idl->idl_lock);
13199 return;
13200 }
13201
13202 /*
13203 * The conn is not in the drain list. Insert it at the
13204 * tail of the drain list. The drain list is circular
13205 * and doubly linked. idl_conn points to the 1st element
13206 * in the list.
13207 */
13208 if (idl->idl_conn == NULL) {
13209 idl->idl_conn = connp;
13210 connp->conn_drain_next = connp;
13211 connp->conn_drain_prev = connp;
13212 } else {
13213 conn_t *head = idl->idl_conn;
13214
13215 connp->conn_drain_next = head;
13216 connp->conn_drain_prev = head->conn_drain_prev;
13217 head->conn_drain_prev->conn_drain_next = connp;
13218 head->conn_drain_prev = connp;
13219 }
13220 /*
13221 * For non streams based sockets assert flow control.
13222 */
13223 conn_setqfull(connp, NULL);
13224 mutex_exit(&idl->idl_lock);
13225 }
13226
13227 static void
13228 conn_drain_remove(conn_t *connp)
13229 {
13230 idl_t *idl = connp->conn_idl;
13231
13232 if (idl != NULL) {
13233 /*
13234 * Remove ourself from the drain list.
13235 */
13236 if (connp->conn_drain_next == connp) {
13237 /* Singleton in the list */
13238 ASSERT(connp->conn_drain_prev == connp);
13239 idl->idl_conn = NULL;
13240 } else {
13241 connp->conn_drain_prev->conn_drain_next =
13242 connp->conn_drain_next;
13243 connp->conn_drain_next->conn_drain_prev =
13244 connp->conn_drain_prev;
13245 if (idl->idl_conn == connp)
13246 idl->idl_conn = connp->conn_drain_next;
13247 }
13248
13249 /*
13250 * NOTE: because conn_idl is associated with a specific drain
13251 * list which in turn is tied to the index the TX ring
13252 * (txl_cookie) hashes to, and because the TX ring can change
13253 * over the lifetime of the conn_t, we must clear conn_idl so
13254 * a subsequent conn_drain_insert() will set conn_idl again
13255 * based on the latest txl_cookie.
13256 */
13257 connp->conn_idl = NULL;
13258 }
13259 connp->conn_drain_next = NULL;
13260 connp->conn_drain_prev = NULL;
13261
13262 conn_clrqfull(connp, NULL);
13263 /*
13264 * For streams based sockets open up flow control.
13265 */
13266 if (!IPCL_IS_NONSTR(connp))
13267 enableok(connp->conn_wq);
13268 }
13269
13270 /*
13271 * This conn is closing, and we are called from ip_close. OR
13272 * this conn is draining because flow-control on the ill has been relieved.
13273 *
13274 * We must also need to remove conn's on this idl from the list, and also
13275 * inform the sockfs upcalls about the change in flow-control.
13276 */
13277 static void
13278 conn_drain(conn_t *connp, boolean_t closing)
13279 {
13280 idl_t *idl;
13281 conn_t *next_connp;
13282
13283 /*
13284 * connp->conn_idl is stable at this point, and no lock is needed
13285 * to check it. If we are called from ip_close, close has already
13286 * set CONN_CLOSING, thus freezing the value of conn_idl, and
13287 * called us only because conn_idl is non-null. If we are called thru
13288 * service, conn_idl could be null, but it cannot change because
13289 * service is single-threaded per queue, and there cannot be another
13290 * instance of service trying to call conn_drain_insert on this conn
13291 * now.
13292 */
13293 ASSERT(!closing || connp == NULL || connp->conn_idl != NULL);
13294
13295 /*
13296 * If the conn doesn't exist or is not on a drain list, bail.
13297 */
13298 if (connp == NULL || connp->conn_idl == NULL ||
13299 connp->conn_drain_prev == NULL) {
13300 return;
13301 }
13302
13303 idl = connp->conn_idl;
13304 ASSERT(MUTEX_HELD(&idl->idl_lock));
13305
13306 if (!closing) {
13307 next_connp = connp->conn_drain_next;
13308 while (next_connp != connp) {
13309 conn_t *delconnp = next_connp;
13310
13311 next_connp = next_connp->conn_drain_next;
13312 conn_drain_remove(delconnp);
13313 }
13314 ASSERT(connp->conn_drain_next == idl->idl_conn);
13315 }
13316 conn_drain_remove(connp);
13317 }
13318
13319 /*
13320 * Write service routine. Shared perimeter entry point.
13321 * The device queue's messages has fallen below the low water mark and STREAMS
13322 * has backenabled the ill_wq. Send sockfs notification about flow-control on
13323 * each waiting conn.
13324 */
13325 void
13326 ip_wsrv(queue_t *q)
13327 {
13328 ill_t *ill;
13329
13330 ill = (ill_t *)q->q_ptr;
13331 if (ill->ill_state_flags == 0) {
13332 ip_stack_t *ipst = ill->ill_ipst;
13333
13334 /*
13335 * The device flow control has opened up.
13336 * Walk through conn drain lists and qenable the
13337 * first conn in each list. This makes sense only
13338 * if the stream is fully plumbed and setup.
13339 * Hence the ill_state_flags check above.
13340 */
13341 ip1dbg(("ip_wsrv: walking\n"));
13342 conn_walk_drain(ipst, &ipst->ips_idl_tx_list[0]);
13343 enableok(ill->ill_wq);
13344 }
13345 }
13346
13347 /*
13348 * Callback to disable flow control in IP.
13349 *
13350 * This is a mac client callback added when the DLD_CAPAB_DIRECT capability
13351 * is enabled.
13352 *
13353 * When MAC_TX() is not able to send any more packets, dld sets its queue
13354 * to QFULL and enable the STREAMS flow control. Later, when the underlying
13355 * driver is able to continue to send packets, it calls mac_tx_(ring_)update()
13356 * function and wakes up corresponding mac worker threads, which in turn
13357 * calls this callback function, and disables flow control.
13358 */
13359 void
13360 ill_flow_enable(void *arg, ip_mac_tx_cookie_t cookie)
13361 {
13362 ill_t *ill = (ill_t *)arg;
13363 ip_stack_t *ipst = ill->ill_ipst;
13364 idl_tx_list_t *idl_txl;
13365
13366 idl_txl = &ipst->ips_idl_tx_list[IDLHASHINDEX(cookie)];
13367 mutex_enter(&idl_txl->txl_lock);
13368 /* add code to to set a flag to indicate idl_txl is enabled */
13369 conn_walk_drain(ipst, idl_txl);
13370 mutex_exit(&idl_txl->txl_lock);
13371 }
13372
13373 /*
13374 * Flow control has been relieved and STREAMS has backenabled us; drain
13375 * all the conn lists on `tx_list'.
13376 */
13377 static void
13378 conn_walk_drain(ip_stack_t *ipst, idl_tx_list_t *tx_list)
13379 {
13380 int i;
13381 idl_t *idl;
13382
13383 IP_STAT(ipst, ip_conn_walk_drain);
13384
13385 for (i = 0; i < ipst->ips_conn_drain_list_cnt; i++) {
13386 idl = &tx_list->txl_drain_list[i];
13387 mutex_enter(&idl->idl_lock);
13388 conn_drain(idl->idl_conn, B_FALSE);
13389 mutex_exit(&idl->idl_lock);
13390 }
13391 }
13392
13393 /*
13394 * Determine if the ill and multicast aspects of that packets
13395 * "matches" the conn.
13396 */
13397 boolean_t
13398 conn_wantpacket(conn_t *connp, ip_recv_attr_t *ira, ipha_t *ipha)
13399 {
13400 ill_t *ill = ira->ira_rill;
13401 zoneid_t zoneid = ira->ira_zoneid;
13402 uint_t in_ifindex;
13403 ipaddr_t dst, src;
13404
13405 dst = ipha->ipha_dst;
13406 src = ipha->ipha_src;
13407
13408 /*
13409 * conn_incoming_ifindex is set by IP_BOUND_IF which limits
13410 * unicast, broadcast and multicast reception to
13411 * conn_incoming_ifindex.
13412 * conn_wantpacket is called for unicast, broadcast and
13413 * multicast packets.
13414 */
13415 in_ifindex = connp->conn_incoming_ifindex;
13416
13417 /* mpathd can bind to the under IPMP interface, which we allow */
13418 if (in_ifindex != 0 && in_ifindex != ill->ill_phyint->phyint_ifindex) {
13419 if (!IS_UNDER_IPMP(ill))
13420 return (B_FALSE);
13421
13422 if (in_ifindex != ipmp_ill_get_ipmp_ifindex(ill))
13423 return (B_FALSE);
13424 }
13425
13426 if (!IPCL_ZONE_MATCH(connp, zoneid))
13427 return (B_FALSE);
13428
13429 if (!(ira->ira_flags & IRAF_MULTICAST))
13430 return (B_TRUE);
13431
13432 if (connp->conn_multi_router) {
13433 /* multicast packet and multicast router socket: send up */
13434 return (B_TRUE);
13435 }
13436
13437 if (ipha->ipha_protocol == IPPROTO_PIM ||
13438 ipha->ipha_protocol == IPPROTO_RSVP)
13439 return (B_TRUE);
13440
13441 return (conn_hasmembers_ill_withsrc_v4(connp, dst, src, ira->ira_ill));
13442 }
13443
13444 void
13445 conn_setqfull(conn_t *connp, boolean_t *flow_stopped)
13446 {
13447 if (IPCL_IS_NONSTR(connp)) {
13448 (*connp->conn_upcalls->su_txq_full)
13449 (connp->conn_upper_handle, B_TRUE);
13450 if (flow_stopped != NULL)
13451 *flow_stopped = B_TRUE;
13452 } else {
13453 queue_t *q = connp->conn_wq;
13454
13455 ASSERT(q != NULL);
13456 if (!(q->q_flag & QFULL)) {
13457 mutex_enter(QLOCK(q));
13458 if (!(q->q_flag & QFULL)) {
13459 /* still need to set QFULL */
13460 q->q_flag |= QFULL;
13461 /* set flow_stopped to true under QLOCK */
13462 if (flow_stopped != NULL)
13463 *flow_stopped = B_TRUE;
13464 mutex_exit(QLOCK(q));
13465 } else {
13466 /* flow_stopped is left unchanged */
13467 mutex_exit(QLOCK(q));
13468 }
13469 }
13470 }
13471 }
13472
13473 void
13474 conn_clrqfull(conn_t *connp, boolean_t *flow_stopped)
13475 {
13476 if (IPCL_IS_NONSTR(connp)) {
13477 (*connp->conn_upcalls->su_txq_full)
13478 (connp->conn_upper_handle, B_FALSE);
13479 if (flow_stopped != NULL)
13480 *flow_stopped = B_FALSE;
13481 } else {
13482 queue_t *q = connp->conn_wq;
13483
13484 ASSERT(q != NULL);
13485 if (q->q_flag & QFULL) {
13486 mutex_enter(QLOCK(q));
13487 if (q->q_flag & QFULL) {
13488 q->q_flag &= ~QFULL;
13489 /* set flow_stopped to false under QLOCK */
13490 if (flow_stopped != NULL)
13491 *flow_stopped = B_FALSE;
13492 mutex_exit(QLOCK(q));
13493 if (q->q_flag & QWANTW)
13494 qbackenable(q, 0);
13495 } else {
13496 /* flow_stopped is left unchanged */
13497 mutex_exit(QLOCK(q));
13498 }
13499 }
13500 }
13501
13502 mutex_enter(&connp->conn_lock);
13503 connp->conn_blocked = B_FALSE;
13504 mutex_exit(&connp->conn_lock);
13505 }
13506
13507 /*
13508 * Return the length in bytes of the IPv4 headers (base header, label, and
13509 * other IP options) that will be needed based on the
13510 * ip_pkt_t structure passed by the caller.
