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  * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
  23  * Use is subject to license terms.
  24  * Copyright (c) 2012 Nexenta Systems, Inc. All rights reserved.
  25  * Copyright (c) 2016 by Delphix. All rights reserved.
  26  */
  27 
  28 /*
  29  * IPsec Security Policy Database.
  30  *
  31  * This module maintains the SPD and provides routines used by ip and ip6
  32  * to apply IPsec policy to inbound and outbound datagrams.
  33  */
  34 
  35 #include <sys/types.h>
  36 #include <sys/stream.h>
  37 #include <sys/stropts.h>
  38 #include <sys/sysmacros.h>
  39 #include <sys/strsubr.h>
  40 #include <sys/strsun.h>
  41 #include <sys/strlog.h>
  42 #include <sys/strsun.h>
  43 #include <sys/cmn_err.h>
  44 #include <sys/zone.h>
  45 
  46 #include <sys/systm.h>
  47 #include <sys/param.h>
  48 #include <sys/kmem.h>
  49 #include <sys/ddi.h>
  50 
  51 #include <sys/crypto/api.h>
  52 
  53 #include <inet/common.h>
  54 #include <inet/mi.h>
  55 
  56 #include <netinet/ip6.h>
  57 #include <netinet/icmp6.h>
  58 #include <netinet/udp.h>
  59 
  60 #include <inet/ip.h>
  61 #include <inet/ip6.h>
  62 
  63 #include <net/pfkeyv2.h>
  64 #include <net/pfpolicy.h>
  65 #include <inet/sadb.h>
  66 #include <inet/ipsec_impl.h>
  67 
  68 #include <inet/ip_impl.h> /* For IP_MOD_ID */
  69 
  70 #include <inet/ipsecah.h>
  71 #include <inet/ipsecesp.h>
  72 #include <inet/ipdrop.h>
  73 #include <inet/ipclassifier.h>
  74 #include <inet/iptun.h>
  75 #include <inet/iptun/iptun_impl.h>
  76 
  77 static void ipsec_update_present_flags(ipsec_stack_t *);
  78 static ipsec_act_t *ipsec_act_wildcard_expand(ipsec_act_t *, uint_t *,
  79     netstack_t *);
  80 static mblk_t *ipsec_check_ipsecin_policy(mblk_t *, ipsec_policy_t *,
  81     ipha_t *, ip6_t *, uint64_t, ip_recv_attr_t *, netstack_t *);
  82 static void ipsec_action_free_table(ipsec_action_t *);
  83 static void ipsec_action_reclaim(void *);
  84 static void ipsec_action_reclaim_stack(ipsec_stack_t *);
  85 static void ipsid_init(netstack_t *);
  86 static void ipsid_fini(netstack_t *);
  87 
  88 /* sel_flags values for ipsec_init_inbound_sel(). */
  89 #define SEL_NONE        0x0000
  90 #define SEL_PORT_POLICY 0x0001
  91 #define SEL_IS_ICMP     0x0002
  92 #define SEL_TUNNEL_MODE 0x0004
  93 #define SEL_POST_FRAG   0x0008
  94 
  95 /* Return values for ipsec_init_inbound_sel(). */
  96 typedef enum { SELRET_NOMEM, SELRET_BADPKT, SELRET_SUCCESS, SELRET_TUNFRAG}
  97     selret_t;
  98 
  99 static selret_t ipsec_init_inbound_sel(ipsec_selector_t *, mblk_t *,
 100     ipha_t *, ip6_t *, uint8_t);
 101 
 102 static boolean_t ipsec_check_ipsecin_action(ip_recv_attr_t *, mblk_t *,
 103     struct ipsec_action_s *, ipha_t *ipha, ip6_t *ip6h, const char **,
 104     kstat_named_t **, netstack_t *);
 105 static void ipsec_unregister_prov_update(void);
 106 static void ipsec_prov_update_callback_stack(uint32_t, void *, netstack_t *);
 107 static boolean_t ipsec_compare_action(ipsec_policy_t *, ipsec_policy_t *);
 108 static uint32_t selector_hash(ipsec_selector_t *, ipsec_policy_root_t *);
 109 static boolean_t ipsec_kstat_init(ipsec_stack_t *);
 110 static void ipsec_kstat_destroy(ipsec_stack_t *);
 111 static int ipsec_free_tables(ipsec_stack_t *);
 112 static int tunnel_compare(const void *, const void *);
 113 static void ipsec_freemsg_chain(mblk_t *);
 114 static void ip_drop_packet_chain(mblk_t *, boolean_t, ill_t *,
 115     struct kstat_named *, ipdropper_t *);
 116 static boolean_t ipsec_kstat_init(ipsec_stack_t *);
 117 static void ipsec_kstat_destroy(ipsec_stack_t *);
 118 static int ipsec_free_tables(ipsec_stack_t *);
 119 static int tunnel_compare(const void *, const void *);
 120 static void ipsec_freemsg_chain(mblk_t *);
 121 
 122 /*
 123  * Selector hash table is statically sized at module load time.
 124  * we default to 251 buckets, which is the largest prime number under 255
 125  */
 126 
 127 #define IPSEC_SPDHASH_DEFAULT 251
 128 
 129 /* SPD hash-size tunable per tunnel. */
 130 #define TUN_SPDHASH_DEFAULT 5
 131 
 132 uint32_t ipsec_spd_hashsize;
 133 uint32_t tun_spd_hashsize;
 134 
 135 #define IPSEC_SEL_NOHASH ((uint32_t)(~0))
 136 
 137 /*
 138  * Handle global across all stack instances
 139  */
 140 static crypto_notify_handle_t prov_update_handle = NULL;
 141 
 142 static kmem_cache_t *ipsec_action_cache;
 143 static kmem_cache_t *ipsec_sel_cache;
 144 static kmem_cache_t *ipsec_pol_cache;
 145 
 146 /* Frag cache prototypes */
 147 static void ipsec_fragcache_clean(ipsec_fragcache_t *, ipsec_stack_t *);
 148 static ipsec_fragcache_entry_t *fragcache_delentry(int,
 149     ipsec_fragcache_entry_t *, ipsec_fragcache_t *, ipsec_stack_t *);
 150 boolean_t ipsec_fragcache_init(ipsec_fragcache_t *);
 151 void ipsec_fragcache_uninit(ipsec_fragcache_t *, ipsec_stack_t *ipss);
 152 mblk_t *ipsec_fragcache_add(ipsec_fragcache_t *, mblk_t *, mblk_t *,
 153     int, ipsec_stack_t *);
 154 
 155 int ipsec_hdr_pullup_needed = 0;
 156 int ipsec_weird_null_inbound_policy = 0;
 157 
 158 #define ALGBITS_ROUND_DOWN(x, align)    (((x)/(align))*(align))
 159 #define ALGBITS_ROUND_UP(x, align)      ALGBITS_ROUND_DOWN((x)+(align)-1, align)
 160 
 161 /*
 162  * Inbound traffic should have matching identities for both SA's.
 163  */
 164 
 165 #define SA_IDS_MATCH(sa1, sa2)                                          \
 166         (((sa1) == NULL) || ((sa2) == NULL) ||                          \
 167         (((sa1)->ipsa_src_cid == (sa2)->ipsa_src_cid) &&          \
 168             (((sa1)->ipsa_dst_cid == (sa2)->ipsa_dst_cid))))
 169 
 170 /*
 171  * IPv6 Fragments
 172  */
 173 #define IS_V6_FRAGMENT(ipp)     (ipp.ipp_fields & IPPF_FRAGHDR)
 174 
 175 /*
 176  * Policy failure messages.
 177  */
 178 static char *ipsec_policy_failure_msgs[] = {
 179 
 180         /* IPSEC_POLICY_NOT_NEEDED */
 181         "%s: Dropping the datagram because the incoming packet "
 182         "is %s, but the recipient expects clear; Source %s, "
 183         "Destination %s.\n",
 184 
 185         /* IPSEC_POLICY_MISMATCH */
 186         "%s: Policy Failure for the incoming packet (%s); Source %s, "
 187         "Destination %s.\n",
 188 
 189         /* IPSEC_POLICY_AUTH_NOT_NEEDED */
 190         "%s: Authentication present while not expected in the "
 191         "incoming %s packet; Source %s, Destination %s.\n",
 192 
 193         /* IPSEC_POLICY_ENCR_NOT_NEEDED */
 194         "%s: Encryption present while not expected in the "
 195         "incoming %s packet; Source %s, Destination %s.\n",
 196 
 197         /* IPSEC_POLICY_SE_NOT_NEEDED */
 198         "%s: Self-Encapsulation present while not expected in the "
 199         "incoming %s packet; Source %s, Destination %s.\n",
 200 };
 201 
 202 /*
 203  * General overviews:
 204  *
 205  * Locking:
 206  *
 207  *      All of the system policy structures are protected by a single
 208  *      rwlock.  These structures are threaded in a
 209  *      fairly complex fashion and are not expected to change on a
 210  *      regular basis, so this should not cause scaling/contention
 211  *      problems.  As a result, policy checks should (hopefully) be MT-hot.
 212  *
 213  * Allocation policy:
 214  *
 215  *      We use custom kmem cache types for the various
 216  *      bits & pieces of the policy data structures.  All allocations
 217  *      use KM_NOSLEEP instead of KM_SLEEP for policy allocation.  The
 218  *      policy table is of potentially unbounded size, so we don't
 219  *      want to provide a way to hog all system memory with policy
 220  *      entries..
 221  */
 222 
 223 /* Convenient functions for freeing or dropping a b_next linked mblk chain */
 224 
 225 /* Free all messages in an mblk chain */
 226 static void
 227 ipsec_freemsg_chain(mblk_t *mp)
 228 {
 229         mblk_t *mpnext;
 230         while (mp != NULL) {
 231                 ASSERT(mp->b_prev == NULL);
 232                 mpnext = mp->b_next;
 233                 mp->b_next = NULL;
 234                 freemsg(mp);
 235                 mp = mpnext;
 236         }
 237 }
 238 
 239 /*
 240  * ip_drop all messages in an mblk chain
 241  * Can handle a b_next chain of ip_recv_attr_t mblks, or just a b_next chain
 242  * of data.
 243  */
 244 static void
 245 ip_drop_packet_chain(mblk_t *mp, boolean_t inbound, ill_t *ill,
 246     struct kstat_named *counter, ipdropper_t *who_called)
 247 {
 248         mblk_t *mpnext;
 249         while (mp != NULL) {
 250                 ASSERT(mp->b_prev == NULL);
 251                 mpnext = mp->b_next;
 252                 mp->b_next = NULL;
 253                 if (ip_recv_attr_is_mblk(mp))
 254                         mp = ip_recv_attr_free_mblk(mp);
 255                 ip_drop_packet(mp, inbound, ill, counter, who_called);
 256                 mp = mpnext;
 257         }
 258 }
 259 
 260 /*
 261  * AVL tree comparison function.
 262  * the in-kernel avl assumes unique keys for all objects.
 263  * Since sometimes policy will duplicate rules, we may insert
 264  * multiple rules with the same rule id, so we need a tie-breaker.
 265  */
 266 static int
 267 ipsec_policy_cmpbyid(const void *a, const void *b)
 268 {
 269         const ipsec_policy_t *ipa, *ipb;
 270         uint64_t idxa, idxb;
 271 
 272         ipa = (const ipsec_policy_t *)a;
 273         ipb = (const ipsec_policy_t *)b;
 274         idxa = ipa->ipsp_index;
 275         idxb = ipb->ipsp_index;
 276 
 277         if (idxa < idxb)
 278                 return (-1);
 279         if (idxa > idxb)
 280                 return (1);
 281         /*
 282          * Tie-breaker #1: All installed policy rules have a non-NULL
 283          * ipsl_sel (selector set), so an entry with a NULL ipsp_sel is not
 284          * actually in-tree but rather a template node being used in
 285          * an avl_find query; see ipsec_policy_delete().  This gives us
 286          * a placeholder in the ordering just before the first entry with
 287          * a key >= the one we're looking for, so we can walk forward from
 288          * that point to get the remaining entries with the same id.
 289          */
 290         if ((ipa->ipsp_sel == NULL) && (ipb->ipsp_sel != NULL))
 291                 return (-1);
 292         if ((ipb->ipsp_sel == NULL) && (ipa->ipsp_sel != NULL))
 293                 return (1);
 294         /*
 295          * At most one of the arguments to the comparison should have a
 296          * NULL selector pointer; if not, the tree is broken.
 297          */
 298         ASSERT(ipa->ipsp_sel != NULL);
 299         ASSERT(ipb->ipsp_sel != NULL);
 300         /*
 301          * Tie-breaker #2: use the virtual address of the policy node
 302          * to arbitrarily break ties.  Since we use the new tree node in
 303          * the avl_find() in ipsec_insert_always, the new node will be
 304          * inserted into the tree in the right place in the sequence.
 305          */
 306         if (ipa < ipb)
 307                 return (-1);
 308         if (ipa > ipb)
 309                 return (1);
 310         return (0);
 311 }
 312 
 313 /*
 314  * Free what ipsec_alloc_table allocated.
 315  */
 316 void
 317 ipsec_polhead_free_table(ipsec_policy_head_t *iph)
 318 {
 319         int dir;
 320         int i;
 321 
 322         for (dir = 0; dir < IPSEC_NTYPES; dir++) {
 323                 ipsec_policy_root_t *ipr = &iph->iph_root[dir];
 324 
 325                 if (ipr->ipr_hash == NULL)
 326                         continue;
 327 
 328                 for (i = 0; i < ipr->ipr_nchains; i++) {
 329                         ASSERT(ipr->ipr_hash[i].hash_head == NULL);
 330                 }
 331                 kmem_free(ipr->ipr_hash, ipr->ipr_nchains *
 332                     sizeof (ipsec_policy_hash_t));
 333                 ipr->ipr_hash = NULL;
 334         }
 335 }
 336 
 337 void
 338 ipsec_polhead_destroy(ipsec_policy_head_t *iph)
 339 {
 340         int dir;
 341 
 342         avl_destroy(&iph->iph_rulebyid);
 343         rw_destroy(&iph->iph_lock);
 344 
 345         for (dir = 0; dir < IPSEC_NTYPES; dir++) {
 346                 ipsec_policy_root_t *ipr = &iph->iph_root[dir];
 347                 int chain;
 348 
 349                 for (chain = 0; chain < ipr->ipr_nchains; chain++)
 350                         mutex_destroy(&(ipr->ipr_hash[chain].hash_lock));
 351 
 352         }
 353         ipsec_polhead_free_table(iph);
 354 }
 355 
 356 /*
 357  * Free the IPsec stack instance.
 358  */
 359 /* ARGSUSED */
 360 static void
 361 ipsec_stack_fini(netstackid_t stackid, void *arg)
 362 {
 363         ipsec_stack_t   *ipss = (ipsec_stack_t *)arg;
 364         void *cookie;
 365         ipsec_tun_pol_t *node;
 366         netstack_t      *ns = ipss->ipsec_netstack;
 367         int             i;
 368         ipsec_algtype_t algtype;
 369 
 370         ipsec_loader_destroy(ipss);
 371 
 372         rw_enter(&ipss->ipsec_tunnel_policy_lock, RW_WRITER);
 373         /*
 374          * It's possible we can just ASSERT() the tree is empty.  After all,
 375          * we aren't called until IP is ready to unload (and presumably all
 376          * tunnels have been unplumbed).  But we'll play it safe for now, the
 377          * loop will just exit immediately if it's empty.
 378          */
 379         cookie = NULL;
 380         while ((node = (ipsec_tun_pol_t *)
 381             avl_destroy_nodes(&ipss->ipsec_tunnel_policies,
 382             &cookie)) != NULL) {
 383                 ITP_REFRELE(node, ns);
 384         }
 385         avl_destroy(&ipss->ipsec_tunnel_policies);
 386         rw_exit(&ipss->ipsec_tunnel_policy_lock);
 387         rw_destroy(&ipss->ipsec_tunnel_policy_lock);
 388 
 389         ipsec_config_flush(ns);
 390 
 391         ipsec_kstat_destroy(ipss);
 392 
 393         ip_drop_unregister(&ipss->ipsec_dropper);
 394 
 395         ip_drop_unregister(&ipss->ipsec_spd_dropper);
 396         ip_drop_destroy(ipss);
 397         /*
 398          * Globals start with ref == 1 to prevent IPPH_REFRELE() from
 399          * attempting to free them, hence they should have 1 now.
 400          */
 401         ipsec_polhead_destroy(&ipss->ipsec_system_policy);
 402         ASSERT(ipss->ipsec_system_policy.iph_refs == 1);
 403         ipsec_polhead_destroy(&ipss->ipsec_inactive_policy);
 404         ASSERT(ipss->ipsec_inactive_policy.iph_refs == 1);
 405 
 406         for (i = 0; i < IPSEC_ACTION_HASH_SIZE; i++) {
 407                 ipsec_action_free_table(ipss->ipsec_action_hash[i].hash_head);
 408                 ipss->ipsec_action_hash[i].hash_head = NULL;
 409                 mutex_destroy(&(ipss->ipsec_action_hash[i].hash_lock));
 410         }
 411 
 412         for (i = 0; i < ipss->ipsec_spd_hashsize; i++) {
 413                 ASSERT(ipss->ipsec_sel_hash[i].hash_head == NULL);
 414                 mutex_destroy(&(ipss->ipsec_sel_hash[i].hash_lock));
 415         }
 416 
 417         rw_enter(&ipss->ipsec_alg_lock, RW_WRITER);
 418         for (algtype = 0; algtype < IPSEC_NALGTYPES; algtype ++) {
 419                 int nalgs = ipss->ipsec_nalgs[algtype];
 420 
 421                 for (i = 0; i < nalgs; i++) {
 422                         if (ipss->ipsec_alglists[algtype][i] != NULL)
 423                                 ipsec_alg_unreg(algtype, i, ns);
 424                 }
 425         }
 426         rw_exit(&ipss->ipsec_alg_lock);
 427         rw_destroy(&ipss->ipsec_alg_lock);
 428 
 429         ipsid_gc(ns);
 430         ipsid_fini(ns);
 431 
 432         (void) ipsec_free_tables(ipss);
 433         kmem_free(ipss, sizeof (*ipss));
 434 }
 435 
 436 void
 437 ipsec_policy_g_destroy(void)
 438 {
 439         kmem_cache_destroy(ipsec_action_cache);
 440         kmem_cache_destroy(ipsec_sel_cache);
 441         kmem_cache_destroy(ipsec_pol_cache);
 442 
 443         ipsec_unregister_prov_update();
 444 
 445         netstack_unregister(NS_IPSEC);
 446 }
 447 
 448 
 449 /*
 450  * Free what ipsec_alloc_tables allocated.
 451  * Called when table allocation fails to free the table.
 452  */
 453 static int
 454 ipsec_free_tables(ipsec_stack_t *ipss)
 455 {
 456         int i;
 457 
 458         if (ipss->ipsec_sel_hash != NULL) {
 459                 for (i = 0; i < ipss->ipsec_spd_hashsize; i++) {
 460                         ASSERT(ipss->ipsec_sel_hash[i].hash_head == NULL);
 461                 }
 462                 kmem_free(ipss->ipsec_sel_hash, ipss->ipsec_spd_hashsize *
 463                     sizeof (*ipss->ipsec_sel_hash));
 464                 ipss->ipsec_sel_hash = NULL;
 465                 ipss->ipsec_spd_hashsize = 0;
 466         }
 467         ipsec_polhead_free_table(&ipss->ipsec_system_policy);
 468         ipsec_polhead_free_table(&ipss->ipsec_inactive_policy);
 469 
 470         return (ENOMEM);
 471 }
 472 
 473 /*
 474  * Attempt to allocate the tables in a single policy head.
 475  * Return nonzero on failure after cleaning up any work in progress.
 476  */
 477 int
 478 ipsec_alloc_table(ipsec_policy_head_t *iph, int nchains, int kmflag,
 479     boolean_t global_cleanup, netstack_t *ns)
 480 {
 481         int dir;
 482 
 483         for (dir = 0; dir < IPSEC_NTYPES; dir++) {
 484                 ipsec_policy_root_t *ipr = &iph->iph_root[dir];
 485 
 486                 ipr->ipr_nchains = nchains;
 487                 ipr->ipr_hash = kmem_zalloc(nchains *
 488                     sizeof (ipsec_policy_hash_t), kmflag);
 489                 if (ipr->ipr_hash == NULL)
 490                         return (global_cleanup ?
 491                             ipsec_free_tables(ns->netstack_ipsec) :
 492                             ENOMEM);
 493         }
 494         return (0);
 495 }
 496 
 497 /*
 498  * Attempt to allocate the various tables.  Return nonzero on failure
 499  * after cleaning up any work in progress.
 500  */
 501 static int
 502 ipsec_alloc_tables(int kmflag, netstack_t *ns)
 503 {
 504         int error;
 505         ipsec_stack_t   *ipss = ns->netstack_ipsec;
 506 
 507         error = ipsec_alloc_table(&ipss->ipsec_system_policy,
 508             ipss->ipsec_spd_hashsize, kmflag, B_TRUE, ns);
 509         if (error != 0)
 510                 return (error);
 511 
 512         error = ipsec_alloc_table(&ipss->ipsec_inactive_policy,
 513             ipss->ipsec_spd_hashsize, kmflag, B_TRUE, ns);
 514         if (error != 0)
 515                 return (error);
 516 
 517         ipss->ipsec_sel_hash = kmem_zalloc(ipss->ipsec_spd_hashsize *
 518             sizeof (*ipss->ipsec_sel_hash), kmflag);
 519 
 520         if (ipss->ipsec_sel_hash == NULL)
 521                 return (ipsec_free_tables(ipss));
 522 
 523         return (0);
 524 }
 525 
 526 /*
 527  * After table allocation, initialize a policy head.
 528  */
 529 void
 530 ipsec_polhead_init(ipsec_policy_head_t *iph, int nchains)
 531 {
 532         int dir, chain;
 533 
 534         rw_init(&iph->iph_lock, NULL, RW_DEFAULT, NULL);
 535         avl_create(&iph->iph_rulebyid, ipsec_policy_cmpbyid,
 536             sizeof (ipsec_policy_t), offsetof(ipsec_policy_t, ipsp_byid));
 537 
 538         for (dir = 0; dir < IPSEC_NTYPES; dir++) {
 539                 ipsec_policy_root_t *ipr = &iph->iph_root[dir];
 540                 ipr->ipr_nchains = nchains;
 541 
 542                 for (chain = 0; chain < nchains; chain++) {
 543                         mutex_init(&(ipr->ipr_hash[chain].hash_lock),
 544                             NULL, MUTEX_DEFAULT, NULL);
 545                 }
 546         }
 547 }
 548 
 549 static boolean_t
 550 ipsec_kstat_init(ipsec_stack_t *ipss)
 551 {
 552         ipss->ipsec_ksp = kstat_create_netstack("ip", 0, "ipsec_stat", "net",
 553             KSTAT_TYPE_NAMED, sizeof (ipsec_kstats_t) / sizeof (kstat_named_t),
 554             KSTAT_FLAG_PERSISTENT, ipss->ipsec_netstack->netstack_stackid);
 555 
 556         if (ipss->ipsec_ksp == NULL || ipss->ipsec_ksp->ks_data == NULL)
 557                 return (B_FALSE);
 558 
 559         ipss->ipsec_kstats = ipss->ipsec_ksp->ks_data;
 560 
 561 #define KI(x) kstat_named_init(&ipss->ipsec_kstats->x, #x, KSTAT_DATA_UINT64)
 562         KI(esp_stat_in_requests);
 563         KI(esp_stat_in_discards);
 564         KI(esp_stat_lookup_failure);
 565         KI(ah_stat_in_requests);
 566         KI(ah_stat_in_discards);
 567         KI(ah_stat_lookup_failure);
 568         KI(sadb_acquire_maxpackets);
 569         KI(sadb_acquire_qhiwater);
 570 #undef KI
 571 
 572         kstat_install(ipss->ipsec_ksp);
 573         return (B_TRUE);
 574 }
 575 
 576 static void
 577 ipsec_kstat_destroy(ipsec_stack_t *ipss)
 578 {
 579         kstat_delete_netstack(ipss->ipsec_ksp,
 580             ipss->ipsec_netstack->netstack_stackid);
 581         ipss->ipsec_kstats = NULL;
 582 
 583 }
 584 
 585 /*
 586  * Initialize the IPsec stack instance.
 587  */
 588 /* ARGSUSED */
 589 static void *
 590 ipsec_stack_init(netstackid_t stackid, netstack_t *ns)
 591 {
 592         ipsec_stack_t   *ipss;
 593         int i;
 594 
 595         ipss = (ipsec_stack_t *)kmem_zalloc(sizeof (*ipss), KM_SLEEP);
 596         ipss->ipsec_netstack = ns;
 597 
 598         /*
 599          * FIXME: netstack_ipsec is used by some of the routines we call
 600          * below, but it isn't set until this routine returns.
 601          * Either we introduce optional xxx_stack_alloc() functions
 602          * that will be called by the netstack framework before xxx_stack_init,
 603          * or we switch spd.c and sadb.c to operate on ipsec_stack_t
 604          * (latter has some include file order issues for sadb.h, but makes
 605          * sense if we merge some of the ipsec related stack_t's together.
 606          */
 607         ns->netstack_ipsec = ipss;
 608 
 609         /*
 610          * Make two attempts to allocate policy hash tables; try it at
 611          * the "preferred" size (may be set in /etc/system) first,
 612          * then fall back to the default size.
 613          */
 614         ipss->ipsec_spd_hashsize = (ipsec_spd_hashsize == 0) ?
 615             IPSEC_SPDHASH_DEFAULT : ipsec_spd_hashsize;
 616 
 617         if (ipsec_alloc_tables(KM_NOSLEEP, ns) != 0) {
 618                 cmn_err(CE_WARN,
 619                     "Unable to allocate %d entry IPsec policy hash table",
 620                     ipss->ipsec_spd_hashsize);
 621                 ipss->ipsec_spd_hashsize = IPSEC_SPDHASH_DEFAULT;
 622                 cmn_err(CE_WARN, "Falling back to %d entries",
 623                     ipss->ipsec_spd_hashsize);
 624                 (void) ipsec_alloc_tables(KM_SLEEP, ns);
 625         }
 626 
 627         /* Just set a default for tunnels. */
 628         ipss->ipsec_tun_spd_hashsize = (tun_spd_hashsize == 0) ?
 629             TUN_SPDHASH_DEFAULT : tun_spd_hashsize;
 630 
 631         ipsid_init(ns);
 632         /*
 633          * Globals need ref == 1 to prevent IPPH_REFRELE() from attempting
 634          * to free them.
 635          */
 636         ipss->ipsec_system_policy.iph_refs = 1;
 637         ipss->ipsec_inactive_policy.iph_refs = 1;
 638         ipsec_polhead_init(&ipss->ipsec_system_policy,
 639             ipss->ipsec_spd_hashsize);
 640         ipsec_polhead_init(&ipss->ipsec_inactive_policy,
 641             ipss->ipsec_spd_hashsize);
 642         rw_init(&ipss->ipsec_tunnel_policy_lock, NULL, RW_DEFAULT, NULL);
 643         avl_create(&ipss->ipsec_tunnel_policies, tunnel_compare,
 644             sizeof (ipsec_tun_pol_t), 0);
 645 
 646         ipss->ipsec_next_policy_index = 1;
 647 
 648         rw_init(&ipss->ipsec_system_policy.iph_lock, NULL, RW_DEFAULT, NULL);
 649         rw_init(&ipss->ipsec_inactive_policy.iph_lock, NULL, RW_DEFAULT, NULL);
 650 
 651         for (i = 0; i < IPSEC_ACTION_HASH_SIZE; i++)
 652                 mutex_init(&(ipss->ipsec_action_hash[i].hash_lock),
 653                     NULL, MUTEX_DEFAULT, NULL);
 654 
 655         for (i = 0; i < ipss->ipsec_spd_hashsize; i++)
 656                 mutex_init(&(ipss->ipsec_sel_hash[i].hash_lock),
 657                     NULL, MUTEX_DEFAULT, NULL);
 658 
 659         rw_init(&ipss->ipsec_alg_lock, NULL, RW_DEFAULT, NULL);
 660         for (i = 0; i < IPSEC_NALGTYPES; i++) {
 661                 ipss->ipsec_nalgs[i] = 0;
 662         }
 663 
 664         ip_drop_init(ipss);
 665         ip_drop_register(&ipss->ipsec_spd_dropper, "IPsec SPD");
 666 
 667         /* IP's IPsec code calls the packet dropper */
 668         ip_drop_register(&ipss->ipsec_dropper, "IP IPsec processing");
 669 
 670         (void) ipsec_kstat_init(ipss);
 671 
 672         ipsec_loader_init(ipss);
 673         ipsec_loader_start(ipss);
 674 
 675         return (ipss);
 676 }
 677 
 678 /* Global across all stack instances */
 679 void
 680 ipsec_policy_g_init(void)
 681 {
 682         ipsec_action_cache = kmem_cache_create("ipsec_actions",
 683             sizeof (ipsec_action_t), _POINTER_ALIGNMENT, NULL, NULL,
 684             ipsec_action_reclaim, NULL, NULL, 0);
 685         ipsec_sel_cache = kmem_cache_create("ipsec_selectors",
 686             sizeof (ipsec_sel_t), _POINTER_ALIGNMENT, NULL, NULL,
 687             NULL, NULL, NULL, 0);
 688         ipsec_pol_cache = kmem_cache_create("ipsec_policy",
 689             sizeof (ipsec_policy_t), _POINTER_ALIGNMENT, NULL, NULL,
 690             NULL, NULL, NULL, 0);
 691 
 692         /*
 693          * We want to be informed each time a stack is created or
 694          * destroyed in the kernel, so we can maintain the
 695          * set of ipsec_stack_t's.
 696          */
 697         netstack_register(NS_IPSEC, ipsec_stack_init, NULL, ipsec_stack_fini);
 698 }
 699 
 700 /*
 701  * Sort algorithm lists.
 702  *
 703  * I may need to split this based on
 704  * authentication/encryption, and I may wish to have an administrator
 705  * configure this list.  Hold on to some NDD variables...
 706  *
 707  * XXX For now, sort on minimum key size (GAG!).  While minimum key size is
 708  * not the ideal metric, it's the only quantifiable measure available.
 709  * We need a better metric for sorting algorithms by preference.
 710  */
 711 static void
 712 alg_insert_sortlist(enum ipsec_algtype at, uint8_t algid, netstack_t *ns)
 713 {
 714         ipsec_stack_t   *ipss = ns->netstack_ipsec;
 715         ipsec_alginfo_t *ai = ipss->ipsec_alglists[at][algid];
 716         uint8_t holder, swap;
 717         uint_t i;
 718         uint_t count = ipss->ipsec_nalgs[at];
 719         ASSERT(ai != NULL);
 720         ASSERT(algid == ai->alg_id);
 721 
 722         ASSERT(RW_WRITE_HELD(&ipss->ipsec_alg_lock));
 723 
 724         holder = algid;
 725 
 726         for (i = 0; i < count - 1; i++) {
 727                 ipsec_alginfo_t *alt;
 728 
 729                 alt = ipss->ipsec_alglists[at][ipss->ipsec_sortlist[at][i]];
 730                 /*
 731                  * If you want to give precedence to newly added algs,
 732                  * add the = in the > comparison.
 733                  */
 734                 if ((holder != algid) || (ai->alg_minbits > alt->alg_minbits)) {
 735                         /* Swap sortlist[i] and holder. */
 736                         swap = ipss->ipsec_sortlist[at][i];
 737                         ipss->ipsec_sortlist[at][i] = holder;
 738                         holder = swap;
 739                         ai = alt;
 740                 } /* Else just continue. */
 741         }
 742 
 743         /* Store holder in last slot. */
 744         ipss->ipsec_sortlist[at][i] = holder;
 745 }
 746 
 747 /*
 748  * Remove an algorithm from a sorted algorithm list.
 749  * This should be considerably easier, even with complex sorting.
 750  */
 751 static void
 752 alg_remove_sortlist(enum ipsec_algtype at, uint8_t algid, netstack_t *ns)
 753 {
 754         boolean_t copyback = B_FALSE;
 755         int i;
 756         ipsec_stack_t   *ipss = ns->netstack_ipsec;
 757         int newcount = ipss->ipsec_nalgs[at];
 758 
 759         ASSERT(RW_WRITE_HELD(&ipss->ipsec_alg_lock));
 760 
 761         for (i = 0; i <= newcount; i++) {
 762                 if (copyback) {
 763                         ipss->ipsec_sortlist[at][i-1] =
 764                             ipss->ipsec_sortlist[at][i];
 765                 } else if (ipss->ipsec_sortlist[at][i] == algid) {
 766                         copyback = B_TRUE;
 767                 }
 768         }
 769 }
 770 
 771 /*
 772  * Add the specified algorithm to the algorithm tables.
 773  * Must be called while holding the algorithm table writer lock.
 774  */
 775 void
 776 ipsec_alg_reg(ipsec_algtype_t algtype, ipsec_alginfo_t *alg, netstack_t *ns)
 777 {
 778         ipsec_stack_t   *ipss = ns->netstack_ipsec;
 779 
 780         ASSERT(RW_WRITE_HELD(&ipss->ipsec_alg_lock));
 781 
 782         ASSERT(ipss->ipsec_alglists[algtype][alg->alg_id] == NULL);
 783         ipsec_alg_fix_min_max(alg, algtype, ns);
 784         ipss->ipsec_alglists[algtype][alg->alg_id] = alg;
 785 
 786         ipss->ipsec_nalgs[algtype]++;
 787         alg_insert_sortlist(algtype, alg->alg_id, ns);
 788 }
 789 
 790 /*
 791  * Remove the specified algorithm from the algorithm tables.
 792  * Must be called while holding the algorithm table writer lock.
 793  */
 794 void
 795 ipsec_alg_unreg(ipsec_algtype_t algtype, uint8_t algid, netstack_t *ns)
 796 {
 797         ipsec_stack_t   *ipss = ns->netstack_ipsec;
 798 
 799         ASSERT(RW_WRITE_HELD(&ipss->ipsec_alg_lock));
 800 
 801         ASSERT(ipss->ipsec_alglists[algtype][algid] != NULL);
 802         ipsec_alg_free(ipss->ipsec_alglists[algtype][algid]);
 803         ipss->ipsec_alglists[algtype][algid] = NULL;
 804 
 805         ipss->ipsec_nalgs[algtype]--;
 806         alg_remove_sortlist(algtype, algid, ns);
 807 }
 808 
 809 /*
 810  * Hooks for spdsock to get a grip on system policy.
 811  */
 812 
 813 ipsec_policy_head_t *
 814 ipsec_system_policy(netstack_t *ns)
 815 {
 816         ipsec_stack_t   *ipss = ns->netstack_ipsec;
 817         ipsec_policy_head_t *h = &ipss->ipsec_system_policy;
 818 
 819         IPPH_REFHOLD(h);
 820         return (h);
 821 }
 822 
 823 ipsec_policy_head_t *
 824 ipsec_inactive_policy(netstack_t *ns)
 825 {
 826         ipsec_stack_t   *ipss = ns->netstack_ipsec;
 827         ipsec_policy_head_t *h = &ipss->ipsec_inactive_policy;
 828 
 829         IPPH_REFHOLD(h);
 830         return (h);
 831 }
 832 
 833 /*
 834  * Lock inactive policy, then active policy, then exchange policy root
 835  * pointers.
 836  */
 837 void
 838 ipsec_swap_policy(ipsec_policy_head_t *active, ipsec_policy_head_t *inactive,
 839     netstack_t *ns)
 840 {
 841         int af, dir;
 842         avl_tree_t r1, r2;
 843 
 844         rw_enter(&inactive->iph_lock, RW_WRITER);
 845         rw_enter(&active->iph_lock, RW_WRITER);
 846 
 847         r1 = active->iph_rulebyid;
 848         r2 = inactive->iph_rulebyid;
 849         active->iph_rulebyid = r2;
 850         inactive->iph_rulebyid = r1;
 851 
 852         for (dir = 0; dir < IPSEC_NTYPES; dir++) {
 853                 ipsec_policy_hash_t *h1, *h2;
 854 
 855                 h1 = active->iph_root[dir].ipr_hash;
 856                 h2 = inactive->iph_root[dir].ipr_hash;
 857                 active->iph_root[dir].ipr_hash = h2;
 858                 inactive->iph_root[dir].ipr_hash = h1;
 859 
 860                 for (af = 0; af < IPSEC_NAF; af++) {
 861                         ipsec_policy_t *t1, *t2;
 862 
 863                         t1 = active->iph_root[dir].ipr_nonhash[af];
 864                         t2 = inactive->iph_root[dir].ipr_nonhash[af];
 865                         active->iph_root[dir].ipr_nonhash[af] = t2;
 866                         inactive->iph_root[dir].ipr_nonhash[af] = t1;
 867                         if (t1 != NULL) {
 868                                 t1->ipsp_hash.hash_pp =
 869                                     &(inactive->iph_root[dir].ipr_nonhash[af]);
 870                         }
 871                         if (t2 != NULL) {
 872                                 t2->ipsp_hash.hash_pp =
 873                                     &(active->iph_root[dir].ipr_nonhash[af]);
 874                         }
 875 
 876                 }
 877         }
 878         active->iph_gen++;
 879         inactive->iph_gen++;
 880         ipsec_update_present_flags(ns->netstack_ipsec);
 881         rw_exit(&active->iph_lock);
 882         rw_exit(&inactive->iph_lock);
 883 }
 884 
 885 /*
 886  * Swap global policy primary/secondary.
 887  */
 888 void
 889 ipsec_swap_global_policy(netstack_t *ns)
 890 {
 891         ipsec_stack_t   *ipss = ns->netstack_ipsec;
 892 
 893         ipsec_swap_policy(&ipss->ipsec_system_policy,
 894             &ipss->ipsec_inactive_policy, ns);
 895 }
 896 
 897 /*
 898  * Clone one policy rule..
