1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 /*
  22  * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
  23  * Use is subject to license terms.
  24  * Copyright (c) 2012 Nexenta Systems, Inc. All rights reserved.
  25  * Copyright (c) 2017 Joyent, Inc.
  26  */
  27 
  28 #include <sys/types.h>
  29 #include <sys/stream.h>
  30 #include <sys/stropts.h>
  31 #include <sys/strsubr.h>
  32 #include <sys/errno.h>
  33 #include <sys/ddi.h>
  34 #include <sys/debug.h>
  35 #include <sys/cmn_err.h>
  36 #include <sys/stream.h>
  37 #include <sys/strlog.h>
  38 #include <sys/kmem.h>
  39 #include <sys/sunddi.h>
  40 #include <sys/tihdr.h>
  41 #include <sys/atomic.h>
  42 #include <sys/socket.h>
  43 #include <sys/sysmacros.h>
  44 #include <sys/crypto/common.h>
  45 #include <sys/crypto/api.h>
  46 #include <sys/zone.h>
  47 #include <netinet/in.h>
  48 #include <net/if.h>
  49 #include <net/pfkeyv2.h>
  50 #include <net/pfpolicy.h>
  51 #include <inet/common.h>
  52 #include <netinet/ip6.h>
  53 #include <inet/ip.h>
  54 #include <inet/ip_ire.h>
  55 #include <inet/ip6.h>
  56 #include <inet/ipsec_info.h>
  57 #include <inet/tcp.h>
  58 #include <inet/sadb.h>
  59 #include <inet/ipsec_impl.h>
  60 #include <inet/ipsecah.h>
  61 #include <inet/ipsecesp.h>
  62 #include <sys/random.h>
  63 #include <sys/dlpi.h>
  64 #include <sys/strsun.h>
  65 #include <sys/strsubr.h>
  66 #include <inet/ip_if.h>
  67 #include <inet/ipdrop.h>
  68 #include <inet/ipclassifier.h>
  69 #include <inet/sctp_ip.h>
  70 #include <sys/tsol/tnet.h>
  71 
  72 /*
  73  * This source file contains Security Association Database (SADB) common
  74  * routines.  They are linked in with the AH module.  Since AH has no chance
  75  * of falling under export control, it was safe to link it in there.
  76  */
  77 
  78 static uint8_t *sadb_action_to_ecomb(uint8_t *, uint8_t *, ipsec_action_t *,
  79     netstack_t *);
  80 static ipsa_t *sadb_torch_assoc(isaf_t *, ipsa_t *);
  81 static void sadb_destroy_acqlist(iacqf_t **, uint_t, boolean_t,
  82                             netstack_t *);
  83 static void sadb_destroy(sadb_t *, netstack_t *);
  84 static mblk_t *sadb_sa2msg(ipsa_t *, sadb_msg_t *);
  85 static ts_label_t *sadb_label_from_sens(sadb_sens_t *, uint64_t *);
  86 
  87 static time_t sadb_add_time(time_t, uint64_t);
  88 static void lifetime_fuzz(ipsa_t *);
  89 static void age_pair_peer_list(templist_t *, sadb_t *, boolean_t);
  90 static int get_ipsa_pair(ipsa_query_t *, ipsap_t *, int *);
  91 static void init_ipsa_pair(ipsap_t *);
  92 static void destroy_ipsa_pair(ipsap_t *);
  93 static int update_pairing(ipsap_t *, ipsa_query_t *, keysock_in_t *, int *);
  94 static void ipsa_set_replay(ipsa_t *ipsa, uint32_t offset);
  95 
  96 /*
  97  * ipsacq_maxpackets is defined here to make it tunable
  98  * from /etc/system.
  99  */
 100 extern uint64_t ipsacq_maxpackets;
 101 
 102 #define SET_EXPIRE(sa, delta, exp) {                            \
 103         if (((sa)->ipsa_ ## delta) != 0) {                           \
 104                 (sa)->ipsa_ ## exp = sadb_add_time((sa)->ipsa_addtime,    \
 105                         (sa)->ipsa_ ## delta);                               \
 106         }                                                               \
 107 }
 108 
 109 #define UPDATE_EXPIRE(sa, delta, exp) {                                 \
 110         if (((sa)->ipsa_ ## delta) != 0) {                           \
 111                 time_t tmp = sadb_add_time((sa)->ipsa_usetime,               \
 112                         (sa)->ipsa_ ## delta);                               \
 113                 if (((sa)->ipsa_ ## exp) == 0)                               \
 114                         (sa)->ipsa_ ## exp = tmp;                    \
 115                 else                                                    \
 116                         (sa)->ipsa_ ## exp =                                 \
 117                             MIN((sa)->ipsa_ ## exp, tmp);            \
 118         }                                                               \
 119 }
 120 
 121 
 122 /* wrap the macro so we can pass it as a function pointer */
 123 void
 124 sadb_sa_refrele(void *target)
 125 {
 126         IPSA_REFRELE(((ipsa_t *)target));
 127 }
 128 
 129 /*
 130  * We presume that sizeof (long) == sizeof (time_t) and that time_t is
 131  * a signed type.
 132  */
 133 #define TIME_MAX LONG_MAX
 134 
 135 /*
 136  * PF_KEY gives us lifetimes in uint64_t seconds.  We presume that
 137  * time_t is defined to be a signed type with the same range as
 138  * "long".  On ILP32 systems, we thus run the risk of wrapping around
 139  * at end of time, as well as "overwrapping" the clock back around
 140  * into a seemingly valid but incorrect future date earlier than the
 141  * desired expiration.
 142  *
 143  * In order to avoid odd behavior (either negative lifetimes or loss
 144  * of high order bits) when someone asks for bizarrely long SA
 145  * lifetimes, we do a saturating add for expire times.
 146  *
 147  * We presume that ILP32 systems will be past end of support life when
 148  * the 32-bit time_t overflows (a dangerous assumption, mind you..).
 149  *
 150  * On LP64, 2^64 seconds are about 5.8e11 years, at which point we
 151  * will hopefully have figured out clever ways to avoid the use of
 152  * fixed-sized integers in computation.
 153  */
 154 static time_t
 155 sadb_add_time(time_t base, uint64_t delta)
 156 {
 157         time_t sum;
 158 
 159         /*
 160          * Clip delta to the maximum possible time_t value to
 161          * prevent "overwrapping" back into a shorter-than-desired
 162          * future time.
 163          */
 164         if (delta > TIME_MAX)
 165                 delta = TIME_MAX;
 166         /*
 167          * This sum may still overflow.
 168          */
 169         sum = base + delta;
 170 
 171         /*
 172          * .. so if the result is less than the base, we overflowed.
 173          */
 174         if (sum < base)
 175                 sum = TIME_MAX;
 176 
 177         return (sum);
 178 }
 179 
 180 /*
 181  * Callers of this function have already created a working security
 182  * association, and have found the appropriate table & hash chain.  All this
 183  * function does is check duplicates, and insert the SA.  The caller needs to
 184  * hold the hash bucket lock and increment the refcnt before insertion.
 185  *
 186  * Return 0 if success, EEXIST if collision.
 187  */
 188 #define SA_UNIQUE_MATCH(sa1, sa2) \
 189         (((sa1)->ipsa_unique_id & (sa1)->ipsa_unique_mask) == \
 190         ((sa2)->ipsa_unique_id & (sa2)->ipsa_unique_mask))
 191 
 192 int
 193 sadb_insertassoc(ipsa_t *ipsa, isaf_t *bucket)
 194 {
 195         ipsa_t **ptpn = NULL;
 196         ipsa_t *walker;
 197         boolean_t unspecsrc;
 198 
 199         ASSERT(MUTEX_HELD(&bucket->isaf_lock));
 200 
 201         unspecsrc = IPSA_IS_ADDR_UNSPEC(ipsa->ipsa_srcaddr, ipsa->ipsa_addrfam);
 202 
 203         walker = bucket->isaf_ipsa;
 204         ASSERT(walker == NULL || ipsa->ipsa_addrfam == walker->ipsa_addrfam);
 205 
 206         /*
 207          * Find insertion point (pointed to with **ptpn).  Insert at the head
 208          * of the list unless there's an unspecified source address, then
 209          * insert it after the last SA with a specified source address.
 210          *
 211          * BTW, you'll have to walk the whole chain, matching on {DST, SPI}
 212          * checking for collisions.
 213          */
 214 
 215         while (walker != NULL) {
 216                 if (IPSA_ARE_ADDR_EQUAL(walker->ipsa_dstaddr,
 217                     ipsa->ipsa_dstaddr, ipsa->ipsa_addrfam)) {
 218                         if (walker->ipsa_spi == ipsa->ipsa_spi)
 219                                 return (EEXIST);
 220 
 221                         mutex_enter(&walker->ipsa_lock);
 222                         if (ipsa->ipsa_state == IPSA_STATE_MATURE &&
 223                             (walker->ipsa_flags & IPSA_F_USED) &&
 224                             SA_UNIQUE_MATCH(walker, ipsa)) {
 225                                 walker->ipsa_flags |= IPSA_F_CINVALID;
 226                         }
 227                         mutex_exit(&walker->ipsa_lock);
 228                 }
 229 
 230                 if (ptpn == NULL && unspecsrc) {
 231                         if (IPSA_IS_ADDR_UNSPEC(walker->ipsa_srcaddr,
 232                             walker->ipsa_addrfam))
 233                                 ptpn = walker->ipsa_ptpn;
 234                         else if (walker->ipsa_next == NULL)
 235                                 ptpn = &walker->ipsa_next;
 236                 }
 237 
 238                 walker = walker->ipsa_next;
 239         }
 240 
 241         if (ptpn == NULL)
 242                 ptpn = &bucket->isaf_ipsa;
 243         ipsa->ipsa_next = *ptpn;
 244         ipsa->ipsa_ptpn = ptpn;
 245         if (ipsa->ipsa_next != NULL)
 246                 ipsa->ipsa_next->ipsa_ptpn = &ipsa->ipsa_next;
 247         *ptpn = ipsa;
 248         ipsa->ipsa_linklock = &bucket->isaf_lock;
 249 
 250         return (0);
 251 }
 252 #undef SA_UNIQUE_MATCH
 253 
 254 /*
 255  * Free a security association.  Its reference count is 0, which means
 256  * I must free it.  The SA must be unlocked and must not be linked into
 257  * any fanout list.
 258  */
 259 static void
 260 sadb_freeassoc(ipsa_t *ipsa)
 261 {
 262         ipsec_stack_t   *ipss = ipsa->ipsa_netstack->netstack_ipsec;
 263         mblk_t          *asyncmp, *mp;
 264 
 265         ASSERT(ipss != NULL);
 266         ASSERT(MUTEX_NOT_HELD(&ipsa->ipsa_lock));
 267         ASSERT(ipsa->ipsa_refcnt == 0);
 268         ASSERT(ipsa->ipsa_next == NULL);
 269         ASSERT(ipsa->ipsa_ptpn == NULL);
 270 
 271 
 272         asyncmp = sadb_clear_lpkt(ipsa);
 273         if (asyncmp != NULL) {
 274                 mp = ip_recv_attr_free_mblk(asyncmp);
 275                 ip_drop_packet(mp, B_TRUE, NULL,
 276                     DROPPER(ipss, ipds_sadb_inlarval_timeout),
 277                     &ipss->ipsec_sadb_dropper);
 278         }
 279         mutex_enter(&ipsa->ipsa_lock);
 280 
 281         if (ipsa->ipsa_tsl != NULL) {
 282                 label_rele(ipsa->ipsa_tsl);
 283                 ipsa->ipsa_tsl = NULL;
 284         }
 285 
 286         if (ipsa->ipsa_otsl != NULL) {
 287                 label_rele(ipsa->ipsa_otsl);
 288                 ipsa->ipsa_otsl = NULL;
 289         }
 290 
 291         ipsec_destroy_ctx_tmpl(ipsa, IPSEC_ALG_AUTH);
 292         ipsec_destroy_ctx_tmpl(ipsa, IPSEC_ALG_ENCR);
 293         mutex_exit(&ipsa->ipsa_lock);
 294 
 295         /* bzero() these fields for paranoia's sake. */
 296         if (ipsa->ipsa_authkey != NULL) {
 297                 bzero(ipsa->ipsa_authkey, ipsa->ipsa_authkeylen);
 298                 kmem_free(ipsa->ipsa_authkey, ipsa->ipsa_authkeylen);
 299         }
 300         if (ipsa->ipsa_encrkey != NULL) {
 301                 bzero(ipsa->ipsa_encrkey, ipsa->ipsa_encrkeylen);
 302                 kmem_free(ipsa->ipsa_encrkey, ipsa->ipsa_encrkeylen);
 303         }
 304         if (ipsa->ipsa_nonce_buf != NULL) {
 305                 bzero(ipsa->ipsa_nonce_buf, sizeof (ipsec_nonce_t));
 306                 kmem_free(ipsa->ipsa_nonce_buf, sizeof (ipsec_nonce_t));
 307         }
 308         if (ipsa->ipsa_src_cid != NULL) {
 309                 IPSID_REFRELE(ipsa->ipsa_src_cid);
 310         }
 311         if (ipsa->ipsa_dst_cid != NULL) {
 312                 IPSID_REFRELE(ipsa->ipsa_dst_cid);
 313         }
 314         if (ipsa->ipsa_emech.cm_param != NULL)
 315                 kmem_free(ipsa->ipsa_emech.cm_param,
 316                     ipsa->ipsa_emech.cm_param_len);
 317 
 318         mutex_destroy(&ipsa->ipsa_lock);
 319         kmem_free(ipsa, sizeof (*ipsa));
 320 }
 321 
 322 /*
 323  * Unlink a security association from a hash bucket.  Assume the hash bucket
 324  * lock is held, but the association's lock is not.