13511 *
13512 * The returned length does not include the length of the upper level
13513 * protocol (ULP) header.
13514 * The caller needs to check that the length doesn't exceed the max for IPv4.
13515 */
13516 int
13517 ip_total_hdrs_len_v4(const ip_pkt_t *ipp)
13518 {
13519 int len;
13520
13521 len = IP_SIMPLE_HDR_LENGTH;
13522 if (ipp->ipp_fields & IPPF_LABEL_V4) {
13523 ASSERT(ipp->ipp_label_len_v4 != 0);
13524 /* We need to round up here */
13525 len += (ipp->ipp_label_len_v4 + 3) & ~3;
13526 }
13527
13528 if (ipp->ipp_fields & IPPF_IPV4_OPTIONS) {
13529 ASSERT(ipp->ipp_ipv4_options_len != 0);
13530 ASSERT((ipp->ipp_ipv4_options_len & 3) == 0);
13531 len += ipp->ipp_ipv4_options_len;
13532 }
13533 return (len);
13534 }
13535
13536 /*
13537 * All-purpose routine to build an IPv4 header with options based
13538 * on the abstract ip_pkt_t.
13539 *
13540 * The caller has to set the source and destination address as well as
13541 * ipha_length. The caller has to massage any source route and compensate
13542 * for the ULP pseudo-header checksum due to the source route.
13543 */
13544 void
13545 ip_build_hdrs_v4(uchar_t *buf, uint_t buf_len, const ip_pkt_t *ipp,
13546 uint8_t protocol)
13547 {
13548 ipha_t *ipha = (ipha_t *)buf;
13549 uint8_t *cp;
13550
13551 /* Initialize IPv4 header */
13552 ipha->ipha_type_of_service = ipp->ipp_type_of_service;
13553 ipha->ipha_length = 0; /* Caller will set later */
13554 ipha->ipha_ident = 0;
13555 ipha->ipha_fragment_offset_and_flags = 0;
13556 ipha->ipha_ttl = ipp->ipp_unicast_hops;
13557 ipha->ipha_protocol = protocol;
13558 ipha->ipha_hdr_checksum = 0;
13559
13560 if ((ipp->ipp_fields & IPPF_ADDR) &&
13561 IN6_IS_ADDR_V4MAPPED(&ipp->ipp_addr))
13562 ipha->ipha_src = ipp->ipp_addr_v4;
13563
13564 cp = (uint8_t *)&ipha[1];
13565 if (ipp->ipp_fields & IPPF_LABEL_V4) {
13566 ASSERT(ipp->ipp_label_len_v4 != 0);
13567 bcopy(ipp->ipp_label_v4, cp, ipp->ipp_label_len_v4);
13568 cp += ipp->ipp_label_len_v4;
13569 /* We need to round up here */
13570 while ((uintptr_t)cp & 0x3) {
13571 *cp++ = IPOPT_NOP;
13572 }
13573 }
13574
13575 if (ipp->ipp_fields & IPPF_IPV4_OPTIONS) {
13576 ASSERT(ipp->ipp_ipv4_options_len != 0);
13577 ASSERT((ipp->ipp_ipv4_options_len & 3) == 0);
13578 bcopy(ipp->ipp_ipv4_options, cp, ipp->ipp_ipv4_options_len);
13579 cp += ipp->ipp_ipv4_options_len;
13580 }
13581 ipha->ipha_version_and_hdr_length =
13582 (uint8_t)((IP_VERSION << 4) + buf_len / 4);
13583
13584 ASSERT((int)(cp - buf) == buf_len);
13585 }
13586
13587 /* Allocate the private structure */
13588 static int
13589 ip_priv_alloc(void **bufp)
13590 {
13591 void *buf;
13592
13593 if ((buf = kmem_alloc(sizeof (ip_priv_t), KM_NOSLEEP)) == NULL)
13594 return (ENOMEM);
13595
13596 *bufp = buf;
13597 return (0);
13598 }
13599
13600 /* Function to delete the private structure */
13601 void
13602 ip_priv_free(void *buf)
13603 {
13604 ASSERT(buf != NULL);
13605 kmem_free(buf, sizeof (ip_priv_t));
13606 }
13607
13608 /*
13609 * The entry point for IPPF processing.
13610 * If the classifier (IPGPC_CLASSIFY) is not loaded and configured, the
13611 * routine just returns.
13612 *
13613 * When called, ip_process generates an ipp_packet_t structure
13614 * which holds the state information for this packet and invokes the
13615 * the classifier (via ipp_packet_process). The classification, depending on
13616 * configured filters, results in a list of actions for this packet. Invoking
13617 * an action may cause the packet to be dropped, in which case we return NULL.
13618 * proc indicates the callout position for
13619 * this packet and ill is the interface this packet arrived on or will leave
13620 * on (inbound and outbound resp.).
13621 *
13622 * We do the processing on the rill (mapped to the upper if ipmp), but MIB
13623 * on the ill corrsponding to the destination IP address.
13624 */
13625 mblk_t *
13626 ip_process(ip_proc_t proc, mblk_t *mp, ill_t *rill, ill_t *ill)
13627 {
13628 ip_priv_t *priv;
13629 ipp_action_id_t aid;
13630 int rc = 0;
13631 ipp_packet_t *pp;
13632
13633 /* If the classifier is not loaded, return */
13634 if ((aid = ipp_action_lookup(IPGPC_CLASSIFY)) == IPP_ACTION_INVAL) {
13635 return (mp);
13636 }
13637
13638 ASSERT(mp != NULL);
13639
13640 /* Allocate the packet structure */
13641 rc = ipp_packet_alloc(&pp, "ip", aid);
13642 if (rc != 0)
13643 goto drop;
13644
13645 /* Allocate the private structure */
13646 rc = ip_priv_alloc((void **)&priv);
13647 if (rc != 0) {
13648 ipp_packet_free(pp);
13649 goto drop;
13650 }
13651 priv->proc = proc;
13652 priv->ill_index = ill_get_upper_ifindex(rill);
13653
13654 ipp_packet_set_private(pp, priv, ip_priv_free);
13655 ipp_packet_set_data(pp, mp);
13656
13657 /* Invoke the classifier */
13658 rc = ipp_packet_process(&pp);
13659 if (pp != NULL) {
13660 mp = ipp_packet_get_data(pp);
13661 ipp_packet_free(pp);
13662 if (rc != 0)
13663 goto drop;
13664 return (mp);
13665 } else {
13666 /* No mp to trace in ip_drop_input/ip_drop_output */
13667 mp = NULL;
13668 }
13669 drop:
13670 if (proc == IPP_LOCAL_IN || proc == IPP_FWD_IN) {
13671 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
13672 ip_drop_input("ip_process", mp, ill);
13673 } else {
13674 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
13675 ip_drop_output("ip_process", mp, ill);
13676 }
13677 freemsg(mp);
13678 return (NULL);
13679 }
13680
13681 /*
13682 * Propagate a multicast group membership operation (add/drop) on
13683 * all the interfaces crossed by the related multirt routes.
13684 * The call is considered successful if the operation succeeds
13685 * on at least one interface.
13686 *
13687 * This assumes that a set of IRE_HOST/RTF_MULTIRT has been created for the
13688 * multicast addresses with the ire argument being the first one.
13689 * We walk the bucket to find all the of those.
13690 *
13691 * Common to IPv4 and IPv6.
13692 */
13693 static int
13694 ip_multirt_apply_membership(int (*fn)(conn_t *, boolean_t,
13695 const in6_addr_t *, ipaddr_t, uint_t, mcast_record_t, const in6_addr_t *),
13696 ire_t *ire, conn_t *connp, boolean_t checkonly, const in6_addr_t *v6group,
13697 mcast_record_t fmode, const in6_addr_t *v6src)
13698 {
13699 ire_t *ire_gw;
13700 irb_t *irb;
13701 int ifindex;
13702 int error = 0;
13703 int result;
13704 ip_stack_t *ipst = ire->ire_ipst;
13705 ipaddr_t group;
13706 boolean_t isv6;
13707 int match_flags;
13708
13709 if (IN6_IS_ADDR_V4MAPPED(v6group)) {
13710 IN6_V4MAPPED_TO_IPADDR(v6group, group);
13711 isv6 = B_FALSE;
13712 } else {
13713 isv6 = B_TRUE;
13714 }
13715
13716 irb = ire->ire_bucket;
13717 ASSERT(irb != NULL);
13718
13719 result = 0;
13720 irb_refhold(irb);
13721 for (; ire != NULL; ire = ire->ire_next) {
13722 if ((ire->ire_flags & RTF_MULTIRT) == 0)
13723 continue;
13724
13725 /* We handle -ifp routes by matching on the ill if set */
13726 match_flags = MATCH_IRE_TYPE;
13727 if (ire->ire_ill != NULL)
13728 match_flags |= MATCH_IRE_ILL;
13729
13730 if (isv6) {
13731 if (!IN6_ARE_ADDR_EQUAL(&ire->ire_addr_v6, v6group))
13732 continue;
13733
13734 ire_gw = ire_ftable_lookup_v6(&ire->ire_gateway_addr_v6,
13735 0, 0, IRE_INTERFACE, ire->ire_ill, ALL_ZONES, NULL,
13736 match_flags, 0, ipst, NULL);
13737 } else {
13738 if (ire->ire_addr != group)
13739 continue;
13740
13741 ire_gw = ire_ftable_lookup_v4(ire->ire_gateway_addr,
13742 0, 0, IRE_INTERFACE, ire->ire_ill, ALL_ZONES, NULL,
13743 match_flags, 0, ipst, NULL);
13744 }
13745 /* No interface route exists for the gateway; skip this ire. */
13746 if (ire_gw == NULL)
13747 continue;
13748 if (ire_gw->ire_flags & (RTF_REJECT|RTF_BLACKHOLE)) {
13749 ire_refrele(ire_gw);
13750 continue;
13751 }
13752 ASSERT(ire_gw->ire_ill != NULL); /* IRE_INTERFACE */
13753 ifindex = ire_gw->ire_ill->ill_phyint->phyint_ifindex;
13754
13755 /*
13756 * The operation is considered a success if
13757 * it succeeds at least once on any one interface.
13758 */
13759 error = fn(connp, checkonly, v6group, INADDR_ANY, ifindex,
13760 fmode, v6src);
13761 if (error == 0)
13762 result = CGTP_MCAST_SUCCESS;
13763
13764 ire_refrele(ire_gw);
13765 }
13766 irb_refrele(irb);
13767 /*
13768 * Consider the call as successful if we succeeded on at least
13769 * one interface. Otherwise, return the last encountered error.
13770 */
13771 return (result == CGTP_MCAST_SUCCESS ? 0 : error);
13772 }
13773
13774 /*
13775 * Return the expected CGTP hooks version number.
13776 */
13777 int
13778 ip_cgtp_filter_supported(void)
13779 {
13780 return (ip_cgtp_filter_rev);
13781 }
13782
13783 /*
13784 * CGTP hooks can be registered by invoking this function.
13785 * Checks that the version number matches.
