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  */
  26 
  27 /*
  28  * IPsec Security Policy Database.
  29  *
  30  * This module maintains the SPD and provides routines used by ip and ip6
  31  * to apply IPsec policy to inbound and outbound datagrams.
  32  */
  33 
  34 #include <sys/types.h>
  35 #include <sys/stream.h>
  36 #include <sys/stropts.h>
  37 #include <sys/sysmacros.h>
  38 #include <sys/strsubr.h>
  39 #include <sys/strsun.h>
  40 #include <sys/strlog.h>
  41 #include <sys/strsun.h>
  42 #include <sys/cmn_err.h>
  43 #include <sys/zone.h>
  44 
  45 #include <sys/systm.h>
  46 #include <sys/param.h>
  47 #include <sys/kmem.h>
  48 #include <sys/ddi.h>
  49 
  50 #include <sys/crypto/api.h>
  51 
  52 #include <inet/common.h>
  53 #include <inet/mi.h>
  54 
  55 #include <netinet/ip6.h>
  56 #include <netinet/icmp6.h>
  57 #include <netinet/udp.h>
  58 
  59 #include <inet/ip.h>
  60 #include <inet/ip6.h>
  61 
  62 #include <net/pfkeyv2.h>
  63 #include <net/pfpolicy.h>
  64 #include <inet/sadb.h>
  65 #include <inet/ipsec_impl.h>
  66 
  67 #include <inet/ip_impl.h> /* For IP_MOD_ID */
  68 
  69 #include <inet/ipsecah.h>
  70 #include <inet/ipsecesp.h>
  71 #include <inet/ipdrop.h>
  72 #include <inet/ipclassifier.h>
  73 #include <inet/iptun.h>
  74 #include <inet/iptun/iptun_impl.h>
  75 
  76 static void ipsec_update_present_flags(ipsec_stack_t *);
  77 static ipsec_act_t *ipsec_act_wildcard_expand(ipsec_act_t *, uint_t *,
  78     netstack_t *);
  79 static mblk_t *ipsec_check_ipsecin_policy(mblk_t *, ipsec_policy_t *,
  80     ipha_t *, ip6_t *, uint64_t, ip_recv_attr_t *, netstack_t *);
  81 static void ipsec_action_free_table(ipsec_action_t *);
  82 static void ipsec_action_reclaim(void *);
  83 static void ipsec_action_reclaim_stack(ipsec_stack_t *);
  84 static void ipsid_init(netstack_t *);
  85 static void ipsid_fini(netstack_t *);
  86 
  87 /* sel_flags values for ipsec_init_inbound_sel(). */
  88 #define SEL_NONE        0x0000
  89 #define SEL_PORT_POLICY 0x0001
  90 #define SEL_IS_ICMP     0x0002
  91 #define SEL_TUNNEL_MODE 0x0004
  92 #define SEL_POST_FRAG   0x0008
  93 
  94 /* Return values for ipsec_init_inbound_sel(). */
  95 typedef enum { SELRET_NOMEM, SELRET_BADPKT, SELRET_SUCCESS, SELRET_TUNFRAG}
  96     selret_t;
  97 
  98 static selret_t ipsec_init_inbound_sel(ipsec_selector_t *, mblk_t *,
  99     ipha_t *, ip6_t *, uint8_t);
 100 
 101 static boolean_t ipsec_check_ipsecin_action(ip_recv_attr_t *, mblk_t *,
 102     struct ipsec_action_s *, ipha_t *ipha, ip6_t *ip6h, const char **,
 103     kstat_named_t **, netstack_t *);
 104 static void ipsec_unregister_prov_update(void);
 105 static void ipsec_prov_update_callback_stack(uint32_t, void *, netstack_t *);
 106 static boolean_t ipsec_compare_action(ipsec_policy_t *, ipsec_policy_t *);
 107 static uint32_t selector_hash(ipsec_selector_t *, ipsec_policy_root_t *);
 108 static boolean_t ipsec_kstat_init(ipsec_stack_t *);
 109 static void ipsec_kstat_destroy(ipsec_stack_t *);
 110 static int ipsec_free_tables(ipsec_stack_t *);
 111 static int tunnel_compare(const void *, const void *);
 112 static void ipsec_freemsg_chain(mblk_t *);
 113 static void ip_drop_packet_chain(mblk_t *, boolean_t, ill_t *,
 114     struct kstat_named *, ipdropper_t *);
 115 static boolean_t ipsec_kstat_init(ipsec_stack_t *);
 116 static void ipsec_kstat_destroy(ipsec_stack_t *);
 117 static int ipsec_free_tables(ipsec_stack_t *);
 118 static int tunnel_compare(const void *, const void *);
 119 static void ipsec_freemsg_chain(mblk_t *);
 120 
 121 /*
 122  * Selector hash table is statically sized at module load time.
 123  * we default to 251 buckets, which is the largest prime number under 255
 124  */
 125 
 126 #define IPSEC_SPDHASH_DEFAULT 251
 127 
 128 /* SPD hash-size tunable per tunnel. */
 129 #define TUN_SPDHASH_DEFAULT 5
 130 
 131 uint32_t ipsec_spd_hashsize;
 132 uint32_t tun_spd_hashsize;
 133 
 134 #define IPSEC_SEL_NOHASH ((uint32_t)(~0))
 135 
 136 /*
 137  * Handle global across all stack instances
 138  */
 139 static crypto_notify_handle_t prov_update_handle = NULL;
 140 
 141 static kmem_cache_t *ipsec_action_cache;
 142 static kmem_cache_t *ipsec_sel_cache;
 143 static kmem_cache_t *ipsec_pol_cache;
 144 
 145 /* Frag cache prototypes */
 146 static void ipsec_fragcache_clean(ipsec_fragcache_t *, ipsec_stack_t *);
 147 static ipsec_fragcache_entry_t *fragcache_delentry(int,
 148     ipsec_fragcache_entry_t *, ipsec_fragcache_t *, ipsec_stack_t *);
 149 boolean_t ipsec_fragcache_init(ipsec_fragcache_t *);
 150 void ipsec_fragcache_uninit(ipsec_fragcache_t *, ipsec_stack_t *ipss);
 151 mblk_t *ipsec_fragcache_add(ipsec_fragcache_t *, mblk_t *, mblk_t *,
 152     int, ipsec_stack_t *);
 153 
 154 int ipsec_hdr_pullup_needed = 0;
 155 int ipsec_weird_null_inbound_policy = 0;
 156 
 157 #define ALGBITS_ROUND_DOWN(x, align)    (((x)/(align))*(align))
 158 #define ALGBITS_ROUND_UP(x, align)      ALGBITS_ROUND_DOWN((x)+(align)-1, align)
 159 
 160 /*
 161  * Inbound traffic should have matching identities for both SA's.
 162  */
 163 
 164 #define SA_IDS_MATCH(sa1, sa2)                                          \
 165         (((sa1) == NULL) || ((sa2) == NULL) ||                          \
 166         (((sa1)->ipsa_src_cid == (sa2)->ipsa_src_cid) &&          \
 167             (((sa1)->ipsa_dst_cid == (sa2)->ipsa_dst_cid))))
 168 
 169 /*
 170  * IPv6 Fragments
 171  */
 172 #define IS_V6_FRAGMENT(ipp)     (ipp.ipp_fields & IPPF_FRAGHDR)
 173 
 174 /*
 175  * Policy failure messages.
 176  */
 177 static char *ipsec_policy_failure_msgs[] = {
 178 
 179         /* IPSEC_POLICY_NOT_NEEDED */
 180         "%s: Dropping the datagram because the incoming packet "
 181         "is %s, but the recipient expects clear; Source %s, "
 182         "Destination %s.\n",
 183 
 184         /* IPSEC_POLICY_MISMATCH */
 185         "%s: Policy Failure for the incoming packet (%s); Source %s, "
 186         "Destination %s.\n",
 187 
 188         /* IPSEC_POLICY_AUTH_NOT_NEEDED */
 189         "%s: Authentication present while not expected in the "
 190         "incoming %s packet; Source %s, Destination %s.\n",
 191 
 192         /* IPSEC_POLICY_ENCR_NOT_NEEDED */
 193         "%s: Encryption present while not expected in the "
 194         "incoming %s packet; Source %s, Destination %s.\n",
 195 
 196         /* IPSEC_POLICY_SE_NOT_NEEDED */
 197         "%s: Self-Encapsulation present while not expected in the "
 198         "incoming %s packet; Source %s, Destination %s.\n",
 199 };
 200 
 201 /*
 202  * General overviews:
 203  *
 204  * Locking:
 205  *
 206  *      All of the system policy structures are protected by a single
 207  *      rwlock.  These structures are threaded in a
 208  *      fairly complex fashion and are not expected to change on a
 209  *      regular basis, so this should not cause scaling/contention
 210  *      problems.  As a result, policy checks should (hopefully) be MT-hot.
 211  *
 212  * Allocation policy:
 213  *
 214  *      We use custom kmem cache types for the various
 215  *      bits & pieces of the policy data structures.  All allocations
 216  *      use KM_NOSLEEP instead of KM_SLEEP for policy allocation.  The
 217  *      policy table is of potentially unbounded size, so we don't
 218  *      want to provide a way to hog all system memory with policy
 219  *      entries..
 220  */
 221 
 222 /* Convenient functions for freeing or dropping a b_next linked mblk chain */
 223 
 224 /* Free all messages in an mblk chain */
 225 static void
 226 ipsec_freemsg_chain(mblk_t *mp)
 227 {
 228         mblk_t *mpnext;
 229         while (mp != NULL) {
 230                 ASSERT(mp->b_prev == NULL);
 231                 mpnext = mp->b_next;
 232                 mp->b_next = NULL;
 233                 freemsg(mp);
 234                 mp = mpnext;
 235         }
 236 }
 237 
 238 /*
 239  * ip_drop all messages in an mblk chain
 240  * Can handle a b_next chain of ip_recv_attr_t mblks, or just a b_next chain
 241  * of data.
 242  */
 243 static void
 244 ip_drop_packet_chain(mblk_t *mp, boolean_t inbound, ill_t *ill,
 245     struct kstat_named *counter, ipdropper_t *who_called)
 246 {
 247         mblk_t *mpnext;
 248         while (mp != NULL) {
 249                 ASSERT(mp->b_prev == NULL);
 250                 mpnext = mp->b_next;
 251                 mp->b_next = NULL;
 252                 if (ip_recv_attr_is_mblk(mp))
 253                         mp = ip_recv_attr_free_mblk(mp);
 254                 ip_drop_packet(mp, inbound, ill, counter, who_called);
 255                 mp = mpnext;
 256         }
 257 }
 258 
 259 /*
 260  * AVL tree comparison function.
 261  * the in-kernel avl assumes unique keys for all objects.
 262  * Since sometimes policy will duplicate rules, we may insert
 263  * multiple rules with the same rule id, so we need a tie-breaker.
 264  */
 265 static int
 266 ipsec_policy_cmpbyid(const void *a, const void *b)
 267 {
 268         const ipsec_policy_t *ipa, *ipb;
 269         uint64_t idxa, idxb;
 270 
 271         ipa = (const ipsec_policy_t *)a;
 272         ipb = (const ipsec_policy_t *)b;
 273         idxa = ipa->ipsp_index;
 274         idxb = ipb->ipsp_index;
 275 
 276         if (idxa < idxb)
 277                 return (-1);
 278         if (idxa > idxb)
 279                 return (1);
 280         /*
 281          * Tie-breaker #1: All installed policy rules have a non-NULL
 282          * ipsl_sel (selector set), so an entry with a NULL ipsp_sel is not
 283          * actually in-tree but rather a template node being used in
 284          * an avl_find query; see ipsec_policy_delete().  This gives us
 285          * a placeholder in the ordering just before the first entry with
 286          * a key >= the one we're looking for, so we can walk forward from
 287          * that point to get the remaining entries with the same id.
 288          */
 289         if ((ipa->ipsp_sel == NULL) && (ipb->ipsp_sel != NULL))
 290                 return (-1);
 291         if ((ipb->ipsp_sel == NULL) && (ipa->ipsp_sel != NULL))
 292                 return (1);
 293         /*
 294          * At most one of the arguments to the comparison should have a
 295          * NULL selector pointer; if not, the tree is broken.
 296          */
 297         ASSERT(ipa->ipsp_sel != NULL);
 298         ASSERT(ipb->ipsp_sel != NULL);
 299         /*
 300          * Tie-breaker #2: use the virtual address of the policy node
 301          * to arbitrarily break ties.  Since we use the new tree node in
 302          * the avl_find() in ipsec_insert_always, the new node will be
 303          * inserted into the tree in the right place in the sequence.
 304          */
 305         if (ipa < ipb)
 306                 return (-1);
 307         if (ipa > ipb)
 308                 return (1);
 309         return (0);
 310 }
 311 
 312 /*
 313  * Free what ipsec_alloc_table allocated.
 314  */
 315 void
 316 ipsec_polhead_free_table(ipsec_policy_head_t *iph)
 317 {
 318         int dir;
 319         int i;
 320 
 321         for (dir = 0; dir < IPSEC_NTYPES; dir++) {
 322                 ipsec_policy_root_t *ipr = &iph->iph_root[dir];
 323 
 324                 if (ipr->ipr_hash == NULL)
 325                         continue;
 326 
 327                 for (i = 0; i < ipr->ipr_nchains; i++) {
 328                         ASSERT(ipr->ipr_hash[i].hash_head == NULL);
 329                 }
 330                 kmem_free(ipr->ipr_hash, ipr->ipr_nchains *
 331                     sizeof (ipsec_policy_hash_t));
 332                 ipr->ipr_hash = NULL;
 333         }
 334 }
 335 
 336 void
 337 ipsec_polhead_destroy(ipsec_policy_head_t *iph)
 338 {
 339         int dir;
 340 
 341         avl_destroy(&iph->iph_rulebyid);
 342         rw_destroy(&iph->iph_lock);
 343 
 344         for (dir = 0; dir < IPSEC_NTYPES; dir++) {
 345                 ipsec_policy_root_t *ipr = &iph->iph_root[dir];
 346                 int chain;
 347 
 348                 for (chain = 0; chain < ipr->ipr_nchains; chain++)
 349                         mutex_destroy(&(ipr->ipr_hash[chain].hash_lock));
 350 
 351         }
 352         ipsec_polhead_free_table(iph);
 353 }
 354 
 355 /*
 356  * Free the IPsec stack instance.
 357  */
 358 /* ARGSUSED */
 359 static void
 360 ipsec_stack_fini(netstackid_t stackid, void *arg)
 361 {
 362         ipsec_stack_t   *ipss = (ipsec_stack_t *)arg;
 363         void *cookie;
 364         ipsec_tun_pol_t *node;
 365         netstack_t      *ns = ipss->ipsec_netstack;
 366         int             i;
 367         ipsec_algtype_t algtype;
 368 
 369         ipsec_loader_destroy(ipss);
 370 
 371         rw_enter(&ipss->ipsec_tunnel_policy_lock, RW_WRITER);
 372         /*
 373          * It's possible we can just ASSERT() the tree is empty.  After all,
 374          * we aren't called until IP is ready to unload (and presumably all
 375          * tunnels have been unplumbed).  But we'll play it safe for now, the
 376          * loop will just exit immediately if it's empty.
 377          */
 378         cookie = NULL;
 379         while ((node = (ipsec_tun_pol_t *)
 380             avl_destroy_nodes(&ipss->ipsec_tunnel_policies,
 381             &cookie)) != NULL) {
 382                 ITP_REFRELE(node, ns);
 383         }
 384         avl_destroy(&ipss->ipsec_tunnel_policies);
 385         rw_exit(&ipss->ipsec_tunnel_policy_lock);
 386         rw_destroy(&ipss->ipsec_tunnel_policy_lock);
 387 
 388         ipsec_config_flush(ns);
 389 
 390         ipsec_kstat_destroy(ipss);
 391 
 392         ip_drop_unregister(&ipss->ipsec_dropper);
 393 
 394         ip_drop_unregister(&ipss->ipsec_spd_dropper);
 395         ip_drop_destroy(ipss);
 396         /*
 397          * Globals start with ref == 1 to prevent IPPH_REFRELE() from
 398          * attempting to free them, hence they should have 1 now.
 399          */
 400         ipsec_polhead_destroy(&ipss->ipsec_system_policy);
 401         ASSERT(ipss->ipsec_system_policy.iph_refs == 1);
 402         ipsec_polhead_destroy(&ipss->ipsec_inactive_policy);
 403         ASSERT(ipss->ipsec_inactive_policy.iph_refs == 1);
 404 
 405         for (i = 0; i < IPSEC_ACTION_HASH_SIZE; i++) {
 406                 ipsec_action_free_table(ipss->ipsec_action_hash[i].hash_head);
 407                 ipss->ipsec_action_hash[i].hash_head = NULL;
 408                 mutex_destroy(&(ipss->ipsec_action_hash[i].hash_lock));
 409         }
 410 
 411         for (i = 0; i < ipss->ipsec_spd_hashsize; i++) {
 412                 ASSERT(ipss->ipsec_sel_hash[i].hash_head == NULL);
 413                 mutex_destroy(&(ipss->ipsec_sel_hash[i].hash_lock));
 414         }
 415 
 416         rw_enter(&ipss->ipsec_alg_lock, RW_WRITER);
 417         for (algtype = 0; algtype < IPSEC_NALGTYPES; algtype ++) {
 418                 int nalgs = ipss->ipsec_nalgs[algtype];
 419 
 420                 for (i = 0; i < nalgs; i++) {
 421                         if (ipss->ipsec_alglists[algtype][i] != NULL)
 422                                 ipsec_alg_unreg(algtype, i, ns);
 423                 }
 424         }
 425         rw_exit(&ipss->ipsec_alg_lock);
 426         rw_destroy(&ipss->ipsec_alg_lock);
 427 
 428         ipsid_gc(ns);
 429         ipsid_fini(ns);
 430 
 431         (void) ipsec_free_tables(ipss);
 432         kmem_free(ipss, sizeof (*ipss));
 433 }
 434 
 435 void
 436 ipsec_policy_g_destroy(void)
 437 {
 438         kmem_cache_destroy(ipsec_action_cache);
 439         kmem_cache_destroy(ipsec_sel_cache);
 440         kmem_cache_destroy(ipsec_pol_cache);
 441 
 442         ipsec_unregister_prov_update();
 443 
 444         netstack_unregister(NS_IPSEC);
 445 }
 446 
 447 
 448 /*
 449  * Free what ipsec_alloc_tables allocated.
 450  * Called when table allocation fails to free the table.
 451  */
 452 static int
 453 ipsec_free_tables(ipsec_stack_t *ipss)
 454 {
 455         int i;
 456 
 457         if (ipss->ipsec_sel_hash != NULL) {
 458                 for (i = 0; i < ipss->ipsec_spd_hashsize; i++) {
 459                         ASSERT(ipss->ipsec_sel_hash[i].hash_head == NULL);
 460                 }
 461                 kmem_free(ipss->ipsec_sel_hash, ipss->ipsec_spd_hashsize *
 462                     sizeof (*ipss->ipsec_sel_hash));
 463                 ipss->ipsec_sel_hash = NULL;
 464                 ipss->ipsec_spd_hashsize = 0;
 465         }
 466         ipsec_polhead_free_table(&ipss->ipsec_system_policy);
 467         ipsec_polhead_free_table(&ipss->ipsec_inactive_policy);
 468 
 469         return (ENOMEM);
 470 }
 471 
 472 /*
 473  * Attempt to allocate the tables in a single policy head.
 474  * Return nonzero on failure after cleaning up any work in progress.
 475  */
 476 int
 477 ipsec_alloc_table(ipsec_policy_head_t *iph, int nchains, int kmflag,
 478     boolean_t global_cleanup, netstack_t *ns)
 479 {
 480         int dir;
 481 
 482         for (dir = 0; dir < IPSEC_NTYPES; dir++) {
 483                 ipsec_policy_root_t *ipr = &iph->iph_root[dir];
 484 
 485                 ipr->ipr_nchains = nchains;
 486                 ipr->ipr_hash = kmem_zalloc(nchains *
 487                     sizeof (ipsec_policy_hash_t), kmflag);
 488                 if (ipr->ipr_hash == NULL)
 489                         return (global_cleanup ?
 490                             ipsec_free_tables(ns->netstack_ipsec) :
 491                             ENOMEM);
 492         }
 493         return (0);
 494 }
 495 
 496 /*
 497  * Attempt to allocate the various tables.  Return nonzero on failure
 498  * after cleaning up any work in progress.
 499  */
 500 static int
 501 ipsec_alloc_tables(int kmflag, netstack_t *ns)
 502 {
 503         int error;
 504         ipsec_stack_t   *ipss = ns->netstack_ipsec;
 505 
 506         error = ipsec_alloc_table(&ipss->ipsec_system_policy,
 507             ipss->ipsec_spd_hashsize, kmflag, B_TRUE, ns);
 508         if (error != 0)
 509                 return (error);
 510 
 511         error = ipsec_alloc_table(&ipss->ipsec_inactive_policy,
 512             ipss->ipsec_spd_hashsize, kmflag, B_TRUE, ns);
 513         if (error != 0)
 514                 return (error);
 515 
 516         ipss->ipsec_sel_hash = kmem_zalloc(ipss->ipsec_spd_hashsize *
 517             sizeof (*ipss->ipsec_sel_hash), kmflag);
 518 
 519         if (ipss->ipsec_sel_hash == NULL)
 520                 return (ipsec_free_tables(ipss));
 521 
 522         return (0);
 523 }
 524 
 525 /*
 526  * After table allocation, initialize a policy head.
 527  */
 528 void
 529 ipsec_polhead_init(ipsec_policy_head_t *iph, int nchains)
 530 {
 531         int dir, chain;
 532 
 533         rw_init(&iph->iph_lock, NULL, RW_DEFAULT, NULL);
 534         avl_create(&iph->iph_rulebyid, ipsec_policy_cmpbyid,
 535             sizeof (ipsec_policy_t), offsetof(ipsec_policy_t, ipsp_byid));
 536 
 537         for (dir = 0; dir < IPSEC_NTYPES; dir++) {
 538                 ipsec_policy_root_t *ipr = &iph->iph_root[dir];
 539                 ipr->ipr_nchains = nchains;
 540 
 541                 for (chain = 0; chain < nchains; chain++) {
 542                         mutex_init(&(ipr->ipr_hash[chain].hash_lock),
 543                             NULL, MUTEX_DEFAULT, NULL);
 544                 }
 545         }
 546 }
 547 
 548 static boolean_t
 549 ipsec_kstat_init(ipsec_stack_t *ipss)
 550 {
 551         ipss->ipsec_ksp = kstat_create_netstack("ip", 0, "ipsec_stat", "net",
 552             KSTAT_TYPE_NAMED, sizeof (ipsec_kstats_t) / sizeof (kstat_named_t),
 553             KSTAT_FLAG_PERSISTENT, ipss->ipsec_netstack->netstack_stackid);
 554 
 555         if (ipss->ipsec_ksp == NULL || ipss->ipsec_ksp->ks_data == NULL)
 556                 return (B_FALSE);
 557 
 558         ipss->ipsec_kstats = ipss->ipsec_ksp->ks_data;
 559 
 560 #define KI(x) kstat_named_init(&ipss->ipsec_kstats->x, #x, KSTAT_DATA_UINT64)
 561         KI(esp_stat_in_requests);
 562         KI(esp_stat_in_discards);
 563         KI(esp_stat_lookup_failure);
 564         KI(ah_stat_in_requests);
 565         KI(ah_stat_in_discards);
 566         KI(ah_stat_lookup_failure);
 567         KI(sadb_acquire_maxpackets);
 568         KI(sadb_acquire_qhiwater);
 569 #undef KI
 570 
 571         kstat_install(ipss->ipsec_ksp);
 572         return (B_TRUE);
 573 }
 574 
 575 static void
 576 ipsec_kstat_destroy(ipsec_stack_t *ipss)
 577 {
 578         kstat_delete_netstack(ipss->ipsec_ksp,
 579             ipss->ipsec_netstack->netstack_stackid);
 580         ipss->ipsec_kstats = NULL;
 581 
 582 }
 583 
 584 /*
 585  * Initialize the IPsec stack instance.
 586  */
 587 /* ARGSUSED */
 588 static void *
 589 ipsec_stack_init(netstackid_t stackid, netstack_t *ns)
 590 {
 591         ipsec_stack_t   *ipss;
 592         int i;
 593 
 594         ipss = (ipsec_stack_t *)kmem_zalloc(sizeof (*ipss), KM_SLEEP);
 595         ipss->ipsec_netstack = ns;
 596 
 597         /*
 598          * FIXME: netstack_ipsec is used by some of the routines we call
 599          * below, but it isn't set until this routine returns.
 600          * Either we introduce optional xxx_stack_alloc() functions
 601          * that will be called by the netstack framework before xxx_stack_init,
 602          * or we switch spd.c and sadb.c to operate on ipsec_stack_t
 603          * (latter has some include file order issues for sadb.h, but makes
 604          * sense if we merge some of the ipsec related stack_t's together.
 605          */
 606         ns->netstack_ipsec = ipss;
 607 
 608         /*
 609          * Make two attempts to allocate policy hash tables; try it at
 610          * the "preferred" size (may be set in /etc/system) first,
 611          * then fall back to the default size.
 612          */
 613         ipss->ipsec_spd_hashsize = (ipsec_spd_hashsize == 0) ?
 614             IPSEC_SPDHASH_DEFAULT : ipsec_spd_hashsize;
 615 
 616         if (ipsec_alloc_tables(KM_NOSLEEP, ns) != 0) {
 617                 cmn_err(CE_WARN,
 618                     "Unable to allocate %d entry IPsec policy hash table",
 619                     ipss->ipsec_spd_hashsize);
 620                 ipss->ipsec_spd_hashsize = IPSEC_SPDHASH_DEFAULT;
 621                 cmn_err(CE_WARN, "Falling back to %d entries",
 622                     ipss->ipsec_spd_hashsize);
 623                 (void) ipsec_alloc_tables(KM_SLEEP, ns);
 624         }
 625 
 626         /* Just set a default for tunnels. */
 627         ipss->ipsec_tun_spd_hashsize = (tun_spd_hashsize == 0) ?
 628             TUN_SPDHASH_DEFAULT : tun_spd_hashsize;
 629 
 630         ipsid_init(ns);
 631         /*
 632          * Globals need ref == 1 to prevent IPPH_REFRELE() from attempting
 633          * to free them.
 634          */
 635         ipss->ipsec_system_policy.iph_refs = 1;
 636         ipss->ipsec_inactive_policy.iph_refs = 1;
 637         ipsec_polhead_init(&ipss->ipsec_system_policy,
 638             ipss->ipsec_spd_hashsize);
 639         ipsec_polhead_init(&ipss->ipsec_inactive_policy,
 640             ipss->ipsec_spd_hashsize);
 641         rw_init(&ipss->ipsec_tunnel_policy_lock, NULL, RW_DEFAULT, NULL);
 642         avl_create(&ipss->ipsec_tunnel_policies, tunnel_compare,
 643             sizeof (ipsec_tun_pol_t), 0);
 644 
 645         ipss->ipsec_next_policy_index = 1;
 646 
 647         rw_init(&ipss->ipsec_system_policy.iph_lock, NULL, RW_DEFAULT, NULL);
 648         rw_init(&ipss->ipsec_inactive_policy.iph_lock, NULL, RW_DEFAULT, NULL);
 649 
 650         for (i = 0; i < IPSEC_ACTION_HASH_SIZE; i++)
 651                 mutex_init(&(ipss->ipsec_action_hash[i].hash_lock),
 652                     NULL, MUTEX_DEFAULT, NULL);
 653 
 654         for (i = 0; i < ipss->ipsec_spd_hashsize; i++)
 655                 mutex_init(&(ipss->ipsec_sel_hash[i].hash_lock),
 656                     NULL, MUTEX_DEFAULT, NULL);
 657 
 658         rw_init(&ipss->ipsec_alg_lock, NULL, RW_DEFAULT, NULL);
 659         for (i = 0; i < IPSEC_NALGTYPES; i++) {
 660                 ipss->ipsec_nalgs[i] = 0;
 661         }
 662 
 663         ip_drop_init(ipss);
 664         ip_drop_register(&ipss->ipsec_spd_dropper, "IPsec SPD");
 665 
 666         /* IP's IPsec code calls the packet dropper */
 667         ip_drop_register(&ipss->ipsec_dropper, "IP IPsec processing");
 668 
 669         (void) ipsec_kstat_init(ipss);
 670 
 671         ipsec_loader_init(ipss);
 672         ipsec_loader_start(ipss);
 673 
 674         return (ipss);
 675 }
 676 
 677 /* Global across all stack instances */
 678 void
 679 ipsec_policy_g_init(void)
 680 {
 681         ipsec_action_cache = kmem_cache_create("ipsec_actions",
 682             sizeof (ipsec_action_t), _POINTER_ALIGNMENT, NULL, NULL,
 683             ipsec_action_reclaim, NULL, NULL, 0);
 684         ipsec_sel_cache = kmem_cache_create("ipsec_selectors",
 685             sizeof (ipsec_sel_t), _POINTER_ALIGNMENT, NULL, NULL,
 686             NULL, NULL, NULL, 0);
 687         ipsec_pol_cache = kmem_cache_create("ipsec_policy",
 688             sizeof (ipsec_policy_t), _POINTER_ALIGNMENT, NULL, NULL,
 689             NULL, NULL, NULL, 0);
 690 
 691         /*
 692          * We want to be informed each time a stack is created or
 693          * destroyed in the kernel, so we can maintain the
 694          * set of ipsec_stack_t's.
 695          */
 696         netstack_register(NS_IPSEC, ipsec_stack_init, NULL, ipsec_stack_fini);
 697 }
 698 
 699 /*
 700  * Sort algorithm lists.
 701  *
 702  * I may need to split this based on
 703  * authentication/encryption, and I may wish to have an administrator
 704  * configure this list.  Hold on to some NDD variables...
 705  *
 706  * XXX For now, sort on minimum key size (GAG!).  While minimum key size is
 707  * not the ideal metric, it's the only quantifiable measure available.
 708  * We need a better metric for sorting algorithms by preference.
 709  */
 710 static void
 711 alg_insert_sortlist(enum ipsec_algtype at, uint8_t algid, netstack_t *ns)
 712 {
 713         ipsec_stack_t   *ipss = ns->netstack_ipsec;
 714         ipsec_alginfo_t *ai = ipss->ipsec_alglists[at][algid];
 715         uint8_t holder, swap;
 716         uint_t i;
 717         uint_t count = ipss->ipsec_nalgs[at];
 718         ASSERT(ai != NULL);
 719         ASSERT(algid == ai->alg_id);
 720 
 721         ASSERT(RW_WRITE_HELD(&ipss->ipsec_alg_lock));
 722 
 723         holder = algid;
 724 
 725         for (i = 0; i < count - 1; i++) {
 726                 ipsec_alginfo_t *alt;
 727 
 728                 alt = ipss->ipsec_alglists[at][ipss->ipsec_sortlist[at][i]];
 729                 /*
 730                  * If you want to give precedence to newly added algs,
 731                  * add the = in the > comparison.
 732                  */
 733                 if ((holder != algid) || (ai->alg_minbits > alt->alg_minbits)) {
 734                         /* Swap sortlist[i] and holder. */
 735                         swap = ipss->ipsec_sortlist[at][i];
 736                         ipss->ipsec_sortlist[at][i] = holder;
 737                         holder = swap;
 738                         ai = alt;
 739                 } /* Else just continue. */
 740         }
 741 
 742         /* Store holder in last slot. */
 743         ipss->ipsec_sortlist[at][i] = holder;
 744 }
 745 
 746 /*
 747  * Remove an algorithm from a sorted algorithm list.
 748  * This should be considerably easier, even with complex sorting.
 749  */
 750 static void
 751 alg_remove_sortlist(enum ipsec_algtype at, uint8_t algid, netstack_t *ns)
 752 {
 753         boolean_t copyback = B_FALSE;
 754         int i;
 755         ipsec_stack_t   *ipss = ns->netstack_ipsec;
 756         int newcount = ipss->ipsec_nalgs[at];
 757 
 758         ASSERT(RW_WRITE_HELD(&ipss->ipsec_alg_lock));
 759 
 760         for (i = 0; i <= newcount; i++) {
 761                 if (copyback) {
 762                         ipss->ipsec_sortlist[at][i-1] =
 763                             ipss->ipsec_sortlist[at][i];
 764                 } else if (ipss->ipsec_sortlist[at][i] == algid) {
 765                         copyback = B_TRUE;
 766                 }
 767         }
 768 }
 769 
 770 /*
 771  * Add the specified algorithm to the algorithm tables.
 772  * Must be called while holding the algorithm table writer lock.
 773  */
 774 void
 775 ipsec_alg_reg(ipsec_algtype_t algtype, ipsec_alginfo_t *alg, netstack_t *ns)
 776 {
 777         ipsec_stack_t   *ipss = ns->netstack_ipsec;
 778 
 779         ASSERT(RW_WRITE_HELD(&ipss->ipsec_alg_lock));
 780 
 781         ASSERT(ipss->ipsec_alglists[algtype][alg->alg_id] == NULL);
 782         ipsec_alg_fix_min_max(alg, algtype, ns);
 783         ipss->ipsec_alglists[algtype][alg->alg_id] = alg;
 784 
 785         ipss->ipsec_nalgs[algtype]++;
 786         alg_insert_sortlist(algtype, alg->alg_id, ns);
 787 }
 788 
 789 /*
 790  * Remove the specified algorithm from the algorithm tables.
 791  * Must be called while holding the algorithm table writer lock.
 792  */
 793 void
 794 ipsec_alg_unreg(ipsec_algtype_t algtype, uint8_t algid, netstack_t *ns)
 795 {
 796         ipsec_stack_t   *ipss = ns->netstack_ipsec;
 797 
 798         ASSERT(RW_WRITE_HELD(&ipss->ipsec_alg_lock));
 799 
 800         ASSERT(ipss->ipsec_alglists[algtype][algid] != NULL);
 801         ipsec_alg_free(ipss->ipsec_alglists[algtype][algid]);
 802         ipss->ipsec_alglists[algtype][algid] = NULL;
 803 
 804         ipss->ipsec_nalgs[algtype]--;
 805         alg_remove_sortlist(algtype, algid, ns);
 806 }
 807 
 808 /*
 809  * Hooks for spdsock to get a grip on system policy.
 810  */
 811 
 812 ipsec_policy_head_t *
 813 ipsec_system_policy(netstack_t *ns)
 814 {
 815         ipsec_stack_t   *ipss = ns->netstack_ipsec;
 816         ipsec_policy_head_t *h = &ipss->ipsec_system_policy;
 817 
 818         IPPH_REFHOLD(h);
 819         return (h);
 820 }
 821 
 822 ipsec_policy_head_t *
 823 ipsec_inactive_policy(netstack_t *ns)
 824 {
 825         ipsec_stack_t   *ipss = ns->netstack_ipsec;
 826         ipsec_policy_head_t *h = &ipss->ipsec_inactive_policy;
 827 
 828         IPPH_REFHOLD(h);
 829         return (h);
 830 }
 831 
 832 /*
 833  * Lock inactive policy, then active policy, then exchange policy root
 834  * pointers.
 835  */
 836 void
 837 ipsec_swap_policy(ipsec_policy_head_t *active, ipsec_policy_head_t *inactive,
 838     netstack_t *ns)
 839 {
 840         int af, dir;
 841         avl_tree_t r1, r2;
 842 
 843         rw_enter(&inactive->iph_lock, RW_WRITER);
 844         rw_enter(&active->iph_lock, RW_WRITER);
 845 
 846         r1 = active->iph_rulebyid;
 847         r2 = inactive->iph_rulebyid;
 848         active->iph_rulebyid = r2;
 849         inactive->iph_rulebyid = r1;
 850 
 851         for (dir = 0; dir < IPSEC_NTYPES; dir++) {
 852                 ipsec_policy_hash_t *h1, *h2;
 853 
 854                 h1 = active->iph_root[dir].ipr_hash;
 855                 h2 = inactive->iph_root[dir].ipr_hash;
 856                 active->iph_root[dir].ipr_hash = h2;
 857                 inactive->iph_root[dir].ipr_hash = h1;
 858 
 859                 for (af = 0; af < IPSEC_NAF; af++) {
 860                         ipsec_policy_t *t1, *t2;
 861 
 862                         t1 = active->iph_root[dir].ipr_nonhash[af];
 863                         t2 = inactive->iph_root[dir].ipr_nonhash[af];
 864                         active->iph_root[dir].ipr_nonhash[af] = t2;
 865                         inactive->iph_root[dir].ipr_nonhash[af] = t1;
 866                         if (t1 != NULL) {
 867                                 t1->ipsp_hash.hash_pp =
 868                                     &(inactive->iph_root[dir].ipr_nonhash[af]);
 869                         }
 870                         if (t2 != NULL) {
 871                                 t2->ipsp_hash.hash_pp =
 872                                     &(active->iph_root[dir].ipr_nonhash[af]);
 873                         }
 874 
 875                 }
 876         }
 877         active->iph_gen++;
 878         inactive->iph_gen++;
 879         ipsec_update_present_flags(ns->netstack_ipsec);
 880         rw_exit(&active->iph_lock);
 881         rw_exit(&inactive->iph_lock);
 882 }
 883 
 884 /*
 885  * Swap global policy primary/secondary.