 325  *
 326  * Note that we do not bump the bucket's generation number here because
 327  * we might not be making a visible change to the set of visible SA's.
 328  * All callers MUST bump the bucket's generation number before they unlock
 329  * the bucket if they use sadb_unlinkassoc to permanetly remove an SA which
 330  * was present in the bucket at the time it was locked.
 331  */
 332 void
 333 sadb_unlinkassoc(ipsa_t *ipsa)
 334 {
 335         ASSERT(ipsa->ipsa_linklock != NULL);
 336         ASSERT(MUTEX_HELD(ipsa->ipsa_linklock));
 337 
 338         /* These fields are protected by the link lock. */
 339         *(ipsa->ipsa_ptpn) = ipsa->ipsa_next;
 340         if (ipsa->ipsa_next != NULL) {
 341                 ipsa->ipsa_next->ipsa_ptpn = ipsa->ipsa_ptpn;
 342                 ipsa->ipsa_next = NULL;
 343         }
 344 
 345         ipsa->ipsa_ptpn = NULL;
 346 
 347         /* This may destroy the SA. */
 348         IPSA_REFRELE(ipsa);
 349 }
 350 
 351 void
 352 sadb_delete_cluster(ipsa_t *assoc)
 353 {
 354         uint8_t protocol;
 355 
 356         if (cl_inet_deletespi &&
 357             ((assoc->ipsa_state == IPSA_STATE_LARVAL) ||
 358             (assoc->ipsa_state == IPSA_STATE_MATURE))) {
 359                 protocol = (assoc->ipsa_type == SADB_SATYPE_AH) ?
 360                     IPPROTO_AH : IPPROTO_ESP;
 361                 cl_inet_deletespi(assoc->ipsa_netstack->netstack_stackid,
 362                     protocol, assoc->ipsa_spi, NULL);
 363         }
 364 }
 365 
 366 /*
 367  * Create a larval security association with the specified SPI.  All other
 368  * fields are zeroed.
 369  */
 370 static ipsa_t *
 371 sadb_makelarvalassoc(uint32_t spi, uint32_t *src, uint32_t *dst, int addrfam,
 372     netstack_t *ns)
 373 {
 374         ipsa_t *newbie;
 375 
 376         /*
 377          * Allocate...
 378          */
 379 
 380         newbie = (ipsa_t *)kmem_zalloc(sizeof (ipsa_t), KM_NOSLEEP);
 381         if (newbie == NULL) {
 382                 /* Can't make new larval SA. */
 383                 return (NULL);
 384         }
 385 
 386         /* Assigned requested SPI, assume caller does SPI allocation magic. */
 387         newbie->ipsa_spi = spi;
 388         newbie->ipsa_netstack = ns;  /* No netstack_hold */
 389 
 390         /*
 391          * Copy addresses...
 392          */
 393 
 394         IPSA_COPY_ADDR(newbie->ipsa_srcaddr, src, addrfam);
 395         IPSA_COPY_ADDR(newbie->ipsa_dstaddr, dst, addrfam);
 396 
 397         newbie->ipsa_addrfam = addrfam;
 398 
 399         /*
 400          * Set common initialization values, including refcnt.
 401          */
 402         mutex_init(&newbie->ipsa_lock, NULL, MUTEX_DEFAULT, NULL);
 403         newbie->ipsa_state = IPSA_STATE_LARVAL;
 404         newbie->ipsa_refcnt = 1;
 405         newbie->ipsa_freefunc = sadb_freeassoc;
 406 
 407         /*
 408          * There aren't a lot of other common initialization values, as
 409          * they are copied in from the PF_KEY message.
 410          */
 411 
 412         return (newbie);
 413 }
 414 
 415 /*
 416  * Call me to initialize a security association fanout.
 417  */
 418 static int
 419 sadb_init_fanout(isaf_t **tablep, uint_t size, int kmflag)
 420 {
 421         isaf_t *table;
 422         int i;
 423 
 424         table = (isaf_t *)kmem_alloc(size * sizeof (*table), kmflag);
 425         *tablep = table;
 426 
 427         if (table == NULL)
 428                 return (ENOMEM);
 429 
 430         for (i = 0; i < size; i++) {
 431                 mutex_init(&(table[i].isaf_lock), NULL, MUTEX_DEFAULT, NULL);
 432                 table[i].isaf_ipsa = NULL;
 433                 table[i].isaf_gen = 0;
 434         }
 435 
 436         return (0);
 437 }
 438 
 439 /*
 440  * Call me to initialize an acquire fanout
 441  */
 442 static int
 443 sadb_init_acfanout(iacqf_t **tablep, uint_t size, int kmflag)
 444 {
 445         iacqf_t *table;
 446         int i;
 447 
 448         table = (iacqf_t *)kmem_alloc(size * sizeof (*table), kmflag);
 449         *tablep = table;
 450 
 451         if (table == NULL)
 452                 return (ENOMEM);
 453 
 454         for (i = 0; i < size; i++) {
 455                 mutex_init(&(table[i].iacqf_lock), NULL, MUTEX_DEFAULT, NULL);
 456                 table[i].iacqf_ipsacq = NULL;
 457         }
 458 
 459         return (0);
 460 }
 461 
 462 /*
 463  * Attempt to initialize an SADB instance.  On failure, return ENOMEM;
 464  * caller must clean up partial allocations.
 465  */
 466 static int
 467 sadb_init_trial(sadb_t *sp, uint_t size, int kmflag)
 468 {
 469         ASSERT(sp->sdb_of == NULL);
 470         ASSERT(sp->sdb_if == NULL);
 471         ASSERT(sp->sdb_acq == NULL);
 472 
 473         sp->sdb_hashsize = size;
 474         if (sadb_init_fanout(&sp->sdb_of, size, kmflag) != 0)
 475                 return (ENOMEM);
 476         if (sadb_init_fanout(&sp->sdb_if, size, kmflag) != 0)
 477                 return (ENOMEM);
 478         if (sadb_init_acfanout(&sp->sdb_acq, size, kmflag) != 0)
 479                 return (ENOMEM);
 480 
 481         return (0);
 482 }
 483 
 484 /*
 485  * Call me to initialize an SADB instance; fall back to default size on failure.
 486  */
 487 static void
 488 sadb_init(const char *name, sadb_t *sp, uint_t size, uint_t ver,
 489     netstack_t *ns)
 490 {
 491         ASSERT(sp->sdb_of == NULL);
 492         ASSERT(sp->sdb_if == NULL);
 493         ASSERT(sp->sdb_acq == NULL);
 494 
 495         if (size < IPSEC_DEFAULT_HASH_SIZE)
 496                 size = IPSEC_DEFAULT_HASH_SIZE;
 497 
 498         if (sadb_init_trial(sp, size, KM_NOSLEEP) != 0) {
 499 
 500                 cmn_err(CE_WARN,
 501                     "Unable to allocate %u entry IPv%u %s SADB hash table",
 502                     size, ver, name);
 503 
 504                 sadb_destroy(sp, ns);
 505                 size = IPSEC_DEFAULT_HASH_SIZE;
 506                 cmn_err(CE_WARN, "Falling back to %d entries", size);
 507                 (void) sadb_init_trial(sp, size, KM_SLEEP);
 508         }
 509 }
 510 
 511 
 512 /*
 513  * Initialize an SADB-pair.
 514  */
 515 void
 516 sadbp_init(const char *name, sadbp_t *sp, int type, int size, netstack_t *ns)
 517 {
 518         sadb_init(name, &sp->s_v4, size, 4, ns);
 519         sadb_init(name, &sp->s_v6, size, 6, ns);
 520 
 521         sp->s_satype = type;
 522 
 523         ASSERT((type == SADB_SATYPE_AH) || (type == SADB_SATYPE_ESP));
 524         if (type == SADB_SATYPE_AH) {
 525                 ipsec_stack_t   *ipss = ns->netstack_ipsec;
 526 
 527                 ip_drop_register(&ipss->ipsec_sadb_dropper, "IPsec SADB");
 528                 sp->s_addflags = AH_ADD_SETTABLE_FLAGS;
 529                 sp->s_updateflags = AH_UPDATE_SETTABLE_FLAGS;
 530         } else {
 531                 sp->s_addflags = ESP_ADD_SETTABLE_FLAGS;
 532                 sp->s_updateflags = ESP_UPDATE_SETTABLE_FLAGS;
 533         }
 534 }
 535 
 536 /*
 537  * Deliver a single SADB_DUMP message representing a single SA.  This is
 538  * called many times by sadb_dump().
 539  *
 540  * If the return value of this is ENOBUFS (not the same as ENOMEM), then
 541  * the caller should take that as a hint that dupb() on the "original answer"
 542  * failed, and that perhaps the caller should try again with a copyb()ed
 543  * "original answer".
 544  */
 545 static int
 546 sadb_dump_deliver(queue_t *pfkey_q, mblk_t *original_answer, ipsa_t *ipsa,
 547     sadb_msg_t *samsg)
 548 {
 549         mblk_t *answer;
 550 
 551         answer = dupb(original_answer);
 552         if (answer == NULL)
 553                 return (ENOBUFS);
 554         answer->b_cont = sadb_sa2msg(ipsa, samsg);
 555         if (answer->b_cont == NULL) {
 556                 freeb(answer);
 557                 return (ENOMEM);
 558         }
 559 
 560         /* Just do a putnext, and let keysock deal with flow control. */
 561         putnext(pfkey_q, answer);
 562         return (0);
 563 }
 564 
 565 /*
 566  * Common function to allocate and prepare a keysock_out_t M_CTL message.
 567  */
 568 mblk_t *
 569 sadb_keysock_out(minor_t serial)
 570 {
 571         mblk_t *mp;
 572         keysock_out_t *kso;
 573 
 574         mp = allocb(sizeof (ipsec_info_t), BPRI_HI);
 575         if (mp != NULL) {
 576                 mp->b_datap->db_type = M_CTL;
 577                 mp->b_wptr += sizeof (ipsec_info_t);
 578                 kso = (keysock_out_t *)mp->b_rptr;
 579                 kso->ks_out_type = KEYSOCK_OUT;
 580                 kso->ks_out_len = sizeof (*kso);
 581                 kso->ks_out_serial = serial;
 582         }
 583 
 584         return (mp);
 585 }
 586 
 587 /*
 588  * Perform an SADB_DUMP, spewing out every SA in an array of SA fanouts
 589  * to keysock.
 590  */
 591 static int
 592 sadb_dump_fanout(queue_t *pfkey_q, mblk_t *mp, minor_t serial, isaf_t *fanout,
 593     int num_entries, boolean_t do_peers, time_t active_time)
 594 {
 595         int i, error = 0;
 596         mblk_t *original_answer;
 597         ipsa_t *walker;
 598         sadb_msg_t *samsg;
 599         time_t  current;
 600 
 601         /*
 602          * For each IPSA hash bucket do:
 603          *      - Hold the mutex
 604          *      - Walk each entry, doing an sadb_dump_deliver() on it.