13786 */
13787 int
13788 ip_cgtp_filter_register(netstackid_t stackid, cgtp_filter_ops_t *ops)
13789 {
13790 netstack_t *ns;
13791 ip_stack_t *ipst;
13792
13793 if (ops->cfo_filter_rev != CGTP_FILTER_REV)
13794 return (ENOTSUP);
13795
13796 ns = netstack_find_by_stackid(stackid);
13797 if (ns == NULL)
13798 return (EINVAL);
13799 ipst = ns->netstack_ip;
13800 ASSERT(ipst != NULL);
13801
13802 if (ipst->ips_ip_cgtp_filter_ops != NULL) {
13803 netstack_rele(ns);
13804 return (EALREADY);
13805 }
13806
13807 ipst->ips_ip_cgtp_filter_ops = ops;
13808
13809 ill_set_inputfn_all(ipst);
13810
13811 netstack_rele(ns);
13812 return (0);
13813 }
13814
13815 /*
13816 * CGTP hooks can be unregistered by invoking this function.
13817 * Returns ENXIO if there was no registration.
13818 * Returns EBUSY if the ndd variable has not been turned off.
13819 */
13820 int
13821 ip_cgtp_filter_unregister(netstackid_t stackid)
13822 {
13823 netstack_t *ns;
13824 ip_stack_t *ipst;
13825
13826 ns = netstack_find_by_stackid(stackid);
13827 if (ns == NULL)
13828 return (EINVAL);
13829 ipst = ns->netstack_ip;
13830 ASSERT(ipst != NULL);
13831
13832 if (ipst->ips_ip_cgtp_filter) {
13833 netstack_rele(ns);
13834 return (EBUSY);
13835 }
13836
13837 if (ipst->ips_ip_cgtp_filter_ops == NULL) {
13838 netstack_rele(ns);
13839 return (ENXIO);
13840 }
13841 ipst->ips_ip_cgtp_filter_ops = NULL;
13842
13843 ill_set_inputfn_all(ipst);
13844
13845 netstack_rele(ns);
13846 return (0);
13847 }
13848
13849 /*
13850 * Check whether there is a CGTP filter registration.
13851 * Returns non-zero if there is a registration, otherwise returns zero.
13852 * Note: returns zero if bad stackid.
13853 */
13854 int
13855 ip_cgtp_filter_is_registered(netstackid_t stackid)
13856 {
13857 netstack_t *ns;
13858 ip_stack_t *ipst;
13859 int ret;
13860
13861 ns = netstack_find_by_stackid(stackid);
13862 if (ns == NULL)
13863 return (0);
13864 ipst = ns->netstack_ip;
13865 ASSERT(ipst != NULL);
13866
13867 if (ipst->ips_ip_cgtp_filter_ops != NULL)
13868 ret = 1;
13869 else
13870 ret = 0;
13871
13872 netstack_rele(ns);
13873 return (ret);
13874 }
13875
13876 static int
13877 ip_squeue_switch(int val)
13878 {
13879 int rval;
13880
13881 switch (val) {
13882 case IP_SQUEUE_ENTER_NODRAIN:
13883 rval = SQ_NODRAIN;
13884 break;
13885 case IP_SQUEUE_ENTER:
13886 rval = SQ_PROCESS;
13887 break;
13888 case IP_SQUEUE_FILL:
13889 default:
13890 rval = SQ_FILL;
13891 break;
13892 }
13893 return (rval);
13894 }
13895
13896 static void *
13897 ip_kstat2_init(netstackid_t stackid, ip_stat_t *ip_statisticsp)
13898 {
13899 kstat_t *ksp;
13900
13901 ip_stat_t template = {
13902 { "ip_udp_fannorm", KSTAT_DATA_UINT64 },
13903 { "ip_udp_fanmb", KSTAT_DATA_UINT64 },
13904 { "ip_recv_pullup", KSTAT_DATA_UINT64 },
13905 { "ip_db_ref", KSTAT_DATA_UINT64 },
13906 { "ip_notaligned", KSTAT_DATA_UINT64 },
13907 { "ip_multimblk", KSTAT_DATA_UINT64 },
13908 { "ip_opt", KSTAT_DATA_UINT64 },
13909 { "ipsec_proto_ahesp", KSTAT_DATA_UINT64 },
13910 { "ip_conn_flputbq", KSTAT_DATA_UINT64 },
13911 { "ip_conn_walk_drain", KSTAT_DATA_UINT64 },
13912 { "ip_out_sw_cksum", KSTAT_DATA_UINT64 },
13913 { "ip_out_sw_cksum_bytes", KSTAT_DATA_UINT64 },
13914 { "ip_in_sw_cksum", KSTAT_DATA_UINT64 },
13915 { "ip_ire_reclaim_calls", KSTAT_DATA_UINT64 },
13916 { "ip_ire_reclaim_deleted", KSTAT_DATA_UINT64 },
13917 { "ip_nce_reclaim_calls", KSTAT_DATA_UINT64 },
13918 { "ip_nce_reclaim_deleted", KSTAT_DATA_UINT64 },
13919 { "ip_dce_reclaim_calls", KSTAT_DATA_UINT64 },
13920 { "ip_dce_reclaim_deleted", KSTAT_DATA_UINT64 },
13921 { "ip_tcp_in_full_hw_cksum_err", KSTAT_DATA_UINT64 },
13922 { "ip_tcp_in_part_hw_cksum_err", KSTAT_DATA_UINT64 },
13923 { "ip_tcp_in_sw_cksum_err", KSTAT_DATA_UINT64 },
13924 { "ip_udp_in_full_hw_cksum_err", KSTAT_DATA_UINT64 },
13925 { "ip_udp_in_part_hw_cksum_err", KSTAT_DATA_UINT64 },
13926 { "ip_udp_in_sw_cksum_err", KSTAT_DATA_UINT64 },
13927 { "conn_in_recvdstaddr", KSTAT_DATA_UINT64 },
13928 { "conn_in_recvopts", KSTAT_DATA_UINT64 },
13929 { "conn_in_recvif", KSTAT_DATA_UINT64 },
13930 { "conn_in_recvslla", KSTAT_DATA_UINT64 },
13931 { "conn_in_recvucred", KSTAT_DATA_UINT64 },
13932 { "conn_in_recvttl", KSTAT_DATA_UINT64 },
13933 { "conn_in_recvhopopts", KSTAT_DATA_UINT64 },
13934 { "conn_in_recvhoplimit", KSTAT_DATA_UINT64 },
13935 { "conn_in_recvdstopts", KSTAT_DATA_UINT64 },
13936 { "conn_in_recvrthdrdstopts", KSTAT_DATA_UINT64 },
13937 { "conn_in_recvrthdr", KSTAT_DATA_UINT64 },
13938 { "conn_in_recvpktinfo", KSTAT_DATA_UINT64 },
13939 { "conn_in_recvtclass", KSTAT_DATA_UINT64 },
13940 { "conn_in_timestamp", KSTAT_DATA_UINT64 },
13941 };
13942
13943 ksp = kstat_create_netstack("ip", 0, "ipstat", "net",
13944 KSTAT_TYPE_NAMED, sizeof (template) / sizeof (kstat_named_t),
13945 KSTAT_FLAG_VIRTUAL, stackid);
13946
13947 if (ksp == NULL)
13948 return (NULL);
13949
13950 bcopy(&template, ip_statisticsp, sizeof (template));
13951 ksp->ks_data = (void *)ip_statisticsp;
13952 ksp->ks_private = (void *)(uintptr_t)stackid;
13953
13954 kstat_install(ksp);
13955 return (ksp);
13956 }
13957
13958 static void
13959 ip_kstat2_fini(netstackid_t stackid, kstat_t *ksp)
13960 {
13961 if (ksp != NULL) {
13962 ASSERT(stackid == (netstackid_t)(uintptr_t)ksp->ks_private);
13963 kstat_delete_netstack(ksp, stackid);
13964 }
13965 }
13966
13967 static void *
13968 ip_kstat_init(netstackid_t stackid, ip_stack_t *ipst)
13969 {
13970 kstat_t *ksp;
13971
13972 ip_named_kstat_t template = {
13973 { "forwarding", KSTAT_DATA_UINT32, 0 },
13974 { "defaultTTL", KSTAT_DATA_UINT32, 0 },
13975 { "inReceives", KSTAT_DATA_UINT64, 0 },
13976 { "inHdrErrors", KSTAT_DATA_UINT32, 0 },
13977 { "inAddrErrors", KSTAT_DATA_UINT32, 0 },
13978 { "forwDatagrams", KSTAT_DATA_UINT64, 0 },
13979 { "inUnknownProtos", KSTAT_DATA_UINT32, 0 },
13980 { "inDiscards", KSTAT_DATA_UINT32, 0 },
13981 { "inDelivers", KSTAT_DATA_UINT64, 0 },
13982 { "outRequests", KSTAT_DATA_UINT64, 0 },
13983 { "outDiscards", KSTAT_DATA_UINT32, 0 },
13984 { "outNoRoutes", KSTAT_DATA_UINT32, 0 },
13985 { "reasmTimeout", KSTAT_DATA_UINT32, 0 },
13986 { "reasmReqds", KSTAT_DATA_UINT32, 0 },
13987 { "reasmOKs", KSTAT_DATA_UINT32, 0 },
13988 { "reasmFails", KSTAT_DATA_UINT32, 0 },
13989 { "fragOKs", KSTAT_DATA_UINT32, 0 },
13990 { "fragFails", KSTAT_DATA_UINT32, 0 },
13991 { "fragCreates", KSTAT_DATA_UINT32, 0 },
13992 { "addrEntrySize", KSTAT_DATA_INT32, 0 },
13993 { "routeEntrySize", KSTAT_DATA_INT32, 0 },
13994 { "netToMediaEntrySize", KSTAT_DATA_INT32, 0 },
13995 { "routingDiscards", KSTAT_DATA_UINT32, 0 },
13996 { "inErrs", KSTAT_DATA_UINT32, 0 },
13997 { "noPorts", KSTAT_DATA_UINT32, 0 },
13998 { "inCksumErrs", KSTAT_DATA_UINT32, 0 },
13999 { "reasmDuplicates", KSTAT_DATA_UINT32, 0 },
14000 { "reasmPartDups", KSTAT_DATA_UINT32, 0 },
14001 { "forwProhibits", KSTAT_DATA_UINT32, 0 },
14002 { "udpInCksumErrs", KSTAT_DATA_UINT32, 0 },
14003 { "udpInOverflows", KSTAT_DATA_UINT32, 0 },
14004 { "rawipInOverflows", KSTAT_DATA_UINT32, 0 },
14005 { "ipsecInSucceeded", KSTAT_DATA_UINT32, 0 },
14006 { "ipsecInFailed", KSTAT_DATA_INT32, 0 },
14007 { "memberEntrySize", KSTAT_DATA_INT32, 0 },
14008 { "inIPv6", KSTAT_DATA_UINT32, 0 },
14009 { "outIPv6", KSTAT_DATA_UINT32, 0 },
14010 { "outSwitchIPv6", KSTAT_DATA_UINT32, 0 },
14011 };
14012
14013 ksp = kstat_create_netstack("ip", 0, "ip", "mib2", KSTAT_TYPE_NAMED,
14014 NUM_OF_FIELDS(ip_named_kstat_t), 0, stackid);
14015 if (ksp == NULL || ksp->ks_data == NULL)
14016 return (NULL);
14017
14018 template.forwarding.value.ui32 = WE_ARE_FORWARDING(ipst) ? 1:2;
14019 template.defaultTTL.value.ui32 = (uint32_t)ipst->ips_ip_def_ttl;
14020 template.reasmTimeout.value.ui32 = ipst->ips_ip_reassembly_timeout;
14021 template.addrEntrySize.value.i32 = sizeof (mib2_ipAddrEntry_t);
14022 template.routeEntrySize.value.i32 = sizeof (mib2_ipRouteEntry_t);
14023
14024 template.netToMediaEntrySize.value.i32 =
14025 sizeof (mib2_ipNetToMediaEntry_t);
14026
14027 template.memberEntrySize.value.i32 = sizeof (ipv6_member_t);
14028
14029 bcopy(&template, ksp->ks_data, sizeof (template));
14030 ksp->ks_update = ip_kstat_update;
14031 ksp->ks_private = (void *)(uintptr_t)stackid;
14032
14033 kstat_install(ksp);
14034 return (ksp);
14035 }
14036
14037 static void
14038 ip_kstat_fini(netstackid_t stackid, kstat_t *ksp)
14039 {
14040 if (ksp != NULL) {
14041 ASSERT(stackid == (netstackid_t)(uintptr_t)ksp->ks_private);
14042 kstat_delete_netstack(ksp, stackid);
14043 }
14044 }
14045
14046 static int
14047 ip_kstat_update(kstat_t *kp, int rw)
14048 {
14049 ip_named_kstat_t *ipkp;
14050 mib2_ipIfStatsEntry_t ipmib;
14051 ill_walk_context_t ctx;
14052 ill_t *ill;
14053 netstackid_t stackid = (zoneid_t)(uintptr_t)kp->ks_private;
14054 netstack_t *ns;
14055 ip_stack_t *ipst;
14056
14057 if (kp == NULL || kp->ks_data == NULL)