 899  */
 900 static ipsec_policy_t *
 901 ipsec_copy_policy(const ipsec_policy_t *src)
 902 {
 903         ipsec_policy_t *dst = kmem_cache_alloc(ipsec_pol_cache, KM_NOSLEEP);
 904 
 905         if (dst == NULL)
 906                 return (NULL);
 907 
 908         /*
 909          * Adjust refcounts of cloned state.
 910          */
 911         IPACT_REFHOLD(src->ipsp_act);
 912         src->ipsp_sel->ipsl_refs++;
 913 
 914         HASH_NULL(dst, ipsp_hash);
 915         dst->ipsp_netstack = src->ipsp_netstack;
 916         dst->ipsp_refs = 1;
 917         dst->ipsp_sel = src->ipsp_sel;
 918         dst->ipsp_act = src->ipsp_act;
 919         dst->ipsp_prio = src->ipsp_prio;
 920         dst->ipsp_index = src->ipsp_index;
 921 
 922         return (dst);
 923 }
 924 
 925 void
 926 ipsec_insert_always(avl_tree_t *tree, void *new_node)
 927 {
 928         void *node;
 929         avl_index_t where;
 930 
 931         node = avl_find(tree, new_node, &where);
 932         ASSERT(node == NULL);
 933         avl_insert(tree, new_node, where);
 934 }
 935 
 936 
 937 static int
 938 ipsec_copy_chain(ipsec_policy_head_t *dph, ipsec_policy_t *src,
 939     ipsec_policy_t **dstp)
 940 {
 941         for (; src != NULL; src = src->ipsp_hash.hash_next) {
 942                 ipsec_policy_t *dst = ipsec_copy_policy(src);
 943                 if (dst == NULL)
 944                         return (ENOMEM);
 945 
 946                 HASHLIST_INSERT(dst, ipsp_hash, *dstp);
 947                 ipsec_insert_always(&dph->iph_rulebyid, dst);
 948         }
 949         return (0);
 950 }
 951 
 952 
 953 
 954 /*
 955  * Make one policy head look exactly like another.
 956  *
 957  * As with ipsec_swap_policy, we lock the destination policy head first, then
 958  * the source policy head. Note that we only need to read-lock the source
 959  * policy head as we are not changing it.
 960  */
 961 int
 962 ipsec_copy_polhead(ipsec_policy_head_t *sph, ipsec_policy_head_t *dph,
 963     netstack_t *ns)
 964 {
 965         int af, dir, chain, nchains;
 966 
 967         rw_enter(&dph->iph_lock, RW_WRITER);
 968 
 969         ipsec_polhead_flush(dph, ns);
 970 
 971         rw_enter(&sph->iph_lock, RW_READER);
 972 
 973         for (dir = 0; dir < IPSEC_NTYPES; dir++) {
 974                 ipsec_policy_root_t *dpr = &dph->iph_root[dir];
 975                 ipsec_policy_root_t *spr = &sph->iph_root[dir];
 976                 nchains = dpr->ipr_nchains;
 977 
 978                 ASSERT(dpr->ipr_nchains == spr->ipr_nchains);
 979 
 980                 for (af = 0; af < IPSEC_NAF; af++) {
 981                         if (ipsec_copy_chain(dph, spr->ipr_nonhash[af],
 982                             &dpr->ipr_nonhash[af]))
 983                                 goto abort_copy;
 984                 }
 985 
 986                 for (chain = 0; chain < nchains; chain++) {
 987                         if (ipsec_copy_chain(dph,
 988                             spr->ipr_hash[chain].hash_head,
 989                             &dpr->ipr_hash[chain].hash_head))
 990                                 goto abort_copy;
 991                 }
 992         }
 993 
 994         dph->iph_gen++;
 995 
 996         rw_exit(&sph->iph_lock);
 997         rw_exit(&dph->iph_lock);
 998         return (0);
 999 
1000 abort_copy:
1001         ipsec_polhead_flush(dph, ns);
1002         rw_exit(&sph->iph_lock);
1003         rw_exit(&dph->iph_lock);
1004         return (ENOMEM);
1005 }
1006 
1007 /*
1008  * Clone currently active policy to the inactive policy list.
1009  */
1010 int
1011 ipsec_clone_system_policy(netstack_t *ns)
1012 {
1013         ipsec_stack_t   *ipss = ns->netstack_ipsec;
1014 
1015         return (ipsec_copy_polhead(&ipss->ipsec_system_policy,
1016             &ipss->ipsec_inactive_policy, ns));
1017 }
1018 
1019 /*
1020  * Extract the string from ipsec_policy_failure_msgs[type] and
1021  * log it.
1022  *
1023  */
1024 void
1025 ipsec_log_policy_failure(int type, char *func_name, ipha_t *ipha, ip6_t *ip6h,
1026     boolean_t secure, netstack_t *ns)
1027 {
1028         char    sbuf[INET6_ADDRSTRLEN];
1029         char    dbuf[INET6_ADDRSTRLEN];
1030         char    *s;
1031         char    *d;
1032         ipsec_stack_t   *ipss = ns->netstack_ipsec;
1033 
1034         ASSERT((ipha == NULL && ip6h != NULL) ||
1035             (ip6h == NULL && ipha != NULL));
1036 
1037         if (ipha != NULL) {
1038                 s = inet_ntop(AF_INET, &ipha->ipha_src, sbuf, sizeof (sbuf));
1039                 d = inet_ntop(AF_INET, &ipha->ipha_dst, dbuf, sizeof (dbuf));
1040         } else {
1041                 s = inet_ntop(AF_INET6, &ip6h->ip6_src, sbuf, sizeof (sbuf));
1042                 d = inet_ntop(AF_INET6, &ip6h->ip6_dst, dbuf, sizeof (dbuf));
1043 
1044         }
1045 
1046         /* Always bump the policy failure counter. */
1047         ipss->ipsec_policy_failure_count[type]++;
1048 
1049         ipsec_rl_strlog(ns, IP_MOD_ID, 0, 0, SL_ERROR|SL_WARN|SL_CONSOLE,
1050             ipsec_policy_failure_msgs[type], func_name,
1051             (secure ? "secure" : "not secure"), s, d);
1052 }
1053 
1054 /*
1055  * Rate-limiting front-end to strlog() for AH and ESP.  Uses the ndd variables
1056  * in /dev/ip and the same rate-limiting clock so that there's a single
1057  * knob to turn to throttle the rate of messages.
1058  */
1059 void
1060 ipsec_rl_strlog(netstack_t *ns, short mid, short sid, char level, ushort_t sl,
1061     char *fmt, ...)
1062 {
1063         va_list adx;
1064         hrtime_t current = gethrtime();
1065         ip_stack_t      *ipst = ns->netstack_ip;
1066         ipsec_stack_t   *ipss = ns->netstack_ipsec;
1067 
1068         sl |= SL_CONSOLE;
1069         /*
1070          * Throttle logging to stop syslog from being swamped. If variable
1071          * 'ipsec_policy_log_interval' is zero, don't log any messages at
1072          * all, otherwise log only one message every 'ipsec_policy_log_interval'
1073          * msec. Convert interval (in msec) to hrtime (in nsec).
1074          */
1075 
1076         if (ipst->ips_ipsec_policy_log_interval) {
1077                 if (ipss->ipsec_policy_failure_last +
1078                     MSEC2NSEC(ipst->ips_ipsec_policy_log_interval) <= current) {
1079                         va_start(adx, fmt);
1080                         (void) vstrlog(mid, sid, level, sl, fmt, adx);
1081                         va_end(adx);
1082                         ipss->ipsec_policy_failure_last = current;
1083                 }
1084         }
1085 }
1086 
1087 void
1088 ipsec_config_flush(netstack_t *ns)
1089 {
1090         ipsec_stack_t   *ipss = ns->netstack_ipsec;
1091 
1092         rw_enter(&ipss->ipsec_system_policy.iph_lock, RW_WRITER);
1093         ipsec_polhead_flush(&ipss->ipsec_system_policy, ns);
1094         ipss->ipsec_next_policy_index = 1;
1095         rw_exit(&ipss->ipsec_system_policy.iph_lock);
1096         ipsec_action_reclaim_stack(ipss);
1097 }
1098 
1099 /*
1100  * Clip a policy's min/max keybits vs. the capabilities of the
1101  * algorithm.
1102  */
1103 static void
1104 act_alg_adjust(uint_t algtype, uint_t algid,
1105     uint16_t *minbits, uint16_t *maxbits, netstack_t *ns)
1106 {
1107         ipsec_stack_t   *ipss = ns->netstack_ipsec;
1108         ipsec_alginfo_t *algp = ipss->ipsec_alglists[algtype][algid];
1109 
1110         if (algp != NULL) {
1111                 /*
1112                  * If passed-in minbits is zero, we assume the caller trusts
1113                  * us with setting the minimum key size.  We pick the
1114                  * algorithms DEFAULT key size for the minimum in this case.
1115                  */
1116                 if (*minbits == 0) {
1117                         *minbits = algp->alg_default_bits;
1118                         ASSERT(*minbits >= algp->alg_minbits);
1119                 } else {
1120                         *minbits = MAX(MIN(*minbits, algp->alg_maxbits),
1121                             algp->alg_minbits);
1122                 }
1123                 if (*maxbits == 0)
1124                         *maxbits = algp->alg_maxbits;
1125                 else
1126                         *maxbits = MIN(MAX(*maxbits, algp->alg_minbits),
1127                             algp->alg_maxbits);
1128                 ASSERT(*minbits <= *maxbits);
1129         } else {
1130                 *minbits = 0;
1131                 *maxbits = 0;
1132         }
1133 }
1134 
1135 /*
1136  * Check an action's requested algorithms against the algorithms currently
1137  * loaded in the system.
1138  */
1139 boolean_t
1140 ipsec_check_action(ipsec_act_t *act, int *diag, netstack_t *ns)
1141 {
1142         ipsec_prot_t *ipp;
1143         ipsec_stack_t   *ipss = ns->netstack_ipsec;
1144 
1145         ipp = &act->ipa_apply;
1146 
1147         if (ipp->ipp_use_ah &&
1148             ipss->ipsec_alglists[IPSEC_ALG_AUTH][ipp->ipp_auth_alg] == NULL) {
1149                 *diag = SPD_DIAGNOSTIC_UNSUPP_AH_ALG;
1150                 return (B_FALSE);
1151         }
1152         if (ipp->ipp_use_espa &&
1153             ipss->ipsec_alglists[IPSEC_ALG_AUTH][ipp->ipp_esp_auth_alg] ==
1154             NULL) {
1155                 *diag = SPD_DIAGNOSTIC_UNSUPP_ESP_AUTH_ALG;
1156                 return (B_FALSE);
1157         }
1158         if (ipp->ipp_use_esp &&
1159             ipss->ipsec_alglists[IPSEC_ALG_ENCR][ipp->ipp_encr_alg] == NULL) {
1160                 *diag = SPD_DIAGNOSTIC_UNSUPP_ESP_ENCR_ALG;
1161                 return (B_FALSE);
1162         }
1163 
1164         act_alg_adjust(IPSEC_ALG_AUTH, ipp->ipp_auth_alg,
1165             &ipp->ipp_ah_minbits, &ipp->ipp_ah_maxbits, ns);
1166         act_alg_adjust(IPSEC_ALG_AUTH, ipp->ipp_esp_auth_alg,
1167             &ipp->ipp_espa_minbits, &ipp->ipp_espa_maxbits, ns);
1168         act_alg_adjust(IPSEC_ALG_ENCR, ipp->ipp_encr_alg,
1169             &ipp->ipp_espe_minbits, &ipp->ipp_espe_maxbits, ns);
1170 
1171         if (ipp->ipp_ah_minbits > ipp->ipp_ah_maxbits) {
1172                 *diag = SPD_DIAGNOSTIC_UNSUPP_AH_KEYSIZE;
1173                 return (B_FALSE);
1174         }
1175         if (ipp->ipp_espa_minbits > ipp->ipp_espa_maxbits) {
1176                 *diag = SPD_DIAGNOSTIC_UNSUPP_ESP_AUTH_KEYSIZE;
1177                 return (B_FALSE);
1178         }
1179         if (ipp->ipp_espe_minbits > ipp->ipp_espe_maxbits) {
1180                 *diag = SPD_DIAGNOSTIC_UNSUPP_ESP_ENCR_KEYSIZE;
1181                 return (B_FALSE);
1182         }
1183         /* TODO: sanity check lifetimes */
1184         return (B_TRUE);
1185 }
1186 
1187 /*
1188  * Set up a single action during wildcard expansion..
1189  */
1190 static void
1191 ipsec_setup_act(ipsec_act_t *outact, ipsec_act_t *act,
1192     uint_t auth_alg, uint_t encr_alg, uint_t eauth_alg, netstack_t *ns)
1193 {
1194         ipsec_prot_t *ipp;
1195 
1196         *outact = *act;
1197         ipp = &outact->ipa_apply;
1198         ipp->ipp_auth_alg = (uint8_t)auth_alg;
1199         ipp->ipp_encr_alg = (uint8_t)encr_alg;
1200         ipp->ipp_esp_auth_alg = (uint8_t)eauth_alg;
1201 
1202         act_alg_adjust(IPSEC_ALG_AUTH, auth_alg,
1203             &ipp->ipp_ah_minbits, &ipp->ipp_ah_maxbits, ns);
1204         act_alg_adjust(IPSEC_ALG_AUTH, eauth_alg,
1205             &ipp->ipp_espa_minbits, &ipp->ipp_espa_maxbits, ns);
1206         act_alg_adjust(IPSEC_ALG_ENCR, encr_alg,
1207             &ipp->ipp_espe_minbits, &ipp->ipp_espe_maxbits, ns);
1208 }
1209 
1210 /*
1211  * combinatoric expansion time: expand a wildcarded action into an
1212  * array of wildcarded actions; we return the exploded action list,
1213  * and return a count in *nact (output only).
1214  */
1215 static ipsec_act_t *
1216 ipsec_act_wildcard_expand(ipsec_act_t *act, uint_t *nact, netstack_t *ns)
1217 {
1218         boolean_t use_ah, use_esp, use_espa;
1219         boolean_t wild_auth, wild_encr, wild_eauth;
1220         uint_t  auth_alg, auth_idx, auth_min, auth_max;
1221         uint_t  eauth_alg, eauth_idx, eauth_min, eauth_max;
1222         uint_t  encr_alg, encr_idx, encr_min, encr_max;
1223         uint_t  action_count, ai;
1224         ipsec_act_t *outact;
1225         ipsec_stack_t   *ipss = ns->netstack_ipsec;
1226 
1227         if (act->ipa_type != IPSEC_ACT_APPLY) {
1228                 outact = kmem_alloc(sizeof (*act), KM_NOSLEEP);
1229                 *nact = 1;
1230                 if (outact != NULL)
1231                         bcopy(act, outact, sizeof (*act));
1232                 return (outact);
1233         }
1234         /*
1235          * compute the combinatoric explosion..
1236          *
1237          * we assume a request for encr if esp_req is PREF_REQUIRED
1238          * we assume a request for ah auth if ah_req is PREF_REQUIRED.
1239          * we assume a request for esp auth if !ah and esp_req is PREF_REQUIRED
1240          */
1241 
1242         use_ah = act->ipa_apply.ipp_use_ah;
1243         use_esp = act->ipa_apply.ipp_use_esp;
1244         use_espa = act->ipa_apply.ipp_use_espa;
1245         auth_alg = act->ipa_apply.ipp_auth_alg;
1246         eauth_alg = act->ipa_apply.ipp_esp_auth_alg;
1247         encr_alg = act->ipa_apply.ipp_encr_alg;
1248 
1249         wild_auth = use_ah && (auth_alg == 0);
1250         wild_eauth = use_espa && (eauth_alg == 0);
1251         wild_encr = use_esp && (encr_alg == 0);
1252 
1253         action_count = 1;
1254         auth_min = auth_max = auth_alg;
1255         eauth_min = eauth_max = eauth_alg;
1256         encr_min = encr_max = encr_alg;
1257 
1258         /*
1259          * set up for explosion.. for each dimension, expand output
1260          * size by the explosion factor.
1261          *
1262          * Don't include the "any" algorithms, if defined, as no
1263          * kernel policies should be set for these algorithms.
1264          */
1265 
1266 #define SET_EXP_MINMAX(type, wild, alg, min, max, ipss)         \
1267         if (wild) {                                             \
1268                 int nalgs = ipss->ipsec_nalgs[type];         \
1269                 if (ipss->ipsec_alglists[type][alg] != NULL) \
1270                         nalgs--;                                \
1271                 action_count *= nalgs;                          \
1272                 min = 0;                                        \
1273                 max = ipss->ipsec_nalgs[type] - 1;           \
1274         }
1275 
1276         SET_EXP_MINMAX(IPSEC_ALG_AUTH, wild_auth, SADB_AALG_NONE,
1277             auth_min, auth_max, ipss);
1278         SET_EXP_MINMAX(IPSEC_ALG_AUTH, wild_eauth, SADB_AALG_NONE,
1279             eauth_min, eauth_max, ipss);
1280         SET_EXP_MINMAX(IPSEC_ALG_ENCR, wild_encr, SADB_EALG_NONE,
1281             encr_min, encr_max, ipss);
1282 
1283 #undef  SET_EXP_MINMAX
1284 
1285         /*
1286          * ok, allocate the whole mess..
1287          */
1288 
1289         outact = kmem_alloc(sizeof (*outact) * action_count, KM_NOSLEEP);
1290         if (outact == NULL)
1291                 return (NULL);
1292 
1293         /*
1294          * Now compute all combinations.  Note that non-wildcarded
1295          * dimensions just get a single value from auth_min, while
1296          * wildcarded dimensions indirect through the sortlist.
1297          *
1298          * We do encryption outermost since, at this time, there's
1299          * greater difference in security and performance between
1300          * encryption algorithms vs. authentication algorithms.
1301          */
1302 
1303         ai = 0;
1304 
1305 #define WHICH_ALG(type, wild, idx, ipss) \
1306         ((wild)?(ipss->ipsec_sortlist[type][idx]):(idx))
1307 
1308         for (encr_idx = encr_min; encr_idx <= encr_max; encr_idx++) {
1309                 encr_alg = WHICH_ALG(IPSEC_ALG_ENCR, wild_encr, encr_idx, ipss);
1310                 if (wild_encr && encr_alg == SADB_EALG_NONE)
1311                         continue;
1312                 for (auth_idx = auth_min; auth_idx <= auth_max; auth_idx++) {
1313                         auth_alg = WHICH_ALG(IPSEC_ALG_AUTH, wild_auth,
1314                             auth_idx, ipss);
1315                         if (wild_auth && auth_alg == SADB_AALG_NONE)
1316                                 continue;
1317                         for (eauth_idx = eauth_min; eauth_idx <= eauth_max;
1318                             eauth_idx++) {
1319                                 eauth_alg = WHICH_ALG(IPSEC_ALG_AUTH,
1320                                     wild_eauth, eauth_idx, ipss);
1321                                 if (wild_eauth && eauth_alg == SADB_AALG_NONE)
1322                                         continue;
1323 
1324                                 ipsec_setup_act(&outact[ai], act,
1325                                     auth_alg, encr_alg, eauth_alg, ns);
1326                                 ai++;
1327                         }
1328                 }
1329         }
1330 
1331 #undef WHICH_ALG
1332 
1333         ASSERT(ai == action_count);
1334         *nact = action_count;
1335         return (outact);
1336 }
1337 
1338 /*
1339  * Extract the parts of an ipsec_prot_t from an old-style ipsec_req_t.
1340  */
1341 static void
1342 ipsec_prot_from_req(const ipsec_req_t *req, ipsec_prot_t *ipp)
1343 {
1344         bzero(ipp, sizeof (*ipp));
1345         /*
1346          * ipp_use_* are bitfields.  Look at "!!" in the following as a
1347          * "boolean canonicalization" operator.
1348          */
1349         ipp->ipp_use_ah = !!(req->ipsr_ah_req & IPSEC_PREF_REQUIRED);
1350         ipp->ipp_use_esp = !!(req->ipsr_esp_req & IPSEC_PREF_REQUIRED);
1351         ipp->ipp_use_espa = !!(req->ipsr_esp_auth_alg);
1352         ipp->ipp_use_se = !!(req->ipsr_self_encap_req & IPSEC_PREF_REQUIRED);
1353         ipp->ipp_use_unique = !!((req->ipsr_ah_req|req->ipsr_esp_req) &
1354             IPSEC_PREF_UNIQUE);
1355         ipp->ipp_encr_alg = req->ipsr_esp_alg;
1356         /*
1357          * SADB_AALG_ANY is a placeholder to distinguish "any" from
1358          * "none" above.  If auth is required, as determined above,
1359          * SADB_AALG_ANY becomes 0, which is the representation
1360          * of "any" and "none" in PF_KEY v2.
1361          */
1362         ipp->ipp_auth_alg = (req->ipsr_auth_alg != SADB_AALG_ANY) ?
1363             req->ipsr_auth_alg : 0;
1364         ipp->ipp_esp_auth_alg = (req->ipsr_esp_auth_alg != SADB_AALG_ANY) ?
1365             req->ipsr_esp_auth_alg : 0;
1366 }
1367 
1368 /*
1369  * Extract a new-style action from a request.
1370  */
1371 void
1372 ipsec_actvec_from_req(const ipsec_req_t *req, ipsec_act_t **actp, uint_t *nactp,
1373     netstack_t *ns)
1374 {
1375         struct ipsec_act act;
1376 
1377         bzero(&act, sizeof (act));
1378         if ((req->ipsr_ah_req & IPSEC_PREF_NEVER) &&
1379             (req->ipsr_esp_req & IPSEC_PREF_NEVER)) {
1380                 act.ipa_type = IPSEC_ACT_BYPASS;
1381         } else {
1382                 act.ipa_type = IPSEC_ACT_APPLY;
1383                 ipsec_prot_from_req(req, &act.ipa_apply);
1384         }
1385         *actp = ipsec_act_wildcard_expand(&act, nactp, ns);
1386 }
1387 
1388 /*
1389  * Convert a new-style "prot" back to an ipsec_req_t (more backwards compat).
1390  * We assume caller has already zero'ed *req for us.
1391  */
1392 static int
1393 ipsec_req_from_prot(ipsec_prot_t *ipp, ipsec_req_t *req)
1394 {
1395         req->ipsr_esp_alg = ipp->ipp_encr_alg;
1396         req->ipsr_auth_alg = ipp->ipp_auth_alg;
1397         req->ipsr_esp_auth_alg = ipp->ipp_esp_auth_alg;
1398 
1399         if (ipp->ipp_use_unique) {
1400                 req->ipsr_ah_req |= IPSEC_PREF_UNIQUE;
1401                 req->ipsr_esp_req |= IPSEC_PREF_UNIQUE;
1402         }
1403         if (ipp->ipp_use_se)
1404                 req->ipsr_self_encap_req |= IPSEC_PREF_REQUIRED;
1405         if (ipp->ipp_use_ah)
1406                 req->ipsr_ah_req |= IPSEC_PREF_REQUIRED;
1407         if (ipp->ipp_use_esp)
1408                 req->ipsr_esp_req |= IPSEC_PREF_REQUIRED;
1409         return (sizeof (*req));
1410 }
1411 
1412 /*
1413  * Convert a new-style action back to an ipsec_req_t (more backwards compat).
1414  * We assume caller has already zero'ed *req for us.
1415  */
1416 static int
1417 ipsec_req_from_act(ipsec_action_t *ap, ipsec_req_t *req)
1418 {
1419         switch (ap->ipa_act.ipa_type) {
1420         case IPSEC_ACT_BYPASS:
1421                 req->ipsr_ah_req = IPSEC_PREF_NEVER;
1422                 req->ipsr_esp_req = IPSEC_PREF_NEVER;
1423                 return (sizeof (*req));
1424         case IPSEC_ACT_APPLY:
1425                 return (ipsec_req_from_prot(&ap->ipa_act.ipa_apply, req));
1426         }
1427         return (sizeof (*req));
1428 }
1429 
1430 /*
1431  * Convert a new-style action back to an ipsec_req_t (more backwards compat).
1432  * We assume caller has already zero'ed *req for us.
1433  */
1434 int
1435 ipsec_req_from_head(ipsec_policy_head_t *ph, ipsec_req_t *req, int af)
1436 {
1437         ipsec_policy_t *p;
1438 
1439         /*
1440          * FULL-PERSOCK: consult hash table, too?
1441          */
1442         for (p = ph->iph_root[IPSEC_INBOUND].ipr_nonhash[af];
1443             p != NULL;
1444             p = p->ipsp_hash.hash_next) {
1445                 if ((p->ipsp_sel->ipsl_key.ipsl_valid & IPSL_WILDCARD) == 0)
1446                         return (ipsec_req_from_act(p->ipsp_act, req));
1447         }
1448         return (sizeof (*req));
1449 }
1450 
1451 /*
1452  * Based on per-socket or latched policy, convert to an appropriate
1453  * IP_SEC_OPT ipsec_req_t for the socket option; return size so we can
1454  * be tail-called from ip.
1455  */
1456 int
1457 ipsec_req_from_conn(conn_t *connp, ipsec_req_t *req, int af)
1458 {
1459         ipsec_latch_t *ipl;
1460         int rv = sizeof (ipsec_req_t);
1461 
1462         bzero(req, sizeof (*req));
1463 
1464         ASSERT(MUTEX_HELD(&connp->conn_lock));
1465         ipl = connp->conn_latch;
1466 
1467         /*
1468          * Find appropriate policy.  First choice is latched action;
1469          * failing that, see latched policy; failing that,
1470          * look at configured policy.
1471          */
1472         if (ipl != NULL) {
1473                 if (connp->conn_latch_in_action != NULL) {
1474                         rv = ipsec_req_from_act(connp->conn_latch_in_action,
1475                             req);
1476                         goto done;
1477                 }
1478                 if (connp->conn_latch_in_policy != NULL) {
1479                         rv = ipsec_req_from_act(
1480                             connp->conn_latch_in_policy->ipsp_act, req);
1481                         goto done;
1482                 }
1483         }
1484         if (connp->conn_policy != NULL)
1485                 rv = ipsec_req_from_head(connp->conn_policy, req, af);
1486 done:
1487         return (rv);
1488 }
1489 
1490 void
1491 ipsec_actvec_free(ipsec_act_t *act, uint_t nact)
1492 {
1493         kmem_free(act, nact * sizeof (*act));
1494 }
1495 
1496 /*
1497  * Consumes a reference to ipsp.
1498  */
1499 static mblk_t *
1500 ipsec_check_loopback_policy(mblk_t *data_mp, ip_recv_attr_t *ira,
1501     ipsec_policy_t *ipsp)
1502 {
1503         if (!(ira->ira_flags & IRAF_IPSEC_SECURE))
1504                 return (data_mp);
1505 
1506         ASSERT(ira->ira_flags & IRAF_LOOPBACK);
1507 
1508         IPPOL_REFRELE(ipsp);
1509 
1510         /*
1511          * We should do an actual policy check here.  Revisit this
1512          * when we revisit the IPsec API.  (And pass a conn_t in when we
1513          * get there.)
1514          */
1515 
1516         return (data_mp);
1517 }
1518 
1519 /*
1520  * Check that packet's inbound ports & proto match the selectors
1521  * expected by the SAs it traversed on the way in.
1522  */
1523 static boolean_t
1524 ipsec_check_ipsecin_unique(ip_recv_attr_t *ira, const char **reason,
1525     kstat_named_t **counter, uint64_t pkt_unique, netstack_t *ns)
1526 {
1527         uint64_t ah_mask, esp_mask;
1528         ipsa_t *ah_assoc;
1529         ipsa_t *esp_assoc;
1530         ipsec_stack_t   *ipss = ns->netstack_ipsec;
1531 
1532         ASSERT(ira->ira_flags & IRAF_IPSEC_SECURE);
1533         ASSERT(!(ira->ira_flags & IRAF_LOOPBACK));
1534 
1535         ah_assoc = ira->ira_ipsec_ah_sa;
1536         esp_assoc = ira->ira_ipsec_esp_sa;
1537         ASSERT((ah_assoc != NULL) || (esp_assoc != NULL));
1538 
1539         ah_mask = (ah_assoc != NULL) ? ah_assoc->ipsa_unique_mask : 0;
1540         esp_mask = (esp_assoc != NULL) ? esp_assoc->ipsa_unique_mask : 0;
1541 
1542         if ((ah_mask == 0) && (esp_mask == 0))
1543                 return (B_TRUE);
1544 
1545         /*
1546          * The pkt_unique check will also check for tunnel mode on the SA
1547          * vs. the tunneled_packet boolean.  "Be liberal in what you receive"
1548          * should not apply in this case.  ;)
1549          */
1550 
1551         if (ah_mask != 0 &&
1552             ah_assoc->ipsa_unique_id != (pkt_unique & ah_mask)) {
1553                 *reason = "AH inner header mismatch";
1554                 *counter = DROPPER(ipss, ipds_spd_ah_innermismatch);
1555                 return (B_FALSE);
1556         }
1557         if (esp_mask != 0 &&
1558             esp_assoc->ipsa_unique_id != (pkt_unique & esp_mask)) {
1559                 *reason = "ESP inner header mismatch";
1560                 *counter = DROPPER(ipss, ipds_spd_esp_innermismatch);
1561                 return (B_FALSE);
1562         }
1563         return (B_TRUE);
1564 }
1565 
1566 static boolean_t
1567 ipsec_check_ipsecin_action(ip_recv_attr_t *ira, mblk_t *mp, ipsec_action_t *ap,
1568     ipha_t *ipha, ip6_t *ip6h, const char **reason, kstat_named_t **counter,
1569     netstack_t *ns)
1570 {
1571         boolean_t ret = B_TRUE;
1572         ipsec_prot_t *ipp;
1573         ipsa_t *ah_assoc;
1574         ipsa_t *esp_assoc;
1575         boolean_t decaps;
1576         ipsec_stack_t   *ipss = ns->netstack_ipsec;
1577 
1578         ASSERT((ipha == NULL && ip6h != NULL) ||
1579             (ip6h == NULL && ipha != NULL));
1580 
1581         if (ira->ira_flags & IRAF_LOOPBACK) {
1582                 /*
1583                  * Besides accepting pointer-equivalent actions, we also
1584                  * accept any ICMP errors we generated for ourselves,
1585                  * regardless of policy.  If we do not wish to make this
1586                  * assumption in the future, check here, and where
1587                  * IXAF_TRUSTED_ICMP is initialized in ip.c and ip6.c.
1588                  */
1589                 if (ap == ira->ira_ipsec_action ||
1590                     (ira->ira_flags & IRAF_TRUSTED_ICMP))
1591                         return (B_TRUE);
1592 
1593                 /* Deep compare necessary here?? */
1594                 *counter = DROPPER(ipss, ipds_spd_loopback_mismatch);
1595                 *reason = "loopback policy mismatch";
1596                 return (B_FALSE);
1597         }
1598         ASSERT(!(ira->ira_flags & IRAF_TRUSTED_ICMP));
1599         ASSERT(ira->ira_flags & IRAF_IPSEC_SECURE);
1600 
1601         ah_assoc = ira->ira_ipsec_ah_sa;
1602         esp_assoc = ira->ira_ipsec_esp_sa;
1603 
1604         decaps = (ira->ira_flags & IRAF_IPSEC_DECAPS);
1605 
1606         switch (ap->ipa_act.ipa_type) {
1607         case IPSEC_ACT_DISCARD:
1608         case IPSEC_ACT_REJECT:
1609                 /* Should "fail hard" */
1610                 *counter = DROPPER(ipss, ipds_spd_explicit);
1611                 *reason = "blocked by policy";
1612                 return (B_FALSE);
1613 
1614         case IPSEC_ACT_BYPASS:
1615         case IPSEC_ACT_CLEAR:
1616                 *counter = DROPPER(ipss, ipds_spd_got_secure);
1617                 *reason = "expected clear, got protected";
1618                 return (B_FALSE);
1619 
1620         case IPSEC_ACT_APPLY:
1621                 ipp = &ap->ipa_act.ipa_apply;
1622                 /*
1623                  * As of now we do the simple checks of whether
1624                  * the datagram has gone through the required IPSEC
1625                  * protocol constraints or not. We might have more
1626                  * in the future like sensitive levels, key bits, etc.
1627                  * If it fails the constraints, check whether we would
1628                  * have accepted this if it had come in clear.
1629                  */
1630                 if (ipp->ipp_use_ah) {
1631                         if (ah_assoc == NULL) {
1632                                 ret = ipsec_inbound_accept_clear(mp, ipha,
1633                                     ip6h);
1634                                 *counter = DROPPER(ipss, ipds_spd_got_clear);
1635                                 *reason = "unprotected not accepted";
1636                                 break;
1637                         }
1638                         ASSERT(ah_assoc != NULL);
1639                         ASSERT(ipp->ipp_auth_alg != 0);
1640 
1641                         if (ah_assoc->ipsa_auth_alg !=
1642                             ipp->ipp_auth_alg) {
1643                                 *counter = DROPPER(ipss, ipds_spd_bad_ahalg);
1644                                 *reason = "unacceptable ah alg";
1645                                 ret = B_FALSE;
1646                                 break;
1647                         }
1648                 } else if (ah_assoc != NULL) {
1649                         /*
1650                          * Don't allow this. Check IPSEC NOTE above
1651                          * ip_fanout_proto().
1652                          */
1653                         *counter = DROPPER(ipss, ipds_spd_got_ah);
1654                         *reason = "unexpected AH";
1655                         ret = B_FALSE;
1656                         break;
1657                 }
1658                 if (ipp->ipp_use_esp) {
1659                         if (esp_assoc == NULL) {
1660                                 ret = ipsec_inbound_accept_clear(mp, ipha,
1661                                     ip6h);
1662                                 *counter = DROPPER(ipss, ipds_spd_got_clear);
1663                                 *reason = "unprotected not accepted";
1664                                 break;
1665                         }
1666                         ASSERT(esp_assoc != NULL);
1667                         ASSERT(ipp->ipp_encr_alg != 0);
1668 
1669                         if (esp_assoc->ipsa_encr_alg !=
1670                             ipp->ipp_encr_alg) {
1671                                 *counter = DROPPER(ipss, ipds_spd_bad_espealg);
1672                                 *reason = "unacceptable esp alg";
1673                                 ret = B_FALSE;
1674                                 break;
1675                         }
1676                         /*
1677                          * If the client does not need authentication,
1678                          * we don't verify the alogrithm.
1679                          */
1680                         if (ipp->ipp_use_espa) {
1681                                 if (esp_assoc->ipsa_auth_alg !=
1682                                     ipp->ipp_esp_auth_alg) {
1683                                         *counter = DROPPER(ipss,
1684                                             ipds_spd_bad_espaalg);
1685                                         *reason = "unacceptable esp auth alg";
1686                                         ret = B_FALSE;
1687                                         break;
1688                                 }
1689                         }
1690                 } else if (esp_assoc != NULL) {
1691                         /*
1692                          * Don't allow this. Check IPSEC NOTE above
1693                          * ip_fanout_proto().
1694                          */
1695                         *counter = DROPPER(ipss, ipds_spd_got_esp);
1696                         *reason = "unexpected ESP";
1697                         ret = B_FALSE;
1698                         break;
1699                 }
1700                 if (ipp->ipp_use_se) {
1701                         if (!decaps) {
1702                                 ret = ipsec_inbound_accept_clear(mp, ipha,
1703                                     ip6h);
1704                                 if (!ret) {
1705                                         /* XXX mutant? */
1706                                         *counter = DROPPER(ipss,
1707                                             ipds_spd_bad_selfencap);
1708                                         *reason = "self encap not found";
1709                                         break;
1710                                 }
1711                         }
1712                 } else if (decaps) {
1713                         /*
1714                          * XXX If the packet comes in tunneled and the
1715                          * recipient does not expect it to be tunneled, it
1716                          * is okay. But we drop to be consistent with the
1717                          * other cases.
1718                          */
1719                         *counter = DROPPER(ipss, ipds_spd_got_selfencap);
1720                         *reason = "unexpected self encap";
1721                         ret = B_FALSE;
1722                         break;
1723                 }
1724                 if (ira->ira_ipsec_action != NULL) {
1725                         /*
1726                          * This can happen if we do a double policy-check on
1727                          * a packet
1728                          * XXX XXX should fix this case!
1729                          */
1730                         IPACT_REFRELE(ira->ira_ipsec_action);
1731                 }
1732                 ASSERT(ira->ira_flags & IRAF_IPSEC_SECURE);
1733                 ASSERT(ira->ira_ipsec_action == NULL);
1734                 IPACT_REFHOLD(ap);
1735                 ira->ira_ipsec_action = ap;
1736                 break;  /* from switch */
1737         }
1738         return (ret);
1739 }
1740 
1741 static boolean_t
1742 spd_match_inbound_ids(ipsec_latch_t *ipl, ipsa_t *sa)
1743 {
1744         ASSERT(ipl->ipl_ids_latched == B_TRUE);
1745         return ipsid_equal(ipl->ipl_remote_cid, sa->ipsa_src_cid) &&
1746             ipsid_equal(ipl->ipl_local_cid, sa->ipsa_dst_cid);
1747 }
1748 
1749 /*
1750  * Takes a latched conn and an inbound packet and returns a unique_id suitable
1751  * for SA comparisons.  Most of the time we will copy from the conn_t, but
1752  * there are cases when the conn_t is latched but it has wildcard selectors,
1753  * and then we need to fallback to scooping them out of the packet.
1754  *
1755  * Assume we'll never have 0 with a conn_t present, so use 0 as a failure.  We
1756  * can get away with this because we only have non-zero ports/proto for
1757  * latched conn_ts.