 886  */
 887 void
 888 ipsec_swap_global_policy(netstack_t *ns)
 889 {
 890         ipsec_stack_t   *ipss = ns->netstack_ipsec;
 891 
 892         ipsec_swap_policy(&ipss->ipsec_system_policy,
 893             &ipss->ipsec_inactive_policy, ns);
 894 }
 895 
 896 /*
 897  * Clone one policy rule..
 898  */
 899 static ipsec_policy_t *
 900 ipsec_copy_policy(const ipsec_policy_t *src)
 901 {
 902         ipsec_policy_t *dst = kmem_cache_alloc(ipsec_pol_cache, KM_NOSLEEP);
 903 
 904         if (dst == NULL)
 905                 return (NULL);
 906 
 907         /*
 908          * Adjust refcounts of cloned state.
 909          */
 910         IPACT_REFHOLD(src->ipsp_act);
 911         src->ipsp_sel->ipsl_refs++;
 912 
 913         HASH_NULL(dst, ipsp_hash);
 914         dst->ipsp_netstack = src->ipsp_netstack;
 915         dst->ipsp_refs = 1;
 916         dst->ipsp_sel = src->ipsp_sel;
 917         dst->ipsp_act = src->ipsp_act;
 918         dst->ipsp_prio = src->ipsp_prio;
 919         dst->ipsp_index = src->ipsp_index;
 920 
 921         return (dst);
 922 }
 923 
 924 void
 925 ipsec_insert_always(avl_tree_t *tree, void *new_node)
 926 {
 927         void *node;
 928         avl_index_t where;
 929 
 930         node = avl_find(tree, new_node, &where);
 931         ASSERT(node == NULL);
 932         avl_insert(tree, new_node, where);
 933 }
 934 
 935 
 936 static int
 937 ipsec_copy_chain(ipsec_policy_head_t *dph, ipsec_policy_t *src,
 938     ipsec_policy_t **dstp)
 939 {
 940         for (; src != NULL; src = src->ipsp_hash.hash_next) {
 941                 ipsec_policy_t *dst = ipsec_copy_policy(src);
 942                 if (dst == NULL)
 943                         return (ENOMEM);
 944 
 945                 HASHLIST_INSERT(dst, ipsp_hash, *dstp);
 946                 ipsec_insert_always(&dph->iph_rulebyid, dst);
 947         }
 948         return (0);
 949 }
 950 
 951 
 952 
 953 /*
 954  * Make one policy head look exactly like another.
 955  *
 956  * As with ipsec_swap_policy, we lock the destination policy head first, then
 957  * the source policy head. Note that we only need to read-lock the source
 958  * policy head as we are not changing it.
 959  */
 960 int
 961 ipsec_copy_polhead(ipsec_policy_head_t *sph, ipsec_policy_head_t *dph,
 962     netstack_t *ns)
 963 {
 964         int af, dir, chain, nchains;
 965 
 966         rw_enter(&dph->iph_lock, RW_WRITER);
 967 
 968         ipsec_polhead_flush(dph, ns);
 969 
 970         rw_enter(&sph->iph_lock, RW_READER);
 971 
 972         for (dir = 0; dir < IPSEC_NTYPES; dir++) {
 973                 ipsec_policy_root_t *dpr = &dph->iph_root[dir];
 974                 ipsec_policy_root_t *spr = &sph->iph_root[dir];
 975                 nchains = dpr->ipr_nchains;
 976 
 977                 ASSERT(dpr->ipr_nchains == spr->ipr_nchains);
 978 
 979                 for (af = 0; af < IPSEC_NAF; af++) {
 980                         if (ipsec_copy_chain(dph, spr->ipr_nonhash[af],
 981                             &dpr->ipr_nonhash[af]))
 982                                 goto abort_copy;
 983                 }
 984 
 985                 for (chain = 0; chain < nchains; chain++) {
 986                         if (ipsec_copy_chain(dph,
 987                             spr->ipr_hash[chain].hash_head,
 988                             &dpr->ipr_hash[chain].hash_head))
 989                                 goto abort_copy;
 990                 }
 991         }
 992 
 993         dph->iph_gen++;
 994 
 995         rw_exit(&sph->iph_lock);
 996         rw_exit(&dph->iph_lock);
 997         return (0);
 998 
 999 abort_copy:
1000         ipsec_polhead_flush(dph, ns);
1001         rw_exit(&sph->iph_lock);
1002         rw_exit(&dph->iph_lock);
1003         return (ENOMEM);
1004 }
1005 
1006 /*
1007  * Clone currently active policy to the inactive policy list.
1008  */
1009 int
1010 ipsec_clone_system_policy(netstack_t *ns)
1011 {
1012         ipsec_stack_t   *ipss = ns->netstack_ipsec;
1013 
1014         return (ipsec_copy_polhead(&ipss->ipsec_system_policy,
1015             &ipss->ipsec_inactive_policy, ns));
1016 }
1017 
1018 /*
1019  * Extract the string from ipsec_policy_failure_msgs[type] and
1020  * log it.
1021  *
1022  */
1023 void
1024 ipsec_log_policy_failure(int type, char *func_name, ipha_t *ipha, ip6_t *ip6h,
1025     boolean_t secure, netstack_t *ns)
1026 {
1027         char    sbuf[INET6_ADDRSTRLEN];
1028         char    dbuf[INET6_ADDRSTRLEN];
1029         char    *s;
1030         char    *d;
1031         ipsec_stack_t   *ipss = ns->netstack_ipsec;
1032 
1033         ASSERT((ipha == NULL && ip6h != NULL) ||
1034             (ip6h == NULL && ipha != NULL));
1035 
1036         if (ipha != NULL) {
1037                 s = inet_ntop(AF_INET, &ipha->ipha_src, sbuf, sizeof (sbuf));
1038                 d = inet_ntop(AF_INET, &ipha->ipha_dst, dbuf, sizeof (dbuf));
1039         } else {
1040                 s = inet_ntop(AF_INET6, &ip6h->ip6_src, sbuf, sizeof (sbuf));
1041                 d = inet_ntop(AF_INET6, &ip6h->ip6_dst, dbuf, sizeof (dbuf));
1042 
1043         }
1044 
1045         /* Always bump the policy failure counter. */
1046         ipss->ipsec_policy_failure_count[type]++;
1047 
1048         ipsec_rl_strlog(ns, IP_MOD_ID, 0, 0, SL_ERROR|SL_WARN|SL_CONSOLE,
1049             ipsec_policy_failure_msgs[type], func_name,
1050             (secure ? "secure" : "not secure"), s, d);
1051 }
1052 
1053 /*
1054  * Rate-limiting front-end to strlog() for AH and ESP.  Uses the ndd variables
1055  * in /dev/ip and the same rate-limiting clock so that there's a single
1056  * knob to turn to throttle the rate of messages.
1057  */
1058 void
1059 ipsec_rl_strlog(netstack_t *ns, short mid, short sid, char level, ushort_t sl,
1060     char *fmt, ...)
1061 {
1062         va_list adx;
1063         hrtime_t current = gethrtime();
1064         ip_stack_t      *ipst = ns->netstack_ip;
1065         ipsec_stack_t   *ipss = ns->netstack_ipsec;
1066 
1067         sl |= SL_CONSOLE;
1068         /*
1069          * Throttle logging to stop syslog from being swamped. If variable
1070          * 'ipsec_policy_log_interval' is zero, don't log any messages at
1071          * all, otherwise log only one message every 'ipsec_policy_log_interval'
1072          * msec. Convert interval (in msec) to hrtime (in nsec).
1073          */
1074 
1075         if (ipst->ips_ipsec_policy_log_interval) {
1076                 if (ipss->ipsec_policy_failure_last +
1077                     MSEC2NSEC(ipst->ips_ipsec_policy_log_interval) <= current) {
1078                         va_start(adx, fmt);
1079                         (void) vstrlog(mid, sid, level, sl, fmt, adx);
1080                         va_end(adx);
1081                         ipss->ipsec_policy_failure_last = current;
1082                 }
1083         }
1084 }
1085 
1086 void
1087 ipsec_config_flush(netstack_t *ns)
1088 {
1089         ipsec_stack_t   *ipss = ns->netstack_ipsec;
1090 
1091         rw_enter(&ipss->ipsec_system_policy.iph_lock, RW_WRITER);
1092         ipsec_polhead_flush(&ipss->ipsec_system_policy, ns);
1093         ipss->ipsec_next_policy_index = 1;
1094         rw_exit(&ipss->ipsec_system_policy.iph_lock);
1095         ipsec_action_reclaim_stack(ipss);
1096 }
1097 
1098 /*
1099  * Clip a policy's min/max keybits vs. the capabilities of the
1100  * algorithm.
1101  */
1102 static void
1103 act_alg_adjust(uint_t algtype, uint_t algid,
1104     uint16_t *minbits, uint16_t *maxbits, netstack_t *ns)
1105 {
1106         ipsec_stack_t   *ipss = ns->netstack_ipsec;
1107         ipsec_alginfo_t *algp = ipss->ipsec_alglists[algtype][algid];
1108 
1109         if (algp != NULL) {
1110                 /*
1111                  * If passed-in minbits is zero, we assume the caller trusts
1112                  * us with setting the minimum key size.  We pick the
1113                  * algorithms DEFAULT key size for the minimum in this case.
1114                  */
1115                 if (*minbits == 0) {
1116                         *minbits = algp->alg_default_bits;
1117                         ASSERT(*minbits >= algp->alg_minbits);
1118                 } else {
1119                         *minbits = MAX(MIN(*minbits, algp->alg_maxbits),
1120                             algp->alg_minbits);
1121                 }
1122                 if (*maxbits == 0)
1123                         *maxbits = algp->alg_maxbits;
1124                 else
1125                         *maxbits = MIN(MAX(*maxbits, algp->alg_minbits),
1126                             algp->alg_maxbits);
1127                 ASSERT(*minbits <= *maxbits);
1128         } else {
1129                 *minbits = 0;
1130                 *maxbits = 0;
1131         }
1132 }
1133 
1134 /*
1135  * Check an action's requested algorithms against the algorithms currently
1136  * loaded in the system.
1137  */
1138 boolean_t
1139 ipsec_check_action(ipsec_act_t *act, int *diag, netstack_t *ns)
1140 {
1141         ipsec_prot_t *ipp;
1142         ipsec_stack_t   *ipss = ns->netstack_ipsec;
1143 
1144         ipp = &act->ipa_apply;
1145 
1146         if (ipp->ipp_use_ah &&
1147             ipss->ipsec_alglists[IPSEC_ALG_AUTH][ipp->ipp_auth_alg] == NULL) {
1148                 *diag = SPD_DIAGNOSTIC_UNSUPP_AH_ALG;
1149                 return (B_FALSE);
1150         }
1151         if (ipp->ipp_use_espa &&
1152             ipss->ipsec_alglists[IPSEC_ALG_AUTH][ipp->ipp_esp_auth_alg] ==
1153             NULL) {
1154                 *diag = SPD_DIAGNOSTIC_UNSUPP_ESP_AUTH_ALG;
1155                 return (B_FALSE);
1156         }
1157         if (ipp->ipp_use_esp &&
1158             ipss->ipsec_alglists[IPSEC_ALG_ENCR][ipp->ipp_encr_alg] == NULL) {
1159                 *diag = SPD_DIAGNOSTIC_UNSUPP_ESP_ENCR_ALG;
1160                 return (B_FALSE);
1161         }
1162 
1163         act_alg_adjust(IPSEC_ALG_AUTH, ipp->ipp_auth_alg,
1164             &ipp->ipp_ah_minbits, &ipp->ipp_ah_maxbits, ns);
1165         act_alg_adjust(IPSEC_ALG_AUTH, ipp->ipp_esp_auth_alg,
1166             &ipp->ipp_espa_minbits, &ipp->ipp_espa_maxbits, ns);
1167         act_alg_adjust(IPSEC_ALG_ENCR, ipp->ipp_encr_alg,
1168             &ipp->ipp_espe_minbits, &ipp->ipp_espe_maxbits, ns);
1169 
1170         if (ipp->ipp_ah_minbits > ipp->ipp_ah_maxbits) {
1171                 *diag = SPD_DIAGNOSTIC_UNSUPP_AH_KEYSIZE;
1172                 return (B_FALSE);
1173         }
1174         if (ipp->ipp_espa_minbits > ipp->ipp_espa_maxbits) {
1175                 *diag = SPD_DIAGNOSTIC_UNSUPP_ESP_AUTH_KEYSIZE;
1176                 return (B_FALSE);
1177         }
1178         if (ipp->ipp_espe_minbits > ipp->ipp_espe_maxbits) {
1179                 *diag = SPD_DIAGNOSTIC_UNSUPP_ESP_ENCR_KEYSIZE;
1180                 return (B_FALSE);
1181         }
1182         /* TODO: sanity check lifetimes */
1183         return (B_TRUE);
1184 }
1185 
1186 /*
1187  * Set up a single action during wildcard expansion..
1188  */
1189 static void
1190 ipsec_setup_act(ipsec_act_t *outact, ipsec_act_t *act,
1191     uint_t auth_alg, uint_t encr_alg, uint_t eauth_alg, netstack_t *ns)
1192 {
1193         ipsec_prot_t *ipp;
1194 
1195         *outact = *act;
1196         ipp = &outact->ipa_apply;
1197         ipp->ipp_auth_alg = (uint8_t)auth_alg;
1198         ipp->ipp_encr_alg = (uint8_t)encr_alg;
1199         ipp->ipp_esp_auth_alg = (uint8_t)eauth_alg;
1200 
1201         act_alg_adjust(IPSEC_ALG_AUTH, auth_alg,
1202             &ipp->ipp_ah_minbits, &ipp->ipp_ah_maxbits, ns);
1203         act_alg_adjust(IPSEC_ALG_AUTH, eauth_alg,
1204             &ipp->ipp_espa_minbits, &ipp->ipp_espa_maxbits, ns);
1205         act_alg_adjust(IPSEC_ALG_ENCR, encr_alg,
1206             &ipp->ipp_espe_minbits, &ipp->ipp_espe_maxbits, ns);
1207 }
1208 
1209 /*
1210  * combinatoric expansion time: expand a wildcarded action into an
1211  * array of wildcarded actions; we return the exploded action list,
1212  * and return a count in *nact (output only).
1213  */
1214 static ipsec_act_t *
1215 ipsec_act_wildcard_expand(ipsec_act_t *act, uint_t *nact, netstack_t *ns)
1216 {
1217         boolean_t use_ah, use_esp, use_espa;
1218         boolean_t wild_auth, wild_encr, wild_eauth;
1219         uint_t  auth_alg, auth_idx, auth_min, auth_max;
1220         uint_t  eauth_alg, eauth_idx, eauth_min, eauth_max;
1221         uint_t  encr_alg, encr_idx, encr_min, encr_max;
1222         uint_t  action_count, ai;
1223         ipsec_act_t *outact;
1224         ipsec_stack_t   *ipss = ns->netstack_ipsec;
1225 
1226         if (act->ipa_type != IPSEC_ACT_APPLY) {
1227                 outact = kmem_alloc(sizeof (*act), KM_NOSLEEP);
1228                 *nact = 1;
1229                 if (outact != NULL)
1230                         bcopy(act, outact, sizeof (*act));
1231                 return (outact);
1232         }
1233         /*
1234          * compute the combinatoric explosion..
1235          *
1236          * we assume a request for encr if esp_req is PREF_REQUIRED
1237          * we assume a request for ah auth if ah_req is PREF_REQUIRED.
1238          * we assume a request for esp auth if !ah and esp_req is PREF_REQUIRED
1239          */
1240 
1241         use_ah = act->ipa_apply.ipp_use_ah;
1242         use_esp = act->ipa_apply.ipp_use_esp;
1243         use_espa = act->ipa_apply.ipp_use_espa;
1244         auth_alg = act->ipa_apply.ipp_auth_alg;
1245         eauth_alg = act->ipa_apply.ipp_esp_auth_alg;
1246         encr_alg = act->ipa_apply.ipp_encr_alg;
1247 
1248         wild_auth = use_ah && (auth_alg == 0);
1249         wild_eauth = use_espa && (eauth_alg == 0);
1250         wild_encr = use_esp && (encr_alg == 0);
1251 
1252         action_count = 1;
1253         auth_min = auth_max = auth_alg;
1254         eauth_min = eauth_max = eauth_alg;
1255         encr_min = encr_max = encr_alg;
1256 
1257         /*
1258          * set up for explosion.. for each dimension, expand output
1259          * size by the explosion factor.
1260          *
1261          * Don't include the "any" algorithms, if defined, as no
1262          * kernel policies should be set for these algorithms.
1263          */
1264 
1265 #define SET_EXP_MINMAX(type, wild, alg, min, max, ipss)         \
1266         if (wild) {                                             \
1267                 int nalgs = ipss->ipsec_nalgs[type];         \
1268                 if (ipss->ipsec_alglists[type][alg] != NULL) \
1269                         nalgs--;                                \
1270                 action_count *= nalgs;                          \
1271                 min = 0;                                        \
1272                 max = ipss->ipsec_nalgs[type] - 1;           \
1273         }
1274 
1275         SET_EXP_MINMAX(IPSEC_ALG_AUTH, wild_auth, SADB_AALG_NONE,
1276             auth_min, auth_max, ipss);
1277         SET_EXP_MINMAX(IPSEC_ALG_AUTH, wild_eauth, SADB_AALG_NONE,
1278             eauth_min, eauth_max, ipss);
1279         SET_EXP_MINMAX(IPSEC_ALG_ENCR, wild_encr, SADB_EALG_NONE,
1280             encr_min, encr_max, ipss);
1281 
1282 #undef  SET_EXP_MINMAX
1283 
1284         /*
1285          * ok, allocate the whole mess..
1286          */
1287 
1288         outact = kmem_alloc(sizeof (*outact) * action_count, KM_NOSLEEP);
1289         if (outact == NULL)
1290                 return (NULL);
1291 
1292         /*
1293          * Now compute all combinations.  Note that non-wildcarded
1294          * dimensions just get a single value from auth_min, while
1295          * wildcarded dimensions indirect through the sortlist.
1296          *
1297          * We do encryption outermost since, at this time, there's
1298          * greater difference in security and performance between
1299          * encryption algorithms vs. authentication algorithms.
1300          */
1301 
1302         ai = 0;
1303 
1304 #define WHICH_ALG(type, wild, idx, ipss) \
1305         ((wild)?(ipss->ipsec_sortlist[type][idx]):(idx))
1306 
1307         for (encr_idx = encr_min; encr_idx <= encr_max; encr_idx++) {
1308                 encr_alg = WHICH_ALG(IPSEC_ALG_ENCR, wild_encr, encr_idx, ipss);
1309                 if (wild_encr && encr_alg == SADB_EALG_NONE)
1310                         continue;
1311                 for (auth_idx = auth_min; auth_idx <= auth_max; auth_idx++) {
1312                         auth_alg = WHICH_ALG(IPSEC_ALG_AUTH, wild_auth,
1313                             auth_idx, ipss);
1314                         if (wild_auth && auth_alg == SADB_AALG_NONE)
1315                                 continue;
1316                         for (eauth_idx = eauth_min; eauth_idx <= eauth_max;
1317                             eauth_idx++) {
1318                                 eauth_alg = WHICH_ALG(IPSEC_ALG_AUTH,
1319                                     wild_eauth, eauth_idx, ipss);
1320                                 if (wild_eauth && eauth_alg == SADB_AALG_NONE)
1321                                         continue;
1322 
1323                                 ipsec_setup_act(&outact[ai], act,
1324                                     auth_alg, encr_alg, eauth_alg, ns);
1325                                 ai++;
1326                         }
1327                 }
1328         }
1329 
1330 #undef WHICH_ALG
1331 
1332         ASSERT(ai == action_count);
1333         *nact = action_count;
1334         return (outact);
1335 }
1336 
1337 /*
1338  * Extract the parts of an ipsec_prot_t from an old-style ipsec_req_t.
1339  */
1340 static void
1341 ipsec_prot_from_req(const ipsec_req_t *req, ipsec_prot_t *ipp)
1342 {
1343         bzero(ipp, sizeof (*ipp));
1344         /*
1345          * ipp_use_* are bitfields.  Look at "!!" in the following as a
1346          * "boolean canonicalization" operator.
1347          */
1348         ipp->ipp_use_ah = !!(req->ipsr_ah_req & IPSEC_PREF_REQUIRED);
1349         ipp->ipp_use_esp = !!(req->ipsr_esp_req & IPSEC_PREF_REQUIRED);
1350         ipp->ipp_use_espa = !!(req->ipsr_esp_auth_alg);
1351         ipp->ipp_use_se = !!(req->ipsr_self_encap_req & IPSEC_PREF_REQUIRED);
1352         ipp->ipp_use_unique = !!((req->ipsr_ah_req|req->ipsr_esp_req) &
1353             IPSEC_PREF_UNIQUE);
1354         ipp->ipp_encr_alg = req->ipsr_esp_alg;
1355         /*
1356          * SADB_AALG_ANY is a placeholder to distinguish "any" from
1357          * "none" above.  If auth is required, as determined above,
1358          * SADB_AALG_ANY becomes 0, which is the representation
1359          * of "any" and "none" in PF_KEY v2.
1360          */
1361         ipp->ipp_auth_alg = (req->ipsr_auth_alg != SADB_AALG_ANY) ?
1362             req->ipsr_auth_alg : 0;
1363         ipp->ipp_esp_auth_alg = (req->ipsr_esp_auth_alg != SADB_AALG_ANY) ?
1364             req->ipsr_esp_auth_alg : 0;
1365 }
1366 
1367 /*
1368  * Extract a new-style action from a request.
1369  */
1370 void
1371 ipsec_actvec_from_req(const ipsec_req_t *req, ipsec_act_t **actp, uint_t *nactp,
1372     netstack_t *ns)
1373 {
1374         struct ipsec_act act;
1375 
1376         bzero(&act, sizeof (act));
1377         if ((req->ipsr_ah_req & IPSEC_PREF_NEVER) &&
1378             (req->ipsr_esp_req & IPSEC_PREF_NEVER)) {
1379                 act.ipa_type = IPSEC_ACT_BYPASS;
1380         } else {
1381                 act.ipa_type = IPSEC_ACT_APPLY;
1382                 ipsec_prot_from_req(req, &act.ipa_apply);
1383         }
1384         *actp = ipsec_act_wildcard_expand(&act, nactp, ns);
1385 }
1386 
1387 /*
1388  * Convert a new-style "prot" back to an ipsec_req_t (more backwards compat).
1389  * We assume caller has already zero'ed *req for us.
1390  */
1391 static int
1392 ipsec_req_from_prot(ipsec_prot_t *ipp, ipsec_req_t *req)
1393 {
1394         req->ipsr_esp_alg = ipp->ipp_encr_alg;
1395         req->ipsr_auth_alg = ipp->ipp_auth_alg;
1396         req->ipsr_esp_auth_alg = ipp->ipp_esp_auth_alg;
1397 
1398         if (ipp->ipp_use_unique) {
1399                 req->ipsr_ah_req |= IPSEC_PREF_UNIQUE;
1400                 req->ipsr_esp_req |= IPSEC_PREF_UNIQUE;
1401         }
1402         if (ipp->ipp_use_se)
1403                 req->ipsr_self_encap_req |= IPSEC_PREF_REQUIRED;
1404         if (ipp->ipp_use_ah)
1405                 req->ipsr_ah_req |= IPSEC_PREF_REQUIRED;
1406         if (ipp->ipp_use_esp)
1407                 req->ipsr_esp_req |= IPSEC_PREF_REQUIRED;
1408         return (sizeof (*req));
1409 }
1410 
1411 /*
1412  * Convert a new-style action back to an ipsec_req_t (more backwards compat).
1413  * We assume caller has already zero'ed *req for us.
1414  */
1415 static int
1416 ipsec_req_from_act(ipsec_action_t *ap, ipsec_req_t *req)
1417 {
1418         switch (ap->ipa_act.ipa_type) {
1419         case IPSEC_ACT_BYPASS:
1420                 req->ipsr_ah_req = IPSEC_PREF_NEVER;
1421                 req->ipsr_esp_req = IPSEC_PREF_NEVER;
1422                 return (sizeof (*req));
1423         case IPSEC_ACT_APPLY:
1424                 return (ipsec_req_from_prot(&ap->ipa_act.ipa_apply, req));
1425         }
1426         return (sizeof (*req));
1427 }
1428 
1429 /*
1430  * Convert a new-style action back to an ipsec_req_t (more backwards compat).
1431  * We assume caller has already zero'ed *req for us.
1432  */
1433 int
1434 ipsec_req_from_head(ipsec_policy_head_t *ph, ipsec_req_t *req, int af)
1435 {
1436         ipsec_policy_t *p;
1437 
1438         /*
1439          * FULL-PERSOCK: consult hash table, too?
1440          */
1441         for (p = ph->iph_root[IPSEC_INBOUND].ipr_nonhash[af];
1442             p != NULL;
1443             p = p->ipsp_hash.hash_next) {
1444                 if ((p->ipsp_sel->ipsl_key.ipsl_valid & IPSL_WILDCARD) == 0)
1445                         return (ipsec_req_from_act(p->ipsp_act, req));
1446         }
1447         return (sizeof (*req));
1448 }
1449 
1450 /*
1451  * Based on per-socket or latched policy, convert to an appropriate
1452  * IP_SEC_OPT ipsec_req_t for the socket option; return size so we can
1453  * be tail-called from ip.
1454  */
1455 int
1456 ipsec_req_from_conn(conn_t *connp, ipsec_req_t *req, int af)
1457 {
1458         ipsec_latch_t *ipl;
1459         int rv = sizeof (ipsec_req_t);
1460 
1461         bzero(req, sizeof (*req));
1462 
1463         ASSERT(MUTEX_HELD(&connp->conn_lock));
1464         ipl = connp->conn_latch;
1465 
1466         /*
1467          * Find appropriate policy.  First choice is latched action;
1468          * failing that, see latched policy; failing that,
1469          * look at configured policy.
1470          */
1471         if (ipl != NULL) {
1472                 if (connp->conn_latch_in_action != NULL) {
1473                         rv = ipsec_req_from_act(connp->conn_latch_in_action,
1474                             req);
1475                         goto done;
1476                 }
1477                 if (connp->conn_latch_in_policy != NULL) {
1478                         rv = ipsec_req_from_act(
1479                             connp->conn_latch_in_policy->ipsp_act, req);
1480                         goto done;
1481                 }
1482         }
1483         if (connp->conn_policy != NULL)
1484                 rv = ipsec_req_from_head(connp->conn_policy, req, af);
1485 done:
1486         return (rv);
1487 }
1488 
1489 void
1490 ipsec_actvec_free(ipsec_act_t *act, uint_t nact)
1491 {
1492         kmem_free(act, nact * sizeof (*act));
1493 }
1494 
1495 /*
1496  * Consumes a reference to ipsp.
1497  */
1498 static mblk_t *
1499 ipsec_check_loopback_policy(mblk_t *data_mp, ip_recv_attr_t *ira,
1500     ipsec_policy_t *ipsp)
1501 {
1502         if (!(ira->ira_flags & IRAF_IPSEC_SECURE))
1503                 return (data_mp);
1504 
1505         ASSERT(ira->ira_flags & IRAF_LOOPBACK);
1506 
1507         IPPOL_REFRELE(ipsp);
1508 
1509         /*
1510          * We should do an actual policy check here.  Revisit this
1511          * when we revisit the IPsec API.  (And pass a conn_t in when we
1512          * get there.)
1513          */
1514 
1515         return (data_mp);
1516 }
1517 
1518 /*
1519  * Check that packet's inbound ports & proto match the selectors
1520  * expected by the SAs it traversed on the way in.
1521  */
1522 static boolean_t
1523 ipsec_check_ipsecin_unique(ip_recv_attr_t *ira, const char **reason,
1524     kstat_named_t **counter, uint64_t pkt_unique, netstack_t *ns)
1525 {
1526         uint64_t ah_mask, esp_mask;
1527         ipsa_t *ah_assoc;
1528         ipsa_t *esp_assoc;
1529         ipsec_stack_t   *ipss = ns->netstack_ipsec;
1530 
1531         ASSERT(ira->ira_flags & IRAF_IPSEC_SECURE);
1532         ASSERT(!(ira->ira_flags & IRAF_LOOPBACK));
1533 
1534         ah_assoc = ira->ira_ipsec_ah_sa;
1535         esp_assoc = ira->ira_ipsec_esp_sa;
1536         ASSERT((ah_assoc != NULL) || (esp_assoc != NULL));
1537 
1538         ah_mask = (ah_assoc != NULL) ? ah_assoc->ipsa_unique_mask : 0;
1539         esp_mask = (esp_assoc != NULL) ? esp_assoc->ipsa_unique_mask : 0;
1540 
1541         if ((ah_mask == 0) && (esp_mask == 0))
1542                 return (B_TRUE);
1543 
1544         /*
1545          * The pkt_unique check will also check for tunnel mode on the SA
1546          * vs. the tunneled_packet boolean.  "Be liberal in what you receive"
1547          * should not apply in this case.  ;)
1548          */
1549 
1550         if (ah_mask != 0 &&
1551             ah_assoc->ipsa_unique_id != (pkt_unique & ah_mask)) {
1552                 *reason = "AH inner header mismatch";
1553                 *counter = DROPPER(ipss, ipds_spd_ah_innermismatch);
1554                 return (B_FALSE);
1555         }
1556         if (esp_mask != 0 &&
1557             esp_assoc->ipsa_unique_id != (pkt_unique & esp_mask)) {
1558                 *reason = "ESP inner header mismatch";
1559                 *counter = DROPPER(ipss, ipds_spd_esp_innermismatch);
1560                 return (B_FALSE);
1561         }
1562         return (B_TRUE);
1563 }
1564 
1565 static boolean_t
1566 ipsec_check_ipsecin_action(ip_recv_attr_t *ira, mblk_t *mp, ipsec_action_t *ap,
1567     ipha_t *ipha, ip6_t *ip6h, const char **reason, kstat_named_t **counter,
1568     netstack_t *ns)
1569 {
1570         boolean_t ret = B_TRUE;
1571         ipsec_prot_t *ipp;
1572         ipsa_t *ah_assoc;
1573         ipsa_t *esp_assoc;
1574         boolean_t decaps;
1575         ipsec_stack_t   *ipss = ns->netstack_ipsec;
1576 
1577         ASSERT((ipha == NULL && ip6h != NULL) ||
1578             (ip6h == NULL && ipha != NULL));
1579 
1580         if (ira->ira_flags & IRAF_LOOPBACK) {
1581                 /*
1582                  * Besides accepting pointer-equivalent actions, we also
1583                  * accept any ICMP errors we generated for ourselves,
1584                  * regardless of policy.  If we do not wish to make this
1585                  * assumption in the future, check here, and where
1586                  * IXAF_TRUSTED_ICMP is initialized in ip.c and ip6.c.
1587                  */
1588                 if (ap == ira->ira_ipsec_action ||
1589                     (ira->ira_flags & IRAF_TRUSTED_ICMP))
1590                         return (B_TRUE);
1591 
1592                 /* Deep compare necessary here?? */
1593                 *counter = DROPPER(ipss, ipds_spd_loopback_mismatch);
1594                 *reason = "loopback policy mismatch";
1595                 return (B_FALSE);
1596         }
1597         ASSERT(!(ira->ira_flags & IRAF_TRUSTED_ICMP));
1598         ASSERT(ira->ira_flags & IRAF_IPSEC_SECURE);
1599 
1600         ah_assoc = ira->ira_ipsec_ah_sa;
1601         esp_assoc = ira->ira_ipsec_esp_sa;
1602 
1603         decaps = (ira->ira_flags & IRAF_IPSEC_DECAPS);
1604 
1605         switch (ap->ipa_act.ipa_type) {
1606         case IPSEC_ACT_DISCARD:
1607         case IPSEC_ACT_REJECT:
1608                 /* Should "fail hard" */
1609                 *counter = DROPPER(ipss, ipds_spd_explicit);
1610                 *reason = "blocked by policy";
1611                 return (B_FALSE);
1612 
1613         case IPSEC_ACT_BYPASS:
1614         case IPSEC_ACT_CLEAR:
1615                 *counter = DROPPER(ipss, ipds_spd_got_secure);
1616                 *reason = "expected clear, got protected";
1617                 return (B_FALSE);
1618 
1619         case IPSEC_ACT_APPLY:
1620                 ipp = &ap->ipa_act.ipa_apply;
1621                 /*
1622                  * As of now we do the simple checks of whether
1623                  * the datagram has gone through the required IPSEC
1624                  * protocol constraints or not. We might have more
1625                  * in the future like sensitive levels, key bits, etc.
1626                  * If it fails the constraints, check whether we would
1627                  * have accepted this if it had come in clear.
1628                  */
1629                 if (ipp->ipp_use_ah) {
1630                         if (ah_assoc == NULL) {
1631                                 ret = ipsec_inbound_accept_clear(mp, ipha,
1632                                     ip6h);
1633                                 *counter = DROPPER(ipss, ipds_spd_got_clear);
1634                                 *reason = "unprotected not accepted";
1635                                 break;
1636                         }
1637                         ASSERT(ah_assoc != NULL);
1638                         ASSERT(ipp->ipp_auth_alg != 0);
1639 
1640                         if (ah_assoc->ipsa_auth_alg !=
1641                             ipp->ipp_auth_alg) {
1642                                 *counter = DROPPER(ipss, ipds_spd_bad_ahalg);
1643                                 *reason = "unacceptable ah alg";
1644                                 ret = B_FALSE;
1645                                 break;
1646                         }
1647                 } else if (ah_assoc != NULL) {
1648                         /*
1649                          * Don't allow this. Check IPSEC NOTE above
1650                          * ip_fanout_proto().
1651                          */
1652                         *counter = DROPPER(ipss, ipds_spd_got_ah);
1653                         *reason = "unexpected AH";
1654                         ret = B_FALSE;
1655                         break;
1656                 }
1657                 if (ipp->ipp_use_esp) {
1658                         if (esp_assoc == NULL) {
1659                                 ret = ipsec_inbound_accept_clear(mp, ipha,
1660                                     ip6h);
1661                                 *counter = DROPPER(ipss, ipds_spd_got_clear);
1662                                 *reason = "unprotected not accepted";
1663                                 break;
1664                         }
1665                         ASSERT(esp_assoc != NULL);
1666                         ASSERT(ipp->ipp_encr_alg != 0);
1667 
1668                         if (esp_assoc->ipsa_encr_alg !=
1669                             ipp->ipp_encr_alg) {
1670                                 *counter = DROPPER(ipss, ipds_spd_bad_espealg);
1671                                 *reason = "unacceptable esp alg";
1672                                 ret = B_FALSE;
1673                                 break;
1674                         }
1675                         /*
1676                          * If the client does not need authentication,
1677                          * we don't verify the alogrithm.
1678                          */
1679                         if (ipp->ipp_use_espa) {
1680                                 if (esp_assoc->ipsa_auth_alg !=
1681                                     ipp->ipp_esp_auth_alg) {
1682                                         *counter = DROPPER(ipss,
1683                                             ipds_spd_bad_espaalg);
1684                                         *reason = "unacceptable esp auth alg";
1685                                         ret = B_FALSE;
1686                                         break;
1687                                 }
1688                         }
1689                 } else if (esp_assoc != NULL) {
1690                         /*
1691                          * Don't allow this. Check IPSEC NOTE above
1692                          * ip_fanout_proto().
1693                          */
1694                         *counter = DROPPER(ipss, ipds_spd_got_esp);
1695                         *reason = "unexpected ESP";
1696                         ret = B_FALSE;
1697                         break;
1698                 }
1699                 if (ipp->ipp_use_se) {
1700                         if (!decaps) {
1701                                 ret = ipsec_inbound_accept_clear(mp, ipha,
1702                                     ip6h);
1703                                 if (!ret) {
1704                                         /* XXX mutant? */
1705                                         *counter = DROPPER(ipss,
1706                                             ipds_spd_bad_selfencap);
1707                                         *reason = "self encap not found";
1708                                         break;
1709                                 }
1710                         }
1711                 } else if (decaps) {
1712                         /*
1713                          * XXX If the packet comes in tunneled and the
1714                          * recipient does not expect it to be tunneled, it
1715                          * is okay. But we drop to be consistent with the
1716                          * other cases.
1717                          */
1718                         *counter = DROPPER(ipss, ipds_spd_got_selfencap);
1719                         *reason = "unexpected self encap";
1720                         ret = B_FALSE;
1721                         break;
1722                 }
1723                 if (ira->ira_ipsec_action != NULL) {
1724                         /*
1725                          * This can happen if we do a double policy-check on
1726                          * a packet
1727                          * XXX XXX should fix this case!