 605          */
 606         ASSERT(mp->b_cont != NULL);
 607         samsg = (sadb_msg_t *)mp->b_cont->b_rptr;
 608 
 609         original_answer = sadb_keysock_out(serial);
 610         if (original_answer == NULL)
 611                 return (ENOMEM);
 612 
 613         current = gethrestime_sec();
 614         for (i = 0; i < num_entries; i++) {
 615                 mutex_enter(&fanout[i].isaf_lock);
 616                 for (walker = fanout[i].isaf_ipsa; walker != NULL;
 617                     walker = walker->ipsa_next) {
 618                         if (!do_peers && walker->ipsa_haspeer)
 619                                 continue;
 620                         if ((active_time != 0) &&
 621                             ((current - walker->ipsa_lastuse) > active_time))
 622                                 continue;
 623                         error = sadb_dump_deliver(pfkey_q, original_answer,
 624                             walker, samsg);
 625                         if (error == ENOBUFS) {
 626                                 mblk_t *new_original_answer;
 627 
 628                                 /* Ran out of dupb's.  Try a copyb. */
 629                                 new_original_answer = copyb(original_answer);
 630                                 if (new_original_answer == NULL) {
 631                                         error = ENOMEM;
 632                                 } else {
 633                                         freeb(original_answer);
 634                                         original_answer = new_original_answer;
 635                                         error = sadb_dump_deliver(pfkey_q,
 636                                             original_answer, walker, samsg);
 637                                 }
 638                         }
 639                         if (error != 0)
 640                                 break;  /* out of for loop. */
 641                 }
 642                 mutex_exit(&fanout[i].isaf_lock);
 643                 if (error != 0)
 644                         break;  /* out of for loop. */
 645         }
 646 
 647         freeb(original_answer);
 648         return (error);
 649 }
 650 
 651 /*
 652  * Dump an entire SADB; outbound first, then inbound.
 653  */
 654 
 655 int
 656 sadb_dump(queue_t *pfkey_q, mblk_t *mp, keysock_in_t *ksi, sadb_t *sp)
 657 {
 658         int error;
 659         time_t  active_time = 0;
 660         sadb_x_edump_t  *edump =
 661             (sadb_x_edump_t *)ksi->ks_in_extv[SADB_X_EXT_EDUMP];
 662 
 663         if (edump != NULL) {
 664                 active_time = edump->sadb_x_edump_timeout;
 665         }
 666 
 667         /* Dump outbound */
 668         error = sadb_dump_fanout(pfkey_q, mp, ksi->ks_in_serial, sp->sdb_of,
 669             sp->sdb_hashsize, B_TRUE, active_time);
 670         if (error)
 671                 return (error);
 672 
 673         /* Dump inbound */
 674         return sadb_dump_fanout(pfkey_q, mp, ksi->ks_in_serial, sp->sdb_if,
 675             sp->sdb_hashsize, B_FALSE, active_time);
 676 }
 677 
 678 /*
 679  * Generic sadb table walker.
 680  *
 681  * Call "walkfn" for each SA in each bucket in "table"; pass the
 682  * bucket, the entry and "cookie" to the callback function.
 683  * Take care to ensure that walkfn can delete the SA without screwing
 684  * up our traverse.
 685  *
 686  * The bucket is locked for the duration of the callback, both so that the
 687  * callback can just call sadb_unlinkassoc() when it wants to delete something,
 688  * and so that no new entries are added while we're walking the list.
 689  */
 690 static void
 691 sadb_walker(isaf_t *table, uint_t numentries,
 692     void (*walkfn)(isaf_t *head, ipsa_t *entry, void *cookie),
 693     void *cookie)
 694 {
 695         int i;
 696         for (i = 0; i < numentries; i++) {
 697                 ipsa_t *entry, *next;
 698 
 699                 mutex_enter(&table[i].isaf_lock);
 700 
 701                 for (entry = table[i].isaf_ipsa; entry != NULL;
 702                     entry = next) {
 703                         next = entry->ipsa_next;
 704                         (*walkfn)(&table[i], entry, cookie);
 705                 }
 706                 mutex_exit(&table[i].isaf_lock);
 707         }
 708 }
 709 
 710 /*
 711  * Call me to free up a security association fanout.  Use the forever
 712  * variable to indicate freeing up the SAs (forever == B_FALSE, e.g.
 713  * an SADB_FLUSH message), or destroying everything (forever == B_TRUE,
 714  * when a module is unloaded).
 715  */
 716 static void
 717 sadb_destroyer(isaf_t **tablep, uint_t numentries, boolean_t forever,
 718     boolean_t inbound)
 719 {
 720         int i;
 721         isaf_t *table = *tablep;
 722         uint8_t protocol;
 723         ipsa_t *sa;
 724         netstackid_t sid;
 725 
 726         if (table == NULL)
 727                 return;
 728 
 729         for (i = 0; i < numentries; i++) {
 730                 mutex_enter(&table[i].isaf_lock);
 731                 while ((sa = table[i].isaf_ipsa) != NULL) {
 732                         if (inbound && cl_inet_deletespi &&
 733                             (sa->ipsa_state != IPSA_STATE_ACTIVE_ELSEWHERE) &&
 734                             (sa->ipsa_state != IPSA_STATE_IDLE)) {
 735                                 protocol = (sa->ipsa_type == SADB_SATYPE_AH) ?
 736                                     IPPROTO_AH : IPPROTO_ESP;
 737                                 sid = sa->ipsa_netstack->netstack_stackid;
 738                                 cl_inet_deletespi(sid, protocol, sa->ipsa_spi,
 739                                     NULL);
 740                         }
 741                         sadb_unlinkassoc(sa);
 742                 }
 743                 table[i].isaf_gen++;
 744                 mutex_exit(&table[i].isaf_lock);
 745                 if (forever)
 746                         mutex_destroy(&(table[i].isaf_lock));
 747         }
 748 
 749         if (forever) {
 750                 *tablep = NULL;
 751                 kmem_free(table, numentries * sizeof (*table));
 752         }
 753 }
 754 
 755 /*
 756  * Entry points to sadb_destroyer().
 757  */
 758 static void
 759 sadb_flush(sadb_t *sp, netstack_t *ns)
 760 {
 761         /*
 762          * Flush out each bucket, one at a time.  Were it not for keysock's
 763          * enforcement, there would be a subtlety where I could add on the
 764          * heels of a flush.  With keysock's enforcement, however, this
 765          * makes ESP's job easy.
 766          */
 767         sadb_destroyer(&sp->sdb_of, sp->sdb_hashsize, B_FALSE, B_FALSE);
 768         sadb_destroyer(&sp->sdb_if, sp->sdb_hashsize, B_FALSE, B_TRUE);
 769 
 770         /* For each acquire, destroy it; leave the bucket mutex alone. */
 771         sadb_destroy_acqlist(&sp->sdb_acq, sp->sdb_hashsize, B_FALSE, ns);
 772 }
 773 
 774 static void
 775 sadb_destroy(sadb_t *sp, netstack_t *ns)
 776 {
 777         sadb_destroyer(&sp->sdb_of, sp->sdb_hashsize, B_TRUE, B_FALSE);
 778         sadb_destroyer(&sp->sdb_if, sp->sdb_hashsize, B_TRUE, B_TRUE);
 779 
 780         /* For each acquire, destroy it, including the bucket mutex. */
 781         sadb_destroy_acqlist(&sp->sdb_acq, sp->sdb_hashsize, B_TRUE, ns);
 782 
 783         ASSERT(sp->sdb_of == NULL);
 784         ASSERT(sp->sdb_if == NULL);
 785         ASSERT(sp->sdb_acq == NULL);
 786 }
 787 
 788 void
 789 sadbp_flush(sadbp_t *spp, netstack_t *ns)
 790 {
 791         sadb_flush(&spp->s_v4, ns);
 792         sadb_flush(&spp->s_v6, ns);
 793 }
 794 
 795 void
 796 sadbp_destroy(sadbp_t *spp, netstack_t *ns)
 797 {
 798         sadb_destroy(&spp->s_v4, ns);
 799         sadb_destroy(&spp->s_v6, ns);
 800 
 801         if (spp->s_satype == SADB_SATYPE_AH) {
 802                 ipsec_stack_t   *ipss = ns->netstack_ipsec;
 803 
 804                 ip_drop_unregister(&ipss->ipsec_sadb_dropper);
 805         }
 806 }
 807 
 808 
 809 /*
 810  * Check hard vs. soft lifetimes.  If there's a reality mismatch (e.g.
 811  * soft lifetimes > hard lifetimes) return an appropriate diagnostic for
 812  * EINVAL.
 813  */
 814 int
 815 sadb_hardsoftchk(sadb_lifetime_t *hard, sadb_lifetime_t *soft,
 816     sadb_lifetime_t *idle)
 817 {
 818         if (hard == NULL || soft == NULL)
 819                 return (0);
 820 
 821         if (hard->sadb_lifetime_allocations != 0 &&
 822             soft->sadb_lifetime_allocations != 0 &&
 823             hard->sadb_lifetime_allocations < soft->sadb_lifetime_allocations)
 824                 return (SADB_X_DIAGNOSTIC_ALLOC_HSERR);
 825 
 826         if (hard->sadb_lifetime_bytes != 0 &&
 827             soft->sadb_lifetime_bytes != 0 &&
 828             hard->sadb_lifetime_bytes < soft->sadb_lifetime_bytes)
 829                 return (SADB_X_DIAGNOSTIC_BYTES_HSERR);
 830 
 831         if (hard->sadb_lifetime_addtime != 0 &&
 832             soft->sadb_lifetime_addtime != 0 &&
 833             hard->sadb_lifetime_addtime < soft->sadb_lifetime_addtime)
 834                 return (SADB_X_DIAGNOSTIC_ADDTIME_HSERR);
 835 
 836         if (hard->sadb_lifetime_usetime != 0 &&
 837             soft->sadb_lifetime_usetime != 0 &&
 838             hard->sadb_lifetime_usetime < soft->sadb_lifetime_usetime)
 839                 return (SADB_X_DIAGNOSTIC_USETIME_HSERR);
 840 
 841         if (idle != NULL) {
 842                 if (hard->sadb_lifetime_addtime != 0 &&
 843                     idle->sadb_lifetime_addtime != 0 &&
 844                     hard->sadb_lifetime_addtime < idle->sadb_lifetime_addtime)
 845                         return (SADB_X_DIAGNOSTIC_ADDTIME_HSERR);
 846 
 847                 if (soft->sadb_lifetime_addtime != 0 &&
 848                     idle->sadb_lifetime_addtime != 0 &&
 849                     soft->sadb_lifetime_addtime < idle->sadb_lifetime_addtime)
 850                         return (SADB_X_DIAGNOSTIC_ADDTIME_HSERR);
 851 
 852                 if (hard->sadb_lifetime_usetime != 0 &&
 853                     idle->sadb_lifetime_usetime != 0 &&
 854                     hard->sadb_lifetime_usetime < idle->sadb_lifetime_usetime)
 855                         return (SADB_X_DIAGNOSTIC_USETIME_HSERR);
 856 
 857                 if (soft->sadb_lifetime_usetime != 0 &&
 858                     idle->sadb_lifetime_usetime != 0 &&
 859                     soft->sadb_lifetime_usetime < idle->sadb_lifetime_usetime)
 860                         return (SADB_X_DIAGNOSTIC_USETIME_HSERR);
 861         }
 862 
 863         return (0);
 864 }
 865 
 866 /*
 867  * Sanity check sensitivity labels.
 868  *
 869  * For now, just reject labels on unlabeled systems.
 870  */
 871 int
 872 sadb_labelchk(keysock_in_t *ksi)
 873 {
 874         if (!is_system_labeled()) {
 875                 if (ksi->ks_in_extv[SADB_EXT_SENSITIVITY] != NULL)
 876                         return (SADB_X_DIAGNOSTIC_BAD_LABEL);
 877 
 878                 if (ksi->ks_in_extv[SADB_X_EXT_OUTER_SENS] != NULL)
 879                         return (SADB_X_DIAGNOSTIC_BAD_LABEL);
 880         }
 881 
 882         return (0);
 883 }
 884 
 885 /*
 886  * Clone a security association for the purposes of inserting a single SA
 887  * into inbound and outbound tables respectively. This function should only
 888  * be called from sadb_common_add().