14058 return (EIO);
14059
14060 if (rw == KSTAT_WRITE)
14061 return (EACCES);
14062
14063 ns = netstack_find_by_stackid(stackid);
14064 if (ns == NULL)
14065 return (-1);
14066 ipst = ns->netstack_ip;
14067 if (ipst == NULL) {
14068 netstack_rele(ns);
14069 return (-1);
14070 }
14071 ipkp = (ip_named_kstat_t *)kp->ks_data;
14072
14073 bcopy(&ipst->ips_ip_mib, &ipmib, sizeof (ipmib));
14074 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
14075 ill = ILL_START_WALK_V4(&ctx, ipst);
14076 for (; ill != NULL; ill = ill_next(&ctx, ill))
14077 ip_mib2_add_ip_stats(&ipmib, ill->ill_ip_mib);
14078 rw_exit(&ipst->ips_ill_g_lock);
14079
14080 ipkp->forwarding.value.ui32 = ipmib.ipIfStatsForwarding;
14081 ipkp->defaultTTL.value.ui32 = ipmib.ipIfStatsDefaultTTL;
14082 ipkp->inReceives.value.ui64 = ipmib.ipIfStatsHCInReceives;
14083 ipkp->inHdrErrors.value.ui32 = ipmib.ipIfStatsInHdrErrors;
14084 ipkp->inAddrErrors.value.ui32 = ipmib.ipIfStatsInAddrErrors;
14085 ipkp->forwDatagrams.value.ui64 = ipmib.ipIfStatsHCOutForwDatagrams;
14086 ipkp->inUnknownProtos.value.ui32 = ipmib.ipIfStatsInUnknownProtos;
14087 ipkp->inDiscards.value.ui32 = ipmib.ipIfStatsInDiscards;
14088 ipkp->inDelivers.value.ui64 = ipmib.ipIfStatsHCInDelivers;
14089 ipkp->outRequests.value.ui64 = ipmib.ipIfStatsHCOutRequests;
14090 ipkp->outDiscards.value.ui32 = ipmib.ipIfStatsOutDiscards;
14091 ipkp->outNoRoutes.value.ui32 = ipmib.ipIfStatsOutNoRoutes;
14092 ipkp->reasmTimeout.value.ui32 = ipst->ips_ip_reassembly_timeout;
14093 ipkp->reasmReqds.value.ui32 = ipmib.ipIfStatsReasmReqds;
14094 ipkp->reasmOKs.value.ui32 = ipmib.ipIfStatsReasmOKs;
14095 ipkp->reasmFails.value.ui32 = ipmib.ipIfStatsReasmFails;
14096 ipkp->fragOKs.value.ui32 = ipmib.ipIfStatsOutFragOKs;
14097 ipkp->fragFails.value.ui32 = ipmib.ipIfStatsOutFragFails;
14098 ipkp->fragCreates.value.ui32 = ipmib.ipIfStatsOutFragCreates;
14099
14100 ipkp->routingDiscards.value.ui32 = 0;
14101 ipkp->inErrs.value.ui32 = ipmib.tcpIfStatsInErrs;
14102 ipkp->noPorts.value.ui32 = ipmib.udpIfStatsNoPorts;
14103 ipkp->inCksumErrs.value.ui32 = ipmib.ipIfStatsInCksumErrs;
14104 ipkp->reasmDuplicates.value.ui32 = ipmib.ipIfStatsReasmDuplicates;
14105 ipkp->reasmPartDups.value.ui32 = ipmib.ipIfStatsReasmPartDups;
14106 ipkp->forwProhibits.value.ui32 = ipmib.ipIfStatsForwProhibits;
14107 ipkp->udpInCksumErrs.value.ui32 = ipmib.udpIfStatsInCksumErrs;
14108 ipkp->udpInOverflows.value.ui32 = ipmib.udpIfStatsInOverflows;
14109 ipkp->rawipInOverflows.value.ui32 = ipmib.rawipIfStatsInOverflows;
14110 ipkp->ipsecInSucceeded.value.ui32 = ipmib.ipsecIfStatsInSucceeded;
14111 ipkp->ipsecInFailed.value.i32 = ipmib.ipsecIfStatsInFailed;
14112
14113 ipkp->inIPv6.value.ui32 = ipmib.ipIfStatsInWrongIPVersion;
14114 ipkp->outIPv6.value.ui32 = ipmib.ipIfStatsOutWrongIPVersion;
14115 ipkp->outSwitchIPv6.value.ui32 = ipmib.ipIfStatsOutSwitchIPVersion;
14116
14117 netstack_rele(ns);
14118
14119 return (0);
14120 }
14121
14122 static void *
14123 icmp_kstat_init(netstackid_t stackid)
14124 {
14125 kstat_t *ksp;
14126
14127 icmp_named_kstat_t template = {
14128 { "inMsgs", KSTAT_DATA_UINT32 },
14129 { "inErrors", KSTAT_DATA_UINT32 },
14130 { "inDestUnreachs", KSTAT_DATA_UINT32 },
14131 { "inTimeExcds", KSTAT_DATA_UINT32 },
14132 { "inParmProbs", KSTAT_DATA_UINT32 },
14133 { "inSrcQuenchs", KSTAT_DATA_UINT32 },
14134 { "inRedirects", KSTAT_DATA_UINT32 },
14135 { "inEchos", KSTAT_DATA_UINT32 },
14136 { "inEchoReps", KSTAT_DATA_UINT32 },
14137 { "inTimestamps", KSTAT_DATA_UINT32 },
14138 { "inTimestampReps", KSTAT_DATA_UINT32 },
14139 { "inAddrMasks", KSTAT_DATA_UINT32 },
14140 { "inAddrMaskReps", KSTAT_DATA_UINT32 },
14141 { "outMsgs", KSTAT_DATA_UINT32 },
14142 { "outErrors", KSTAT_DATA_UINT32 },
14143 { "outDestUnreachs", KSTAT_DATA_UINT32 },
14144 { "outTimeExcds", KSTAT_DATA_UINT32 },
14145 { "outParmProbs", KSTAT_DATA_UINT32 },
14146 { "outSrcQuenchs", KSTAT_DATA_UINT32 },
14147 { "outRedirects", KSTAT_DATA_UINT32 },
14148 { "outEchos", KSTAT_DATA_UINT32 },
14149 { "outEchoReps", KSTAT_DATA_UINT32 },
14150 { "outTimestamps", KSTAT_DATA_UINT32 },
14151 { "outTimestampReps", KSTAT_DATA_UINT32 },
14152 { "outAddrMasks", KSTAT_DATA_UINT32 },
14153 { "outAddrMaskReps", KSTAT_DATA_UINT32 },
14154 { "inChksumErrs", KSTAT_DATA_UINT32 },
14155 { "inUnknowns", KSTAT_DATA_UINT32 },
14156 { "inFragNeeded", KSTAT_DATA_UINT32 },
14157 { "outFragNeeded", KSTAT_DATA_UINT32 },
14158 { "outDrops", KSTAT_DATA_UINT32 },
14159 { "inOverFlows", KSTAT_DATA_UINT32 },
14160 { "inBadRedirects", KSTAT_DATA_UINT32 },
14161 };
14162
14163 ksp = kstat_create_netstack("ip", 0, "icmp", "mib2", KSTAT_TYPE_NAMED,
14164 NUM_OF_FIELDS(icmp_named_kstat_t), 0, stackid);
14165 if (ksp == NULL || ksp->ks_data == NULL)
14166 return (NULL);
14167
14168 bcopy(&template, ksp->ks_data, sizeof (template));
14169
14170 ksp->ks_update = icmp_kstat_update;
14171 ksp->ks_private = (void *)(uintptr_t)stackid;
14172
14173 kstat_install(ksp);
14174 return (ksp);
14175 }
14176
14177 static void
14178 icmp_kstat_fini(netstackid_t stackid, kstat_t *ksp)
14179 {
14180 if (ksp != NULL) {
14181 ASSERT(stackid == (netstackid_t)(uintptr_t)ksp->ks_private);
14182 kstat_delete_netstack(ksp, stackid);
14183 }
14184 }
14185
14186 static int
14187 icmp_kstat_update(kstat_t *kp, int rw)
14188 {
14189 icmp_named_kstat_t *icmpkp;
14190 netstackid_t stackid = (zoneid_t)(uintptr_t)kp->ks_private;
14191 netstack_t *ns;
14192 ip_stack_t *ipst;
14193
14194 if ((kp == NULL) || (kp->ks_data == NULL))
14195 return (EIO);
14196
14197 if (rw == KSTAT_WRITE)
14198 return (EACCES);
14199
14200 ns = netstack_find_by_stackid(stackid);
14201 if (ns == NULL)
14202 return (-1);
14203 ipst = ns->netstack_ip;
14204 if (ipst == NULL) {
14205 netstack_rele(ns);
14206 return (-1);
14207 }
14208 icmpkp = (icmp_named_kstat_t *)kp->ks_data;
14209
14210 icmpkp->inMsgs.value.ui32 = ipst->ips_icmp_mib.icmpInMsgs;
14211 icmpkp->inErrors.value.ui32 = ipst->ips_icmp_mib.icmpInErrors;
14212 icmpkp->inDestUnreachs.value.ui32 =
14213 ipst->ips_icmp_mib.icmpInDestUnreachs;
14214 icmpkp->inTimeExcds.value.ui32 = ipst->ips_icmp_mib.icmpInTimeExcds;
14215 icmpkp->inParmProbs.value.ui32 = ipst->ips_icmp_mib.icmpInParmProbs;
14216 icmpkp->inSrcQuenchs.value.ui32 = ipst->ips_icmp_mib.icmpInSrcQuenchs;
14217 icmpkp->inRedirects.value.ui32 = ipst->ips_icmp_mib.icmpInRedirects;
14218 icmpkp->inEchos.value.ui32 = ipst->ips_icmp_mib.icmpInEchos;
14219 icmpkp->inEchoReps.value.ui32 = ipst->ips_icmp_mib.icmpInEchoReps;
14220 icmpkp->inTimestamps.value.ui32 = ipst->ips_icmp_mib.icmpInTimestamps;
14221 icmpkp->inTimestampReps.value.ui32 =
14222 ipst->ips_icmp_mib.icmpInTimestampReps;
14223 icmpkp->inAddrMasks.value.ui32 = ipst->ips_icmp_mib.icmpInAddrMasks;
14224 icmpkp->inAddrMaskReps.value.ui32 =
14225 ipst->ips_icmp_mib.icmpInAddrMaskReps;
14226 icmpkp->outMsgs.value.ui32 = ipst->ips_icmp_mib.icmpOutMsgs;
14227 icmpkp->outErrors.value.ui32 = ipst->ips_icmp_mib.icmpOutErrors;
14228 icmpkp->outDestUnreachs.value.ui32 =
14229 ipst->ips_icmp_mib.icmpOutDestUnreachs;
14230 icmpkp->outTimeExcds.value.ui32 = ipst->ips_icmp_mib.icmpOutTimeExcds;
14231 icmpkp->outParmProbs.value.ui32 = ipst->ips_icmp_mib.icmpOutParmProbs;
14232 icmpkp->outSrcQuenchs.value.ui32 =
14233 ipst->ips_icmp_mib.icmpOutSrcQuenchs;
14234 icmpkp->outRedirects.value.ui32 = ipst->ips_icmp_mib.icmpOutRedirects;
14235 icmpkp->outEchos.value.ui32 = ipst->ips_icmp_mib.icmpOutEchos;
14236 icmpkp->outEchoReps.value.ui32 = ipst->ips_icmp_mib.icmpOutEchoReps;
14237 icmpkp->outTimestamps.value.ui32 =
14238 ipst->ips_icmp_mib.icmpOutTimestamps;
14239 icmpkp->outTimestampReps.value.ui32 =
14240 ipst->ips_icmp_mib.icmpOutTimestampReps;
14241 icmpkp->outAddrMasks.value.ui32 =
14242 ipst->ips_icmp_mib.icmpOutAddrMasks;
14243 icmpkp->outAddrMaskReps.value.ui32 =
14244 ipst->ips_icmp_mib.icmpOutAddrMaskReps;
14245 icmpkp->inCksumErrs.value.ui32 = ipst->ips_icmp_mib.icmpInCksumErrs;
14246 icmpkp->inUnknowns.value.ui32 = ipst->ips_icmp_mib.icmpInUnknowns;
14247 icmpkp->inFragNeeded.value.ui32 = ipst->ips_icmp_mib.icmpInFragNeeded;
14248 icmpkp->outFragNeeded.value.ui32 =
14249 ipst->ips_icmp_mib.icmpOutFragNeeded;
14250 icmpkp->outDrops.value.ui32 = ipst->ips_icmp_mib.icmpOutDrops;
14251 icmpkp->inOverflows.value.ui32 = ipst->ips_icmp_mib.icmpInOverflows;
14252 icmpkp->inBadRedirects.value.ui32 =
14253 ipst->ips_icmp_mib.icmpInBadRedirects;
14254
14255 netstack_rele(ns);
14256 return (0);
14257 }
14258
14259 /*
14260 * This is the fanout function for raw socket opened for SCTP. Note
14261 * that it is called after SCTP checks that there is no socket which
14262 * wants a packet. Then before SCTP handles this out of the blue packet,
14263 * this function is called to see if there is any raw socket for SCTP.