1728                          */
1729                         IPACT_REFRELE(ira->ira_ipsec_action);
1730                 }
1731                 ASSERT(ira->ira_flags & IRAF_IPSEC_SECURE);
1732                 ASSERT(ira->ira_ipsec_action == NULL);
1733                 IPACT_REFHOLD(ap);
1734                 ira->ira_ipsec_action = ap;
1735                 break;  /* from switch */
1736         }
1737         return (ret);
1738 }
1739 
1740 static boolean_t
1741 spd_match_inbound_ids(ipsec_latch_t *ipl, ipsa_t *sa)
1742 {
1743         ASSERT(ipl->ipl_ids_latched == B_TRUE);
1744         return ipsid_equal(ipl->ipl_remote_cid, sa->ipsa_src_cid) &&
1745             ipsid_equal(ipl->ipl_local_cid, sa->ipsa_dst_cid);
1746 }
1747 
1748 /*
1749  * Takes a latched conn and an inbound packet and returns a unique_id suitable
1750  * for SA comparisons.  Most of the time we will copy from the conn_t, but
1751  * there are cases when the conn_t is latched but it has wildcard selectors,
1752  * and then we need to fallback to scooping them out of the packet.
1753  *
1754  * Assume we'll never have 0 with a conn_t present, so use 0 as a failure.  We
1755  * can get away with this because we only have non-zero ports/proto for
1756  * latched conn_ts.
1757  *
1758  * Ideal candidate for an "inline" keyword, as we're JUST convoluted enough
1759  * to not be a nice macro.
1760  */
1761 static uint64_t
1762 conn_to_unique(conn_t *connp, mblk_t *data_mp, ipha_t *ipha, ip6_t *ip6h)
1763 {
1764         ipsec_selector_t sel;
1765         uint8_t ulp = connp->conn_proto;
1766 
1767         ASSERT(connp->conn_latch_in_policy != NULL);
1768 
1769         if ((ulp == IPPROTO_TCP || ulp == IPPROTO_UDP || ulp == IPPROTO_SCTP) &&
1770             (connp->conn_fport == 0 || connp->conn_lport == 0)) {
1771                 /* Slow path - we gotta grab from the packet. */
1772                 if (ipsec_init_inbound_sel(&sel, data_mp, ipha, ip6h,
1773                     SEL_NONE) != SELRET_SUCCESS) {
1774                         /* Failure -> have caller free packet with ENOMEM. */
1775                         return (0);
1776                 }
1777                 return (SA_UNIQUE_ID(sel.ips_remote_port, sel.ips_local_port,
1778                     sel.ips_protocol, 0));
1779         }
1780 
1781 #ifdef DEBUG_NOT_UNTIL_6478464
1782         if (ipsec_init_inbound_sel(&sel, data_mp, ipha, ip6h, SEL_NONE) ==
1783             SELRET_SUCCESS) {
1784                 ASSERT(sel.ips_local_port == connp->conn_lport);
1785                 ASSERT(sel.ips_remote_port == connp->conn_fport);
1786                 ASSERT(sel.ips_protocol == connp->conn_proto);
1787         }
1788         ASSERT(connp->conn_proto != 0);
1789 #endif
1790 
1791         return (SA_UNIQUE_ID(connp->conn_fport, connp->conn_lport, ulp, 0));
1792 }
1793 
1794 /*
1795  * Called to check policy on a latched connection.
1796  * Note that we don't dereference conn_latch or conn_ihere since the conn might
1797  * be closing. The caller passes a held ipsec_latch_t instead.
1798  */
1799 static boolean_t
1800 ipsec_check_ipsecin_latch(ip_recv_attr_t *ira, mblk_t *mp, ipsec_latch_t *ipl,
1801     ipsec_action_t *ap, ipha_t *ipha, ip6_t *ip6h, const char **reason,
1802     kstat_named_t **counter, conn_t *connp, netstack_t *ns)
1803 {
1804         ipsec_stack_t   *ipss = ns->netstack_ipsec;
1805 
1806         ASSERT(ipl->ipl_ids_latched == B_TRUE);
1807         ASSERT(ira->ira_flags & IRAF_IPSEC_SECURE);
1808 
1809         if (!(ira->ira_flags & IRAF_LOOPBACK)) {
1810                 /*
1811                  * Over loopback, there aren't real security associations,
1812                  * so there are neither identities nor "unique" values
1813                  * for us to check the packet against.
1814                  */
1815                 if (ira->ira_ipsec_ah_sa != NULL) {
1816                         if (!spd_match_inbound_ids(ipl,
1817                             ira->ira_ipsec_ah_sa)) {
1818                                 *counter = DROPPER(ipss, ipds_spd_ah_badid);
1819                                 *reason = "AH identity mismatch";
1820                                 return (B_FALSE);
1821                         }
1822                 }
1823 
1824                 if (ira->ira_ipsec_esp_sa != NULL) {
1825                         if (!spd_match_inbound_ids(ipl,
1826                             ira->ira_ipsec_esp_sa)) {
1827                                 *counter = DROPPER(ipss, ipds_spd_esp_badid);
1828                                 *reason = "ESP identity mismatch";
1829                                 return (B_FALSE);
1830                         }
1831                 }
1832 
1833                 /*
1834                  * Can fudge pkt_unique from connp because we're latched.
1835                  * In DEBUG kernels (see conn_to_unique()'s implementation),
1836                  * verify this even if it REALLY slows things down.
1837                  */
1838                 if (!ipsec_check_ipsecin_unique(ira, reason, counter,
1839                     conn_to_unique(connp, mp, ipha, ip6h), ns)) {
1840                         return (B_FALSE);
1841                 }
1842         }
1843         return (ipsec_check_ipsecin_action(ira, mp, ap, ipha, ip6h, reason,
1844             counter, ns));
1845 }
1846 
1847 /*
1848  * Check to see whether this secured datagram meets the policy
1849  * constraints specified in ipsp.
1850  *
1851  * Called from ipsec_check_global_policy, and ipsec_check_inbound_policy.
1852  *
1853  * Consumes a reference to ipsp.
1854  * Returns the mblk if ok.
1855  */
1856 static mblk_t *
1857 ipsec_check_ipsecin_policy(mblk_t *data_mp, ipsec_policy_t *ipsp,
1858     ipha_t *ipha, ip6_t *ip6h, uint64_t pkt_unique, ip_recv_attr_t *ira,
1859     netstack_t *ns)
1860 {
1861         ipsec_action_t *ap;
1862         const char *reason = "no policy actions found";
1863         ip_stack_t      *ipst = ns->netstack_ip;
1864         ipsec_stack_t   *ipss = ns->netstack_ipsec;
1865         kstat_named_t *counter;
1866 
1867         counter = DROPPER(ipss, ipds_spd_got_secure);
1868 
1869         ASSERT(ipsp != NULL);
1870 
1871         ASSERT((ipha == NULL && ip6h != NULL) ||
1872             (ip6h == NULL && ipha != NULL));
1873 
1874         if (ira->ira_flags & IRAF_LOOPBACK)
1875                 return (ipsec_check_loopback_policy(data_mp, ira, ipsp));
1876 
1877         ASSERT(ira->ira_flags & IRAF_IPSEC_SECURE);
1878 
1879         if (ira->ira_ipsec_action != NULL) {
1880                 /*
1881                  * this can happen if we do a double policy-check on a packet
1882                  * Would be nice to be able to delete this test..
1883                  */
1884                 IPACT_REFRELE(ira->ira_ipsec_action);
1885         }
1886         ASSERT(ira->ira_ipsec_action == NULL);
1887 
1888         if (!SA_IDS_MATCH(ira->ira_ipsec_ah_sa, ira->ira_ipsec_esp_sa)) {
1889                 reason = "inbound AH and ESP identities differ";
1890                 counter = DROPPER(ipss, ipds_spd_ahesp_diffid);
1891                 goto drop;
1892         }
1893 
1894         if (!ipsec_check_ipsecin_unique(ira, &reason, &counter, pkt_unique,
1895             ns))
1896                 goto drop;
1897 
1898         /*
1899          * Ok, now loop through the possible actions and see if any
1900          * of them work for us.
1901          */
1902 
1903         for (ap = ipsp->ipsp_act; ap != NULL; ap = ap->ipa_next) {
1904                 if (ipsec_check_ipsecin_action(ira, data_mp, ap,
1905                     ipha, ip6h, &reason, &counter, ns)) {
1906                         BUMP_MIB(&ipst->ips_ip_mib, ipsecInSucceeded);
1907                         IPPOL_REFRELE(ipsp);
1908                         return (data_mp);
1909                 }
1910         }
1911 drop:
1912         ipsec_rl_strlog(ns, IP_MOD_ID, 0, 0, SL_ERROR|SL_WARN|SL_CONSOLE,
1913             "ipsec inbound policy mismatch: %s, packet dropped\n",
1914             reason);
1915         IPPOL_REFRELE(ipsp);
1916         ASSERT(ira->ira_ipsec_action == NULL);
1917         BUMP_MIB(&ipst->ips_ip_mib, ipsecInFailed);
1918         ip_drop_packet(data_mp, B_TRUE, NULL, counter,
1919             &ipss->ipsec_spd_dropper);
1920         return (NULL);
1921 }
1922 
1923 /*
1924  * sleazy prefix-length-based compare.
1925  * another inlining candidate..
1926  */
1927 boolean_t
1928 ip_addr_match(uint8_t *addr1, int pfxlen, in6_addr_t *addr2p)
1929 {
1930         int offset = pfxlen>>3;
1931         int bitsleft = pfxlen & 7;
1932         uint8_t *addr2 = (uint8_t *)addr2p;
1933 
1934         /*
1935          * and there was much evil..
1936          * XXX should inline-expand the bcmp here and do this 32 bits
1937          * or 64 bits at a time..
1938          */
1939         return ((bcmp(addr1, addr2, offset) == 0) &&
1940             ((bitsleft == 0) ||
1941             (((addr1[offset] ^ addr2[offset]) & (0xff<<(8-bitsleft))) == 0)));
1942 }
1943 
1944 static ipsec_policy_t *
1945 ipsec_find_policy_chain(ipsec_policy_t *best, ipsec_policy_t *chain,
1946     ipsec_selector_t *sel, boolean_t is_icmp_inv_acq)
1947 {
1948         ipsec_selkey_t *isel;
1949         ipsec_policy_t *p;
1950         int bpri = best ? best->ipsp_prio : 0;
1951 
1952         for (p = chain; p != NULL; p = p->ipsp_hash.hash_next) {
1953                 uint32_t valid;
1954 
1955                 if (p->ipsp_prio <= bpri)
1956                         continue;
1957                 isel = &p->ipsp_sel->ipsl_key;
1958                 valid = isel->ipsl_valid;
1959 
1960                 if ((valid & IPSL_PROTOCOL) &&
1961                     (isel->ipsl_proto != sel->ips_protocol))
1962                         continue;
1963 
1964                 if ((valid & IPSL_REMOTE_ADDR) &&
1965                     !ip_addr_match((uint8_t *)&isel->ipsl_remote,
1966                     isel->ipsl_remote_pfxlen, &sel->ips_remote_addr_v6))
1967                         continue;
1968 
1969                 if ((valid & IPSL_LOCAL_ADDR) &&
1970                     !ip_addr_match((uint8_t *)&isel->ipsl_local,
1971                     isel->ipsl_local_pfxlen, &sel->ips_local_addr_v6))
1972                         continue;
1973 
1974                 if ((valid & IPSL_REMOTE_PORT) &&
1975                     isel->ipsl_rport != sel->ips_remote_port)
1976                         continue;
1977 
1978                 if ((valid & IPSL_LOCAL_PORT) &&
1979                     isel->ipsl_lport != sel->ips_local_port)
1980                         continue;
1981 
1982                 if (!is_icmp_inv_acq) {
1983                         if ((valid & IPSL_ICMP_TYPE) &&
1984                             (isel->ipsl_icmp_type > sel->ips_icmp_type ||
1985                             isel->ipsl_icmp_type_end < sel->ips_icmp_type)) {
1986                                 continue;
1987                         }
1988 
1989                         if ((valid & IPSL_ICMP_CODE) &&
1990                             (isel->ipsl_icmp_code > sel->ips_icmp_code ||
1991                             isel->ipsl_icmp_code_end <
1992                             sel->ips_icmp_code)) {
1993                                 continue;
1994                         }
1995                 } else {
1996                         /*
1997                          * special case for icmp inverse acquire
1998                          * we only want policies that aren't drop/pass
1999                          */
2000                         if (p->ipsp_act->ipa_act.ipa_type != IPSEC_ACT_APPLY)
2001                                 continue;
2002                 }
2003 
2004                 /* we matched all the packet-port-field selectors! */
2005                 best = p;
2006                 bpri = p->ipsp_prio;
2007         }
2008 
2009         return (best);
2010 }
2011 
2012 /*
2013  * Try to find and return the best policy entry under a given policy
2014  * root for a given set of selectors; the first parameter "best" is
2015  * the current best policy so far.  If "best" is non-null, we have a
2016  * reference to it.  We return a reference to a policy; if that policy
2017  * is not the original "best", we need to release that reference
2018  * before returning.
2019  */
2020 ipsec_policy_t *
2021 ipsec_find_policy_head(ipsec_policy_t *best, ipsec_policy_head_t *head,
2022     int direction, ipsec_selector_t *sel)
2023 {
2024         ipsec_policy_t *curbest;
2025         ipsec_policy_root_t *root;
2026         uint8_t is_icmp_inv_acq = sel->ips_is_icmp_inv_acq;
2027         int af = sel->ips_isv4 ? IPSEC_AF_V4 : IPSEC_AF_V6;
2028 
2029         curbest = best;
2030         root = &head->iph_root[direction];
2031 
2032 #ifdef DEBUG
2033         if (is_icmp_inv_acq) {
2034                 if (sel->ips_isv4) {
2035                         if (sel->ips_protocol != IPPROTO_ICMP) {
2036                                 cmn_err(CE_WARN, "ipsec_find_policy_head:"
2037                                     " expecting icmp, got %d",
2038                                     sel->ips_protocol);
2039                         }
2040                 } else {
2041                         if (sel->ips_protocol != IPPROTO_ICMPV6) {
2042                                 cmn_err(CE_WARN, "ipsec_find_policy_head:"
2043                                     " expecting icmpv6, got %d",
2044                                     sel->ips_protocol);
2045                         }
2046                 }
2047         }
2048 #endif
2049 
2050         rw_enter(&head->iph_lock, RW_READER);
2051 
2052         if (root->ipr_nchains > 0) {
2053                 curbest = ipsec_find_policy_chain(curbest,
2054                     root->ipr_hash[selector_hash(sel, root)].hash_head, sel,
2055                     is_icmp_inv_acq);
2056         }
2057         curbest = ipsec_find_policy_chain(curbest, root->ipr_nonhash[af], sel,
2058             is_icmp_inv_acq);
2059 
2060         /*
2061          * Adjust reference counts if we found anything new.
2062          */
2063         if (curbest != best) {
2064                 ASSERT(curbest != NULL);
2065                 IPPOL_REFHOLD(curbest);
2066 
2067                 if (best != NULL) {
2068                         IPPOL_REFRELE(best);
2069                 }
2070         }
2071 
2072         rw_exit(&head->iph_lock);
2073 
2074         return (curbest);
2075 }
2076 
2077 /*
2078  * Find the best system policy (either global or per-interface) which
2079  * applies to the given selector; look in all the relevant policy roots
2080  * to figure out which policy wins.
2081  *
2082  * Returns a reference to a policy; caller must release this
2083  * reference when done.
2084  */
2085 ipsec_policy_t *
2086 ipsec_find_policy(int direction, const conn_t *connp, ipsec_selector_t *sel,
2087     netstack_t *ns)
2088 {
2089         ipsec_policy_t *p;
2090         ipsec_stack_t   *ipss = ns->netstack_ipsec;
2091 
2092         p = ipsec_find_policy_head(NULL, &ipss->ipsec_system_policy,
2093             direction, sel);
2094         if ((connp != NULL) && (connp->conn_policy != NULL)) {
2095                 p = ipsec_find_policy_head(p, connp->conn_policy,
2096                     direction, sel);
2097         }
2098 
2099         return (p);
2100 }
2101 
2102 /*
2103  * Check with global policy and see whether this inbound
2104  * packet meets the policy constraints.
2105  *
2106  * Locate appropriate policy from global policy, supplemented by the
2107  * conn's configured and/or cached policy if the conn is supplied.
2108  *
2109  * Dispatch to ipsec_check_ipsecin_policy if we have policy and an
2110  * encrypted packet to see if they match.
2111  *
2112  * Otherwise, see if the policy allows cleartext; if not, drop it on the
2113  * floor.
2114  */
2115 mblk_t *
2116 ipsec_check_global_policy(mblk_t *data_mp, conn_t *connp,
2117     ipha_t *ipha, ip6_t *ip6h, ip_recv_attr_t *ira, netstack_t *ns)
2118 {
2119         ipsec_policy_t *p;
2120         ipsec_selector_t sel;
2121         boolean_t policy_present;
2122         kstat_named_t *counter;
2123         uint64_t pkt_unique;
2124         ip_stack_t      *ipst = ns->netstack_ip;
2125         ipsec_stack_t   *ipss = ns->netstack_ipsec;
2126 
2127         sel.ips_is_icmp_inv_acq = 0;
2128 
2129         ASSERT((ipha == NULL && ip6h != NULL) ||
2130             (ip6h == NULL && ipha != NULL));
2131 
2132         if (ipha != NULL)
2133                 policy_present = ipss->ipsec_inbound_v4_policy_present;
2134         else
2135                 policy_present = ipss->ipsec_inbound_v6_policy_present;
2136 
2137         if (!policy_present && connp == NULL) {
2138                 /*
2139                  * No global policy and no per-socket policy;
2140                  * just pass it back (but we shouldn't get here in that case)
2141                  */
2142                 return (data_mp);
2143         }
2144 
2145         /*
2146          * If we have cached policy, use it.
2147          * Otherwise consult system policy.
2148          */
2149         if ((connp != NULL) && (connp->conn_latch != NULL)) {
2150                 p = connp->conn_latch_in_policy;
2151                 if (p != NULL) {
2152                         IPPOL_REFHOLD(p);
2153                 }
2154                 /*
2155                  * Fudge sel for UNIQUE_ID setting below.
2156                  */
2157                 pkt_unique = conn_to_unique(connp, data_mp, ipha, ip6h);
2158         } else {
2159                 /* Initialize the ports in the selector */
2160                 if (ipsec_init_inbound_sel(&sel, data_mp, ipha, ip6h,
2161                     SEL_NONE) == SELRET_NOMEM) {
2162                         /*
2163                          * Technically not a policy mismatch, but it is
2164                          * an internal failure.
2165                          */
2166                         ipsec_log_policy_failure(IPSEC_POLICY_MISMATCH,
2167                             "ipsec_init_inbound_sel", ipha, ip6h, B_TRUE, ns);
2168                         counter = DROPPER(ipss, ipds_spd_nomem);
2169                         goto fail;
2170                 }
2171 
2172                 /*
2173                  * Find the policy which best applies.
2174                  *
2175                  * If we find global policy, we should look at both
2176                  * local policy and global policy and see which is
2177                  * stronger and match accordingly.
2178                  *
2179                  * If we don't find a global policy, check with
2180                  * local policy alone.
2181                  */
2182 
2183                 p = ipsec_find_policy(IPSEC_TYPE_INBOUND, connp, &sel, ns);
2184                 pkt_unique = SA_UNIQUE_ID(sel.ips_remote_port,
2185                     sel.ips_local_port, sel.ips_protocol, 0);
2186         }
2187 
2188         if (p == NULL) {
2189                 if (!(ira->ira_flags & IRAF_IPSEC_SECURE)) {
2190                         /*
2191                          * We have no policy; default to succeeding.
2192                          * XXX paranoid system design doesn't do this.
2193                          */
2194                         BUMP_MIB(&ipst->ips_ip_mib, ipsecInSucceeded);
2195                         return (data_mp);
2196                 } else {
2197                         counter = DROPPER(ipss, ipds_spd_got_secure);
2198                         ipsec_log_policy_failure(IPSEC_POLICY_NOT_NEEDED,
2199                             "ipsec_check_global_policy", ipha, ip6h, B_TRUE,
2200                             ns);
2201                         goto fail;
2202                 }
2203         }
2204         if (ira->ira_flags & IRAF_IPSEC_SECURE) {
2205                 return (ipsec_check_ipsecin_policy(data_mp, p, ipha, ip6h,
2206                     pkt_unique, ira, ns));
2207         }
2208         if (p->ipsp_act->ipa_allow_clear) {
2209                 BUMP_MIB(&ipst->ips_ip_mib, ipsecInSucceeded);
2210                 IPPOL_REFRELE(p);
2211                 return (data_mp);
2212         }
2213         IPPOL_REFRELE(p);
2214         /*
2215          * If we reach here, we will drop the packet because it failed the
2216          * global policy check because the packet was cleartext, and it
2217          * should not have been.
2218          */
2219         ipsec_log_policy_failure(IPSEC_POLICY_MISMATCH,
2220             "ipsec_check_global_policy", ipha, ip6h, B_FALSE, ns);
2221         counter = DROPPER(ipss, ipds_spd_got_clear);
2222 
2223 fail:
2224         ip_drop_packet(data_mp, B_TRUE, NULL, counter,
2225             &ipss->ipsec_spd_dropper);
2226         BUMP_MIB(&ipst->ips_ip_mib, ipsecInFailed);
2227         return (NULL);
2228 }
2229 
2230 /*
2231  * We check whether an inbound datagram is a valid one
2232  * to accept in clear. If it is secure, it is the job
2233  * of IPSEC to log information appropriately if it
2234  * suspects that it may not be the real one.
2235  *
2236  * It is called only while fanning out to the ULP
2237  * where ULP accepts only secure data and the incoming
2238  * is clear. Usually we never accept clear datagrams in
2239  * such cases. ICMP is the only exception.
2240  *
2241  * NOTE : We don't call this function if the client (ULP)
2242  * is willing to accept things in clear.
2243  */
2244 boolean_t
2245 ipsec_inbound_accept_clear(mblk_t *mp, ipha_t *ipha, ip6_t *ip6h)
2246 {
2247         ushort_t iph_hdr_length;
2248         icmph_t *icmph;
2249         icmp6_t *icmp6;
2250         uint8_t *nexthdrp;
2251 
2252         ASSERT((ipha != NULL && ip6h == NULL) ||
2253             (ipha == NULL && ip6h != NULL));
2254 
2255         if (ip6h != NULL) {
2256                 iph_hdr_length = ip_hdr_length_v6(mp, ip6h);
2257                 if (!ip_hdr_length_nexthdr_v6(mp, ip6h, &iph_hdr_length,
2258                     &nexthdrp)) {
2259                         return (B_FALSE);
2260                 }
2261                 if (*nexthdrp != IPPROTO_ICMPV6)
2262                         return (B_FALSE);
2263                 icmp6 = (icmp6_t *)(&mp->b_rptr[iph_hdr_length]);
2264                 /* Match IPv6 ICMP policy as closely as IPv4 as possible. */
2265                 switch (icmp6->icmp6_type) {
2266                 case ICMP6_PARAM_PROB:
2267                         /* Corresponds to port/proto unreach in IPv4. */
2268                 case ICMP6_ECHO_REQUEST:
2269                         /* Just like IPv4. */
2270                         return (B_FALSE);
2271 
2272                 case MLD_LISTENER_QUERY:
2273                 case MLD_LISTENER_REPORT:
2274                 case MLD_LISTENER_REDUCTION:
2275                         /*
2276                          * XXX Seperate NDD in IPv4 what about here?
2277                          * Plus, mcast is important to ND.
2278                          */
2279                 case ICMP6_DST_UNREACH:
2280                         /* Corresponds to HOST/NET unreachable in IPv4. */
2281                 case ICMP6_PACKET_TOO_BIG:
2282                 case ICMP6_ECHO_REPLY:
2283                         /* These are trusted in IPv4. */
2284                 case ND_ROUTER_SOLICIT:
2285                 case ND_ROUTER_ADVERT:
2286                 case ND_NEIGHBOR_SOLICIT:
2287                 case ND_NEIGHBOR_ADVERT:
2288                 case ND_REDIRECT:
2289                         /* Trust ND messages for now. */
2290                 case ICMP6_TIME_EXCEEDED:
2291                 default:
2292                         return (B_TRUE);
2293                 }
2294         } else {
2295                 /*
2296                  * If it is not ICMP, fail this request.
2297                  */
2298                 if (ipha->ipha_protocol != IPPROTO_ICMP) {
2299 #ifdef FRAGCACHE_DEBUG
2300                         cmn_err(CE_WARN, "Dropping - ipha_proto = %d\n",
2301                             ipha->ipha_protocol);
2302 #endif
2303                         return (B_FALSE);
2304                 }
2305                 iph_hdr_length = IPH_HDR_LENGTH(ipha);
2306                 icmph = (icmph_t *)&mp->b_rptr[iph_hdr_length];
2307                 /*
2308                  * It is an insecure icmp message. Check to see whether we are
2309                  * willing to accept this one.
2310                  */
2311 
2312                 switch (icmph->icmph_type) {
2313                 case ICMP_ECHO_REPLY:
2314                 case ICMP_TIME_STAMP_REPLY:
2315                 case ICMP_INFO_REPLY:
2316                 case ICMP_ROUTER_ADVERTISEMENT:
2317                         /*
2318                          * We should not encourage clear replies if this
2319                          * client expects secure. If somebody is replying
2320                          * in clear some mailicious user watching both the
2321                          * request and reply, can do chosen-plain-text attacks.
2322                          * With global policy we might be just expecting secure
2323                          * but sending out clear. We don't know what the right
2324                          * thing is. We can't do much here as we can't control
2325                          * the sender here. Till we are sure of what to do,
2326                          * accept them.
2327                          */
2328                         return (B_TRUE);
2329                 case ICMP_ECHO_REQUEST:
2330                 case ICMP_TIME_STAMP_REQUEST:
2331                 case ICMP_INFO_REQUEST:
2332                 case ICMP_ADDRESS_MASK_REQUEST:
2333                 case ICMP_ROUTER_SOLICITATION:
2334                 case ICMP_ADDRESS_MASK_REPLY:
2335                         /*
2336                          * Don't accept this as somebody could be sending
2337                          * us plain text to get encrypted data. If we reply,
2338                          * it will lead to chosen plain text attack.
2339                          */
2340                         return (B_FALSE);
2341                 case ICMP_DEST_UNREACHABLE:
2342                         switch (icmph->icmph_code) {
2343                         case ICMP_FRAGMENTATION_NEEDED:
2344                                 /*
2345                                  * Be in sync with icmp_inbound, where we have
2346                                  * already set dce_pmtu
2347                                  */
2348 #ifdef FRAGCACHE_DEBUG
2349                         cmn_err(CE_WARN, "ICMP frag needed\n");
2350 #endif
2351                                 return (B_TRUE);
2352                         case ICMP_HOST_UNREACHABLE:
2353                         case ICMP_NET_UNREACHABLE:
2354                                 /*
2355                                  * By accepting, we could reset a connection.
2356                                  * How do we solve the problem of some
2357                                  * intermediate router sending in-secure ICMP
2358                                  * messages ?
2359                                  */
2360                                 return (B_TRUE);
2361                         case ICMP_PORT_UNREACHABLE:
2362                         case ICMP_PROTOCOL_UNREACHABLE:
2363                         default :
2364                                 return (B_FALSE);
2365                         }
2366                 case ICMP_SOURCE_QUENCH:
2367                         /*
2368                          * If this is an attack, TCP will slow start
2369                          * because of this. Is it very harmful ?
2370                          */
2371                         return (B_TRUE);
2372                 case ICMP_PARAM_PROBLEM:
2373                         return (B_FALSE);
2374                 case ICMP_TIME_EXCEEDED:
2375                         return (B_TRUE);
2376                 case ICMP_REDIRECT:
2377                         return (B_FALSE);
2378                 default :
2379                         return (B_FALSE);
2380                 }
2381         }
2382 }
2383 
2384 void
2385 ipsec_latch_ids(ipsec_latch_t *ipl, ipsid_t *local, ipsid_t *remote)
2386 {
2387         mutex_enter(&ipl->ipl_lock);
2388 
2389         if (ipl->ipl_ids_latched) {
2390                 /* I lost, someone else got here before me */
2391                 mutex_exit(&ipl->ipl_lock);
2392                 return;
2393         }
2394 
2395         if (local != NULL)
2396                 IPSID_REFHOLD(local);
2397         if (remote != NULL)
2398                 IPSID_REFHOLD(remote);
2399 
2400         ipl->ipl_local_cid = local;
2401         ipl->ipl_remote_cid = remote;
2402         ipl->ipl_ids_latched = B_TRUE;
2403         mutex_exit(&ipl->ipl_lock);
2404 }
2405 
2406 void
2407 ipsec_latch_inbound(conn_t *connp, ip_recv_attr_t *ira)
2408 {
2409         ipsa_t *sa;
2410         ipsec_latch_t *ipl = connp->conn_latch;
2411 
2412         if (!ipl->ipl_ids_latched) {
2413                 ipsid_t *local = NULL;
2414                 ipsid_t *remote = NULL;
2415 
2416                 if (!(ira->ira_flags & IRAF_LOOPBACK)) {
2417                         ASSERT(ira->ira_flags & IRAF_IPSEC_SECURE);
2418                         if (ira->ira_ipsec_esp_sa != NULL)
2419                                 sa = ira->ira_ipsec_esp_sa;
2420                         else
2421                                 sa = ira->ira_ipsec_ah_sa;
2422                         ASSERT(sa != NULL);
2423                         local = sa->ipsa_dst_cid;
2424                         remote = sa->ipsa_src_cid;
2425                 }
2426                 ipsec_latch_ids(ipl, local, remote);
2427         }
2428         if (ira->ira_flags & IRAF_IPSEC_SECURE) {
2429                 if (connp->conn_latch_in_action != NULL) {
2430                         /*
2431                          * Previously cached action.  This is probably
2432                          * harmless, but in DEBUG kernels, check for
2433                          * action equality.
2434                          *
2435                          * Preserve the existing action to preserve latch
2436                          * invariance.
2437                          */
2438                         ASSERT(connp->conn_latch_in_action ==
2439                             ira->ira_ipsec_action);
2440                         return;
2441                 }
2442                 connp->conn_latch_in_action = ira->ira_ipsec_action;
2443                 IPACT_REFHOLD(connp->conn_latch_in_action);
2444         }
2445 }
2446 
2447 /*
2448  * Check whether the policy constraints are met either for an
2449  * inbound datagram; called from IP in numerous places.
2450  *
2451  * Note that this is not a chokepoint for inbound policy checks;
2452  * see also ipsec_check_ipsecin_latch() and ipsec_check_global_policy()
2453  */
2454 mblk_t *
2455 ipsec_check_inbound_policy(mblk_t *mp, conn_t *connp,
2456     ipha_t *ipha, ip6_t *ip6h, ip_recv_attr_t *ira)
2457 {
2458         boolean_t       ret;
2459         ipsec_latch_t   *ipl;
2460         ipsec_action_t  *ap;
2461         uint64_t        unique_id;
2462         ipsec_stack_t   *ipss;
2463         ip_stack_t      *ipst;
2464         netstack_t      *ns;
2465         ipsec_policy_head_t *policy_head;
2466         ipsec_policy_t  *p = NULL;
2467 
2468         ASSERT(connp != NULL);
2469         ns = connp->conn_netstack;
2470         ipss = ns->netstack_ipsec;
2471         ipst = ns->netstack_ip;
2472 
2473         if (!(ira->ira_flags & IRAF_IPSEC_SECURE)) {
2474                 /*
2475                  * This is the case where the incoming datagram is
2476                  * cleartext and we need to see whether this client
2477                  * would like to receive such untrustworthy things from
2478                  * the wire.
2479                  */
2480                 ASSERT(mp != NULL);
2481 
2482                 mutex_enter(&connp->conn_lock);
2483                 if (connp->conn_state_flags & CONN_CONDEMNED) {
2484                         mutex_exit(&connp->conn_lock);
2485                         ip_drop_packet(mp, B_TRUE, NULL,
2486                             DROPPER(ipss, ipds_spd_got_clear),
2487                             &ipss->ipsec_spd_dropper);
2488                         BUMP_MIB(&ipst->ips_ip_mib, ipsecInFailed);
2489                         return (NULL);
2490                 }
2491                 if (connp->conn_latch != NULL) {
2492                         /* Hold a reference in case the conn is closing */
2493                         p = connp->conn_latch_in_policy;
2494                         if (p != NULL)
2495                                 IPPOL_REFHOLD(p);
2496                         mutex_exit(&connp->conn_lock);
2497                         /*
2498                          * Policy is cached in the conn.
2499                          */
2500                         if (p != NULL && !p->ipsp_act->ipa_allow_clear) {
2501                                 ret = ipsec_inbound_accept_clear(mp,
2502                                     ipha, ip6h);
2503                                 if (ret) {
2504                                         BUMP_MIB(&ipst->ips_ip_mib,
2505                                             ipsecInSucceeded);
2506                                         IPPOL_REFRELE(p);
2507                                         return (mp);
2508                                 } else {
2509                                         ipsec_log_policy_failure(
2510                                             IPSEC_POLICY_MISMATCH,
2511                                             "ipsec_check_inbound_policy", ipha,
2512                                             ip6h, B_FALSE, ns);
2513                                         ip_drop_packet(mp, B_TRUE, NULL,
2514                                             DROPPER(ipss, ipds_spd_got_clear),
2515                                             &ipss->ipsec_spd_dropper);
2516                                         BUMP_MIB(&ipst->ips_ip_mib,
2517                                             ipsecInFailed);
2518                                         IPPOL_REFRELE(p);
2519                                         return (NULL);
2520                                 }
2521                         } else {
2522                                 BUMP_MIB(&ipst->ips_ip_mib, ipsecInSucceeded);
2523                                 if (p != NULL)
2524                                         IPPOL_REFRELE(p);
2525                                 return (mp);
2526                         }
2527                 } else {
2528                         policy_head = connp->conn_policy;
2529 
2530                         /* Hold a reference in case the conn is closing */
2531                         if (policy_head != NULL)
2532                                 IPPH_REFHOLD(policy_head);
2533                         mutex_exit(&connp->conn_lock);
2534                         /*
2535                          * As this is a non-hardbound connection we need
2536                          * to look at both per-socket policy and global
2537                          * policy.
2538                          */
2539                         mp = ipsec_check_global_policy(mp, connp,
2540                             ipha, ip6h, ira, ns);
2541                         if (policy_head != NULL)
2542                                 IPPH_REFRELE(policy_head, ns);
2543                         return (mp);
2544                 }
2545         }
2546 
2547         mutex_enter(&connp->conn_lock);
2548         /* Connection is closing */
2549         if (connp->conn_state_flags & CONN_CONDEMNED) {
2550                 mutex_exit(&connp->conn_lock);
2551                 ip_drop_packet(mp, B_TRUE, NULL,
2552                     DROPPER(ipss, ipds_spd_got_clear),
2553                     &ipss->ipsec_spd_dropper);
2554                 BUMP_MIB(&ipst->ips_ip_mib, ipsecInFailed);
2555                 return (NULL);
2556         }
2557 
2558         /*
2559          * Once a connection is latched it remains so for life, the conn_latch
2560          * pointer on the conn has not changed, simply initializing ipl here
2561          * as the earlier initialization was done only in the cleartext case.
2562          */
2563         if ((ipl = connp->conn_latch) == NULL) {
2564                 mblk_t *retmp;
2565                 policy_head = connp->conn_policy;
2566 
2567                 /* Hold a reference in case the conn is closing */
2568                 if (policy_head != NULL)
2569                         IPPH_REFHOLD(policy_head);
2570                 mutex_exit(&connp->conn_lock);
2571                 /*
2572                  * We don't have policies cached in the conn
2573                  * for this stream. So, look at the global
2574                  * policy. It will check against conn or global
2575                  * depending on whichever is stronger.