 889  */
 890 static ipsa_t *
 891 sadb_cloneassoc(ipsa_t *ipsa)
 892 {
 893         ipsa_t *newbie;
 894         boolean_t error = B_FALSE;
 895 
 896         ASSERT(MUTEX_NOT_HELD(&(ipsa->ipsa_lock)));
 897 
 898         newbie = kmem_alloc(sizeof (ipsa_t), KM_NOSLEEP);
 899         if (newbie == NULL)
 900                 return (NULL);
 901 
 902         /* Copy over what we can. */
 903         *newbie = *ipsa;
 904 
 905         /* bzero and initialize locks, in case *_init() allocates... */
 906         mutex_init(&newbie->ipsa_lock, NULL, MUTEX_DEFAULT, NULL);
 907 
 908         if (newbie->ipsa_tsl != NULL)
 909                 label_hold(newbie->ipsa_tsl);
 910 
 911         if (newbie->ipsa_otsl != NULL)
 912                 label_hold(newbie->ipsa_otsl);
 913 
 914         /*
 915          * While somewhat dain-bramaged, the most graceful way to
 916          * recover from errors is to keep plowing through the
 917          * allocations, and getting what I can.  It's easier to call
 918          * sadb_freeassoc() on the stillborn clone when all the
 919          * pointers aren't pointing to the parent's data.
 920          */
 921 
 922         if (ipsa->ipsa_authkey != NULL) {
 923                 newbie->ipsa_authkey = kmem_alloc(newbie->ipsa_authkeylen,
 924                     KM_NOSLEEP);
 925                 if (newbie->ipsa_authkey == NULL) {
 926                         error = B_TRUE;
 927                 } else {
 928                         bcopy(ipsa->ipsa_authkey, newbie->ipsa_authkey,
 929                             newbie->ipsa_authkeylen);
 930 
 931                         newbie->ipsa_kcfauthkey.ck_data =
 932                             newbie->ipsa_authkey;
 933                 }
 934 
 935                 if (newbie->ipsa_amech.cm_param != NULL) {
 936                         newbie->ipsa_amech.cm_param =
 937                             (char *)&newbie->ipsa_mac_len;
 938                 }
 939         }
 940 
 941         if (ipsa->ipsa_encrkey != NULL) {
 942                 newbie->ipsa_encrkey = kmem_alloc(newbie->ipsa_encrkeylen,
 943                     KM_NOSLEEP);
 944                 if (newbie->ipsa_encrkey == NULL) {
 945                         error = B_TRUE;
 946                 } else {
 947                         bcopy(ipsa->ipsa_encrkey, newbie->ipsa_encrkey,
 948                             newbie->ipsa_encrkeylen);
 949 
 950                         newbie->ipsa_kcfencrkey.ck_data =
 951                             newbie->ipsa_encrkey;
 952                 }
 953         }
 954 
 955         newbie->ipsa_authtmpl = NULL;
 956         newbie->ipsa_encrtmpl = NULL;
 957         newbie->ipsa_haspeer = B_TRUE;
 958 
 959         if (ipsa->ipsa_src_cid != NULL) {
 960                 newbie->ipsa_src_cid = ipsa->ipsa_src_cid;
 961                 IPSID_REFHOLD(ipsa->ipsa_src_cid);
 962         }
 963 
 964         if (ipsa->ipsa_dst_cid != NULL) {
 965                 newbie->ipsa_dst_cid = ipsa->ipsa_dst_cid;
 966                 IPSID_REFHOLD(ipsa->ipsa_dst_cid);
 967         }
 968 
 969         if (error) {
 970                 sadb_freeassoc(newbie);
 971                 return (NULL);
 972         }
 973 
 974         return (newbie);
 975 }
 976 
 977 /*
 978  * Initialize a SADB address extension at the address specified by addrext.
 979  * Return a pointer to the end of the new address extension.
 980  */
 981 static uint8_t *
 982 sadb_make_addr_ext(uint8_t *start, uint8_t *end, uint16_t exttype,
 983     sa_family_t af, uint32_t *addr, uint16_t port, uint8_t proto, int prefix)
 984 {
 985         struct sockaddr_in *sin;
 986         struct sockaddr_in6 *sin6;
 987         uint8_t *cur = start;
 988         int addrext_len;
 989         int sin_len;
 990         sadb_address_t *addrext = (sadb_address_t *)cur;
 991 
 992         if (cur == NULL)
 993                 return (NULL);
 994 
 995         cur += sizeof (*addrext);
 996         if (cur > end)
 997                 return (NULL);
 998 
 999         addrext->sadb_address_proto = proto;
1000         addrext->sadb_address_prefixlen = prefix;
1001         addrext->sadb_address_reserved = 0;
1002         addrext->sadb_address_exttype = exttype;
1003 
1004         switch (af) {
1005         case AF_INET:
1006                 sin = (struct sockaddr_in *)cur;
1007                 sin_len = sizeof (*sin);
1008                 cur += sin_len;
1009                 if (cur > end)
1010                         return (NULL);
1011 
1012                 sin->sin_family = af;
1013                 bzero(sin->sin_zero, sizeof (sin->sin_zero));
1014                 sin->sin_port = port;
1015                 IPSA_COPY_ADDR(&sin->sin_addr, addr, af);
1016                 break;
1017         case AF_INET6:
1018                 sin6 = (struct sockaddr_in6 *)cur;
1019                 sin_len = sizeof (*sin6);
1020                 cur += sin_len;
1021                 if (cur > end)
1022                         return (NULL);
1023 
1024                 bzero(sin6, sizeof (*sin6));
1025                 sin6->sin6_family = af;
1026                 sin6->sin6_port = port;
1027                 IPSA_COPY_ADDR(&sin6->sin6_addr, addr, af);
1028                 break;
1029         }
1030 
1031         addrext_len = roundup(cur - start, sizeof (uint64_t));
1032         addrext->sadb_address_len = SADB_8TO64(addrext_len);
1033 
1034         cur = start + addrext_len;
1035         if (cur > end)
1036                 cur = NULL;
1037 
1038         return (cur);
1039 }
1040 
1041 /*
1042  * Construct a key management cookie extension.
1043  */
1044 
1045 static uint8_t *
1046 sadb_make_kmc_ext(uint8_t *cur, uint8_t *end, uint32_t kmp, uint64_t kmc)
1047 {
1048         sadb_x_kmc_t *kmcext = (sadb_x_kmc_t *)cur;
1049 
1050         if (cur == NULL)
1051                 return (NULL);
1052 
1053         cur += sizeof (*kmcext);
1054 
1055         if (cur > end)
1056                 return (NULL);
1057 
1058         kmcext->sadb_x_kmc_len = SADB_8TO64(sizeof (*kmcext));
1059         kmcext->sadb_x_kmc_exttype = SADB_X_EXT_KM_COOKIE;
1060         kmcext->sadb_x_kmc_proto = kmp;
1061         kmcext->sadb_x_kmc_cookie64 = kmc;
1062 
1063         return (cur);
1064 }
1065 
1066 /*
1067  * Given an original message header with sufficient space following it, and an
1068  * SA, construct a full PF_KEY message with all of the relevant extensions.
1069  * This is mostly used for SADB_GET, and SADB_DUMP.
1070  */
1071 static mblk_t *
1072 sadb_sa2msg(ipsa_t *ipsa, sadb_msg_t *samsg)
1073 {
1074         int alloclen, addrsize, paddrsize, authsize, encrsize;
1075         int srcidsize, dstidsize, senslen, osenslen;
1076         sa_family_t fam, pfam;  /* Address family for SADB_EXT_ADDRESS */
1077                                 /* src/dst and proxy sockaddrs. */
1078         /*
1079          * The following are pointers into the PF_KEY message this PF_KEY
1080          * message creates.
1081          */
1082         sadb_msg_t *newsamsg;
1083         sadb_sa_t *assoc;
1084         sadb_lifetime_t *lt;
1085         sadb_key_t *key;
1086         sadb_ident_t *ident;
1087         sadb_sens_t *sens;
1088         sadb_ext_t *walker;     /* For when we need a generic ext. pointer. */
1089         sadb_x_replay_ctr_t *repl_ctr;
1090         sadb_x_pair_t *pair_ext;
1091 
1092         mblk_t *mp;
1093         uint8_t *cur, *end;
1094         /* These indicate the presence of the above extension fields. */
1095         boolean_t soft = B_FALSE, hard = B_FALSE;
1096         boolean_t isrc = B_FALSE, idst = B_FALSE;
1097         boolean_t auth = B_FALSE, encr = B_FALSE;
1098         boolean_t sensinteg = B_FALSE, osensinteg = B_FALSE;
1099         boolean_t srcid = B_FALSE, dstid = B_FALSE;
1100         boolean_t idle;
1101         boolean_t paired;
1102         uint32_t otherspi;
1103 
1104         /* First off, figure out the allocation length for this message. */
1105         /*
1106          * Constant stuff.  This includes base, SA, address (src, dst),
1107          * and lifetime (current).
1108          */
1109         alloclen = sizeof (sadb_msg_t) + sizeof (sadb_sa_t) +
1110             sizeof (sadb_lifetime_t);
1111 
1112         fam = ipsa->ipsa_addrfam;
1113         switch (fam) {
1114         case AF_INET:
1115                 addrsize = roundup(sizeof (struct sockaddr_in) +
1116                     sizeof (sadb_address_t), sizeof (uint64_t));
1117                 break;
1118         case AF_INET6:
1119                 addrsize = roundup(sizeof (struct sockaddr_in6) +
1120                     sizeof (sadb_address_t), sizeof (uint64_t));
1121                 break;
1122         default:
1123                 return (NULL);
1124         }
1125         /*
1126          * Allocate TWO address extensions, for source and destination.
1127          * (Thus, the * 2.)
1128          */
1129         alloclen += addrsize * 2;
1130         if (ipsa->ipsa_flags & IPSA_F_NATT_REM)
1131                 alloclen += addrsize;
1132         if (ipsa->ipsa_flags & IPSA_F_NATT_LOC)
1133                 alloclen += addrsize;
1134 
1135         if (ipsa->ipsa_flags & IPSA_F_PAIRED) {
1136                 paired = B_TRUE;
1137                 alloclen += sizeof (sadb_x_pair_t);
1138                 otherspi = ipsa->ipsa_otherspi;
1139         } else {
1140                 paired = B_FALSE;
1141         }
1142 
1143         /* How 'bout other lifetimes? */
1144         if (ipsa->ipsa_softaddlt != 0 || ipsa->ipsa_softuselt != 0 ||
1145             ipsa->ipsa_softbyteslt != 0 || ipsa->ipsa_softalloc != 0) {
1146                 alloclen += sizeof (sadb_lifetime_t);
1147                 soft = B_TRUE;
1148         }
1149 
1150         if (ipsa->ipsa_hardaddlt != 0 || ipsa->ipsa_harduselt != 0 ||
1151             ipsa->ipsa_hardbyteslt != 0 || ipsa->ipsa_hardalloc != 0) {
1152                 alloclen += sizeof (sadb_lifetime_t);
1153                 hard = B_TRUE;
1154         }
1155 
1156         if (ipsa->ipsa_idleaddlt != 0 || ipsa->ipsa_idleuselt != 0) {
1157                 alloclen += sizeof (sadb_lifetime_t);
1158                 idle = B_TRUE;
1159         } else {
1160                 idle = B_FALSE;
1161         }
1162 
1163         /* Inner addresses. */
1164         if (ipsa->ipsa_innerfam != 0) {
1165                 pfam = ipsa->ipsa_innerfam;
1166                 switch (pfam) {
1167                 case AF_INET6:
1168                         paddrsize = roundup(sizeof (struct sockaddr_in6) +
1169                             sizeof (sadb_address_t), sizeof (uint64_t));
1170                         break;
1171                 case AF_INET:
1172                         paddrsize = roundup(sizeof (struct sockaddr_in) +
1173                             sizeof (sadb_address_t), sizeof (uint64_t));
1174                         break;
1175                 default:
1176                         cmn_err(CE_PANIC,
1177                             "IPsec SADB: Proxy length failure.\n");
1178                         break;
1179                 }
1180                 isrc = B_TRUE;
1181                 idst = B_TRUE;
1182                 alloclen += 2 * paddrsize;
1183         }
1184 
1185         /* For the following fields, assume that length != 0 ==> stuff */
1186         if (ipsa->ipsa_authkeylen != 0) {
1187                 authsize = roundup(sizeof (sadb_key_t) + ipsa->ipsa_authkeylen,
1188                     sizeof (uint64_t));
1189                 alloclen += authsize;
1190                 auth = B_TRUE;
1191         }
1192 
1193         if (ipsa->ipsa_encrkeylen != 0) {
1194                 encrsize = roundup(sizeof (sadb_key_t) + ipsa->ipsa_encrkeylen +
1195                     ipsa->ipsa_nonce_len, sizeof (uint64_t));
1196                 alloclen += encrsize;
1197                 encr = B_TRUE;
1198         } else {
1199                 encr = B_FALSE;
1200         }
1201 
1202         if (ipsa->ipsa_tsl != NULL) {
1203                 senslen = sadb_sens_len_from_label(ipsa->ipsa_tsl);
1204                 alloclen += senslen;
1205                 sensinteg = B_TRUE;
1206         }
1207 
1208         if (ipsa->ipsa_otsl != NULL) {
1209                 osenslen = sadb_sens_len_from_label(ipsa->ipsa_otsl);
1210                 alloclen += osenslen;
1211                 osensinteg = B_TRUE;
1212         }
1213 
1214         /*
1215          * Must use strlen() here for lengths.  Identities use NULL
1216          * pointers to indicate their nonexistence.