14264 * If there is and it is bound to the correct address, the packet will
14265 * be sent to that socket. Note that only one raw socket can be bound to
14266 * a port. This is assured in ipcl_sctp_hash_insert();
14267 */
14268 void
14269 ip_fanout_sctp_raw(mblk_t *mp, ipha_t *ipha, ip6_t *ip6h, uint32_t ports,
14270 ip_recv_attr_t *ira)
14271 {
14272 conn_t *connp;
14273 queue_t *rq;
14274 boolean_t secure;
14275 ill_t *ill = ira->ira_ill;
14276 ip_stack_t *ipst = ill->ill_ipst;
14277 ipsec_stack_t *ipss = ipst->ips_netstack->netstack_ipsec;
14278 sctp_stack_t *sctps = ipst->ips_netstack->netstack_sctp;
14279 iaflags_t iraflags = ira->ira_flags;
14280 ill_t *rill = ira->ira_rill;
14281
14282 secure = iraflags & IRAF_IPSEC_SECURE;
14283
14284 connp = ipcl_classify_raw(mp, IPPROTO_SCTP, ports, ipha, ip6h,
14285 ira, ipst);
14286 if (connp == NULL) {
14287 /*
14288 * Although raw sctp is not summed, OOB chunks must be.
14289 * Drop the packet here if the sctp checksum failed.
14290 */
14291 if (iraflags & IRAF_SCTP_CSUM_ERR) {
14292 SCTPS_BUMP_MIB(sctps, sctpChecksumError);
14293 freemsg(mp);
14294 return;
14295 }
14296 ira->ira_ill = ira->ira_rill = NULL;
14297 sctp_ootb_input(mp, ira, ipst);
14298 ira->ira_ill = ill;
14299 ira->ira_rill = rill;
14300 return;
14301 }
14302 rq = connp->conn_rq;
14303 if (IPCL_IS_NONSTR(connp) ? connp->conn_flow_cntrld : !canputnext(rq)) {
14304 CONN_DEC_REF(connp);
14305 BUMP_MIB(ill->ill_ip_mib, rawipIfStatsInOverflows);
14306 freemsg(mp);
14307 return;
14308 }
14309 if (((iraflags & IRAF_IS_IPV4) ?
14310 CONN_INBOUND_POLICY_PRESENT(connp, ipss) :
14311 CONN_INBOUND_POLICY_PRESENT_V6(connp, ipss)) ||
14312 secure) {
14313 mp = ipsec_check_inbound_policy(mp, connp, ipha,
14314 ip6h, ira);
14315 if (mp == NULL) {
14316 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
14317 /* Note that mp is NULL */
14318 ip_drop_input("ipIfStatsInDiscards", mp, ill);
14319 CONN_DEC_REF(connp);
14320 return;
14321 }
14322 }
14323
14324 if (iraflags & IRAF_ICMP_ERROR) {
14325 (connp->conn_recvicmp)(connp, mp, NULL, ira);
14326 } else {
14327 ill_t *rill = ira->ira_rill;
14328
14329 BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCInDelivers);
14330 /* This is the SOCK_RAW, IPPROTO_SCTP case. */
14331 ira->ira_ill = ira->ira_rill = NULL;
14332 (connp->conn_recv)(connp, mp, NULL, ira);
14333 ira->ira_ill = ill;
14334 ira->ira_rill = rill;
14335 }
14336 CONN_DEC_REF(connp);
14337 }
14338
14339 /*
14340 * Free a packet that has the link-layer dl_unitdata_req_t or fast-path
14341 * header before the ip payload.
14342 */
14343 static void
14344 ip_xmit_flowctl_drop(ill_t *ill, mblk_t *mp, boolean_t is_fp_mp, int fp_mp_len)
14345 {
14346 int len = (mp->b_wptr - mp->b_rptr);
14347 mblk_t *ip_mp;
14348
14349 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
14350 if (is_fp_mp || len != fp_mp_len) {
14351 if (len > fp_mp_len) {
14352 /*
14353 * fastpath header and ip header in the first mblk
14354 */
14355 mp->b_rptr += fp_mp_len;
14356 } else {
14357 /*
14358 * ip_xmit_attach_llhdr had to prepend an mblk to
14359 * attach the fastpath header before ip header.
14360 */
14361 ip_mp = mp->b_cont;
14362 freeb(mp);
14363 mp = ip_mp;
14364 mp->b_rptr += (fp_mp_len - len);
14365 }
14366 } else {
14367 ip_mp = mp->b_cont;
14368 freeb(mp);
14369 mp = ip_mp;
14370 }
14371 ip_drop_output("ipIfStatsOutDiscards - flow ctl", mp, ill);
14372 freemsg(mp);
14373 }
14374
14375 /*
14376 * Normal post fragmentation function.
14377 *
14378 * Send a packet using the passed in nce. This handles both IPv4 and IPv6
14379 * using the same state machine.
14380 *
14381 * We return an error on failure. In particular we return EWOULDBLOCK
14382 * when the driver flow controls. In that case this ensures that ip_wsrv runs
14383 * (currently by canputnext failure resulting in backenabling from GLD.)
14384 * This allows the callers of conn_ip_output() to use EWOULDBLOCK as an
14385 * indication that they can flow control until ip_wsrv() tells then to restart.
14386 *
14387 * If the nce passed by caller is incomplete, this function
14388 * queues the packet and if necessary, sends ARP request and bails.
14389 * If the Neighbor Cache passed is fully resolved, we simply prepend
14390 * the link-layer header to the packet, do ipsec hw acceleration
14391 * work if necessary, and send the packet out on the wire.
14392 */
14393 /* ARGSUSED6 */
14394 int
14395 ip_xmit(mblk_t *mp, nce_t *nce, iaflags_t ixaflags, uint_t pkt_len,
14396 uint32_t xmit_hint, zoneid_t szone, zoneid_t nolzid, uintptr_t *ixacookie)
14397 {
14398 queue_t *wq;
14399 ill_t *ill = nce->nce_ill;
14400 ip_stack_t *ipst = ill->ill_ipst;
14401 uint64_t delta;
14402 boolean_t isv6 = ill->ill_isv6;
14403 boolean_t fp_mp;
14404 ncec_t *ncec = nce->nce_common;
14405 int64_t now = LBOLT_FASTPATH64;
14406 boolean_t is_probe;
14407
14408 DTRACE_PROBE1(ip__xmit, nce_t *, nce);
14409
14410 ASSERT(mp != NULL);
14411 ASSERT(mp->b_datap->db_type == M_DATA);
14412 ASSERT(pkt_len == msgdsize(mp));
14413
14414 /*
14415 * If we have already been here and are coming back after ARP/ND.
14416 * the IXAF_NO_TRACE flag is set. We skip FW_HOOKS, DTRACE and ipobs
14417 * in that case since they have seen the packet when it came here
14418 * the first time.
14419 */
14420 if (ixaflags & IXAF_NO_TRACE)
14421 goto sendit;
14422
14423 if (ixaflags & IXAF_IS_IPV4) {
14424 ipha_t *ipha = (ipha_t *)mp->b_rptr;
14425
14426 ASSERT(!isv6);
14427 ASSERT(pkt_len == ntohs(((ipha_t *)mp->b_rptr)->ipha_length));
14428 if (HOOKS4_INTERESTED_PHYSICAL_OUT(ipst) &&
14429 !(ixaflags & IXAF_NO_PFHOOK)) {
14430 int error;
14431
14432 FW_HOOKS(ipst->ips_ip4_physical_out_event,
14433 ipst->ips_ipv4firewall_physical_out,
14434 NULL, ill, ipha, mp, mp, 0, ipst, error);
14435 DTRACE_PROBE1(ip4__physical__out__end,
14436 mblk_t *, mp);
14437 if (mp == NULL)
14438 return (error);
14439
14440 /* The length could have changed */
14441 pkt_len = msgdsize(mp);
14442 }
14443 if (ipst->ips_ip4_observe.he_interested) {
14444 /*
14445 * Note that for TX the zoneid is the sending
14446 * zone, whether or not MLP is in play.
14447 * Since the szone argument is the IP zoneid (i.e.,
14448 * zero for exclusive-IP zones) and ipobs wants
14449 * the system zoneid, we map it here.