2576                  */
2577                 retmp = ipsec_check_global_policy(mp, connp,
2578                     ipha, ip6h, ira, ns);
2579                 if (policy_head != NULL)
2580                         IPPH_REFRELE(policy_head, ns);
2581                 return (retmp);
2582         }
2583 
2584         IPLATCH_REFHOLD(ipl);
2585         /* Hold reference on conn_latch_in_action in case conn is closing */
2586         ap = connp->conn_latch_in_action;
2587         if (ap != NULL)
2588                 IPACT_REFHOLD(ap);
2589         mutex_exit(&connp->conn_lock);
2590 
2591         if (ap != NULL) {
2592                 /* Policy is cached & latched; fast(er) path */
2593                 const char *reason;
2594                 kstat_named_t *counter;
2595 
2596                 if (ipsec_check_ipsecin_latch(ira, mp, ipl, ap,
2597                     ipha, ip6h, &reason, &counter, connp, ns)) {
2598                         BUMP_MIB(&ipst->ips_ip_mib, ipsecInSucceeded);
2599                         IPLATCH_REFRELE(ipl);
2600                         IPACT_REFRELE(ap);
2601                         return (mp);
2602                 }
2603                 ipsec_rl_strlog(ns, IP_MOD_ID, 0, 0,
2604                     SL_ERROR|SL_WARN|SL_CONSOLE,
2605                     "ipsec inbound policy mismatch: %s, packet dropped\n",
2606                     reason);
2607                 ip_drop_packet(mp, B_TRUE, NULL, counter,
2608                     &ipss->ipsec_spd_dropper);
2609                 BUMP_MIB(&ipst->ips_ip_mib, ipsecInFailed);
2610                 IPLATCH_REFRELE(ipl);
2611                 IPACT_REFRELE(ap);
2612                 return (NULL);
2613         }
2614         if ((p = connp->conn_latch_in_policy) == NULL) {
2615                 ipsec_weird_null_inbound_policy++;
2616                 IPLATCH_REFRELE(ipl);
2617                 return (mp);
2618         }
2619 
2620         unique_id = conn_to_unique(connp, mp, ipha, ip6h);
2621         IPPOL_REFHOLD(p);
2622         mp = ipsec_check_ipsecin_policy(mp, p, ipha, ip6h, unique_id, ira, ns);
2623         /*
2624          * NOTE: ipsecIn{Failed,Succeeeded} bumped by
2625          * ipsec_check_ipsecin_policy().
2626          */
2627         if (mp != NULL)
2628                 ipsec_latch_inbound(connp, ira);
2629         IPLATCH_REFRELE(ipl);
2630         return (mp);
2631 }
2632 
2633 /*
2634  * Handle all sorts of cases like tunnel-mode and ICMP.
2635  */
2636 static int
2637 prepended_length(mblk_t *mp, uintptr_t hptr)
2638 {
2639         int rc = 0;
2640 
2641         while (mp != NULL) {
2642                 if (hptr >= (uintptr_t)mp->b_rptr && hptr <
2643                     (uintptr_t)mp->b_wptr) {
2644                         rc += (int)(hptr - (uintptr_t)mp->b_rptr);
2645                         break;  /* out of while loop */
2646                 }
2647                 rc += (int)MBLKL(mp);
2648                 mp = mp->b_cont;
2649         }
2650 
2651         if (mp == NULL) {
2652                 /*
2653                  * IF (big IF) we make it here by naturally exiting the loop,
2654                  * then ip6h isn't in the mblk chain "mp" at all.
2655                  *
2656                  * The only case where this happens is with a reversed IP
2657                  * header that gets passed up by inbound ICMP processing.
2658                  * This unfortunately triggers longstanding bug 6478464.  For
2659                  * now, just pass up 0 for the answer.
2660                  */
2661 #ifdef DEBUG_NOT_UNTIL_6478464
2662                 ASSERT(mp != NULL);
2663 #endif
2664                 rc = 0;
2665         }
2666 
2667         return (rc);
2668 }
2669 
2670 /*
2671  * Returns:
2672  *
2673  * SELRET_NOMEM --> msgpullup() needed to gather things failed.
2674  * SELRET_BADPKT --> If we're being called after tunnel-mode fragment
2675  *                   gathering, the initial fragment is too short for
2676  *                   useful data.  Only returned if SEL_TUNNEL_FIRSTFRAG is
2677  *                   set.
2678  * SELRET_SUCCESS --> "sel" now has initialized IPsec selector data.
2679  * SELRET_TUNFRAG --> This is a fragment in a tunnel-mode packet.  Caller
2680  *                    should put this packet in a fragment-gathering queue.
2681  *                    Only returned if SEL_TUNNEL_MODE and SEL_PORT_POLICY
2682  *                    is set.
2683  *
2684  * Note that ipha/ip6h can be in a different mblk (mp->b_cont) in the case
2685  * of tunneled packets.
2686  * Also, mp->b_rptr can be an ICMP error where ipha/ip6h is the packet in
2687  * error past the ICMP error.
2688  */
2689 static selret_t
2690 ipsec_init_inbound_sel(ipsec_selector_t *sel, mblk_t *mp, ipha_t *ipha,
2691     ip6_t *ip6h, uint8_t sel_flags)
2692 {
2693         uint16_t *ports;
2694         int outer_hdr_len = 0;  /* For ICMP or tunnel-mode cases... */
2695         ushort_t hdr_len;
2696         mblk_t *spare_mp = NULL;
2697         uint8_t *nexthdrp, *transportp;
2698         uint8_t nexthdr;
2699         uint8_t icmp_proto;
2700         ip_pkt_t ipp;
2701         boolean_t port_policy_present = (sel_flags & SEL_PORT_POLICY);
2702         boolean_t is_icmp = (sel_flags & SEL_IS_ICMP);
2703         boolean_t tunnel_mode = (sel_flags & SEL_TUNNEL_MODE);
2704         boolean_t post_frag = (sel_flags & SEL_POST_FRAG);
2705 
2706         ASSERT((ipha == NULL && ip6h != NULL) ||
2707             (ipha != NULL && ip6h == NULL));
2708 
2709         if (ip6h != NULL) {
2710                 outer_hdr_len = prepended_length(mp, (uintptr_t)ip6h);
2711                 nexthdr = ip6h->ip6_nxt;
2712                 icmp_proto = IPPROTO_ICMPV6;
2713                 sel->ips_isv4 = B_FALSE;
2714                 sel->ips_local_addr_v6 = ip6h->ip6_dst;
2715                 sel->ips_remote_addr_v6 = ip6h->ip6_src;
2716 
2717                 bzero(&ipp, sizeof (ipp));
2718 
2719                 switch (nexthdr) {
2720                 case IPPROTO_HOPOPTS:
2721                 case IPPROTO_ROUTING:
2722                 case IPPROTO_DSTOPTS:
2723                 case IPPROTO_FRAGMENT:
2724                         /*
2725                          * Use ip_hdr_length_nexthdr_v6().  And have a spare
2726                          * mblk that's contiguous to feed it
2727                          */
2728                         if ((spare_mp = msgpullup(mp, -1)) == NULL)
2729                                 return (SELRET_NOMEM);
2730                         if (!ip_hdr_length_nexthdr_v6(spare_mp,
2731                             (ip6_t *)(spare_mp->b_rptr + outer_hdr_len),
2732                             &hdr_len, &nexthdrp)) {
2733                                 /* Malformed packet - caller frees. */
2734                                 ipsec_freemsg_chain(spare_mp);
2735                                 return (SELRET_BADPKT);
2736                         }
2737                         /* Repopulate now that we have the whole packet */
2738                         ip6h = (ip6_t *)(spare_mp->b_rptr + outer_hdr_len);
2739                         (void) ip_find_hdr_v6(spare_mp, ip6h, B_FALSE, &ipp,
2740                             NULL);
2741                         nexthdr = *nexthdrp;
2742                         /* We can just extract based on hdr_len now. */
2743                         break;
2744                 default:
2745                         (void) ip_find_hdr_v6(mp, ip6h, B_FALSE, &ipp, NULL);
2746                         hdr_len = IPV6_HDR_LEN;
2747                         break;
2748                 }
2749                 if (port_policy_present && IS_V6_FRAGMENT(ipp) && !is_icmp) {
2750                         /* IPv6 Fragment */
2751                         ipsec_freemsg_chain(spare_mp);
2752                         return (SELRET_TUNFRAG);
2753                 }
2754                 transportp = (uint8_t *)ip6h + hdr_len;
2755         } else {
2756                 outer_hdr_len = prepended_length(mp, (uintptr_t)ipha);
2757                 icmp_proto = IPPROTO_ICMP;
2758                 sel->ips_isv4 = B_TRUE;
2759                 sel->ips_local_addr_v4 = ipha->ipha_dst;
2760                 sel->ips_remote_addr_v4 = ipha->ipha_src;
2761                 nexthdr = ipha->ipha_protocol;
2762                 hdr_len = IPH_HDR_LENGTH(ipha);
2763 
2764                 if (port_policy_present &&
2765                     IS_V4_FRAGMENT(ipha->ipha_fragment_offset_and_flags) &&
2766                     !is_icmp) {
2767                         /* IPv4 Fragment */
2768                         ipsec_freemsg_chain(spare_mp);
2769                         return (SELRET_TUNFRAG);
2770                 }
2771                 transportp = (uint8_t *)ipha + hdr_len;
2772         }
2773         sel->ips_protocol = nexthdr;
2774 
2775         if ((nexthdr != IPPROTO_TCP && nexthdr != IPPROTO_UDP &&
2776             nexthdr != IPPROTO_SCTP && nexthdr != icmp_proto) ||
2777             (!port_policy_present && !post_frag && tunnel_mode)) {
2778                 sel->ips_remote_port = sel->ips_local_port = 0;
2779                 ipsec_freemsg_chain(spare_mp);
2780                 return (SELRET_SUCCESS);
2781         }
2782 
2783         if (transportp + 4 > mp->b_wptr) {
2784                 /* If we didn't pullup a copy already, do so now. */
2785                 /*
2786                  * XXX performance, will upper-layers frequently split TCP/UDP
2787                  * apart from IP or options?  If so, perhaps we should revisit
2788                  * the spare_mp strategy.
2789                  */
2790                 ipsec_hdr_pullup_needed++;
2791                 if (spare_mp == NULL &&
2792                     (spare_mp = msgpullup(mp, -1)) == NULL) {
2793                         return (SELRET_NOMEM);
2794                 }
2795                 transportp = &spare_mp->b_rptr[hdr_len + outer_hdr_len];
2796         }
2797 
2798         if (nexthdr == icmp_proto) {
2799                 sel->ips_icmp_type = *transportp++;
2800                 sel->ips_icmp_code = *transportp;
2801                 sel->ips_remote_port = sel->ips_local_port = 0;
2802         } else {
2803                 ports = (uint16_t *)transportp;
2804                 sel->ips_remote_port = *ports++;
2805                 sel->ips_local_port = *ports;
2806         }
2807         ipsec_freemsg_chain(spare_mp);
2808         return (SELRET_SUCCESS);
2809 }
2810 
2811 /*
2812  * This is called with a b_next chain of messages from the fragcache code,
2813  * hence it needs to discard a chain on error.
2814  */
2815 static boolean_t
2816 ipsec_init_outbound_ports(ipsec_selector_t *sel, mblk_t *mp, ipha_t *ipha,
2817     ip6_t *ip6h, int outer_hdr_len, ipsec_stack_t *ipss)
2818 {
2819         /*
2820          * XXX cut&paste shared with ipsec_init_inbound_sel
2821          */
2822         uint16_t *ports;
2823         ushort_t hdr_len;
2824         mblk_t *spare_mp = NULL;
2825         uint8_t *nexthdrp;
2826         uint8_t nexthdr;
2827         uint8_t *typecode;
2828         uint8_t check_proto;
2829 
2830         ASSERT((ipha == NULL && ip6h != NULL) ||
2831             (ipha != NULL && ip6h == NULL));
2832 
2833         if (ip6h != NULL) {
2834                 check_proto = IPPROTO_ICMPV6;
2835                 nexthdr = ip6h->ip6_nxt;
2836                 switch (nexthdr) {
2837                 case IPPROTO_HOPOPTS:
2838                 case IPPROTO_ROUTING:
2839                 case IPPROTO_DSTOPTS:
2840                 case IPPROTO_FRAGMENT:
2841                         /*
2842                          * Use ip_hdr_length_nexthdr_v6().  And have a spare
2843                          * mblk that's contiguous to feed it
2844                          */
2845                         spare_mp = msgpullup(mp, -1);
2846                         if (spare_mp == NULL ||
2847                             !ip_hdr_length_nexthdr_v6(spare_mp,
2848                             (ip6_t *)(spare_mp->b_rptr + outer_hdr_len),
2849                             &hdr_len, &nexthdrp)) {
2850                                 /* Always works, even if NULL. */
2851                                 ipsec_freemsg_chain(spare_mp);
2852                                 ip_drop_packet_chain(mp, B_FALSE, NULL,
2853                                     DROPPER(ipss, ipds_spd_nomem),
2854                                     &ipss->ipsec_spd_dropper);
2855                                 return (B_FALSE);
2856                         } else {
2857                                 nexthdr = *nexthdrp;
2858                                 /* We can just extract based on hdr_len now. */
2859                         }
2860                         break;
2861                 default:
2862                         hdr_len = IPV6_HDR_LEN;
2863                         break;
2864                 }
2865         } else {
2866                 check_proto = IPPROTO_ICMP;
2867                 hdr_len = IPH_HDR_LENGTH(ipha);
2868                 nexthdr = ipha->ipha_protocol;
2869         }
2870 
2871         sel->ips_protocol = nexthdr;
2872         if (nexthdr != IPPROTO_TCP && nexthdr != IPPROTO_UDP &&
2873             nexthdr != IPPROTO_SCTP && nexthdr != check_proto) {
2874                 sel->ips_local_port = sel->ips_remote_port = 0;
2875                 ipsec_freemsg_chain(spare_mp); /* Always works, even if NULL */
2876                 return (B_TRUE);
2877         }
2878 
2879         if (&mp->b_rptr[hdr_len] + 4 + outer_hdr_len > mp->b_wptr) {
2880                 /* If we didn't pullup a copy already, do so now. */
2881                 /*
2882                  * XXX performance, will upper-layers frequently split TCP/UDP
2883                  * apart from IP or options?  If so, perhaps we should revisit
2884                  * the spare_mp strategy.
2885                  *
2886                  * XXX should this be msgpullup(mp, hdr_len+4) ???
2887                  */
2888                 if (spare_mp == NULL &&
2889                     (spare_mp = msgpullup(mp, -1)) == NULL) {
2890                         ip_drop_packet_chain(mp, B_FALSE, NULL,
2891                             DROPPER(ipss, ipds_spd_nomem),
2892                             &ipss->ipsec_spd_dropper);
2893                         return (B_FALSE);
2894                 }
2895                 ports = (uint16_t *)&spare_mp->b_rptr[hdr_len + outer_hdr_len];
2896         } else {
2897                 ports = (uint16_t *)&mp->b_rptr[hdr_len + outer_hdr_len];
2898         }
2899 
2900         if (nexthdr == check_proto) {
2901                 typecode = (uint8_t *)ports;
2902                 sel->ips_icmp_type = *typecode++;
2903                 sel->ips_icmp_code = *typecode;
2904                 sel->ips_remote_port = sel->ips_local_port = 0;
2905         } else {
2906                 sel->ips_local_port = *ports++;
2907                 sel->ips_remote_port = *ports;
2908         }
2909         ipsec_freemsg_chain(spare_mp);  /* Always works, even if NULL */
2910         return (B_TRUE);
2911 }
2912 
2913 /*
2914  * Prepend an mblk with a ipsec_crypto_t to the message chain.
2915  * Frees the argument and returns NULL should the allocation fail.
2916  * Returns the pointer to the crypto data part.
2917  */
2918 mblk_t *
2919 ipsec_add_crypto_data(mblk_t *data_mp, ipsec_crypto_t **icp)
2920 {
2921         mblk_t  *mp;
2922 
2923         mp = allocb(sizeof (ipsec_crypto_t), BPRI_MED);
2924         if (mp == NULL) {
2925                 freemsg(data_mp);
2926                 return (NULL);
2927         }
2928         bzero(mp->b_rptr, sizeof (ipsec_crypto_t));
2929         mp->b_wptr += sizeof (ipsec_crypto_t);
2930         mp->b_cont = data_mp;
2931         mp->b_datap->db_type = M_EVENT;   /* For ASSERT */
2932         *icp = (ipsec_crypto_t *)mp->b_rptr;
2933         return (mp);
2934 }
2935 
2936 /*
2937  * Remove what was prepended above. Return b_cont and a pointer to the
2938  * crypto data.
2939  * The caller must call ipsec_free_crypto_data for mblk once it is done
2940  * with the crypto data.
2941  */
2942 mblk_t *
2943 ipsec_remove_crypto_data(mblk_t *crypto_mp, ipsec_crypto_t **icp)
2944 {
2945         ASSERT(crypto_mp->b_datap->db_type == M_EVENT);
2946         ASSERT(MBLKL(crypto_mp) == sizeof (ipsec_crypto_t));
2947 
2948         *icp = (ipsec_crypto_t *)crypto_mp->b_rptr;
2949         return (crypto_mp->b_cont);
2950 }
2951 
2952 /*
2953  * Free what was prepended above. Return b_cont.
2954  */
2955 mblk_t *
2956 ipsec_free_crypto_data(mblk_t *crypto_mp)
2957 {
2958         mblk_t  *mp;
2959 
2960         ASSERT(crypto_mp->b_datap->db_type == M_EVENT);
2961         ASSERT(MBLKL(crypto_mp) == sizeof (ipsec_crypto_t));
2962 
2963         mp = crypto_mp->b_cont;
2964         freeb(crypto_mp);
2965         return (mp);
2966 }
2967 
2968 /*
2969  * Create an ipsec_action_t based on the way an inbound packet was protected.
2970  * Used to reflect traffic back to a sender.
2971  *
2972  * We don't bother interning the action into the hash table.
2973  */
2974 ipsec_action_t *
2975 ipsec_in_to_out_action(ip_recv_attr_t *ira)
2976 {
2977         ipsa_t *ah_assoc, *esp_assoc;
2978         uint_t auth_alg = 0, encr_alg = 0, espa_alg = 0;
2979         ipsec_action_t *ap;
2980         boolean_t unique;
2981 
2982         ap = kmem_cache_alloc(ipsec_action_cache, KM_NOSLEEP);
2983 
2984         if (ap == NULL)
2985                 return (NULL);
2986 
2987         bzero(ap, sizeof (*ap));
2988         HASH_NULL(ap, ipa_hash);
2989         ap->ipa_next = NULL;
2990         ap->ipa_refs = 1;
2991 
2992         /*
2993          * Get the algorithms that were used for this packet.
2994          */
2995         ap->ipa_act.ipa_type = IPSEC_ACT_APPLY;
2996         ap->ipa_act.ipa_log = 0;
2997         ASSERT(ira->ira_flags & IRAF_IPSEC_SECURE);
2998 
2999         ah_assoc = ira->ira_ipsec_ah_sa;
3000         ap->ipa_act.ipa_apply.ipp_use_ah = (ah_assoc != NULL);
3001 
3002         esp_assoc = ira->ira_ipsec_esp_sa;
3003         ap->ipa_act.ipa_apply.ipp_use_esp = (esp_assoc != NULL);
3004 
3005         if (esp_assoc != NULL) {
3006                 encr_alg = esp_assoc->ipsa_encr_alg;
3007                 espa_alg = esp_assoc->ipsa_auth_alg;
3008                 ap->ipa_act.ipa_apply.ipp_use_espa = (espa_alg != 0);
3009         }
3010         if (ah_assoc != NULL)
3011                 auth_alg = ah_assoc->ipsa_auth_alg;
3012 
3013         ap->ipa_act.ipa_apply.ipp_encr_alg = (uint8_t)encr_alg;
3014         ap->ipa_act.ipa_apply.ipp_auth_alg = (uint8_t)auth_alg;
3015         ap->ipa_act.ipa_apply.ipp_esp_auth_alg = (uint8_t)espa_alg;
3016         ap->ipa_act.ipa_apply.ipp_use_se =
3017             !!(ira->ira_flags & IRAF_IPSEC_DECAPS);
3018         unique = B_FALSE;
3019 
3020         if (esp_assoc != NULL) {
3021                 ap->ipa_act.ipa_apply.ipp_espa_minbits =
3022                     esp_assoc->ipsa_authkeybits;
3023                 ap->ipa_act.ipa_apply.ipp_espa_maxbits =
3024                     esp_assoc->ipsa_authkeybits;
3025                 ap->ipa_act.ipa_apply.ipp_espe_minbits =
3026                     esp_assoc->ipsa_encrkeybits;
3027                 ap->ipa_act.ipa_apply.ipp_espe_maxbits =
3028                     esp_assoc->ipsa_encrkeybits;
3029                 ap->ipa_act.ipa_apply.ipp_km_proto = esp_assoc->ipsa_kmp;
3030                 ap->ipa_act.ipa_apply.ipp_km_cookie = esp_assoc->ipsa_kmc;
3031                 if (esp_assoc->ipsa_flags & IPSA_F_UNIQUE)
3032                         unique = B_TRUE;
3033         }
3034         if (ah_assoc != NULL) {
3035                 ap->ipa_act.ipa_apply.ipp_ah_minbits =
3036                     ah_assoc->ipsa_authkeybits;
3037                 ap->ipa_act.ipa_apply.ipp_ah_maxbits =
3038                     ah_assoc->ipsa_authkeybits;
3039                 ap->ipa_act.ipa_apply.ipp_km_proto = ah_assoc->ipsa_kmp;
3040                 ap->ipa_act.ipa_apply.ipp_km_cookie = ah_assoc->ipsa_kmc;
3041                 if (ah_assoc->ipsa_flags & IPSA_F_UNIQUE)
3042                         unique = B_TRUE;
3043         }
3044         ap->ipa_act.ipa_apply.ipp_use_unique = unique;
3045         ap->ipa_want_unique = unique;
3046         ap->ipa_allow_clear = B_FALSE;
3047         ap->ipa_want_se = !!(ira->ira_flags & IRAF_IPSEC_DECAPS);
3048         ap->ipa_want_ah = (ah_assoc != NULL);
3049         ap->ipa_want_esp = (esp_assoc != NULL);
3050 
3051         ap->ipa_ovhd = ipsec_act_ovhd(&ap->ipa_act);
3052 
3053         ap->ipa_act.ipa_apply.ipp_replay_depth = 0; /* don't care */
3054 
3055         return (ap);
3056 }
3057 
3058 
3059 /*
3060  * Compute the worst-case amount of extra space required by an action.
3061  * Note that, because of the ESP considerations listed below, this is
3062  * actually not the same as the best-case reduction in the MTU; in the
3063  * future, we should pass additional information to this function to
3064  * allow the actual MTU impact to be computed.
3065  *
3066  * AH: Revisit this if we implement algorithms with
3067  * a verifier size of more than 12 bytes.
3068  *
3069  * ESP: A more exact but more messy computation would take into
3070  * account the interaction between the cipher block size and the
3071  * effective MTU, yielding the inner payload size which reflects a
3072  * packet with *minimum* ESP padding..
3073  */
3074 int32_t
3075 ipsec_act_ovhd(const ipsec_act_t *act)
3076 {
3077         int32_t overhead = 0;
3078 
3079         if (act->ipa_type == IPSEC_ACT_APPLY) {
3080                 const ipsec_prot_t *ipp = &act->ipa_apply;
3081 
3082                 if (ipp->ipp_use_ah)
3083                         overhead += IPSEC_MAX_AH_HDR_SIZE;
3084                 if (ipp->ipp_use_esp) {
3085                         overhead += IPSEC_MAX_ESP_HDR_SIZE;
3086                         overhead += sizeof (struct udphdr);
3087                 }
3088                 if (ipp->ipp_use_se)
3089                         overhead += IP_SIMPLE_HDR_LENGTH;
3090         }
3091         return (overhead);
3092 }
3093 
3094 /*
3095  * This hash function is used only when creating policies and thus is not
3096  * performance-critical for packet flows.
3097  *
3098  * Future work: canonicalize the structures hashed with this (i.e.,
3099  * zeroize padding) so the hash works correctly.
3100  */
3101 /* ARGSUSED */
3102 static uint32_t
3103 policy_hash(int size, const void *start, const void *end)
3104 {
3105         return (0);
3106 }
3107 
3108 
3109 /*
3110  * Hash function macros for each address type.
3111  *
3112  * The IPV6 hash function assumes that the low order 32-bits of the
3113  * address (typically containing the low order 24 bits of the mac
3114  * address) are reasonably well-distributed.  Revisit this if we run
3115  * into trouble from lots of collisions on ::1 addresses and the like
3116  * (seems unlikely).
3117  */
3118 #define IPSEC_IPV4_HASH(a, n) ((a) % (n))
3119 #define IPSEC_IPV6_HASH(a, n) (((a).s6_addr32[3]) % (n))
3120 
3121 /*
3122  * These two hash functions should produce coordinated values
3123  * but have slightly different roles.
3124  */
3125 static uint32_t
3126 selkey_hash(const ipsec_selkey_t *selkey, netstack_t *ns)
3127 {
3128         uint32_t valid = selkey->ipsl_valid;
3129         ipsec_stack_t   *ipss = ns->netstack_ipsec;
3130 
3131         if (!(valid & IPSL_REMOTE_ADDR))
3132                 return (IPSEC_SEL_NOHASH);
3133 
3134         if (valid & IPSL_IPV4) {
3135                 if (selkey->ipsl_remote_pfxlen == 32) {
3136                         return (IPSEC_IPV4_HASH(selkey->ipsl_remote.ipsad_v4,
3137                             ipss->ipsec_spd_hashsize));
3138                 }
3139         }
3140         if (valid & IPSL_IPV6) {
3141                 if (selkey->ipsl_remote_pfxlen == 128) {
3142                         return (IPSEC_IPV6_HASH(selkey->ipsl_remote.ipsad_v6,
3143                             ipss->ipsec_spd_hashsize));
3144                 }
3145         }
3146         return (IPSEC_SEL_NOHASH);
3147 }
3148 
3149 static uint32_t
3150 selector_hash(ipsec_selector_t *sel, ipsec_policy_root_t *root)
3151 {
3152         if (sel->ips_isv4) {
3153                 return (IPSEC_IPV4_HASH(sel->ips_remote_addr_v4,
3154                     root->ipr_nchains));
3155         }
3156         return (IPSEC_IPV6_HASH(sel->ips_remote_addr_v6, root->ipr_nchains));
3157 }
3158 
3159 /*
3160  * Intern actions into the action hash table.
3161  */
3162 ipsec_action_t *
3163 ipsec_act_find(const ipsec_act_t *a, int n, netstack_t *ns)
3164 {
3165         int i;
3166         uint32_t hval;
3167         ipsec_action_t *ap;
3168         ipsec_action_t *prev = NULL;
3169         int32_t overhead, maxovhd = 0;
3170         boolean_t allow_clear = B_FALSE;
3171         boolean_t want_ah = B_FALSE;
3172         boolean_t want_esp = B_FALSE;
3173         boolean_t want_se = B_FALSE;
3174         boolean_t want_unique = B_FALSE;
3175         ipsec_stack_t   *ipss = ns->netstack_ipsec;
3176 
3177         /*
3178          * TODO: should canonicalize a[] (i.e., zeroize any padding)
3179          * so we can use a non-trivial policy_hash function.
3180          */
3181         for (i = n-1; i >= 0; i--) {
3182                 hval = policy_hash(IPSEC_ACTION_HASH_SIZE, &a[i], &a[n]);
3183 
3184                 HASH_LOCK(ipss->ipsec_action_hash, hval);
3185 
3186                 for (HASH_ITERATE(ap, ipa_hash,
3187                     ipss->ipsec_action_hash, hval)) {
3188                         if (bcmp(&ap->ipa_act, &a[i], sizeof (*a)) != 0)
3189                                 continue;
3190                         if (ap->ipa_next != prev)
3191                                 continue;
3192                         break;
3193                 }
3194                 if (ap != NULL) {
3195                         HASH_UNLOCK(ipss->ipsec_action_hash, hval);
3196                         prev = ap;
3197                         continue;
3198                 }
3199                 /*
3200                  * need to allocate a new one..
3201                  */
3202                 ap = kmem_cache_alloc(ipsec_action_cache, KM_NOSLEEP);
3203                 if (ap == NULL) {
3204                         HASH_UNLOCK(ipss->ipsec_action_hash, hval);
3205                         if (prev != NULL)
3206                                 ipsec_action_free(prev);
3207                         return (NULL);
3208                 }
3209                 HASH_INSERT(ap, ipa_hash, ipss->ipsec_action_hash, hval);
3210 
3211                 ap->ipa_next = prev;
3212                 ap->ipa_act = a[i];
3213 
3214                 overhead = ipsec_act_ovhd(&a[i]);
3215                 if (maxovhd < overhead)
3216                         maxovhd = overhead;
3217 
3218                 if ((a[i].ipa_type == IPSEC_ACT_BYPASS) ||
3219                     (a[i].ipa_type == IPSEC_ACT_CLEAR))
3220                         allow_clear = B_TRUE;
3221                 if (a[i].ipa_type == IPSEC_ACT_APPLY) {
3222                         const ipsec_prot_t *ipp = &a[i].ipa_apply;
3223 
3224                         ASSERT(ipp->ipp_use_ah || ipp->ipp_use_esp);
3225                         want_ah |= ipp->ipp_use_ah;
3226                         want_esp |= ipp->ipp_use_esp;
3227                         want_se |= ipp->ipp_use_se;
3228                         want_unique |= ipp->ipp_use_unique;
3229                 }
3230                 ap->ipa_allow_clear = allow_clear;
3231                 ap->ipa_want_ah = want_ah;
3232                 ap->ipa_want_esp = want_esp;
3233                 ap->ipa_want_se = want_se;
3234                 ap->ipa_want_unique = want_unique;
3235                 ap->ipa_refs = 1; /* from the hash table */
3236                 ap->ipa_ovhd = maxovhd;
3237                 if (prev)
3238                         prev->ipa_refs++;
3239                 prev = ap;
3240                 HASH_UNLOCK(ipss->ipsec_action_hash, hval);
3241         }
3242 
3243         ap->ipa_refs++;              /* caller's reference */
3244 
3245         return (ap);
3246 }
3247 
3248 /*
3249  * Called when refcount goes to 0, indicating that all references to this
3250  * node are gone.
3251  *
3252  * This does not unchain the action from the hash table.
3253  */
3254 void
3255 ipsec_action_free(ipsec_action_t *ap)
3256 {
3257         for (;;) {
3258                 ipsec_action_t *np = ap->ipa_next;
3259                 ASSERT(ap->ipa_refs == 0);
3260                 ASSERT(ap->ipa_hash.hash_pp == NULL);
3261                 kmem_cache_free(ipsec_action_cache, ap);
3262                 ap = np;
3263                 /* Inlined IPACT_REFRELE -- avoid recursion */
3264                 if (ap == NULL)
3265                         break;
3266                 membar_exit();
3267                 if (atomic_add_32_nv(&(ap)->ipa_refs, -1) != 0)
3268                         break;
3269                 /* End inlined IPACT_REFRELE */
3270         }
3271 }
3272 
3273 /*
3274  * Called when the action hash table goes away.
3275  *
3276  * The actions can be queued on an mblk with ipsec_in or
3277  * ipsec_out, hence the actions might still be around.
3278  * But we decrement ipa_refs here since we no longer have
3279  * a reference to the action from the hash table.
3280  */
3281 static void
3282 ipsec_action_free_table(ipsec_action_t *ap)
3283 {
3284         while (ap != NULL) {
3285                 ipsec_action_t *np = ap->ipa_next;
3286 
3287                 /* FIXME: remove? */
3288                 (void) printf("ipsec_action_free_table(%p) ref %d\n",
3289                     (void *)ap, ap->ipa_refs);
3290                 ASSERT(ap->ipa_refs > 0);
3291                 IPACT_REFRELE(ap);
3292                 ap = np;
3293         }
3294 }
3295 
3296 /*
3297  * Need to walk all stack instances since the reclaim function
3298  * is global for all instances
3299  */
3300 /* ARGSUSED */
3301 static void
3302 ipsec_action_reclaim(void *arg)
3303 {
3304         netstack_handle_t nh;
3305         netstack_t *ns;
3306         ipsec_stack_t *ipss;
3307 
3308         netstack_next_init(&nh);
3309         while ((ns = netstack_next(&nh)) != NULL) {
3310                 /*
3311                  * netstack_next() can return a netstack_t with a NULL
3312                  * netstack_ipsec at boot time.
3313                  */
3314                 if ((ipss = ns->netstack_ipsec) == NULL) {
3315                         netstack_rele(ns);
3316                         continue;
3317                 }
3318                 ipsec_action_reclaim_stack(ipss);
3319                 netstack_rele(ns);
3320         }
3321         netstack_next_fini(&nh);
3322 }
3323 
3324 /*
3325  * Periodically sweep action hash table for actions with refcount==1, and
3326  * nuke them.  We cannot do this "on demand" (i.e., from IPACT_REFRELE)
3327  * because we can't close the race between another thread finding the action
3328  * in the hash table without holding the bucket lock during IPACT_REFRELE.
3329  * Instead, we run this function sporadically to clean up after ourselves;
3330  * we also set it as the "reclaim" function for the action kmem_cache.
3331  *
3332  * Note that it may take several passes of ipsec_action_gc() to free all
3333  * "stale" actions.
3334  */
3335 static void
3336 ipsec_action_reclaim_stack(ipsec_stack_t *ipss)
3337 {
3338         int i;
3339 
3340         for (i = 0; i < IPSEC_ACTION_HASH_SIZE; i++) {
3341                 ipsec_action_t *ap, *np;
3342 
3343                 /* skip the lock if nobody home */
3344                 if (ipss->ipsec_action_hash[i].hash_head == NULL)
3345                         continue;
3346 
3347                 HASH_LOCK(ipss->ipsec_action_hash, i);
3348                 for (ap = ipss->ipsec_action_hash[i].hash_head;
3349                     ap != NULL; ap = np) {
3350                         ASSERT(ap->ipa_refs > 0);
3351                         np = ap->ipa_hash.hash_next;
3352                         if (ap->ipa_refs > 1)
3353                                 continue;
3354                         HASH_UNCHAIN(ap, ipa_hash,
3355                             ipss->ipsec_action_hash, i);
3356                         IPACT_REFRELE(ap);
3357                 }
3358                 HASH_UNLOCK(ipss->ipsec_action_hash, i);
3359         }
3360 }
3361 
3362 /*
3363  * Intern a selector set into the selector set hash table.
3364  * This is simpler than the actions case..
3365  */
3366 static ipsec_sel_t *
3367 ipsec_find_sel(ipsec_selkey_t *selkey, netstack_t *ns)
3368 {
3369         ipsec_sel_t *sp;
3370         uint32_t hval, bucket;
3371         ipsec_stack_t   *ipss = ns->netstack_ipsec;
3372 
3373         /*
3374          * Exactly one AF bit should be set in selkey.
3375          */
3376         ASSERT(!(selkey->ipsl_valid & IPSL_IPV4) ^
3377             !(selkey->ipsl_valid & IPSL_IPV6));
3378 
3379         hval = selkey_hash(selkey, ns);
3380         /* Set pol_hval to uninitialized until we put it in a polhead. */
3381         selkey->ipsl_sel_hval = hval;
3382 
3383         bucket = (hval == IPSEC_SEL_NOHASH) ? 0 : hval;
3384 
3385         ASSERT(!HASH_LOCKED(ipss->ipsec_sel_hash, bucket));
3386         HASH_LOCK(ipss->ipsec_sel_hash, bucket);
3387 
3388         for (HASH_ITERATE(sp, ipsl_hash, ipss->ipsec_sel_hash, bucket)) {
3389                 if (bcmp(&sp->ipsl_key, selkey,
3390                     offsetof(ipsec_selkey_t, ipsl_pol_hval)) == 0)
3391                         break;
3392         }
3393         if (sp != NULL) {
3394                 sp->ipsl_refs++;
3395 
3396                 HASH_UNLOCK(ipss->ipsec_sel_hash, bucket);
3397                 return (sp);
3398         }
3399 
3400         sp = kmem_cache_alloc(ipsec_sel_cache, KM_NOSLEEP);
3401         if (sp == NULL) {
3402                 HASH_UNLOCK(ipss->ipsec_sel_hash, bucket);
3403                 return (NULL);
3404         }
3405 
3406         HASH_INSERT(sp, ipsl_hash, ipss->ipsec_sel_hash, bucket);
3407         sp->ipsl_refs = 2;   /* one for hash table, one for caller */
3408         sp->ipsl_key = *selkey;
3409         /* Set to uninitalized and have insertion into polhead fix things. */
3410         if (selkey->ipsl_sel_hval != IPSEC_SEL_NOHASH)
3411                 sp->ipsl_key.ipsl_pol_hval = 0;
3412         else
3413                 sp->ipsl_key.ipsl_pol_hval = IPSEC_SEL_NOHASH;
3414 
3415         HASH_UNLOCK(ipss->ipsec_sel_hash, bucket);
3416 
3417         return (sp);
3418 }
3419 
3420 static void
3421 ipsec_sel_rel(ipsec_sel_t **spp, netstack_t *ns)
3422 {
3423         ipsec_sel_t *sp = *spp;
3424         int hval = sp->ipsl_key.ipsl_sel_hval;
3425         ipsec_stack_t   *ipss = ns->netstack_ipsec;
3426 
3427         *spp = NULL;
3428 
3429         if (hval == IPSEC_SEL_NOHASH)
3430                 hval = 0;
3431 
3432         ASSERT(!HASH_LOCKED(ipss->ipsec_sel_hash, hval));
3433         HASH_LOCK(ipss->ipsec_sel_hash, hval);
3434         if (--sp->ipsl_refs == 1) {
3435                 HASH_UNCHAIN(sp, ipsl_hash, ipss->ipsec_sel_hash, hval);
3436                 sp->ipsl_refs--;
3437                 HASH_UNLOCK(ipss->ipsec_sel_hash, hval);
3438                 ASSERT(sp->ipsl_refs == 0);
3439                 kmem_cache_free(ipsec_sel_cache, sp);
3440                 /* Caller unlocks */
3441                 return;
3442         }
3443 
3444         HASH_UNLOCK(ipss->ipsec_sel_hash, hval);
3445 }
3446 
3447 /*
3448  * Free a policy rule which we know is no longer being referenced.