1217          */
1218         if (ipsa->ipsa_src_cid != NULL) {
1219                 srcidsize = roundup(sizeof (sadb_ident_t) +
1220                     strlen(ipsa->ipsa_src_cid->ipsid_cid) + 1,
1221                     sizeof (uint64_t));
1222                 alloclen += srcidsize;
1223                 srcid = B_TRUE;
1224         }
1225 
1226         if (ipsa->ipsa_dst_cid != NULL) {
1227                 dstidsize = roundup(sizeof (sadb_ident_t) +
1228                     strlen(ipsa->ipsa_dst_cid->ipsid_cid) + 1,
1229                     sizeof (uint64_t));
1230                 alloclen += dstidsize;
1231                 dstid = B_TRUE;
1232         }
1233 
1234         if ((ipsa->ipsa_kmp != 0) || (ipsa->ipsa_kmc != 0))
1235                 alloclen += sizeof (sadb_x_kmc_t);
1236 
1237         if (ipsa->ipsa_replay != 0) {
1238                 alloclen += sizeof (sadb_x_replay_ctr_t);
1239         }
1240 
1241         /* Make sure the allocation length is a multiple of 8 bytes. */
1242         ASSERT((alloclen & 0x7) == 0);
1243 
1244         /* XXX Possibly make it esballoc, with a bzero-ing free_ftn. */
1245         mp = allocb(alloclen, BPRI_HI);
1246         if (mp == NULL)
1247                 return (NULL);
1248         bzero(mp->b_rptr, alloclen);
1249 
1250         mp->b_wptr += alloclen;
1251         end = mp->b_wptr;
1252         newsamsg = (sadb_msg_t *)mp->b_rptr;
1253         *newsamsg = *samsg;
1254         newsamsg->sadb_msg_len = (uint16_t)SADB_8TO64(alloclen);
1255 
1256         mutex_enter(&ipsa->ipsa_lock);   /* Since I'm grabbing SA fields... */
1257 
1258         newsamsg->sadb_msg_satype = ipsa->ipsa_type;
1259 
1260         assoc = (sadb_sa_t *)(newsamsg + 1);
1261         assoc->sadb_sa_len = SADB_8TO64(sizeof (*assoc));
1262         assoc->sadb_sa_exttype = SADB_EXT_SA;
1263         assoc->sadb_sa_spi = ipsa->ipsa_spi;
1264         assoc->sadb_sa_replay = ipsa->ipsa_replay_wsize;
1265         assoc->sadb_sa_state = ipsa->ipsa_state;
1266         assoc->sadb_sa_auth = ipsa->ipsa_auth_alg;
1267         assoc->sadb_sa_encrypt = ipsa->ipsa_encr_alg;
1268         assoc->sadb_sa_flags = ipsa->ipsa_flags;
1269 
1270         lt = (sadb_lifetime_t *)(assoc + 1);
1271         lt->sadb_lifetime_len = SADB_8TO64(sizeof (*lt));
1272         lt->sadb_lifetime_exttype = SADB_EXT_LIFETIME_CURRENT;
1273         /* We do not support the concept. */
1274         lt->sadb_lifetime_allocations = 0;
1275         lt->sadb_lifetime_bytes = ipsa->ipsa_bytes;
1276         lt->sadb_lifetime_addtime = ipsa->ipsa_addtime;
1277         lt->sadb_lifetime_usetime = ipsa->ipsa_usetime;
1278 
1279         if (hard) {
1280                 lt++;
1281                 lt->sadb_lifetime_len = SADB_8TO64(sizeof (*lt));
1282                 lt->sadb_lifetime_exttype = SADB_EXT_LIFETIME_HARD;
1283                 lt->sadb_lifetime_allocations = ipsa->ipsa_hardalloc;
1284                 lt->sadb_lifetime_bytes = ipsa->ipsa_hardbyteslt;
1285                 lt->sadb_lifetime_addtime = ipsa->ipsa_hardaddlt;
1286                 lt->sadb_lifetime_usetime = ipsa->ipsa_harduselt;
1287         }
1288 
1289         if (soft) {
1290                 lt++;
1291                 lt->sadb_lifetime_len = SADB_8TO64(sizeof (*lt));
1292                 lt->sadb_lifetime_exttype = SADB_EXT_LIFETIME_SOFT;
1293                 lt->sadb_lifetime_allocations = ipsa->ipsa_softalloc;
1294                 lt->sadb_lifetime_bytes = ipsa->ipsa_softbyteslt;
1295                 lt->sadb_lifetime_addtime = ipsa->ipsa_softaddlt;
1296                 lt->sadb_lifetime_usetime = ipsa->ipsa_softuselt;
1297         }
1298 
1299         if (idle) {
1300                 lt++;
1301                 lt->sadb_lifetime_len = SADB_8TO64(sizeof (*lt));
1302                 lt->sadb_lifetime_exttype = SADB_X_EXT_LIFETIME_IDLE;
1303                 lt->sadb_lifetime_addtime = ipsa->ipsa_idleaddlt;
1304                 lt->sadb_lifetime_usetime = ipsa->ipsa_idleuselt;
1305         }
1306 
1307         cur = (uint8_t *)(lt + 1);
1308 
1309         /* NOTE:  Don't fill in ports here if we are a tunnel-mode SA. */
1310         cur = sadb_make_addr_ext(cur, end, SADB_EXT_ADDRESS_SRC, fam,
1311             ipsa->ipsa_srcaddr, (!isrc && !idst) ? SA_SRCPORT(ipsa) : 0,
1312             SA_PROTO(ipsa), 0);
1313         if (cur == NULL) {
1314                 freemsg(mp);
1315                 mp = NULL;
1316                 goto bail;
1317         }
1318 
1319         cur = sadb_make_addr_ext(cur, end, SADB_EXT_ADDRESS_DST, fam,
1320             ipsa->ipsa_dstaddr, (!isrc && !idst) ? SA_DSTPORT(ipsa) : 0,
1321             SA_PROTO(ipsa), 0);
1322         if (cur == NULL) {
1323                 freemsg(mp);
1324                 mp = NULL;
1325                 goto bail;
1326         }
1327 
1328         if (ipsa->ipsa_flags & IPSA_F_NATT_LOC) {
1329                 cur = sadb_make_addr_ext(cur, end, SADB_X_EXT_ADDRESS_NATT_LOC,
1330                     fam, &ipsa->ipsa_natt_addr_loc, ipsa->ipsa_local_nat_port,
1331                     IPPROTO_UDP, 0);
1332                 if (cur == NULL) {
1333                         freemsg(mp);
1334                         mp = NULL;
1335                         goto bail;
1336                 }
1337         }
1338 
1339         if (ipsa->ipsa_flags & IPSA_F_NATT_REM) {
1340                 cur = sadb_make_addr_ext(cur, end, SADB_X_EXT_ADDRESS_NATT_REM,
1341                     fam, &ipsa->ipsa_natt_addr_rem, ipsa->ipsa_remote_nat_port,
1342                     IPPROTO_UDP, 0);
1343                 if (cur == NULL) {
1344                         freemsg(mp);
1345                         mp = NULL;
1346                         goto bail;
1347                 }
1348         }
1349 
1350         /* If we are a tunnel-mode SA, fill in the inner-selectors. */
1351         if (isrc) {
1352                 cur = sadb_make_addr_ext(cur, end, SADB_X_EXT_ADDRESS_INNER_SRC,
1353                     pfam, ipsa->ipsa_innersrc, SA_SRCPORT(ipsa),
1354                     SA_IPROTO(ipsa), ipsa->ipsa_innersrcpfx);
1355                 if (cur == NULL) {
1356                         freemsg(mp);
1357                         mp = NULL;
1358                         goto bail;
1359                 }
1360         }
1361 
1362         if (idst) {
1363                 cur = sadb_make_addr_ext(cur, end, SADB_X_EXT_ADDRESS_INNER_DST,
1364                     pfam, ipsa->ipsa_innerdst, SA_DSTPORT(ipsa),
1365                     SA_IPROTO(ipsa), ipsa->ipsa_innerdstpfx);
1366                 if (cur == NULL) {
1367                         freemsg(mp);
1368                         mp = NULL;
1369                         goto bail;
1370                 }
1371         }
1372 
1373         if ((ipsa->ipsa_kmp != 0) || (ipsa->ipsa_kmc != 0)) {
1374                 cur = sadb_make_kmc_ext(cur, end,
1375                     ipsa->ipsa_kmp, ipsa->ipsa_kmc);
1376                 if (cur == NULL) {
1377                         freemsg(mp);
1378                         mp = NULL;
1379                         goto bail;
1380                 }
1381         }
1382 
1383         walker = (sadb_ext_t *)cur;
1384         if (auth) {
1385                 key = (sadb_key_t *)walker;
1386                 key->sadb_key_len = SADB_8TO64(authsize);
1387                 key->sadb_key_exttype = SADB_EXT_KEY_AUTH;
1388                 key->sadb_key_bits = ipsa->ipsa_authkeybits;
1389                 key->sadb_key_reserved = 0;
1390                 bcopy(ipsa->ipsa_authkey, key + 1, ipsa->ipsa_authkeylen);
1391                 walker = (sadb_ext_t *)((uint64_t *)walker +
1392                     walker->sadb_ext_len);
1393         }
1394 
1395         if (encr) {
1396                 uint8_t *buf_ptr;
1397                 key = (sadb_key_t *)walker;
1398                 key->sadb_key_len = SADB_8TO64(encrsize);
1399                 key->sadb_key_exttype = SADB_EXT_KEY_ENCRYPT;
1400                 key->sadb_key_bits = ipsa->ipsa_encrkeybits;
1401                 key->sadb_key_reserved = ipsa->ipsa_saltbits;
1402                 buf_ptr = (uint8_t *)(key + 1);
1403                 bcopy(ipsa->ipsa_encrkey, buf_ptr, ipsa->ipsa_encrkeylen);
1404                 if (ipsa->ipsa_salt != NULL) {
1405                         buf_ptr += ipsa->ipsa_encrkeylen;
1406                         bcopy(ipsa->ipsa_salt, buf_ptr, ipsa->ipsa_saltlen);
1407                 }
1408                 walker = (sadb_ext_t *)((uint64_t *)walker +
1409                     walker->sadb_ext_len);
1410         }
1411 
1412         if (srcid) {
1413                 ident = (sadb_ident_t *)walker;
1414                 ident->sadb_ident_len = SADB_8TO64(srcidsize);
1415                 ident->sadb_ident_exttype = SADB_EXT_IDENTITY_SRC;
1416                 ident->sadb_ident_type = ipsa->ipsa_src_cid->ipsid_type;
1417                 ident->sadb_ident_id = 0;
1418                 ident->sadb_ident_reserved = 0;
1419                 (void) strcpy((char *)(ident + 1),
1420                     ipsa->ipsa_src_cid->ipsid_cid);
1421                 walker = (sadb_ext_t *)((uint64_t *)walker +
1422                     walker->sadb_ext_len);
1423         }
1424 
1425         if (dstid) {
1426                 ident = (sadb_ident_t *)walker;
1427                 ident->sadb_ident_len = SADB_8TO64(dstidsize);
1428                 ident->sadb_ident_exttype = SADB_EXT_IDENTITY_DST;
1429                 ident->sadb_ident_type = ipsa->ipsa_dst_cid->ipsid_type;
1430                 ident->sadb_ident_id = 0;
1431                 ident->sadb_ident_reserved = 0;
1432                 (void) strcpy((char *)(ident + 1),
1433                     ipsa->ipsa_dst_cid->ipsid_cid);
1434                 walker = (sadb_ext_t *)((uint64_t *)walker +
1435                     walker->sadb_ext_len);
1436         }
1437 
1438         if (sensinteg) {
1439                 sens = (sadb_sens_t *)walker;
1440                 sadb_sens_from_label(sens, SADB_EXT_SENSITIVITY,
1441                     ipsa->ipsa_tsl, senslen);
1442 
1443                 walker = (sadb_ext_t *)((uint64_t *)walker +
1444                     walker->sadb_ext_len);
1445         }
1446 
1447         if (osensinteg) {
1448                 sens = (sadb_sens_t *)walker;
1449 
1450                 sadb_sens_from_label(sens, SADB_X_EXT_OUTER_SENS,
1451                     ipsa->ipsa_otsl, osenslen);
1452                 if (ipsa->ipsa_mac_exempt)
1453                         sens->sadb_x_sens_flags = SADB_X_SENS_IMPLICIT;
1454 
1455                 walker = (sadb_ext_t *)((uint64_t *)walker +
1456                     walker->sadb_ext_len);
1457         }
1458 
1459         if (paired) {
1460                 pair_ext = (sadb_x_pair_t *)walker;
1461 
1462                 pair_ext->sadb_x_pair_len = SADB_8TO64(sizeof (sadb_x_pair_t));
1463                 pair_ext->sadb_x_pair_exttype = SADB_X_EXT_PAIR;
1464                 pair_ext->sadb_x_pair_spi = otherspi;
1465 
1466                 walker = (sadb_ext_t *)((uint64_t *)walker +
1467                     walker->sadb_ext_len);
1468         }
1469 
1470         if (ipsa->ipsa_replay != 0) {
1471                 repl_ctr = (sadb_x_replay_ctr_t *)walker;
1472                 repl_ctr->sadb_x_rc_len = SADB_8TO64(sizeof (*repl_ctr));
1473                 repl_ctr->sadb_x_rc_exttype = SADB_X_EXT_REPLAY_VALUE;
1474                 repl_ctr->sadb_x_rc_replay32 = ipsa->ipsa_replay;
1475                 repl_ctr->sadb_x_rc_replay64 = 0;
1476                 walker = (sadb_ext_t *)(repl_ctr + 1);
1477         }
1478 
1479 bail:
1480         /* Pardon any delays... */
1481         mutex_exit(&ipsa->ipsa_lock);
1482 
1483         return (mp);
1484 }
1485 
1486 /*
1487  * Strip out key headers or unmarked headers (SADB_EXT_KEY_*, SADB_EXT_UNKNOWN)
1488  * and adjust base message accordingly.