1758  *
1759  * Ideal candidate for an "inline" keyword, as we're JUST convoluted enough
1760  * to not be a nice macro.
1761  */
1762 static uint64_t
1763 conn_to_unique(conn_t *connp, mblk_t *data_mp, ipha_t *ipha, ip6_t *ip6h)
1764 {
1765         ipsec_selector_t sel;
1766         uint8_t ulp = connp->conn_proto;
1767 
1768         ASSERT(connp->conn_latch_in_policy != NULL);
1769 
1770         if ((ulp == IPPROTO_TCP || ulp == IPPROTO_UDP || ulp == IPPROTO_SCTP) &&
1771             (connp->conn_fport == 0 || connp->conn_lport == 0)) {
1772                 /* Slow path - we gotta grab from the packet. */
1773                 if (ipsec_init_inbound_sel(&sel, data_mp, ipha, ip6h,
1774                     SEL_NONE) != SELRET_SUCCESS) {
1775                         /* Failure -> have caller free packet with ENOMEM. */
1776                         return (0);
1777                 }
1778                 return (SA_UNIQUE_ID(sel.ips_remote_port, sel.ips_local_port,
1779                     sel.ips_protocol, 0));
1780         }
1781 
1782 #ifdef DEBUG_NOT_UNTIL_6478464
1783         if (ipsec_init_inbound_sel(&sel, data_mp, ipha, ip6h, SEL_NONE) ==
1784             SELRET_SUCCESS) {
1785                 ASSERT(sel.ips_local_port == connp->conn_lport);
1786                 ASSERT(sel.ips_remote_port == connp->conn_fport);
1787                 ASSERT(sel.ips_protocol == connp->conn_proto);
1788         }
1789         ASSERT(connp->conn_proto != 0);
1790 #endif
1791 
1792         return (SA_UNIQUE_ID(connp->conn_fport, connp->conn_lport, ulp, 0));
1793 }
1794 
1795 /*
1796  * Called to check policy on a latched connection.
1797  * Note that we don't dereference conn_latch or conn_ihere since the conn might
1798  * be closing. The caller passes a held ipsec_latch_t instead.
1799  */
1800 static boolean_t
1801 ipsec_check_ipsecin_latch(ip_recv_attr_t *ira, mblk_t *mp, ipsec_latch_t *ipl,
1802     ipsec_action_t *ap, ipha_t *ipha, ip6_t *ip6h, const char **reason,
1803     kstat_named_t **counter, conn_t *connp, netstack_t *ns)
1804 {
1805         ipsec_stack_t   *ipss = ns->netstack_ipsec;
1806 
1807         ASSERT(ipl->ipl_ids_latched == B_TRUE);
1808         ASSERT(ira->ira_flags & IRAF_IPSEC_SECURE);
1809 
1810         if (!(ira->ira_flags & IRAF_LOOPBACK)) {
1811                 /*
1812                  * Over loopback, there aren't real security associations,
1813                  * so there are neither identities nor "unique" values
1814                  * for us to check the packet against.
1815                  */
1816                 if (ira->ira_ipsec_ah_sa != NULL) {
1817                         if (!spd_match_inbound_ids(ipl,
1818                             ira->ira_ipsec_ah_sa)) {
1819                                 *counter = DROPPER(ipss, ipds_spd_ah_badid);
1820                                 *reason = "AH identity mismatch";
1821                                 return (B_FALSE);
1822                         }
1823                 }
1824 
1825                 if (ira->ira_ipsec_esp_sa != NULL) {
1826                         if (!spd_match_inbound_ids(ipl,
1827                             ira->ira_ipsec_esp_sa)) {
1828                                 *counter = DROPPER(ipss, ipds_spd_esp_badid);
1829                                 *reason = "ESP identity mismatch";
1830                                 return (B_FALSE);
1831                         }
1832                 }
1833 
1834                 /*
1835                  * Can fudge pkt_unique from connp because we're latched.
1836                  * In DEBUG kernels (see conn_to_unique()'s implementation),
1837                  * verify this even if it REALLY slows things down.
1838                  */
1839                 if (!ipsec_check_ipsecin_unique(ira, reason, counter,
1840                     conn_to_unique(connp, mp, ipha, ip6h), ns)) {
1841                         return (B_FALSE);
1842                 }
1843         }
1844         return (ipsec_check_ipsecin_action(ira, mp, ap, ipha, ip6h, reason,
1845             counter, ns));
1846 }
1847 
1848 /*
1849  * Check to see whether this secured datagram meets the policy
1850  * constraints specified in ipsp.
1851  *
1852  * Called from ipsec_check_global_policy, and ipsec_check_inbound_policy.
1853  *
1854  * Consumes a reference to ipsp.
1855  * Returns the mblk if ok.
1856  */
1857 static mblk_t *
1858 ipsec_check_ipsecin_policy(mblk_t *data_mp, ipsec_policy_t *ipsp,
1859     ipha_t *ipha, ip6_t *ip6h, uint64_t pkt_unique, ip_recv_attr_t *ira,
1860     netstack_t *ns)
1861 {
1862         ipsec_action_t *ap;
1863         const char *reason = "no policy actions found";
1864         ip_stack_t      *ipst = ns->netstack_ip;
1865         ipsec_stack_t   *ipss = ns->netstack_ipsec;
1866         kstat_named_t *counter;
1867 
1868         counter = DROPPER(ipss, ipds_spd_got_secure);
1869 
1870         ASSERT(ipsp != NULL);
1871 
1872         ASSERT((ipha == NULL && ip6h != NULL) ||
1873             (ip6h == NULL && ipha != NULL));
1874 
1875         if (ira->ira_flags & IRAF_LOOPBACK)
1876                 return (ipsec_check_loopback_policy(data_mp, ira, ipsp));
1877 
1878         ASSERT(ira->ira_flags & IRAF_IPSEC_SECURE);
1879 
1880         if (ira->ira_ipsec_action != NULL) {
1881                 /*
1882                  * this can happen if we do a double policy-check on a packet
1883                  * Would be nice to be able to delete this test..
1884                  */
1885                 IPACT_REFRELE(ira->ira_ipsec_action);
1886         }
1887         ASSERT(ira->ira_ipsec_action == NULL);
1888 
1889         if (!SA_IDS_MATCH(ira->ira_ipsec_ah_sa, ira->ira_ipsec_esp_sa)) {
1890                 reason = "inbound AH and ESP identities differ";
1891                 counter = DROPPER(ipss, ipds_spd_ahesp_diffid);
1892                 goto drop;
1893         }
1894 
1895         if (!ipsec_check_ipsecin_unique(ira, &reason, &counter, pkt_unique,
1896             ns))
1897                 goto drop;
1898 
1899         /*
1900          * Ok, now loop through the possible actions and see if any
1901          * of them work for us.
1902          */
1903 
1904         for (ap = ipsp->ipsp_act; ap != NULL; ap = ap->ipa_next) {
1905                 if (ipsec_check_ipsecin_action(ira, data_mp, ap,
1906                     ipha, ip6h, &reason, &counter, ns)) {
1907                         BUMP_MIB(&ipst->ips_ip_mib, ipsecInSucceeded);
1908                         IPPOL_REFRELE(ipsp);
1909                         return (data_mp);
1910                 }
1911         }
1912 drop:
1913         ipsec_rl_strlog(ns, IP_MOD_ID, 0, 0, SL_ERROR|SL_WARN|SL_CONSOLE,
1914             "ipsec inbound policy mismatch: %s, packet dropped\n",
1915             reason);
1916         IPPOL_REFRELE(ipsp);
1917         ASSERT(ira->ira_ipsec_action == NULL);
1918         BUMP_MIB(&ipst->ips_ip_mib, ipsecInFailed);
1919         ip_drop_packet(data_mp, B_TRUE, NULL, counter,
1920             &ipss->ipsec_spd_dropper);
1921         return (NULL);
1922 }
1923 
1924 /*
1925  * sleazy prefix-length-based compare.
1926  * another inlining candidate..
1927  */
1928 boolean_t
1929 ip_addr_match(uint8_t *addr1, int pfxlen, in6_addr_t *addr2p)
1930 {
1931         int offset = pfxlen>>3;
1932         int bitsleft = pfxlen & 7;
1933         uint8_t *addr2 = (uint8_t *)addr2p;
1934 
1935         /*
1936          * and there was much evil..
1937          * XXX should inline-expand the bcmp here and do this 32 bits
1938          * or 64 bits at a time..
1939          */
1940         return ((bcmp(addr1, addr2, offset) == 0) &&
1941             ((bitsleft == 0) ||
1942             (((addr1[offset] ^ addr2[offset]) & (0xff<<(8-bitsleft))) == 0)));
1943 }
1944 
1945 static ipsec_policy_t *
1946 ipsec_find_policy_chain(ipsec_policy_t *best, ipsec_policy_t *chain,
1947     ipsec_selector_t *sel, boolean_t is_icmp_inv_acq)
1948 {
1949         ipsec_selkey_t *isel;
1950         ipsec_policy_t *p;
1951         int bpri = best ? best->ipsp_prio : 0;
1952 
1953         for (p = chain; p != NULL; p = p->ipsp_hash.hash_next) {
1954                 uint32_t valid;
1955 
1956                 if (p->ipsp_prio <= bpri)
1957                         continue;
1958                 isel = &p->ipsp_sel->ipsl_key;
1959                 valid = isel->ipsl_valid;
1960 
1961                 if ((valid & IPSL_PROTOCOL) &&
1962                     (isel->ipsl_proto != sel->ips_protocol))
1963                         continue;
1964 
1965                 if ((valid & IPSL_REMOTE_ADDR) &&
1966                     !ip_addr_match((uint8_t *)&isel->ipsl_remote,
1967                     isel->ipsl_remote_pfxlen, &sel->ips_remote_addr_v6))
1968                         continue;
1969 
1970                 if ((valid & IPSL_LOCAL_ADDR) &&
1971                     !ip_addr_match((uint8_t *)&isel->ipsl_local,
1972                     isel->ipsl_local_pfxlen, &sel->ips_local_addr_v6))
1973                         continue;
1974 
1975                 if ((valid & IPSL_REMOTE_PORT) &&
1976                     isel->ipsl_rport != sel->ips_remote_port)
1977                         continue;
1978 
1979                 if ((valid & IPSL_LOCAL_PORT) &&
1980                     isel->ipsl_lport != sel->ips_local_port)
1981                         continue;
1982 
1983                 if (!is_icmp_inv_acq) {
1984                         if ((valid & IPSL_ICMP_TYPE) &&
1985                             (isel->ipsl_icmp_type > sel->ips_icmp_type ||
1986                             isel->ipsl_icmp_type_end < sel->ips_icmp_type)) {
1987                                 continue;
1988                         }
1989 
1990                         if ((valid & IPSL_ICMP_CODE) &&
1991                             (isel->ipsl_icmp_code > sel->ips_icmp_code ||
1992                             isel->ipsl_icmp_code_end <
1993                             sel->ips_icmp_code)) {
1994                                 continue;
1995                         }
1996                 } else {
1997                         /*
1998                          * special case for icmp inverse acquire
1999                          * we only want policies that aren't drop/pass
2000                          */
2001                         if (p->ipsp_act->ipa_act.ipa_type != IPSEC_ACT_APPLY)
2002                                 continue;
2003                 }
2004 
2005                 /* we matched all the packet-port-field selectors! */
2006                 best = p;
2007                 bpri = p->ipsp_prio;
2008         }
2009 
2010         return (best);
2011 }
2012 
2013 /*
2014  * Try to find and return the best policy entry under a given policy
2015  * root for a given set of selectors; the first parameter "best" is
2016  * the current best policy so far.  If "best" is non-null, we have a
2017  * reference to it.  We return a reference to a policy; if that policy
2018  * is not the original "best", we need to release that reference
2019  * before returning.
2020  */
2021 ipsec_policy_t *
2022 ipsec_find_policy_head(ipsec_policy_t *best, ipsec_policy_head_t *head,
2023     int direction, ipsec_selector_t *sel)
2024 {
2025         ipsec_policy_t *curbest;
2026         ipsec_policy_root_t *root;
2027         uint8_t is_icmp_inv_acq = sel->ips_is_icmp_inv_acq;
2028         int af = sel->ips_isv4 ? IPSEC_AF_V4 : IPSEC_AF_V6;
2029 
2030         curbest = best;
2031         root = &head->iph_root[direction];
2032 
2033 #ifdef DEBUG
2034         if (is_icmp_inv_acq) {
2035                 if (sel->ips_isv4) {
2036                         if (sel->ips_protocol != IPPROTO_ICMP) {
2037                                 cmn_err(CE_WARN, "ipsec_find_policy_head:"
2038                                     " expecting icmp, got %d",
2039                                     sel->ips_protocol);
2040                         }
2041                 } else {
2042                         if (sel->ips_protocol != IPPROTO_ICMPV6) {
2043                                 cmn_err(CE_WARN, "ipsec_find_policy_head:"
2044                                     " expecting icmpv6, got %d",
2045                                     sel->ips_protocol);
2046                         }
2047                 }
2048         }
2049 #endif
2050 
2051         rw_enter(&head->iph_lock, RW_READER);
2052 
2053         if (root->ipr_nchains > 0) {
2054                 curbest = ipsec_find_policy_chain(curbest,
2055                     root->ipr_hash[selector_hash(sel, root)].hash_head, sel,
2056                     is_icmp_inv_acq);
2057         }
2058         curbest = ipsec_find_policy_chain(curbest, root->ipr_nonhash[af], sel,
2059             is_icmp_inv_acq);
2060 
2061         /*
2062          * Adjust reference counts if we found anything new.
2063          */
2064         if (curbest != best) {
2065                 ASSERT(curbest != NULL);
2066                 IPPOL_REFHOLD(curbest);
2067 
2068                 if (best != NULL) {
2069                         IPPOL_REFRELE(best);
2070                 }
2071         }
2072 
2073         rw_exit(&head->iph_lock);
2074 
2075         return (curbest);
2076 }
2077 
2078 /*
2079  * Find the best system policy (either global or per-interface) which
2080  * applies to the given selector; look in all the relevant policy roots
2081  * to figure out which policy wins.
2082  *
2083  * Returns a reference to a policy; caller must release this
2084  * reference when done.
2085  */
2086 ipsec_policy_t *
2087 ipsec_find_policy(int direction, const conn_t *connp, ipsec_selector_t *sel,
2088     netstack_t *ns)
2089 {
2090         ipsec_policy_t *p;
2091         ipsec_stack_t   *ipss = ns->netstack_ipsec;
2092 
2093         p = ipsec_find_policy_head(NULL, &ipss->ipsec_system_policy,
2094             direction, sel);
2095         if ((connp != NULL) && (connp->conn_policy != NULL)) {
2096                 p = ipsec_find_policy_head(p, connp->conn_policy,
2097                     direction, sel);
2098         }
2099 
2100         return (p);
2101 }
2102 
2103 /*
2104  * Check with global policy and see whether this inbound
2105  * packet meets the policy constraints.
2106  *
2107  * Locate appropriate policy from global policy, supplemented by the
2108  * conn's configured and/or cached policy if the conn is supplied.
2109  *
2110  * Dispatch to ipsec_check_ipsecin_policy if we have policy and an
2111  * encrypted packet to see if they match.
2112  *
2113  * Otherwise, see if the policy allows cleartext; if not, drop it on the
2114  * floor.
2115  */
2116 mblk_t *
2117 ipsec_check_global_policy(mblk_t *data_mp, conn_t *connp,
2118     ipha_t *ipha, ip6_t *ip6h, ip_recv_attr_t *ira, netstack_t *ns)
2119 {
2120         ipsec_policy_t *p;
2121         ipsec_selector_t sel;
2122         boolean_t policy_present;
2123         kstat_named_t *counter;
2124         uint64_t pkt_unique;
2125         ip_stack_t      *ipst = ns->netstack_ip;
2126         ipsec_stack_t   *ipss = ns->netstack_ipsec;
2127 
2128         sel.ips_is_icmp_inv_acq = 0;
2129 
2130         ASSERT((ipha == NULL && ip6h != NULL) ||
2131             (ip6h == NULL && ipha != NULL));
2132 
2133         if (ipha != NULL)
2134                 policy_present = ipss->ipsec_inbound_v4_policy_present;
2135         else
2136                 policy_present = ipss->ipsec_inbound_v6_policy_present;
2137 
2138         if (!policy_present && connp == NULL) {
2139                 /*
2140                  * No global policy and no per-socket policy;
2141                  * just pass it back (but we shouldn't get here in that case)
2142                  */
2143                 return (data_mp);
2144         }
2145 
2146         /*
2147          * If we have cached policy, use it.
2148          * Otherwise consult system policy.
2149          */
2150         if ((connp != NULL) && (connp->conn_latch != NULL)) {
2151                 p = connp->conn_latch_in_policy;
2152                 if (p != NULL) {
2153                         IPPOL_REFHOLD(p);
2154                 }
2155                 /*
2156                  * Fudge sel for UNIQUE_ID setting below.
2157                  */
2158                 pkt_unique = conn_to_unique(connp, data_mp, ipha, ip6h);
2159         } else {
2160                 /* Initialize the ports in the selector */
2161                 if (ipsec_init_inbound_sel(&sel, data_mp, ipha, ip6h,
2162                     SEL_NONE) == SELRET_NOMEM) {
2163                         /*
2164                          * Technically not a policy mismatch, but it is
2165                          * an internal failure.
2166                          */
2167                         ipsec_log_policy_failure(IPSEC_POLICY_MISMATCH,
2168                             "ipsec_init_inbound_sel", ipha, ip6h, B_TRUE, ns);
2169                         counter = DROPPER(ipss, ipds_spd_nomem);
2170                         goto fail;
2171                 }
2172 
2173                 /*
2174                  * Find the policy which best applies.
2175                  *
2176                  * If we find global policy, we should look at both
2177                  * local policy and global policy and see which is
2178                  * stronger and match accordingly.
2179                  *
2180                  * If we don't find a global policy, check with
2181                  * local policy alone.
2182                  */
2183 
2184                 p = ipsec_find_policy(IPSEC_TYPE_INBOUND, connp, &sel, ns);
2185                 pkt_unique = SA_UNIQUE_ID(sel.ips_remote_port,
2186                     sel.ips_local_port, sel.ips_protocol, 0);
2187         }
2188 
2189         if (p == NULL) {
2190                 if (!(ira->ira_flags & IRAF_IPSEC_SECURE)) {
2191                         /*
2192                          * We have no policy; default to succeeding.
2193                          * XXX paranoid system design doesn't do this.
2194                          */
2195                         BUMP_MIB(&ipst->ips_ip_mib, ipsecInSucceeded);
2196                         return (data_mp);
2197                 } else {
2198                         counter = DROPPER(ipss, ipds_spd_got_secure);
2199                         ipsec_log_policy_failure(IPSEC_POLICY_NOT_NEEDED,
2200                             "ipsec_check_global_policy", ipha, ip6h, B_TRUE,
2201                             ns);
2202                         goto fail;
2203                 }
2204         }
2205         if (ira->ira_flags & IRAF_IPSEC_SECURE) {
2206                 return (ipsec_check_ipsecin_policy(data_mp, p, ipha, ip6h,
2207                     pkt_unique, ira, ns));
2208         }
2209         if (p->ipsp_act->ipa_allow_clear) {
2210                 BUMP_MIB(&ipst->ips_ip_mib, ipsecInSucceeded);
2211                 IPPOL_REFRELE(p);
2212                 return (data_mp);
2213         }
2214         IPPOL_REFRELE(p);
2215         /*
2216          * If we reach here, we will drop the packet because it failed the
2217          * global policy check because the packet was cleartext, and it
2218          * should not have been.
2219          */
2220         ipsec_log_policy_failure(IPSEC_POLICY_MISMATCH,
2221             "ipsec_check_global_policy", ipha, ip6h, B_FALSE, ns);
2222         counter = DROPPER(ipss, ipds_spd_got_clear);
2223 
2224 fail:
2225         ip_drop_packet(data_mp, B_TRUE, NULL, counter,
2226             &ipss->ipsec_spd_dropper);
2227         BUMP_MIB(&ipst->ips_ip_mib, ipsecInFailed);
2228         return (NULL);
2229 }
2230 
2231 /*
2232  * We check whether an inbound datagram is a valid one
2233  * to accept in clear. If it is secure, it is the job
2234  * of IPSEC to log information appropriately if it
2235  * suspects that it may not be the real one.
2236  *
2237  * It is called only while fanning out to the ULP
2238  * where ULP accepts only secure data and the incoming
2239  * is clear. Usually we never accept clear datagrams in
2240  * such cases. ICMP is the only exception.
2241  *
2242  * NOTE : We don't call this function if the client (ULP)
2243  * is willing to accept things in clear.
2244  */
2245 boolean_t
2246 ipsec_inbound_accept_clear(mblk_t *mp, ipha_t *ipha, ip6_t *ip6h)
2247 {
2248         ushort_t iph_hdr_length;
2249         icmph_t *icmph;
2250         icmp6_t *icmp6;
2251         uint8_t *nexthdrp;
2252 
2253         ASSERT((ipha != NULL && ip6h == NULL) ||
2254             (ipha == NULL && ip6h != NULL));
2255 
2256         if (ip6h != NULL) {
2257                 iph_hdr_length = ip_hdr_length_v6(mp, ip6h);
2258                 if (!ip_hdr_length_nexthdr_v6(mp, ip6h, &iph_hdr_length,
2259                     &nexthdrp)) {
2260                         return (B_FALSE);
2261                 }
2262                 if (*nexthdrp != IPPROTO_ICMPV6)
2263                         return (B_FALSE);
2264                 icmp6 = (icmp6_t *)(&mp->b_rptr[iph_hdr_length]);
2265                 /* Match IPv6 ICMP policy as closely as IPv4 as possible. */
2266                 switch (icmp6->icmp6_type) {
2267                 case ICMP6_PARAM_PROB:
2268                         /* Corresponds to port/proto unreach in IPv4. */
2269                 case ICMP6_ECHO_REQUEST:
2270                         /* Just like IPv4. */
2271                         return (B_FALSE);
2272 
2273                 case MLD_LISTENER_QUERY:
2274                 case MLD_LISTENER_REPORT:
2275                 case MLD_LISTENER_REDUCTION:
2276                         /*
2277                          * XXX Seperate NDD in IPv4 what about here?
2278                          * Plus, mcast is important to ND.
2279                          */
2280                 case ICMP6_DST_UNREACH:
2281                         /* Corresponds to HOST/NET unreachable in IPv4. */
2282                 case ICMP6_PACKET_TOO_BIG:
2283                 case ICMP6_ECHO_REPLY:
2284                         /* These are trusted in IPv4. */
2285                 case ND_ROUTER_SOLICIT:
2286                 case ND_ROUTER_ADVERT:
2287                 case ND_NEIGHBOR_SOLICIT:
2288                 case ND_NEIGHBOR_ADVERT:
2289                 case ND_REDIRECT:
2290                         /* Trust ND messages for now. */
2291                 case ICMP6_TIME_EXCEEDED:
2292                 default:
2293                         return (B_TRUE);
2294                 }
2295         } else {
2296                 /*
2297                  * If it is not ICMP, fail this request.
2298                  */
2299                 if (ipha->ipha_protocol != IPPROTO_ICMP) {
2300 #ifdef FRAGCACHE_DEBUG
2301                         cmn_err(CE_WARN, "Dropping - ipha_proto = %d\n",
2302                             ipha->ipha_protocol);
2303 #endif
2304                         return (B_FALSE);
2305                 }
2306                 iph_hdr_length = IPH_HDR_LENGTH(ipha);
2307                 icmph = (icmph_t *)&mp->b_rptr[iph_hdr_length];
2308                 /*
2309                  * It is an insecure icmp message. Check to see whether we are
2310                  * willing to accept this one.
2311                  */
2312 
2313                 switch (icmph->icmph_type) {
2314                 case ICMP_ECHO_REPLY:
2315                 case ICMP_TIME_STAMP_REPLY:
2316                 case ICMP_INFO_REPLY:
2317                 case ICMP_ROUTER_ADVERTISEMENT:
2318                         /*
2319                          * We should not encourage clear replies if this
2320                          * client expects secure. If somebody is replying
2321                          * in clear some mailicious user watching both the
2322                          * request and reply, can do chosen-plain-text attacks.
2323                          * With global policy we might be just expecting secure
2324                          * but sending out clear. We don't know what the right
2325                          * thing is. We can't do much here as we can't control
2326                          * the sender here. Till we are sure of what to do,
2327                          * accept them.
2328                          */
2329                         return (B_TRUE);
2330                 case ICMP_ECHO_REQUEST:
2331                 case ICMP_TIME_STAMP_REQUEST:
2332                 case ICMP_INFO_REQUEST:
2333                 case ICMP_ADDRESS_MASK_REQUEST:
2334                 case ICMP_ROUTER_SOLICITATION:
2335                 case ICMP_ADDRESS_MASK_REPLY:
2336                         /*
2337                          * Don't accept this as somebody could be sending
2338                          * us plain text to get encrypted data. If we reply,
2339                          * it will lead to chosen plain text attack.
2340                          */
2341                         return (B_FALSE);
2342                 case ICMP_DEST_UNREACHABLE:
2343                         switch (icmph->icmph_code) {
2344                         case ICMP_FRAGMENTATION_NEEDED:
2345                                 /*
2346                                  * Be in sync with icmp_inbound, where we have
2347                                  * already set dce_pmtu
2348                                  */
2349 #ifdef FRAGCACHE_DEBUG
2350                         cmn_err(CE_WARN, "ICMP frag needed\n");
2351 #endif
2352                                 return (B_TRUE);
2353                         case ICMP_HOST_UNREACHABLE:
2354                         case ICMP_NET_UNREACHABLE:
2355                                 /*
2356                                  * By accepting, we could reset a connection.
2357                                  * How do we solve the problem of some
2358                                  * intermediate router sending in-secure ICMP
2359                                  * messages ?
2360                                  */
2361                                 return (B_TRUE);
2362                         case ICMP_PORT_UNREACHABLE:
2363                         case ICMP_PROTOCOL_UNREACHABLE:
2364                         default :
2365                                 return (B_FALSE);
2366                         }
2367                 case ICMP_SOURCE_QUENCH:
2368                         /*
2369                          * If this is an attack, TCP will slow start
2370                          * because of this. Is it very harmful ?
2371                          */
2372                         return (B_TRUE);
2373                 case ICMP_PARAM_PROBLEM:
2374                         return (B_FALSE);
2375                 case ICMP_TIME_EXCEEDED:
2376                         return (B_TRUE);
2377                 case ICMP_REDIRECT:
2378                         return (B_FALSE);
2379                 default :
2380                         return (B_FALSE);
2381                 }
2382         }
2383 }
2384 
2385 void
2386 ipsec_latch_ids(ipsec_latch_t *ipl, ipsid_t *local, ipsid_t *remote)
2387 {
2388         mutex_enter(&ipl->ipl_lock);
2389 
2390         if (ipl->ipl_ids_latched) {
2391                 /* I lost, someone else got here before me */
2392                 mutex_exit(&ipl->ipl_lock);
2393                 return;
2394         }
2395 
2396         if (local != NULL)
2397                 IPSID_REFHOLD(local);
2398         if (remote != NULL)
2399                 IPSID_REFHOLD(remote);
2400 
2401         ipl->ipl_local_cid = local;
2402         ipl->ipl_remote_cid = remote;
2403         ipl->ipl_ids_latched = B_TRUE;
2404         mutex_exit(&ipl->ipl_lock);
2405 }
2406 
2407 void
2408 ipsec_latch_inbound(conn_t *connp, ip_recv_attr_t *ira)
2409 {
2410         ipsa_t *sa;
2411         ipsec_latch_t *ipl = connp->conn_latch;
2412 
2413         if (!ipl->ipl_ids_latched) {
2414                 ipsid_t *local = NULL;
2415                 ipsid_t *remote = NULL;
2416 
2417                 if (!(ira->ira_flags & IRAF_LOOPBACK)) {
2418                         ASSERT(ira->ira_flags & IRAF_IPSEC_SECURE);
2419                         if (ira->ira_ipsec_esp_sa != NULL)
2420                                 sa = ira->ira_ipsec_esp_sa;
2421                         else
2422                                 sa = ira->ira_ipsec_ah_sa;
2423                         ASSERT(sa != NULL);
2424                         local = sa->ipsa_dst_cid;
2425                         remote = sa->ipsa_src_cid;
2426                 }
2427                 ipsec_latch_ids(ipl, local, remote);
2428         }
2429         if (ira->ira_flags & IRAF_IPSEC_SECURE) {
2430                 if (connp->conn_latch_in_action != NULL) {
2431                         /*
2432                          * Previously cached action.  This is probably
2433                          * harmless, but in DEBUG kernels, check for
2434                          * action equality.
2435                          *
2436                          * Preserve the existing action to preserve latch
2437                          * invariance.
2438                          */
2439                         ASSERT(connp->conn_latch_in_action ==
2440                             ira->ira_ipsec_action);
2441                         return;
2442                 }
2443                 connp->conn_latch_in_action = ira->ira_ipsec_action;
2444                 IPACT_REFHOLD(connp->conn_latch_in_action);
2445         }
2446 }
2447 
2448 /*
2449  * Check whether the policy constraints are met either for an
2450  * inbound datagram; called from IP in numerous places.
2451  *
2452  * Note that this is not a chokepoint for inbound policy checks;
2453  * see also ipsec_check_ipsecin_latch() and ipsec_check_global_policy()
2454  */
2455 mblk_t *
2456 ipsec_check_inbound_policy(mblk_t *mp, conn_t *connp,
2457     ipha_t *ipha, ip6_t *ip6h, ip_recv_attr_t *ira)
2458 {
2459         boolean_t       ret;
2460         ipsec_latch_t   *ipl;
2461         ipsec_action_t  *ap;
2462         uint64_t        unique_id;
2463         ipsec_stack_t   *ipss;
2464         ip_stack_t      *ipst;
2465         netstack_t      *ns;
2466         ipsec_policy_head_t *policy_head;
2467         ipsec_policy_t  *p = NULL;
2468 
2469         ASSERT(connp != NULL);
2470         ns = connp->conn_netstack;
2471         ipss = ns->netstack_ipsec;
2472         ipst = ns->netstack_ip;
2473 
2474         if (!(ira->ira_flags & IRAF_IPSEC_SECURE)) {
2475                 /*
2476                  * This is the case where the incoming datagram is
2477                  * cleartext and we need to see whether this client
2478                  * would like to receive such untrustworthy things from
2479                  * the wire.
2480                  */
2481                 ASSERT(mp != NULL);
2482 
2483                 mutex_enter(&connp->conn_lock);
2484                 if (connp->conn_state_flags & CONN_CONDEMNED) {
2485                         mutex_exit(&connp->conn_lock);
2486                         ip_drop_packet(mp, B_TRUE, NULL,
2487                             DROPPER(ipss, ipds_spd_got_clear),
2488                             &ipss->ipsec_spd_dropper);
2489                         BUMP_MIB(&ipst->ips_ip_mib, ipsecInFailed);
2490                         return (NULL);
2491                 }
2492                 if (connp->conn_latch != NULL) {
2493                         /* Hold a reference in case the conn is closing */
2494                         p = connp->conn_latch_in_policy;
2495                         if (p != NULL)
2496                                 IPPOL_REFHOLD(p);
2497                         mutex_exit(&connp->conn_lock);
2498                         /*
2499                          * Policy is cached in the conn.
2500                          */
2501                         if (p != NULL && !p->ipsp_act->ipa_allow_clear) {
2502                                 ret = ipsec_inbound_accept_clear(mp,
2503                                     ipha, ip6h);
2504                                 if (ret) {
2505                                         BUMP_MIB(&ipst->ips_ip_mib,
2506                                             ipsecInSucceeded);
2507                                         IPPOL_REFRELE(p);
2508                                         return (mp);
2509                                 } else {
2510                                         ipsec_log_policy_failure(
2511                                             IPSEC_POLICY_MISMATCH,
2512                                             "ipsec_check_inbound_policy", ipha,
2513                                             ip6h, B_FALSE, ns);
2514                                         ip_drop_packet(mp, B_TRUE, NULL,
2515                                             DROPPER(ipss, ipds_spd_got_clear),
2516                                             &ipss->ipsec_spd_dropper);
2517                                         BUMP_MIB(&ipst->ips_ip_mib,
2518                                             ipsecInFailed);
2519                                         IPPOL_REFRELE(p);
2520                                         return (NULL);
2521                                 }
2522                         } else {
2523                                 BUMP_MIB(&ipst->ips_ip_mib, ipsecInSucceeded);
2524                                 if (p != NULL)
2525                                         IPPOL_REFRELE(p);
2526                                 return (mp);
2527                         }
2528                 } else {
2529                         policy_head = connp->conn_policy;
2530 
2531                         /* Hold a reference in case the conn is closing */
2532                         if (policy_head != NULL)
2533                                 IPPH_REFHOLD(policy_head);
2534                         mutex_exit(&connp->conn_lock);
2535                         /*
2536                          * As this is a non-hardbound connection we need
2537                          * to look at both per-socket policy and global
2538                          * policy.
2539                          */
2540                         mp = ipsec_check_global_policy(mp, connp,
2541                             ipha, ip6h, ira, ns);
2542                         if (policy_head != NULL)
2543                                 IPPH_REFRELE(policy_head, ns);
2544                         return (mp);
2545                 }
2546         }
2547 
2548         mutex_enter(&connp->conn_lock);
2549         /* Connection is closing */
2550         if (connp->conn_state_flags & CONN_CONDEMNED) {
2551                 mutex_exit(&connp->conn_lock);
2552                 ip_drop_packet(mp, B_TRUE, NULL,
2553                     DROPPER(ipss, ipds_spd_got_clear),
2554                     &ipss->ipsec_spd_dropper);
2555                 BUMP_MIB(&ipst->ips_ip_mib, ipsecInFailed);
2556                 return (NULL);
2557         }
2558 
2559         /*
2560          * Once a connection is latched it remains so for life, the conn_latch
2561          * pointer on the conn has not changed, simply initializing ipl here
2562          * as the earlier initialization was done only in the cleartext case.
2563          */
2564         if ((ipl = connp->conn_latch) == NULL) {
2565                 mblk_t *retmp;
2566                 policy_head = connp->conn_policy;
2567 
2568                 /* Hold a reference in case the conn is closing */
2569                 if (policy_head != NULL)
2570                         IPPH_REFHOLD(policy_head);
2571                 mutex_exit(&connp->conn_lock);
2572                 /*
2573                  * We don't have policies cached in the conn
2574                  * for this stream. So, look at the global
2575                  * policy. It will check against conn or global
2576                  * depending on whichever is stronger.
2577                  */
2578                 retmp = ipsec_check_global_policy(mp, connp,
2579                     ipha, ip6h, ira, ns);
2580                 if (policy_head != NULL)
2581                         IPPH_REFRELE(policy_head, ns);
2582                 return (retmp);
2583         }
2584 
2585         IPLATCH_REFHOLD(ipl);
2586         /* Hold reference on conn_latch_in_action in case conn is closing */
2587         ap = connp->conn_latch_in_action;
2588         if (ap != NULL)
2589                 IPACT_REFHOLD(ap);
2590         mutex_exit(&connp->conn_lock);
2591 
2592         if (ap != NULL) {
2593                 /* Policy is cached & latched; fast(er) path */
2594                 const char *reason;
2595                 kstat_named_t *counter;
2596 
2597                 if (ipsec_check_ipsecin_latch(ira, mp, ipl, ap,
2598                     ipha, ip6h, &reason, &counter, connp, ns)) {
2599                         BUMP_MIB(&ipst->ips_ip_mib, ipsecInSucceeded);
2600                         IPLATCH_REFRELE(ipl);
2601                         IPACT_REFRELE(ap);
2602                         return (mp);
2603                 }
2604                 ipsec_rl_strlog(ns, IP_MOD_ID, 0, 0,
2605                     SL_ERROR|SL_WARN|SL_CONSOLE,
2606                     "ipsec inbound policy mismatch: %s, packet dropped\n",
2607                     reason);
2608                 ip_drop_packet(mp, B_TRUE, NULL, counter,
2609                     &ipss->ipsec_spd_dropper);
2610                 BUMP_MIB(&ipst->ips_ip_mib, ipsecInFailed);
2611                 IPLATCH_REFRELE(ipl);
2612                 IPACT_REFRELE(ap);
2613                 return (NULL);
2614         }
2615         if ((p = connp->conn_latch_in_policy) == NULL) {
2616                 ipsec_weird_null_inbound_policy++;
2617                 IPLATCH_REFRELE(ipl);
2618                 return (mp);
2619         }
2620 
2621         unique_id = conn_to_unique(connp, mp, ipha, ip6h);
2622         IPPOL_REFHOLD(p);
2623         mp = ipsec_check_ipsecin_policy(mp, p, ipha, ip6h, unique_id, ira, ns);
2624         /*
2625          * NOTE: ipsecIn{Failed,Succeeeded} bumped by
2626          * ipsec_check_ipsecin_policy().
2627          */
2628         if (mp != NULL)
2629                 ipsec_latch_inbound(connp, ira);
2630         IPLATCH_REFRELE(ipl);
2631         return (mp);
2632 }
2633 
2634 /*
2635  * Handle all sorts of cases like tunnel-mode and ICMP.
2636  */
2637 static int
2638 prepended_length(mblk_t *mp, uintptr_t hptr)
2639 {
2640         int rc = 0;
2641 
2642         while (mp != NULL) {
2643                 if (hptr >= (uintptr_t)mp->b_rptr && hptr <
2644                     (uintptr_t)mp->b_wptr) {
2645                         rc += (int)(hptr - (uintptr_t)mp->b_rptr);
2646                         break;  /* out of while loop */
2647                 }
2648                 rc += (int)MBLKL(mp);
2649                 mp = mp->b_cont;
2650         }
2651 
2652         if (mp == NULL) {
2653                 /*
2654                  * IF (big IF) we make it here by naturally exiting the loop,
2655                  * then ip6h isn't in the mblk chain "mp" at all.