3449  */
3450 void
3451 ipsec_policy_free(ipsec_policy_t *ipp)
3452 {
3453         ASSERT(ipp->ipsp_refs == 0);
3454         ASSERT(ipp->ipsp_sel != NULL);
3455         ASSERT(ipp->ipsp_act != NULL);
3456         ASSERT(ipp->ipsp_netstack != NULL);
3457 
3458         ipsec_sel_rel(&ipp->ipsp_sel, ipp->ipsp_netstack);
3459         IPACT_REFRELE(ipp->ipsp_act);
3460         kmem_cache_free(ipsec_pol_cache, ipp);
3461 }
3462 
3463 /*
3464  * Construction of new policy rules; construct a policy, and add it to
3465  * the appropriate tables.
3466  */
3467 ipsec_policy_t *
3468 ipsec_policy_create(ipsec_selkey_t *keys, const ipsec_act_t *a,
3469     int nacts, int prio, uint64_t *index_ptr, netstack_t *ns)
3470 {
3471         ipsec_action_t *ap;
3472         ipsec_sel_t *sp;
3473         ipsec_policy_t *ipp;
3474         ipsec_stack_t   *ipss = ns->netstack_ipsec;
3475 
3476         if (index_ptr == NULL)
3477                 index_ptr = &ipss->ipsec_next_policy_index;
3478 
3479         ipp = kmem_cache_alloc(ipsec_pol_cache, KM_NOSLEEP);
3480         ap = ipsec_act_find(a, nacts, ns);
3481         sp = ipsec_find_sel(keys, ns);
3482 
3483         if ((ap == NULL) || (sp == NULL) || (ipp == NULL)) {
3484                 if (ap != NULL) {
3485                         IPACT_REFRELE(ap);
3486                 }
3487                 if (sp != NULL)
3488                         ipsec_sel_rel(&sp, ns);
3489                 if (ipp != NULL)
3490                         kmem_cache_free(ipsec_pol_cache, ipp);
3491                 return (NULL);
3492         }
3493 
3494         HASH_NULL(ipp, ipsp_hash);
3495 
3496         ipp->ipsp_netstack = ns;     /* Needed for ipsec_policy_free */
3497         ipp->ipsp_refs = 1;  /* caller's reference */
3498         ipp->ipsp_sel = sp;
3499         ipp->ipsp_act = ap;
3500         ipp->ipsp_prio = prio;       /* rule priority */
3501         ipp->ipsp_index = *index_ptr;
3502         (*index_ptr)++;
3503 
3504         return (ipp);
3505 }
3506 
3507 static void
3508 ipsec_update_present_flags(ipsec_stack_t *ipss)
3509 {
3510         boolean_t hashpol;
3511 
3512         hashpol = (avl_numnodes(&ipss->ipsec_system_policy.iph_rulebyid) > 0);
3513 
3514         if (hashpol) {
3515                 ipss->ipsec_outbound_v4_policy_present = B_TRUE;
3516                 ipss->ipsec_outbound_v6_policy_present = B_TRUE;
3517                 ipss->ipsec_inbound_v4_policy_present = B_TRUE;
3518                 ipss->ipsec_inbound_v6_policy_present = B_TRUE;
3519                 return;
3520         }
3521 
3522         ipss->ipsec_outbound_v4_policy_present = (NULL !=
3523             ipss->ipsec_system_policy.iph_root[IPSEC_TYPE_OUTBOUND].
3524             ipr_nonhash[IPSEC_AF_V4]);
3525         ipss->ipsec_outbound_v6_policy_present = (NULL !=
3526             ipss->ipsec_system_policy.iph_root[IPSEC_TYPE_OUTBOUND].
3527             ipr_nonhash[IPSEC_AF_V6]);
3528         ipss->ipsec_inbound_v4_policy_present = (NULL !=
3529             ipss->ipsec_system_policy.iph_root[IPSEC_TYPE_INBOUND].
3530             ipr_nonhash[IPSEC_AF_V4]);
3531         ipss->ipsec_inbound_v6_policy_present = (NULL !=
3532             ipss->ipsec_system_policy.iph_root[IPSEC_TYPE_INBOUND].
3533             ipr_nonhash[IPSEC_AF_V6]);
3534 }
3535 
3536 boolean_t
3537 ipsec_policy_delete(ipsec_policy_head_t *php, ipsec_selkey_t *keys, int dir,
3538         netstack_t *ns)
3539 {
3540         ipsec_sel_t *sp;
3541         ipsec_policy_t *ip, *nip, *head;
3542         int af;
3543         ipsec_policy_root_t *pr = &php->iph_root[dir];
3544 
3545         sp = ipsec_find_sel(keys, ns);
3546 
3547         if (sp == NULL)
3548                 return (B_FALSE);
3549 
3550         af = (sp->ipsl_key.ipsl_valid & IPSL_IPV4) ? IPSEC_AF_V4 : IPSEC_AF_V6;
3551 
3552         rw_enter(&php->iph_lock, RW_WRITER);
3553 
3554         if (sp->ipsl_key.ipsl_pol_hval == IPSEC_SEL_NOHASH) {
3555                 head = pr->ipr_nonhash[af];
3556         } else {
3557                 head = pr->ipr_hash[sp->ipsl_key.ipsl_pol_hval].hash_head;
3558         }
3559 
3560         for (ip = head; ip != NULL; ip = nip) {
3561                 nip = ip->ipsp_hash.hash_next;
3562                 if (ip->ipsp_sel != sp) {
3563                         continue;
3564                 }
3565 
3566                 IPPOL_UNCHAIN(php, ip);
3567 
3568                 php->iph_gen++;
3569                 ipsec_update_present_flags(ns->netstack_ipsec);
3570 
3571                 rw_exit(&php->iph_lock);
3572 
3573                 ipsec_sel_rel(&sp, ns);
3574 
3575                 return (B_TRUE);
3576         }
3577 
3578         rw_exit(&php->iph_lock);
3579         ipsec_sel_rel(&sp, ns);
3580         return (B_FALSE);
3581 }
3582 
3583 int
3584 ipsec_policy_delete_index(ipsec_policy_head_t *php, uint64_t policy_index,
3585     netstack_t *ns)
3586 {
3587         boolean_t found = B_FALSE;
3588         ipsec_policy_t ipkey;
3589         ipsec_policy_t *ip;
3590         avl_index_t where;
3591 
3592         bzero(&ipkey, sizeof (ipkey));
3593         ipkey.ipsp_index = policy_index;
3594 
3595         rw_enter(&php->iph_lock, RW_WRITER);
3596 
3597         /*
3598          * We could be cleverer here about the walk.
3599          * but well, (k+1)*log(N) will do for now (k==number of matches,
3600          * N==number of table entries
3601          */
3602         for (;;) {
3603                 ip = (ipsec_policy_t *)avl_find(&php->iph_rulebyid,
3604                     (void *)&ipkey, &where);
3605                 ASSERT(ip == NULL);
3606 
3607                 ip = avl_nearest(&php->iph_rulebyid, where, AVL_AFTER);
3608 
3609                 if (ip == NULL)
3610                         break;
3611 
3612                 if (ip->ipsp_index != policy_index) {
3613                         ASSERT(ip->ipsp_index > policy_index);
3614                         break;
3615                 }
3616 
3617                 IPPOL_UNCHAIN(php, ip);
3618                 found = B_TRUE;
3619         }
3620 
3621         if (found) {
3622                 php->iph_gen++;
3623                 ipsec_update_present_flags(ns->netstack_ipsec);
3624         }
3625 
3626         rw_exit(&php->iph_lock);
3627 
3628         return (found ? 0 : ENOENT);
3629 }
3630 
3631 /*
3632  * Given a constructed ipsec_policy_t policy rule, see if it can be entered
3633  * into the correct policy ruleset.  As a side-effect, it sets the hash
3634  * entries on "ipp"'s ipsp_pol_hval.
3635  *
3636  * Returns B_TRUE if it can be entered, B_FALSE if it can't be (because a
3637  * duplicate policy exists with exactly the same selectors), or an icmp
3638  * rule exists with a different encryption/authentication action.
3639  */
3640 boolean_t
3641 ipsec_check_policy(ipsec_policy_head_t *php, ipsec_policy_t *ipp, int direction)
3642 {
3643         ipsec_policy_root_t *pr = &php->iph_root[direction];
3644         int af = -1;
3645         ipsec_policy_t *p2, *head;
3646         uint8_t check_proto;
3647         ipsec_selkey_t *selkey = &ipp->ipsp_sel->ipsl_key;
3648         uint32_t        valid = selkey->ipsl_valid;
3649 
3650         if (valid & IPSL_IPV6) {
3651                 ASSERT(!(valid & IPSL_IPV4));
3652                 af = IPSEC_AF_V6;
3653                 check_proto = IPPROTO_ICMPV6;
3654         } else {
3655                 ASSERT(valid & IPSL_IPV4);
3656                 af = IPSEC_AF_V4;
3657                 check_proto = IPPROTO_ICMP;
3658         }
3659 
3660         ASSERT(RW_WRITE_HELD(&php->iph_lock));
3661 
3662         /*
3663          * Double-check that we don't have any duplicate selectors here.
3664          * Because selectors are interned below, we need only compare pointers
3665          * for equality.
3666          */
3667         if (selkey->ipsl_sel_hval == IPSEC_SEL_NOHASH) {
3668                 head = pr->ipr_nonhash[af];
3669         } else {
3670                 selkey->ipsl_pol_hval =
3671                     (selkey->ipsl_valid & IPSL_IPV4) ?
3672                     IPSEC_IPV4_HASH(selkey->ipsl_remote.ipsad_v4,
3673                     pr->ipr_nchains) :
3674                     IPSEC_IPV6_HASH(selkey->ipsl_remote.ipsad_v6,
3675                     pr->ipr_nchains);
3676 
3677                 head = pr->ipr_hash[selkey->ipsl_pol_hval].hash_head;
3678         }
3679 
3680         for (p2 = head; p2 != NULL; p2 = p2->ipsp_hash.hash_next) {
3681                 if (p2->ipsp_sel == ipp->ipsp_sel)
3682                         return (B_FALSE);
3683         }
3684 
3685         /*
3686          * If it's ICMP and not a drop or pass rule, run through the ICMP
3687          * rules and make sure the action is either new or the same as any
3688          * other actions.  We don't have to check the full chain because
3689          * discard and bypass will override all other actions
3690          */
3691 
3692         if (valid & IPSL_PROTOCOL &&
3693             selkey->ipsl_proto == check_proto &&
3694             (ipp->ipsp_act->ipa_act.ipa_type == IPSEC_ACT_APPLY)) {
3695 
3696                 for (p2 = head; p2 != NULL; p2 = p2->ipsp_hash.hash_next) {
3697 
3698                         if (p2->ipsp_sel->ipsl_key.ipsl_valid & IPSL_PROTOCOL &&
3699                             p2->ipsp_sel->ipsl_key.ipsl_proto == check_proto &&
3700                             (p2->ipsp_act->ipa_act.ipa_type ==
3701                             IPSEC_ACT_APPLY)) {
3702                                 return (ipsec_compare_action(p2, ipp));
3703                         }
3704                 }
3705         }
3706 
3707         return (B_TRUE);
3708 }
3709 
3710 /*
3711  * compare the action chains of two policies for equality
3712  * B_TRUE -> effective equality
3713  */
3714 
3715 static boolean_t
3716 ipsec_compare_action(ipsec_policy_t *p1, ipsec_policy_t *p2)
3717 {
3718 
3719         ipsec_action_t *act1, *act2;
3720 
3721         /* We have a valid rule. Let's compare the actions */
3722         if (p1->ipsp_act == p2->ipsp_act) {
3723                 /* same action. We are good */
3724                 return (B_TRUE);
3725         }
3726 
3727         /* we have to walk the chain */
3728 
3729         act1 = p1->ipsp_act;
3730         act2 = p2->ipsp_act;
3731 
3732         while (act1 != NULL && act2 != NULL) {
3733 
3734                 /* otherwise, Are we close enough? */
3735                 if (act1->ipa_allow_clear != act2->ipa_allow_clear ||
3736                     act1->ipa_want_ah != act2->ipa_want_ah ||
3737                     act1->ipa_want_esp != act2->ipa_want_esp ||
3738                     act1->ipa_want_se != act2->ipa_want_se) {
3739                         /* Nope, we aren't */
3740                         return (B_FALSE);
3741                 }
3742 
3743                 if (act1->ipa_want_ah) {
3744                         if (act1->ipa_act.ipa_apply.ipp_auth_alg !=
3745                             act2->ipa_act.ipa_apply.ipp_auth_alg) {
3746                                 return (B_FALSE);
3747                         }
3748 
3749                         if (act1->ipa_act.ipa_apply.ipp_ah_minbits !=
3750                             act2->ipa_act.ipa_apply.ipp_ah_minbits ||
3751                             act1->ipa_act.ipa_apply.ipp_ah_maxbits !=
3752                             act2->ipa_act.ipa_apply.ipp_ah_maxbits) {
3753                                 return (B_FALSE);
3754                         }
3755                 }
3756 
3757                 if (act1->ipa_want_esp) {
3758                         if (act1->ipa_act.ipa_apply.ipp_use_esp !=
3759                             act2->ipa_act.ipa_apply.ipp_use_esp ||
3760                             act1->ipa_act.ipa_apply.ipp_use_espa !=
3761                             act2->ipa_act.ipa_apply.ipp_use_espa) {
3762                                 return (B_FALSE);
3763                         }
3764 
3765                         if (act1->ipa_act.ipa_apply.ipp_use_esp) {
3766                                 if (act1->ipa_act.ipa_apply.ipp_encr_alg !=
3767                                     act2->ipa_act.ipa_apply.ipp_encr_alg) {
3768                                         return (B_FALSE);
3769                                 }
3770 
3771                                 if (act1->ipa_act.ipa_apply.ipp_espe_minbits !=
3772                                     act2->ipa_act.ipa_apply.ipp_espe_minbits ||
3773                                     act1->ipa_act.ipa_apply.ipp_espe_maxbits !=
3774                                     act2->ipa_act.ipa_apply.ipp_espe_maxbits) {
3775                                         return (B_FALSE);
3776                                 }
3777                         }
3778 
3779                         if (act1->ipa_act.ipa_apply.ipp_use_espa) {
3780                                 if (act1->ipa_act.ipa_apply.ipp_esp_auth_alg !=
3781                                     act2->ipa_act.ipa_apply.ipp_esp_auth_alg) {
3782                                         return (B_FALSE);
3783                                 }
3784 
3785                                 if (act1->ipa_act.ipa_apply.ipp_espa_minbits !=
3786                                     act2->ipa_act.ipa_apply.ipp_espa_minbits ||
3787                                     act1->ipa_act.ipa_apply.ipp_espa_maxbits !=
3788                                     act2->ipa_act.ipa_apply.ipp_espa_maxbits) {
3789                                         return (B_FALSE);
3790                                 }
3791                         }
3792 
3793                 }
3794 
3795                 act1 = act1->ipa_next;
3796                 act2 = act2->ipa_next;
3797         }
3798 
3799         if (act1 != NULL || act2 != NULL) {
3800                 return (B_FALSE);
3801         }
3802 
3803         return (B_TRUE);
3804 }
3805 
3806 
3807 /*
3808  * Given a constructed ipsec_policy_t policy rule, enter it into
3809  * the correct policy ruleset.
3810  *
3811  * ipsec_check_policy() is assumed to have succeeded first (to check for
3812  * duplicates).
3813  */
3814 void
3815 ipsec_enter_policy(ipsec_policy_head_t *php, ipsec_policy_t *ipp, int direction,
3816     netstack_t *ns)
3817 {
3818         ipsec_policy_root_t *pr = &php->iph_root[direction];
3819         ipsec_selkey_t *selkey = &ipp->ipsp_sel->ipsl_key;
3820         uint32_t valid = selkey->ipsl_valid;
3821         uint32_t hval = selkey->ipsl_pol_hval;
3822         int af = -1;
3823 
3824         ASSERT(RW_WRITE_HELD(&php->iph_lock));
3825 
3826         if (valid & IPSL_IPV6) {
3827                 ASSERT(!(valid & IPSL_IPV4));
3828                 af = IPSEC_AF_V6;
3829         } else {
3830                 ASSERT(valid & IPSL_IPV4);
3831                 af = IPSEC_AF_V4;
3832         }
3833 
3834         php->iph_gen++;
3835 
3836         if (hval == IPSEC_SEL_NOHASH) {
3837                 HASHLIST_INSERT(ipp, ipsp_hash, pr->ipr_nonhash[af]);
3838         } else {
3839                 HASH_LOCK(pr->ipr_hash, hval);
3840                 HASH_INSERT(ipp, ipsp_hash, pr->ipr_hash, hval);
3841                 HASH_UNLOCK(pr->ipr_hash, hval);
3842         }
3843 
3844         ipsec_insert_always(&php->iph_rulebyid, ipp);
3845 
3846         ipsec_update_present_flags(ns->netstack_ipsec);
3847 }
3848 
3849 static void
3850 ipsec_ipr_flush(ipsec_policy_head_t *php, ipsec_policy_root_t *ipr)
3851 {
3852         ipsec_policy_t *ip, *nip;
3853         int af, chain, nchain;
3854 
3855         for (af = 0; af < IPSEC_NAF; af++) {
3856                 for (ip = ipr->ipr_nonhash[af]; ip != NULL; ip = nip) {
3857                         nip = ip->ipsp_hash.hash_next;
3858                         IPPOL_UNCHAIN(php, ip);
3859                 }
3860                 ipr->ipr_nonhash[af] = NULL;
3861         }
3862         nchain = ipr->ipr_nchains;
3863 
3864         for (chain = 0; chain < nchain; chain++) {
3865                 for (ip = ipr->ipr_hash[chain].hash_head; ip != NULL;
3866                     ip = nip) {
3867                         nip = ip->ipsp_hash.hash_next;
3868                         IPPOL_UNCHAIN(php, ip);
3869                 }
3870                 ipr->ipr_hash[chain].hash_head = NULL;
3871         }
3872 }
3873 
3874 /*
3875  * Create and insert inbound or outbound policy associated with actp for the
3876  * address family fam into the policy head ph.  Returns B_TRUE if policy was
3877  * inserted, and B_FALSE otherwise.
3878  */
3879 boolean_t
3880 ipsec_polhead_insert(ipsec_policy_head_t *ph, ipsec_act_t *actp, uint_t nact,
3881     int fam, int ptype, netstack_t *ns)
3882 {
3883         ipsec_selkey_t          sel;
3884         ipsec_policy_t          *pol;
3885         ipsec_policy_root_t     *pr;
3886 
3887         bzero(&sel, sizeof (sel));
3888         sel.ipsl_valid = (fam == IPSEC_AF_V4 ? IPSL_IPV4 : IPSL_IPV6);
3889         if ((pol = ipsec_policy_create(&sel, actp, nact, IPSEC_PRIO_SOCKET,
3890             NULL, ns)) != NULL) {
3891                 pr = &ph->iph_root[ptype];
3892                 HASHLIST_INSERT(pol, ipsp_hash, pr->ipr_nonhash[fam]);
3893                 ipsec_insert_always(&ph->iph_rulebyid, pol);
3894         }
3895         return (pol != NULL);
3896 }
3897 
3898 void
3899 ipsec_polhead_flush(ipsec_policy_head_t *php, netstack_t *ns)
3900 {
3901         int dir;
3902 
3903         ASSERT(RW_WRITE_HELD(&php->iph_lock));
3904 
3905         for (dir = 0; dir < IPSEC_NTYPES; dir++)
3906                 ipsec_ipr_flush(php, &php->iph_root[dir]);
3907 
3908         php->iph_gen++;
3909         ipsec_update_present_flags(ns->netstack_ipsec);
3910 }
3911 
3912 void
3913 ipsec_polhead_free(ipsec_policy_head_t *php, netstack_t *ns)
3914 {
3915         int dir;
3916 
3917         ASSERT(php->iph_refs == 0);
3918 
3919         rw_enter(&php->iph_lock, RW_WRITER);
3920         ipsec_polhead_flush(php, ns);
3921         rw_exit(&php->iph_lock);
3922         rw_destroy(&php->iph_lock);
3923         for (dir = 0; dir < IPSEC_NTYPES; dir++) {
3924                 ipsec_policy_root_t *ipr = &php->iph_root[dir];
3925                 int chain;
3926 
3927                 for (chain = 0; chain < ipr->ipr_nchains; chain++)
3928                         mutex_destroy(&(ipr->ipr_hash[chain].hash_lock));
3929 
3930         }
3931         ipsec_polhead_free_table(php);
3932         kmem_free(php, sizeof (*php));
3933 }
3934 
3935 static void
3936 ipsec_ipr_init(ipsec_policy_root_t *ipr)
3937 {
3938         int af;
3939 
3940         ipr->ipr_nchains = 0;
3941         ipr->ipr_hash = NULL;
3942 
3943         for (af = 0; af < IPSEC_NAF; af++) {
3944                 ipr->ipr_nonhash[af] = NULL;
3945         }
3946 }
3947 
3948 ipsec_policy_head_t *
3949 ipsec_polhead_create(void)
3950 {
3951         ipsec_policy_head_t *php;
3952 
3953         php = kmem_alloc(sizeof (*php), KM_NOSLEEP);
3954         if (php == NULL)
3955                 return (php);
3956 
3957         rw_init(&php->iph_lock, NULL, RW_DEFAULT, NULL);
3958         php->iph_refs = 1;
3959         php->iph_gen = 0;
3960 
3961         ipsec_ipr_init(&php->iph_root[IPSEC_TYPE_INBOUND]);
3962         ipsec_ipr_init(&php->iph_root[IPSEC_TYPE_OUTBOUND]);
3963 
3964         avl_create(&php->iph_rulebyid, ipsec_policy_cmpbyid,
3965             sizeof (ipsec_policy_t), offsetof(ipsec_policy_t, ipsp_byid));
3966 
3967         return (php);
3968 }
3969 
3970 /*
3971  * Clone the policy head into a new polhead; release one reference to the
3972  * old one and return the only reference to the new one.
3973  * If the old one had a refcount of 1, just return it.
3974  */
3975 ipsec_policy_head_t *
3976 ipsec_polhead_split(ipsec_policy_head_t *php, netstack_t *ns)
3977 {
3978         ipsec_policy_head_t *nphp;
3979 
3980         if (php == NULL)
3981                 return (ipsec_polhead_create());
3982         else if (php->iph_refs == 1)
3983                 return (php);
3984 
3985         nphp = ipsec_polhead_create();
3986         if (nphp == NULL)
3987                 return (NULL);
3988 
3989         if (ipsec_copy_polhead(php, nphp, ns) != 0) {
3990                 ipsec_polhead_free(nphp, ns);
3991                 return (NULL);
3992         }
3993         IPPH_REFRELE(php, ns);
3994         return (nphp);
3995 }
3996 
3997 /*
3998  * When sending a response to a ICMP request or generating a RST
3999  * in the TCP case, the outbound packets need to go at the same level
4000  * of protection as the incoming ones i.e we associate our outbound
4001  * policy with how the packet came in. We call this after we have
4002  * accepted the incoming packet which may or may not have been in
4003  * clear and hence we are sending the reply back with the policy
4004  * matching the incoming datagram's policy.
4005  *
4006  * NOTE : This technology serves two purposes :
4007  *
4008  * 1) If we have multiple outbound policies, we send out a reply
4009  *    matching with how it came in rather than matching the outbound
4010  *    policy.
4011  *
4012  * 2) For assymetric policies, we want to make sure that incoming
4013  *    and outgoing has the same level of protection. Assymetric
4014  *    policies exist only with global policy where we may not have
4015  *    both outbound and inbound at the same time.
4016  *
4017  * NOTE2:       This function is called by cleartext cases, so it needs to be
4018  *              in IP proper.
4019  *
4020  * Note: the caller has moved other parts of ira into ixa already.
4021  */
4022 boolean_t
4023 ipsec_in_to_out(ip_recv_attr_t *ira, ip_xmit_attr_t *ixa, mblk_t *data_mp,
4024     ipha_t *ipha, ip6_t *ip6h)
4025 {
4026         ipsec_selector_t sel;
4027         ipsec_action_t  *reflect_action = NULL;
4028         netstack_t      *ns = ixa->ixa_ipst->ips_netstack;
4029 
4030         bzero((void*)&sel, sizeof (sel));
4031 
4032         if (ira->ira_ipsec_action != NULL) {
4033                 /* transfer reference.. */
4034                 reflect_action = ira->ira_ipsec_action;
4035                 ira->ira_ipsec_action = NULL;
4036         } else if (!(ira->ira_flags & IRAF_LOOPBACK))
4037                 reflect_action = ipsec_in_to_out_action(ira);
4038 
4039         /*
4040          * The caller is going to send the datagram out which might
4041          * go on the wire or delivered locally through ire_send_local.
4042          *
4043          * 1) If it goes out on the wire, new associations will be
4044          *    obtained.
4045          * 2) If it is delivered locally, ire_send_local will convert
4046          *    this ip_xmit_attr_t back to a ip_recv_attr_t looking at the
4047          *    requests.
4048          */
4049         ixa->ixa_ipsec_action = reflect_action;
4050 
4051         if (!ipsec_init_outbound_ports(&sel, data_mp, ipha, ip6h, 0,
4052             ns->netstack_ipsec)) {
4053                 /* Note: data_mp already consumed and ip_drop_packet done */
4054                 return (B_FALSE);
4055         }
4056         ixa->ixa_ipsec_src_port = sel.ips_local_port;
4057         ixa->ixa_ipsec_dst_port = sel.ips_remote_port;
4058         ixa->ixa_ipsec_proto = sel.ips_protocol;
4059         ixa->ixa_ipsec_icmp_type = sel.ips_icmp_type;
4060         ixa->ixa_ipsec_icmp_code = sel.ips_icmp_code;
4061 
4062         /*
4063          * Don't use global policy for this, as we want
4064          * to use the same protection that was applied to the inbound packet.
4065          * Thus we set IXAF_NO_IPSEC is it arrived in the clear to make
4066          * it be sent in the clear.
4067          */
4068         if (ira->ira_flags & IRAF_IPSEC_SECURE)
4069                 ixa->ixa_flags |= IXAF_IPSEC_SECURE;
4070         else
4071                 ixa->ixa_flags |= IXAF_NO_IPSEC;
4072 
4073         return (B_TRUE);
4074 }
4075 
4076 void
4077 ipsec_out_release_refs(ip_xmit_attr_t *ixa)
4078 {
4079         if (!(ixa->ixa_flags & IXAF_IPSEC_SECURE))
4080                 return;
4081 
4082         if (ixa->ixa_ipsec_ah_sa != NULL) {
4083                 IPSA_REFRELE(ixa->ixa_ipsec_ah_sa);
4084                 ixa->ixa_ipsec_ah_sa = NULL;
4085         }
4086         if (ixa->ixa_ipsec_esp_sa != NULL) {
4087                 IPSA_REFRELE(ixa->ixa_ipsec_esp_sa);
4088                 ixa->ixa_ipsec_esp_sa = NULL;
4089         }
4090         if (ixa->ixa_ipsec_policy != NULL) {
4091                 IPPOL_REFRELE(ixa->ixa_ipsec_policy);
4092                 ixa->ixa_ipsec_policy = NULL;
4093         }
4094         if (ixa->ixa_ipsec_action != NULL) {
4095                 IPACT_REFRELE(ixa->ixa_ipsec_action);
4096                 ixa->ixa_ipsec_action = NULL;
4097         }
4098         if (ixa->ixa_ipsec_latch) {
4099                 IPLATCH_REFRELE(ixa->ixa_ipsec_latch);
4100                 ixa->ixa_ipsec_latch = NULL;
4101         }
4102         /* Clear the soft references to the SAs */
4103         ixa->ixa_ipsec_ref[0].ipsr_sa = NULL;
4104         ixa->ixa_ipsec_ref[0].ipsr_bucket = NULL;
4105         ixa->ixa_ipsec_ref[0].ipsr_gen = 0;
4106         ixa->ixa_ipsec_ref[1].ipsr_sa = NULL;
4107         ixa->ixa_ipsec_ref[1].ipsr_bucket = NULL;
4108         ixa->ixa_ipsec_ref[1].ipsr_gen = 0;
4109         ixa->ixa_flags &= ~IXAF_IPSEC_SECURE;
4110 }
4111 
4112 void
4113 ipsec_in_release_refs(ip_recv_attr_t *ira)
4114 {
4115         if (!(ira->ira_flags & IRAF_IPSEC_SECURE))
4116                 return;
4117 
4118         if (ira->ira_ipsec_ah_sa != NULL) {
4119                 IPSA_REFRELE(ira->ira_ipsec_ah_sa);
4120                 ira->ira_ipsec_ah_sa = NULL;
4121         }
4122         if (ira->ira_ipsec_esp_sa != NULL) {
4123                 IPSA_REFRELE(ira->ira_ipsec_esp_sa);
4124                 ira->ira_ipsec_esp_sa = NULL;
4125         }
4126         ira->ira_flags &= ~IRAF_IPSEC_SECURE;
4127 }
4128 
4129 /*
4130  * This is called from ire_send_local when a packet
4131  * is looped back. We setup the ip_recv_attr_t "borrowing" the references
4132  * held by the callers.
4133  * Note that we don't do any IPsec but we carry the actions and IPSEC flags
4134  * across so that the fanout policy checks see that IPsec was applied.
4135  *
4136  * The caller should do ipsec_in_release_refs() on the ira by calling
4137  * ira_cleanup().
4138  */
4139 void
4140 ipsec_out_to_in(ip_xmit_attr_t *ixa, ill_t *ill, ip_recv_attr_t *ira)
4141 {
4142         ipsec_policy_t *pol;
4143         ipsec_action_t *act;
4144 
4145         /* Non-IPsec operations */
4146         ira->ira_free_flags = 0;
4147         ira->ira_zoneid = ixa->ixa_zoneid;
4148         ira->ira_cred = ixa->ixa_cred;
4149         ira->ira_cpid = ixa->ixa_cpid;
4150         ira->ira_tsl = ixa->ixa_tsl;
4151         ira->ira_ill = ira->ira_rill = ill;
4152         ira->ira_flags = ixa->ixa_flags & IAF_MASK;
4153         ira->ira_no_loop_zoneid = ixa->ixa_no_loop_zoneid;
4154         ira->ira_pktlen = ixa->ixa_pktlen;
4155         ira->ira_ip_hdr_length = ixa->ixa_ip_hdr_length;
4156         ira->ira_protocol = ixa->ixa_protocol;
4157         ira->ira_mhip = NULL;
4158 
4159         ira->ira_flags |= IRAF_LOOPBACK | IRAF_L2SRC_LOOPBACK;
4160 
4161         ira->ira_sqp = ixa->ixa_sqp;
4162         ira->ira_ring = NULL;
4163 
4164         ira->ira_ruifindex = ill->ill_phyint->phyint_ifindex;
4165         ira->ira_rifindex = ira->ira_ruifindex;
4166 
4167         if (!(ixa->ixa_flags & IXAF_IPSEC_SECURE))
4168                 return;
4169 
4170         ira->ira_flags |= IRAF_IPSEC_SECURE;
4171 
4172         ira->ira_ipsec_ah_sa = NULL;
4173         ira->ira_ipsec_esp_sa = NULL;
4174 
4175         act = ixa->ixa_ipsec_action;
4176         if (act == NULL) {
4177                 pol = ixa->ixa_ipsec_policy;
4178                 if (pol != NULL) {
4179                         act = pol->ipsp_act;
4180                         IPACT_REFHOLD(act);
4181                 }
4182         }
4183         ixa->ixa_ipsec_action = NULL;
4184         ira->ira_ipsec_action = act;
4185 }
4186 
4187 /*
4188  * Consults global policy and per-socket policy to see whether this datagram
4189  * should go out secure. If so it updates the ip_xmit_attr_t
4190  * Should not be used when connecting, since then we want to latch the policy.
4191  *
4192  * If connp is NULL we just look at the global policy.
4193  *
4194  * Returns NULL if the packet was dropped, in which case the MIB has
4195  * been incremented and ip_drop_packet done.
4196  */
4197 mblk_t *
4198 ip_output_attach_policy(mblk_t *mp, ipha_t *ipha, ip6_t *ip6h,
4199     const conn_t *connp, ip_xmit_attr_t *ixa)
4200 {
4201         ipsec_selector_t sel;
4202         boolean_t       policy_present;
4203         ip_stack_t      *ipst = ixa->ixa_ipst;
4204         netstack_t      *ns = ipst->ips_netstack;
4205         ipsec_stack_t   *ipss = ns->netstack_ipsec;
4206         ipsec_policy_t  *p;
4207 
4208         ixa->ixa_ipsec_policy_gen = ipss->ipsec_system_policy.iph_gen;
4209         ASSERT((ipha != NULL && ip6h == NULL) ||
4210             (ip6h != NULL && ipha == NULL));
4211 
4212         if (ipha != NULL)
4213                 policy_present = ipss->ipsec_outbound_v4_policy_present;
4214         else
4215                 policy_present = ipss->ipsec_outbound_v6_policy_present;
4216 
4217         if (!policy_present && (connp == NULL || connp->conn_policy == NULL))
4218                 return (mp);
4219 
4220         bzero((void*)&sel, sizeof (sel));
4221 
4222         if (ipha != NULL) {
4223                 sel.ips_local_addr_v4 = ipha->ipha_src;
4224                 sel.ips_remote_addr_v4 = ip_get_dst(ipha);
4225                 sel.ips_isv4 = B_TRUE;
4226         } else {
4227                 sel.ips_isv4 = B_FALSE;
4228                 sel.ips_local_addr_v6 = ip6h->ip6_src;
4229                 sel.ips_remote_addr_v6 = ip_get_dst_v6(ip6h, mp, NULL);
4230         }
4231         sel.ips_protocol = ixa->ixa_protocol;
4232 
4233         if (!ipsec_init_outbound_ports(&sel, mp, ipha, ip6h, 0, ipss)) {
4234                 if (ipha != NULL) {
4235                         BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsOutDiscards);
4236                 } else {
4237                         BUMP_MIB(&ipst->ips_ip6_mib, ipIfStatsOutDiscards);
4238                 }
4239                 /* Note: mp already consumed and ip_drop_packet done */
4240                 return (NULL);
4241         }
4242 
4243         ASSERT(ixa->ixa_ipsec_policy == NULL);
4244         p = ipsec_find_policy(IPSEC_TYPE_OUTBOUND, connp, &sel, ns);
4245         ixa->ixa_ipsec_policy = p;
4246         if (p != NULL) {
4247                 ixa->ixa_flags |= IXAF_IPSEC_SECURE;
4248                 if (connp == NULL || connp->conn_policy == NULL)
4249                         ixa->ixa_flags |= IXAF_IPSEC_GLOBAL_POLICY;
4250         } else {
4251                 ixa->ixa_flags &= ~IXAF_IPSEC_SECURE;
4252         }
4253 
4254         /*
4255          * Copy the right port information.
4256          */
4257         ixa->ixa_ipsec_src_port = sel.ips_local_port;
4258         ixa->ixa_ipsec_dst_port = sel.ips_remote_port;
4259         ixa->ixa_ipsec_icmp_type = sel.ips_icmp_type;
4260         ixa->ixa_ipsec_icmp_code = sel.ips_icmp_code;
4261         ixa->ixa_ipsec_proto = sel.ips_protocol;
4262         return (mp);
4263 }
4264 
4265 /*
4266  * When appropriate, this function caches inbound and outbound policy
4267  * for this connection. The outbound policy is stored in conn_ixa.
4268  * Note that it can not be used for SCTP since conn_faddr isn't set for SCTP.
4269  *
4270  * XXX need to work out more details about per-interface policy and
4271  * caching here!
4272  *
4273  * XXX may want to split inbound and outbound caching for ill..
4274  */
4275 int
4276 ipsec_conn_cache_policy(conn_t *connp, boolean_t isv4)
4277 {
4278         boolean_t global_policy_present;
4279         netstack_t      *ns = connp->conn_netstack;
4280         ipsec_stack_t   *ipss = ns->netstack_ipsec;
4281 
4282         connp->conn_ixa->ixa_ipsec_policy_gen =
4283             ipss->ipsec_system_policy.iph_gen;
4284         /*
4285          * There is no policy latching for ICMP sockets because we can't
4286          * decide on which policy to use until we see the packet and get
4287          * type/code selectors.