1489  *
1490  * Assume message is pulled up in one piece of contiguous memory.
1491  *
1492  * Say if we start off with:
1493  *
1494  * +------+----+-------------+-----------+---------------+---------------+
1495  * | base | SA | source addr | dest addr | rsrvd. or key | soft lifetime |
1496  * +------+----+-------------+-----------+---------------+---------------+
1497  *
1498  * we will end up with
1499  *
1500  * +------+----+-------------+-----------+---------------+
1501  * | base | SA | source addr | dest addr | soft lifetime |
1502  * +------+----+-------------+-----------+---------------+
1503  */
1504 static void
1505 sadb_strip(sadb_msg_t *samsg)
1506 {
1507         sadb_ext_t *ext;
1508         uint8_t *target = NULL;
1509         uint8_t *msgend;
1510         int sofar = SADB_8TO64(sizeof (*samsg));
1511         int copylen;
1512 
1513         ext = (sadb_ext_t *)(samsg + 1);
1514         msgend = (uint8_t *)samsg;
1515         msgend += SADB_64TO8(samsg->sadb_msg_len);
1516         while ((uint8_t *)ext < msgend) {
1517                 if (ext->sadb_ext_type == SADB_EXT_RESERVED ||
1518                     ext->sadb_ext_type == SADB_EXT_KEY_AUTH ||
1519                     ext->sadb_ext_type == SADB_X_EXT_EDUMP ||
1520                     ext->sadb_ext_type == SADB_EXT_KEY_ENCRYPT) {
1521                         /*
1522                          * Aha!  I found a header to be erased.
1523                          */
1524 
1525                         if (target != NULL) {
1526                                 /*
1527                                  * If I had a previous header to be erased,
1528                                  * copy over it.  I can get away with just
1529                                  * copying backwards because the target will
1530                                  * always be 8 bytes behind the source.
1531                                  */
1532                                 copylen = ((uint8_t *)ext) - (target +
1533                                     SADB_64TO8(
1534                                     ((sadb_ext_t *)target)->sadb_ext_len));
1535                                 ovbcopy(((uint8_t *)ext - copylen), target,
1536                                     copylen);
1537                                 target += copylen;
1538                                 ((sadb_ext_t *)target)->sadb_ext_len =
1539                                     SADB_8TO64(((uint8_t *)ext) - target +
1540                                     SADB_64TO8(ext->sadb_ext_len));
1541                         } else {
1542                                 target = (uint8_t *)ext;
1543                         }
1544                 } else {
1545                         sofar += ext->sadb_ext_len;
1546                 }
1547 
1548                 ext = (sadb_ext_t *)(((uint64_t *)ext) + ext->sadb_ext_len);
1549         }
1550 
1551         ASSERT((uint8_t *)ext == msgend);
1552 
1553         if (target != NULL) {
1554                 copylen = ((uint8_t *)ext) - (target +
1555                     SADB_64TO8(((sadb_ext_t *)target)->sadb_ext_len));
1556                 if (copylen != 0)
1557                         ovbcopy(((uint8_t *)ext - copylen), target, copylen);
1558         }
1559 
1560         /* Adjust samsg. */
1561         samsg->sadb_msg_len = (uint16_t)sofar;
1562 
1563         /* Assume all of the rest is cleared by caller in sadb_pfkey_echo(). */
1564 }
1565 
1566 /*
1567  * AH needs to send an error to PF_KEY.  Assume mp points to an M_CTL
1568  * followed by an M_DATA with a PF_KEY message in it.  The serial of
1569  * the sending keysock instance is included.
1570  */
1571 void
1572 sadb_pfkey_error(queue_t *pfkey_q, mblk_t *mp, int error, int diagnostic,
1573     uint_t serial)
1574 {
1575         mblk_t *msg = mp->b_cont;
1576         sadb_msg_t *samsg;
1577         keysock_out_t *kso;
1578 
1579         /*
1580          * Enough functions call this to merit a NULL queue check.
1581          */
1582         if (pfkey_q == NULL) {
1583                 freemsg(mp);
1584                 return;
1585         }
1586 
1587         ASSERT(msg != NULL);
1588         ASSERT((mp->b_wptr - mp->b_rptr) == sizeof (ipsec_info_t));
1589         ASSERT((msg->b_wptr - msg->b_rptr) >= sizeof (sadb_msg_t));
1590         samsg = (sadb_msg_t *)msg->b_rptr;
1591         kso = (keysock_out_t *)mp->b_rptr;
1592 
1593         kso->ks_out_type = KEYSOCK_OUT;
1594         kso->ks_out_len = sizeof (*kso);
1595         kso->ks_out_serial = serial;
1596 
1597         /*
1598          * Only send the base message up in the event of an error.
1599          * Don't worry about bzero()-ing, because it was probably bogus
1600          * anyway.
1601          */
1602         msg->b_wptr = msg->b_rptr + sizeof (*samsg);
1603         samsg = (sadb_msg_t *)msg->b_rptr;
1604         samsg->sadb_msg_len = SADB_8TO64(sizeof (*samsg));
1605         samsg->sadb_msg_errno = (uint8_t)error;
1606         if (diagnostic != SADB_X_DIAGNOSTIC_PRESET)
1607                 samsg->sadb_x_msg_diagnostic = (uint16_t)diagnostic;
1608 
1609         putnext(pfkey_q, mp);
1610 }
1611 
1612 /*
1613  * Send a successful return packet back to keysock via the queue in pfkey_q.
1614  *
1615  * Often, an SA is associated with the reply message, it's passed in if needed,
1616  * and NULL if not.  BTW, that ipsa will have its refcnt appropriately held,
1617  * and the caller will release said refcnt.
1618  */
1619 void
1620 sadb_pfkey_echo(queue_t *pfkey_q, mblk_t *mp, sadb_msg_t *samsg,
1621     keysock_in_t *ksi, ipsa_t *ipsa)
1622 {
1623         keysock_out_t *kso;
1624         mblk_t *mp1;
1625         sadb_msg_t *newsamsg;
1626         uint8_t *oldend;
1627 
1628         ASSERT((mp->b_cont != NULL) &&
1629             ((void *)samsg == (void *)mp->b_cont->b_rptr) &&
1630             ((void *)mp->b_rptr == (void *)ksi));
1631 
1632         switch (samsg->sadb_msg_type) {
1633         case SADB_ADD:
1634         case SADB_UPDATE:
1635         case SADB_X_UPDATEPAIR:
1636         case SADB_X_DELPAIR_STATE:
1637         case SADB_FLUSH:
1638         case SADB_DUMP:
1639                 /*
1640                  * I have all of the message already.  I just need to strip
1641                  * out the keying material and echo the message back.
1642                  *
1643                  * NOTE: for SADB_DUMP, the function sadb_dump() did the
1644                  * work.  When DUMP reaches here, it should only be a base
1645                  * message.
1646                  */
1647         justecho:
1648                 if (ksi->ks_in_extv[SADB_EXT_KEY_AUTH] != NULL ||
1649                     ksi->ks_in_extv[SADB_EXT_KEY_ENCRYPT] != NULL ||
1650                     ksi->ks_in_extv[SADB_X_EXT_EDUMP] != NULL) {
1651                         sadb_strip(samsg);
1652                         /* Assume PF_KEY message is contiguous. */
1653                         ASSERT(mp->b_cont->b_cont == NULL);
1654                         oldend = mp->b_cont->b_wptr;
1655                         mp->b_cont->b_wptr = mp->b_cont->b_rptr +
1656                             SADB_64TO8(samsg->sadb_msg_len);
1657                         bzero(mp->b_cont->b_wptr, oldend - mp->b_cont->b_wptr);
1658                 }
1659                 break;
1660         case SADB_GET:
1661                 /*
1662                  * Do a lot of work here, because of the ipsa I just found.
1663                  * First construct the new PF_KEY message, then abandon
1664                  * the old one.
1665                  */
1666                 mp1 = sadb_sa2msg(ipsa, samsg);
1667                 if (mp1 == NULL) {
1668                         sadb_pfkey_error(pfkey_q, mp, ENOMEM,
1669                             SADB_X_DIAGNOSTIC_NONE, ksi->ks_in_serial);
1670                         return;
1671                 }
1672                 freemsg(mp->b_cont);
1673                 mp->b_cont = mp1;
1674                 break;
1675         case SADB_DELETE:
1676         case SADB_X_DELPAIR:
1677                 if (ipsa == NULL)
1678                         goto justecho;
1679                 /*
1680                  * Because listening KMds may require more info, treat
1681                  * DELETE like a special case of GET.