2656                  *
2657                  * The only case where this happens is with a reversed IP
2658                  * header that gets passed up by inbound ICMP processing.
2659                  * This unfortunately triggers longstanding bug 6478464.  For
2660                  * now, just pass up 0 for the answer.
2661                  */
2662 #ifdef DEBUG_NOT_UNTIL_6478464
2663                 ASSERT(mp != NULL);
2664 #endif
2665                 rc = 0;
2666         }
2667 
2668         return (rc);
2669 }
2670 
2671 /*
2672  * Returns:
2673  *
2674  * SELRET_NOMEM --> msgpullup() needed to gather things failed.
2675  * SELRET_BADPKT --> If we're being called after tunnel-mode fragment
2676  *                   gathering, the initial fragment is too short for
2677  *                   useful data.  Only returned if SEL_TUNNEL_FIRSTFRAG is
2678  *                   set.
2679  * SELRET_SUCCESS --> "sel" now has initialized IPsec selector data.
2680  * SELRET_TUNFRAG --> This is a fragment in a tunnel-mode packet.  Caller
2681  *                    should put this packet in a fragment-gathering queue.
2682  *                    Only returned if SEL_TUNNEL_MODE and SEL_PORT_POLICY
2683  *                    is set.
2684  *
2685  * Note that ipha/ip6h can be in a different mblk (mp->b_cont) in the case
2686  * of tunneled packets.
2687  * Also, mp->b_rptr can be an ICMP error where ipha/ip6h is the packet in
2688  * error past the ICMP error.
2689  */
2690 static selret_t
2691 ipsec_init_inbound_sel(ipsec_selector_t *sel, mblk_t *mp, ipha_t *ipha,
2692     ip6_t *ip6h, uint8_t sel_flags)
2693 {
2694         uint16_t *ports;
2695         int outer_hdr_len = 0;  /* For ICMP or tunnel-mode cases... */
2696         ushort_t hdr_len;
2697         mblk_t *spare_mp = NULL;
2698         uint8_t *nexthdrp, *transportp;
2699         uint8_t nexthdr;
2700         uint8_t icmp_proto;
2701         ip_pkt_t ipp;
2702         boolean_t port_policy_present = (sel_flags & SEL_PORT_POLICY);
2703         boolean_t is_icmp = (sel_flags & SEL_IS_ICMP);
2704         boolean_t tunnel_mode = (sel_flags & SEL_TUNNEL_MODE);
2705         boolean_t post_frag = (sel_flags & SEL_POST_FRAG);
2706 
2707         ASSERT((ipha == NULL && ip6h != NULL) ||
2708             (ipha != NULL && ip6h == NULL));
2709 
2710         if (ip6h != NULL) {
2711                 outer_hdr_len = prepended_length(mp, (uintptr_t)ip6h);
2712                 nexthdr = ip6h->ip6_nxt;
2713                 icmp_proto = IPPROTO_ICMPV6;
2714                 sel->ips_isv4 = B_FALSE;
2715                 sel->ips_local_addr_v6 = ip6h->ip6_dst;
2716                 sel->ips_remote_addr_v6 = ip6h->ip6_src;
2717 
2718                 bzero(&ipp, sizeof (ipp));
2719 
2720                 switch (nexthdr) {
2721                 case IPPROTO_HOPOPTS:
2722                 case IPPROTO_ROUTING:
2723                 case IPPROTO_DSTOPTS:
2724                 case IPPROTO_FRAGMENT:
2725                         /*
2726                          * Use ip_hdr_length_nexthdr_v6().  And have a spare
2727                          * mblk that's contiguous to feed it
2728                          */
2729                         if ((spare_mp = msgpullup(mp, -1)) == NULL)
2730                                 return (SELRET_NOMEM);
2731                         if (!ip_hdr_length_nexthdr_v6(spare_mp,
2732                             (ip6_t *)(spare_mp->b_rptr + outer_hdr_len),
2733                             &hdr_len, &nexthdrp)) {
2734                                 /* Malformed packet - caller frees. */
2735                                 ipsec_freemsg_chain(spare_mp);
2736                                 return (SELRET_BADPKT);
2737                         }
2738                         /* Repopulate now that we have the whole packet */
2739                         ip6h = (ip6_t *)(spare_mp->b_rptr + outer_hdr_len);
2740                         (void) ip_find_hdr_v6(spare_mp, ip6h, B_FALSE, &ipp,
2741                             NULL);
2742                         nexthdr = *nexthdrp;
2743                         /* We can just extract based on hdr_len now. */
2744                         break;
2745                 default:
2746                         (void) ip_find_hdr_v6(mp, ip6h, B_FALSE, &ipp, NULL);
2747                         hdr_len = IPV6_HDR_LEN;
2748                         break;
2749                 }
2750                 if (port_policy_present && IS_V6_FRAGMENT(ipp) && !is_icmp) {
2751                         /* IPv6 Fragment */
2752                         ipsec_freemsg_chain(spare_mp);
2753                         return (SELRET_TUNFRAG);
2754                 }
2755                 transportp = (uint8_t *)ip6h + hdr_len;
2756         } else {
2757                 outer_hdr_len = prepended_length(mp, (uintptr_t)ipha);
2758                 icmp_proto = IPPROTO_ICMP;
2759                 sel->ips_isv4 = B_TRUE;
2760                 sel->ips_local_addr_v4 = ipha->ipha_dst;
2761                 sel->ips_remote_addr_v4 = ipha->ipha_src;
2762                 nexthdr = ipha->ipha_protocol;
2763                 hdr_len = IPH_HDR_LENGTH(ipha);
2764 
2765                 if (port_policy_present &&
2766                     IS_V4_FRAGMENT(ipha->ipha_fragment_offset_and_flags) &&
2767                     !is_icmp) {
2768                         /* IPv4 Fragment */
2769                         ipsec_freemsg_chain(spare_mp);
2770                         return (SELRET_TUNFRAG);
2771                 }
2772                 transportp = (uint8_t *)ipha + hdr_len;
2773         }
2774         sel->ips_protocol = nexthdr;
2775 
2776         if ((nexthdr != IPPROTO_TCP && nexthdr != IPPROTO_UDP &&
2777             nexthdr != IPPROTO_SCTP && nexthdr != icmp_proto) ||
2778             (!port_policy_present && !post_frag && tunnel_mode)) {
2779                 sel->ips_remote_port = sel->ips_local_port = 0;
2780                 ipsec_freemsg_chain(spare_mp);
2781                 return (SELRET_SUCCESS);
2782         }
2783 
2784         if (transportp + 4 > mp->b_wptr) {
2785                 /* If we didn't pullup a copy already, do so now. */
2786                 /*
2787                  * XXX performance, will upper-layers frequently split TCP/UDP
2788                  * apart from IP or options?  If so, perhaps we should revisit
2789                  * the spare_mp strategy.
2790                  */
2791                 ipsec_hdr_pullup_needed++;
2792                 if (spare_mp == NULL &&
2793                     (spare_mp = msgpullup(mp, -1)) == NULL) {
2794                         return (SELRET_NOMEM);
2795                 }
2796                 transportp = &spare_mp->b_rptr[hdr_len + outer_hdr_len];
2797         }
2798 
2799         if (nexthdr == icmp_proto) {
2800                 sel->ips_icmp_type = *transportp++;
2801                 sel->ips_icmp_code = *transportp;
2802                 sel->ips_remote_port = sel->ips_local_port = 0;
2803         } else {
2804                 ports = (uint16_t *)transportp;
2805                 sel->ips_remote_port = *ports++;
2806                 sel->ips_local_port = *ports;
2807         }
2808         ipsec_freemsg_chain(spare_mp);
2809         return (SELRET_SUCCESS);
2810 }
2811 
2812 /*
2813  * This is called with a b_next chain of messages from the fragcache code,
2814  * hence it needs to discard a chain on error.
2815  */
2816 static boolean_t
2817 ipsec_init_outbound_ports(ipsec_selector_t *sel, mblk_t *mp, ipha_t *ipha,
2818     ip6_t *ip6h, int outer_hdr_len, ipsec_stack_t *ipss)
2819 {
2820         /*
2821          * XXX cut&paste shared with ipsec_init_inbound_sel
2822          */
2823         uint16_t *ports;
2824         ushort_t hdr_len;
2825         mblk_t *spare_mp = NULL;
2826         uint8_t *nexthdrp;
2827         uint8_t nexthdr;
2828         uint8_t *typecode;
2829         uint8_t check_proto;
2830 
2831         ASSERT((ipha == NULL && ip6h != NULL) ||
2832             (ipha != NULL && ip6h == NULL));
2833 
2834         if (ip6h != NULL) {
2835                 check_proto = IPPROTO_ICMPV6;
2836                 nexthdr = ip6h->ip6_nxt;
2837                 switch (nexthdr) {
2838                 case IPPROTO_HOPOPTS:
2839                 case IPPROTO_ROUTING:
2840                 case IPPROTO_DSTOPTS:
2841                 case IPPROTO_FRAGMENT:
2842                         /*
2843                          * Use ip_hdr_length_nexthdr_v6().  And have a spare
2844                          * mblk that's contiguous to feed it
2845                          */
2846                         spare_mp = msgpullup(mp, -1);
2847                         if (spare_mp == NULL ||
2848                             !ip_hdr_length_nexthdr_v6(spare_mp,
2849                             (ip6_t *)(spare_mp->b_rptr + outer_hdr_len),
2850                             &hdr_len, &nexthdrp)) {
2851                                 /* Always works, even if NULL. */
2852                                 ipsec_freemsg_chain(spare_mp);
2853                                 ip_drop_packet_chain(mp, B_FALSE, NULL,
2854                                     DROPPER(ipss, ipds_spd_nomem),
2855                                     &ipss->ipsec_spd_dropper);
2856                                 return (B_FALSE);
2857                         } else {
2858                                 nexthdr = *nexthdrp;
2859                                 /* We can just extract based on hdr_len now. */
2860                         }
2861                         break;
2862                 default:
2863                         hdr_len = IPV6_HDR_LEN;
2864                         break;
2865                 }
2866         } else {
2867                 check_proto = IPPROTO_ICMP;
2868                 hdr_len = IPH_HDR_LENGTH(ipha);
2869                 nexthdr = ipha->ipha_protocol;
2870         }
2871 
2872         sel->ips_protocol = nexthdr;
2873         if (nexthdr != IPPROTO_TCP && nexthdr != IPPROTO_UDP &&
2874             nexthdr != IPPROTO_SCTP && nexthdr != check_proto) {
2875                 sel->ips_local_port = sel->ips_remote_port = 0;
2876                 ipsec_freemsg_chain(spare_mp); /* Always works, even if NULL */
2877                 return (B_TRUE);
2878         }
2879 
2880         if (&mp->b_rptr[hdr_len] + 4 + outer_hdr_len > mp->b_wptr) {
2881                 /* If we didn't pullup a copy already, do so now. */
2882                 /*
2883                  * XXX performance, will upper-layers frequently split TCP/UDP
2884                  * apart from IP or options?  If so, perhaps we should revisit
2885                  * the spare_mp strategy.
2886                  *
2887                  * XXX should this be msgpullup(mp, hdr_len+4) ???
2888                  */
2889                 if (spare_mp == NULL &&
2890                     (spare_mp = msgpullup(mp, -1)) == NULL) {
2891                         ip_drop_packet_chain(mp, B_FALSE, NULL,
2892                             DROPPER(ipss, ipds_spd_nomem),
2893                             &ipss->ipsec_spd_dropper);
2894                         return (B_FALSE);
2895                 }
2896                 ports = (uint16_t *)&spare_mp->b_rptr[hdr_len + outer_hdr_len];
2897         } else {
2898                 ports = (uint16_t *)&mp->b_rptr[hdr_len + outer_hdr_len];
2899         }
2900 
2901         if (nexthdr == check_proto) {
2902                 typecode = (uint8_t *)ports;
2903                 sel->ips_icmp_type = *typecode++;
2904                 sel->ips_icmp_code = *typecode;
2905                 sel->ips_remote_port = sel->ips_local_port = 0;
2906         } else {
2907                 sel->ips_local_port = *ports++;
2908                 sel->ips_remote_port = *ports;
2909         }
2910         ipsec_freemsg_chain(spare_mp);  /* Always works, even if NULL */
2911         return (B_TRUE);
2912 }
2913 
2914 /*
2915  * Prepend an mblk with a ipsec_crypto_t to the message chain.
2916  * Frees the argument and returns NULL should the allocation fail.
2917  * Returns the pointer to the crypto data part.
2918  */
2919 mblk_t *
2920 ipsec_add_crypto_data(mblk_t *data_mp, ipsec_crypto_t **icp)
2921 {
2922         mblk_t  *mp;
2923 
2924         mp = allocb(sizeof (ipsec_crypto_t), BPRI_MED);
2925         if (mp == NULL) {
2926                 freemsg(data_mp);
2927                 return (NULL);
2928         }
2929         bzero(mp->b_rptr, sizeof (ipsec_crypto_t));
2930         mp->b_wptr += sizeof (ipsec_crypto_t);
2931         mp->b_cont = data_mp;
2932         mp->b_datap->db_type = M_EVENT;   /* For ASSERT */
2933         *icp = (ipsec_crypto_t *)mp->b_rptr;
2934         return (mp);
2935 }
2936 
2937 /*
2938  * Remove what was prepended above. Return b_cont and a pointer to the
2939  * crypto data.
2940  * The caller must call ipsec_free_crypto_data for mblk once it is done
2941  * with the crypto data.
2942  */
2943 mblk_t *
2944 ipsec_remove_crypto_data(mblk_t *crypto_mp, ipsec_crypto_t **icp)
2945 {
2946         ASSERT(crypto_mp->b_datap->db_type == M_EVENT);
2947         ASSERT(MBLKL(crypto_mp) == sizeof (ipsec_crypto_t));
2948 
2949         *icp = (ipsec_crypto_t *)crypto_mp->b_rptr;
2950         return (crypto_mp->b_cont);
2951 }
2952 
2953 /*
2954  * Free what was prepended above. Return b_cont.
2955  */
2956 mblk_t *
2957 ipsec_free_crypto_data(mblk_t *crypto_mp)
2958 {
2959         mblk_t  *mp;
2960 
2961         ASSERT(crypto_mp->b_datap->db_type == M_EVENT);
2962         ASSERT(MBLKL(crypto_mp) == sizeof (ipsec_crypto_t));
2963 
2964         mp = crypto_mp->b_cont;
2965         freeb(crypto_mp);
2966         return (mp);
2967 }
2968 
2969 /*
2970  * Create an ipsec_action_t based on the way an inbound packet was protected.
2971  * Used to reflect traffic back to a sender.
2972  *
2973  * We don't bother interning the action into the hash table.
2974  */
2975 ipsec_action_t *
2976 ipsec_in_to_out_action(ip_recv_attr_t *ira)
2977 {
2978         ipsa_t *ah_assoc, *esp_assoc;
2979         uint_t auth_alg = 0, encr_alg = 0, espa_alg = 0;
2980         ipsec_action_t *ap;
2981         boolean_t unique;
2982 
2983         ap = kmem_cache_alloc(ipsec_action_cache, KM_NOSLEEP);
2984 
2985         if (ap == NULL)
2986                 return (NULL);
2987 
2988         bzero(ap, sizeof (*ap));
2989         HASH_NULL(ap, ipa_hash);
2990         ap->ipa_next = NULL;
2991         ap->ipa_refs = 1;
2992 
2993         /*
2994          * Get the algorithms that were used for this packet.
2995          */
2996         ap->ipa_act.ipa_type = IPSEC_ACT_APPLY;
2997         ap->ipa_act.ipa_log = 0;
2998         ASSERT(ira->ira_flags & IRAF_IPSEC_SECURE);
2999 
3000         ah_assoc = ira->ira_ipsec_ah_sa;
3001         ap->ipa_act.ipa_apply.ipp_use_ah = (ah_assoc != NULL);
3002 
3003         esp_assoc = ira->ira_ipsec_esp_sa;
3004         ap->ipa_act.ipa_apply.ipp_use_esp = (esp_assoc != NULL);
3005 
3006         if (esp_assoc != NULL) {
3007                 encr_alg = esp_assoc->ipsa_encr_alg;
3008                 espa_alg = esp_assoc->ipsa_auth_alg;
3009                 ap->ipa_act.ipa_apply.ipp_use_espa = (espa_alg != 0);
3010         }
3011         if (ah_assoc != NULL)
3012                 auth_alg = ah_assoc->ipsa_auth_alg;
3013 
3014         ap->ipa_act.ipa_apply.ipp_encr_alg = (uint8_t)encr_alg;
3015         ap->ipa_act.ipa_apply.ipp_auth_alg = (uint8_t)auth_alg;
3016         ap->ipa_act.ipa_apply.ipp_esp_auth_alg = (uint8_t)espa_alg;
3017         ap->ipa_act.ipa_apply.ipp_use_se =
3018             !!(ira->ira_flags & IRAF_IPSEC_DECAPS);
3019         unique = B_FALSE;
3020 
3021         if (esp_assoc != NULL) {
3022                 ap->ipa_act.ipa_apply.ipp_espa_minbits =
3023                     esp_assoc->ipsa_authkeybits;
3024                 ap->ipa_act.ipa_apply.ipp_espa_maxbits =
3025                     esp_assoc->ipsa_authkeybits;
3026                 ap->ipa_act.ipa_apply.ipp_espe_minbits =
3027                     esp_assoc->ipsa_encrkeybits;
3028                 ap->ipa_act.ipa_apply.ipp_espe_maxbits =
3029                     esp_assoc->ipsa_encrkeybits;
3030                 ap->ipa_act.ipa_apply.ipp_km_proto = esp_assoc->ipsa_kmp;
3031                 ap->ipa_act.ipa_apply.ipp_km_cookie = esp_assoc->ipsa_kmc;
3032                 if (esp_assoc->ipsa_flags & IPSA_F_UNIQUE)
3033                         unique = B_TRUE;
3034         }
3035         if (ah_assoc != NULL) {
3036                 ap->ipa_act.ipa_apply.ipp_ah_minbits =
3037                     ah_assoc->ipsa_authkeybits;
3038                 ap->ipa_act.ipa_apply.ipp_ah_maxbits =
3039                     ah_assoc->ipsa_authkeybits;
3040                 ap->ipa_act.ipa_apply.ipp_km_proto = ah_assoc->ipsa_kmp;
3041                 ap->ipa_act.ipa_apply.ipp_km_cookie = ah_assoc->ipsa_kmc;
3042                 if (ah_assoc->ipsa_flags & IPSA_F_UNIQUE)
3043                         unique = B_TRUE;
3044         }
3045         ap->ipa_act.ipa_apply.ipp_use_unique = unique;
3046         ap->ipa_want_unique = unique;
3047         ap->ipa_allow_clear = B_FALSE;
3048         ap->ipa_want_se = !!(ira->ira_flags & IRAF_IPSEC_DECAPS);
3049         ap->ipa_want_ah = (ah_assoc != NULL);
3050         ap->ipa_want_esp = (esp_assoc != NULL);
3051 
3052         ap->ipa_ovhd = ipsec_act_ovhd(&ap->ipa_act);
3053 
3054         ap->ipa_act.ipa_apply.ipp_replay_depth = 0; /* don't care */
3055 
3056         return (ap);
3057 }
3058 
3059 
3060 /*
3061  * Compute the worst-case amount of extra space required by an action.
3062  * Note that, because of the ESP considerations listed below, this is
3063  * actually not the same as the best-case reduction in the MTU; in the
3064  * future, we should pass additional information to this function to
3065  * allow the actual MTU impact to be computed.
3066  *
3067  * AH: Revisit this if we implement algorithms with
3068  * a verifier size of more than 12 bytes.
3069  *
3070  * ESP: A more exact but more messy computation would take into
3071  * account the interaction between the cipher block size and the
3072  * effective MTU, yielding the inner payload size which reflects a
3073  * packet with *minimum* ESP padding..
3074  */
3075 int32_t
3076 ipsec_act_ovhd(const ipsec_act_t *act)
3077 {
3078         int32_t overhead = 0;
3079 
3080         if (act->ipa_type == IPSEC_ACT_APPLY) {
3081                 const ipsec_prot_t *ipp = &act->ipa_apply;
3082 
3083                 if (ipp->ipp_use_ah)
3084                         overhead += IPSEC_MAX_AH_HDR_SIZE;
3085                 if (ipp->ipp_use_esp) {
3086                         overhead += IPSEC_MAX_ESP_HDR_SIZE;
3087                         overhead += sizeof (struct udphdr);
3088                 }
3089                 if (ipp->ipp_use_se)
3090                         overhead += IP_SIMPLE_HDR_LENGTH;
3091         }
3092         return (overhead);
3093 }
3094 
3095 /*
3096  * This hash function is used only when creating policies and thus is not
3097  * performance-critical for packet flows.
3098  *
3099  * Future work: canonicalize the structures hashed with this (i.e.,
3100  * zeroize padding) so the hash works correctly.
3101  */
3102 /* ARGSUSED */
3103 static uint32_t
3104 policy_hash(int size, const void *start, const void *end)
3105 {
3106         return (0);
3107 }
3108 
3109 
3110 /*
3111  * Hash function macros for each address type.
3112  *
3113  * The IPV6 hash function assumes that the low order 32-bits of the
3114  * address (typically containing the low order 24 bits of the mac
3115  * address) are reasonably well-distributed.  Revisit this if we run
3116  * into trouble from lots of collisions on ::1 addresses and the like
3117  * (seems unlikely).
3118  */
3119 #define IPSEC_IPV4_HASH(a, n) ((a) % (n))
3120 #define IPSEC_IPV6_HASH(a, n) (((a).s6_addr32[3]) % (n))
3121 
3122 /*
3123  * These two hash functions should produce coordinated values
3124  * but have slightly different roles.
3125  */
3126 static uint32_t
3127 selkey_hash(const ipsec_selkey_t *selkey, netstack_t *ns)
3128 {
3129         uint32_t valid = selkey->ipsl_valid;
3130         ipsec_stack_t   *ipss = ns->netstack_ipsec;
3131 
3132         if (!(valid & IPSL_REMOTE_ADDR))
3133                 return (IPSEC_SEL_NOHASH);
3134 
3135         if (valid & IPSL_IPV4) {
3136                 if (selkey->ipsl_remote_pfxlen == 32) {
3137                         return (IPSEC_IPV4_HASH(selkey->ipsl_remote.ipsad_v4,
3138                             ipss->ipsec_spd_hashsize));
3139                 }
3140         }
3141         if (valid & IPSL_IPV6) {
3142                 if (selkey->ipsl_remote_pfxlen == 128) {
3143                         return (IPSEC_IPV6_HASH(selkey->ipsl_remote.ipsad_v6,
3144                             ipss->ipsec_spd_hashsize));
3145                 }
3146         }
3147         return (IPSEC_SEL_NOHASH);
3148 }
3149 
3150 static uint32_t
3151 selector_hash(ipsec_selector_t *sel, ipsec_policy_root_t *root)
3152 {
3153         if (sel->ips_isv4) {
3154                 return (IPSEC_IPV4_HASH(sel->ips_remote_addr_v4,
3155                     root->ipr_nchains));
3156         }
3157         return (IPSEC_IPV6_HASH(sel->ips_remote_addr_v6, root->ipr_nchains));
3158 }
3159 
3160 /*
3161  * Intern actions into the action hash table.
3162  */
3163 ipsec_action_t *
3164 ipsec_act_find(const ipsec_act_t *a, int n, netstack_t *ns)
3165 {
3166         int i;
3167         uint32_t hval;
3168         ipsec_action_t *ap;
3169         ipsec_action_t *prev = NULL;
3170         int32_t overhead, maxovhd = 0;
3171         boolean_t allow_clear = B_FALSE;
3172         boolean_t want_ah = B_FALSE;
3173         boolean_t want_esp = B_FALSE;
3174         boolean_t want_se = B_FALSE;
3175         boolean_t want_unique = B_FALSE;
3176         ipsec_stack_t   *ipss = ns->netstack_ipsec;
3177 
3178         /*
3179          * TODO: should canonicalize a[] (i.e., zeroize any padding)
3180          * so we can use a non-trivial policy_hash function.
3181          */
3182         for (i = n-1; i >= 0; i--) {
3183                 hval = policy_hash(IPSEC_ACTION_HASH_SIZE, &a[i], &a[n]);
3184 
3185                 HASH_LOCK(ipss->ipsec_action_hash, hval);
3186 
3187                 for (HASH_ITERATE(ap, ipa_hash,
3188                     ipss->ipsec_action_hash, hval)) {
3189                         if (bcmp(&ap->ipa_act, &a[i], sizeof (*a)) != 0)
3190                                 continue;
3191                         if (ap->ipa_next != prev)
3192                                 continue;
3193                         break;
3194                 }
3195                 if (ap != NULL) {
3196                         HASH_UNLOCK(ipss->ipsec_action_hash, hval);
3197                         prev = ap;
3198                         continue;
3199                 }
3200                 /*
3201                  * need to allocate a new one..
3202                  */
3203                 ap = kmem_cache_alloc(ipsec_action_cache, KM_NOSLEEP);
3204                 if (ap == NULL) {
3205                         HASH_UNLOCK(ipss->ipsec_action_hash, hval);
3206                         if (prev != NULL)
3207                                 ipsec_action_free(prev);
3208                         return (NULL);
3209                 }
3210                 HASH_INSERT(ap, ipa_hash, ipss->ipsec_action_hash, hval);
3211 
3212                 ap->ipa_next = prev;
3213                 ap->ipa_act = a[i];
3214 
3215                 overhead = ipsec_act_ovhd(&a[i]);
3216                 if (maxovhd < overhead)
3217                         maxovhd = overhead;
3218 
3219                 if ((a[i].ipa_type == IPSEC_ACT_BYPASS) ||
3220                     (a[i].ipa_type == IPSEC_ACT_CLEAR))
3221                         allow_clear = B_TRUE;
3222                 if (a[i].ipa_type == IPSEC_ACT_APPLY) {
3223                         const ipsec_prot_t *ipp = &a[i].ipa_apply;
3224 
3225                         ASSERT(ipp->ipp_use_ah || ipp->ipp_use_esp);
3226                         want_ah |= ipp->ipp_use_ah;
3227                         want_esp |= ipp->ipp_use_esp;
3228                         want_se |= ipp->ipp_use_se;
3229                         want_unique |= ipp->ipp_use_unique;
3230                 }
3231                 ap->ipa_allow_clear = allow_clear;
3232                 ap->ipa_want_ah = want_ah;
3233                 ap->ipa_want_esp = want_esp;
3234                 ap->ipa_want_se = want_se;
3235                 ap->ipa_want_unique = want_unique;
3236                 ap->ipa_refs = 1; /* from the hash table */
3237                 ap->ipa_ovhd = maxovhd;
3238                 if (prev)
3239                         prev->ipa_refs++;
3240                 prev = ap;
3241                 HASH_UNLOCK(ipss->ipsec_action_hash, hval);
3242         }
3243 
3244         ap->ipa_refs++;              /* caller's reference */
3245 
3246         return (ap);
3247 }
3248 
3249 /*
3250  * Called when refcount goes to 0, indicating that all references to this
3251  * node are gone.
3252  *
3253  * This does not unchain the action from the hash table.
3254  */
3255 void
3256 ipsec_action_free(ipsec_action_t *ap)
3257 {
3258         for (;;) {
3259                 ipsec_action_t *np = ap->ipa_next;
3260                 ASSERT(ap->ipa_refs == 0);
3261                 ASSERT(ap->ipa_hash.hash_pp == NULL);
3262                 kmem_cache_free(ipsec_action_cache, ap);
3263                 ap = np;
3264                 /* Inlined IPACT_REFRELE -- avoid recursion */
3265                 if (ap == NULL)
3266                         break;
3267                 membar_exit();
3268                 if (atomic_dec_32_nv(&(ap)->ipa_refs) != 0)
3269                         break;
3270                 /* End inlined IPACT_REFRELE */
3271         }
3272 }
3273 
3274 /*
3275  * Called when the action hash table goes away.
3276  *
3277  * The actions can be queued on an mblk with ipsec_in or
3278  * ipsec_out, hence the actions might still be around.
3279  * But we decrement ipa_refs here since we no longer have
3280  * a reference to the action from the hash table.
3281  */
3282 static void
3283 ipsec_action_free_table(ipsec_action_t *ap)
3284 {
3285         while (ap != NULL) {
3286                 ipsec_action_t *np = ap->ipa_next;
3287 
3288                 /* FIXME: remove? */
3289                 (void) printf("ipsec_action_free_table(%p) ref %d\n",
3290                     (void *)ap, ap->ipa_refs);
3291                 ASSERT(ap->ipa_refs > 0);
3292                 IPACT_REFRELE(ap);
3293                 ap = np;
3294         }
3295 }
3296 
3297 /*
3298  * Need to walk all stack instances since the reclaim function
3299  * is global for all instances
3300  */
3301 /* ARGSUSED */
3302 static void
3303 ipsec_action_reclaim(void *arg)
3304 {
3305         netstack_handle_t nh;
3306         netstack_t *ns;
3307         ipsec_stack_t *ipss;
3308 
3309         netstack_next_init(&nh);
3310         while ((ns = netstack_next(&nh)) != NULL) {
3311                 /*
3312                  * netstack_next() can return a netstack_t with a NULL
3313                  * netstack_ipsec at boot time.
3314                  */
3315                 if ((ipss = ns->netstack_ipsec) == NULL) {
3316                         netstack_rele(ns);
3317                         continue;
3318                 }
3319                 ipsec_action_reclaim_stack(ipss);
3320                 netstack_rele(ns);
3321         }
3322         netstack_next_fini(&nh);
3323 }
3324 
3325 /*
3326  * Periodically sweep action hash table for actions with refcount==1, and
3327  * nuke them.  We cannot do this "on demand" (i.e., from IPACT_REFRELE)
3328  * because we can't close the race between another thread finding the action
3329  * in the hash table without holding the bucket lock during IPACT_REFRELE.
3330  * Instead, we run this function sporadically to clean up after ourselves;
3331  * we also set it as the "reclaim" function for the action kmem_cache.
3332  *
3333  * Note that it may take several passes of ipsec_action_gc() to free all
3334  * "stale" actions.
3335  */
3336 static void
3337 ipsec_action_reclaim_stack(ipsec_stack_t *ipss)
3338 {
3339         int i;
3340 
3341         for (i = 0; i < IPSEC_ACTION_HASH_SIZE; i++) {
3342                 ipsec_action_t *ap, *np;
3343 
3344                 /* skip the lock if nobody home */
3345                 if (ipss->ipsec_action_hash[i].hash_head == NULL)
3346                         continue;
3347 
3348                 HASH_LOCK(ipss->ipsec_action_hash, i);
3349                 for (ap = ipss->ipsec_action_hash[i].hash_head;
3350                     ap != NULL; ap = np) {
3351                         ASSERT(ap->ipa_refs > 0);
3352                         np = ap->ipa_hash.hash_next;
3353                         if (ap->ipa_refs > 1)
3354                                 continue;
3355                         HASH_UNCHAIN(ap, ipa_hash,
3356                             ipss->ipsec_action_hash, i);
3357                         IPACT_REFRELE(ap);
3358                 }
3359                 HASH_UNLOCK(ipss->ipsec_action_hash, i);
3360         }
3361 }
3362 
3363 /*
3364  * Intern a selector set into the selector set hash table.
3365  * This is simpler than the actions case..
3366  */
3367 static ipsec_sel_t *
3368 ipsec_find_sel(ipsec_selkey_t *selkey, netstack_t *ns)
3369 {
3370         ipsec_sel_t *sp;
3371         uint32_t hval, bucket;
3372         ipsec_stack_t   *ipss = ns->netstack_ipsec;
3373 
3374         /*
3375          * Exactly one AF bit should be set in selkey.
3376          */
3377         ASSERT(!(selkey->ipsl_valid & IPSL_IPV4) ^
3378             !(selkey->ipsl_valid & IPSL_IPV6));
3379 
3380         hval = selkey_hash(selkey, ns);
3381         /* Set pol_hval to uninitialized until we put it in a polhead. */
3382         selkey->ipsl_sel_hval = hval;
3383 
3384         bucket = (hval == IPSEC_SEL_NOHASH) ? 0 : hval;
3385 
3386         ASSERT(!HASH_LOCKED(ipss->ipsec_sel_hash, bucket));
3387         HASH_LOCK(ipss->ipsec_sel_hash, bucket);
3388 
3389         for (HASH_ITERATE(sp, ipsl_hash, ipss->ipsec_sel_hash, bucket)) {
3390                 if (bcmp(&sp->ipsl_key, selkey,
3391                     offsetof(ipsec_selkey_t, ipsl_pol_hval)) == 0)
3392                         break;
3393         }
3394         if (sp != NULL) {
3395                 sp->ipsl_refs++;
3396 
3397                 HASH_UNLOCK(ipss->ipsec_sel_hash, bucket);
3398                 return (sp);
3399         }
3400 
3401         sp = kmem_cache_alloc(ipsec_sel_cache, KM_NOSLEEP);
3402         if (sp == NULL) {
3403                 HASH_UNLOCK(ipss->ipsec_sel_hash, bucket);
3404                 return (NULL);
3405         }
3406 
3407         HASH_INSERT(sp, ipsl_hash, ipss->ipsec_sel_hash, bucket);
3408         sp->ipsl_refs = 2;   /* one for hash table, one for caller */
3409         sp->ipsl_key = *selkey;
3410         /* Set to uninitalized and have insertion into polhead fix things. */
3411         if (selkey->ipsl_sel_hval != IPSEC_SEL_NOHASH)
3412                 sp->ipsl_key.ipsl_pol_hval = 0;
3413         else
3414                 sp->ipsl_key.ipsl_pol_hval = IPSEC_SEL_NOHASH;
3415 
3416         HASH_UNLOCK(ipss->ipsec_sel_hash, bucket);
3417 
3418         return (sp);
3419 }
3420 
3421 static void
3422 ipsec_sel_rel(ipsec_sel_t **spp, netstack_t *ns)
3423 {
3424         ipsec_sel_t *sp = *spp;
3425         int hval = sp->ipsl_key.ipsl_sel_hval;
3426         ipsec_stack_t   *ipss = ns->netstack_ipsec;
3427 
3428         *spp = NULL;
3429 
3430         if (hval == IPSEC_SEL_NOHASH)
3431                 hval = 0;
3432 
3433         ASSERT(!HASH_LOCKED(ipss->ipsec_sel_hash, hval));
3434         HASH_LOCK(ipss->ipsec_sel_hash, hval);
3435         if (--sp->ipsl_refs == 1) {
3436                 HASH_UNCHAIN(sp, ipsl_hash, ipss->ipsec_sel_hash, hval);
3437                 sp->ipsl_refs--;
3438                 HASH_UNLOCK(ipss->ipsec_sel_hash, hval);
3439                 ASSERT(sp->ipsl_refs == 0);
3440                 kmem_cache_free(ipsec_sel_cache, sp);
3441                 /* Caller unlocks */
3442                 return;
3443         }
3444 
3445         HASH_UNLOCK(ipss->ipsec_sel_hash, hval);
3446 }
3447 
3448 /*
3449  * Free a policy rule which we know is no longer being referenced.
3450  */
3451 void
3452 ipsec_policy_free(ipsec_policy_t *ipp)
3453 {
3454         ASSERT(ipp->ipsp_refs == 0);
3455         ASSERT(ipp->ipsp_sel != NULL);
3456         ASSERT(ipp->ipsp_act != NULL);
3457         ASSERT(ipp->ipsp_netstack != NULL);
3458 
3459         ipsec_sel_rel(&ipp->ipsp_sel, ipp->ipsp_netstack);
3460         IPACT_REFRELE(ipp->ipsp_act);
3461         kmem_cache_free(ipsec_pol_cache, ipp);
3462 }
3463 
3464 /*
3465  * Construction of new policy rules; construct a policy, and add it to
3466  * the appropriate tables.
3467  */
3468 ipsec_policy_t *
3469 ipsec_policy_create(ipsec_selkey_t *keys, const ipsec_act_t *a,
3470     int nacts, int prio, uint64_t *index_ptr, netstack_t *ns)
3471 {
3472         ipsec_action_t *ap;
3473         ipsec_sel_t *sp;
3474         ipsec_policy_t *ipp;
3475         ipsec_stack_t   *ipss = ns->netstack_ipsec;
3476 
3477         if (index_ptr == NULL)
3478                 index_ptr = &ipss->ipsec_next_policy_index;
3479 
3480         ipp = kmem_cache_alloc(ipsec_pol_cache, KM_NOSLEEP);
3481         ap = ipsec_act_find(a, nacts, ns);
3482         sp = ipsec_find_sel(keys, ns);
3483 
3484         if ((ap == NULL) || (sp == NULL) || (ipp == NULL)) {
3485                 if (ap != NULL) {
3486                         IPACT_REFRELE(ap);
3487                 }
3488                 if (sp != NULL)
3489                         ipsec_sel_rel(&sp, ns);
3490                 if (ipp != NULL)
3491                         kmem_cache_free(ipsec_pol_cache, ipp);
3492                 return (NULL);
3493         }
3494 
3495         HASH_NULL(ipp, ipsp_hash);
3496 
3497         ipp->ipsp_netstack = ns;     /* Needed for ipsec_policy_free */
3498         ipp->ipsp_refs = 1;  /* caller's reference */
3499         ipp->ipsp_sel = sp;
3500         ipp->ipsp_act = ap;
3501         ipp->ipsp_prio = prio;       /* rule priority */
3502         ipp->ipsp_index = *index_ptr;
3503         (*index_ptr)++;
3504 
3505         return (ipp);
3506 }
3507 
3508 static void
3509 ipsec_update_present_flags(ipsec_stack_t *ipss)
3510 {
3511         boolean_t hashpol;
3512 
3513         hashpol = (avl_numnodes(&ipss->ipsec_system_policy.iph_rulebyid) > 0);
3514 
3515         if (hashpol) {
3516                 ipss->ipsec_outbound_v4_policy_present = B_TRUE;
3517                 ipss->ipsec_outbound_v6_policy_present = B_TRUE;
3518                 ipss->ipsec_inbound_v4_policy_present = B_TRUE;
3519                 ipss->ipsec_inbound_v6_policy_present = B_TRUE;
3520                 return;
3521         }
3522 
3523         ipss->ipsec_outbound_v4_policy_present = (NULL !=
3524             ipss->ipsec_system_policy.iph_root[IPSEC_TYPE_OUTBOUND].