4288          */
4289         if (connp->conn_proto == IPPROTO_ICMP ||
4290             connp->conn_proto == IPPROTO_ICMPV6) {
4291                 connp->conn_in_enforce_policy =
4292                     connp->conn_out_enforce_policy = B_TRUE;
4293                 if (connp->conn_latch != NULL) {
4294                         IPLATCH_REFRELE(connp->conn_latch);
4295                         connp->conn_latch = NULL;
4296                 }
4297                 if (connp->conn_latch_in_policy != NULL) {
4298                         IPPOL_REFRELE(connp->conn_latch_in_policy);
4299                         connp->conn_latch_in_policy = NULL;
4300                 }
4301                 if (connp->conn_latch_in_action != NULL) {
4302                         IPACT_REFRELE(connp->conn_latch_in_action);
4303                         connp->conn_latch_in_action = NULL;
4304                 }
4305                 if (connp->conn_ixa->ixa_ipsec_policy != NULL) {
4306                         IPPOL_REFRELE(connp->conn_ixa->ixa_ipsec_policy);
4307                         connp->conn_ixa->ixa_ipsec_policy = NULL;
4308                 }
4309                 if (connp->conn_ixa->ixa_ipsec_action != NULL) {
4310                         IPACT_REFRELE(connp->conn_ixa->ixa_ipsec_action);
4311                         connp->conn_ixa->ixa_ipsec_action = NULL;
4312                 }
4313                 connp->conn_ixa->ixa_flags &= ~IXAF_IPSEC_SECURE;
4314                 return (0);
4315         }
4316 
4317         global_policy_present = isv4 ?
4318             (ipss->ipsec_outbound_v4_policy_present ||
4319             ipss->ipsec_inbound_v4_policy_present) :
4320             (ipss->ipsec_outbound_v6_policy_present ||
4321             ipss->ipsec_inbound_v6_policy_present);
4322 
4323         if ((connp->conn_policy != NULL) || global_policy_present) {
4324                 ipsec_selector_t sel;
4325                 ipsec_policy_t  *p;
4326 
4327                 if (connp->conn_latch == NULL &&
4328                     (connp->conn_latch = iplatch_create()) == NULL) {
4329                         return (ENOMEM);
4330                 }
4331 
4332                 bzero((void*)&sel, sizeof (sel));
4333 
4334                 sel.ips_protocol = connp->conn_proto;
4335                 sel.ips_local_port = connp->conn_lport;
4336                 sel.ips_remote_port = connp->conn_fport;
4337                 sel.ips_is_icmp_inv_acq = 0;
4338                 sel.ips_isv4 = isv4;
4339                 if (isv4) {
4340                         sel.ips_local_addr_v4 = connp->conn_laddr_v4;
4341                         sel.ips_remote_addr_v4 = connp->conn_faddr_v4;
4342                 } else {
4343                         sel.ips_local_addr_v6 = connp->conn_laddr_v6;
4344                         sel.ips_remote_addr_v6 = connp->conn_faddr_v6;
4345                 }
4346 
4347                 p = ipsec_find_policy(IPSEC_TYPE_INBOUND, connp, &sel, ns);
4348                 if (connp->conn_latch_in_policy != NULL)
4349                         IPPOL_REFRELE(connp->conn_latch_in_policy);
4350                 connp->conn_latch_in_policy = p;
4351                 connp->conn_in_enforce_policy = (p != NULL);
4352 
4353                 p = ipsec_find_policy(IPSEC_TYPE_OUTBOUND, connp, &sel, ns);
4354                 if (connp->conn_ixa->ixa_ipsec_policy != NULL)
4355                         IPPOL_REFRELE(connp->conn_ixa->ixa_ipsec_policy);
4356                 connp->conn_ixa->ixa_ipsec_policy = p;
4357                 connp->conn_out_enforce_policy = (p != NULL);
4358                 if (p != NULL) {
4359                         connp->conn_ixa->ixa_flags |= IXAF_IPSEC_SECURE;
4360                         if (connp->conn_policy == NULL) {
4361                                 connp->conn_ixa->ixa_flags |=
4362                                     IXAF_IPSEC_GLOBAL_POLICY;
4363                         }
4364                 } else {
4365                         connp->conn_ixa->ixa_flags &= ~IXAF_IPSEC_SECURE;
4366                 }
4367                 /* Clear the latched actions too, in case we're recaching. */
4368                 if (connp->conn_ixa->ixa_ipsec_action != NULL) {
4369                         IPACT_REFRELE(connp->conn_ixa->ixa_ipsec_action);
4370                         connp->conn_ixa->ixa_ipsec_action = NULL;
4371                 }
4372                 if (connp->conn_latch_in_action != NULL) {
4373                         IPACT_REFRELE(connp->conn_latch_in_action);
4374                         connp->conn_latch_in_action = NULL;
4375                 }
4376                 connp->conn_ixa->ixa_ipsec_src_port = sel.ips_local_port;
4377                 connp->conn_ixa->ixa_ipsec_dst_port = sel.ips_remote_port;
4378                 connp->conn_ixa->ixa_ipsec_icmp_type = sel.ips_icmp_type;
4379                 connp->conn_ixa->ixa_ipsec_icmp_code = sel.ips_icmp_code;
4380                 connp->conn_ixa->ixa_ipsec_proto = sel.ips_protocol;
4381         } else {
4382                 connp->conn_ixa->ixa_flags &= ~IXAF_IPSEC_SECURE;
4383         }
4384 
4385         /*
4386          * We may or may not have policy for this endpoint.  We still set
4387          * conn_policy_cached so that inbound datagrams don't have to look
4388          * at global policy as policy is considered latched for these
4389          * endpoints.  We should not set conn_policy_cached until the conn
4390          * reflects the actual policy. If we *set* this before inheriting
4391          * the policy there is a window where the check
4392          * CONN_INBOUND_POLICY_PRESENT, will neither check with the policy
4393          * on the conn (because we have not yet copied the policy on to
4394          * conn and hence not set conn_in_enforce_policy) nor with the
4395          * global policy (because conn_policy_cached is already set).
4396          */
4397         connp->conn_policy_cached = B_TRUE;
4398         return (0);
4399 }
4400 
4401 /*
4402  * When appropriate, this function caches outbound policy for faddr/fport.
4403  * It is used when we are not connected i.e., when we can not latch the
4404  * policy.
4405  */
4406 void
4407 ipsec_cache_outbound_policy(const conn_t *connp, const in6_addr_t *v6src,
4408     const in6_addr_t *v6dst, in_port_t dstport, ip_xmit_attr_t *ixa)
4409 {
4410         boolean_t       isv4 = (ixa->ixa_flags & IXAF_IS_IPV4) != 0;
4411         boolean_t       global_policy_present;
4412         netstack_t      *ns = connp->conn_netstack;
4413         ipsec_stack_t   *ipss = ns->netstack_ipsec;
4414 
4415         ixa->ixa_ipsec_policy_gen = ipss->ipsec_system_policy.iph_gen;
4416 
4417         /*
4418          * There is no policy caching for ICMP sockets because we can't
4419          * decide on which policy to use until we see the packet and get
4420          * type/code selectors.
4421          */
4422         if (connp->conn_proto == IPPROTO_ICMP ||
4423             connp->conn_proto == IPPROTO_ICMPV6) {
4424                 ixa->ixa_flags &= ~IXAF_IPSEC_SECURE;
4425                 if (ixa->ixa_ipsec_policy != NULL) {
4426                         IPPOL_REFRELE(ixa->ixa_ipsec_policy);
4427                         ixa->ixa_ipsec_policy = NULL;
4428                 }
4429                 if (ixa->ixa_ipsec_action != NULL) {
4430                         IPACT_REFRELE(ixa->ixa_ipsec_action);
4431                         ixa->ixa_ipsec_action = NULL;
4432                 }
4433                 return;
4434         }
4435 
4436         global_policy_present = isv4 ?
4437             (ipss->ipsec_outbound_v4_policy_present ||
4438             ipss->ipsec_inbound_v4_policy_present) :
4439             (ipss->ipsec_outbound_v6_policy_present ||
4440             ipss->ipsec_inbound_v6_policy_present);
4441 
4442         if ((connp->conn_policy != NULL) || global_policy_present) {
4443                 ipsec_selector_t sel;
4444                 ipsec_policy_t  *p;
4445 
4446                 bzero((void*)&sel, sizeof (sel));
4447 
4448                 sel.ips_protocol = connp->conn_proto;
4449                 sel.ips_local_port = connp->conn_lport;
4450                 sel.ips_remote_port = dstport;
4451                 sel.ips_is_icmp_inv_acq = 0;
4452                 sel.ips_isv4 = isv4;
4453                 if (isv4) {
4454                         IN6_V4MAPPED_TO_IPADDR(v6src, sel.ips_local_addr_v4);
4455                         IN6_V4MAPPED_TO_IPADDR(v6dst, sel.ips_remote_addr_v4);
4456                 } else {
4457                         sel.ips_local_addr_v6 = *v6src;
4458                         sel.ips_remote_addr_v6 = *v6dst;
4459                 }
4460 
4461                 p = ipsec_find_policy(IPSEC_TYPE_OUTBOUND, connp, &sel, ns);
4462                 if (ixa->ixa_ipsec_policy != NULL)
4463                         IPPOL_REFRELE(ixa->ixa_ipsec_policy);
4464                 ixa->ixa_ipsec_policy = p;
4465                 if (p != NULL) {
4466                         ixa->ixa_flags |= IXAF_IPSEC_SECURE;
4467                         if (connp->conn_policy == NULL)
4468                                 ixa->ixa_flags |= IXAF_IPSEC_GLOBAL_POLICY;
4469                 } else {
4470                         ixa->ixa_flags &= ~IXAF_IPSEC_SECURE;
4471                 }
4472                 /* Clear the latched actions too, in case we're recaching. */
4473                 if (ixa->ixa_ipsec_action != NULL) {
4474                         IPACT_REFRELE(ixa->ixa_ipsec_action);
4475                         ixa->ixa_ipsec_action = NULL;
4476                 }
4477 
4478                 ixa->ixa_ipsec_src_port = sel.ips_local_port;
4479                 ixa->ixa_ipsec_dst_port = sel.ips_remote_port;
4480                 ixa->ixa_ipsec_icmp_type = sel.ips_icmp_type;
4481                 ixa->ixa_ipsec_icmp_code = sel.ips_icmp_code;
4482                 ixa->ixa_ipsec_proto = sel.ips_protocol;
4483         } else {
4484                 ixa->ixa_flags &= ~IXAF_IPSEC_SECURE;
4485                 if (ixa->ixa_ipsec_policy != NULL) {
4486                         IPPOL_REFRELE(ixa->ixa_ipsec_policy);
4487                         ixa->ixa_ipsec_policy = NULL;
4488                 }
4489                 if (ixa->ixa_ipsec_action != NULL) {
4490                         IPACT_REFRELE(ixa->ixa_ipsec_action);
4491                         ixa->ixa_ipsec_action = NULL;
4492                 }
4493         }
4494 }
4495 
4496 /*
4497  * Returns B_FALSE if the policy has gone stale.
4498  */
4499 boolean_t
4500 ipsec_outbound_policy_current(ip_xmit_attr_t *ixa)
4501 {
4502         ipsec_stack_t   *ipss = ixa->ixa_ipst->ips_netstack->netstack_ipsec;
4503 
4504         if (!(ixa->ixa_flags & IXAF_IPSEC_GLOBAL_POLICY))
4505                 return (B_TRUE);
4506 
4507         return (ixa->ixa_ipsec_policy_gen == ipss->ipsec_system_policy.iph_gen);
4508 }
4509 
4510 void
4511 iplatch_free(ipsec_latch_t *ipl)
4512 {
4513         if (ipl->ipl_local_cid != NULL)
4514                 IPSID_REFRELE(ipl->ipl_local_cid);
4515         if (ipl->ipl_remote_cid != NULL)
4516                 IPSID_REFRELE(ipl->ipl_remote_cid);
4517         mutex_destroy(&ipl->ipl_lock);
4518         kmem_free(ipl, sizeof (*ipl));
4519 }
4520 
4521 ipsec_latch_t *
4522 iplatch_create()
4523 {
4524         ipsec_latch_t *ipl = kmem_alloc(sizeof (*ipl), KM_NOSLEEP);
4525         if (ipl == NULL)
4526                 return (ipl);
4527         bzero(ipl, sizeof (*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 specified
4692  * algorithm, which is either a member of a netstack alg array or about to be,
4693  * and therefore must be called holding ipsec_alg_lock for write.
4694  */
4695 void
4696 ipsec_alg_fix_min_max(ipsec_alginfo_t *alg, ipsec_algtype_t alg_type,
4697     netstack_t *ns)
4698 {
4699         size_t crypto_min = (size_t)-1, crypto_max = 0;
4700         size_t cur_crypto_min, cur_crypto_max;
4701         boolean_t is_valid;
4702         crypto_mechanism_info_t *mech_infos;
4703         uint_t nmech_infos;
4704         int crypto_rc, i;
4705         crypto_mech_usage_t mask;
4706         ipsec_stack_t   *ipss = ns->netstack_ipsec;
4707 
4708         ASSERT(RW_WRITE_HELD(&ipss->ipsec_alg_lock));
4709 
4710         /*
4711          * Compute the min, max, and default key sizes (in number of
4712          * increments to the default key size in bits) as defined
4713          * by the algorithm mappings. This range of key sizes is used
4714          * for policy related operations. The effective key sizes
4715          * supported by the framework could be more limited than
4716          * those defined for an algorithm.
4717          */
4718         alg->alg_default_bits = alg->alg_key_sizes[0];
4719         alg->alg_default = 0;
4720         if (alg->alg_increment != 0) {
4721                 /* key sizes are defined by range & increment */
4722                 alg->alg_minbits = alg->alg_key_sizes[1];
4723                 alg->alg_maxbits = alg->alg_key_sizes[2];
4724         } else if (alg->alg_nkey_sizes == 0) {
4725                 /* no specified key size for algorithm */
4726                 alg->alg_minbits = alg->alg_maxbits = 0;
4727         } else {
4728                 /* key sizes are defined by enumeration */
4729                 alg->alg_minbits = (uint16_t)-1;
4730                 alg->alg_maxbits = 0;
4731 
4732                 for (i = 0; i < alg->alg_nkey_sizes; i++) {
4733                         if (alg->alg_key_sizes[i] < alg->alg_minbits)
4734                                 alg->alg_minbits = alg->alg_key_sizes[i];
4735                         if (alg->alg_key_sizes[i] > alg->alg_maxbits)
4736                                 alg->alg_maxbits = alg->alg_key_sizes[i];
4737                 }
4738         }
4739 
4740         if (!(alg->alg_flags & ALG_FLAG_VALID))
4741                 return;
4742 
4743         /*
4744          * Mechanisms do not apply to the NULL encryption
4745          * algorithm, so simply return for this case.
4746          */
4747         if (alg->alg_id == SADB_EALG_NULL)
4748                 return;
4749 
4750         /*
4751          * Find the min and max key sizes supported by the cryptographic
4752          * framework providers.
4753          */
4754 
4755         /* get the key sizes supported by the framework */
4756         crypto_rc = crypto_get_all_mech_info(alg->alg_mech_type,
4757             &mech_infos, &nmech_infos, KM_SLEEP);
4758         if (crypto_rc != CRYPTO_SUCCESS || nmech_infos == 0) {
4759                 alg->alg_flags &= ~ALG_FLAG_VALID;
4760                 return;
4761         }
4762 
4763         /* min and max key sizes supported by framework */
4764         for (i = 0, is_valid = B_FALSE; i < nmech_infos; i++) {
4765                 int unit_bits;
4766 
4767                 /*
4768                  * Ignore entries that do not support the operations
4769                  * needed for the algorithm type.
4770                  */
4771                 if (alg_type == IPSEC_ALG_AUTH) {
4772                         mask = CRYPTO_MECH_USAGE_MAC;
4773                 } else {
4774                         mask = CRYPTO_MECH_USAGE_ENCRYPT |
4775                             CRYPTO_MECH_USAGE_DECRYPT;
4776                 }
4777                 if ((mech_infos[i].mi_usage & mask) != mask)
4778                         continue;
4779 
4780                 unit_bits = (mech_infos[i].mi_keysize_unit ==
4781                     CRYPTO_KEYSIZE_UNIT_IN_BYTES)  ? 8 : 1;
4782                 /* adjust min/max supported by framework */
4783                 cur_crypto_min = mech_infos[i].mi_min_key_size * unit_bits;
4784                 cur_crypto_max = mech_infos[i].mi_max_key_size * unit_bits;
4785 
4786                 if (cur_crypto_min < crypto_min)
4787                         crypto_min = cur_crypto_min;
4788 
4789                 /*
4790                  * CRYPTO_EFFECTIVELY_INFINITE is a special value of
4791                  * the crypto framework which means "no upper limit".
4792                  */
4793                 if (mech_infos[i].mi_max_key_size ==
4794                     CRYPTO_EFFECTIVELY_INFINITE) {
4795                         crypto_max = (size_t)-1;
4796                 } else if (cur_crypto_max > crypto_max) {
4797                         crypto_max = cur_crypto_max;
4798                 }
4799 
4800                 is_valid = B_TRUE;
4801         }
4802 
4803         kmem_free(mech_infos, sizeof (crypto_mechanism_info_t) *
4804             nmech_infos);
4805 
4806         if (!is_valid) {
4807                 /* no key sizes supported by framework */
4808                 alg->alg_flags &= ~ALG_FLAG_VALID;
4809                 return;
4810         }
4811 
4812         /*
4813          * Determine min and max key sizes from alg_key_sizes[].
4814          * defined for the algorithm entry. Adjust key sizes based on
4815          * those supported by the framework.
4816          */
4817         alg->alg_ef_default_bits = alg->alg_key_sizes[0];
4818 
4819         /*
4820          * For backwards compatability, assume that the IV length
4821          * is the same as the data length.
4822          */
4823         alg->alg_ivlen = alg->alg_datalen;
4824 
4825         /*
4826          * Copy any algorithm parameters (if provided) into dedicated
4827          * elements in the ipsec_alginfo_t structure.
4828          * There may be a better place to put this code.
4829          */
4830         for (i = 0; i < alg->alg_nparams; i++) {
4831                 switch (i) {
4832                 case 0:
4833                         /* Initialisation Vector length (bytes) */
4834                         alg->alg_ivlen =  alg->alg_params[0];
4835                         break;
4836                 case 1:
4837                         /* Integrity Check Vector length (bytes) */
4838                         alg->alg_icvlen = alg->alg_params[1];
4839                         break;
4840                 case 2:
4841                         /* Salt length (bytes) */
4842                         alg->alg_saltlen = (uint8_t)alg->alg_params[2];
4843                         break;
4844                 default:
4845                         break;
4846                 }
4847         }
4848 
4849         /* Default if the IV length is not specified. */
4850         if (alg_type == IPSEC_ALG_ENCR && alg->alg_ivlen == 0)
4851                 alg->alg_ivlen = alg->alg_datalen;
4852 
4853         alg_flag_check(alg);
4854 
4855         if (alg->alg_increment != 0) {
4856                 /* supported key sizes are defined by range  & increment */
4857                 crypto_min = ALGBITS_ROUND_UP(crypto_min, alg->alg_increment);
4858                 crypto_max = ALGBITS_ROUND_DOWN(crypto_max, alg->alg_increment);
4859 
4860                 alg->alg_ef_minbits = MAX(alg->alg_minbits,
4861                     (uint16_t)crypto_min);
4862                 alg->alg_ef_maxbits = MIN(alg->alg_maxbits,
4863                     (uint16_t)crypto_max);
4864 
4865                 /*
4866                  * If the sizes supported by the framework are outside
4867                  * the range of sizes defined by the algorithm mappings,
4868                  * the algorithm cannot be used. Check for this
4869                  * condition here.
4870                  */
4871                 if (alg->alg_ef_minbits > alg->alg_ef_maxbits) {
4872                         alg->alg_flags &= ~ALG_FLAG_VALID;
4873                         return;
4874                 }
4875                 if (alg->alg_ef_default_bits < alg->alg_ef_minbits)
4876                         alg->alg_ef_default_bits = alg->alg_ef_minbits;
4877                 if (alg->alg_ef_default_bits > alg->alg_ef_maxbits)
4878                         alg->alg_ef_default_bits = alg->alg_ef_maxbits;
4879         } else if (alg->alg_nkey_sizes == 0) {
4880                 /* no specified key size for algorithm */
4881                 alg->alg_ef_minbits = alg->alg_ef_maxbits = 0;
4882         } else {
4883                 /* supported key sizes are defined by enumeration */
4884                 alg->alg_ef_minbits = (uint16_t)-1;
4885                 alg->alg_ef_maxbits = 0;
4886 
4887                 for (i = 0, is_valid = B_FALSE; i < alg->alg_nkey_sizes; i++) {
4888                         /*
4889                          * Ignore the current key size if it is not in the
4890                          * range of sizes supported by the framework.
4891                          */
4892                         if (alg->alg_key_sizes[i] < crypto_min ||
4893                             alg->alg_key_sizes[i] > crypto_max)
4894                                 continue;
4895                         if (alg->alg_key_sizes[i] < alg->alg_ef_minbits)
4896                                 alg->alg_ef_minbits = alg->alg_key_sizes[i];
4897                         if (alg->alg_key_sizes[i] > alg->alg_ef_maxbits)
4898                                 alg->alg_ef_maxbits = alg->alg_key_sizes[i];
4899                         is_valid = B_TRUE;
4900                 }
4901 
4902                 if (!is_valid) {
4903                         alg->alg_flags &= ~ALG_FLAG_VALID;
4904                         return;
4905                 }
4906                 alg->alg_ef_default = 0;
4907         }
4908 }
4909 
4910 /*
4911  * Sanity check parameters provided by ipsecalgs(1m). Assume that
4912  * the algoritm is marked as valid, there is a check at the top
4913  * of this function. If any of the checks below fail, the algorithm
4914  * entry is invalid.
4915  */
4916 void
4917 alg_flag_check(ipsec_alginfo_t *alg)
4918 {
4919         alg->alg_flags &= ~ALG_FLAG_VALID;
4920 
4921         /*
4922          * Can't have the algorithm marked as CCM and GCM.
4923          * Check the ALG_FLAG_COMBINED and ALG_FLAG_COUNTERMODE
4924          * flags are set for CCM & GCM.
4925          */
4926         if ((alg->alg_flags & (ALG_FLAG_CCM|ALG_FLAG_GCM)) ==
4927             (ALG_FLAG_CCM|ALG_FLAG_GCM))
4928                 return;
4929         if (alg->alg_flags & (ALG_FLAG_CCM|ALG_FLAG_GCM)) {
4930                 if (!(alg->alg_flags & ALG_FLAG_COUNTERMODE))
4931                         return;
4932                 if (!(alg->alg_flags & ALG_FLAG_COMBINED))
4933                         return;
4934         }
4935 
4936         /*
4937          * For ALG_FLAG_COUNTERMODE, check the parameters
4938          * fit in the ipsec_nonce_t structure.
4939          */
4940         if (alg->alg_flags & ALG_FLAG_COUNTERMODE) {
4941                 if (alg->alg_ivlen != sizeof (((ipsec_nonce_t *)NULL)->iv))
4942                         return;
4943                 if (alg->alg_saltlen > sizeof (((ipsec_nonce_t *)NULL)->salt))
4944                         return;
4945         }
4946         if ((alg->alg_flags & ALG_FLAG_COMBINED) &&
4947             (alg->alg_icvlen == 0))
4948                 return;
4949 
4950         /* all is well. */
4951         alg->alg_flags |= ALG_FLAG_VALID;
4952 }
4953 
4954 /*
4955  * Free the memory used by the specified algorithm.
4956  */
4957 void
4958 ipsec_alg_free(ipsec_alginfo_t *alg)
4959 {
4960         if (alg == NULL)
4961                 return;
4962 
4963         if (alg->alg_key_sizes != NULL) {
4964                 kmem_free(alg->alg_key_sizes,
4965                     (alg->alg_nkey_sizes + 1) * sizeof (uint16_t));
4966                 alg->alg_key_sizes = NULL;
4967         }
4968         if (alg->alg_block_sizes != NULL) {
4969                 kmem_free(alg->alg_block_sizes,
4970                     (alg->alg_nblock_sizes + 1) * sizeof (uint16_t));
4971                 alg->alg_block_sizes = NULL;
4972         }
4973         if (alg->alg_params != NULL) {
4974                 kmem_free(alg->alg_params,
4975                     (alg->alg_nparams + 1) * sizeof (uint16_t));
4976                 alg->alg_params = NULL;
4977         }
4978         kmem_free(alg, sizeof (*alg));
4979 }
4980 
4981 /*
4982  * Check the validity of the specified key size for an algorithm.
4983  * Returns B_TRUE if key size is valid, B_FALSE otherwise.
4984  */
4985 boolean_t
4986 ipsec_valid_key_size(uint16_t key_size, ipsec_alginfo_t *alg)
4987 {
4988         if (key_size < alg->alg_ef_minbits || key_size > alg->alg_ef_maxbits)
4989                 return (B_FALSE);
4990 
4991         if (alg->alg_increment == 0 && alg->alg_nkey_sizes != 0) {
4992                 /*
4993                  * If the key sizes are defined by enumeration, the new
4994                  * key size must be equal to one of the supported values.
4995                  */
4996                 int i;
4997 
4998                 for (i = 0; i < alg->alg_nkey_sizes; i++)
4999                         if (key_size == alg->alg_key_sizes[i])
5000                                 break;
5001                 if (i == alg->alg_nkey_sizes)
5002                         return (B_FALSE);
5003         }
5004 
5005         return (B_TRUE);
5006 }
5007 
5008 /*
5009  * Callback function invoked by the crypto framework when a provider
5010  * registers or unregisters. This callback updates the algorithms
5011  * tables when a crypto algorithm is no longer available or becomes
5012  * available, and triggers the freeing/creation of context templates
5013  * associated with existing SAs, if needed.
5014  *
5015  * Need to walk all stack instances since the callback is global
5016  * for all instances
5017  */
5018 void
5019 ipsec_prov_update_callback(uint32_t event, void *event_arg)
5020 {
5021         netstack_handle_t nh;
5022         netstack_t *ns;
5023 
5024         netstack_next_init(&nh);
5025         while ((ns = netstack_next(&nh)) != NULL) {
5026                 ipsec_prov_update_callback_stack(event, event_arg, ns);
5027                 netstack_rele(ns);
5028         }
5029         netstack_next_fini(&nh);
5030 }
5031 
5032 static void
5033 ipsec_prov_update_callback_stack(uint32_t event, void *event_arg,
5034     netstack_t *ns)
5035 {
5036         crypto_notify_event_change_t *prov_change =
5037             (crypto_notify_event_change_t *)event_arg;
5038         uint_t algidx, algid, algtype, mech_count, mech_idx;
5039         ipsec_alginfo_t *alg;
5040         ipsec_alginfo_t oalg;
5041         crypto_mech_name_t *mechs;
5042         boolean_t alg_changed = B_FALSE;
5043         ipsec_stack_t   *ipss = ns->netstack_ipsec;
5044 
5045         /* ignore events for which we didn't register */
5046         if (event != CRYPTO_EVENT_MECHS_CHANGED) {
5047                 ip1dbg(("ipsec_prov_update_callback: unexpected event 0x%x "
5048                     " received from crypto framework\n", event));
5049                 return;
5050         }
5051 
5052         mechs = crypto_get_mech_list(&mech_count, KM_SLEEP);
5053         if (mechs == NULL)
5054                 return;
5055 
5056         /*
5057          * Walk the list of currently defined IPsec algorithm. Update
5058          * the algorithm valid flag and trigger an update of the
5059          * SAs that depend on that algorithm.
5060          */
5061         rw_enter(&ipss->ipsec_alg_lock, RW_WRITER);
5062         for (algtype = 0; algtype < IPSEC_NALGTYPES; algtype++) {
5063                 for (algidx = 0; algidx < ipss->ipsec_nalgs[algtype];
5064                     algidx++) {
5065 
5066                         algid = ipss->ipsec_sortlist[algtype][algidx];
5067                         alg = ipss->ipsec_alglists[algtype][algid];
5068                         ASSERT(alg != NULL);
5069 
5070                         /*
5071                          * Skip the algorithms which do not map to the
5072                          * crypto framework provider being added or removed.
5073                          */
5074                         if (strncmp(alg->alg_mech_name,
5075                             prov_change->ec_mech_name,
5076                             CRYPTO_MAX_MECH_NAME) != 0)
5077                                 continue;
5078 
5079                         /*
5080                          * Determine if the mechanism is valid. If it
5081                          * is not, mark the algorithm as being invalid. If
5082                          * it is, mark the algorithm as being valid.
5083                          */
5084                         for (mech_idx = 0; mech_idx < mech_count; mech_idx++)
5085                                 if (strncmp(alg->alg_mech_name,
5086                                     mechs[mech_idx], CRYPTO_MAX_MECH_NAME) == 0)
5087                                         break;
5088                         if (mech_idx == mech_count &&
5089                             alg->alg_flags & ALG_FLAG_VALID) {
5090                                 alg->alg_flags &= ~ALG_FLAG_VALID;
5091                                 alg_changed = B_TRUE;
5092                         } else if (mech_idx < mech_count &&
5093                             !(alg->alg_flags & ALG_FLAG_VALID)) {
5094                                 alg->alg_flags |= ALG_FLAG_VALID;
5095                                 alg_changed = B_TRUE;
5096                         }
5097 
5098                         /*
5099                          * Update the supported key sizes, regardless
5100                          * of whether a crypto provider was added or
5101                          * removed.
5102                          */
5103                         oalg = *alg;
5104                         ipsec_alg_fix_min_max(alg, algtype, ns);
5105                         if (!alg_changed &&
5106                             alg->alg_ef_minbits != oalg.alg_ef_minbits ||
5107                             alg->alg_ef_maxbits != oalg.alg_ef_maxbits ||
5108                             alg->alg_ef_default != oalg.alg_ef_default ||
5109                             alg->alg_ef_default_bits !=
5110                             oalg.alg_ef_default_bits)
5111                                 alg_changed = B_TRUE;
5112 
5113                         /*
5114                          * Update the affected SAs if a software provider is
5115                          * being added or removed.
5116                          */
5117                         if (prov_change->ec_provider_type ==
5118                             CRYPTO_SW_PROVIDER)
5119                                 sadb_alg_update(algtype, alg->alg_id,
5120                                     prov_change->ec_change ==
5121                                     CRYPTO_MECH_ADDED, ns);
5122                 }
5123         }
5124         rw_exit(&ipss->ipsec_alg_lock);
5125         crypto_free_mech_list(mechs, mech_count);
5126 
5127         if (alg_changed) {
5128                 /*
5129                  * An algorithm has changed, i.e. it became valid or
5130                  * invalid, or its support key sizes have changed.
5131                  * Notify ipsecah and ipsecesp of this change so
5132                  * that they can send a SADB_REGISTER to their consumers.
5133                  */
5134                 ipsecah_algs_changed(ns);
5135                 ipsecesp_algs_changed(ns);
5136         }
5137 }
5138 
5139 /*
5140  * Registers with the crypto framework to be notified of crypto
5141  * providers changes. Used to update the algorithm tables and
5142  * to free or create context templates if needed. Invoked after IPsec
5143  * is loaded successfully.
5144  *
5145  * This is called separately for each IP instance, so we ensure we only
5146  * register once.
5147  */
5148 void
5149 ipsec_register_prov_update(void)
5150 {
5151         if (prov_update_handle != NULL)
5152                 return;
5153 
5154         prov_update_handle = crypto_notify_events(
5155             ipsec_prov_update_callback, CRYPTO_EVENT_MECHS_CHANGED);
5156 }
5157 
5158 /*
5159  * Unregisters from the framework to be notified of crypto providers
5160  * changes. Called from ipsec_policy_g_destroy().
5161  */
5162 static void
5163 ipsec_unregister_prov_update(void)
5164 {
5165         if (prov_update_handle != NULL)
5166                 crypto_unnotify_events(prov_update_handle);
5167 }
5168 
5169 /*
5170  * Tunnel-mode support routines.
5171  */
5172 
5173 /*
5174  * Returns an mblk chain suitable for putnext() if policies match and IPsec
5175  * SAs are available.  If there's no per-tunnel policy, or a match comes back
5176  * with no match, then still return the packet and have global policy take
5177  * a crack at it in IP.
5178  * This updates the ip_xmit_attr with the IPsec policy.
5179  *
5180  * Remember -> we can be forwarding packets.  Keep that in mind w.r.t.
5181  * inner-packet contents.
5182  */
5183 mblk_t *
5184 ipsec_tun_outbound(mblk_t *mp, iptun_t *iptun, ipha_t *inner_ipv4,
5185     ip6_t *inner_ipv6, ipha_t *outer_ipv4, ip6_t *outer_ipv6, int outer_hdr_len,
5186     ip_xmit_attr_t *ixa)
5187 {
5188         ipsec_policy_head_t *polhead;
5189         ipsec_selector_t sel;
5190         mblk_t *nmp;
5191         boolean_t is_fragment;
5192         ipsec_policy_t *pol;
5193         ipsec_tun_pol_t *itp = iptun->iptun_itp;
5194         netstack_t *ns = iptun->iptun_ns;
5195         ipsec_stack_t *ipss = ns->netstack_ipsec;
5196 
5197         ASSERT(outer_ipv6 != NULL && outer_ipv4 == NULL ||
5198             outer_ipv4 != NULL && outer_ipv6 == NULL);
5199         /* We take care of inners in a bit. */
5200 
5201         /* Are the IPsec fields initialized at all? */
5202         if (!(ixa->ixa_flags & IXAF_IPSEC_SECURE)) {
5203                 ASSERT(ixa->ixa_ipsec_policy == NULL);
5204                 ASSERT(ixa->ixa_ipsec_latch == NULL);
5205                 ASSERT(ixa->ixa_ipsec_action == NULL);
5206                 ASSERT(ixa->ixa_ipsec_ah_sa == NULL);
5207                 ASSERT(ixa->ixa_ipsec_esp_sa == NULL);
5208         }
5209 
5210         ASSERT(itp != NULL && (itp->itp_flags & ITPF_P_ACTIVE));
5211         polhead = itp->itp_policy;
5212 
5213         bzero(&sel, sizeof (sel));
5214         if (inner_ipv4 != NULL) {
5215                 ASSERT(inner_ipv6 == NULL);
5216                 sel.ips_isv4 = B_TRUE;
5217                 sel.ips_local_addr_v4 = inner_ipv4->ipha_src;
5218                 sel.ips_remote_addr_v4 = inner_ipv4->ipha_dst;
5219                 sel.ips_protocol = (uint8_t)inner_ipv4->ipha_protocol;
5220         } else {
5221                 ASSERT(inner_ipv6 != NULL);
5222                 sel.ips_isv4 = B_FALSE;
5223                 sel.ips_local_addr_v6 = inner_ipv6->ip6_src;
5224                 /*
5225                  * We don't care about routing-header dests in the
5226                  * forwarding/tunnel path, so just grab ip6_dst.
5227                  */
5228                 sel.ips_remote_addr_v6 = inner_ipv6->ip6_dst;
5229         }
5230 
5231         if (itp->itp_flags & ITPF_P_PER_PORT_SECURITY) {
5232                 /*
5233                  * Caller can prepend the outer header, which means
5234                  * inner_ipv[46] may be stuck in the middle.  Pullup the whole
5235                  * mess now if need-be, for easier processing later.  Don't
5236                  * forget to rewire the outer header too.
5237                  */
5238                 if (mp->b_cont != NULL) {
5239                         nmp = msgpullup(mp, -1);
5240                         if (nmp == NULL) {
5241                                 ip_drop_packet(mp, B_FALSE, NULL,
5242                                     DROPPER(ipss, ipds_spd_nomem),
5243                                     &ipss->ipsec_spd_dropper);
5244                                 return (NULL);
5245                         }
5246                         freemsg(mp);
5247                         mp = nmp;
5248                         if (outer_ipv4 != NULL)
5249                                 outer_ipv4 = (ipha_t *)mp->b_rptr;
5250                         else
5251                                 outer_ipv6 = (ip6_t *)mp->b_rptr;
5252                         if (inner_ipv4 != NULL) {
5253                                 inner_ipv4 =
5254                                     (ipha_t *)(mp->b_rptr + outer_hdr_len);
5255                         } else {
5256                                 inner_ipv6 =
5257                                     (ip6_t *)(mp->b_rptr + outer_hdr_len);
5258                         }
5259                 }
5260                 if (inner_ipv4 != NULL) {
5261                         is_fragment = IS_V4_FRAGMENT(
5262                             inner_ipv4->ipha_fragment_offset_and_flags);
5263                 } else {
5264                         sel.ips_remote_addr_v6 = ip_get_dst_v6(inner_ipv6, mp,
5265                             &is_fragment);
5266                 }
5267 
5268                 if (is_fragment) {
5269                         ipha_t *oiph;
5270                         ipha_t *iph = NULL;
5271                         ip6_t *ip6h = NULL;
5272                         int hdr_len;
5273                         uint16_t ip6_hdr_length;
5274                         uint8_t v6_proto;
5275                         uint8_t *v6_proto_p;
5276 
5277                         /*
5278                          * We have a fragment we need to track!