1682                  */
1683                 mp1 = sadb_sa2msg(ipsa, samsg);
1684                 if (mp1 == NULL) {
1685                         sadb_pfkey_error(pfkey_q, mp, ENOMEM,
1686                             SADB_X_DIAGNOSTIC_NONE, ksi->ks_in_serial);
1687                         return;
1688                 }
1689                 newsamsg = (sadb_msg_t *)mp1->b_rptr;
1690                 sadb_strip(newsamsg);
1691                 oldend = mp1->b_wptr;
1692                 mp1->b_wptr = mp1->b_rptr + SADB_64TO8(newsamsg->sadb_msg_len);
1693                 bzero(mp1->b_wptr, oldend - mp1->b_wptr);
1694                 freemsg(mp->b_cont);
1695                 mp->b_cont = mp1;
1696                 break;
1697         default:
1698                 if (mp != NULL)
1699                         freemsg(mp);
1700                 return;
1701         }
1702 
1703         /* ksi is now null and void. */
1704         kso = (keysock_out_t *)ksi;
1705         kso->ks_out_type = KEYSOCK_OUT;
1706         kso->ks_out_len = sizeof (*kso);
1707         kso->ks_out_serial = ksi->ks_in_serial;
1708         /* We're ready to send... */
1709         putnext(pfkey_q, mp);
1710 }
1711 
1712 /*
1713  * Set up a global pfkey_q instance for AH, ESP, or some other consumer.
1714  */
1715 void
1716 sadb_keysock_hello(queue_t **pfkey_qp, queue_t *q, mblk_t *mp,
1717     void (*ager)(void *), void *agerarg, timeout_id_t *top, int satype)
1718 {
1719         keysock_hello_ack_t *kha;
1720         queue_t *oldq;
1721 
1722         ASSERT(OTHERQ(q) != NULL);
1723 
1724         /*
1725          * First, check atomically that I'm the first and only keysock
1726          * instance.
1727          *
1728          * Use OTHERQ(q), because qreply(q, mp) == putnext(OTHERQ(q), mp),
1729          * and I want this module to say putnext(*_pfkey_q, mp) for PF_KEY
1730          * messages.
1731          */
1732 
1733         oldq = atomic_cas_ptr((void **)pfkey_qp, NULL, OTHERQ(q));
1734         if (oldq != NULL) {
1735                 ASSERT(oldq != q);
1736                 cmn_err(CE_WARN, "Danger!  Multiple keysocks on top of %s.\n",
1737                     (satype == SADB_SATYPE_ESP)? "ESP" : "AH or other");
1738                 freemsg(mp);
1739                 return;
1740         }
1741 
1742         kha = (keysock_hello_ack_t *)mp->b_rptr;
1743         kha->ks_hello_len = sizeof (keysock_hello_ack_t);
1744         kha->ks_hello_type = KEYSOCK_HELLO_ACK;
1745         kha->ks_hello_satype = (uint8_t)satype;
1746 
1747         /*
1748          * If we made it past the atomic_cas_ptr, then we have "exclusive"
1749          * access to the timeout handle.  Fire it off after the default ager
1750          * interval.
1751          */
1752         *top = qtimeout(*pfkey_qp, ager, agerarg,
1753             drv_usectohz(SADB_AGE_INTERVAL_DEFAULT * 1000));
1754 
1755         putnext(*pfkey_qp, mp);
1756 }
1757 
1758 /*
1759  * Normalize IPv4-mapped IPv6 addresses (and prefixes) as appropriate.
1760  *
1761  * Check addresses themselves for wildcard or multicast.
1762  * Check ire table for local/non-local/broadcast.
1763  */
1764 int
1765 sadb_addrcheck(queue_t *pfkey_q, mblk_t *mp, sadb_ext_t *ext, uint_t serial,
1766     netstack_t *ns)
1767 {
1768         sadb_address_t *addr = (sadb_address_t *)ext;
1769         struct sockaddr_in *sin;
1770         struct sockaddr_in6 *sin6;
1771         int diagnostic, type;
1772         boolean_t normalized = B_FALSE;
1773 
1774         ASSERT(ext != NULL);
1775         ASSERT((ext->sadb_ext_type == SADB_EXT_ADDRESS_SRC) ||
1776             (ext->sadb_ext_type == SADB_EXT_ADDRESS_DST) ||
1777             (ext->sadb_ext_type == SADB_X_EXT_ADDRESS_INNER_SRC) ||
1778             (ext->sadb_ext_type == SADB_X_EXT_ADDRESS_INNER_DST) ||
1779             (ext->sadb_ext_type == SADB_X_EXT_ADDRESS_NATT_LOC) ||
1780             (ext->sadb_ext_type == SADB_X_EXT_ADDRESS_NATT_REM));
1781 
1782         /* Assign both sockaddrs, the compiler will do the right thing. */
1783         sin = (struct sockaddr_in *)(addr + 1);
1784         sin6 = (struct sockaddr_in6 *)(addr + 1);
1785 
1786         if (sin6->sin6_family == AF_INET6) {
1787                 if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
1788                         /*
1789                          * Convert to an AF_INET sockaddr.  This means the
1790                          * return messages will have the extra space, but have
1791                          * AF_INET sockaddrs instead of AF_INET6.
1792                          *
1793                          * Yes, RFC 2367 isn't clear on what to do here w.r.t.
1794                          * mapped addresses, but since AF_INET6 ::ffff:<v4> is
1795                          * equal to AF_INET <v4>, it shouldnt be a huge
1796                          * problem.
1797                          */
1798                         sin->sin_family = AF_INET;
1799                         IN6_V4MAPPED_TO_INADDR(&sin6->sin6_addr,
1800                             &sin->sin_addr);
1801                         bzero(&sin->sin_zero, sizeof (sin->sin_zero));
1802                         normalized = B_TRUE;
1803                 }
1804         } else if (sin->sin_family != AF_INET) {
1805                 switch (ext->sadb_ext_type) {
1806                 case SADB_EXT_ADDRESS_SRC:
1807                         diagnostic = SADB_X_DIAGNOSTIC_BAD_SRC_AF;
1808                         break;
1809                 case SADB_EXT_ADDRESS_DST:
1810                         diagnostic = SADB_X_DIAGNOSTIC_BAD_DST_AF;
1811                         break;
1812                 case SADB_X_EXT_ADDRESS_INNER_SRC:
1813                         diagnostic = SADB_X_DIAGNOSTIC_BAD_PROXY_AF;
1814                         break;
1815                 case SADB_X_EXT_ADDRESS_INNER_DST:
1816                         diagnostic = SADB_X_DIAGNOSTIC_BAD_INNER_DST_AF;
1817                         break;
1818                 case SADB_X_EXT_ADDRESS_NATT_LOC:
1819                         diagnostic = SADB_X_DIAGNOSTIC_BAD_NATT_LOC_AF;
1820                         break;
1821                 case SADB_X_EXT_ADDRESS_NATT_REM:
1822                         diagnostic = SADB_X_DIAGNOSTIC_BAD_NATT_REM_AF;
1823                         break;
1824                         /* There is no default, see above ASSERT. */
1825                 }
1826 bail:
1827                 if (pfkey_q != NULL) {
1828                         sadb_pfkey_error(pfkey_q, mp, EINVAL, diagnostic,
1829                             serial);
1830                 } else {
1831                         /*
1832                          * Scribble in sadb_msg that we got passed in.
1833                          * Overload "mp" to be an sadb_msg pointer.
1834                          */
1835                         sadb_msg_t *samsg = (sadb_msg_t *)mp;
1836 
1837                         samsg->sadb_msg_errno = EINVAL;
1838                         samsg->sadb_x_msg_diagnostic = diagnostic;
1839                 }
1840                 return (KS_IN_ADDR_UNKNOWN);
1841         }
1842 
1843         if (ext->sadb_ext_type == SADB_X_EXT_ADDRESS_INNER_SRC ||
1844             ext->sadb_ext_type == SADB_X_EXT_ADDRESS_INNER_DST) {
1845                 /*
1846                  * We need only check for prefix issues.
1847                  */
1848 
1849                 /* Set diagnostic now, in case we need it later. */
1850                 diagnostic =
1851                     (ext->sadb_ext_type == SADB_X_EXT_ADDRESS_INNER_SRC) ?
1852                     SADB_X_DIAGNOSTIC_PREFIX_INNER_SRC :
1853                     SADB_X_DIAGNOSTIC_PREFIX_INNER_DST;
1854 
1855                 if (normalized)
1856                         addr->sadb_address_prefixlen -= 96;
1857 
1858                 /*
1859                  * Verify and mask out inner-addresses based on prefix length.
1860                  */
1861                 if (sin->sin_family == AF_INET) {
1862                         if (addr->sadb_address_prefixlen > 32)
1863                                 goto bail;
1864                         sin->sin_addr.s_addr &=
1865                             ip_plen_to_mask(addr->sadb_address_prefixlen);
1866                 } else {
1867                         in6_addr_t mask;
1868 
1869                         ASSERT(sin->sin_family == AF_INET6);
1870                         /*
1871                          * ip_plen_to_mask_v6() returns NULL if the value in
1872                          * question is out of range.
1873                          */
1874                         if (ip_plen_to_mask_v6(addr->sadb_address_prefixlen,
1875                             &mask) == NULL)
1876                                 goto bail;
1877                         sin6->sin6_addr.s6_addr32[0] &= mask.s6_addr32[0];
1878                         sin6->sin6_addr.s6_addr32[1] &= mask.s6_addr32[1];
1879                         sin6->sin6_addr.s6_addr32[2] &= mask.s6_addr32[2];
1880                         sin6->sin6_addr.s6_addr32[3] &= mask.s6_addr32[3];
1881                 }
1882 
1883                 /* We don't care in these cases. */
1884                 return (KS_IN_ADDR_DONTCARE);
1885         }
1886 
1887         if (sin->sin_family == AF_INET6) {
1888                 /* Check the easy ones now. */
1889                 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
1890                         return (KS_IN_ADDR_MBCAST);
1891                 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
1892                         return (KS_IN_ADDR_UNSPEC);
1893                 /*
1894                  * At this point, we're a unicast IPv6 address.
1895                  *
1896                  * XXX Zones alert -> me/notme decision needs to be tempered
1897                  * by what zone we're in when we go to zone-aware IPsec.
1898                  */
1899                 if (ip_type_v6(&sin6->sin6_addr, ns->netstack_ip) ==
1900                     IRE_LOCAL) {
1901                         /* Hey hey, it's local. */
1902                         return (KS_IN_ADDR_ME);
1903                 }
1904         } else {
1905                 ASSERT(sin->sin_family == AF_INET);
1906                 if (sin->sin_addr.s_addr == INADDR_ANY)
1907                         return (KS_IN_ADDR_UNSPEC);
1908                 if (CLASSD(sin->sin_addr.s_addr))
1909                         return (KS_IN_ADDR_MBCAST);
1910                 /*
1911                  * At this point we're a unicast or broadcast IPv4 address.
1912                  *
1913                  * Check if the address is IRE_BROADCAST or IRE_LOCAL.
1914                  *
1915                  * XXX Zones alert -> me/notme decision needs to be tempered
1916                  * by what zone we're in when we go to zone-aware IPsec.
1917                  */
1918                 type = ip_type_v4(sin->sin_addr.s_addr, ns->netstack_ip);
1919                 switch (type) {
1920                 case IRE_LOCAL:
1921                         return (KS_IN_ADDR_ME);
1922                 case IRE_BROADCAST:
1923                         return (KS_IN_ADDR_MBCAST);
1924                 }
1925         }
1926 
1927         return (KS_IN_ADDR_NOTME);
1928 }
1929 
1930 /*
1931  * Address normalizations and reality checks for inbound PF_KEY messages.
1932  *
1933  * For the case of src == unspecified AF_INET6, and dst == AF_INET, convert
1934  * the source to AF_INET.  Do the same for the inner sources.