3525             ipr_nonhash[IPSEC_AF_V4]);
3526         ipss->ipsec_outbound_v6_policy_present = (NULL !=
3527             ipss->ipsec_system_policy.iph_root[IPSEC_TYPE_OUTBOUND].
3528             ipr_nonhash[IPSEC_AF_V6]);
3529         ipss->ipsec_inbound_v4_policy_present = (NULL !=
3530             ipss->ipsec_system_policy.iph_root[IPSEC_TYPE_INBOUND].
3531             ipr_nonhash[IPSEC_AF_V4]);
3532         ipss->ipsec_inbound_v6_policy_present = (NULL !=
3533             ipss->ipsec_system_policy.iph_root[IPSEC_TYPE_INBOUND].
3534             ipr_nonhash[IPSEC_AF_V6]);
3535 }
3536 
3537 boolean_t
3538 ipsec_policy_delete(ipsec_policy_head_t *php, ipsec_selkey_t *keys, int dir,
3539         netstack_t *ns)
3540 {
3541         ipsec_sel_t *sp;
3542         ipsec_policy_t *ip, *nip, *head;
3543         int af;
3544         ipsec_policy_root_t *pr = &php->iph_root[dir];
3545 
3546         sp = ipsec_find_sel(keys, ns);
3547 
3548         if (sp == NULL)
3549                 return (B_FALSE);
3550 
3551         af = (sp->ipsl_key.ipsl_valid & IPSL_IPV4) ? IPSEC_AF_V4 : IPSEC_AF_V6;
3552 
3553         rw_enter(&php->iph_lock, RW_WRITER);
3554 
3555         if (sp->ipsl_key.ipsl_pol_hval == IPSEC_SEL_NOHASH) {
3556                 head = pr->ipr_nonhash[af];
3557         } else {
3558                 head = pr->ipr_hash[sp->ipsl_key.ipsl_pol_hval].hash_head;
3559         }
3560 
3561         for (ip = head; ip != NULL; ip = nip) {
3562                 nip = ip->ipsp_hash.hash_next;
3563                 if (ip->ipsp_sel != sp) {
3564                         continue;
3565                 }
3566 
3567                 IPPOL_UNCHAIN(php, ip);
3568 
3569                 php->iph_gen++;
3570                 ipsec_update_present_flags(ns->netstack_ipsec);
3571 
3572                 rw_exit(&php->iph_lock);
3573 
3574                 ipsec_sel_rel(&sp, ns);
3575 
3576                 return (B_TRUE);
3577         }
3578 
3579         rw_exit(&php->iph_lock);
3580         ipsec_sel_rel(&sp, ns);
3581         return (B_FALSE);
3582 }
3583 
3584 int
3585 ipsec_policy_delete_index(ipsec_policy_head_t *php, uint64_t policy_index,
3586     netstack_t *ns)
3587 {
3588         boolean_t found = B_FALSE;
3589         ipsec_policy_t ipkey;
3590         ipsec_policy_t *ip;
3591         avl_index_t where;
3592 
3593         bzero(&ipkey, sizeof (ipkey));
3594         ipkey.ipsp_index = policy_index;
3595 
3596         rw_enter(&php->iph_lock, RW_WRITER);
3597 
3598         /*
3599          * We could be cleverer here about the walk.
3600          * but well, (k+1)*log(N) will do for now (k==number of matches,
3601          * N==number of table entries
3602          */
3603         for (;;) {
3604                 ip = (ipsec_policy_t *)avl_find(&php->iph_rulebyid,
3605                     (void *)&ipkey, &where);
3606                 ASSERT(ip == NULL);
3607 
3608                 ip = avl_nearest(&php->iph_rulebyid, where, AVL_AFTER);
3609 
3610                 if (ip == NULL)
3611                         break;
3612 
3613                 if (ip->ipsp_index != policy_index) {
3614                         ASSERT(ip->ipsp_index > policy_index);
3615                         break;
3616                 }
3617 
3618                 IPPOL_UNCHAIN(php, ip);
3619                 found = B_TRUE;
3620         }
3621 
3622         if (found) {
3623                 php->iph_gen++;
3624                 ipsec_update_present_flags(ns->netstack_ipsec);
3625         }
3626 
3627         rw_exit(&php->iph_lock);
3628 
3629         return (found ? 0 : ENOENT);
3630 }
3631 
3632 /*
3633  * Given a constructed ipsec_policy_t policy rule, see if it can be entered
3634  * into the correct policy ruleset.  As a side-effect, it sets the hash
3635  * entries on "ipp"'s ipsp_pol_hval.
3636  *
3637  * Returns B_TRUE if it can be entered, B_FALSE if it can't be (because a
3638  * duplicate policy exists with exactly the same selectors), or an icmp
3639  * rule exists with a different encryption/authentication action.
3640  */
3641 boolean_t
3642 ipsec_check_policy(ipsec_policy_head_t *php, ipsec_policy_t *ipp, int direction)
3643 {
3644         ipsec_policy_root_t *pr = &php->iph_root[direction];
3645         int af = -1;
3646         ipsec_policy_t *p2, *head;
3647         uint8_t check_proto;
3648         ipsec_selkey_t *selkey = &ipp->ipsp_sel->ipsl_key;
3649         uint32_t        valid = selkey->ipsl_valid;
3650 
3651         if (valid & IPSL_IPV6) {
3652                 ASSERT(!(valid & IPSL_IPV4));
3653                 af = IPSEC_AF_V6;
3654                 check_proto = IPPROTO_ICMPV6;
3655         } else {
3656                 ASSERT(valid & IPSL_IPV4);
3657                 af = IPSEC_AF_V4;
3658                 check_proto = IPPROTO_ICMP;
3659         }
3660 
3661         ASSERT(RW_WRITE_HELD(&php->iph_lock));
3662 
3663         /*
3664          * Double-check that we don't have any duplicate selectors here.
3665          * Because selectors are interned below, we need only compare pointers
3666          * for equality.
3667          */
3668         if (selkey->ipsl_sel_hval == IPSEC_SEL_NOHASH) {
3669                 head = pr->ipr_nonhash[af];
3670         } else {
3671                 selkey->ipsl_pol_hval =
3672                     (selkey->ipsl_valid & IPSL_IPV4) ?
3673                     IPSEC_IPV4_HASH(selkey->ipsl_remote.ipsad_v4,
3674                     pr->ipr_nchains) :
3675                     IPSEC_IPV6_HASH(selkey->ipsl_remote.ipsad_v6,
3676                     pr->ipr_nchains);
3677 
3678                 head = pr->ipr_hash[selkey->ipsl_pol_hval].hash_head;
3679         }
3680 
3681         for (p2 = head; p2 != NULL; p2 = p2->ipsp_hash.hash_next) {
3682                 if (p2->ipsp_sel == ipp->ipsp_sel)
3683                         return (B_FALSE);
3684         }
3685 
3686         /*
3687          * If it's ICMP and not a drop or pass rule, run through the ICMP
3688          * rules and make sure the action is either new or the same as any
3689          * other actions.  We don't have to check the full chain because
3690          * discard and bypass will override all other actions
3691          */
3692 
3693         if (valid & IPSL_PROTOCOL &&
3694             selkey->ipsl_proto == check_proto &&
3695             (ipp->ipsp_act->ipa_act.ipa_type == IPSEC_ACT_APPLY)) {
3696 
3697                 for (p2 = head; p2 != NULL; p2 = p2->ipsp_hash.hash_next) {
3698 
3699                         if (p2->ipsp_sel->ipsl_key.ipsl_valid & IPSL_PROTOCOL &&
3700                             p2->ipsp_sel->ipsl_key.ipsl_proto == check_proto &&
3701                             (p2->ipsp_act->ipa_act.ipa_type ==
3702                             IPSEC_ACT_APPLY)) {
3703                                 return (ipsec_compare_action(p2, ipp));
3704                         }
3705                 }
3706         }
3707 
3708         return (B_TRUE);
3709 }
3710 
3711 /*
3712  * compare the action chains of two policies for equality
3713  * B_TRUE -> effective equality
3714  */
3715 
3716 static boolean_t
3717 ipsec_compare_action(ipsec_policy_t *p1, ipsec_policy_t *p2)
3718 {
3719 
3720         ipsec_action_t *act1, *act2;
3721 
3722         /* We have a valid rule. Let's compare the actions */
3723         if (p1->ipsp_act == p2->ipsp_act) {
3724                 /* same action. We are good */
3725                 return (B_TRUE);
3726         }
3727 
3728         /* we have to walk the chain */
3729 
3730         act1 = p1->ipsp_act;
3731         act2 = p2->ipsp_act;
3732 
3733         while (act1 != NULL && act2 != NULL) {
3734 
3735                 /* otherwise, Are we close enough? */
3736                 if (act1->ipa_allow_clear != act2->ipa_allow_clear ||
3737                     act1->ipa_want_ah != act2->ipa_want_ah ||
3738                     act1->ipa_want_esp != act2->ipa_want_esp ||
3739                     act1->ipa_want_se != act2->ipa_want_se) {
3740                         /* Nope, we aren't */
3741                         return (B_FALSE);
3742                 }
3743 
3744                 if (act1->ipa_want_ah) {
3745                         if (act1->ipa_act.ipa_apply.ipp_auth_alg !=
3746                             act2->ipa_act.ipa_apply.ipp_auth_alg) {
3747                                 return (B_FALSE);
3748                         }
3749 
3750                         if (act1->ipa_act.ipa_apply.ipp_ah_minbits !=
3751                             act2->ipa_act.ipa_apply.ipp_ah_minbits ||
3752                             act1->ipa_act.ipa_apply.ipp_ah_maxbits !=
3753                             act2->ipa_act.ipa_apply.ipp_ah_maxbits) {
3754                                 return (B_FALSE);
3755                         }
3756                 }
3757 
3758                 if (act1->ipa_want_esp) {
3759                         if (act1->ipa_act.ipa_apply.ipp_use_esp !=
3760                             act2->ipa_act.ipa_apply.ipp_use_esp ||
3761                             act1->ipa_act.ipa_apply.ipp_use_espa !=
3762                             act2->ipa_act.ipa_apply.ipp_use_espa) {
3763                                 return (B_FALSE);
3764                         }
3765 
3766                         if (act1->ipa_act.ipa_apply.ipp_use_esp) {
3767                                 if (act1->ipa_act.ipa_apply.ipp_encr_alg !=
3768                                     act2->ipa_act.ipa_apply.ipp_encr_alg) {
3769                                         return (B_FALSE);
3770                                 }
3771 
3772                                 if (act1->ipa_act.ipa_apply.ipp_espe_minbits !=
3773                                     act2->ipa_act.ipa_apply.ipp_espe_minbits ||
3774                                     act1->ipa_act.ipa_apply.ipp_espe_maxbits !=
3775                                     act2->ipa_act.ipa_apply.ipp_espe_maxbits) {
3776                                         return (B_FALSE);
3777                                 }
3778                         }
3779 
3780                         if (act1->ipa_act.ipa_apply.ipp_use_espa) {
3781                                 if (act1->ipa_act.ipa_apply.ipp_esp_auth_alg !=
3782                                     act2->ipa_act.ipa_apply.ipp_esp_auth_alg) {
3783                                         return (B_FALSE);
3784                                 }
3785 
3786                                 if (act1->ipa_act.ipa_apply.ipp_espa_minbits !=
3787                                     act2->ipa_act.ipa_apply.ipp_espa_minbits ||
3788                                     act1->ipa_act.ipa_apply.ipp_espa_maxbits !=
3789                                     act2->ipa_act.ipa_apply.ipp_espa_maxbits) {
3790                                         return (B_FALSE);
3791                                 }
3792                         }
3793 
3794                 }
3795 
3796                 act1 = act1->ipa_next;
3797                 act2 = act2->ipa_next;
3798         }
3799 
3800         if (act1 != NULL || act2 != NULL) {
3801                 return (B_FALSE);
3802         }
3803 
3804         return (B_TRUE);
3805 }
3806 
3807 
3808 /*
3809  * Given a constructed ipsec_policy_t policy rule, enter it into
3810  * the correct policy ruleset.
3811  *
3812  * ipsec_check_policy() is assumed to have succeeded first (to check for
3813  * duplicates).
3814  */
3815 void
3816 ipsec_enter_policy(ipsec_policy_head_t *php, ipsec_policy_t *ipp, int direction,
3817     netstack_t *ns)
3818 {
3819         ipsec_policy_root_t *pr = &php->iph_root[direction];
3820         ipsec_selkey_t *selkey = &ipp->ipsp_sel->ipsl_key;
3821         uint32_t valid = selkey->ipsl_valid;
3822         uint32_t hval = selkey->ipsl_pol_hval;
3823         int af = -1;
3824 
3825         ASSERT(RW_WRITE_HELD(&php->iph_lock));
3826 
3827         if (valid & IPSL_IPV6) {
3828                 ASSERT(!(valid & IPSL_IPV4));
3829                 af = IPSEC_AF_V6;
3830         } else {
3831                 ASSERT(valid & IPSL_IPV4);
3832                 af = IPSEC_AF_V4;
3833         }
3834 
3835         php->iph_gen++;
3836 
3837         if (hval == IPSEC_SEL_NOHASH) {
3838                 HASHLIST_INSERT(ipp, ipsp_hash, pr->ipr_nonhash[af]);
3839         } else {
3840                 HASH_LOCK(pr->ipr_hash, hval);
3841                 HASH_INSERT(ipp, ipsp_hash, pr->ipr_hash, hval);
3842                 HASH_UNLOCK(pr->ipr_hash, hval);
3843         }
3844 
3845         ipsec_insert_always(&php->iph_rulebyid, ipp);
3846 
3847         ipsec_update_present_flags(ns->netstack_ipsec);
3848 }
3849 
3850 static void
3851 ipsec_ipr_flush(ipsec_policy_head_t *php, ipsec_policy_root_t *ipr)
3852 {
3853         ipsec_policy_t *ip, *nip;
3854         int af, chain, nchain;
3855 
3856         for (af = 0; af < IPSEC_NAF; af++) {
3857                 for (ip = ipr->ipr_nonhash[af]; ip != NULL; ip = nip) {
3858                         nip = ip->ipsp_hash.hash_next;
3859                         IPPOL_UNCHAIN(php, ip);
3860                 }
3861                 ipr->ipr_nonhash[af] = NULL;
3862         }
3863         nchain = ipr->ipr_nchains;
3864 
3865         for (chain = 0; chain < nchain; chain++) {
3866                 for (ip = ipr->ipr_hash[chain].hash_head; ip != NULL;
3867                     ip = nip) {
3868                         nip = ip->ipsp_hash.hash_next;
3869                         IPPOL_UNCHAIN(php, ip);
3870                 }
3871                 ipr->ipr_hash[chain].hash_head = NULL;
3872         }
3873 }
3874 
3875 /*
3876  * Create and insert inbound or outbound policy associated with actp for the
3877  * address family fam into the policy head ph.  Returns B_TRUE if policy was
3878  * inserted, and B_FALSE otherwise.
3879  */
3880 boolean_t
3881 ipsec_polhead_insert(ipsec_policy_head_t *ph, ipsec_act_t *actp, uint_t nact,
3882     int fam, int ptype, netstack_t *ns)
3883 {
3884         ipsec_selkey_t          sel;
3885         ipsec_policy_t          *pol;
3886         ipsec_policy_root_t     *pr;
3887 
3888         bzero(&sel, sizeof (sel));
3889         sel.ipsl_valid = (fam == IPSEC_AF_V4 ? IPSL_IPV4 : IPSL_IPV6);
3890         if ((pol = ipsec_policy_create(&sel, actp, nact, IPSEC_PRIO_SOCKET,
3891             NULL, ns)) != NULL) {
3892                 pr = &ph->iph_root[ptype];
3893                 HASHLIST_INSERT(pol, ipsp_hash, pr->ipr_nonhash[fam]);
3894                 ipsec_insert_always(&ph->iph_rulebyid, pol);
3895         }
3896         return (pol != NULL);
3897 }
3898 
3899 void
3900 ipsec_polhead_flush(ipsec_policy_head_t *php, netstack_t *ns)
3901 {
3902         int dir;
3903 
3904         ASSERT(RW_WRITE_HELD(&php->iph_lock));
3905 
3906         for (dir = 0; dir < IPSEC_NTYPES; dir++)
3907                 ipsec_ipr_flush(php, &php->iph_root[dir]);
3908 
3909         php->iph_gen++;
3910         ipsec_update_present_flags(ns->netstack_ipsec);
3911 }
3912 
3913 void
3914 ipsec_polhead_free(ipsec_policy_head_t *php, netstack_t *ns)
3915 {
3916         int dir;
3917 
3918         ASSERT(php->iph_refs == 0);
3919 
3920         rw_enter(&php->iph_lock, RW_WRITER);
3921         ipsec_polhead_flush(php, ns);
3922         rw_exit(&php->iph_lock);
3923         rw_destroy(&php->iph_lock);
3924         for (dir = 0; dir < IPSEC_NTYPES; dir++) {
3925                 ipsec_policy_root_t *ipr = &php->iph_root[dir];
3926                 int chain;
3927 
3928                 for (chain = 0; chain < ipr->ipr_nchains; chain++)
3929                         mutex_destroy(&(ipr->ipr_hash[chain].hash_lock));
3930 
3931         }
3932         ipsec_polhead_free_table(php);
3933         kmem_free(php, sizeof (*php));
3934 }
3935 
3936 static void
3937 ipsec_ipr_init(ipsec_policy_root_t *ipr)
3938 {
3939         int af;
3940 
3941         ipr->ipr_nchains = 0;
3942         ipr->ipr_hash = NULL;
3943 
3944         for (af = 0; af < IPSEC_NAF; af++) {
3945                 ipr->ipr_nonhash[af] = NULL;
3946         }
3947 }
3948 
3949 ipsec_policy_head_t *
3950 ipsec_polhead_create(void)
3951 {
3952         ipsec_policy_head_t *php;
3953 
3954         php = kmem_alloc(sizeof (*php), KM_NOSLEEP);
3955         if (php == NULL)
3956                 return (php);
3957 
3958         rw_init(&php->iph_lock, NULL, RW_DEFAULT, NULL);
3959         php->iph_refs = 1;
3960         php->iph_gen = 0;
3961 
3962         ipsec_ipr_init(&php->iph_root[IPSEC_TYPE_INBOUND]);
3963         ipsec_ipr_init(&php->iph_root[IPSEC_TYPE_OUTBOUND]);
3964 
3965         avl_create(&php->iph_rulebyid, ipsec_policy_cmpbyid,
3966             sizeof (ipsec_policy_t), offsetof(ipsec_policy_t, ipsp_byid));
3967 
3968         return (php);
3969 }
3970 
3971 /*
3972  * Clone the policy head into a new polhead; release one reference to the
3973  * old one and return the only reference to the new one.
3974  * If the old one had a refcount of 1, just return it.
3975  */
3976 ipsec_policy_head_t *
3977 ipsec_polhead_split(ipsec_policy_head_t *php, netstack_t *ns)
3978 {
3979         ipsec_policy_head_t *nphp;
3980 
3981         if (php == NULL)
3982                 return (ipsec_polhead_create());
3983         else if (php->iph_refs == 1)
3984                 return (php);
3985 
3986         nphp = ipsec_polhead_create();
3987         if (nphp == NULL)
3988                 return (NULL);
3989 
3990         if (ipsec_copy_polhead(php, nphp, ns) != 0) {
3991                 ipsec_polhead_free(nphp, ns);
3992                 return (NULL);
3993         }
3994         IPPH_REFRELE(php, ns);
3995         return (nphp);
3996 }
3997 
3998 /*
3999  * When sending a response to a ICMP request or generating a RST
4000  * in the TCP case, the outbound packets need to go at the same level
4001  * of protection as the incoming ones i.e we associate our outbound
4002  * policy with how the packet came in. We call this after we have
4003  * accepted the incoming packet which may or may not have been in
4004  * clear and hence we are sending the reply back with the policy
4005  * matching the incoming datagram's policy.
4006  *
4007  * NOTE : This technology serves two purposes :
4008  *
4009  * 1) If we have multiple outbound policies, we send out a reply
4010  *    matching with how it came in rather than matching the outbound
4011  *    policy.
4012  *
4013  * 2) For assymetric policies, we want to make sure that incoming
4014  *    and outgoing has the same level of protection. Assymetric
4015  *    policies exist only with global policy where we may not have
4016  *    both outbound and inbound at the same time.
4017  *
4018  * NOTE2:       This function is called by cleartext cases, so it needs to be
4019  *              in IP proper.
4020  *
4021  * Note: the caller has moved other parts of ira into ixa already.
4022  */
4023 boolean_t
4024 ipsec_in_to_out(ip_recv_attr_t *ira, ip_xmit_attr_t *ixa, mblk_t *data_mp,
4025     ipha_t *ipha, ip6_t *ip6h)
4026 {
4027         ipsec_selector_t sel;
4028         ipsec_action_t  *reflect_action = NULL;
4029         netstack_t      *ns = ixa->ixa_ipst->ips_netstack;
4030 
4031         bzero((void*)&sel, sizeof (sel));
4032 
4033         if (ira->ira_ipsec_action != NULL) {
4034                 /* transfer reference.. */
4035                 reflect_action = ira->ira_ipsec_action;
4036                 ira->ira_ipsec_action = NULL;
4037         } else if (!(ira->ira_flags & IRAF_LOOPBACK))
4038                 reflect_action = ipsec_in_to_out_action(ira);
4039 
4040         /*
4041          * The caller is going to send the datagram out which might
4042          * go on the wire or delivered locally through ire_send_local.
4043          *
4044          * 1) If it goes out on the wire, new associations will be
4045          *    obtained.
4046          * 2) If it is delivered locally, ire_send_local will convert
4047          *    this ip_xmit_attr_t back to a ip_recv_attr_t looking at the
4048          *    requests.
4049          */
4050         ixa->ixa_ipsec_action = reflect_action;
4051 
4052         if (!ipsec_init_outbound_ports(&sel, data_mp, ipha, ip6h, 0,
4053             ns->netstack_ipsec)) {
4054                 /* Note: data_mp already consumed and ip_drop_packet done */
4055                 return (B_FALSE);
4056         }
4057         ixa->ixa_ipsec_src_port = sel.ips_local_port;
4058         ixa->ixa_ipsec_dst_port = sel.ips_remote_port;
4059         ixa->ixa_ipsec_proto = sel.ips_protocol;
4060         ixa->ixa_ipsec_icmp_type = sel.ips_icmp_type;
4061         ixa->ixa_ipsec_icmp_code = sel.ips_icmp_code;
4062 
4063         /*
4064          * Don't use global policy for this, as we want
4065          * to use the same protection that was applied to the inbound packet.
4066          * Thus we set IXAF_NO_IPSEC is it arrived in the clear to make
4067          * it be sent in the clear.
4068          */
4069         if (ira->ira_flags & IRAF_IPSEC_SECURE)
4070                 ixa->ixa_flags |= IXAF_IPSEC_SECURE;
4071         else
4072                 ixa->ixa_flags |= IXAF_NO_IPSEC;
4073 
4074         return (B_TRUE);
4075 }
4076 
4077 void
4078 ipsec_out_release_refs(ip_xmit_attr_t *ixa)
4079 {
4080         if (!(ixa->ixa_flags & IXAF_IPSEC_SECURE))
4081                 return;
4082 
4083         if (ixa->ixa_ipsec_ah_sa != NULL) {
4084                 IPSA_REFRELE(ixa->ixa_ipsec_ah_sa);
4085                 ixa->ixa_ipsec_ah_sa = NULL;
4086         }
4087         if (ixa->ixa_ipsec_esp_sa != NULL) {
4088                 IPSA_REFRELE(ixa->ixa_ipsec_esp_sa);
4089                 ixa->ixa_ipsec_esp_sa = NULL;
4090         }
4091         if (ixa->ixa_ipsec_policy != NULL) {
4092                 IPPOL_REFRELE(ixa->ixa_ipsec_policy);
4093                 ixa->ixa_ipsec_policy = NULL;
4094         }
4095         if (ixa->ixa_ipsec_action != NULL) {
4096                 IPACT_REFRELE(ixa->ixa_ipsec_action);
4097                 ixa->ixa_ipsec_action = NULL;
4098         }
4099         if (ixa->ixa_ipsec_latch) {
4100                 IPLATCH_REFRELE(ixa->ixa_ipsec_latch);
4101                 ixa->ixa_ipsec_latch = NULL;
4102         }
4103         /* Clear the soft references to the SAs */
4104         ixa->ixa_ipsec_ref[0].ipsr_sa = NULL;
4105         ixa->ixa_ipsec_ref[0].ipsr_bucket = NULL;
4106         ixa->ixa_ipsec_ref[0].ipsr_gen = 0;
4107         ixa->ixa_ipsec_ref[1].ipsr_sa = NULL;
4108         ixa->ixa_ipsec_ref[1].ipsr_bucket = NULL;
4109         ixa->ixa_ipsec_ref[1].ipsr_gen = 0;
4110         ixa->ixa_flags &= ~IXAF_IPSEC_SECURE;
4111 }
4112 
4113 void
4114 ipsec_in_release_refs(ip_recv_attr_t *ira)
4115 {
4116         if (!(ira->ira_flags & IRAF_IPSEC_SECURE))
4117                 return;
4118 
4119         if (ira->ira_ipsec_ah_sa != NULL) {
4120                 IPSA_REFRELE(ira->ira_ipsec_ah_sa);
4121                 ira->ira_ipsec_ah_sa = NULL;
4122         }
4123         if (ira->ira_ipsec_esp_sa != NULL) {
4124                 IPSA_REFRELE(ira->ira_ipsec_esp_sa);
4125                 ira->ira_ipsec_esp_sa = NULL;
4126         }
4127         ira->ira_flags &= ~IRAF_IPSEC_SECURE;
4128 }
4129 
4130 /*
4131  * This is called from ire_send_local when a packet
4132  * is looped back. We setup the ip_recv_attr_t "borrowing" the references
4133  * held by the callers.
4134  * Note that we don't do any IPsec but we carry the actions and IPSEC flags
4135  * across so that the fanout policy checks see that IPsec was applied.
4136  *
4137  * The caller should do ipsec_in_release_refs() on the ira by calling
4138  * ira_cleanup().
4139  */
4140 void
4141 ipsec_out_to_in(ip_xmit_attr_t *ixa, ill_t *ill, ip_recv_attr_t *ira)
4142 {
4143         ipsec_policy_t *pol;
4144         ipsec_action_t *act;
4145 
4146         /* Non-IPsec operations */
4147         ira->ira_free_flags = 0;
4148         ira->ira_zoneid = ixa->ixa_zoneid;
4149         ira->ira_cred = ixa->ixa_cred;
4150         ira->ira_cpid = ixa->ixa_cpid;
4151         ira->ira_tsl = ixa->ixa_tsl;
4152         ira->ira_ill = ira->ira_rill = ill;
4153         ira->ira_flags = ixa->ixa_flags & IAF_MASK;
4154         ira->ira_no_loop_zoneid = ixa->ixa_no_loop_zoneid;
4155         ira->ira_pktlen = ixa->ixa_pktlen;
4156         ira->ira_ip_hdr_length = ixa->ixa_ip_hdr_length;
4157         ira->ira_protocol = ixa->ixa_protocol;
4158         ira->ira_mhip = NULL;
4159 
4160         ira->ira_flags |= IRAF_LOOPBACK | IRAF_L2SRC_LOOPBACK;
4161 
4162         ira->ira_sqp = ixa->ixa_sqp;
4163         ira->ira_ring = NULL;
4164 
4165         ira->ira_ruifindex = ill->ill_phyint->phyint_ifindex;
4166         ira->ira_rifindex = ira->ira_ruifindex;
4167 
4168         if (!(ixa->ixa_flags & IXAF_IPSEC_SECURE))
4169                 return;
4170 
4171         ira->ira_flags |= IRAF_IPSEC_SECURE;
4172 
4173         ira->ira_ipsec_ah_sa = NULL;
4174         ira->ira_ipsec_esp_sa = NULL;
4175 
4176         act = ixa->ixa_ipsec_action;
4177         if (act == NULL) {
4178                 pol = ixa->ixa_ipsec_policy;
4179                 if (pol != NULL) {
4180                         act = pol->ipsp_act;
4181                         IPACT_REFHOLD(act);
4182                 }
4183         }
4184         ixa->ixa_ipsec_action = NULL;
4185         ira->ira_ipsec_action = act;
4186 }
4187 
4188 /*
4189  * Consults global policy and per-socket policy to see whether this datagram
4190  * should go out secure. If so it updates the ip_xmit_attr_t
4191  * Should not be used when connecting, since then we want to latch the policy.
4192  *
4193  * If connp is NULL we just look at the global policy.
4194  *
4195  * Returns NULL if the packet was dropped, in which case the MIB has
4196  * been incremented and ip_drop_packet done.
4197  */
4198 mblk_t *
4199 ip_output_attach_policy(mblk_t *mp, ipha_t *ipha, ip6_t *ip6h,
4200     const conn_t *connp, ip_xmit_attr_t *ixa)
4201 {
4202         ipsec_selector_t sel;
4203         boolean_t       policy_present;
4204         ip_stack_t      *ipst = ixa->ixa_ipst;
4205         netstack_t      *ns = ipst->ips_netstack;
4206         ipsec_stack_t   *ipss = ns->netstack_ipsec;
4207         ipsec_policy_t  *p;
4208 
4209         ixa->ixa_ipsec_policy_gen = ipss->ipsec_system_policy.iph_gen;
4210         ASSERT((ipha != NULL && ip6h == NULL) ||
4211             (ip6h != NULL && ipha == NULL));
4212 
4213         if (ipha != NULL)
4214                 policy_present = ipss->ipsec_outbound_v4_policy_present;
4215         else
4216                 policy_present = ipss->ipsec_outbound_v6_policy_present;
4217 
4218         if (!policy_present && (connp == NULL || connp->conn_policy == NULL))
4219                 return (mp);
4220 
4221         bzero((void*)&sel, sizeof (sel));
4222 
4223         if (ipha != NULL) {
4224                 sel.ips_local_addr_v4 = ipha->ipha_src;
4225                 sel.ips_remote_addr_v4 = ip_get_dst(ipha);
4226                 sel.ips_isv4 = B_TRUE;
4227         } else {
4228                 sel.ips_isv4 = B_FALSE;
4229                 sel.ips_local_addr_v6 = ip6h->ip6_src;
4230                 sel.ips_remote_addr_v6 = ip_get_dst_v6(ip6h, mp, NULL);
4231         }
4232         sel.ips_protocol = ixa->ixa_protocol;
4233 
4234         if (!ipsec_init_outbound_ports(&sel, mp, ipha, ip6h, 0, ipss)) {
4235                 if (ipha != NULL) {
4236                         BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsOutDiscards);
4237                 } else {
4238                         BUMP_MIB(&ipst->ips_ip6_mib, ipIfStatsOutDiscards);
4239                 }
4240                 /* Note: mp already consumed and ip_drop_packet done */
4241                 return (NULL);
4242         }
4243 
4244         ASSERT(ixa->ixa_ipsec_policy == NULL);
4245         p = ipsec_find_policy(IPSEC_TYPE_OUTBOUND, connp, &sel, ns);
4246         ixa->ixa_ipsec_policy = p;
4247         if (p != NULL) {
4248                 ixa->ixa_flags |= IXAF_IPSEC_SECURE;
4249                 if (connp == NULL || connp->conn_policy == NULL)
4250                         ixa->ixa_flags |= IXAF_IPSEC_GLOBAL_POLICY;
4251         } else {
4252                 ixa->ixa_flags &= ~IXAF_IPSEC_SECURE;
4253         }
4254 
4255         /*
4256          * Copy the right port information.
4257          */
4258         ixa->ixa_ipsec_src_port = sel.ips_local_port;
4259         ixa->ixa_ipsec_dst_port = sel.ips_remote_port;
4260         ixa->ixa_ipsec_icmp_type = sel.ips_icmp_type;
4261         ixa->ixa_ipsec_icmp_code = sel.ips_icmp_code;
4262         ixa->ixa_ipsec_proto = sel.ips_protocol;
4263         return (mp);
4264 }
4265 
4266 /*
4267  * When appropriate, this function caches inbound and outbound policy
4268  * for this connection. The outbound policy is stored in conn_ixa.
4269  * Note that it can not be used for SCTP since conn_faddr isn't set for SCTP.
4270  *
4271  * XXX need to work out more details about per-interface policy and
4272  * caching here!
4273  *
4274  * XXX may want to split inbound and outbound caching for ill..
4275  */
4276 int
4277 ipsec_conn_cache_policy(conn_t *connp, boolean_t isv4)
4278 {
4279         boolean_t global_policy_present;
4280         netstack_t      *ns = connp->conn_netstack;
4281         ipsec_stack_t   *ipss = ns->netstack_ipsec;
4282 
4283         connp->conn_ixa->ixa_ipsec_policy_gen =
4284             ipss->ipsec_system_policy.iph_gen;
4285         /*
4286          * There is no policy latching for ICMP sockets because we can't
4287          * decide on which policy to use until we see the packet and get
4288          * type/code selectors.
4289          */
4290         if (connp->conn_proto == IPPROTO_ICMP ||
4291             connp->conn_proto == IPPROTO_ICMPV6) {
4292                 connp->conn_in_enforce_policy =
4293                     connp->conn_out_enforce_policy = B_TRUE;
4294                 if (connp->conn_latch != NULL) {
4295                         IPLATCH_REFRELE(connp->conn_latch);
4296                         connp->conn_latch = NULL;
4297                 }
4298                 if (connp->conn_latch_in_policy != NULL) {
4299                         IPPOL_REFRELE(connp->conn_latch_in_policy);
4300                         connp->conn_latch_in_policy = NULL;
4301                 }
4302                 if (connp->conn_latch_in_action != NULL) {
4303                         IPACT_REFRELE(connp->conn_latch_in_action);
4304                         connp->conn_latch_in_action = NULL;
4305                 }
4306                 if (connp->conn_ixa->ixa_ipsec_policy != NULL) {
4307                         IPPOL_REFRELE(connp->conn_ixa->ixa_ipsec_policy);
4308                         connp->conn_ixa->ixa_ipsec_policy = NULL;
4309                 }
4310                 if (connp->conn_ixa->ixa_ipsec_action != NULL) {
4311                         IPACT_REFRELE(connp->conn_ixa->ixa_ipsec_action);
4312                         connp->conn_ixa->ixa_ipsec_action = NULL;
4313                 }
4314                 connp->conn_ixa->ixa_flags &= ~IXAF_IPSEC_SECURE;
4315                 return (0);
4316         }
4317 
4318         global_policy_present = isv4 ?
4319             (ipss->ipsec_outbound_v4_policy_present ||
4320             ipss->ipsec_inbound_v4_policy_present) :
4321             (ipss->ipsec_outbound_v6_policy_present ||
4322             ipss->ipsec_inbound_v6_policy_present);
4323 
4324         if ((connp->conn_policy != NULL) || global_policy_present) {
4325                 ipsec_selector_t sel;
4326                 ipsec_policy_t  *p;
4327 
4328                 if (connp->conn_latch == NULL &&
4329                     (connp->conn_latch = iplatch_create()) == NULL) {
4330                         return (ENOMEM);
4331                 }
4332 
4333                 bzero((void*)&sel, sizeof (sel));
4334 
4335                 sel.ips_protocol = connp->conn_proto;
4336                 sel.ips_local_port = connp->conn_lport;
4337                 sel.ips_remote_port = connp->conn_fport;
4338                 sel.ips_is_icmp_inv_acq = 0;
4339                 sel.ips_isv4 = isv4;
4340                 if (isv4) {
4341                         sel.ips_local_addr_v4 = connp->conn_laddr_v4;
4342                         sel.ips_remote_addr_v4 = connp->conn_faddr_v4;
4343                 } else {
4344                         sel.ips_local_addr_v6 = connp->conn_laddr_v6;
4345                         sel.ips_remote_addr_v6 = connp->conn_faddr_v6;
4346                 }
4347 
4348                 p = ipsec_find_policy(IPSEC_TYPE_INBOUND, connp, &sel, ns);
4349                 if (connp->conn_latch_in_policy != NULL)
4350                         IPPOL_REFRELE(connp->conn_latch_in_policy);
4351                 connp->conn_latch_in_policy = p;
4352                 connp->conn_in_enforce_policy = (p != NULL);
4353 
4354                 p = ipsec_find_policy(IPSEC_TYPE_OUTBOUND, connp, &sel, ns);
4355                 if (connp->conn_ixa->ixa_ipsec_policy != NULL)
4356                         IPPOL_REFRELE(connp->conn_ixa->ixa_ipsec_policy);
4357                 connp->conn_ixa->ixa_ipsec_policy = p;
4358                 connp->conn_out_enforce_policy = (p != NULL);
4359                 if (p != NULL) {
4360                         connp->conn_ixa->ixa_flags |= IXAF_IPSEC_SECURE;
4361                         if (connp->conn_policy == NULL) {
4362                                 connp->conn_ixa->ixa_flags |=
4363                                     IXAF_IPSEC_GLOBAL_POLICY;
4364                         }
4365                 } else {
4366                         connp->conn_ixa->ixa_flags &= ~IXAF_IPSEC_SECURE;
4367                 }
4368                 /* Clear the latched actions too, in case we're recaching. */
4369                 if (connp->conn_ixa->ixa_ipsec_action != NULL) {
4370                         IPACT_REFRELE(connp->conn_ixa->ixa_ipsec_action);
4371                         connp->conn_ixa->ixa_ipsec_action = NULL;
4372                 }
4373                 if (connp->conn_latch_in_action != NULL) {
4374                         IPACT_REFRELE(connp->conn_latch_in_action);
4375                         connp->conn_latch_in_action = NULL;
4376                 }
4377                 connp->conn_ixa->ixa_ipsec_src_port = sel.ips_local_port;
4378                 connp->conn_ixa->ixa_ipsec_dst_port = sel.ips_remote_port;
4379                 connp->conn_ixa->ixa_ipsec_icmp_type = sel.ips_icmp_type;
4380                 connp->conn_ixa->ixa_ipsec_icmp_code = sel.ips_icmp_code;
4381                 connp->conn_ixa->ixa_ipsec_proto = sel.ips_protocol;
4382         } else {
4383                 connp->conn_ixa->ixa_flags &= ~IXAF_IPSEC_SECURE;
4384         }
4385 
4386         /*
4387          * We may or may not have policy for this endpoint.  We still set
4388          * conn_policy_cached so that inbound datagrams don't have to look
4389          * at global policy as policy is considered latched for these
4390          * endpoints.  We should not set conn_policy_cached until the conn
4391          * reflects the actual policy. If we *set* this before inheriting
4392          * the policy there is a window where the check
4393          * CONN_INBOUND_POLICY_PRESENT, will neither check with the policy
4394          * on the conn (because we have not yet copied the policy on to
4395          * conn and hence not set conn_in_enforce_policy) nor with the
4396          * global policy (because conn_policy_cached is already set).