5279                          */
5280                         mp = ipsec_fragcache_add(&itp->itp_fragcache, NULL, mp,
5281                             outer_hdr_len, ipss);
5282                         if (mp == NULL)
5283                                 return (NULL);
5284                         ASSERT(mp->b_cont == NULL);
5285 
5286                         /*
5287                          * If we get here, we have a full fragment chain
5288                          */
5289 
5290                         oiph = (ipha_t *)mp->b_rptr;
5291                         if (IPH_HDR_VERSION(oiph) == IPV4_VERSION) {
5292                                 hdr_len = ((outer_hdr_len != 0) ?
5293                                     IPH_HDR_LENGTH(oiph) : 0);
5294                                 iph = (ipha_t *)(mp->b_rptr + hdr_len);
5295                         } else {
5296                                 ASSERT(IPH_HDR_VERSION(oiph) == IPV6_VERSION);
5297                                 ip6h = (ip6_t *)mp->b_rptr;
5298                                 if (!ip_hdr_length_nexthdr_v6(mp, ip6h,
5299                                     &ip6_hdr_length, &v6_proto_p)) {
5300                                         ip_drop_packet_chain(mp, B_FALSE, NULL,
5301                                             DROPPER(ipss,
5302                                             ipds_spd_malformed_packet),
5303                                             &ipss->ipsec_spd_dropper);
5304                                         return (NULL);
5305                                 }
5306                                 hdr_len = ip6_hdr_length;
5307                         }
5308                         outer_hdr_len = hdr_len;
5309 
5310                         if (sel.ips_isv4) {
5311                                 if (iph == NULL) {
5312                                         /* Was v6 outer */
5313                                         iph = (ipha_t *)(mp->b_rptr + hdr_len);
5314                                 }
5315                                 inner_ipv4 = iph;
5316                                 sel.ips_local_addr_v4 = inner_ipv4->ipha_src;
5317                                 sel.ips_remote_addr_v4 = inner_ipv4->ipha_dst;
5318                                 sel.ips_protocol =
5319                                     (uint8_t)inner_ipv4->ipha_protocol;
5320                         } else {
5321                                 inner_ipv6 = (ip6_t *)(mp->b_rptr +
5322                                     hdr_len);
5323                                 sel.ips_local_addr_v6 = inner_ipv6->ip6_src;
5324                                 sel.ips_remote_addr_v6 = inner_ipv6->ip6_dst;
5325                                 if (!ip_hdr_length_nexthdr_v6(mp,
5326                                     inner_ipv6, &ip6_hdr_length, &v6_proto_p)) {
5327                                         ip_drop_packet_chain(mp, B_FALSE, NULL,
5328                                             DROPPER(ipss,
5329                                             ipds_spd_malformed_frag),
5330                                             &ipss->ipsec_spd_dropper);
5331                                         return (NULL);
5332                                 }
5333                                 v6_proto = *v6_proto_p;
5334                                 sel.ips_protocol = v6_proto;
5335 #ifdef FRAGCACHE_DEBUG
5336                                 cmn_err(CE_WARN, "v6_sel.ips_protocol = %d\n",
5337                                     sel.ips_protocol);
5338 #endif
5339                         }
5340                         /* Ports are extracted below */
5341                 }
5342 
5343                 /* Get ports... */
5344                 if (!ipsec_init_outbound_ports(&sel, mp,
5345                     inner_ipv4, inner_ipv6, outer_hdr_len, ipss)) {
5346                         /* callee did ip_drop_packet_chain() on mp. */
5347                         return (NULL);
5348                 }
5349 #ifdef FRAGCACHE_DEBUG
5350                 if (inner_ipv4 != NULL)
5351                         cmn_err(CE_WARN,
5352                             "(v4) sel.ips_protocol = %d, "
5353                             "sel.ips_local_port = %d, "
5354                             "sel.ips_remote_port = %d\n",
5355                             sel.ips_protocol, ntohs(sel.ips_local_port),
5356                             ntohs(sel.ips_remote_port));
5357                 if (inner_ipv6 != NULL)
5358                         cmn_err(CE_WARN,
5359                             "(v6) sel.ips_protocol = %d, "
5360                             "sel.ips_local_port = %d, "
5361                             "sel.ips_remote_port = %d\n",
5362                             sel.ips_protocol, ntohs(sel.ips_local_port),
5363                             ntohs(sel.ips_remote_port));
5364 #endif
5365                 /* Success so far! */
5366         }
5367         rw_enter(&polhead->iph_lock, RW_READER);
5368         pol = ipsec_find_policy_head(NULL, polhead, IPSEC_TYPE_OUTBOUND, &sel);
5369         rw_exit(&polhead->iph_lock);
5370         if (pol == NULL) {
5371                 /*
5372                  * No matching policy on this tunnel, drop the packet.
5373                  *
5374                  * NOTE:  Tunnel-mode tunnels are different from the
5375                  * IP global transport mode policy head.  For a tunnel-mode
5376                  * tunnel, we drop the packet in lieu of passing it
5377                  * along accepted the way a global-policy miss would.
5378                  *
5379                  * NOTE2:  "negotiate transport" tunnels should match ALL
5380                  * inbound packets, but we do not uncomment the ASSERT()
5381                  * below because if/when we open PF_POLICY, a user can
5382                  * shoot him/her-self in the foot with a 0 priority.
5383                  */
5384 
5385                 /* ASSERT(itp->itp_flags & ITPF_P_TUNNEL); */
5386 #ifdef FRAGCACHE_DEBUG
5387                 cmn_err(CE_WARN, "ipsec_tun_outbound(): No matching tunnel "
5388                     "per-port policy\n");
5389 #endif
5390                 ip_drop_packet_chain(mp, B_FALSE, NULL,
5391                     DROPPER(ipss, ipds_spd_explicit),
5392                     &ipss->ipsec_spd_dropper);
5393                 return (NULL);
5394         }
5395 
5396 #ifdef FRAGCACHE_DEBUG
5397         cmn_err(CE_WARN, "Having matching tunnel per-port policy\n");
5398 #endif
5399 
5400         /*
5401          * NOTE: ixa_cleanup() function will release pol references.
5402          */
5403         ixa->ixa_ipsec_policy = pol;
5404         /*
5405          * NOTE: There is a subtle difference between iptun_zoneid and
5406          * iptun_connp->conn_zoneid explained in iptun_conn_create().  When
5407          * interacting with the ip module, we must use conn_zoneid.
5408          */
5409         ixa->ixa_zoneid = iptun->iptun_connp->conn_zoneid;
5410 
5411         ASSERT((outer_ipv4 != NULL) ? (ixa->ixa_flags & IXAF_IS_IPV4) :
5412             !(ixa->ixa_flags & IXAF_IS_IPV4));
5413         ASSERT(ixa->ixa_ipsec_policy != NULL);
5414         ixa->ixa_flags |= IXAF_IPSEC_SECURE;
5415 
5416         if (!(itp->itp_flags & ITPF_P_TUNNEL)) {
5417                 /* Set up transport mode for tunnelled packets. */
5418                 ixa->ixa_ipsec_proto = (inner_ipv4 != NULL) ? IPPROTO_ENCAP :
5419                     IPPROTO_IPV6;
5420                 return (mp);
5421         }
5422 
5423         /* Fill in tunnel-mode goodies here. */
5424         ixa->ixa_flags |= IXAF_IPSEC_TUNNEL;
5425         /* XXX Do I need to fill in all of the goodies here? */
5426         if (inner_ipv4) {
5427                 ixa->ixa_ipsec_inaf = AF_INET;
5428                 ixa->ixa_ipsec_insrc[0] =
5429                     pol->ipsp_sel->ipsl_key.ipsl_local.ipsad_v4;
5430                 ixa->ixa_ipsec_indst[0] =
5431                     pol->ipsp_sel->ipsl_key.ipsl_remote.ipsad_v4;
5432         } else {
5433                 ixa->ixa_ipsec_inaf = AF_INET6;
5434                 ixa->ixa_ipsec_insrc[0] =
5435                     pol->ipsp_sel->ipsl_key.ipsl_local.ipsad_v6.s6_addr32[0];
5436                 ixa->ixa_ipsec_insrc[1] =
5437                     pol->ipsp_sel->ipsl_key.ipsl_local.ipsad_v6.s6_addr32[1];
5438                 ixa->ixa_ipsec_insrc[2] =
5439                     pol->ipsp_sel->ipsl_key.ipsl_local.ipsad_v6.s6_addr32[2];
5440                 ixa->ixa_ipsec_insrc[3] =
5441                     pol->ipsp_sel->ipsl_key.ipsl_local.ipsad_v6.s6_addr32[3];
5442                 ixa->ixa_ipsec_indst[0] =
5443                     pol->ipsp_sel->ipsl_key.ipsl_remote.ipsad_v6.s6_addr32[0];
5444                 ixa->ixa_ipsec_indst[1] =
5445                     pol->ipsp_sel->ipsl_key.ipsl_remote.ipsad_v6.s6_addr32[1];
5446                 ixa->ixa_ipsec_indst[2] =
5447                     pol->ipsp_sel->ipsl_key.ipsl_remote.ipsad_v6.s6_addr32[2];
5448                 ixa->ixa_ipsec_indst[3] =
5449                     pol->ipsp_sel->ipsl_key.ipsl_remote.ipsad_v6.s6_addr32[3];
5450         }
5451         ixa->ixa_ipsec_insrcpfx = pol->ipsp_sel->ipsl_key.ipsl_local_pfxlen;
5452         ixa->ixa_ipsec_indstpfx = pol->ipsp_sel->ipsl_key.ipsl_remote_pfxlen;
5453         /* NOTE:  These are used for transport mode too. */
5454         ixa->ixa_ipsec_src_port = pol->ipsp_sel->ipsl_key.ipsl_lport;
5455         ixa->ixa_ipsec_dst_port = pol->ipsp_sel->ipsl_key.ipsl_rport;
5456         ixa->ixa_ipsec_proto = pol->ipsp_sel->ipsl_key.ipsl_proto;
5457 
5458         return (mp);
5459 }
5460 
5461 /*
5462  * NOTE: The following releases pol's reference and
5463  * calls ip_drop_packet() for me on NULL returns.
5464  */
5465 mblk_t *
5466 ipsec_check_ipsecin_policy_reasm(mblk_t *attr_mp, ipsec_policy_t *pol,
5467     ipha_t *inner_ipv4, ip6_t *inner_ipv6, uint64_t pkt_unique, netstack_t *ns)
5468 {
5469         /* Assume attr_mp is a chain of b_next-linked ip_recv_attr mblk. */
5470         mblk_t *data_chain = NULL, *data_tail = NULL;
5471         mblk_t *next;
5472         mblk_t *data_mp;
5473         ip_recv_attr_t  iras;
5474 
5475         while (attr_mp != NULL) {
5476                 ASSERT(ip_recv_attr_is_mblk(attr_mp));
5477                 next = attr_mp->b_next;
5478                 attr_mp->b_next = NULL;  /* No tripping asserts. */
5479 
5480                 data_mp = attr_mp->b_cont;
5481                 attr_mp->b_cont = NULL;
5482                 if (!ip_recv_attr_from_mblk(attr_mp, &iras)) {
5483                         /* The ill or ip_stack_t disappeared on us */
5484                         freemsg(data_mp);       /* ip_drop_packet?? */
5485                         ira_cleanup(&iras, B_TRUE);
5486                         goto fail;
5487                 }
5488 
5489                 /*
5490                  * Need IPPOL_REFHOLD(pol) for extras because
5491                  * ipsecin_policy does the refrele.
5492                  */
5493                 IPPOL_REFHOLD(pol);
5494 
5495                 data_mp = ipsec_check_ipsecin_policy(data_mp, pol, inner_ipv4,
5496                     inner_ipv6, pkt_unique, &iras, ns);
5497                 ira_cleanup(&iras, B_TRUE);
5498 
5499                 if (data_mp == NULL)
5500                         goto fail;
5501 
5502                 if (data_tail == NULL) {
5503                         /* First one */
5504                         data_chain = data_tail = data_mp;
5505                 } else {
5506                         data_tail->b_next = data_mp;
5507                         data_tail = data_mp;
5508                 }
5509                 attr_mp = next;
5510         }
5511         /*
5512          * One last release because either the loop bumped it up, or we never
5513          * called ipsec_check_ipsecin_policy().
5514          */
5515         IPPOL_REFRELE(pol);
5516 
5517         /* data_chain is ready for return to tun module. */
5518         return (data_chain);
5519 
5520 fail:
5521         /*
5522          * Need to get rid of any extra pol
5523          * references, and any remaining bits as well.
5524          */
5525         IPPOL_REFRELE(pol);
5526         ipsec_freemsg_chain(data_chain);
5527         ipsec_freemsg_chain(next);      /* ipdrop stats? */
5528         return (NULL);
5529 }
5530 
5531 /*
5532  * Return a message if the inbound packet passed an IPsec policy check.  Returns
5533  * NULL if it failed or if it is a fragment needing its friends before a
5534  * policy check can be performed.
5535  *
5536  * Expects a non-NULL data_mp, and a non-NULL polhead.
5537  * The returned mblk may be a b_next chain of packets if fragments
5538  * neeeded to be collected for a proper policy check.
5539  *
5540  * This function calls ip_drop_packet() on data_mp if need be.
5541  *
5542  * NOTE:  outer_hdr_len is signed.  If it's a negative value, the caller
5543  * is inspecting an ICMP packet.
5544  */
5545 mblk_t *
5546 ipsec_tun_inbound(ip_recv_attr_t *ira, mblk_t *data_mp, ipsec_tun_pol_t *itp,
5547     ipha_t *inner_ipv4, ip6_t *inner_ipv6, ipha_t *outer_ipv4,
5548     ip6_t *outer_ipv6, int outer_hdr_len, netstack_t *ns)
5549 {
5550         ipsec_policy_head_t *polhead;
5551         ipsec_selector_t sel;
5552         ipsec_policy_t *pol;
5553         uint16_t tmpport;
5554         selret_t rc;
5555         boolean_t port_policy_present, is_icmp, global_present;
5556         in6_addr_t tmpaddr;
5557         ipaddr_t tmp4;
5558         uint8_t flags, *inner_hdr;
5559         ipsec_stack_t *ipss = ns->netstack_ipsec;
5560 
5561         sel.ips_is_icmp_inv_acq = 0;
5562 
5563         if (outer_ipv4 != NULL) {
5564                 ASSERT(outer_ipv6 == NULL);
5565                 global_present = ipss->ipsec_inbound_v4_policy_present;
5566         } else {
5567                 ASSERT(outer_ipv6 != NULL);
5568                 global_present = ipss->ipsec_inbound_v6_policy_present;
5569         }
5570 
5571         ASSERT(inner_ipv4 != NULL && inner_ipv6 == NULL ||
5572             inner_ipv4 == NULL && inner_ipv6 != NULL);
5573 
5574         if (outer_hdr_len < 0) {
5575                 outer_hdr_len = (-outer_hdr_len);
5576                 is_icmp = B_TRUE;
5577         } else {
5578                 is_icmp = B_FALSE;
5579         }
5580 
5581         if (itp != NULL && (itp->itp_flags & ITPF_P_ACTIVE)) {
5582                 mblk_t *mp = data_mp;
5583 
5584                 polhead = itp->itp_policy;
5585                 /*
5586                  * We need to perform full Tunnel-Mode enforcement,
5587                  * and we need to have inner-header data for such enforcement.
5588                  *
5589                  * See ipsec_init_inbound_sel() for the 0x80000000 on inbound
5590                  * and on return.
5591                  */
5592 
5593                 port_policy_present = ((itp->itp_flags &
5594                     ITPF_P_PER_PORT_SECURITY) ? B_TRUE : B_FALSE);
5595                 /*
5596                  * NOTE:  Even if our policy is transport mode, set the
5597                  * SEL_TUNNEL_MODE flag so ipsec_init_inbound_sel() can
5598                  * do the right thing w.r.t. outer headers.
5599                  */
5600                 flags = ((port_policy_present ? SEL_PORT_POLICY : SEL_NONE) |
5601                     (is_icmp ? SEL_IS_ICMP : SEL_NONE) | SEL_TUNNEL_MODE);
5602 
5603                 rc = ipsec_init_inbound_sel(&sel, data_mp, inner_ipv4,
5604                     inner_ipv6, flags);
5605 
5606                 switch (rc) {
5607                 case SELRET_NOMEM:
5608                         ip_drop_packet(data_mp, B_TRUE, NULL,
5609                             DROPPER(ipss, ipds_spd_nomem),
5610                             &ipss->ipsec_spd_dropper);
5611                         return (NULL);
5612                 case SELRET_TUNFRAG:
5613                         /*
5614                          * At this point, if we're cleartext, we don't want
5615                          * to go there.
5616                          */
5617                         if (!(ira->ira_flags & IRAF_IPSEC_SECURE)) {
5618                                 ip_drop_packet(data_mp, B_TRUE, NULL,
5619                                     DROPPER(ipss, ipds_spd_got_clear),
5620                                     &ipss->ipsec_spd_dropper);
5621                                 return (NULL);
5622                         }
5623 
5624                         /*
5625                          * Inner and outer headers may not be contiguous.
5626                          * Pullup the data_mp now to satisfy assumptions of
5627                          * ipsec_fragcache_add()
5628                          */
5629                         if (data_mp->b_cont != NULL) {
5630                                 mblk_t *nmp;
5631 
5632                                 nmp = msgpullup(data_mp, -1);
5633                                 if (nmp == NULL) {
5634                                         ip_drop_packet(data_mp, B_TRUE, NULL,
5635                                             DROPPER(ipss, ipds_spd_nomem),
5636                                             &ipss->ipsec_spd_dropper);
5637                                         return (NULL);
5638                                 }
5639                                 freemsg(data_mp);
5640                                 data_mp = nmp;
5641                                 if (outer_ipv4 != NULL)
5642                                         outer_ipv4 =
5643                                             (ipha_t *)data_mp->b_rptr;
5644                                 else
5645                                         outer_ipv6 =
5646                                             (ip6_t *)data_mp->b_rptr;
5647                                 if (inner_ipv4 != NULL) {
5648                                         inner_ipv4 =
5649                                             (ipha_t *)(data_mp->b_rptr +
5650                                             outer_hdr_len);
5651                                 } else {
5652                                         inner_ipv6 =
5653                                             (ip6_t *)(data_mp->b_rptr +
5654                                             outer_hdr_len);
5655                                 }
5656                         }
5657 
5658                         /*
5659                          * If we need to queue the packet. First we
5660                          * get an mblk with the attributes. ipsec_fragcache_add
5661                          * will prepend that to the queued data and return
5662                          * a list of b_next messages each of which starts with
5663                          * the attribute mblk.
5664                          */
5665                         mp = ip_recv_attr_to_mblk(ira);
5666                         if (mp == NULL) {
5667                                 ip_drop_packet(data_mp, B_TRUE, NULL,
5668                                     DROPPER(ipss, ipds_spd_nomem),
5669                                     &ipss->ipsec_spd_dropper);
5670                                 return (NULL);
5671                         }
5672 
5673                         mp = ipsec_fragcache_add(&itp->itp_fragcache,
5674                             mp, data_mp, outer_hdr_len, ipss);
5675 
5676                         if (mp == NULL) {
5677                                 /*
5678                                  * Data is cached, fragment chain is not
5679                                  * complete.
5680                                  */
5681                                 return (NULL);
5682                         }
5683 
5684                         /*
5685                          * If we get here, we have a full fragment chain.
5686                          * Reacquire headers and selectors from first fragment.
5687                          */
5688                         ASSERT(ip_recv_attr_is_mblk(mp));
5689                         data_mp = mp->b_cont;
5690                         inner_hdr = data_mp->b_rptr;
5691                         if (outer_ipv4 != NULL) {
5692                                 inner_hdr += IPH_HDR_LENGTH(
5693                                     (ipha_t *)data_mp->b_rptr);
5694                         } else {
5695                                 inner_hdr += ip_hdr_length_v6(data_mp,
5696                                     (ip6_t *)data_mp->b_rptr);
5697                         }
5698                         ASSERT(inner_hdr <= data_mp->b_wptr);
5699 
5700                         if (inner_ipv4 != NULL) {
5701                                 inner_ipv4 = (ipha_t *)inner_hdr;
5702                                 inner_ipv6 = NULL;
5703                         } else {
5704                                 inner_ipv6 = (ip6_t *)inner_hdr;
5705                                 inner_ipv4 = NULL;
5706                         }
5707 
5708                         /*
5709                          * Use SEL_TUNNEL_MODE to take into account the outer
5710                          * header.  Use SEL_POST_FRAG so we always get ports.
5711                          */
5712                         rc = ipsec_init_inbound_sel(&sel, data_mp,
5713                             inner_ipv4, inner_ipv6,
5714                             SEL_TUNNEL_MODE | SEL_POST_FRAG);
5715                         switch (rc) {
5716                         case SELRET_SUCCESS:
5717                                 /*
5718                                  * Get to same place as first caller's
5719                                  * SELRET_SUCCESS case.
5720                                  */
5721                                 break;
5722                         case SELRET_NOMEM:
5723                                 ip_drop_packet_chain(mp, B_TRUE, NULL,
5724                                     DROPPER(ipss, ipds_spd_nomem),
5725                                     &ipss->ipsec_spd_dropper);
5726                                 return (NULL);
5727                         case SELRET_BADPKT:
5728                                 ip_drop_packet_chain(mp, B_TRUE, NULL,
5729                                     DROPPER(ipss, ipds_spd_malformed_frag),
5730                                     &ipss->ipsec_spd_dropper);
5731                                 return (NULL);
5732                         case SELRET_TUNFRAG:
5733                                 cmn_err(CE_WARN, "(TUNFRAG on 2nd call...)");
5734                                 /* FALLTHRU */
5735                         default:
5736                                 cmn_err(CE_WARN, "ipsec_init_inbound_sel(mark2)"
5737                                     " returns bizarro 0x%x", rc);
5738                                 /* Guaranteed panic! */
5739                                 ASSERT(rc == SELRET_NOMEM);
5740                                 return (NULL);
5741                         }
5742                         /* FALLTHRU */
5743                 case SELRET_SUCCESS:
5744                         /*
5745                          * Common case:
5746                          * No per-port policy or a non-fragment.  Keep going.
5747                          */
5748                         break;
5749                 case SELRET_BADPKT:
5750                         /*
5751                          * We may receive ICMP (with IPv6 inner) packets that
5752                          * trigger this return value.  Send 'em in for
5753                          * enforcement checking.
5754                          */
5755                         cmn_err(CE_NOTE, "ipsec_tun_inbound(): "
5756                             "sending 'bad packet' in for enforcement");
5757                         break;
5758                 default:
5759                         cmn_err(CE_WARN,
5760                             "ipsec_init_inbound_sel() returns bizarro 0x%x",
5761                             rc);
5762                         ASSERT(rc == SELRET_NOMEM);     /* Guaranteed panic! */
5763                         return (NULL);
5764                 }
5765 
5766                 if (is_icmp) {
5767                         /*
5768                          * Swap local/remote because this is an ICMP packet.
5769                          */
5770                         tmpaddr = sel.ips_local_addr_v6;
5771                         sel.ips_local_addr_v6 = sel.ips_remote_addr_v6;
5772                         sel.ips_remote_addr_v6 = tmpaddr;
5773                         tmpport = sel.ips_local_port;
5774                         sel.ips_local_port = sel.ips_remote_port;
5775                         sel.ips_remote_port = tmpport;
5776                 }
5777 
5778                 /* find_policy_head() */
5779                 rw_enter(&polhead->iph_lock, RW_READER);
5780                 pol = ipsec_find_policy_head(NULL, polhead, IPSEC_TYPE_INBOUND,
5781                     &sel);
5782                 rw_exit(&polhead->iph_lock);
5783                 if (pol != NULL) {
5784                         uint64_t pkt_unique;
5785 
5786                         if (!(ira->ira_flags & IRAF_IPSEC_SECURE)) {
5787                                 if (!pol->ipsp_act->ipa_allow_clear) {
5788                                         /*
5789                                          * XXX should never get here with
5790                                          * tunnel reassembled fragments?
5791                                          */
5792                                         ASSERT(mp == data_mp);
5793                                         ip_drop_packet(data_mp, B_TRUE, NULL,
5794                                             DROPPER(ipss, ipds_spd_got_clear),
5795                                             &ipss->ipsec_spd_dropper);
5796                                         IPPOL_REFRELE(pol);
5797                                         return (NULL);
5798                                 } else {
5799                                         IPPOL_REFRELE(pol);
5800                                         return (mp);
5801                                 }
5802                         }
5803                         pkt_unique = SA_UNIQUE_ID(sel.ips_remote_port,
5804                             sel.ips_local_port,
5805                             (inner_ipv4 == NULL) ? IPPROTO_IPV6 :
5806                             IPPROTO_ENCAP, sel.ips_protocol);
5807 
5808                         /*
5809                          * NOTE: The following releases pol's reference and
5810                          * calls ip_drop_packet() for me on NULL returns.
5811                          *
5812                          * "sel" is still good here, so let's use it!
5813                          */
5814                         if (data_mp == mp) {
5815                                 /* A single packet without attributes */
5816                                 data_mp = ipsec_check_ipsecin_policy(data_mp,
5817                                     pol, inner_ipv4, inner_ipv6, pkt_unique,
5818                                     ira, ns);
5819                         } else {
5820                                 /*
5821                                  * We pass in the b_next chain of attr_mp's
5822                                  * and get back a b_next chain of data_mp's.
5823                                  */
5824                                 data_mp = ipsec_check_ipsecin_policy_reasm(mp,
5825                                     pol, inner_ipv4, inner_ipv6, pkt_unique,
5826                                     ns);
5827                         }
5828                         return (data_mp);
5829                 }
5830 
5831                 /*
5832                  * Else fallthru and check the global policy on the outer
5833                  * header(s) if this tunnel is an old-style transport-mode
5834                  * one.  Drop the packet explicitly (no policy entry) for
5835                  * a new-style tunnel-mode tunnel.
5836                  */
5837                 if ((itp->itp_flags & ITPF_P_TUNNEL) && !is_icmp) {
5838                         ip_drop_packet_chain(data_mp, B_TRUE, NULL,
5839                             DROPPER(ipss, ipds_spd_explicit),
5840                             &ipss->ipsec_spd_dropper);
5841                         return (NULL);
5842                 }
5843         }
5844 
5845         /*
5846          * NOTE:  If we reach here, we will not have packet chains from
5847          * fragcache_add(), because the only way I get chains is on a
5848          * tunnel-mode tunnel, which either returns with a pass, or gets
5849          * hit by the ip_drop_packet_chain() call right above here.
5850          */
5851         ASSERT(data_mp->b_next == NULL);
5852 
5853         /* If no per-tunnel security, check global policy now. */
5854         if ((ira->ira_flags & IRAF_IPSEC_SECURE) && !global_present) {
5855                 if (ira->ira_flags & IRAF_TRUSTED_ICMP) {
5856                         /*
5857                          * This is an ICMP message that was geenrated locally.
5858                          * We should accept it.
5859                          */
5860                         return (data_mp);
5861                 }
5862 
5863                 ip_drop_packet(data_mp, B_TRUE, NULL,
5864                     DROPPER(ipss, ipds_spd_got_secure),
5865                     &ipss->ipsec_spd_dropper);
5866                 return (NULL);
5867         }
5868 
5869         if (is_icmp) {
5870                 /*
5871                  * For ICMP packets, "outer_ipvN" is set to the outer header
5872                  * that is *INSIDE* the ICMP payload.  For global policy
5873                  * checking, we need to reverse src/dst on the payload in
5874                  * order to construct selectors appropriately.  See "ripha"
5875                  * constructions in ip.c.  To avoid a bug like 6478464 (see
5876                  * earlier in this file), we will actually exchange src/dst
5877                  * in the packet, and reverse if after the call to
5878                  * ipsec_check_global_policy().
5879                  */
5880                 if (outer_ipv4 != NULL) {
5881                         tmp4 = outer_ipv4->ipha_src;
5882                         outer_ipv4->ipha_src = outer_ipv4->ipha_dst;
5883                         outer_ipv4->ipha_dst = tmp4;
5884                 } else {
5885                         ASSERT(outer_ipv6 != NULL);
5886                         tmpaddr = outer_ipv6->ip6_src;
5887                         outer_ipv6->ip6_src = outer_ipv6->ip6_dst;
5888                         outer_ipv6->ip6_dst = tmpaddr;
5889                 }
5890         }
5891 
5892         data_mp = ipsec_check_global_policy(data_mp, NULL, outer_ipv4,
5893             outer_ipv6, ira, ns);
5894         if (data_mp == NULL)
5895                 return (NULL);
5896 
5897         if (is_icmp) {
5898                 /* Set things back to normal. */
5899                 if (outer_ipv4 != NULL) {
5900                         tmp4 = outer_ipv4->ipha_src;
5901                         outer_ipv4->ipha_src = outer_ipv4->ipha_dst;
5902                         outer_ipv4->ipha_dst = tmp4;
5903                 } else {
5904                         /* No need for ASSERT()s now. */
5905                         tmpaddr = outer_ipv6->ip6_src;
5906                         outer_ipv6->ip6_src = outer_ipv6->ip6_dst;
5907                         outer_ipv6->ip6_dst = tmpaddr;
5908                 }
5909         }
5910 
5911         /*
5912          * At this point, we pretend it's a cleartext accepted
5913          * packet.
5914          */
5915         return (data_mp);
5916 }
5917 
5918 /*
5919  * AVL comparison routine for our list of tunnel polheads.
5920  */
5921 static int
5922 tunnel_compare(const void *arg1, const void *arg2)
5923 {
5924         ipsec_tun_pol_t *left, *right;
5925         int rc;
5926 
5927         left = (ipsec_tun_pol_t *)arg1;
5928         right = (ipsec_tun_pol_t *)arg2;
5929 
5930         rc = strncmp(left->itp_name, right->itp_name, LIFNAMSIZ);
5931         return (rc == 0 ? rc : (rc > 0 ? 1 : -1));
5932 }
5933 
5934 /*
5935  * Free a tunnel policy node.
5936  */
5937 void
5938 itp_free(ipsec_tun_pol_t *node, netstack_t *ns)
5939 {
5940         if (node->itp_policy != NULL) {
5941                 IPPH_REFRELE(node->itp_policy, ns);
5942                 node->itp_policy = NULL;
5943         }
5944         if (node->itp_inactive != NULL) {
5945                 IPPH_REFRELE(node->itp_inactive, ns);
5946                 node->itp_inactive = NULL;
5947         }
5948         mutex_destroy(&node->itp_lock);
5949         kmem_free(node, sizeof (*node));
5950 }
5951 
5952 void
5953 itp_unlink(ipsec_tun_pol_t *node, netstack_t *ns)
5954 {
5955         ipsec_stack_t *ipss = ns->netstack_ipsec;
5956 
5957         rw_enter(&ipss->ipsec_tunnel_policy_lock, RW_WRITER);
5958         ipss->ipsec_tunnel_policy_gen++;
5959         ipsec_fragcache_uninit(&node->itp_fragcache, ipss);
5960         avl_remove(&ipss->ipsec_tunnel_policies, node);
5961         rw_exit(&ipss->ipsec_tunnel_policy_lock);
5962         ITP_REFRELE(node, ns);
5963 }
5964 
5965 /*
5966  * Public interface to look up a tunnel security policy by name.  Used by
5967  * spdsock mostly.  Returns "node" with a bumped refcnt.
5968  */
5969 ipsec_tun_pol_t *
5970 get_tunnel_policy(char *name, netstack_t *ns)
5971 {
5972         ipsec_tun_pol_t *node, lookup;
5973         ipsec_stack_t *ipss = ns->netstack_ipsec;
5974 
5975         (void) strncpy(lookup.itp_name, name, LIFNAMSIZ);
5976 
5977         rw_enter(&ipss->ipsec_tunnel_policy_lock, RW_READER);
5978         node = (ipsec_tun_pol_t *)avl_find(&ipss->ipsec_tunnel_policies,
5979             &lookup, NULL);
5980         if (node != NULL) {
5981                 ITP_REFHOLD(node);
5982         }
5983         rw_exit(&ipss->ipsec_tunnel_policy_lock);
5984 
5985         return (node);
5986 }
5987 
5988 /*
5989  * Public interface to walk all tunnel security polcies.  Useful for spdsock
5990  * DUMP operations.  iterator() will not consume a reference.
5991  */
5992 void
5993 itp_walk(void (*iterator)(ipsec_tun_pol_t *, void *, netstack_t *),
5994     void *arg, netstack_t *ns)
5995 {
5996         ipsec_tun_pol_t *node;
5997         ipsec_stack_t *ipss = ns->netstack_ipsec;
5998 
5999         rw_enter(&ipss->ipsec_tunnel_policy_lock, RW_READER);
6000         for (node = avl_first(&ipss->ipsec_tunnel_policies); node != NULL;
6001             node = AVL_NEXT(&ipss->ipsec_tunnel_policies, node)) {
6002                 iterator(node, arg, ns);
6003         }
6004         rw_exit(&ipss->ipsec_tunnel_policy_lock);
6005 }
6006 
6007 /*
6008  * Initialize policy head.  This can only fail if there's a memory problem.
6009  */
6010 static boolean_t
6011 tunnel_polhead_init(ipsec_policy_head_t *iph, netstack_t *ns)
6012 {
6013         ipsec_stack_t *ipss = ns->netstack_ipsec;
6014 
6015         rw_init(&iph->iph_lock, NULL, RW_DEFAULT, NULL);
6016         iph->iph_refs = 1;
6017         iph->iph_gen = 0;
6018         if (ipsec_alloc_table(iph, ipss->ipsec_tun_spd_hashsize,
6019             KM_SLEEP, B_FALSE, ns) != 0) {
6020                 ipsec_polhead_free_table(iph);
6021                 return (B_FALSE);
6022         }
6023         ipsec_polhead_init(iph, ipss->ipsec_tun_spd_hashsize);
6024         return (B_TRUE);
6025 }
6026 
6027 /*
6028  * Create a tunnel policy node with "name".  Set errno with
6029  * ENOMEM if there's a memory problem, and EEXIST if there's an existing
6030  * node.
6031  */
6032 ipsec_tun_pol_t *
6033 create_tunnel_policy(char *name, int *errno, uint64_t *gen, netstack_t *ns)
6034 {
6035         ipsec_tun_pol_t *newbie, *existing;
6036         avl_index_t where;
6037         ipsec_stack_t *ipss = ns->netstack_ipsec;
6038 
6039         newbie = kmem_zalloc(sizeof (*newbie), KM_NOSLEEP);
6040         if (newbie == NULL) {
6041                 *errno = ENOMEM;
6042                 return (NULL);
6043         }
6044         if (!ipsec_fragcache_init(&newbie->itp_fragcache)) {
6045                 kmem_free(newbie, sizeof (*newbie));
6046                 *errno = ENOMEM;
6047                 return (NULL);
6048         }
6049 
6050         (void) strncpy(newbie->itp_name, name, LIFNAMSIZ);
6051 
6052         rw_enter(&ipss->ipsec_tunnel_policy_lock, RW_WRITER);
6053         existing = (ipsec_tun_pol_t *)avl_find(&ipss->ipsec_tunnel_policies,
6054             newbie, &where);
6055         if (existing != NULL) {
6056                 itp_free(newbie, ns);
6057                 *errno = EEXIST;
6058                 rw_exit(&ipss->ipsec_tunnel_policy_lock);
6059                 return (NULL);
6060         }
6061         ipss->ipsec_tunnel_policy_gen++;
6062         *gen = ipss->ipsec_tunnel_policy_gen;
6063         newbie->itp_refcnt = 2;      /* One for the caller, one for the tree. */
6064         newbie->itp_next_policy_index = 1;
6065         avl_insert(&ipss->ipsec_tunnel_policies, newbie, where);
6066         mutex_init(&newbie->itp_lock, NULL, MUTEX_DEFAULT, NULL);
6067         newbie->itp_policy = kmem_zalloc(sizeof (ipsec_policy_head_t),
6068             KM_NOSLEEP);
6069         if (newbie->itp_policy == NULL)
6070                 goto nomem;
6071         newbie->itp_inactive = kmem_zalloc(sizeof (ipsec_policy_head_t),
6072             KM_NOSLEEP);
6073         if (newbie->itp_inactive == NULL) {
6074                 kmem_free(newbie->itp_policy, sizeof (ipsec_policy_head_t));
6075                 goto nomem;
6076         }
6077 
6078         if (!tunnel_polhead_init(newbie->itp_policy, ns)) {
6079                 kmem_free(newbie->itp_policy, sizeof (ipsec_policy_head_t));
6080                 kmem_free(newbie->itp_inactive, sizeof (ipsec_policy_head_t));
6081                 goto nomem;
6082         } else if (!tunnel_polhead_init(newbie->itp_inactive, ns)) {
6083                 IPPH_REFRELE(newbie->itp_policy, ns);
6084                 kmem_free(newbie->itp_inactive, sizeof (ipsec_policy_head_t));
6085                 goto nomem;
6086         }
6087         rw_exit(&ipss->ipsec_tunnel_policy_lock);
6088 
6089         return (newbie);
6090 nomem:
6091         *errno = ENOMEM;
6092         kmem_free(newbie, sizeof (*newbie));
6093         return (NULL);
6094 }
6095 
6096 /*
6097  * Given two addresses, find a tunnel instance's IPsec policy heads.