14450 */
14451 szone = IP_REAL_ZONEID(szone, ipst);
14452
14453 /*
14454 * On the outbound path the destination zone will be
14455 * unknown as we're sending this packet out on the
14456 * wire.
14457 */
14458 ipobs_hook(mp, IPOBS_HOOK_OUTBOUND, szone, ALL_ZONES,
14459 ill, ipst);
14460 }
14461 DTRACE_IP7(send, mblk_t *, mp, conn_t *, NULL,
14462 void_ip_t *, ipha, __dtrace_ipsr_ill_t *, ill,
14463 ipha_t *, ipha, ip6_t *, NULL, int, 0);
14464 } else {
14465 ip6_t *ip6h = (ip6_t *)mp->b_rptr;
14466
14467 ASSERT(isv6);
14468 ASSERT(pkt_len ==
14469 ntohs(((ip6_t *)mp->b_rptr)->ip6_plen) + IPV6_HDR_LEN);
14470 if (HOOKS6_INTERESTED_PHYSICAL_OUT(ipst) &&
14471 !(ixaflags & IXAF_NO_PFHOOK)) {
14472 int error;
14473
14474 FW_HOOKS6(ipst->ips_ip6_physical_out_event,
14475 ipst->ips_ipv6firewall_physical_out,
14476 NULL, ill, ip6h, mp, mp, 0, ipst, error);
14477 DTRACE_PROBE1(ip6__physical__out__end,
14478 mblk_t *, mp);
14479 if (mp == NULL)
14480 return (error);
14481
14482 /* The length could have changed */
14483 pkt_len = msgdsize(mp);
14484 }
14485 if (ipst->ips_ip6_observe.he_interested) {
14486 /* See above */
14487 szone = IP_REAL_ZONEID(szone, ipst);
14488
14489 ipobs_hook(mp, IPOBS_HOOK_OUTBOUND, szone, ALL_ZONES,
14490 ill, ipst);
14491 }
14492 DTRACE_IP7(send, mblk_t *, mp, conn_t *, NULL,
14493 void_ip_t *, ip6h, __dtrace_ipsr_ill_t *, ill,
14494 ipha_t *, NULL, ip6_t *, ip6h, int, 0);
14495 }
14496
14497 sendit:
14498 /*
14499 * We check the state without a lock because the state can never
14500 * move "backwards" to initial or incomplete.
14501 */
14502 switch (ncec->ncec_state) {
14503 case ND_REACHABLE:
14504 case ND_STALE:
14505 case ND_DELAY:
14506 case ND_PROBE:
14507 mp = ip_xmit_attach_llhdr(mp, nce);
14508 if (mp == NULL) {
14509 /*
14510 * ip_xmit_attach_llhdr has increased
14511 * ipIfStatsOutDiscards and called ip_drop_output()
14512 */
14513 return (ENOBUFS);
14514 }
14515 /*
14516 * check if nce_fastpath completed and we tagged on a
14517 * copy of nce_fp_mp in ip_xmit_attach_llhdr().
14518 */
14519 fp_mp = (mp->b_datap->db_type == M_DATA);
14520
14521 if (fp_mp &&
14522 (ill->ill_capabilities & ILL_CAPAB_DLD_DIRECT)) {
14523 ill_dld_direct_t *idd;
14524
14525 idd = &ill->ill_dld_capab->idc_direct;
14526 /*
14527 * Send the packet directly to DLD, where it
14528 * may be queued depending on the availability
14529 * of transmit resources at the media layer.
14530 * Return value should be taken into
14531 * account and flow control the TCP.
14532 */
14533 BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCOutTransmits);
14534 UPDATE_MIB(ill->ill_ip_mib, ipIfStatsHCOutOctets,
14535 pkt_len);
14536
14537 if (ixaflags & IXAF_NO_DEV_FLOW_CTL) {
14538 (void) idd->idd_tx_df(idd->idd_tx_dh, mp,
14539 (uintptr_t)xmit_hint, IP_DROP_ON_NO_DESC);
14540 } else {
14541 uintptr_t cookie;
14542
14543 if ((cookie = idd->idd_tx_df(idd->idd_tx_dh,
14544 mp, (uintptr_t)xmit_hint, 0)) != 0) {
14545 if (ixacookie != NULL)
14546 *ixacookie = cookie;
14547 return (EWOULDBLOCK);
14548 }
14549 }
14550 } else {
14551 wq = ill->ill_wq;
14552
14553 if (!(ixaflags & IXAF_NO_DEV_FLOW_CTL) &&
14554 !canputnext(wq)) {
14555 if (ixacookie != NULL)
14556 *ixacookie = 0;
14557 ip_xmit_flowctl_drop(ill, mp, fp_mp,
14558 nce->nce_fp_mp != NULL ?
14559 MBLKL(nce->nce_fp_mp) : 0);
14560 return (EWOULDBLOCK);
14561 }
14562 BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCOutTransmits);
14563 UPDATE_MIB(ill->ill_ip_mib, ipIfStatsHCOutOctets,
14564 pkt_len);
14565 putnext(wq, mp);
14566 }
14567
14568 /*
14569 * The rest of this function implements Neighbor Unreachability
14570 * detection. Determine if the ncec is eligible for NUD.
14571 */
14572 if (ncec->ncec_flags & NCE_F_NONUD)
14573 return (0);
14574
14575 ASSERT(ncec->ncec_state != ND_INCOMPLETE);
14576
14577 /*
14578 * Check for upper layer advice
14579 */
14580 if (ixaflags & IXAF_REACH_CONF) {
14581 timeout_id_t tid;
14582
14583 /*
14584 * It should be o.k. to check the state without
14585 * a lock here, at most we lose an advice.
14586 */
14587 ncec->ncec_last = TICK_TO_MSEC(now);
14588 if (ncec->ncec_state != ND_REACHABLE) {
14589 mutex_enter(&ncec->ncec_lock);
14590 ncec->ncec_state = ND_REACHABLE;
14591 tid = ncec->ncec_timeout_id;
14592 ncec->ncec_timeout_id = 0;
14593 mutex_exit(&ncec->ncec_lock);
14594 (void) untimeout(tid);
14595 if (ip_debug > 2) {
14596 /* ip1dbg */
14597 pr_addr_dbg("ip_xmit: state"
14598 " for %s changed to"
14599 " REACHABLE\n", AF_INET6,
14600 &ncec->ncec_addr);
14601 }
14602 }
14603 return (0);
14604 }
14605
14606 delta = TICK_TO_MSEC(now) - ncec->ncec_last;
14607 ip1dbg(("ip_xmit: delta = %" PRId64
14608 " ill_reachable_time = %d \n", delta,
14609 ill->ill_reachable_time));
14610 if (delta > (uint64_t)ill->ill_reachable_time) {
14611 mutex_enter(&ncec->ncec_lock);
14612 switch (ncec->ncec_state) {
14613 case ND_REACHABLE:
14614 ASSERT((ncec->ncec_flags & NCE_F_NONUD) == 0);
14615 /* FALLTHROUGH */
14616 case ND_STALE:
14617 /*
14618 * ND_REACHABLE is identical to
14619 * ND_STALE in this specific case. If
14620 * reachable time has expired for this
14621 * neighbor (delta is greater than
14622 * reachable time), conceptually, the
14623 * neighbor cache is no longer in
14624 * REACHABLE state, but already in
14625 * STALE state. So the correct
14626 * transition here is to ND_DELAY.
14627 */
14628 ncec->ncec_state = ND_DELAY;
14629 mutex_exit(&ncec->ncec_lock);
14630 nce_restart_timer(ncec,
14631 ipst->ips_delay_first_probe_time);
14632 if (ip_debug > 3) {
14633 /* ip2dbg */
14634 pr_addr_dbg("ip_xmit: state"
14635 " for %s changed to"
14636 " DELAY\n", AF_INET6,
14637 &ncec->ncec_addr);
14638 }
14639 break;
14640 case ND_DELAY:
14641 case ND_PROBE:
14642 mutex_exit(&ncec->ncec_lock);
14643 /* Timers have already started */
14644 break;
14645 case ND_UNREACHABLE:
14646 /*
14647 * nce_timer has detected that this ncec
14648 * is unreachable and initiated deleting
14649 * this ncec.
14650 * This is a harmless race where we found the
14651 * ncec before it was deleted and have
14652 * just sent out a packet using this
14653 * unreachable ncec.
14654 */
14655 mutex_exit(&ncec->ncec_lock);
14656 break;
14657 default:
14658 ASSERT(0);
14659 mutex_exit(&ncec->ncec_lock);
14660 }
14661 }
14662 return (0);
14663
14664 case ND_INCOMPLETE:
14665 /*
14666 * the state could have changed since we didn't hold the lock.
14667 * Re-verify state under lock.
14668 */
14669 is_probe = ipmp_packet_is_probe(mp, nce->nce_ill);
14670 mutex_enter(&ncec->ncec_lock);
14671 if (NCE_ISREACHABLE(ncec)) {
14672 mutex_exit(&ncec->ncec_lock);
14673 goto sendit;
14674 }
14675 /* queue the packet */
14676 nce_queue_mp(ncec, mp, is_probe);
14677 mutex_exit(&ncec->ncec_lock);
14678 DTRACE_PROBE2(ip__xmit__incomplete,
14679 (ncec_t *), ncec, (mblk_t *), mp);
14680 return (0);
14681
14682 case ND_INITIAL:
14683 /*
14684 * State could have changed since we didn't hold the lock, so
14685 * re-verify state.
14686 */
14687 is_probe = ipmp_packet_is_probe(mp, nce->nce_ill);
14688 mutex_enter(&ncec->ncec_lock);
14689 if (NCE_ISREACHABLE(ncec)) {
14690 mutex_exit(&ncec->ncec_lock);
14691 goto sendit;
14692 }
14693 nce_queue_mp(ncec, mp, is_probe);
14694 if (ncec->ncec_state == ND_INITIAL) {
14695 ncec->ncec_state = ND_INCOMPLETE;
14696 mutex_exit(&ncec->ncec_lock);
14697 /*
14698 * figure out the source we want to use
14699 * and resolve it.
14700 */
14701 ip_ndp_resolve(ncec);
14702 } else {
14703 mutex_exit(&ncec->ncec_lock);
14704 }
14705 return (0);
14706
14707 case ND_UNREACHABLE:
14708 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
14709 ip_drop_output("ipIfStatsOutDiscards - ND_UNREACHABLE",
14710 mp, ill);
14711 freemsg(mp);
14712 return (0);
14713
14714 default:
14715 ASSERT(0);
14716 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
14717 ip_drop_output("ipIfStatsOutDiscards - ND_other",
14718 mp, ill);
14719 freemsg(mp);
14720 return (ENETUNREACH);
14721 }
14722 }
14723
14724 /*
14725 * Return B_TRUE if the buffers differ in length or content.
14726 * This is used for comparing extension header buffers.
14727 * Note that an extension header would be declared different
14728 * even if all that changed was the next header value in that header i.e.
14729 * what really changed is the next extension header.
14730 */
14731 boolean_t
14732 ip_cmpbuf(const void *abuf, uint_t alen, boolean_t b_valid, const void *bbuf,
14733 uint_t blen)
14734 {
14735 if (!b_valid)
14736 blen = 0;
14737
14738 if (alen != blen)
14739 return (B_TRUE);
14740 if (alen == 0)
14741 return (B_FALSE); /* Both zero length */
14742 return (bcmp(abuf, bbuf, alen));
14743 }
14744
14745 /*
14746 * Preallocate memory for ip_savebuf(). Returns B_TRUE if ok.