4397          */
4398         connp->conn_policy_cached = B_TRUE;
4399         return (0);
4400 }
4401 
4402 /*
4403  * When appropriate, this function caches outbound policy for faddr/fport.
4404  * It is used when we are not connected i.e., when we can not latch the
4405  * policy.
4406  */
4407 void
4408 ipsec_cache_outbound_policy(const conn_t *connp, const in6_addr_t *v6src,
4409     const in6_addr_t *v6dst, in_port_t dstport, ip_xmit_attr_t *ixa)
4410 {
4411         boolean_t       isv4 = (ixa->ixa_flags & IXAF_IS_IPV4) != 0;
4412         boolean_t       global_policy_present;
4413         netstack_t      *ns = connp->conn_netstack;
4414         ipsec_stack_t   *ipss = ns->netstack_ipsec;
4415 
4416         ixa->ixa_ipsec_policy_gen = ipss->ipsec_system_policy.iph_gen;
4417 
4418         /*
4419          * There is no policy caching for ICMP sockets because we can't
4420          * decide on which policy to use until we see the packet and get
4421          * type/code selectors.
4422          */
4423         if (connp->conn_proto == IPPROTO_ICMP ||
4424             connp->conn_proto == IPPROTO_ICMPV6) {
4425                 ixa->ixa_flags &= ~IXAF_IPSEC_SECURE;
4426                 if (ixa->ixa_ipsec_policy != NULL) {
4427                         IPPOL_REFRELE(ixa->ixa_ipsec_policy);
4428                         ixa->ixa_ipsec_policy = NULL;
4429                 }
4430                 if (ixa->ixa_ipsec_action != NULL) {
4431                         IPACT_REFRELE(ixa->ixa_ipsec_action);
4432                         ixa->ixa_ipsec_action = NULL;
4433                 }
4434                 return;
4435         }
4436 
4437         global_policy_present = isv4 ?
4438             (ipss->ipsec_outbound_v4_policy_present ||
4439             ipss->ipsec_inbound_v4_policy_present) :
4440             (ipss->ipsec_outbound_v6_policy_present ||
4441             ipss->ipsec_inbound_v6_policy_present);
4442 
4443         if ((connp->conn_policy != NULL) || global_policy_present) {
4444                 ipsec_selector_t sel;
4445                 ipsec_policy_t  *p;
4446 
4447                 bzero((void*)&sel, sizeof (sel));
4448 
4449                 sel.ips_protocol = connp->conn_proto;
4450                 sel.ips_local_port = connp->conn_lport;
4451                 sel.ips_remote_port = dstport;
4452                 sel.ips_is_icmp_inv_acq = 0;
4453                 sel.ips_isv4 = isv4;
4454                 if (isv4) {
4455                         IN6_V4MAPPED_TO_IPADDR(v6src, sel.ips_local_addr_v4);
4456                         IN6_V4MAPPED_TO_IPADDR(v6dst, sel.ips_remote_addr_v4);
4457                 } else {
4458                         sel.ips_local_addr_v6 = *v6src;
4459                         sel.ips_remote_addr_v6 = *v6dst;
4460                 }
4461 
4462                 p = ipsec_find_policy(IPSEC_TYPE_OUTBOUND, connp, &sel, ns);
4463                 if (ixa->ixa_ipsec_policy != NULL)
4464                         IPPOL_REFRELE(ixa->ixa_ipsec_policy);
4465                 ixa->ixa_ipsec_policy = p;
4466                 if (p != NULL) {
4467                         ixa->ixa_flags |= IXAF_IPSEC_SECURE;
4468                         if (connp->conn_policy == NULL)
4469                                 ixa->ixa_flags |= IXAF_IPSEC_GLOBAL_POLICY;
4470                 } else {
4471                         ixa->ixa_flags &= ~IXAF_IPSEC_SECURE;
4472                 }
4473                 /* Clear the latched actions too, in case we're recaching. */
4474                 if (ixa->ixa_ipsec_action != NULL) {
4475                         IPACT_REFRELE(ixa->ixa_ipsec_action);
4476                         ixa->ixa_ipsec_action = NULL;
4477                 }
4478 
4479                 ixa->ixa_ipsec_src_port = sel.ips_local_port;
4480                 ixa->ixa_ipsec_dst_port = sel.ips_remote_port;
4481                 ixa->ixa_ipsec_icmp_type = sel.ips_icmp_type;
4482                 ixa->ixa_ipsec_icmp_code = sel.ips_icmp_code;
4483                 ixa->ixa_ipsec_proto = sel.ips_protocol;
4484         } else {
4485                 ixa->ixa_flags &= ~IXAF_IPSEC_SECURE;
4486                 if (ixa->ixa_ipsec_policy != NULL) {
4487                         IPPOL_REFRELE(ixa->ixa_ipsec_policy);
4488                         ixa->ixa_ipsec_policy = NULL;
4489                 }
4490                 if (ixa->ixa_ipsec_action != NULL) {
4491                         IPACT_REFRELE(ixa->ixa_ipsec_action);
4492                         ixa->ixa_ipsec_action = NULL;
4493                 }
4494         }
4495 }
4496 
4497 /*
4498  * Returns B_FALSE if the policy has gone stale.
4499  */
4500 boolean_t
4501 ipsec_outbound_policy_current(ip_xmit_attr_t *ixa)
4502 {
4503         ipsec_stack_t   *ipss = ixa->ixa_ipst->ips_netstack->netstack_ipsec;
4504 
4505         if (!(ixa->ixa_flags & IXAF_IPSEC_GLOBAL_POLICY))
4506                 return (B_TRUE);
4507 
4508         return (ixa->ixa_ipsec_policy_gen == ipss->ipsec_system_policy.iph_gen);
4509 }
4510 
4511 void
4512 iplatch_free(ipsec_latch_t *ipl)
4513 {
4514         if (ipl->ipl_local_cid != NULL)
4515                 IPSID_REFRELE(ipl->ipl_local_cid);
4516         if (ipl->ipl_remote_cid != NULL)
4517                 IPSID_REFRELE(ipl->ipl_remote_cid);
4518         mutex_destroy(&ipl->ipl_lock);
4519         kmem_free(ipl, sizeof (*ipl));
4520 }
4521 
4522 ipsec_latch_t *
4523 iplatch_create()
4524 {
4525         ipsec_latch_t *ipl = kmem_zalloc(sizeof (*ipl), KM_NOSLEEP);
4526         if (ipl == NULL)
4527                 return (ipl);
4528         mutex_init(&ipl->ipl_lock, NULL, MUTEX_DEFAULT, NULL);
4529         ipl->ipl_refcnt = 1;
4530         return (ipl);
4531 }
4532 
4533 /*
4534  * Hash function for ID hash table.
4535  */
4536 static uint32_t
4537 ipsid_hash(int idtype, char *idstring)
4538 {
4539         uint32_t hval = idtype;
4540         unsigned char c;
4541 
4542         while ((c = *idstring++) != 0) {
4543                 hval = (hval << 4) | (hval >> 28);
4544                 hval ^= c;
4545         }
4546         hval = hval ^ (hval >> 16);
4547         return (hval & (IPSID_HASHSIZE-1));
4548 }
4549 
4550 /*
4551  * Look up identity string in hash table.  Return identity object
4552  * corresponding to the name -- either preexisting, or newly allocated.
4553  *
4554  * Return NULL if we need to allocate a new one and can't get memory.
4555  */
4556 ipsid_t *
4557 ipsid_lookup(int idtype, char *idstring, netstack_t *ns)
4558 {
4559         ipsid_t *retval;
4560         char *nstr;
4561         int idlen = strlen(idstring) + 1;
4562         ipsec_stack_t   *ipss = ns->netstack_ipsec;
4563         ipsif_t *bucket;
4564 
4565         bucket = &ipss->ipsec_ipsid_buckets[ipsid_hash(idtype, idstring)];
4566 
4567         mutex_enter(&bucket->ipsif_lock);
4568 
4569         for (retval = bucket->ipsif_head; retval != NULL;
4570             retval = retval->ipsid_next) {
4571                 if (idtype != retval->ipsid_type)
4572                         continue;
4573                 if (bcmp(idstring, retval->ipsid_cid, idlen) != 0)
4574                         continue;
4575 
4576                 IPSID_REFHOLD(retval);
4577                 mutex_exit(&bucket->ipsif_lock);
4578                 return (retval);
4579         }
4580 
4581         retval = kmem_alloc(sizeof (*retval), KM_NOSLEEP);
4582         if (!retval) {
4583                 mutex_exit(&bucket->ipsif_lock);
4584                 return (NULL);
4585         }
4586 
4587         nstr = kmem_alloc(idlen, KM_NOSLEEP);
4588         if (!nstr) {
4589                 mutex_exit(&bucket->ipsif_lock);
4590                 kmem_free(retval, sizeof (*retval));
4591                 return (NULL);
4592         }
4593 
4594         retval->ipsid_refcnt = 1;
4595         retval->ipsid_next = bucket->ipsif_head;
4596         if (retval->ipsid_next != NULL)
4597                 retval->ipsid_next->ipsid_ptpn = &retval->ipsid_next;
4598         retval->ipsid_ptpn = &bucket->ipsif_head;
4599         retval->ipsid_type = idtype;
4600         retval->ipsid_cid = nstr;
4601         bucket->ipsif_head = retval;
4602         bcopy(idstring, nstr, idlen);
4603         mutex_exit(&bucket->ipsif_lock);
4604 
4605         return (retval);
4606 }
4607 
4608 /*
4609  * Garbage collect the identity hash table.
4610  */
4611 void
4612 ipsid_gc(netstack_t *ns)
4613 {
4614         int i, len;
4615         ipsid_t *id, *nid;
4616         ipsif_t *bucket;
4617         ipsec_stack_t   *ipss = ns->netstack_ipsec;
4618 
4619         for (i = 0; i < IPSID_HASHSIZE; i++) {
4620                 bucket = &ipss->ipsec_ipsid_buckets[i];
4621                 mutex_enter(&bucket->ipsif_lock);
4622                 for (id = bucket->ipsif_head; id != NULL; id = nid) {
4623                         nid = id->ipsid_next;
4624                         if (id->ipsid_refcnt == 0) {
4625                                 *id->ipsid_ptpn = nid;
4626                                 if (nid != NULL)
4627                                         nid->ipsid_ptpn = id->ipsid_ptpn;
4628                                 len = strlen(id->ipsid_cid) + 1;
4629                                 kmem_free(id->ipsid_cid, len);
4630                                 kmem_free(id, sizeof (*id));
4631                         }
4632                 }
4633                 mutex_exit(&bucket->ipsif_lock);
4634         }
4635 }
4636 
4637 /*
4638  * Return true if two identities are the same.
4639  */
4640 boolean_t
4641 ipsid_equal(ipsid_t *id1, ipsid_t *id2)
4642 {
4643         if (id1 == id2)
4644                 return (B_TRUE);
4645 #ifdef DEBUG
4646         if ((id1 == NULL) || (id2 == NULL))
4647                 return (B_FALSE);
4648         /*
4649          * test that we're interning id's correctly..
4650          */
4651         ASSERT((strcmp(id1->ipsid_cid, id2->ipsid_cid) != 0) ||
4652             (id1->ipsid_type != id2->ipsid_type));
4653 #endif
4654         return (B_FALSE);
4655 }
4656 
4657 /*
4658  * Initialize identity table; called during module initialization.
4659  */
4660 static void
4661 ipsid_init(netstack_t *ns)
4662 {
4663         ipsif_t *bucket;
4664         int i;
4665         ipsec_stack_t   *ipss = ns->netstack_ipsec;
4666 
4667         for (i = 0; i < IPSID_HASHSIZE; i++) {
4668                 bucket = &ipss->ipsec_ipsid_buckets[i];
4669                 mutex_init(&bucket->ipsif_lock, NULL, MUTEX_DEFAULT, NULL);
4670         }
4671 }
4672 
4673 /*
4674  * Free identity table (preparatory to module unload)
4675  */
4676 static void
4677 ipsid_fini(netstack_t *ns)
4678 {
4679         ipsif_t *bucket;
4680         int i;
4681         ipsec_stack_t   *ipss = ns->netstack_ipsec;
4682 
4683         for (i = 0; i < IPSID_HASHSIZE; i++) {
4684                 bucket = &ipss->ipsec_ipsid_buckets[i];
4685                 ASSERT(bucket->ipsif_head == NULL);
4686                 mutex_destroy(&bucket->ipsif_lock);
4687         }
4688 }
4689 
4690 /*
4691  * Update the minimum and maximum supported key sizes for the
4692  * specified algorithm. Must be called while holding the algorithms lock.
4693  */
4694 void
4695 ipsec_alg_fix_min_max(ipsec_alginfo_t *alg, ipsec_algtype_t alg_type,
4696     netstack_t *ns)
4697 {
4698         size_t crypto_min = (size_t)-1, crypto_max = 0;
4699         size_t cur_crypto_min, cur_crypto_max;
4700         boolean_t is_valid;
4701         crypto_mechanism_info_t *mech_infos;
4702         uint_t nmech_infos;
4703         int crypto_rc, i;
4704         crypto_mech_usage_t mask;
4705         ipsec_stack_t   *ipss = ns->netstack_ipsec;
4706 
4707         ASSERT(RW_WRITE_HELD(&ipss->ipsec_alg_lock));
4708 
4709         /*
4710          * Compute the min, max, and default key sizes (in number of
4711          * increments to the default key size in bits) as defined
4712          * by the algorithm mappings. This range of key sizes is used
4713          * for policy related operations. The effective key sizes
4714          * supported by the framework could be more limited than
4715          * those defined for an algorithm.
4716          */
4717         alg->alg_default_bits = alg->alg_key_sizes[0];
4718         alg->alg_default = 0;
4719         if (alg->alg_increment != 0) {
4720                 /* key sizes are defined by range & increment */
4721                 alg->alg_minbits = alg->alg_key_sizes[1];
4722                 alg->alg_maxbits = alg->alg_key_sizes[2];
4723         } else if (alg->alg_nkey_sizes == 0) {
4724                 /* no specified key size for algorithm */
4725                 alg->alg_minbits = alg->alg_maxbits = 0;
4726         } else {
4727                 /* key sizes are defined by enumeration */
4728                 alg->alg_minbits = (uint16_t)-1;
4729                 alg->alg_maxbits = 0;
4730 
4731                 for (i = 0; i < alg->alg_nkey_sizes; i++) {
4732                         if (alg->alg_key_sizes[i] < alg->alg_minbits)
4733                                 alg->alg_minbits = alg->alg_key_sizes[i];
4734                         if (alg->alg_key_sizes[i] > alg->alg_maxbits)
4735                                 alg->alg_maxbits = alg->alg_key_sizes[i];
4736                 }
4737         }
4738 
4739         if (!(alg->alg_flags & ALG_FLAG_VALID))
4740                 return;
4741 
4742         /*
4743          * Mechanisms do not apply to the NULL encryption
4744          * algorithm, so simply return for this case.
4745          */
4746         if (alg->alg_id == SADB_EALG_NULL)
4747                 return;
4748 
4749         /*
4750          * Find the min and max key sizes supported by the cryptographic
4751          * framework providers.
4752          */
4753 
4754         /* get the key sizes supported by the framework */
4755         crypto_rc = crypto_get_all_mech_info(alg->alg_mech_type,
4756             &mech_infos, &nmech_infos, KM_SLEEP);
4757         if (crypto_rc != CRYPTO_SUCCESS || nmech_infos == 0) {
4758                 alg->alg_flags &= ~ALG_FLAG_VALID;
4759                 return;
4760         }
4761 
4762         /* min and max key sizes supported by framework */
4763         for (i = 0, is_valid = B_FALSE; i < nmech_infos; i++) {
4764                 int unit_bits;
4765 
4766                 /*
4767                  * Ignore entries that do not support the operations
4768                  * needed for the algorithm type.
4769                  */
4770                 if (alg_type == IPSEC_ALG_AUTH) {
4771                         mask = CRYPTO_MECH_USAGE_MAC;
4772                 } else {
4773                         mask = CRYPTO_MECH_USAGE_ENCRYPT |
4774                             CRYPTO_MECH_USAGE_DECRYPT;
4775                 }
4776                 if ((mech_infos[i].mi_usage & mask) != mask)
4777                         continue;
4778 
4779                 unit_bits = (mech_infos[i].mi_keysize_unit ==
4780                     CRYPTO_KEYSIZE_UNIT_IN_BYTES)  ? 8 : 1;
4781                 /* adjust min/max supported by framework */
4782                 cur_crypto_min = mech_infos[i].mi_min_key_size * unit_bits;
4783                 cur_crypto_max = mech_infos[i].mi_max_key_size * unit_bits;
4784 
4785                 if (cur_crypto_min < crypto_min)
4786                         crypto_min = cur_crypto_min;
4787 
4788                 /*
4789                  * CRYPTO_EFFECTIVELY_INFINITE is a special value of
4790                  * the crypto framework which means "no upper limit".
4791                  */
4792                 if (mech_infos[i].mi_max_key_size ==
4793                     CRYPTO_EFFECTIVELY_INFINITE) {
4794                         crypto_max = (size_t)-1;
4795                 } else if (cur_crypto_max > crypto_max) {
4796                         crypto_max = cur_crypto_max;
4797                 }
4798 
4799                 is_valid = B_TRUE;
4800         }
4801 
4802         kmem_free(mech_infos, sizeof (crypto_mechanism_info_t) *
4803             nmech_infos);
4804 
4805         if (!is_valid) {
4806                 /* no key sizes supported by framework */
4807                 alg->alg_flags &= ~ALG_FLAG_VALID;
4808                 return;
4809         }
4810 
4811         /*
4812          * Determine min and max key sizes from alg_key_sizes[].
4813          * defined for the algorithm entry. Adjust key sizes based on
4814          * those supported by the framework.
4815          */
4816         alg->alg_ef_default_bits = alg->alg_key_sizes[0];
4817 
4818         /*
4819          * For backwards compatability, assume that the IV length
4820          * is the same as the data length.
4821          */
4822         alg->alg_ivlen = alg->alg_datalen;
4823 
4824         /*
4825          * Copy any algorithm parameters (if provided) into dedicated
4826          * elements in the ipsec_alginfo_t structure.
4827          * There may be a better place to put this code.
4828          */
4829         for (i = 0; i < alg->alg_nparams; i++) {
4830                 switch (i) {
4831                 case 0:
4832                         /* Initialisation Vector length (bytes) */
4833                         alg->alg_ivlen =  alg->alg_params[0];
4834                         break;
4835                 case 1:
4836                         /* Integrity Check Vector length (bytes) */
4837                         alg->alg_icvlen = alg->alg_params[1];
4838                         break;
4839                 case 2:
4840                         /* Salt length (bytes) */
4841                         alg->alg_saltlen = (uint8_t)alg->alg_params[2];
4842                         break;
4843                 default:
4844                         break;
4845                 }
4846         }
4847 
4848         /* Default if the IV length is not specified. */
4849         if (alg_type == IPSEC_ALG_ENCR && alg->alg_ivlen == 0)
4850                 alg->alg_ivlen = alg->alg_datalen;
4851 
4852         alg_flag_check(alg);
4853 
4854         if (alg->alg_increment != 0) {
4855                 /* supported key sizes are defined by range  & increment */
4856                 crypto_min = ALGBITS_ROUND_UP(crypto_min, alg->alg_increment);
4857                 crypto_max = ALGBITS_ROUND_DOWN(crypto_max, alg->alg_increment);
4858 
4859                 alg->alg_ef_minbits = MAX(alg->alg_minbits,
4860                     (uint16_t)crypto_min);
4861                 alg->alg_ef_maxbits = MIN(alg->alg_maxbits,
4862                     (uint16_t)crypto_max);
4863 
4864                 /*
4865                  * If the sizes supported by the framework are outside
4866                  * the range of sizes defined by the algorithm mappings,
4867                  * the algorithm cannot be used. Check for this
4868                  * condition here.
4869                  */
4870                 if (alg->alg_ef_minbits > alg->alg_ef_maxbits) {
4871                         alg->alg_flags &= ~ALG_FLAG_VALID;
4872                         return;
4873                 }
4874                 if (alg->alg_ef_default_bits < alg->alg_ef_minbits)
4875                         alg->alg_ef_default_bits = alg->alg_ef_minbits;
4876                 if (alg->alg_ef_default_bits > alg->alg_ef_maxbits)
4877                         alg->alg_ef_default_bits = alg->alg_ef_maxbits;
4878         } else if (alg->alg_nkey_sizes == 0) {
4879                 /* no specified key size for algorithm */
4880                 alg->alg_ef_minbits = alg->alg_ef_maxbits = 0;
4881         } else {
4882                 /* supported key sizes are defined by enumeration */
4883                 alg->alg_ef_minbits = (uint16_t)-1;
4884                 alg->alg_ef_maxbits = 0;
4885 
4886                 for (i = 0, is_valid = B_FALSE; i < alg->alg_nkey_sizes; i++) {
4887                         /*
4888                          * Ignore the current key size if it is not in the
4889                          * range of sizes supported by the framework.
4890                          */
4891                         if (alg->alg_key_sizes[i] < crypto_min ||
4892                             alg->alg_key_sizes[i] > crypto_max)
4893                                 continue;
4894                         if (alg->alg_key_sizes[i] < alg->alg_ef_minbits)
4895                                 alg->alg_ef_minbits = alg->alg_key_sizes[i];
4896                         if (alg->alg_key_sizes[i] > alg->alg_ef_maxbits)
4897                                 alg->alg_ef_maxbits = alg->alg_key_sizes[i];
4898                         is_valid = B_TRUE;
4899                 }
4900 
4901                 if (!is_valid) {
4902                         alg->alg_flags &= ~ALG_FLAG_VALID;
4903                         return;
4904                 }
4905                 alg->alg_ef_default = 0;
4906         }
4907 }
4908 
4909 /*
4910  * Sanity check parameters provided by ipsecalgs(1m). Assume that
4911  * the algoritm is marked as valid, there is a check at the top
4912  * of this function. If any of the checks below fail, the algorithm
4913  * entry is invalid.
4914  */
4915 void
4916 alg_flag_check(ipsec_alginfo_t *alg)
4917 {
4918         alg->alg_flags &= ~ALG_FLAG_VALID;
4919 
4920         /*
4921          * Can't have the algorithm marked as CCM and GCM.
4922          * Check the ALG_FLAG_COMBINED and ALG_FLAG_COUNTERMODE
4923          * flags are set for CCM & GCM.
4924          */
4925         if ((alg->alg_flags & (ALG_FLAG_CCM|ALG_FLAG_GCM)) ==
4926             (ALG_FLAG_CCM|ALG_FLAG_GCM))
4927                 return;
4928         if (alg->alg_flags & (ALG_FLAG_CCM|ALG_FLAG_GCM)) {
4929                 if (!(alg->alg_flags & ALG_FLAG_COUNTERMODE))
4930                         return;
4931                 if (!(alg->alg_flags & ALG_FLAG_COMBINED))
4932                         return;
4933         }
4934 
4935         /*
4936          * For ALG_FLAG_COUNTERMODE, check the parameters
4937          * fit in the ipsec_nonce_t structure.
4938          */
4939         if (alg->alg_flags & ALG_FLAG_COUNTERMODE) {
4940                 if (alg->alg_ivlen != sizeof (((ipsec_nonce_t *)NULL)->iv))
4941                         return;
4942                 if (alg->alg_saltlen > sizeof (((ipsec_nonce_t *)NULL)->salt))
4943                         return;
4944         }
4945         if ((alg->alg_flags & ALG_FLAG_COMBINED) &&
4946             (alg->alg_icvlen == 0))
4947                 return;
4948 
4949         /* all is well. */
4950         alg->alg_flags |= ALG_FLAG_VALID;
4951 }
4952 
4953 /*
4954  * Free the memory used by the specified algorithm.
4955  */
4956 void
4957 ipsec_alg_free(ipsec_alginfo_t *alg)
4958 {
4959         if (alg == NULL)
4960                 return;
4961 
4962         if (alg->alg_key_sizes != NULL) {
4963                 kmem_free(alg->alg_key_sizes,
4964                     (alg->alg_nkey_sizes + 1) * sizeof (uint16_t));
4965                 alg->alg_key_sizes = NULL;
4966         }
4967         if (alg->alg_block_sizes != NULL) {
4968                 kmem_free(alg->alg_block_sizes,
4969                     (alg->alg_nblock_sizes + 1) * sizeof (uint16_t));
4970                 alg->alg_block_sizes = NULL;
4971         }
4972         if (alg->alg_params != NULL) {
4973                 kmem_free(alg->alg_params,
4974                     (alg->alg_nparams + 1) * sizeof (uint16_t));
4975                 alg->alg_params = NULL;
4976         }
4977         kmem_free(alg, sizeof (*alg));
4978 }
4979 
4980 /*
4981  * Check the validity of the specified key size for an algorithm.
4982  * Returns B_TRUE if key size is valid, B_FALSE otherwise.
4983  */
4984 boolean_t
4985 ipsec_valid_key_size(uint16_t key_size, ipsec_alginfo_t *alg)
4986 {
4987         if (key_size < alg->alg_ef_minbits || key_size > alg->alg_ef_maxbits)
4988                 return (B_FALSE);
4989 
4990         if (alg->alg_increment == 0 && alg->alg_nkey_sizes != 0) {
4991                 /*
4992                  * If the key sizes are defined by enumeration, the new
4993                  * key size must be equal to one of the supported values.
4994                  */
4995                 int i;
4996 
4997                 for (i = 0; i < alg->alg_nkey_sizes; i++)
4998                         if (key_size == alg->alg_key_sizes[i])
4999                                 break;
5000                 if (i == alg->alg_nkey_sizes)
5001                         return (B_FALSE);
5002         }
5003 
5004         return (B_TRUE);
5005 }
5006 
5007 /*
5008  * Callback function invoked by the crypto framework when a provider
5009  * registers or unregisters. This callback updates the algorithms
5010  * tables when a crypto algorithm is no longer available or becomes
5011  * available, and triggers the freeing/creation of context templates
5012  * associated with existing SAs, if needed.
5013  *
5014  * Need to walk all stack instances since the callback is global
5015  * for all instances
5016  */
5017 void
5018 ipsec_prov_update_callback(uint32_t event, void *event_arg)
5019 {
5020         netstack_handle_t nh;
5021         netstack_t *ns;
5022 
5023         netstack_next_init(&nh);
5024         while ((ns = netstack_next(&nh)) != NULL) {
5025                 ipsec_prov_update_callback_stack(event, event_arg, ns);
5026                 netstack_rele(ns);
5027         }
5028         netstack_next_fini(&nh);
5029 }
5030 
5031 static void
5032 ipsec_prov_update_callback_stack(uint32_t event, void *event_arg,
5033     netstack_t *ns)
5034 {
5035         crypto_notify_event_change_t *prov_change =
5036             (crypto_notify_event_change_t *)event_arg;
5037         uint_t algidx, algid, algtype, mech_count, mech_idx;
5038         ipsec_alginfo_t *alg;
5039         ipsec_alginfo_t oalg;
5040         crypto_mech_name_t *mechs;
5041         boolean_t alg_changed = B_FALSE;
5042         ipsec_stack_t   *ipss = ns->netstack_ipsec;
5043 
5044         /* ignore events for which we didn't register */
5045         if (event != CRYPTO_EVENT_MECHS_CHANGED) {
5046                 ip1dbg(("ipsec_prov_update_callback: unexpected event 0x%x "
5047                     " received from crypto framework\n", event));
5048                 return;
5049         }
5050 
5051         mechs = crypto_get_mech_list(&mech_count, KM_SLEEP);
5052         if (mechs == NULL)
5053                 return;
5054 
5055         /*
5056          * Walk the list of currently defined IPsec algorithm. Update
5057          * the algorithm valid flag and trigger an update of the
5058          * SAs that depend on that algorithm.
5059          */
5060         rw_enter(&ipss->ipsec_alg_lock, RW_WRITER);
5061         for (algtype = 0; algtype < IPSEC_NALGTYPES; algtype++) {
5062                 for (algidx = 0; algidx < ipss->ipsec_nalgs[algtype];
5063                     algidx++) {
5064 
5065                         algid = ipss->ipsec_sortlist[algtype][algidx];
5066                         alg = ipss->ipsec_alglists[algtype][algid];
5067                         ASSERT(alg != NULL);
5068 
5069                         /*
5070                          * Skip the algorithms which do not map to the
5071                          * crypto framework provider being added or removed.
5072                          */
5073                         if (strncmp(alg->alg_mech_name,
5074                             prov_change->ec_mech_name,
5075                             CRYPTO_MAX_MECH_NAME) != 0)
5076                                 continue;
5077 
5078                         /*
5079                          * Determine if the mechanism is valid. If it
5080                          * is not, mark the algorithm as being invalid. If
5081                          * it is, mark the algorithm as being valid.
5082                          */
5083                         for (mech_idx = 0; mech_idx < mech_count; mech_idx++)
5084                                 if (strncmp(alg->alg_mech_name,
5085                                     mechs[mech_idx], CRYPTO_MAX_MECH_NAME) == 0)
5086                                         break;
5087                         if (mech_idx == mech_count &&
5088                             alg->alg_flags & ALG_FLAG_VALID) {
5089                                 alg->alg_flags &= ~ALG_FLAG_VALID;
5090                                 alg_changed = B_TRUE;
5091                         } else if (mech_idx < mech_count &&
5092                             !(alg->alg_flags & ALG_FLAG_VALID)) {
5093                                 alg->alg_flags |= ALG_FLAG_VALID;
5094                                 alg_changed = B_TRUE;
5095                         }
5096 
5097                         /*
5098                          * Update the supported key sizes, regardless
5099                          * of whether a crypto provider was added or
5100                          * removed.
5101                          */
5102                         oalg = *alg;
5103                         ipsec_alg_fix_min_max(alg, algtype, ns);
5104                         if (!alg_changed &&
5105                             alg->alg_ef_minbits != oalg.alg_ef_minbits ||
5106                             alg->alg_ef_maxbits != oalg.alg_ef_maxbits ||
5107                             alg->alg_ef_default != oalg.alg_ef_default ||
5108                             alg->alg_ef_default_bits !=
5109                             oalg.alg_ef_default_bits)
5110                                 alg_changed = B_TRUE;
5111 
5112                         /*
5113                          * Update the affected SAs if a software provider is
5114                          * being added or removed.
5115                          */
5116                         if (prov_change->ec_provider_type ==
5117                             CRYPTO_SW_PROVIDER)
5118                                 sadb_alg_update(algtype, alg->alg_id,
5119                                     prov_change->ec_change ==
5120                                     CRYPTO_MECH_ADDED, ns);
5121                 }
5122         }
5123         rw_exit(&ipss->ipsec_alg_lock);
5124         crypto_free_mech_list(mechs, mech_count);
5125 
5126         if (alg_changed) {
5127                 /*
5128                  * An algorithm has changed, i.e. it became valid or
5129                  * invalid, or its support key sizes have changed.
5130                  * Notify ipsecah and ipsecesp of this change so
5131                  * that they can send a SADB_REGISTER to their consumers.
5132                  */
5133                 ipsecah_algs_changed(ns);
5134                 ipsecesp_algs_changed(ns);
5135         }
5136 }
5137 
5138 /*
5139  * Registers with the crypto framework to be notified of crypto
5140  * providers changes. Used to update the algorithm tables and
5141  * to free or create context templates if needed. Invoked after IPsec
5142  * is loaded successfully.
5143  *
5144  * This is called separately for each IP instance, so we ensure we only
5145  * register once.
5146  */
5147 void
5148 ipsec_register_prov_update(void)
5149 {
5150         if (prov_update_handle != NULL)
5151                 return;
5152 
5153         prov_update_handle = crypto_notify_events(
5154             ipsec_prov_update_callback, CRYPTO_EVENT_MECHS_CHANGED);
5155 }
5156 
5157 /*
5158  * Unregisters from the framework to be notified of crypto providers
5159  * changes. Called from ipsec_policy_g_destroy().
5160  */
5161 static void
5162 ipsec_unregister_prov_update(void)
5163 {
5164         if (prov_update_handle != NULL)
5165                 crypto_unnotify_events(prov_update_handle);
5166 }
5167 
5168 /*
5169  * Tunnel-mode support routines.
5170  */
5171 
5172 /*
5173  * Returns an mblk chain suitable for putnext() if policies match and IPsec
5174  * SAs are available.  If there's no per-tunnel policy, or a match comes back
5175  * with no match, then still return the packet and have global policy take
5176  * a crack at it in IP.
5177  * This updates the ip_xmit_attr with the IPsec policy.
5178  *
5179  * Remember -> we can be forwarding packets.  Keep that in mind w.r.t.
5180  * inner-packet contents.
5181  */
5182 mblk_t *
5183 ipsec_tun_outbound(mblk_t *mp, iptun_t *iptun, ipha_t *inner_ipv4,
5184     ip6_t *inner_ipv6, ipha_t *outer_ipv4, ip6_t *outer_ipv6, int outer_hdr_len,
5185     ip_xmit_attr_t *ixa)
5186 {
5187         ipsec_policy_head_t *polhead;
5188         ipsec_selector_t sel;
5189         mblk_t *nmp;
5190         boolean_t is_fragment;
5191         ipsec_policy_t *pol;
5192         ipsec_tun_pol_t *itp = iptun->iptun_itp;
5193         netstack_t *ns = iptun->iptun_ns;
5194         ipsec_stack_t *ipss = ns->netstack_ipsec;
5195 
5196         ASSERT(outer_ipv6 != NULL && outer_ipv4 == NULL ||
5197             outer_ipv4 != NULL && outer_ipv6 == NULL);
5198         /* We take care of inners in a bit. */
5199 
5200         /* Are the IPsec fields initialized at all? */
5201         if (!(ixa->ixa_flags & IXAF_IPSEC_SECURE)) {
5202                 ASSERT(ixa->ixa_ipsec_policy == NULL);
5203                 ASSERT(ixa->ixa_ipsec_latch == NULL);
5204                 ASSERT(ixa->ixa_ipsec_action == NULL);
5205                 ASSERT(ixa->ixa_ipsec_ah_sa == NULL);
5206                 ASSERT(ixa->ixa_ipsec_esp_sa == NULL);
5207         }
5208 
5209         ASSERT(itp != NULL && (itp->itp_flags & ITPF_P_ACTIVE));
5210         polhead = itp->itp_policy;
5211 
5212         bzero(&sel, sizeof (sel));
5213         if (inner_ipv4 != NULL) {
5214                 ASSERT(inner_ipv6 == NULL);
5215                 sel.ips_isv4 = B_TRUE;
5216                 sel.ips_local_addr_v4 = inner_ipv4->ipha_src;
5217                 sel.ips_remote_addr_v4 = inner_ipv4->ipha_dst;
5218                 sel.ips_protocol = (uint8_t)inner_ipv4->ipha_protocol;
5219         } else {
5220                 ASSERT(inner_ipv6 != NULL);
5221                 sel.ips_isv4 = B_FALSE;
5222                 sel.ips_local_addr_v6 = inner_ipv6->ip6_src;
5223                 /*
5224                  * We don't care about routing-header dests in the
5225                  * forwarding/tunnel path, so just grab ip6_dst.
5226                  */
5227                 sel.ips_remote_addr_v6 = inner_ipv6->ip6_dst;
5228         }
5229 
5230         if (itp->itp_flags & ITPF_P_PER_PORT_SECURITY) {
5231                 /*
5232                  * Caller can prepend the outer header, which means
5233                  * inner_ipv[46] may be stuck in the middle.  Pullup the whole
5234                  * mess now if need-be, for easier processing later.  Don't
5235                  * forget to rewire the outer header too.