6098  * Returns NULL on failure.
6099  */
6100 ipsec_tun_pol_t *
6101 itp_get_byaddr(uint32_t *laddr, uint32_t *faddr, int af, ip_stack_t *ipst)
6102 {
6103         conn_t *connp;
6104         iptun_t *iptun;
6105         ipsec_tun_pol_t *itp = NULL;
6106 
6107         /* Classifiers are used to "src" being foreign. */
6108         if (af == AF_INET) {
6109                 connp = ipcl_iptun_classify_v4((ipaddr_t *)faddr,
6110                     (ipaddr_t *)laddr, ipst);
6111         } else {
6112                 ASSERT(af == AF_INET6);
6113                 ASSERT(!IN6_IS_ADDR_V4MAPPED((in6_addr_t *)laddr));
6114                 ASSERT(!IN6_IS_ADDR_V4MAPPED((in6_addr_t *)faddr));
6115                 connp = ipcl_iptun_classify_v6((in6_addr_t *)faddr,
6116                     (in6_addr_t *)laddr, ipst);
6117         }
6118 
6119         if (connp == NULL)
6120                 return (NULL);
6121 
6122         if (IPCL_IS_IPTUN(connp)) {
6123                 iptun = connp->conn_iptun;
6124                 if (iptun != NULL) {
6125                         itp = iptun->iptun_itp;
6126                         if (itp != NULL) {
6127                                 /* Braces due to the macro's nature... */
6128                                 ITP_REFHOLD(itp);
6129                         }
6130                 }  /* Else itp is already NULL. */
6131         }
6132 
6133         CONN_DEC_REF(connp);
6134         return (itp);
6135 }
6136 
6137 /*
6138  * Frag cache code, based on SunScreen 3.2 source
6139  *      screen/kernel/common/screen_fragcache.c
6140  */
6141 
6142 #define IPSEC_FRAG_TTL_MAX      5
6143 /*
6144  * Note that the following parameters create 256 hash buckets
6145  * with 1024 free entries to be distributed.  Things are cleaned
6146  * periodically and are attempted to be cleaned when there is no
6147  * free space, but this system errs on the side of dropping packets
6148  * over creating memory exhaustion.  We may decide to make hash
6149  * factor a tunable if this proves to be a bad decision.
6150  */
6151 #define IPSEC_FRAG_HASH_SLOTS   (1<<8)
6152 #define IPSEC_FRAG_HASH_FACTOR  4
6153 #define IPSEC_FRAG_HASH_SIZE    (IPSEC_FRAG_HASH_SLOTS * IPSEC_FRAG_HASH_FACTOR)
6154 
6155 #define IPSEC_FRAG_HASH_MASK            (IPSEC_FRAG_HASH_SLOTS - 1)
6156 #define IPSEC_FRAG_HASH_FUNC(id)        (((id) & IPSEC_FRAG_HASH_MASK) ^ \
6157                                             (((id) / \
6158                                             (ushort_t)IPSEC_FRAG_HASH_SLOTS) & \
6159                                             IPSEC_FRAG_HASH_MASK))
6160 
6161 /* Maximum fragments per packet.  48 bytes payload x 1366 packets > 64KB */
6162 #define IPSEC_MAX_FRAGS         1366
6163 
6164 #define V4_FRAG_OFFSET(ipha) ((ntohs(ipha->ipha_fragment_offset_and_flags) & \
6165                                     IPH_OFFSET) << 3)
6166 #define V4_MORE_FRAGS(ipha) (ntohs(ipha->ipha_fragment_offset_and_flags) & \
6167                 IPH_MF)
6168 
6169 /*
6170  * Initialize an ipsec fragcache instance.
6171  * Returns B_FALSE if memory allocation fails.
6172  */
6173 boolean_t
6174 ipsec_fragcache_init(ipsec_fragcache_t *frag)
6175 {
6176         ipsec_fragcache_entry_t *ftemp;
6177         int i;
6178 
6179         mutex_init(&frag->itpf_lock, NULL, MUTEX_DEFAULT, NULL);
6180         frag->itpf_ptr = (ipsec_fragcache_entry_t **)
6181             kmem_zalloc(sizeof (ipsec_fragcache_entry_t *) *
6182             IPSEC_FRAG_HASH_SLOTS, KM_NOSLEEP);
6183         if (frag->itpf_ptr == NULL)
6184                 return (B_FALSE);
6185 
6186         ftemp = (ipsec_fragcache_entry_t *)
6187             kmem_zalloc(sizeof (ipsec_fragcache_entry_t) *
6188             IPSEC_FRAG_HASH_SIZE, KM_NOSLEEP);
6189         if (ftemp == NULL) {
6190                 kmem_free(frag->itpf_ptr, sizeof (ipsec_fragcache_entry_t *) *
6191                     IPSEC_FRAG_HASH_SLOTS);
6192                 return (B_FALSE);
6193         }
6194 
6195         frag->itpf_freelist = NULL;
6196 
6197         for (i = 0; i < IPSEC_FRAG_HASH_SIZE; i++) {
6198                 ftemp->itpfe_next = frag->itpf_freelist;
6199                 frag->itpf_freelist = ftemp;
6200                 ftemp++;
6201         }
6202 
6203         frag->itpf_expire_hint = 0;
6204 
6205         return (B_TRUE);
6206 }
6207 
6208 void
6209 ipsec_fragcache_uninit(ipsec_fragcache_t *frag, ipsec_stack_t *ipss)
6210 {
6211         ipsec_fragcache_entry_t *fep;
6212         int i;
6213 
6214         mutex_enter(&frag->itpf_lock);
6215         if (frag->itpf_ptr) {
6216                 /* Delete any existing fragcache entry chains */
6217                 for (i = 0; i < IPSEC_FRAG_HASH_SLOTS; i++) {
6218                         fep = (frag->itpf_ptr)[i];
6219                         while (fep != NULL) {
6220                                 /* Returned fep is next in chain or NULL */
6221                                 fep = fragcache_delentry(i, fep, frag, ipss);
6222                         }
6223                 }
6224                 /*
6225                  * Chase the pointers back to the beginning
6226                  * of the memory allocation and then
6227                  * get rid of the allocated freelist
6228                  */
6229                 while (frag->itpf_freelist->itpfe_next != NULL)
6230                         frag->itpf_freelist = frag->itpf_freelist->itpfe_next;
6231                 /*
6232                  * XXX - If we ever dynamically grow the freelist
6233                  * then we'll have to free entries individually
6234                  * or determine how many entries or chunks we have
6235                  * grown since the initial allocation.
6236                  */
6237                 kmem_free(frag->itpf_freelist,
6238                     sizeof (ipsec_fragcache_entry_t) *
6239                     IPSEC_FRAG_HASH_SIZE);
6240                 /* Free the fragcache structure */
6241                 kmem_free(frag->itpf_ptr,
6242                     sizeof (ipsec_fragcache_entry_t *) *
6243                     IPSEC_FRAG_HASH_SLOTS);
6244         }
6245         mutex_exit(&frag->itpf_lock);
6246         mutex_destroy(&frag->itpf_lock);
6247 }
6248 
6249 /*
6250  * Add a fragment to the fragment cache.   Consumes mp if NULL is returned.
6251  * Returns mp if a whole fragment has been assembled, NULL otherwise
6252  * The returned mp could be a b_next chain of fragments.
6253  *
6254  * The iramp argument is set on inbound; NULL if outbound.
6255  */
6256 mblk_t *
6257 ipsec_fragcache_add(ipsec_fragcache_t *frag, mblk_t *iramp, mblk_t *mp,
6258     int outer_hdr_len, ipsec_stack_t *ipss)
6259 {
6260         boolean_t is_v4;
6261         time_t itpf_time;
6262         ipha_t *iph;
6263         ipha_t *oiph;
6264         ip6_t *ip6h = NULL;
6265         uint8_t v6_proto;
6266         uint8_t *v6_proto_p;
6267         uint16_t ip6_hdr_length;
6268         ip_pkt_t ipp;
6269         ip6_frag_t *fraghdr;
6270         ipsec_fragcache_entry_t *fep;
6271         int i;
6272         mblk_t *nmp, *prevmp;
6273         int firstbyte, lastbyte;
6274         int offset;
6275         int last;
6276         boolean_t inbound = (iramp != NULL);
6277 
6278 #ifdef FRAGCACHE_DEBUG
6279         cmn_err(CE_WARN, "Fragcache: %s\n", inbound ? "INBOUND" : "OUTBOUND");
6280 #endif
6281         /*
6282          * You're on the slow path, so insure that every packet in the
6283          * cache is a single-mblk one.
6284          */
6285         if (mp->b_cont != NULL) {
6286                 nmp = msgpullup(mp, -1);
6287                 if (nmp == NULL) {
6288                         ip_drop_packet(mp, inbound, NULL,
6289                             DROPPER(ipss, ipds_spd_nomem),
6290                             &ipss->ipsec_spd_dropper);
6291                         if (inbound)
6292                                 (void) ip_recv_attr_free_mblk(iramp);
6293                         return (NULL);
6294                 }
6295                 freemsg(mp);
6296                 mp = nmp;
6297         }
6298 
6299         mutex_enter(&frag->itpf_lock);
6300 
6301         oiph  = (ipha_t *)mp->b_rptr;
6302         iph  = (ipha_t *)(mp->b_rptr + outer_hdr_len);
6303 
6304         if (IPH_HDR_VERSION(iph) == IPV4_VERSION) {
6305                 is_v4 = B_TRUE;
6306         } else {
6307                 ASSERT(IPH_HDR_VERSION(iph) == IPV6_VERSION);
6308                 ip6h = (ip6_t *)(mp->b_rptr + outer_hdr_len);
6309 
6310                 if (!ip_hdr_length_nexthdr_v6(mp, ip6h, &ip6_hdr_length,
6311                     &v6_proto_p)) {
6312                         /*
6313                          * Find upper layer protocol.
6314                          * If it fails we have a malformed packet
6315                          */
6316                         mutex_exit(&frag->itpf_lock);
6317                         ip_drop_packet(mp, inbound, NULL,
6318                             DROPPER(ipss, ipds_spd_malformed_packet),
6319                             &ipss->ipsec_spd_dropper);
6320                         if (inbound)
6321                                 (void) ip_recv_attr_free_mblk(iramp);
6322                         return (NULL);
6323                 } else {
6324                         v6_proto = *v6_proto_p;
6325                 }
6326 
6327 
6328                 bzero(&ipp, sizeof (ipp));
6329                 (void) ip_find_hdr_v6(mp, ip6h, B_FALSE, &ipp, NULL);
6330                 if (!(ipp.ipp_fields & IPPF_FRAGHDR)) {
6331                         /*
6332                          * We think this is a fragment, but didn't find
6333                          * a fragment header.  Something is wrong.
6334                          */
6335                         mutex_exit(&frag->itpf_lock);
6336                         ip_drop_packet(mp, inbound, NULL,
6337                             DROPPER(ipss, ipds_spd_malformed_frag),
6338                             &ipss->ipsec_spd_dropper);
6339                         if (inbound)
6340                                 (void) ip_recv_attr_free_mblk(iramp);
6341                         return (NULL);
6342                 }
6343                 fraghdr = ipp.ipp_fraghdr;
6344                 is_v4 = B_FALSE;
6345         }
6346 
6347         /* Anything to cleanup? */
6348 
6349         /*
6350          * This cleanup call could be put in a timer loop
6351          * but it may actually be just as reasonable a decision to
6352          * leave it here.  The disadvantage is this only gets called when
6353          * frags are added.  The advantage is that it is not
6354          * susceptible to race conditions like a time-based cleanup
6355          * may be.
6356          */
6357         itpf_time = gethrestime_sec();
6358         if (itpf_time >= frag->itpf_expire_hint)
6359                 ipsec_fragcache_clean(frag, ipss);
6360 
6361         /* Lookup to see if there is an existing entry */
6362 
6363         if (is_v4)
6364                 i = IPSEC_FRAG_HASH_FUNC(iph->ipha_ident);
6365         else
6366                 i = IPSEC_FRAG_HASH_FUNC(fraghdr->ip6f_ident);
6367 
6368         for (fep = (frag->itpf_ptr)[i]; fep; fep = fep->itpfe_next) {
6369                 if (is_v4) {
6370                         ASSERT(iph != NULL);
6371                         if ((fep->itpfe_id == iph->ipha_ident) &&
6372                             (fep->itpfe_src == iph->ipha_src) &&
6373                             (fep->itpfe_dst == iph->ipha_dst) &&
6374                             (fep->itpfe_proto == iph->ipha_protocol))
6375                                 break;
6376                 } else {
6377                         ASSERT(fraghdr != NULL);
6378                         ASSERT(fep != NULL);
6379                         if ((fep->itpfe_id == fraghdr->ip6f_ident) &&
6380                             IN6_ARE_ADDR_EQUAL(&fep->itpfe_src6,
6381                             &ip6h->ip6_src) &&
6382                             IN6_ARE_ADDR_EQUAL(&fep->itpfe_dst6,
6383                             &ip6h->ip6_dst) && (fep->itpfe_proto == v6_proto))
6384                                 break;
6385                 }
6386         }
6387 
6388         if (is_v4) {
6389                 firstbyte = V4_FRAG_OFFSET(iph);
6390                 lastbyte  = firstbyte + ntohs(iph->ipha_length) -
6391                     IPH_HDR_LENGTH(iph);
6392                 last = (V4_MORE_FRAGS(iph) == 0);
6393 #ifdef FRAGCACHE_DEBUG
6394                 cmn_err(CE_WARN, "V4 fragcache: firstbyte = %d, lastbyte = %d, "
6395                     "is_last_frag = %d, id = %d, mp = %p\n", firstbyte,
6396                     lastbyte, last, iph->ipha_ident, mp);
6397 #endif
6398         } else {
6399                 firstbyte = ntohs(fraghdr->ip6f_offlg & IP6F_OFF_MASK);
6400                 lastbyte  = firstbyte + ntohs(ip6h->ip6_plen) +
6401                     sizeof (ip6_t) - ip6_hdr_length;
6402                 last = (fraghdr->ip6f_offlg & IP6F_MORE_FRAG) == 0;
6403 #ifdef FRAGCACHE_DEBUG
6404                 cmn_err(CE_WARN, "V6 fragcache: firstbyte = %d, lastbyte = %d, "
6405                     "is_last_frag = %d, id = %d, fraghdr = %p, mp = %p\n",
6406                     firstbyte, lastbyte, last, fraghdr->ip6f_ident, fraghdr,
6407                     mp);
6408 #endif
6409         }
6410 
6411         /* check for bogus fragments and delete the entry */
6412         if (firstbyte > 0 && firstbyte <= 8) {
6413                 if (fep != NULL)
6414                         (void) fragcache_delentry(i, fep, frag, ipss);
6415                 mutex_exit(&frag->itpf_lock);
6416                 ip_drop_packet(mp, inbound, NULL,
6417                     DROPPER(ipss, ipds_spd_malformed_frag),
6418                     &ipss->ipsec_spd_dropper);
6419                 if (inbound)
6420                         (void) ip_recv_attr_free_mblk(iramp);
6421                 return (NULL);
6422         }
6423 
6424         /* Not found, allocate a new entry */
6425         if (fep == NULL) {
6426                 if (frag->itpf_freelist == NULL) {
6427                         /* see if there is some space */
6428                         ipsec_fragcache_clean(frag, ipss);
6429                         if (frag->itpf_freelist == NULL) {
6430                                 mutex_exit(&frag->itpf_lock);
6431                                 ip_drop_packet(mp, inbound, NULL,
6432                                     DROPPER(ipss, ipds_spd_nomem),
6433                                     &ipss->ipsec_spd_dropper);
6434                                 if (inbound)
6435                                         (void) ip_recv_attr_free_mblk(iramp);
6436                                 return (NULL);
6437                         }
6438                 }
6439 
6440                 fep = frag->itpf_freelist;
6441                 frag->itpf_freelist = fep->itpfe_next;
6442 
6443                 if (is_v4) {
6444                         bcopy((caddr_t)&iph->ipha_src, (caddr_t)&fep->itpfe_src,
6445                             sizeof (struct in_addr));
6446                         bcopy((caddr_t)&iph->ipha_dst, (caddr_t)&fep->itpfe_dst,
6447                             sizeof (struct in_addr));
6448                         fep->itpfe_id = iph->ipha_ident;
6449                         fep->itpfe_proto = iph->ipha_protocol;
6450                         i = IPSEC_FRAG_HASH_FUNC(fep->itpfe_id);
6451                 } else {
6452                         bcopy((in6_addr_t *)&ip6h->ip6_src,
6453                             (in6_addr_t *)&fep->itpfe_src6,
6454                             sizeof (struct in6_addr));
6455                         bcopy((in6_addr_t *)&ip6h->ip6_dst,
6456                             (in6_addr_t *)&fep->itpfe_dst6,
6457                             sizeof (struct in6_addr));
6458                         fep->itpfe_id = fraghdr->ip6f_ident;
6459                         fep->itpfe_proto = v6_proto;
6460                         i = IPSEC_FRAG_HASH_FUNC(fep->itpfe_id);
6461                 }
6462                 itpf_time = gethrestime_sec();
6463                 fep->itpfe_exp = itpf_time + IPSEC_FRAG_TTL_MAX + 1;
6464                 fep->itpfe_last = 0;
6465                 fep->itpfe_fraglist = NULL;
6466                 fep->itpfe_depth = 0;
6467                 fep->itpfe_next = (frag->itpf_ptr)[i];
6468                 (frag->itpf_ptr)[i] = fep;
6469 
6470                 if (frag->itpf_expire_hint > fep->itpfe_exp)
6471                         frag->itpf_expire_hint = fep->itpfe_exp;
6472 
6473         }
6474 
6475         /* Insert it in the frag list */
6476         /* List is in order by starting offset of fragments */
6477 
6478         prevmp = NULL;
6479         for (nmp = fep->itpfe_fraglist; nmp; nmp = nmp->b_next) {
6480                 ipha_t *niph;
6481                 ipha_t *oniph;
6482                 ip6_t *nip6h;
6483                 ip_pkt_t nipp;
6484                 ip6_frag_t *nfraghdr;
6485                 uint16_t nip6_hdr_length;
6486                 uint8_t *nv6_proto_p;
6487                 int nfirstbyte, nlastbyte;
6488                 char *data, *ndata;
6489                 mblk_t *ndata_mp = (inbound ? nmp->b_cont : nmp);
6490                 int hdr_len;
6491 
6492                 oniph  = (ipha_t *)mp->b_rptr;
6493                 nip6h = NULL;
6494                 niph = NULL;
6495 
6496                 /*
6497                  * Determine outer header type and length and set
6498                  * pointers appropriately
6499                  */
6500 
6501                 if (IPH_HDR_VERSION(oniph) == IPV4_VERSION) {
6502                         hdr_len = ((outer_hdr_len != 0) ?
6503                             IPH_HDR_LENGTH(oiph) : 0);
6504                         niph = (ipha_t *)(ndata_mp->b_rptr + hdr_len);
6505                 } else {
6506                         ASSERT(IPH_HDR_VERSION(oniph) == IPV6_VERSION);
6507                         ASSERT(ndata_mp->b_cont == NULL);
6508                         nip6h = (ip6_t *)ndata_mp->b_rptr;
6509                         (void) ip_hdr_length_nexthdr_v6(ndata_mp, nip6h,
6510                             &nip6_hdr_length, &v6_proto_p);
6511                         hdr_len = ((outer_hdr_len != 0) ? nip6_hdr_length : 0);
6512                 }
6513 
6514                 /*
6515                  * Determine inner header type and length and set
6516                  * pointers appropriately
6517                  */
6518 
6519                 if (is_v4) {
6520                         if (niph == NULL) {
6521                                 /* Was v6 outer */
6522                                 niph = (ipha_t *)(ndata_mp->b_rptr + hdr_len);
6523                         }
6524                         nfirstbyte = V4_FRAG_OFFSET(niph);
6525                         nlastbyte = nfirstbyte + ntohs(niph->ipha_length) -
6526                             IPH_HDR_LENGTH(niph);
6527                 } else {
6528                         ASSERT(ndata_mp->b_cont == NULL);
6529                         nip6h = (ip6_t *)(ndata_mp->b_rptr + hdr_len);
6530                         if (!ip_hdr_length_nexthdr_v6(ndata_mp, nip6h,
6531                             &nip6_hdr_length, &nv6_proto_p)) {
6532                                 mutex_exit(&frag->itpf_lock);
6533                                 ip_drop_packet_chain(nmp, inbound, NULL,
6534                                     DROPPER(ipss, ipds_spd_malformed_frag),
6535                                     &ipss->ipsec_spd_dropper);
6536                                 ipsec_freemsg_chain(ndata_mp);
6537                                 if (inbound)
6538                                         (void) ip_recv_attr_free_mblk(iramp);
6539                                 return (NULL);
6540                         }
6541                         bzero(&nipp, sizeof (nipp));
6542                         (void) ip_find_hdr_v6(ndata_mp, nip6h, B_FALSE, &nipp,
6543                             NULL);
6544                         nfraghdr = nipp.ipp_fraghdr;
6545                         nfirstbyte = ntohs(nfraghdr->ip6f_offlg &
6546                             IP6F_OFF_MASK);
6547                         nlastbyte  = nfirstbyte + ntohs(nip6h->ip6_plen) +
6548                             sizeof (ip6_t) - nip6_hdr_length;
6549                 }
6550 
6551                 /* Check for overlapping fragments */
6552                 if (firstbyte >= nfirstbyte && firstbyte < nlastbyte) {
6553                         /*
6554                          * Overlap Check:
6555                          *  ~~~~---------               # Check if the newly
6556                          * ~    ndata_mp|               # received fragment
6557                          *  ~~~~---------               # overlaps with the
6558                          *       ---------~~~~~~        # current fragment.
6559                          *      |    mp         ~
6560                          *       ---------~~~~~~
6561                          */
6562                         if (is_v4) {
6563                                 data  = (char *)iph  + IPH_HDR_LENGTH(iph) +
6564                                     firstbyte - nfirstbyte;
6565                                 ndata = (char *)niph + IPH_HDR_LENGTH(niph);
6566                         } else {
6567                                 data  = (char *)ip6h  +
6568                                     nip6_hdr_length + firstbyte -
6569                                     nfirstbyte;
6570                                 ndata = (char *)nip6h + nip6_hdr_length;
6571                         }
6572                         if (bcmp(data, ndata, MIN(lastbyte, nlastbyte) -
6573                             firstbyte)) {
6574                                 /* Overlapping data does not match */
6575                                 (void) fragcache_delentry(i, fep, frag, ipss);
6576                                 mutex_exit(&frag->itpf_lock);
6577                                 ip_drop_packet(mp, inbound, NULL,
6578                                     DROPPER(ipss, ipds_spd_overlap_frag),
6579                                     &ipss->ipsec_spd_dropper);
6580                                 if (inbound)
6581                                         (void) ip_recv_attr_free_mblk(iramp);
6582                                 return (NULL);
6583                         }
6584                         /* Part of defense for jolt2.c fragmentation attack */
6585                         if (firstbyte >= nfirstbyte && lastbyte <= nlastbyte) {
6586                                 /*
6587                                  * Check for identical or subset fragments:
6588                                  *  ----------      ~~~~--------~~~~~
6589                                  * |    nmp   | or  ~      nmp      ~
6590                                  *  ----------      ~~~~--------~~~~~
6591                                  *  ----------            ------
6592                                  * |    mp    |          |  mp  |
6593                                  *  ----------            ------
6594                                  */
6595                                 mutex_exit(&frag->itpf_lock);
6596                                 ip_drop_packet(mp, inbound, NULL,
6597                                     DROPPER(ipss, ipds_spd_evil_frag),
6598                                     &ipss->ipsec_spd_dropper);
6599                                 if (inbound)
6600                                         (void) ip_recv_attr_free_mblk(iramp);
6601                                 return (NULL);
6602                         }
6603 
6604                 }
6605 
6606                 /* Correct location for this fragment? */
6607                 if (firstbyte <= nfirstbyte) {
6608                         /*
6609                          * Check if the tail end of the new fragment overlaps
6610                          * with the head of the current fragment.
6611                          *        --------~~~~~~~
6612                          *       |    nmp       ~
6613                          *        --------~~~~~~~
6614                          *  ~~~~~--------
6615                          *  ~   mp       |
6616                          *  ~~~~~--------
6617                          */
6618                         if (lastbyte > nfirstbyte) {
6619                                 /* Fragments overlap */
6620                                 data  = (char *)iph  + IPH_HDR_LENGTH(iph) +
6621                                     firstbyte - nfirstbyte;
6622                                 ndata = (char *)niph + IPH_HDR_LENGTH(niph);
6623                                 if (is_v4) {
6624                                         data  = (char *)iph +
6625                                             IPH_HDR_LENGTH(iph) + firstbyte -
6626                                             nfirstbyte;
6627                                         ndata = (char *)niph +
6628                                             IPH_HDR_LENGTH(niph);
6629                                 } else {
6630                                         data  = (char *)ip6h  +
6631                                             nip6_hdr_length + firstbyte -
6632                                             nfirstbyte;
6633                                         ndata = (char *)nip6h + nip6_hdr_length;
6634                                 }
6635                                 if (bcmp(data, ndata, MIN(lastbyte, nlastbyte)
6636                                     - nfirstbyte)) {
6637                                         /* Overlap mismatch */
6638                                         (void) fragcache_delentry(i, fep, frag,
6639                                             ipss);
6640                                         mutex_exit(&frag->itpf_lock);
6641                                         ip_drop_packet(mp, inbound, NULL,
6642                                             DROPPER(ipss,
6643                                             ipds_spd_overlap_frag),
6644                                             &ipss->ipsec_spd_dropper);
6645                                         if (inbound) {
6646                                                 (void) ip_recv_attr_free_mblk(
6647                                                     iramp);
6648                                         }
6649                                         return (NULL);
6650                                 }
6651                         }
6652 
6653                         /*
6654                          * Fragment does not illegally overlap and can now
6655                          * be inserted into the chain
6656                          */
6657                         break;
6658                 }
6659 
6660                 prevmp = nmp;
6661         }
6662         /* Prepend the attributes before we link it in */
6663         if (iramp != NULL) {
6664                 ASSERT(iramp->b_cont == NULL);
6665                 iramp->b_cont = mp;
6666                 mp = iramp;
6667                 iramp = NULL;
6668         }
6669         mp->b_next = nmp;
6670 
6671         if (prevmp == NULL) {
6672                 fep->itpfe_fraglist = mp;
6673         } else {
6674                 prevmp->b_next = mp;
6675         }
6676         if (last)
6677                 fep->itpfe_last = 1;
6678 
6679         /* Part of defense for jolt2.c fragmentation attack */
6680         if (++(fep->itpfe_depth) > IPSEC_MAX_FRAGS) {
6681                 (void) fragcache_delentry(i, fep, frag, ipss);
6682                 mutex_exit(&frag->itpf_lock);
6683                 if (inbound)
6684                         mp = ip_recv_attr_free_mblk(mp);
6685 
6686                 ip_drop_packet(mp, inbound, NULL,
6687                     DROPPER(ipss, ipds_spd_max_frags),
6688                     &ipss->ipsec_spd_dropper);
6689                 return (NULL);
6690         }
6691 
6692         /* Check for complete packet */
6693 
6694         if (!fep->itpfe_last) {
6695                 mutex_exit(&frag->itpf_lock);
6696 #ifdef FRAGCACHE_DEBUG
6697                 cmn_err(CE_WARN, "Fragment cached, last not yet seen.\n");
6698 #endif
6699                 return (NULL);
6700         }
6701 
6702         offset = 0;
6703         for (mp = fep->itpfe_fraglist; mp; mp = mp->b_next) {
6704                 mblk_t *data_mp = (inbound ? mp->b_cont : mp);
6705                 int hdr_len;
6706 
6707                 oiph  = (ipha_t *)data_mp->b_rptr;
6708                 ip6h = NULL;
6709                 iph = NULL;
6710 
6711                 if (IPH_HDR_VERSION(oiph) == IPV4_VERSION) {
6712                         hdr_len = ((outer_hdr_len != 0) ?
6713                             IPH_HDR_LENGTH(oiph) : 0);
6714                         iph = (ipha_t *)(data_mp->b_rptr + hdr_len);
6715                 } else {
6716                         ASSERT(IPH_HDR_VERSION(oiph) == IPV6_VERSION);
6717                         ASSERT(data_mp->b_cont == NULL);
6718                         ip6h = (ip6_t *)data_mp->b_rptr;
6719                         (void) ip_hdr_length_nexthdr_v6(data_mp, ip6h,
6720                             &ip6_hdr_length, &v6_proto_p);
6721                         hdr_len = ((outer_hdr_len != 0) ? ip6_hdr_length : 0);
6722                 }
6723 
6724                 /* Calculate current fragment start/end */
6725                 if (is_v4) {
6726                         if (iph == NULL) {
6727                                 /* Was v6 outer */
6728                                 iph = (ipha_t *)(data_mp->b_rptr + hdr_len);
6729                         }
6730                         firstbyte = V4_FRAG_OFFSET(iph);
6731                         lastbyte = firstbyte + ntohs(iph->ipha_length) -
6732                             IPH_HDR_LENGTH(iph);
6733                 } else {
6734                         ASSERT(data_mp->b_cont == NULL);
6735                         ip6h = (ip6_t *)(data_mp->b_rptr + hdr_len);
6736                         if (!ip_hdr_length_nexthdr_v6(data_mp, ip6h,
6737                             &ip6_hdr_length, &v6_proto_p)) {
6738                                 mutex_exit(&frag->itpf_lock);
6739                                 ip_drop_packet_chain(mp, inbound, NULL,
6740                                     DROPPER(ipss, ipds_spd_malformed_frag),
6741                                     &ipss->ipsec_spd_dropper);
6742                                 return (NULL);
6743                         }
6744                         v6_proto = *v6_proto_p;
6745                         bzero(&ipp, sizeof (ipp));
6746                         (void) ip_find_hdr_v6(data_mp, ip6h, B_FALSE, &ipp,
6747                             NULL);
6748                         fraghdr = ipp.ipp_fraghdr;
6749                         firstbyte = ntohs(fraghdr->ip6f_offlg &
6750                             IP6F_OFF_MASK);
6751                         lastbyte  = firstbyte + ntohs(ip6h->ip6_plen) +
6752                             sizeof (ip6_t) - ip6_hdr_length;
6753                 }
6754 
6755                 /*
6756                  * If this fragment is greater than current offset,
6757                  * we have a missing fragment so return NULL
6758                  */
6759                 if (firstbyte > offset) {
6760                         mutex_exit(&frag->itpf_lock);
6761 #ifdef FRAGCACHE_DEBUG
6762                         /*
6763                          * Note, this can happen when the last frag
6764                          * gets sent through because it is smaller
6765                          * than the MTU.  It is not necessarily an
6766                          * error condition.
6767                          */
6768                         cmn_err(CE_WARN, "Frag greater than offset! : "
6769                             "missing fragment: firstbyte = %d, offset = %d, "
6770                             "mp = %p\n", firstbyte, offset, mp);
6771 #endif
6772                         return (NULL);
6773                 }
6774 #ifdef FRAGCACHE_DEBUG
6775                 cmn_err(CE_WARN, "Frag offsets : "
6776                     "firstbyte = %d, offset = %d, mp = %p\n",
6777                     firstbyte, offset, mp);
6778 #endif
6779 
6780                 /*
6781                  * If we are at the last fragment, we have the complete
6782                  * packet, so rechain things and return it to caller
6783                  * for processing
6784                  */
6785 
6786                 if ((is_v4 && !V4_MORE_FRAGS(iph)) ||
6787                     (!is_v4 && !(fraghdr->ip6f_offlg & IP6F_MORE_FRAG))) {
6788                         mp = fep->itpfe_fraglist;
6789                         fep->itpfe_fraglist = NULL;
6790                         (void) fragcache_delentry(i, fep, frag, ipss);
6791                         mutex_exit(&frag->itpf_lock);
6792 
6793                         if ((is_v4 && (firstbyte + ntohs(iph->ipha_length) >
6794                             65535)) || (!is_v4 && (firstbyte +
6795                             ntohs(ip6h->ip6_plen) > 65535))) {
6796                                 /* It is an invalid "ping-o-death" packet */
6797                                 /* Discard it */
6798                                 ip_drop_packet_chain(mp, inbound, NULL,
6799                                     DROPPER(ipss, ipds_spd_evil_frag),
6800                                     &ipss->ipsec_spd_dropper);
6801                                 return (NULL);
6802                         }
6803 #ifdef FRAGCACHE_DEBUG
6804                         cmn_err(CE_WARN, "Fragcache returning mp = %p, "
6805                             "mp->b_next = %p", mp, mp->b_next);
6806 #endif
6807                         /*
6808                          * For inbound case, mp has attrmp b_next'd chain
6809                          * For outbound case, it is just data mp chain
6810                          */
6811                         return (mp);
6812                 }
6813 
6814                 /*
6815                  * Update new ending offset if this
6816                  * fragment extends the packet
6817                  */
6818                 if (offset < lastbyte)
6819                         offset = lastbyte;
6820         }
6821 
6822         mutex_exit(&frag->itpf_lock);
6823 
6824         /* Didn't find last fragment, so return NULL */
6825         return (NULL);
6826 }
6827 
6828 static void
6829 ipsec_fragcache_clean(ipsec_fragcache_t *frag, ipsec_stack_t *ipss)
6830 {
6831         ipsec_fragcache_entry_t *fep;
6832         int i;
6833         ipsec_fragcache_entry_t *earlyfep = NULL;
6834         time_t itpf_time;
6835         int earlyexp;
6836         int earlyi = 0;
6837 
6838         ASSERT(MUTEX_HELD(&frag->itpf_lock));
6839 
6840         itpf_time = gethrestime_sec();
6841         earlyexp = itpf_time + 10000;
6842 
6843         for (i = 0; i < IPSEC_FRAG_HASH_SLOTS; i++) {
6844                 fep = (frag->itpf_ptr)[i];
6845                 while (fep) {
6846                         if (fep->itpfe_exp < itpf_time) {
6847                                 /* found */
6848                                 fep = fragcache_delentry(i, fep, frag, ipss);
6849                         } else {
6850                                 if (fep->itpfe_exp < earlyexp) {
6851                                         earlyfep = fep;
6852                                         earlyexp = fep->itpfe_exp;
6853                                         earlyi = i;
6854                                 }
6855                                 fep = fep->itpfe_next;
6856                         }
6857                 }
6858         }
6859 
6860         frag->itpf_expire_hint = earlyexp;
6861 
6862         /* if (!found) */
6863         if (frag->itpf_freelist == NULL)
6864                 (void) fragcache_delentry(earlyi, earlyfep, frag, ipss);
6865 }
6866 
6867 static ipsec_fragcache_entry_t *
6868 fragcache_delentry(int slot, ipsec_fragcache_entry_t *fep,
6869     ipsec_fragcache_t *frag, ipsec_stack_t *ipss)
6870 {
6871         ipsec_fragcache_entry_t *targp;
6872         ipsec_fragcache_entry_t *nextp = fep->itpfe_next;
6873 
6874         ASSERT(MUTEX_HELD(&frag->itpf_lock));
6875 
6876         /* Free up any fragment list still in cache entry */
6877         if (fep->itpfe_fraglist != NULL) {
6878                 ip_drop_packet_chain(fep->itpfe_fraglist,
6879                     ip_recv_attr_is_mblk(fep->itpfe_fraglist), NULL,
6880                     DROPPER(ipss, ipds_spd_expired_frags),
6881                     &ipss->ipsec_spd_dropper);
6882         }
6883         fep->itpfe_fraglist = NULL;
6884 
6885         targp = (frag->itpf_ptr)[slot];
6886         ASSERT(targp != 0);
6887 
6888         if (targp == fep) {
6889                 /* unlink from head of hash chain */
6890                 (frag->itpf_ptr)[slot] = nextp;
6891                 /* link into free list */
6892                 fep->itpfe_next = frag->itpf_freelist;
6893                 frag->itpf_freelist = fep;
6894                 return (nextp);
6895         }
6896 
6897         /* maybe should use double linked list to make update faster */
6898         /* must be past front of chain */
6899         while (targp) {
6900                 if (targp->itpfe_next == fep) {
6901                         /* unlink from hash chain */
6902                         targp->itpfe_next = nextp;
6903                         /* link into free list */
6904                         fep->itpfe_next = frag->itpf_freelist;
6905                         frag->itpf_freelist = fep;
6906                         return (nextp);
6907                 }
6908                 targp = targp->itpfe_next;
6909                 ASSERT(targp != 0);
6910         }
6911         /* NOTREACHED */
6912         return (NULL);
6913 }