14747 * Return B_FALSE if memory allocation fails - don't change any state!
14748 */
14749 boolean_t
14750 ip_allocbuf(void **dstp, uint_t *dstlenp, boolean_t src_valid,
14751 const void *src, uint_t srclen)
14752 {
14753 void *dst;
14754
14755 if (!src_valid)
14756 srclen = 0;
14757
14758 ASSERT(*dstlenp == 0);
14759 if (src != NULL && srclen != 0) {
14760 dst = mi_alloc(srclen, BPRI_MED);
14761 if (dst == NULL)
14762 return (B_FALSE);
14763 } else {
14764 dst = NULL;
14765 }
14766 if (*dstp != NULL)
14767 mi_free(*dstp);
14768 *dstp = dst;
14769 *dstlenp = dst == NULL ? 0 : srclen;
14770 return (B_TRUE);
14771 }
14772
14773 /*
14774 * Replace what is in *dst, *dstlen with the source.
14775 * Assumes ip_allocbuf has already been called.
14776 */
14777 void
14778 ip_savebuf(void **dstp, uint_t *dstlenp, boolean_t src_valid,
14779 const void *src, uint_t srclen)
14780 {
14781 if (!src_valid)
14782 srclen = 0;
14783
14784 ASSERT(*dstlenp == srclen);
14785 if (src != NULL && srclen != 0)
14786 bcopy(src, *dstp, srclen);
14787 }
14788
14789 /*
14790 * Free the storage pointed to by the members of an ip_pkt_t.
14791 */
14792 void
14793 ip_pkt_free(ip_pkt_t *ipp)
14794 {
14795 uint_t fields = ipp->ipp_fields;
14796
14797 if (fields & IPPF_HOPOPTS) {
14798 kmem_free(ipp->ipp_hopopts, ipp->ipp_hopoptslen);
14799 ipp->ipp_hopopts = NULL;
14800 ipp->ipp_hopoptslen = 0;
14801 }
14802 if (fields & IPPF_RTHDRDSTOPTS) {
14803 kmem_free(ipp->ipp_rthdrdstopts, ipp->ipp_rthdrdstoptslen);
14804 ipp->ipp_rthdrdstopts = NULL;
14805 ipp->ipp_rthdrdstoptslen = 0;
14806 }
14807 if (fields & IPPF_DSTOPTS) {
14808 kmem_free(ipp->ipp_dstopts, ipp->ipp_dstoptslen);
14809 ipp->ipp_dstopts = NULL;
14810 ipp->ipp_dstoptslen = 0;
14811 }
14812 if (fields & IPPF_RTHDR) {
14813 kmem_free(ipp->ipp_rthdr, ipp->ipp_rthdrlen);
14814 ipp->ipp_rthdr = NULL;
14815 ipp->ipp_rthdrlen = 0;
14816 }
14817 if (fields & IPPF_IPV4_OPTIONS) {
14818 kmem_free(ipp->ipp_ipv4_options, ipp->ipp_ipv4_options_len);
14819 ipp->ipp_ipv4_options = NULL;
14820 ipp->ipp_ipv4_options_len = 0;
14821 }
14822 if (fields & IPPF_LABEL_V4) {
14823 kmem_free(ipp->ipp_label_v4, ipp->ipp_label_len_v4);
14824 ipp->ipp_label_v4 = NULL;
14825 ipp->ipp_label_len_v4 = 0;
14826 }
14827 if (fields & IPPF_LABEL_V6) {
14828 kmem_free(ipp->ipp_label_v6, ipp->ipp_label_len_v6);
14829 ipp->ipp_label_v6 = NULL;
14830 ipp->ipp_label_len_v6 = 0;
14831 }
14832 ipp->ipp_fields &= ~(IPPF_HOPOPTS | IPPF_RTHDRDSTOPTS | IPPF_DSTOPTS |
14833 IPPF_RTHDR | IPPF_IPV4_OPTIONS | IPPF_LABEL_V4 | IPPF_LABEL_V6);
14834 }
14835
14836 /*
14837 * Copy from src to dst and allocate as needed.
14838 * Returns zero or ENOMEM.
14839 *
14840 * The caller must initialize dst to zero.
14841 */
14842 int
14843 ip_pkt_copy(ip_pkt_t *src, ip_pkt_t *dst, int kmflag)
14844 {
14845 uint_t fields = src->ipp_fields;
14846
14847 /* Start with fields that don't require memory allocation */
14848 dst->ipp_fields = fields &
14849 ~(IPPF_HOPOPTS | IPPF_RTHDRDSTOPTS | IPPF_DSTOPTS |
14850 IPPF_RTHDR | IPPF_IPV4_OPTIONS | IPPF_LABEL_V4 | IPPF_LABEL_V6);
14851
14852 dst->ipp_addr = src->ipp_addr;
14853 dst->ipp_unicast_hops = src->ipp_unicast_hops;
14854 dst->ipp_hoplimit = src->ipp_hoplimit;
14855 dst->ipp_tclass = src->ipp_tclass;
14856 dst->ipp_type_of_service = src->ipp_type_of_service;
14857
14858 if (!(fields & (IPPF_HOPOPTS | IPPF_RTHDRDSTOPTS | IPPF_DSTOPTS |
14859 IPPF_RTHDR | IPPF_IPV4_OPTIONS | IPPF_LABEL_V4 | IPPF_LABEL_V6)))
14860 return (0);
14861
14862 if (fields & IPPF_HOPOPTS) {
14863 dst->ipp_hopopts = kmem_alloc(src->ipp_hopoptslen, kmflag);
14864 if (dst->ipp_hopopts == NULL) {
14865 ip_pkt_free(dst);
14866 return (ENOMEM);
14867 }
14868 dst->ipp_fields |= IPPF_HOPOPTS;
14869 bcopy(src->ipp_hopopts, dst->ipp_hopopts,
14870 src->ipp_hopoptslen);
14871 dst->ipp_hopoptslen = src->ipp_hopoptslen;
14872 }
14873 if (fields & IPPF_RTHDRDSTOPTS) {
14874 dst->ipp_rthdrdstopts = kmem_alloc(src->ipp_rthdrdstoptslen,
14875 kmflag);
14876 if (dst->ipp_rthdrdstopts == NULL) {
14877 ip_pkt_free(dst);
14878 return (ENOMEM);
14879 }
14880 dst->ipp_fields |= IPPF_RTHDRDSTOPTS;
14881 bcopy(src->ipp_rthdrdstopts, dst->ipp_rthdrdstopts,
14882 src->ipp_rthdrdstoptslen);
14883 dst->ipp_rthdrdstoptslen = src->ipp_rthdrdstoptslen;
14884 }
14885 if (fields & IPPF_DSTOPTS) {
14886 dst->ipp_dstopts = kmem_alloc(src->ipp_dstoptslen, kmflag);
14887 if (dst->ipp_dstopts == NULL) {
14888 ip_pkt_free(dst);
14889 return (ENOMEM);
14890 }
14891 dst->ipp_fields |= IPPF_DSTOPTS;
14892 bcopy(src->ipp_dstopts, dst->ipp_dstopts,
14893 src->ipp_dstoptslen);
14894 dst->ipp_dstoptslen = src->ipp_dstoptslen;
14895 }
14896 if (fields & IPPF_RTHDR) {
14897 dst->ipp_rthdr = kmem_alloc(src->ipp_rthdrlen, kmflag);
14898 if (dst->ipp_rthdr == NULL) {
14899 ip_pkt_free(dst);
14900 return (ENOMEM);
14901 }
14902 dst->ipp_fields |= IPPF_RTHDR;
14903 bcopy(src->ipp_rthdr, dst->ipp_rthdr,
14904 src->ipp_rthdrlen);
14905 dst->ipp_rthdrlen = src->ipp_rthdrlen;
14906 }
14907 if (fields & IPPF_IPV4_OPTIONS) {
14908 dst->ipp_ipv4_options = kmem_alloc(src->ipp_ipv4_options_len,
14909 kmflag);
14910 if (dst->ipp_ipv4_options == NULL) {
14911 ip_pkt_free(dst);
14912 return (ENOMEM);
14913 }
14914 dst->ipp_fields |= IPPF_IPV4_OPTIONS;
14915 bcopy(src->ipp_ipv4_options, dst->ipp_ipv4_options,
14916 src->ipp_ipv4_options_len);
14917 dst->ipp_ipv4_options_len = src->ipp_ipv4_options_len;
14918 }
14919 if (fields & IPPF_LABEL_V4) {
14920 dst->ipp_label_v4 = kmem_alloc(src->ipp_label_len_v4, kmflag);
14921 if (dst->ipp_label_v4 == NULL) {
14922 ip_pkt_free(dst);
14923 return (ENOMEM);
14924 }
14925 dst->ipp_fields |= IPPF_LABEL_V4;
14926 bcopy(src->ipp_label_v4, dst->ipp_label_v4,
14927 src->ipp_label_len_v4);
14928 dst->ipp_label_len_v4 = src->ipp_label_len_v4;
14929 }
14930 if (fields & IPPF_LABEL_V6) {
14931 dst->ipp_label_v6 = kmem_alloc(src->ipp_label_len_v6, kmflag);
14932 if (dst->ipp_label_v6 == NULL) {
14933 ip_pkt_free(dst);
14934 return (ENOMEM);
14935 }
14936 dst->ipp_fields |= IPPF_LABEL_V6;
14937 bcopy(src->ipp_label_v6, dst->ipp_label_v6,
14938 src->ipp_label_len_v6);
14939 dst->ipp_label_len_v6 = src->ipp_label_len_v6;
14940 }
14941 if (fields & IPPF_FRAGHDR) {
14942 dst->ipp_fraghdr = kmem_alloc(src->ipp_fraghdrlen, kmflag);
14943 if (dst->ipp_fraghdr == NULL) {
14944 ip_pkt_free(dst);
14945 return (ENOMEM);
14946 }
14947 dst->ipp_fields |= IPPF_FRAGHDR;
14948 bcopy(src->ipp_fraghdr, dst->ipp_fraghdr,
14949 src->ipp_fraghdrlen);
14950 dst->ipp_fraghdrlen = src->ipp_fraghdrlen;
14951 }
14952 return (0);
14953 }
14954
14955 /*
14956 * Returns INADDR_ANY if no source route
14957 */
14958 ipaddr_t
14959 ip_pkt_source_route_v4(const ip_pkt_t *ipp)
14960 {
14961 ipaddr_t nexthop = INADDR_ANY;
14962 ipoptp_t opts;
14963 uchar_t *opt;
14964 uint8_t optval;
14965 uint8_t optlen;
14966 uint32_t totallen;
14967
14968 if (!(ipp->ipp_fields & IPPF_IPV4_OPTIONS))
14969 return (INADDR_ANY);
14970
14971 totallen = ipp->ipp_ipv4_options_len;
14972 if (totallen & 0x3)
14973 return (INADDR_ANY);
14974
14975 for (optval = ipoptp_first2(&opts, totallen, ipp->ipp_ipv4_options);
14976 optval != IPOPT_EOL;
14977 optval = ipoptp_next(&opts)) {
14978 opt = opts.ipoptp_cur;
14979 switch (optval) {
14980 uint8_t off;
14981 case IPOPT_SSRR:
14982 case IPOPT_LSRR:
14983 if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
14984 break;
14985 }
14986 optlen = opts.ipoptp_len;
14987 off = opt[IPOPT_OFFSET];
14988 off--;
14989 if (optlen < IP_ADDR_LEN ||
14990 off > optlen - IP_ADDR_LEN) {
14991 /* End of source route */
14992 break;
14993 }
14994 bcopy((char *)opt + off, &nexthop, IP_ADDR_LEN);
14995 if (nexthop == htonl(INADDR_LOOPBACK)) {
14996 /* Ignore */
14997 nexthop = INADDR_ANY;
14998 break;
14999 }
15000 break;
15001 }
15002 }
15003 return (nexthop);
15004 }
15005
15006 /*
15007 * Reverse a source route.