5236                  */
5237                 if (mp->b_cont != NULL) {
5238                         nmp = msgpullup(mp, -1);
5239                         if (nmp == NULL) {
5240                                 ip_drop_packet(mp, B_FALSE, NULL,
5241                                     DROPPER(ipss, ipds_spd_nomem),
5242                                     &ipss->ipsec_spd_dropper);
5243                                 return (NULL);
5244                         }
5245                         freemsg(mp);
5246                         mp = nmp;
5247                         if (outer_ipv4 != NULL)
5248                                 outer_ipv4 = (ipha_t *)mp->b_rptr;
5249                         else
5250                                 outer_ipv6 = (ip6_t *)mp->b_rptr;
5251                         if (inner_ipv4 != NULL) {
5252                                 inner_ipv4 =
5253                                     (ipha_t *)(mp->b_rptr + outer_hdr_len);
5254                         } else {
5255                                 inner_ipv6 =
5256                                     (ip6_t *)(mp->b_rptr + outer_hdr_len);
5257                         }
5258                 }
5259                 if (inner_ipv4 != NULL) {
5260                         is_fragment = IS_V4_FRAGMENT(
5261                             inner_ipv4->ipha_fragment_offset_and_flags);
5262                 } else {
5263                         sel.ips_remote_addr_v6 = ip_get_dst_v6(inner_ipv6, mp,
5264                             &is_fragment);
5265                 }
5266 
5267                 if (is_fragment) {
5268                         ipha_t *oiph;
5269                         ipha_t *iph = NULL;
5270                         ip6_t *ip6h = NULL;
5271                         int hdr_len;
5272                         uint16_t ip6_hdr_length;
5273                         uint8_t v6_proto;
5274                         uint8_t *v6_proto_p;
5275 
5276                         /*
5277                          * We have a fragment we need to track!
5278                          */
5279                         mp = ipsec_fragcache_add(&itp->itp_fragcache, NULL, mp,
5280                             outer_hdr_len, ipss);
5281                         if (mp == NULL)
5282                                 return (NULL);
5283                         ASSERT(mp->b_cont == NULL);
5284 
5285                         /*
5286                          * If we get here, we have a full fragment chain
5287                          */
5288 
5289                         oiph = (ipha_t *)mp->b_rptr;
5290                         if (IPH_HDR_VERSION(oiph) == IPV4_VERSION) {
5291                                 hdr_len = ((outer_hdr_len != 0) ?
5292                                     IPH_HDR_LENGTH(oiph) : 0);
5293                                 iph = (ipha_t *)(mp->b_rptr + hdr_len);
5294                         } else {
5295                                 ASSERT(IPH_HDR_VERSION(oiph) == IPV6_VERSION);
5296                                 ip6h = (ip6_t *)mp->b_rptr;
5297                                 if (!ip_hdr_length_nexthdr_v6(mp, ip6h,
5298                                     &ip6_hdr_length, &v6_proto_p)) {
5299                                         ip_drop_packet_chain(mp, B_FALSE, NULL,
5300                                             DROPPER(ipss,
5301                                             ipds_spd_malformed_packet),
5302                                             &ipss->ipsec_spd_dropper);
5303                                         return (NULL);
5304                                 }
5305                                 hdr_len = ip6_hdr_length;
5306                         }
5307                         outer_hdr_len = hdr_len;
5308 
5309                         if (sel.ips_isv4) {
5310                                 if (iph == NULL) {
5311                                         /* Was v6 outer */
5312                                         iph = (ipha_t *)(mp->b_rptr + hdr_len);
5313                                 }
5314                                 inner_ipv4 = iph;
5315                                 sel.ips_local_addr_v4 = inner_ipv4->ipha_src;
5316                                 sel.ips_remote_addr_v4 = inner_ipv4->ipha_dst;
5317                                 sel.ips_protocol =
5318                                     (uint8_t)inner_ipv4->ipha_protocol;
5319                         } else {
5320                                 inner_ipv6 = (ip6_t *)(mp->b_rptr +
5321                                     hdr_len);
5322                                 sel.ips_local_addr_v6 = inner_ipv6->ip6_src;
5323                                 sel.ips_remote_addr_v6 = inner_ipv6->ip6_dst;
5324                                 if (!ip_hdr_length_nexthdr_v6(mp,
5325                                     inner_ipv6, &ip6_hdr_length, &v6_proto_p)) {
5326                                         ip_drop_packet_chain(mp, B_FALSE, NULL,
5327                                             DROPPER(ipss,
5328                                             ipds_spd_malformed_frag),
5329                                             &ipss->ipsec_spd_dropper);
5330                                         return (NULL);
5331                                 }
5332                                 v6_proto = *v6_proto_p;
5333                                 sel.ips_protocol = v6_proto;
5334 #ifdef FRAGCACHE_DEBUG
5335                                 cmn_err(CE_WARN, "v6_sel.ips_protocol = %d\n",
5336                                     sel.ips_protocol);
5337 #endif
5338                         }
5339                         /* Ports are extracted below */
5340                 }
5341 
5342                 /* Get ports... */
5343                 if (!ipsec_init_outbound_ports(&sel, mp,
5344                     inner_ipv4, inner_ipv6, outer_hdr_len, ipss)) {
5345                         /* callee did ip_drop_packet_chain() on mp. */
5346                         return (NULL);
5347                 }
5348 #ifdef FRAGCACHE_DEBUG
5349                 if (inner_ipv4 != NULL)
5350                         cmn_err(CE_WARN,
5351                             "(v4) sel.ips_protocol = %d, "
5352                             "sel.ips_local_port = %d, "
5353                             "sel.ips_remote_port = %d\n",
5354                             sel.ips_protocol, ntohs(sel.ips_local_port),
5355                             ntohs(sel.ips_remote_port));
5356                 if (inner_ipv6 != NULL)
5357                         cmn_err(CE_WARN,
5358                             "(v6) sel.ips_protocol = %d, "
5359                             "sel.ips_local_port = %d, "
5360                             "sel.ips_remote_port = %d\n",
5361                             sel.ips_protocol, ntohs(sel.ips_local_port),
5362                             ntohs(sel.ips_remote_port));
5363 #endif
5364                 /* Success so far! */
5365         }
5366         rw_enter(&polhead->iph_lock, RW_READER);
5367         pol = ipsec_find_policy_head(NULL, polhead, IPSEC_TYPE_OUTBOUND, &sel);
5368         rw_exit(&polhead->iph_lock);
5369         if (pol == NULL) {
5370                 /*
5371                  * No matching policy on this tunnel, drop the packet.
5372                  *
5373                  * NOTE:  Tunnel-mode tunnels are different from the
5374                  * IP global transport mode policy head.  For a tunnel-mode
5375                  * tunnel, we drop the packet in lieu of passing it
5376                  * along accepted the way a global-policy miss would.
5377                  *
5378                  * NOTE2:  "negotiate transport" tunnels should match ALL
5379                  * inbound packets, but we do not uncomment the ASSERT()
5380                  * below because if/when we open PF_POLICY, a user can
5381                  * shoot themself in the foot with a 0 priority.
5382                  */
5383 
5384                 /* ASSERT(itp->itp_flags & ITPF_P_TUNNEL); */
5385 #ifdef FRAGCACHE_DEBUG
5386                 cmn_err(CE_WARN, "ipsec_tun_outbound(): No matching tunnel "
5387                     "per-port policy\n");
5388 #endif
5389                 ip_drop_packet_chain(mp, B_FALSE, NULL,
5390                     DROPPER(ipss, ipds_spd_explicit),
5391                     &ipss->ipsec_spd_dropper);
5392                 return (NULL);
5393         }
5394 
5395 #ifdef FRAGCACHE_DEBUG
5396         cmn_err(CE_WARN, "Having matching tunnel per-port policy\n");
5397 #endif
5398 
5399         /*
5400          * NOTE: ixa_cleanup() function will release pol references.
5401          */
5402         ixa->ixa_ipsec_policy = pol;
5403         /*
5404          * NOTE: There is a subtle difference between iptun_zoneid and
5405          * iptun_connp->conn_zoneid explained in iptun_conn_create().  When
5406          * interacting with the ip module, we must use conn_zoneid.
5407          */
5408         ixa->ixa_zoneid = iptun->iptun_connp->conn_zoneid;
5409 
5410         ASSERT((outer_ipv4 != NULL) ? (ixa->ixa_flags & IXAF_IS_IPV4) :
5411             !(ixa->ixa_flags & IXAF_IS_IPV4));
5412         ASSERT(ixa->ixa_ipsec_policy != NULL);
5413         ixa->ixa_flags |= IXAF_IPSEC_SECURE;
5414 
5415         if (!(itp->itp_flags & ITPF_P_TUNNEL)) {
5416                 /* Set up transport mode for tunnelled packets. */
5417                 ixa->ixa_ipsec_proto = (inner_ipv4 != NULL) ? IPPROTO_ENCAP :
5418                     IPPROTO_IPV6;
5419                 return (mp);
5420         }
5421 
5422         /* Fill in tunnel-mode goodies here. */
5423         ixa->ixa_flags |= IXAF_IPSEC_TUNNEL;
5424         /* XXX Do I need to fill in all of the goodies here? */
5425         if (inner_ipv4) {
5426                 ixa->ixa_ipsec_inaf = AF_INET;
5427                 ixa->ixa_ipsec_insrc[0] =
5428                     pol->ipsp_sel->ipsl_key.ipsl_local.ipsad_v4;
5429                 ixa->ixa_ipsec_indst[0] =
5430                     pol->ipsp_sel->ipsl_key.ipsl_remote.ipsad_v4;
5431         } else {
5432                 ixa->ixa_ipsec_inaf = AF_INET6;
5433                 ixa->ixa_ipsec_insrc[0] =
5434                     pol->ipsp_sel->ipsl_key.ipsl_local.ipsad_v6.s6_addr32[0];
5435                 ixa->ixa_ipsec_insrc[1] =
5436                     pol->ipsp_sel->ipsl_key.ipsl_local.ipsad_v6.s6_addr32[1];
5437                 ixa->ixa_ipsec_insrc[2] =
5438                     pol->ipsp_sel->ipsl_key.ipsl_local.ipsad_v6.s6_addr32[2];
5439                 ixa->ixa_ipsec_insrc[3] =
5440                     pol->ipsp_sel->ipsl_key.ipsl_local.ipsad_v6.s6_addr32[3];
5441                 ixa->ixa_ipsec_indst[0] =
5442                     pol->ipsp_sel->ipsl_key.ipsl_remote.ipsad_v6.s6_addr32[0];
5443                 ixa->ixa_ipsec_indst[1] =
5444                     pol->ipsp_sel->ipsl_key.ipsl_remote.ipsad_v6.s6_addr32[1];
5445                 ixa->ixa_ipsec_indst[2] =
5446                     pol->ipsp_sel->ipsl_key.ipsl_remote.ipsad_v6.s6_addr32[2];
5447                 ixa->ixa_ipsec_indst[3] =
5448                     pol->ipsp_sel->ipsl_key.ipsl_remote.ipsad_v6.s6_addr32[3];
5449         }
5450         ixa->ixa_ipsec_insrcpfx = pol->ipsp_sel->ipsl_key.ipsl_local_pfxlen;
5451         ixa->ixa_ipsec_indstpfx = pol->ipsp_sel->ipsl_key.ipsl_remote_pfxlen;
5452         /* NOTE:  These are used for transport mode too. */
5453         ixa->ixa_ipsec_src_port = pol->ipsp_sel->ipsl_key.ipsl_lport;
5454         ixa->ixa_ipsec_dst_port = pol->ipsp_sel->ipsl_key.ipsl_rport;
5455         ixa->ixa_ipsec_proto = pol->ipsp_sel->ipsl_key.ipsl_proto;
5456 
5457         return (mp);
5458 }
5459 
5460 /*
5461  * NOTE: The following releases pol's reference and
5462  * calls ip_drop_packet() for me on NULL returns.
5463  */
5464 mblk_t *
5465 ipsec_check_ipsecin_policy_reasm(mblk_t *attr_mp, ipsec_policy_t *pol,
5466     ipha_t *inner_ipv4, ip6_t *inner_ipv6, uint64_t pkt_unique, netstack_t *ns)
5467 {
5468         /* Assume attr_mp is a chain of b_next-linked ip_recv_attr mblk. */
5469         mblk_t *data_chain = NULL, *data_tail = NULL;
5470         mblk_t *next;
5471         mblk_t *data_mp;
5472         ip_recv_attr_t  iras;
5473 
5474         while (attr_mp != NULL) {
5475                 ASSERT(ip_recv_attr_is_mblk(attr_mp));
5476                 next = attr_mp->b_next;
5477                 attr_mp->b_next = NULL;  /* No tripping asserts. */
5478 
5479                 data_mp = attr_mp->b_cont;
5480                 attr_mp->b_cont = NULL;
5481                 if (!ip_recv_attr_from_mblk(attr_mp, &iras)) {
5482                         /* The ill or ip_stack_t disappeared on us */
5483                         freemsg(data_mp);       /* ip_drop_packet?? */
5484                         ira_cleanup(&iras, B_TRUE);
5485                         goto fail;
5486                 }
5487 
5488                 /*
5489                  * Need IPPOL_REFHOLD(pol) for extras because
5490                  * ipsecin_policy does the refrele.
5491                  */
5492                 IPPOL_REFHOLD(pol);
5493 
5494                 data_mp = ipsec_check_ipsecin_policy(data_mp, pol, inner_ipv4,
5495                     inner_ipv6, pkt_unique, &iras, ns);
5496                 ira_cleanup(&iras, B_TRUE);
5497 
5498                 if (data_mp == NULL)
5499                         goto fail;
5500 
5501                 if (data_tail == NULL) {
5502                         /* First one */
5503                         data_chain = data_tail = data_mp;
5504                 } else {
5505                         data_tail->b_next = data_mp;
5506                         data_tail = data_mp;
5507                 }
5508                 attr_mp = next;
5509         }
5510         /*
5511          * One last release because either the loop bumped it up, or we never
5512          * called ipsec_check_ipsecin_policy().
5513          */
5514         IPPOL_REFRELE(pol);
5515 
5516         /* data_chain is ready for return to tun module. */
5517         return (data_chain);
5518 
5519 fail:
5520         /*
5521          * Need to get rid of any extra pol
5522          * references, and any remaining bits as well.
5523          */
5524         IPPOL_REFRELE(pol);
5525         ipsec_freemsg_chain(data_chain);
5526         ipsec_freemsg_chain(next);      /* ipdrop stats? */
5527         return (NULL);
5528 }
5529 
5530 /*
5531  * Return a message if the inbound packet passed an IPsec policy check.  Returns
5532  * NULL if it failed or if it is a fragment needing its friends before a
5533  * policy check can be performed.
5534  *
5535  * Expects a non-NULL data_mp, and a non-NULL polhead.
5536  * The returned mblk may be a b_next chain of packets if fragments
5537  * neeeded to be collected for a proper policy check.
5538  *
5539  * This function calls ip_drop_packet() on data_mp if need be.
5540  *
5541  * NOTE:  outer_hdr_len is signed.  If it's a negative value, the caller
5542  * is inspecting an ICMP packet.
5543  */
5544 mblk_t *
5545 ipsec_tun_inbound(ip_recv_attr_t *ira, mblk_t *data_mp, ipsec_tun_pol_t *itp,
5546     ipha_t *inner_ipv4, ip6_t *inner_ipv6, ipha_t *outer_ipv4,
5547     ip6_t *outer_ipv6, int outer_hdr_len, netstack_t *ns)
5548 {
5549         ipsec_policy_head_t *polhead;
5550         ipsec_selector_t sel;
5551         ipsec_policy_t *pol;
5552         uint16_t tmpport;
5553         selret_t rc;
5554         boolean_t port_policy_present, is_icmp, global_present;
5555         in6_addr_t tmpaddr;
5556         ipaddr_t tmp4;
5557         uint8_t flags, *inner_hdr;
5558         ipsec_stack_t *ipss = ns->netstack_ipsec;
5559 
5560         sel.ips_is_icmp_inv_acq = 0;
5561 
5562         if (outer_ipv4 != NULL) {
5563                 ASSERT(outer_ipv6 == NULL);
5564                 global_present = ipss->ipsec_inbound_v4_policy_present;
5565         } else {
5566                 ASSERT(outer_ipv6 != NULL);
5567                 global_present = ipss->ipsec_inbound_v6_policy_present;
5568         }
5569 
5570         ASSERT(inner_ipv4 != NULL && inner_ipv6 == NULL ||
5571             inner_ipv4 == NULL && inner_ipv6 != NULL);
5572 
5573         if (outer_hdr_len < 0) {
5574                 outer_hdr_len = (-outer_hdr_len);
5575                 is_icmp = B_TRUE;
5576         } else {
5577                 is_icmp = B_FALSE;
5578         }
5579 
5580         if (itp != NULL && (itp->itp_flags & ITPF_P_ACTIVE)) {
5581                 mblk_t *mp = data_mp;
5582 
5583                 polhead = itp->itp_policy;
5584                 /*
5585                  * We need to perform full Tunnel-Mode enforcement,
5586                  * and we need to have inner-header data for such enforcement.
5587                  *
5588                  * See ipsec_init_inbound_sel() for the 0x80000000 on inbound
5589                  * and on return.
5590                  */
5591 
5592                 port_policy_present = ((itp->itp_flags &
5593                     ITPF_P_PER_PORT_SECURITY) ? B_TRUE : B_FALSE);
5594                 /*
5595                  * NOTE:  Even if our policy is transport mode, set the
5596                  * SEL_TUNNEL_MODE flag so ipsec_init_inbound_sel() can
5597                  * do the right thing w.r.t. outer headers.
5598                  */
5599                 flags = ((port_policy_present ? SEL_PORT_POLICY : SEL_NONE) |
5600                     (is_icmp ? SEL_IS_ICMP : SEL_NONE) | SEL_TUNNEL_MODE);
5601 
5602                 rc = ipsec_init_inbound_sel(&sel, data_mp, inner_ipv4,
5603                     inner_ipv6, flags);
5604 
5605                 switch (rc) {
5606                 case SELRET_NOMEM:
5607                         ip_drop_packet(data_mp, B_TRUE, NULL,
5608                             DROPPER(ipss, ipds_spd_nomem),
5609                             &ipss->ipsec_spd_dropper);
5610                         return (NULL);
5611                 case SELRET_TUNFRAG:
5612                         /*
5613                          * At this point, if we're cleartext, we don't want
5614                          * to go there.
5615                          */
5616                         if (!(ira->ira_flags & IRAF_IPSEC_SECURE)) {
5617                                 ip_drop_packet(data_mp, B_TRUE, NULL,
5618                                     DROPPER(ipss, ipds_spd_got_clear),
5619                                     &ipss->ipsec_spd_dropper);
5620                                 return (NULL);
5621                         }
5622 
5623                         /*
5624                          * Inner and outer headers may not be contiguous.
5625                          * Pullup the data_mp now to satisfy assumptions of
5626                          * ipsec_fragcache_add()
5627                          */
5628                         if (data_mp->b_cont != NULL) {
5629                                 mblk_t *nmp;
5630 
5631                                 nmp = msgpullup(data_mp, -1);
5632                                 if (nmp == NULL) {
5633                                         ip_drop_packet(data_mp, B_TRUE, NULL,
5634                                             DROPPER(ipss, ipds_spd_nomem),
5635                                             &ipss->ipsec_spd_dropper);
5636                                         return (NULL);
5637                                 }
5638                                 freemsg(data_mp);
5639                                 data_mp = nmp;
5640                                 if (outer_ipv4 != NULL)
5641                                         outer_ipv4 =
5642                                             (ipha_t *)data_mp->b_rptr;
5643                                 else
5644                                         outer_ipv6 =
5645                                             (ip6_t *)data_mp->b_rptr;
5646                                 if (inner_ipv4 != NULL) {
5647                                         inner_ipv4 =
5648                                             (ipha_t *)(data_mp->b_rptr +
5649                                             outer_hdr_len);
5650                                 } else {
5651                                         inner_ipv6 =
5652                                             (ip6_t *)(data_mp->b_rptr +
5653                                             outer_hdr_len);
5654                                 }
5655                         }
5656 
5657                         /*
5658                          * If we need to queue the packet. First we
5659                          * get an mblk with the attributes. ipsec_fragcache_add
5660                          * will prepend that to the queued data and return
5661                          * a list of b_next messages each of which starts with
5662                          * the attribute mblk.
5663                          */
5664                         mp = ip_recv_attr_to_mblk(ira);
5665                         if (mp == NULL) {
5666                                 ip_drop_packet(data_mp, B_TRUE, NULL,
5667                                     DROPPER(ipss, ipds_spd_nomem),
5668                                     &ipss->ipsec_spd_dropper);
5669                                 return (NULL);
5670                         }
5671 
5672                         mp = ipsec_fragcache_add(&itp->itp_fragcache,
5673                             mp, data_mp, outer_hdr_len, ipss);
5674 
5675                         if (mp == NULL) {
5676                                 /*
5677                                  * Data is cached, fragment chain is not
5678                                  * complete.
5679                                  */
5680                                 return (NULL);
5681                         }
5682 
5683                         /*
5684                          * If we get here, we have a full fragment chain.
5685                          * Reacquire headers and selectors from first fragment.
5686                          */
5687                         ASSERT(ip_recv_attr_is_mblk(mp));
5688                         data_mp = mp->b_cont;
5689                         inner_hdr = data_mp->b_rptr;
5690                         if (outer_ipv4 != NULL) {
5691                                 inner_hdr += IPH_HDR_LENGTH(
5692                                     (ipha_t *)data_mp->b_rptr);
5693                         } else {
5694                                 inner_hdr += ip_hdr_length_v6(data_mp,
5695                                     (ip6_t *)data_mp->b_rptr);
5696                         }
5697                         ASSERT(inner_hdr <= data_mp->b_wptr);
5698 
5699                         if (inner_ipv4 != NULL) {
5700                                 inner_ipv4 = (ipha_t *)inner_hdr;
5701                                 inner_ipv6 = NULL;
5702                         } else {
5703                                 inner_ipv6 = (ip6_t *)inner_hdr;
5704                                 inner_ipv4 = NULL;
5705                         }
5706 
5707                         /*
5708                          * Use SEL_TUNNEL_MODE to take into account the outer
5709                          * header.  Use SEL_POST_FRAG so we always get ports.
5710                          */
5711                         rc = ipsec_init_inbound_sel(&sel, data_mp,
5712                             inner_ipv4, inner_ipv6,
5713                             SEL_TUNNEL_MODE | SEL_POST_FRAG);
5714                         switch (rc) {
5715                         case SELRET_SUCCESS:
5716                                 /*
5717                                  * Get to same place as first caller's
5718                                  * SELRET_SUCCESS case.
5719                                  */
5720                                 break;
5721                         case SELRET_NOMEM:
5722                                 ip_drop_packet_chain(mp, B_TRUE, NULL,
5723                                     DROPPER(ipss, ipds_spd_nomem),
5724                                     &ipss->ipsec_spd_dropper);
5725                                 return (NULL);
5726                         case SELRET_BADPKT:
5727                                 ip_drop_packet_chain(mp, B_TRUE, NULL,
5728                                     DROPPER(ipss, ipds_spd_malformed_frag),
5729                                     &ipss->ipsec_spd_dropper);
5730                                 return (NULL);
5731                         case SELRET_TUNFRAG:
5732                                 cmn_err(CE_WARN, "(TUNFRAG on 2nd call...)");
5733                                 /* FALLTHRU */
5734                         default:
5735                                 cmn_err(CE_WARN, "ipsec_init_inbound_sel(mark2)"
5736                                     " returns bizarro 0x%x", rc);
5737                                 /* Guaranteed panic! */
5738                                 ASSERT(rc == SELRET_NOMEM);
5739                                 return (NULL);
5740                         }
5741                         /* FALLTHRU */
5742                 case SELRET_SUCCESS:
5743                         /*
5744                          * Common case:
5745                          * No per-port policy or a non-fragment.  Keep going.
5746                          */
5747                         break;
5748                 case SELRET_BADPKT:
5749                         /*
5750                          * We may receive ICMP (with IPv6 inner) packets that
5751                          * trigger this return value.  Send 'em in for
5752                          * enforcement checking.
5753                          */
5754                         cmn_err(CE_NOTE, "ipsec_tun_inbound(): "
5755                             "sending 'bad packet' in for enforcement");
5756                         break;
5757                 default:
5758                         cmn_err(CE_WARN,
5759                             "ipsec_init_inbound_sel() returns bizarro 0x%x",
5760                             rc);
5761                         ASSERT(rc == SELRET_NOMEM);     /* Guaranteed panic! */
5762                         return (NULL);
5763                 }
5764 
5765                 if (is_icmp) {
5766                         /*
5767                          * Swap local/remote because this is an ICMP packet.
5768                          */
5769                         tmpaddr = sel.ips_local_addr_v6;
5770                         sel.ips_local_addr_v6 = sel.ips_remote_addr_v6;
5771                         sel.ips_remote_addr_v6 = tmpaddr;
5772                         tmpport = sel.ips_local_port;
5773                         sel.ips_local_port = sel.ips_remote_port;
5774                         sel.ips_remote_port = tmpport;
5775                 }
5776 
5777                 /* find_policy_head() */
5778                 rw_enter(&polhead->iph_lock, RW_READER);
5779                 pol = ipsec_find_policy_head(NULL, polhead, IPSEC_TYPE_INBOUND,
5780                     &sel);
5781                 rw_exit(&polhead->iph_lock);
5782                 if (pol != NULL) {
5783                         uint64_t pkt_unique;
5784 
5785                         if (!(ira->ira_flags & IRAF_IPSEC_SECURE)) {
5786                                 if (!pol->ipsp_act->ipa_allow_clear) {
5787                                         /*
5788                                          * XXX should never get here with
5789                                          * tunnel reassembled fragments?
5790                                          */
5791                                         ASSERT(mp == data_mp);
5792                                         ip_drop_packet(data_mp, B_TRUE, NULL,
5793                                             DROPPER(ipss, ipds_spd_got_clear),
5794                                             &ipss->ipsec_spd_dropper);
5795                                         IPPOL_REFRELE(pol);
5796                                         return (NULL);
5797                                 } else {
5798                                         IPPOL_REFRELE(pol);
5799                                         return (mp);
5800                                 }
5801                         }
5802                         pkt_unique = SA_UNIQUE_ID(sel.ips_remote_port,
5803                             sel.ips_local_port,
5804                             (inner_ipv4 == NULL) ? IPPROTO_IPV6 :
5805                             IPPROTO_ENCAP, sel.ips_protocol);
5806 
5807                         /*
5808                          * NOTE: The following releases pol's reference and
5809                          * calls ip_drop_packet() for me on NULL returns.
5810                          *
5811                          * "sel" is still good here, so let's use it!
5812                          */
5813                         if (data_mp == mp) {
5814                                 /* A single packet without attributes */
5815                                 data_mp = ipsec_check_ipsecin_policy(data_mp,
5816                                     pol, inner_ipv4, inner_ipv6, pkt_unique,
5817                                     ira, ns);
5818                         } else {
5819                                 /*
5820                                  * We pass in the b_next chain of attr_mp's
5821                                  * and get back a b_next chain of data_mp's.
5822                                  */
5823                                 data_mp = ipsec_check_ipsecin_policy_reasm(mp,
5824                                     pol, inner_ipv4, inner_ipv6, pkt_unique,
5825                                     ns);
5826                         }
5827                         return (data_mp);
5828                 }
5829 
5830                 /*
5831                  * Else fallthru and check the global policy on the outer
5832                  * header(s) if this tunnel is an old-style transport-mode
5833                  * one.  Drop the packet explicitly (no policy entry) for
5834                  * a new-style tunnel-mode tunnel.
5835                  */
5836                 if ((itp->itp_flags & ITPF_P_TUNNEL) && !is_icmp) {
5837                         ip_drop_packet_chain(data_mp, B_TRUE, NULL,
5838                             DROPPER(ipss, ipds_spd_explicit),
5839                             &ipss->ipsec_spd_dropper);
5840                         return (NULL);
5841                 }
5842         }
5843 
5844         /*
5845          * NOTE:  If we reach here, we will not have packet chains from
5846          * fragcache_add(), because the only way I get chains is on a
5847          * tunnel-mode tunnel, which either returns with a pass, or gets
5848          * hit by the ip_drop_packet_chain() call right above here.
5849          */
5850         ASSERT(data_mp->b_next == NULL);
5851 
5852         /* If no per-tunnel security, check global policy now. */
5853         if ((ira->ira_flags & IRAF_IPSEC_SECURE) && !global_present) {
5854                 if (ira->ira_flags & IRAF_TRUSTED_ICMP) {
5855                         /*
5856                          * This is an ICMP message that was geenrated locally.
5857                          * We should accept it.
5858                          */
5859                         return (data_mp);
5860                 }
5861 
5862                 ip_drop_packet(data_mp, B_TRUE, NULL,
5863                     DROPPER(ipss, ipds_spd_got_secure),
5864                     &ipss->ipsec_spd_dropper);
5865                 return (NULL);
5866         }
5867 
5868         if (is_icmp) {
5869                 /*
5870                  * For ICMP packets, "outer_ipvN" is set to the outer header
5871                  * that is *INSIDE* the ICMP payload.  For global policy
5872                  * checking, we need to reverse src/dst on the payload in
5873                  * order to construct selectors appropriately.  See "ripha"
5874                  * constructions in ip.c.  To avoid a bug like 6478464 (see
5875                  * earlier in this file), we will actually exchange src/dst
5876                  * in the packet, and reverse if after the call to
5877                  * ipsec_check_global_policy().
5878                  */
5879                 if (outer_ipv4 != NULL) {
5880                         tmp4 = outer_ipv4->ipha_src;
5881                         outer_ipv4->ipha_src = outer_ipv4->ipha_dst;
5882                         outer_ipv4->ipha_dst = tmp4;
5883                 } else {
5884                         ASSERT(outer_ipv6 != NULL);
5885                         tmpaddr = outer_ipv6->ip6_src;
5886                         outer_ipv6->ip6_src = outer_ipv6->ip6_dst;
5887                         outer_ipv6->ip6_dst = tmpaddr;
5888                 }
5889         }
5890 
5891         data_mp = ipsec_check_global_policy(data_mp, NULL, outer_ipv4,
5892             outer_ipv6, ira, ns);
5893         if (data_mp == NULL)
5894                 return (NULL);
5895 
5896         if (is_icmp) {
5897                 /* Set things back to normal. */
5898                 if (outer_ipv4 != NULL) {
5899                         tmp4 = outer_ipv4->ipha_src;
5900                         outer_ipv4->ipha_src = outer_ipv4->ipha_dst;
5901                         outer_ipv4->ipha_dst = tmp4;
5902                 } else {
5903                         /* No need for ASSERT()s now. */
5904                         tmpaddr = outer_ipv6->ip6_src;
5905                         outer_ipv6->ip6_src = outer_ipv6->ip6_dst;
5906                         outer_ipv6->ip6_dst = tmpaddr;
5907                 }
5908         }
5909 
5910         /*
5911          * At this point, we pretend it's a cleartext accepted
5912          * packet.
5913          */
5914         return (data_mp);
5915 }
5916 
5917 /*
5918  * AVL comparison routine for our list of tunnel polheads.
5919  */
5920 static int
5921 tunnel_compare(const void *arg1, const void *arg2)
5922 {
5923         ipsec_tun_pol_t *left, *right;
5924         int rc;
5925 
5926         left = (ipsec_tun_pol_t *)arg1;
5927         right = (ipsec_tun_pol_t *)arg2;
5928 
5929         rc = strncmp(left->itp_name, right->itp_name, LIFNAMSIZ);
5930         return (rc == 0 ? rc : (rc > 0 ? 1 : -1));
5931 }
5932 
5933 /*
5934  * Free a tunnel policy node.
5935  */
5936 void
5937 itp_free(ipsec_tun_pol_t *node, netstack_t *ns)
5938 {
5939         if (node->itp_policy != NULL) {
5940                 IPPH_REFRELE(node->itp_policy, ns);
5941                 node->itp_policy = NULL;
5942         }
5943         if (node->itp_inactive != NULL) {
5944                 IPPH_REFRELE(node->itp_inactive, ns);
5945                 node->itp_inactive = NULL;
5946         }
5947         mutex_destroy(&node->itp_lock);
5948         kmem_free(node, sizeof (*node));
5949 }
5950 
5951 void
5952 itp_unlink(ipsec_tun_pol_t *node, netstack_t *ns)
5953 {
5954         ipsec_stack_t *ipss = ns->netstack_ipsec;
5955 
5956         rw_enter(&ipss->ipsec_tunnel_policy_lock, RW_WRITER);
5957         ipss->ipsec_tunnel_policy_gen++;
5958         ipsec_fragcache_uninit(&node->itp_fragcache, ipss);
5959         avl_remove(&ipss->ipsec_tunnel_policies, node);
5960         rw_exit(&ipss->ipsec_tunnel_policy_lock);
5961         ITP_REFRELE(node, ns);
5962 }
5963 
5964 /*
5965  * Public interface to look up a tunnel security policy by name.  Used by
5966  * spdsock mostly.  Returns "node" with a bumped refcnt.
5967  */
5968 ipsec_tun_pol_t *
5969 get_tunnel_policy(char *name, netstack_t *ns)
5970 {
5971         ipsec_tun_pol_t *node, lookup;
5972         ipsec_stack_t *ipss = ns->netstack_ipsec;
5973 
5974         (void) strncpy(lookup.itp_name, name, LIFNAMSIZ);
5975 
5976         rw_enter(&ipss->ipsec_tunnel_policy_lock, RW_READER);
5977         node = (ipsec_tun_pol_t *)avl_find(&ipss->ipsec_tunnel_policies,
5978             &lookup, NULL);
5979         if (node != NULL) {
5980                 ITP_REFHOLD(node);
5981         }
5982         rw_exit(&ipss->ipsec_tunnel_policy_lock);
5983 
5984         return (node);
5985 }
5986 
5987 /*
5988  * Public interface to walk all tunnel security polcies.  Useful for spdsock
5989  * DUMP operations.  iterator() will not consume a reference.
5990  */
5991 void
5992 itp_walk(void (*iterator)(ipsec_tun_pol_t *, void *, netstack_t *),
5993     void *arg, netstack_t *ns)
5994 {
5995         ipsec_tun_pol_t *node;
5996         ipsec_stack_t *ipss = ns->netstack_ipsec;
5997 
5998         rw_enter(&ipss->ipsec_tunnel_policy_lock, RW_READER);
5999         for (node = avl_first(&ipss->ipsec_tunnel_policies); node != NULL;
6000             node = AVL_NEXT(&ipss->ipsec_tunnel_policies, node)) {
6001                 iterator(node, arg, ns);
6002         }
6003         rw_exit(&ipss->ipsec_tunnel_policy_lock);
6004 }
6005 
6006 /*
6007  * Initialize policy head.  This can only fail if there's a memory problem.
6008  */
6009 static boolean_t
6010 tunnel_polhead_init(ipsec_policy_head_t *iph, netstack_t *ns)
6011 {
6012         ipsec_stack_t *ipss = ns->netstack_ipsec;
6013 
6014         rw_init(&iph->iph_lock, NULL, RW_DEFAULT, NULL);
6015         iph->iph_refs = 1;
6016         iph->iph_gen = 0;
6017         if (ipsec_alloc_table(iph, ipss->ipsec_tun_spd_hashsize,
6018             KM_SLEEP, B_FALSE, ns) != 0) {
6019                 ipsec_polhead_free_table(iph);
6020                 return (B_FALSE);
6021         }
6022         ipsec_polhead_init(iph, ipss->ipsec_tun_spd_hashsize);
6023         return (B_TRUE);
6024 }
6025 
6026 /*
6027  * Create a tunnel policy node with "name".  Set errno with
6028  * ENOMEM if there's a memory problem, and EEXIST if there's an existing
6029  * node.
6030  */
6031 ipsec_tun_pol_t *
6032 create_tunnel_policy(char *name, int *errno, uint64_t *gen, netstack_t *ns)
6033 {
6034         ipsec_tun_pol_t *newbie, *existing;
6035         avl_index_t where;
6036         ipsec_stack_t *ipss = ns->netstack_ipsec;
6037 
6038         newbie = kmem_zalloc(sizeof (*newbie), KM_NOSLEEP);
6039         if (newbie == NULL) {
6040                 *errno = ENOMEM;
6041                 return (NULL);
6042         }
6043         if (!ipsec_fragcache_init(&newbie->itp_fragcache)) {
6044                 kmem_free(newbie, sizeof (*newbie));
6045                 *errno = ENOMEM;
6046                 return (NULL);
6047         }
6048 
6049         (void) strncpy(newbie->itp_name, name, LIFNAMSIZ);
6050 
6051         rw_enter(&ipss->ipsec_tunnel_policy_lock, RW_WRITER);
6052         existing = (ipsec_tun_pol_t *)avl_find(&ipss->ipsec_tunnel_policies,
6053             newbie, &where);
6054         if (existing != NULL) {
6055                 itp_free(newbie, ns);
6056                 *errno = EEXIST;
6057                 rw_exit(&ipss->ipsec_tunnel_policy_lock);
6058                 return (NULL);
6059         }
6060         ipss->ipsec_tunnel_policy_gen++;
6061         *gen = ipss->ipsec_tunnel_policy_gen;
6062         newbie->itp_refcnt = 2;      /* One for the caller, one for the tree. */
6063         newbie->itp_next_policy_index = 1;
6064         avl_insert(&ipss->ipsec_tunnel_policies, newbie, where);
6065         mutex_init(&newbie->itp_lock, NULL, MUTEX_DEFAULT, NULL);
6066         newbie->itp_policy = kmem_zalloc(sizeof (ipsec_policy_head_t),
6067             KM_NOSLEEP);
6068         if (newbie->itp_policy == NULL)
6069                 goto nomem;
6070         newbie->itp_inactive = kmem_zalloc(sizeof (ipsec_policy_head_t),
6071             KM_NOSLEEP);
6072         if (newbie->itp_inactive == NULL) {
6073                 kmem_free(newbie->itp_policy, sizeof (ipsec_policy_head_t));
6074                 goto nomem;
6075         }
6076 
6077         if (!tunnel_polhead_init(newbie->itp_policy, ns)) {
6078                 kmem_free(newbie->itp_policy, sizeof (ipsec_policy_head_t));
6079                 kmem_free(newbie->itp_inactive, sizeof (ipsec_policy_head_t));
6080                 goto nomem;
6081         } else if (!tunnel_polhead_init(newbie->itp_inactive, ns)) {
6082                 IPPH_REFRELE(newbie->itp_policy, ns);
6083                 kmem_free(newbie->itp_inactive, sizeof (ipsec_policy_head_t));
6084                 goto nomem;
6085         }
6086         rw_exit(&ipss->ipsec_tunnel_policy_lock);
6087 
6088         return (newbie);
6089 nomem:
6090         *errno = ENOMEM;
6091         kmem_free(newbie, sizeof (*newbie));
6092         return (NULL);
6093 }
6094 
6095 /*
6096  * Given two addresses, find a tunnel instance's IPsec policy heads.