15008 */
15009 void
15010 ip_pkt_source_route_reverse_v4(ip_pkt_t *ipp)
15011 {
15012 ipaddr_t tmp;
15013 ipoptp_t opts;
15014 uchar_t *opt;
15015 uint8_t optval;
15016 uint32_t totallen;
15017
15018 if (!(ipp->ipp_fields & IPPF_IPV4_OPTIONS))
15019 return;
15020
15021 totallen = ipp->ipp_ipv4_options_len;
15022 if (totallen & 0x3)
15023 return;
15024
15025 for (optval = ipoptp_first2(&opts, totallen, ipp->ipp_ipv4_options);
15026 optval != IPOPT_EOL;
15027 optval = ipoptp_next(&opts)) {
15028 uint8_t off1, off2;
15029
15030 opt = opts.ipoptp_cur;
15031 switch (optval) {
15032 case IPOPT_SSRR:
15033 case IPOPT_LSRR:
15034 if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
15035 break;
15036 }
15037 off1 = IPOPT_MINOFF_SR - 1;
15038 off2 = opt[IPOPT_OFFSET] - IP_ADDR_LEN - 1;
15039 while (off2 > off1) {
15040 bcopy(opt + off2, &tmp, IP_ADDR_LEN);
15041 bcopy(opt + off1, opt + off2, IP_ADDR_LEN);
15042 bcopy(&tmp, opt + off2, IP_ADDR_LEN);
15043 off2 -= IP_ADDR_LEN;
15044 off1 += IP_ADDR_LEN;
15045 }
15046 opt[IPOPT_OFFSET] = IPOPT_MINOFF_SR;
15047 break;
15048 }
15049 }
15050 }
15051
15052 /*
15053 * Returns NULL if no routing header
15054 */
15055 in6_addr_t *
15056 ip_pkt_source_route_v6(const ip_pkt_t *ipp)
15057 {
15058 in6_addr_t *nexthop = NULL;
15059 ip6_rthdr0_t *rthdr;
15060
15061 if (!(ipp->ipp_fields & IPPF_RTHDR))
15062 return (NULL);
15063
15064 rthdr = (ip6_rthdr0_t *)ipp->ipp_rthdr;
15065 if (rthdr->ip6r0_segleft == 0)
15066 return (NULL);
15067
15068 nexthop = (in6_addr_t *)((char *)rthdr + sizeof (*rthdr));
15069 return (nexthop);
15070 }
15071
15072 zoneid_t
15073 ip_get_zoneid_v4(ipaddr_t addr, mblk_t *mp, ip_recv_attr_t *ira,
15074 zoneid_t lookup_zoneid)
15075 {
15076 ip_stack_t *ipst = ira->ira_ill->ill_ipst;
15077 ire_t *ire;
15078 int ire_flags = MATCH_IRE_TYPE;
15079 zoneid_t zoneid = ALL_ZONES;
15080
15081 if (is_system_labeled() && !tsol_can_accept_raw(mp, ira, B_FALSE))
15082 return (ALL_ZONES);
15083
15084 if (lookup_zoneid != ALL_ZONES)
15085 ire_flags |= MATCH_IRE_ZONEONLY;
15086 ire = ire_ftable_lookup_v4(addr, NULL, NULL, IRE_LOCAL | IRE_LOOPBACK,
15087 NULL, lookup_zoneid, NULL, ire_flags, 0, ipst, NULL);
15088 if (ire != NULL) {
15089 zoneid = IP_REAL_ZONEID(ire->ire_zoneid, ipst);
15090 ire_refrele(ire);
15091 }
15092 return (zoneid);
15093 }
15094
15095 zoneid_t
15096 ip_get_zoneid_v6(in6_addr_t *addr, mblk_t *mp, const ill_t *ill,
15097 ip_recv_attr_t *ira, zoneid_t lookup_zoneid)
15098 {
15099 ip_stack_t *ipst = ira->ira_ill->ill_ipst;
15100 ire_t *ire;
15101 int ire_flags = MATCH_IRE_TYPE;
15102 zoneid_t zoneid = ALL_ZONES;
15103
15104 if (is_system_labeled() && !tsol_can_accept_raw(mp, ira, B_FALSE))
15105 return (ALL_ZONES);
15106
15107 if (IN6_IS_ADDR_LINKLOCAL(addr))
15108 ire_flags |= MATCH_IRE_ILL;
15109
15110 if (lookup_zoneid != ALL_ZONES)
15111 ire_flags |= MATCH_IRE_ZONEONLY;
15112 ire = ire_ftable_lookup_v6(addr, NULL, NULL, IRE_LOCAL | IRE_LOOPBACK,
15113 ill, lookup_zoneid, NULL, ire_flags, 0, ipst, NULL);
15114 if (ire != NULL) {
15115 zoneid = IP_REAL_ZONEID(ire->ire_zoneid, ipst);
15116 ire_refrele(ire);
15117 }
15118 return (zoneid);
15119 }
15120
15121 /*
15122 * IP obserability hook support functions.
15123 */
15124 static void
15125 ipobs_init(ip_stack_t *ipst)
15126 {
15127 netid_t id;
15128
15129 id = net_getnetidbynetstackid(ipst->ips_netstack->netstack_stackid);
15130
15131 ipst->ips_ip4_observe_pr = net_protocol_lookup(id, NHF_INET);
15132 VERIFY(ipst->ips_ip4_observe_pr != NULL);
15133
15134 ipst->ips_ip6_observe_pr = net_protocol_lookup(id, NHF_INET6);
15135 VERIFY(ipst->ips_ip6_observe_pr != NULL);
15136 }
15137
15138 static void
15139 ipobs_fini(ip_stack_t *ipst)
15140 {
15141
15142 VERIFY(net_protocol_release(ipst->ips_ip4_observe_pr) == 0);
15143 VERIFY(net_protocol_release(ipst->ips_ip6_observe_pr) == 0);
15144 }
15145
15146 /*
15147 * hook_pkt_observe_t is composed in network byte order so that the
15148 * entire mblk_t chain handed into hook_run can be used as-is.
15149 * The caveat is that use of the fields, such as the zone fields,
15150 * requires conversion into host byte order first.
15151 */
15152 void
15153 ipobs_hook(mblk_t *mp, int htype, zoneid_t zsrc, zoneid_t zdst,
15154 const ill_t *ill, ip_stack_t *ipst)
15155 {
15156 hook_pkt_observe_t *hdr;
15157 uint64_t grifindex;
15158 mblk_t *imp;
15159
15160 imp = allocb(sizeof (*hdr), BPRI_HI);
15161 if (imp == NULL)
15162 return;
15163
15164 hdr = (hook_pkt_observe_t *)imp->b_rptr;
15165 /*
15166 * b_wptr is set to make the apparent size of the data in the mblk_t
15167 * to exclude the pointers at the end of hook_pkt_observer_t.
15168 */
15169 imp->b_wptr = imp->b_rptr + sizeof (dl_ipnetinfo_t);
15170 imp->b_cont = mp;
15171
15172 ASSERT(DB_TYPE(mp) == M_DATA);
15173
15174 if (IS_UNDER_IPMP(ill))
15175 grifindex = ipmp_ill_get_ipmp_ifindex(ill);
15176 else
15177 grifindex = 0;
15178
15179 hdr->hpo_version = 1;
15180 hdr->hpo_htype = htons(htype);
15181 hdr->hpo_pktlen = htonl((ulong_t)msgdsize(mp));
15182 hdr->hpo_ifindex = htonl(ill->ill_phyint->phyint_ifindex);
15183 hdr->hpo_grifindex = htonl(grifindex);
15184 hdr->hpo_zsrc = htonl(zsrc);
15185 hdr->hpo_zdst = htonl(zdst);
15186 hdr->hpo_pkt = imp;
15187 hdr->hpo_ctx = ipst->ips_netstack;
15188
15189 if (ill->ill_isv6) {
15190 hdr->hpo_family = AF_INET6;
15191 (void) hook_run(ipst->ips_ipv6_net_data->netd_hooks,
15192 ipst->ips_ipv6observing, (hook_data_t)hdr);
15193 } else {
15194 hdr->hpo_family = AF_INET;
15195 (void) hook_run(ipst->ips_ipv4_net_data->netd_hooks,
15196 ipst->ips_ipv4observing, (hook_data_t)hdr);
15197 }
15198
15199 imp->b_cont = NULL;
15200 freemsg(imp);
15201 }
15202
15203 /*
15204 * Utility routine that checks if `v4srcp' is a valid address on underlying
15205 * interface `ill'. If `ipifp' is non-NULL, it's set to a held ipif
15206 * associated with `v4srcp' on success. NOTE: if this is not called from
15207 * inside the IPSQ (ill_g_lock is not held), `ill' may be removed from the
15208 * group during or after this lookup.
15209 */
15210 boolean_t
15211 ipif_lookup_testaddr_v4(ill_t *ill, const in_addr_t *v4srcp, ipif_t **ipifp)
15212 {
15213 ipif_t *ipif;
15214
15215 ipif = ipif_lookup_addr_exact(*v4srcp, ill, ill->ill_ipst);
15216 if (ipif != NULL) {
15217 if (ipifp != NULL)
15218 *ipifp = ipif;
15219 else
15220 ipif_refrele(ipif);
15221 return (B_TRUE);
15222 }
15223
15224 ip1dbg(("ipif_lookup_testaddr_v4: cannot find ipif for src %x\n",
15225 *v4srcp));
15226 return (B_FALSE);
15227 }
15228
15229 /*
15230 * Transport protocol call back function for CPU state change.
15231 */
15232 /* ARGSUSED */
15233 static int
15234 ip_tp_cpu_update(cpu_setup_t what, int id, void *arg)
15235 {
15236 processorid_t cpu_seqid;
15237 netstack_handle_t nh;
15238 netstack_t *ns;
15239
15240 ASSERT(MUTEX_HELD(&cpu_lock));
15241
15242 switch (what) {
15243 case CPU_CONFIG:
15244 case CPU_ON:
15245 case CPU_INIT:
15246 case CPU_CPUPART_IN:
15247 cpu_seqid = cpu[id]->cpu_seqid;
15248 netstack_next_init(&nh);
15249 while ((ns = netstack_next(&nh)) != NULL) {
15250 tcp_stack_cpu_add(ns->netstack_tcp, cpu_seqid);
15251 sctp_stack_cpu_add(ns->netstack_sctp, cpu_seqid);
15252 udp_stack_cpu_add(ns->netstack_udp, cpu_seqid);
15253 netstack_rele(ns);
15254 }
15255 netstack_next_fini(&nh);
15256 break;
15257 case CPU_UNCONFIG:
15258 case CPU_OFF:
15259 case CPU_CPUPART_OUT:
15260 /*
15261 * Nothing to do. We don't remove the per CPU stats from
15262 * the IP stack even when the CPU goes offline.
15263 */
15264 break;
15265 default:
15266 break;
15267 }
15268 return (0);
15269 }