6097  * Returns NULL on failure.
6098  */
6099 ipsec_tun_pol_t *
6100 itp_get_byaddr(uint32_t *laddr, uint32_t *faddr, int af, ip_stack_t *ipst)
6101 {
6102         conn_t *connp;
6103         iptun_t *iptun;
6104         ipsec_tun_pol_t *itp = NULL;
6105 
6106         /* Classifiers are used to "src" being foreign. */
6107         if (af == AF_INET) {
6108                 connp = ipcl_iptun_classify_v4((ipaddr_t *)faddr,
6109                     (ipaddr_t *)laddr, ipst);
6110         } else {
6111                 ASSERT(af == AF_INET6);
6112                 ASSERT(!IN6_IS_ADDR_V4MAPPED((in6_addr_t *)laddr));
6113                 ASSERT(!IN6_IS_ADDR_V4MAPPED((in6_addr_t *)faddr));
6114                 connp = ipcl_iptun_classify_v6((in6_addr_t *)faddr,
6115                     (in6_addr_t *)laddr, ipst);
6116         }
6117 
6118         if (connp == NULL)
6119                 return (NULL);
6120 
6121         if (IPCL_IS_IPTUN(connp)) {
6122                 iptun = connp->conn_iptun;
6123                 if (iptun != NULL) {
6124                         itp = iptun->iptun_itp;
6125                         if (itp != NULL) {
6126                                 /* Braces due to the macro's nature... */
6127                                 ITP_REFHOLD(itp);
6128                         }
6129                 }  /* Else itp is already NULL. */
6130         }
6131 
6132         CONN_DEC_REF(connp);
6133         return (itp);
6134 }
6135 
6136 /*
6137  * Frag cache code, based on SunScreen 3.2 source
6138  *      screen/kernel/common/screen_fragcache.c
6139  */
6140 
6141 #define IPSEC_FRAG_TTL_MAX      5
6142 /*
6143  * Note that the following parameters create 256 hash buckets
6144  * with 1024 free entries to be distributed.  Things are cleaned
6145  * periodically and are attempted to be cleaned when there is no
6146  * free space, but this system errs on the side of dropping packets
6147  * over creating memory exhaustion.  We may decide to make hash
6148  * factor a tunable if this proves to be a bad decision.
6149  */
6150 #define IPSEC_FRAG_HASH_SLOTS   (1<<8)
6151 #define IPSEC_FRAG_HASH_FACTOR  4
6152 #define IPSEC_FRAG_HASH_SIZE    (IPSEC_FRAG_HASH_SLOTS * IPSEC_FRAG_HASH_FACTOR)
6153 
6154 #define IPSEC_FRAG_HASH_MASK            (IPSEC_FRAG_HASH_SLOTS - 1)
6155 #define IPSEC_FRAG_HASH_FUNC(id)        (((id) & IPSEC_FRAG_HASH_MASK) ^ \
6156                                             (((id) / \
6157                                             (ushort_t)IPSEC_FRAG_HASH_SLOTS) & \
6158                                             IPSEC_FRAG_HASH_MASK))
6159 
6160 /* Maximum fragments per packet.  48 bytes payload x 1366 packets > 64KB */
6161 #define IPSEC_MAX_FRAGS         1366
6162 
6163 #define V4_FRAG_OFFSET(ipha) ((ntohs(ipha->ipha_fragment_offset_and_flags) & \
6164                                     IPH_OFFSET) << 3)
6165 #define V4_MORE_FRAGS(ipha) (ntohs(ipha->ipha_fragment_offset_and_flags) & \
6166                 IPH_MF)
6167 
6168 /*
6169  * Initialize an ipsec fragcache instance.
6170  * Returns B_FALSE if memory allocation fails.
6171  */
6172 boolean_t
6173 ipsec_fragcache_init(ipsec_fragcache_t *frag)
6174 {
6175         ipsec_fragcache_entry_t *ftemp;
6176         int i;
6177 
6178         mutex_init(&frag->itpf_lock, NULL, MUTEX_DEFAULT, NULL);
6179         frag->itpf_ptr = (ipsec_fragcache_entry_t **)
6180             kmem_zalloc(sizeof (ipsec_fragcache_entry_t *) *
6181             IPSEC_FRAG_HASH_SLOTS, KM_NOSLEEP);
6182         if (frag->itpf_ptr == NULL)
6183                 return (B_FALSE);
6184 
6185         ftemp = (ipsec_fragcache_entry_t *)
6186             kmem_zalloc(sizeof (ipsec_fragcache_entry_t) *
6187             IPSEC_FRAG_HASH_SIZE, KM_NOSLEEP);
6188         if (ftemp == NULL) {
6189                 kmem_free(frag->itpf_ptr, sizeof (ipsec_fragcache_entry_t *) *
6190                     IPSEC_FRAG_HASH_SLOTS);
6191                 return (B_FALSE);
6192         }
6193 
6194         frag->itpf_freelist = NULL;
6195 
6196         for (i = 0; i < IPSEC_FRAG_HASH_SIZE; i++) {
6197                 ftemp->itpfe_next = frag->itpf_freelist;
6198                 frag->itpf_freelist = ftemp;
6199                 ftemp++;
6200         }
6201 
6202         frag->itpf_expire_hint = 0;
6203 
6204         return (B_TRUE);
6205 }
6206 
6207 void
6208 ipsec_fragcache_uninit(ipsec_fragcache_t *frag, ipsec_stack_t *ipss)
6209 {
6210         ipsec_fragcache_entry_t *fep;
6211         int i;
6212 
6213         mutex_enter(&frag->itpf_lock);
6214         if (frag->itpf_ptr) {
6215                 /* Delete any existing fragcache entry chains */
6216                 for (i = 0; i < IPSEC_FRAG_HASH_SLOTS; i++) {
6217                         fep = (frag->itpf_ptr)[i];
6218                         while (fep != NULL) {
6219                                 /* Returned fep is next in chain or NULL */
6220                                 fep = fragcache_delentry(i, fep, frag, ipss);
6221                         }
6222                 }
6223                 /*
6224                  * Chase the pointers back to the beginning
6225                  * of the memory allocation and then
6226                  * get rid of the allocated freelist
6227                  */
6228                 while (frag->itpf_freelist->itpfe_next != NULL)
6229                         frag->itpf_freelist = frag->itpf_freelist->itpfe_next;
6230                 /*
6231                  * XXX - If we ever dynamically grow the freelist
6232                  * then we'll have to free entries individually
6233                  * or determine how many entries or chunks we have
6234                  * grown since the initial allocation.
6235                  */
6236                 kmem_free(frag->itpf_freelist,
6237                     sizeof (ipsec_fragcache_entry_t) *
6238                     IPSEC_FRAG_HASH_SIZE);
6239                 /* Free the fragcache structure */
6240                 kmem_free(frag->itpf_ptr,
6241                     sizeof (ipsec_fragcache_entry_t *) *
6242                     IPSEC_FRAG_HASH_SLOTS);
6243         }
6244         mutex_exit(&frag->itpf_lock);
6245         mutex_destroy(&frag->itpf_lock);
6246 }
6247 
6248 /*
6249  * Add a fragment to the fragment cache.   Consumes mp if NULL is returned.
6250  * Returns mp if a whole fragment has been assembled, NULL otherwise
6251  * The returned mp could be a b_next chain of fragments.
6252  *
6253  * The iramp argument is set on inbound; NULL if outbound.
6254  */
6255 mblk_t *
6256 ipsec_fragcache_add(ipsec_fragcache_t *frag, mblk_t *iramp, mblk_t *mp,
6257     int outer_hdr_len, ipsec_stack_t *ipss)
6258 {
6259         boolean_t is_v4;
6260         time_t itpf_time;
6261         ipha_t *iph;
6262         ipha_t *oiph;
6263         ip6_t *ip6h = NULL;
6264         uint8_t v6_proto;
6265         uint8_t *v6_proto_p;
6266         uint16_t ip6_hdr_length;
6267         ip_pkt_t ipp;
6268         ip6_frag_t *fraghdr;
6269         ipsec_fragcache_entry_t *fep;
6270         int i;
6271         mblk_t *nmp, *prevmp;
6272         int firstbyte, lastbyte;
6273         int offset;
6274         int last;
6275         boolean_t inbound = (iramp != NULL);
6276 
6277 #ifdef FRAGCACHE_DEBUG
6278         cmn_err(CE_WARN, "Fragcache: %s\n", inbound ? "INBOUND" : "OUTBOUND");
6279 #endif
6280         /*
6281          * You're on the slow path, so insure that every packet in the
6282          * cache is a single-mblk one.
6283          */
6284         if (mp->b_cont != NULL) {
6285                 nmp = msgpullup(mp, -1);
6286                 if (nmp == NULL) {
6287                         ip_drop_packet(mp, inbound, NULL,
6288                             DROPPER(ipss, ipds_spd_nomem),
6289                             &ipss->ipsec_spd_dropper);
6290                         if (inbound)
6291                                 (void) ip_recv_attr_free_mblk(iramp);
6292                         return (NULL);
6293                 }
6294                 freemsg(mp);
6295                 mp = nmp;
6296         }
6297 
6298         mutex_enter(&frag->itpf_lock);
6299 
6300         oiph  = (ipha_t *)mp->b_rptr;
6301         iph  = (ipha_t *)(mp->b_rptr + outer_hdr_len);
6302 
6303         if (IPH_HDR_VERSION(iph) == IPV4_VERSION) {
6304                 is_v4 = B_TRUE;
6305         } else {
6306                 ASSERT(IPH_HDR_VERSION(iph) == IPV6_VERSION);
6307                 ip6h = (ip6_t *)(mp->b_rptr + outer_hdr_len);
6308 
6309                 if (!ip_hdr_length_nexthdr_v6(mp, ip6h, &ip6_hdr_length,
6310                     &v6_proto_p)) {
6311                         /*
6312                          * Find upper layer protocol.
6313                          * If it fails we have a malformed packet
6314                          */
6315                         mutex_exit(&frag->itpf_lock);
6316                         ip_drop_packet(mp, inbound, NULL,
6317                             DROPPER(ipss, ipds_spd_malformed_packet),
6318                             &ipss->ipsec_spd_dropper);
6319                         if (inbound)
6320                                 (void) ip_recv_attr_free_mblk(iramp);
6321                         return (NULL);
6322                 } else {
6323                         v6_proto = *v6_proto_p;
6324                 }
6325 
6326 
6327                 bzero(&ipp, sizeof (ipp));
6328                 (void) ip_find_hdr_v6(mp, ip6h, B_FALSE, &ipp, NULL);
6329                 if (!(ipp.ipp_fields & IPPF_FRAGHDR)) {
6330                         /*
6331                          * We think this is a fragment, but didn't find
6332                          * a fragment header.  Something is wrong.
6333                          */
6334                         mutex_exit(&frag->itpf_lock);
6335                         ip_drop_packet(mp, inbound, NULL,
6336                             DROPPER(ipss, ipds_spd_malformed_frag),
6337                             &ipss->ipsec_spd_dropper);
6338                         if (inbound)
6339                                 (void) ip_recv_attr_free_mblk(iramp);
6340                         return (NULL);
6341                 }
6342                 fraghdr = ipp.ipp_fraghdr;
6343                 is_v4 = B_FALSE;
6344         }
6345 
6346         /* Anything to cleanup? */
6347 
6348         /*
6349          * This cleanup call could be put in a timer loop
6350          * but it may actually be just as reasonable a decision to
6351          * leave it here.  The disadvantage is this only gets called when
6352          * frags are added.  The advantage is that it is not
6353          * susceptible to race conditions like a time-based cleanup
6354          * may be.
6355          */
6356         itpf_time = gethrestime_sec();
6357         if (itpf_time >= frag->itpf_expire_hint)
6358                 ipsec_fragcache_clean(frag, ipss);
6359 
6360         /* Lookup to see if there is an existing entry */
6361 
6362         if (is_v4)
6363                 i = IPSEC_FRAG_HASH_FUNC(iph->ipha_ident);
6364         else
6365                 i = IPSEC_FRAG_HASH_FUNC(fraghdr->ip6f_ident);
6366 
6367         for (fep = (frag->itpf_ptr)[i]; fep; fep = fep->itpfe_next) {
6368                 if (is_v4) {
6369                         ASSERT(iph != NULL);
6370                         if ((fep->itpfe_id == iph->ipha_ident) &&
6371                             (fep->itpfe_src == iph->ipha_src) &&
6372                             (fep->itpfe_dst == iph->ipha_dst) &&
6373                             (fep->itpfe_proto == iph->ipha_protocol))
6374                                 break;
6375                 } else {
6376                         ASSERT(fraghdr != NULL);
6377                         ASSERT(fep != NULL);
6378                         if ((fep->itpfe_id == fraghdr->ip6f_ident) &&
6379                             IN6_ARE_ADDR_EQUAL(&fep->itpfe_src6,
6380                             &ip6h->ip6_src) &&
6381                             IN6_ARE_ADDR_EQUAL(&fep->itpfe_dst6,
6382                             &ip6h->ip6_dst) && (fep->itpfe_proto == v6_proto))
6383                                 break;
6384                 }
6385         }
6386 
6387         if (is_v4) {
6388                 firstbyte = V4_FRAG_OFFSET(iph);
6389                 lastbyte  = firstbyte + ntohs(iph->ipha_length) -
6390                     IPH_HDR_LENGTH(iph);
6391                 last = (V4_MORE_FRAGS(iph) == 0);
6392 #ifdef FRAGCACHE_DEBUG
6393                 cmn_err(CE_WARN, "V4 fragcache: firstbyte = %d, lastbyte = %d, "
6394                     "is_last_frag = %d, id = %d, mp = %p\n", firstbyte,
6395                     lastbyte, last, iph->ipha_ident, mp);
6396 #endif
6397         } else {
6398                 firstbyte = ntohs(fraghdr->ip6f_offlg & IP6F_OFF_MASK);
6399                 lastbyte  = firstbyte + ntohs(ip6h->ip6_plen) +
6400                     sizeof (ip6_t) - ip6_hdr_length;
6401                 last = (fraghdr->ip6f_offlg & IP6F_MORE_FRAG) == 0;
6402 #ifdef FRAGCACHE_DEBUG
6403                 cmn_err(CE_WARN, "V6 fragcache: firstbyte = %d, lastbyte = %d, "
6404                     "is_last_frag = %d, id = %d, fraghdr = %p, mp = %p\n",
6405                     firstbyte, lastbyte, last, fraghdr->ip6f_ident, fraghdr,
6406                     mp);
6407 #endif
6408         }
6409 
6410         /* check for bogus fragments and delete the entry */
6411         if (firstbyte > 0 && firstbyte <= 8) {
6412                 if (fep != NULL)
6413                         (void) fragcache_delentry(i, fep, frag, ipss);
6414                 mutex_exit(&frag->itpf_lock);
6415                 ip_drop_packet(mp, inbound, NULL,
6416                     DROPPER(ipss, ipds_spd_malformed_frag),
6417                     &ipss->ipsec_spd_dropper);
6418                 if (inbound)
6419                         (void) ip_recv_attr_free_mblk(iramp);
6420                 return (NULL);
6421         }
6422 
6423         /* Not found, allocate a new entry */
6424         if (fep == NULL) {
6425                 if (frag->itpf_freelist == NULL) {
6426                         /* see if there is some space */
6427                         ipsec_fragcache_clean(frag, ipss);
6428                         if (frag->itpf_freelist == NULL) {
6429                                 mutex_exit(&frag->itpf_lock);
6430                                 ip_drop_packet(mp, inbound, NULL,
6431                                     DROPPER(ipss, ipds_spd_nomem),
6432                                     &ipss->ipsec_spd_dropper);
6433                                 if (inbound)
6434                                         (void) ip_recv_attr_free_mblk(iramp);
6435                                 return (NULL);
6436                         }
6437                 }
6438 
6439                 fep = frag->itpf_freelist;
6440                 frag->itpf_freelist = fep->itpfe_next;
6441 
6442                 if (is_v4) {
6443                         bcopy((caddr_t)&iph->ipha_src, (caddr_t)&fep->itpfe_src,
6444                             sizeof (struct in_addr));
6445                         bcopy((caddr_t)&iph->ipha_dst, (caddr_t)&fep->itpfe_dst,
6446                             sizeof (struct in_addr));
6447                         fep->itpfe_id = iph->ipha_ident;
6448                         fep->itpfe_proto = iph->ipha_protocol;
6449                         i = IPSEC_FRAG_HASH_FUNC(fep->itpfe_id);
6450                 } else {
6451                         bcopy((in6_addr_t *)&ip6h->ip6_src,
6452                             (in6_addr_t *)&fep->itpfe_src6,
6453                             sizeof (struct in6_addr));
6454                         bcopy((in6_addr_t *)&ip6h->ip6_dst,
6455                             (in6_addr_t *)&fep->itpfe_dst6,
6456                             sizeof (struct in6_addr));
6457                         fep->itpfe_id = fraghdr->ip6f_ident;
6458                         fep->itpfe_proto = v6_proto;
6459                         i = IPSEC_FRAG_HASH_FUNC(fep->itpfe_id);
6460                 }
6461                 itpf_time = gethrestime_sec();
6462                 fep->itpfe_exp = itpf_time + IPSEC_FRAG_TTL_MAX + 1;
6463                 fep->itpfe_last = 0;
6464                 fep->itpfe_fraglist = NULL;
6465                 fep->itpfe_depth = 0;
6466                 fep->itpfe_next = (frag->itpf_ptr)[i];
6467                 (frag->itpf_ptr)[i] = fep;
6468 
6469                 if (frag->itpf_expire_hint > fep->itpfe_exp)
6470                         frag->itpf_expire_hint = fep->itpfe_exp;
6471 
6472         }
6473 
6474         /* Insert it in the frag list */
6475         /* List is in order by starting offset of fragments */
6476 
6477         prevmp = NULL;
6478         for (nmp = fep->itpfe_fraglist; nmp; nmp = nmp->b_next) {
6479                 ipha_t *niph;
6480                 ipha_t *oniph;
6481                 ip6_t *nip6h;
6482                 ip_pkt_t nipp;
6483                 ip6_frag_t *nfraghdr;
6484                 uint16_t nip6_hdr_length;
6485                 uint8_t *nv6_proto_p;
6486                 int nfirstbyte, nlastbyte;
6487                 char *data, *ndata;
6488                 mblk_t *ndata_mp = (inbound ? nmp->b_cont : nmp);
6489                 int hdr_len;
6490 
6491                 oniph  = (ipha_t *)mp->b_rptr;
6492                 nip6h = NULL;
6493                 niph = NULL;
6494 
6495                 /*
6496                  * Determine outer header type and length and set
6497                  * pointers appropriately
6498                  */
6499 
6500                 if (IPH_HDR_VERSION(oniph) == IPV4_VERSION) {
6501                         hdr_len = ((outer_hdr_len != 0) ?
6502                             IPH_HDR_LENGTH(oiph) : 0);
6503                         niph = (ipha_t *)(ndata_mp->b_rptr + hdr_len);
6504                 } else {
6505                         ASSERT(IPH_HDR_VERSION(oniph) == IPV6_VERSION);
6506                         ASSERT(ndata_mp->b_cont == NULL);
6507                         nip6h = (ip6_t *)ndata_mp->b_rptr;
6508                         (void) ip_hdr_length_nexthdr_v6(ndata_mp, nip6h,
6509                             &nip6_hdr_length, &v6_proto_p);
6510                         hdr_len = ((outer_hdr_len != 0) ? nip6_hdr_length : 0);
6511                 }
6512 
6513                 /*
6514                  * Determine inner header type and length and set
6515                  * pointers appropriately
6516                  */
6517 
6518                 if (is_v4) {
6519                         if (niph == NULL) {
6520                                 /* Was v6 outer */
6521                                 niph = (ipha_t *)(ndata_mp->b_rptr + hdr_len);
6522                         }
6523                         nfirstbyte = V4_FRAG_OFFSET(niph);
6524                         nlastbyte = nfirstbyte + ntohs(niph->ipha_length) -
6525                             IPH_HDR_LENGTH(niph);
6526                 } else {
6527                         ASSERT(ndata_mp->b_cont == NULL);
6528                         nip6h = (ip6_t *)(ndata_mp->b_rptr + hdr_len);
6529                         if (!ip_hdr_length_nexthdr_v6(ndata_mp, nip6h,
6530                             &nip6_hdr_length, &nv6_proto_p)) {
6531                                 mutex_exit(&frag->itpf_lock);
6532                                 ip_drop_packet_chain(nmp, inbound, NULL,
6533                                     DROPPER(ipss, ipds_spd_malformed_frag),
6534                                     &ipss->ipsec_spd_dropper);
6535                                 ipsec_freemsg_chain(ndata_mp);
6536                                 if (inbound)
6537                                         (void) ip_recv_attr_free_mblk(iramp);
6538                                 return (NULL);
6539                         }
6540                         bzero(&nipp, sizeof (nipp));
6541                         (void) ip_find_hdr_v6(ndata_mp, nip6h, B_FALSE, &nipp,
6542                             NULL);
6543                         nfraghdr = nipp.ipp_fraghdr;
6544                         nfirstbyte = ntohs(nfraghdr->ip6f_offlg &
6545                             IP6F_OFF_MASK);
6546                         nlastbyte  = nfirstbyte + ntohs(nip6h->ip6_plen) +
6547                             sizeof (ip6_t) - nip6_hdr_length;
6548                 }
6549 
6550                 /* Check for overlapping fragments */
6551                 if (firstbyte >= nfirstbyte && firstbyte < nlastbyte) {
6552                         /*
6553                          * Overlap Check:
6554                          *  ~~~~---------               # Check if the newly
6555                          * ~    ndata_mp|               # received fragment
6556                          *  ~~~~---------               # overlaps with the
6557                          *       ---------~~~~~~        # current fragment.
6558                          *      |    mp         ~
6559                          *       ---------~~~~~~
6560                          */
6561                         if (is_v4) {
6562                                 data  = (char *)iph  + IPH_HDR_LENGTH(iph) +
6563                                     firstbyte - nfirstbyte;
6564                                 ndata = (char *)niph + IPH_HDR_LENGTH(niph);
6565                         } else {
6566                                 data  = (char *)ip6h  +
6567                                     nip6_hdr_length + firstbyte -
6568                                     nfirstbyte;
6569                                 ndata = (char *)nip6h + nip6_hdr_length;
6570                         }
6571                         if (bcmp(data, ndata, MIN(lastbyte, nlastbyte) -
6572                             firstbyte)) {
6573                                 /* Overlapping data does not match */
6574                                 (void) fragcache_delentry(i, fep, frag, ipss);
6575                                 mutex_exit(&frag->itpf_lock);
6576                                 ip_drop_packet(mp, inbound, NULL,
6577                                     DROPPER(ipss, ipds_spd_overlap_frag),
6578                                     &ipss->ipsec_spd_dropper);
6579                                 if (inbound)
6580                                         (void) ip_recv_attr_free_mblk(iramp);
6581                                 return (NULL);
6582                         }
6583                         /* Part of defense for jolt2.c fragmentation attack */
6584                         if (firstbyte >= nfirstbyte && lastbyte <= nlastbyte) {
6585                                 /*
6586                                  * Check for identical or subset fragments:
6587                                  *  ----------      ~~~~--------~~~~~
6588                                  * |    nmp   | or  ~      nmp      ~
6589                                  *  ----------      ~~~~--------~~~~~
6590                                  *  ----------            ------
6591                                  * |    mp    |          |  mp  |
6592                                  *  ----------            ------
6593                                  */
6594                                 mutex_exit(&frag->itpf_lock);
6595                                 ip_drop_packet(mp, inbound, NULL,
6596                                     DROPPER(ipss, ipds_spd_evil_frag),
6597                                     &ipss->ipsec_spd_dropper);
6598                                 if (inbound)
6599                                         (void) ip_recv_attr_free_mblk(iramp);
6600                                 return (NULL);
6601                         }
6602 
6603                 }
6604 
6605                 /* Correct location for this fragment? */
6606                 if (firstbyte <= nfirstbyte) {
6607                         /*
6608                          * Check if the tail end of the new fragment overlaps
6609                          * with the head of the current fragment.
6610                          *        --------~~~~~~~
6611                          *       |    nmp       ~
6612                          *        --------~~~~~~~
6613                          *  ~~~~~--------
6614                          *  ~   mp       |
6615                          *  ~~~~~--------
6616                          */
6617                         if (lastbyte > nfirstbyte) {
6618                                 /* Fragments overlap */
6619                                 data  = (char *)iph  + IPH_HDR_LENGTH(iph) +
6620                                     firstbyte - nfirstbyte;
6621                                 ndata = (char *)niph + IPH_HDR_LENGTH(niph);
6622                                 if (is_v4) {
6623                                         data  = (char *)iph +
6624                                             IPH_HDR_LENGTH(iph) + firstbyte -
6625                                             nfirstbyte;
6626                                         ndata = (char *)niph +
6627                                             IPH_HDR_LENGTH(niph);
6628                                 } else {
6629                                         data  = (char *)ip6h  +
6630                                             nip6_hdr_length + firstbyte -
6631                                             nfirstbyte;
6632                                         ndata = (char *)nip6h + nip6_hdr_length;
6633                                 }
6634                                 if (bcmp(data, ndata, MIN(lastbyte, nlastbyte)
6635                                     - nfirstbyte)) {
6636                                         /* Overlap mismatch */
6637                                         (void) fragcache_delentry(i, fep, frag,
6638                                             ipss);
6639                                         mutex_exit(&frag->itpf_lock);
6640                                         ip_drop_packet(mp, inbound, NULL,
6641                                             DROPPER(ipss,
6642                                             ipds_spd_overlap_frag),
6643                                             &ipss->ipsec_spd_dropper);
6644                                         if (inbound) {
6645                                                 (void) ip_recv_attr_free_mblk(
6646                                                     iramp);
6647                                         }
6648                                         return (NULL);
6649                                 }
6650                         }
6651 
6652                         /*
6653                          * Fragment does not illegally overlap and can now
6654                          * be inserted into the chain
6655                          */
6656                         break;
6657                 }
6658 
6659                 prevmp = nmp;
6660         }
6661         /* Prepend the attributes before we link it in */
6662         if (iramp != NULL) {
6663                 ASSERT(iramp->b_cont == NULL);
6664                 iramp->b_cont = mp;
6665                 mp = iramp;
6666                 iramp = NULL;
6667         }
6668         mp->b_next = nmp;
6669 
6670         if (prevmp == NULL) {
6671                 fep->itpfe_fraglist = mp;
6672         } else {
6673                 prevmp->b_next = mp;
6674         }
6675         if (last)
6676                 fep->itpfe_last = 1;
6677 
6678         /* Part of defense for jolt2.c fragmentation attack */
6679         if (++(fep->itpfe_depth) > IPSEC_MAX_FRAGS) {
6680                 (void) fragcache_delentry(i, fep, frag, ipss);
6681                 mutex_exit(&frag->itpf_lock);
6682                 if (inbound)
6683                         mp = ip_recv_attr_free_mblk(mp);
6684 
6685                 ip_drop_packet(mp, inbound, NULL,
6686                     DROPPER(ipss, ipds_spd_max_frags),
6687                     &ipss->ipsec_spd_dropper);
6688                 return (NULL);
6689         }
6690 
6691         /* Check for complete packet */
6692 
6693         if (!fep->itpfe_last) {
6694                 mutex_exit(&frag->itpf_lock);
6695 #ifdef FRAGCACHE_DEBUG
6696                 cmn_err(CE_WARN, "Fragment cached, last not yet seen.\n");
6697 #endif
6698                 return (NULL);
6699         }
6700 
6701         offset = 0;
6702         for (mp = fep->itpfe_fraglist; mp; mp = mp->b_next) {
6703                 mblk_t *data_mp = (inbound ? mp->b_cont : mp);
6704                 int hdr_len;
6705 
6706                 oiph  = (ipha_t *)data_mp->b_rptr;
6707                 ip6h = NULL;
6708                 iph = NULL;
6709 
6710                 if (IPH_HDR_VERSION(oiph) == IPV4_VERSION) {
6711                         hdr_len = ((outer_hdr_len != 0) ?
6712                             IPH_HDR_LENGTH(oiph) : 0);
6713                         iph = (ipha_t *)(data_mp->b_rptr + hdr_len);
6714                 } else {
6715                         ASSERT(IPH_HDR_VERSION(oiph) == IPV6_VERSION);
6716                         ASSERT(data_mp->b_cont == NULL);
6717                         ip6h = (ip6_t *)data_mp->b_rptr;
6718                         (void) ip_hdr_length_nexthdr_v6(data_mp, ip6h,
6719                             &ip6_hdr_length, &v6_proto_p);
6720                         hdr_len = ((outer_hdr_len != 0) ? ip6_hdr_length : 0);
6721                 }
6722 
6723                 /* Calculate current fragment start/end */
6724                 if (is_v4) {
6725                         if (iph == NULL) {
6726                                 /* Was v6 outer */
6727                                 iph = (ipha_t *)(data_mp->b_rptr + hdr_len);
6728                         }
6729                         firstbyte = V4_FRAG_OFFSET(iph);
6730                         lastbyte = firstbyte + ntohs(iph->ipha_length) -
6731                             IPH_HDR_LENGTH(iph);
6732                 } else {
6733                         ASSERT(data_mp->b_cont == NULL);
6734                         ip6h = (ip6_t *)(data_mp->b_rptr + hdr_len);
6735                         if (!ip_hdr_length_nexthdr_v6(data_mp, ip6h,
6736                             &ip6_hdr_length, &v6_proto_p)) {
6737                                 mutex_exit(&frag->itpf_lock);
6738                                 ip_drop_packet_chain(mp, inbound, NULL,
6739                                     DROPPER(ipss, ipds_spd_malformed_frag),
6740                                     &ipss->ipsec_spd_dropper);
6741                                 return (NULL);
6742                         }
6743                         v6_proto = *v6_proto_p;
6744                         bzero(&ipp, sizeof (ipp));
6745                         (void) ip_find_hdr_v6(data_mp, ip6h, B_FALSE, &ipp,
6746                             NULL);
6747                         fraghdr = ipp.ipp_fraghdr;
6748                         firstbyte = ntohs(fraghdr->ip6f_offlg &
6749                             IP6F_OFF_MASK);
6750                         lastbyte  = firstbyte + ntohs(ip6h->ip6_plen) +
6751                             sizeof (ip6_t) - ip6_hdr_length;
6752                 }
6753 
6754                 /*
6755                  * If this fragment is greater than current offset,
6756                  * we have a missing fragment so return NULL
6757                  */
6758                 if (firstbyte > offset) {
6759                         mutex_exit(&frag->itpf_lock);
6760 #ifdef FRAGCACHE_DEBUG
6761                         /*
6762                          * Note, this can happen when the last frag
6763                          * gets sent through because it is smaller
6764                          * than the MTU.  It is not necessarily an
6765                          * error condition.
6766                          */
6767                         cmn_err(CE_WARN, "Frag greater than offset! : "
6768                             "missing fragment: firstbyte = %d, offset = %d, "
6769                             "mp = %p\n", firstbyte, offset, mp);
6770 #endif
6771                         return (NULL);
6772                 }
6773 #ifdef FRAGCACHE_DEBUG
6774                 cmn_err(CE_WARN, "Frag offsets : "
6775                     "firstbyte = %d, offset = %d, mp = %p\n",
6776                     firstbyte, offset, mp);
6777 #endif
6778 
6779                 /*
6780                  * If we are at the last fragment, we have the complete
6781                  * packet, so rechain things and return it to caller
6782                  * for processing
6783                  */
6784 
6785                 if ((is_v4 && !V4_MORE_FRAGS(iph)) ||
6786                     (!is_v4 && !(fraghdr->ip6f_offlg & IP6F_MORE_FRAG))) {
6787                         mp = fep->itpfe_fraglist;
6788                         fep->itpfe_fraglist = NULL;
6789                         (void) fragcache_delentry(i, fep, frag, ipss);
6790                         mutex_exit(&frag->itpf_lock);
6791 
6792                         if ((is_v4 && (firstbyte + ntohs(iph->ipha_length) >
6793                             65535)) || (!is_v4 && (firstbyte +
6794                             ntohs(ip6h->ip6_plen) > 65535))) {
6795                                 /* It is an invalid "ping-o-death" packet */
6796                                 /* Discard it */
6797                                 ip_drop_packet_chain(mp, inbound, NULL,
6798                                     DROPPER(ipss, ipds_spd_evil_frag),
6799                                     &ipss->ipsec_spd_dropper);
6800                                 return (NULL);
6801                         }
6802 #ifdef FRAGCACHE_DEBUG
6803                         cmn_err(CE_WARN, "Fragcache returning mp = %p, "
6804                             "mp->b_next = %p", mp, mp->b_next);
6805 #endif
6806                         /*
6807                          * For inbound case, mp has attrmp b_next'd chain
6808                          * For outbound case, it is just data mp chain
6809                          */
6810                         return (mp);
6811                 }
6812 
6813                 /*
6814                  * Update new ending offset if this
6815                  * fragment extends the packet
6816                  */
6817                 if (offset < lastbyte)
6818                         offset = lastbyte;
6819         }
6820 
6821         mutex_exit(&frag->itpf_lock);
6822 
6823         /* Didn't find last fragment, so return NULL */
6824         return (NULL);
6825 }
6826 
6827 static void
6828 ipsec_fragcache_clean(ipsec_fragcache_t *frag, ipsec_stack_t *ipss)
6829 {
6830         ipsec_fragcache_entry_t *fep;
6831         int i;
6832         ipsec_fragcache_entry_t *earlyfep = NULL;
6833         time_t itpf_time;
6834         int earlyexp;
6835         int earlyi = 0;
6836 
6837         ASSERT(MUTEX_HELD(&frag->itpf_lock));
6838 
6839         itpf_time = gethrestime_sec();
6840         earlyexp = itpf_time + 10000;
6841 
6842         for (i = 0; i < IPSEC_FRAG_HASH_SLOTS; i++) {
6843                 fep = (frag->itpf_ptr)[i];
6844                 while (fep) {
6845                         if (fep->itpfe_exp < itpf_time) {
6846                                 /* found */
6847                                 fep = fragcache_delentry(i, fep, frag, ipss);
6848                         } else {
6849                                 if (fep->itpfe_exp < earlyexp) {
6850                                         earlyfep = fep;
6851                                         earlyexp = fep->itpfe_exp;
6852                                         earlyi = i;
6853                                 }
6854                                 fep = fep->itpfe_next;
6855                         }
6856                 }
6857         }
6858 
6859         frag->itpf_expire_hint = earlyexp;
6860 
6861         /* if (!found) */
6862         if (frag->itpf_freelist == NULL)
6863                 (void) fragcache_delentry(earlyi, earlyfep, frag, ipss);
6864 }
6865 
6866 static ipsec_fragcache_entry_t *
6867 fragcache_delentry(int slot, ipsec_fragcache_entry_t *fep,
6868     ipsec_fragcache_t *frag, ipsec_stack_t *ipss)
6869 {
6870         ipsec_fragcache_entry_t *targp;
6871         ipsec_fragcache_entry_t *nextp = fep->itpfe_next;
6872 
6873         ASSERT(MUTEX_HELD(&frag->itpf_lock));
6874 
6875         /* Free up any fragment list still in cache entry */
6876         if (fep->itpfe_fraglist != NULL) {
6877                 ip_drop_packet_chain(fep->itpfe_fraglist,
6878                     ip_recv_attr_is_mblk(fep->itpfe_fraglist), NULL,
6879                     DROPPER(ipss, ipds_spd_expired_frags),
6880                     &ipss->ipsec_spd_dropper);
6881         }
6882         fep->itpfe_fraglist = NULL;
6883 
6884         targp = (frag->itpf_ptr)[slot];
6885         ASSERT(targp != 0);
6886 
6887         if (targp == fep) {
6888                 /* unlink from head of hash chain */
6889                 (frag->itpf_ptr)[slot] = nextp;
6890                 /* link into free list */
6891                 fep->itpfe_next = frag->itpf_freelist;
6892                 frag->itpf_freelist = fep;
6893                 return (nextp);
6894         }
6895 
6896         /* maybe should use double linked list to make update faster */
6897         /* must be past front of chain */
6898         while (targp) {
6899                 if (targp->itpfe_next == fep) {
6900                         /* unlink from hash chain */
6901                         targp->itpfe_next = nextp;
6902                         /* link into free list */
6903                         fep->itpfe_next = frag->itpf_freelist;
6904                         frag->itpf_freelist = fep;
6905                         return (nextp);
6906                 }
6907                 targp = targp->itpfe_next;
6908                 ASSERT(targp != 0);
6909         }
6910         /* NOTREACHED */
6911         return (NULL);
6912 }