1935  */
1936 boolean_t
1937 sadb_addrfix(keysock_in_t *ksi, queue_t *pfkey_q, mblk_t *mp, netstack_t *ns)
1938 {
1939         struct sockaddr_in *src, *isrc;
1940         struct sockaddr_in6 *dst, *idst;
1941         sadb_address_t *srcext, *dstext;
1942         uint16_t sport;
1943         sadb_ext_t **extv = ksi->ks_in_extv;
1944         int rc;
1945 
1946         if (extv[SADB_EXT_ADDRESS_SRC] != NULL) {
1947                 rc = sadb_addrcheck(pfkey_q, mp, extv[SADB_EXT_ADDRESS_SRC],
1948                     ksi->ks_in_serial, ns);
1949                 if (rc == KS_IN_ADDR_UNKNOWN)
1950                         return (B_FALSE);
1951                 if (rc == KS_IN_ADDR_MBCAST) {
1952                         sadb_pfkey_error(pfkey_q, mp, EINVAL,
1953                             SADB_X_DIAGNOSTIC_BAD_SRC, ksi->ks_in_serial);
1954                         return (B_FALSE);
1955                 }
1956                 ksi->ks_in_srctype = rc;
1957         }
1958 
1959         if (extv[SADB_EXT_ADDRESS_DST] != NULL) {
1960                 rc = sadb_addrcheck(pfkey_q, mp, extv[SADB_EXT_ADDRESS_DST],
1961                     ksi->ks_in_serial, ns);
1962                 if (rc == KS_IN_ADDR_UNKNOWN)
1963                         return (B_FALSE);
1964                 if (rc == KS_IN_ADDR_UNSPEC) {
1965                         sadb_pfkey_error(pfkey_q, mp, EINVAL,
1966                             SADB_X_DIAGNOSTIC_BAD_DST, ksi->ks_in_serial);
1967                         return (B_FALSE);
1968                 }
1969                 ksi->ks_in_dsttype = rc;
1970         }
1971 
1972         /*
1973          * NAT-Traversal addrs are simple enough to not require all of
1974          * the checks in sadb_addrcheck().  Just normalize or reject if not
1975          * AF_INET.
1976          */
1977         if (extv[SADB_X_EXT_ADDRESS_NATT_LOC] != NULL) {
1978                 rc = sadb_addrcheck(pfkey_q, mp,
1979                     extv[SADB_X_EXT_ADDRESS_NATT_LOC], ksi->ks_in_serial, ns);
1980 
1981                 /*
1982                  * Local NAT-T addresses never use an IRE_LOCAL, so it should
1983                  * always be NOTME, or UNSPEC (to handle both tunnel mode
1984                  * AND local-port flexibility).
1985                  */
1986                 if (rc != KS_IN_ADDR_NOTME && rc != KS_IN_ADDR_UNSPEC) {
1987                         sadb_pfkey_error(pfkey_q, mp, EINVAL,
1988                             SADB_X_DIAGNOSTIC_MALFORMED_NATT_LOC,
1989                             ksi->ks_in_serial);
1990                         return (B_FALSE);
1991                 }
1992                 src = (struct sockaddr_in *)
1993                     (((sadb_address_t *)extv[SADB_X_EXT_ADDRESS_NATT_LOC]) + 1);
1994                 if (src->sin_family != AF_INET) {
1995                         sadb_pfkey_error(pfkey_q, mp, EINVAL,
1996                             SADB_X_DIAGNOSTIC_BAD_NATT_LOC_AF,
1997                             ksi->ks_in_serial);
1998                         return (B_FALSE);
1999                 }
2000         }
2001 
2002         if (extv[SADB_X_EXT_ADDRESS_NATT_REM] != NULL) {
2003                 rc = sadb_addrcheck(pfkey_q, mp,
2004                     extv[SADB_X_EXT_ADDRESS_NATT_REM], ksi->ks_in_serial, ns);
2005 
2006                 /*
2007                  * Remote NAT-T addresses never use an IRE_LOCAL, so it should
2008                  * always be NOTME, or UNSPEC if it's a tunnel-mode SA.
2009                  */
2010                 if (rc != KS_IN_ADDR_NOTME &&
2011                     !(extv[SADB_X_EXT_ADDRESS_INNER_SRC] != NULL &&
2012                     rc == KS_IN_ADDR_UNSPEC)) {
2013                         sadb_pfkey_error(pfkey_q, mp, EINVAL,
2014                             SADB_X_DIAGNOSTIC_MALFORMED_NATT_REM,
2015                             ksi->ks_in_serial);
2016                         return (B_FALSE);
2017                 }
2018                 src = (struct sockaddr_in *)
2019                     (((sadb_address_t *)extv[SADB_X_EXT_ADDRESS_NATT_REM]) + 1);
2020                 if (src->sin_family != AF_INET) {
2021                         sadb_pfkey_error(pfkey_q, mp, EINVAL,
2022                             SADB_X_DIAGNOSTIC_BAD_NATT_REM_AF,
2023                             ksi->ks_in_serial);
2024                         return (B_FALSE);
2025                 }
2026         }
2027 
2028         if (extv[SADB_X_EXT_ADDRESS_INNER_SRC] != NULL) {
2029                 if (extv[SADB_X_EXT_ADDRESS_INNER_DST] == NULL) {
2030                         sadb_pfkey_error(pfkey_q, mp, EINVAL,
2031                             SADB_X_DIAGNOSTIC_MISSING_INNER_DST,
2032                             ksi->ks_in_serial);
2033                         return (B_FALSE);
2034                 }
2035 
2036                 if (sadb_addrcheck(pfkey_q, mp,
2037                     extv[SADB_X_EXT_ADDRESS_INNER_DST], ksi->ks_in_serial, ns)
2038                     == KS_IN_ADDR_UNKNOWN ||
2039                     sadb_addrcheck(pfkey_q, mp,
2040                     extv[SADB_X_EXT_ADDRESS_INNER_SRC], ksi->ks_in_serial, ns)
2041                     == KS_IN_ADDR_UNKNOWN)
2042                         return (B_FALSE);
2043 
2044                 isrc = (struct sockaddr_in *)
2045                     (((sadb_address_t *)extv[SADB_X_EXT_ADDRESS_INNER_SRC]) +
2046                     1);
2047                 idst = (struct sockaddr_in6 *)
2048                     (((sadb_address_t *)extv[SADB_X_EXT_ADDRESS_INNER_DST]) +
2049                     1);
2050                 if (isrc->sin_family != idst->sin6_family) {
2051                         sadb_pfkey_error(pfkey_q, mp, EINVAL,
2052                             SADB_X_DIAGNOSTIC_INNER_AF_MISMATCH,
2053                             ksi->ks_in_serial);
2054                         return (B_FALSE);
2055                 }
2056         } else if (extv[SADB_X_EXT_ADDRESS_INNER_DST] != NULL) {
2057                         sadb_pfkey_error(pfkey_q, mp, EINVAL,
2058                             SADB_X_DIAGNOSTIC_MISSING_INNER_SRC,
2059                             ksi->ks_in_serial);
2060                         return (B_FALSE);
2061         } else {
2062                 isrc = NULL;    /* For inner/outer port check below. */
2063         }
2064 
2065         dstext = (sadb_address_t *)extv[SADB_EXT_ADDRESS_DST];
2066         srcext = (sadb_address_t *)extv[SADB_EXT_ADDRESS_SRC];
2067 
2068         if (dstext == NULL || srcext == NULL)
2069                 return (B_TRUE);
2070 
2071         dst = (struct sockaddr_in6 *)(dstext + 1);
2072         src = (struct sockaddr_in *)(srcext + 1);
2073 
2074         if (isrc != NULL &&
2075             (isrc->sin_port != 0 || idst->sin6_port != 0) &&
2076             (src->sin_port != 0 || dst->sin6_port != 0)) {
2077                 /* Can't set inner and outer ports in one SA. */
2078                 sadb_pfkey_error(pfkey_q, mp, EINVAL,
2079                     SADB_X_DIAGNOSTIC_DUAL_PORT_SETS,
2080                     ksi->ks_in_serial);
2081                 return (B_FALSE);
2082         }
2083 
2084         if (dst->sin6_family == src->sin_family)
2085                 return (B_TRUE);
2086 
2087         if (srcext->sadb_address_proto != dstext->sadb_address_proto) {
2088                 if (srcext->sadb_address_proto == 0) {
2089                         srcext->sadb_address_proto = dstext->sadb_address_proto;
2090                 } else if (dstext->sadb_address_proto == 0) {
2091                         dstext->sadb_address_proto = srcext->sadb_address_proto;
2092                 } else {
2093                         /* Inequal protocols, neither were 0.  Report error. */
2094                         sadb_pfkey_error(pfkey_q, mp, EINVAL,
2095                             SADB_X_DIAGNOSTIC_PROTO_MISMATCH,
2096                             ksi->ks_in_serial);
2097                         return (B_FALSE);
2098                 }
2099         }
2100 
2101         /*
2102          * With the exception of an unspec IPv6 source and an IPv4
2103          * destination, address families MUST me matched.
2104          */
2105         if (src->sin_family == AF_INET ||
2106             ksi->ks_in_srctype != KS_IN_ADDR_UNSPEC) {
2107                 sadb_pfkey_error(pfkey_q, mp, EINVAL,
2108                     SADB_X_DIAGNOSTIC_AF_MISMATCH, ksi->ks_in_serial);
2109                 return (B_FALSE);
2110         }
2111 
2112         /*
2113          * Convert "src" to AF_INET INADDR_ANY.  We rely on sin_port being
2114          * in the same place for sockaddr_in and sockaddr_in6.
2115          */
2116         sport = src->sin_port;
2117         bzero(src, sizeof (*src));
2118         src->sin_family = AF_INET;
2119         src->sin_port = sport;
2120 
2121         return (B_TRUE);
2122 }
2123 
2124 /*
2125  * Set the results in "addrtype", given an IRE as requested by
2126  * sadb_addrcheck().
2127  */
2128 int
2129 sadb_addrset(ire_t *ire)
2130 {
2131         if ((ire->ire_type & IRE_BROADCAST) ||
2132             (ire->ire_ipversion == IPV4_VERSION && CLASSD(ire->ire_addr)) ||
2133             (ire->ire_ipversion == IPV6_VERSION &&
2134             IN6_IS_ADDR_MULTICAST(&(ire->ire_addr_v6))))
2135                 return (KS_IN_ADDR_MBCAST);
2136         if (ire->ire_type & (IRE_LOCAL | IRE_LOOPBACK))
2137                 return (KS_IN_ADDR_ME);
2138         return (KS_IN_ADDR_NOTME);
2139 }
2140 
2141 /*
2142  * Match primitives..
2143  * !!! TODO: short term: inner selectors
2144  *              ipv6 scope id (ifindex)
2145  * longer term:  zone id.  sensitivity label. uid.
2146  */
2147 boolean_t
2148 sadb_match_spi(ipsa_query_t *sq, ipsa_t *sa)
2149 {
2150         return (sq->spi == sa->ipsa_spi);
2151 }
2152 
2153 boolean_t
2154 sadb_match_dst_v6(ipsa_query_t *sq, ipsa_t *sa)
2155 {
2156         return (IPSA_ARE_ADDR_EQUAL(sa->ipsa_dstaddr, sq->dstaddr, AF_INET6));
2157 }
2158 
2159 boolean_t
2160 sadb_match_src_v6(ipsa_query_t *sq, ipsa_t *sa)
2161 {
2162         return (IPSA_ARE_ADDR_EQUAL(sa->ipsa_srcaddr, sq->srcaddr, AF_INET6));
2163 }
2164 
2165 boolean_t
2166 sadb_match_dst_v4(ipsa_query_t *sq, ipsa_t *sa)
2167 {
2168         return (sq->dstaddr[0] == sa->ipsa_dstaddr[0]);
2169 }
2170 
2171 boolean_t
2172 sadb_match_src_v4(ipsa_query_t *sq, ipsa_t *sa)
2173 {
2174         return (sq->srcaddr[0] == sa->ipsa_srcaddr[0]);
2175 }
2176 
2177 boolean_t
2178 sadb_match_dstid(ipsa_query_t *sq, ipsa_t *sa)
2179 {
2180         return ((sa->ipsa_dst_cid != NULL) &&
2181             (sq->didtype == sa->ipsa_dst_cid->ipsid_type) &&
2182             (strcmp(sq->didstr, sa->ipsa_dst_cid->ipsid_cid) == 0));
2183 
2184 }
2185 boolean_t
2186 sadb_match_srcid(ipsa_query_t *sq, ipsa_t *sa)
2187 {
2188         return ((sa->ipsa_src_cid != NULL) &&
2189             (sq->sidtype == sa->ipsa_src_cid->ipsid_type) &&
2190             (strcmp(sq->sidstr, sa->ipsa_src_cid->ipsid_cid) == 0));
2191 }
2192 
2193 boolean_t
2194 sadb_match_kmc(ipsa_query_t *sq, ipsa_t *sa)
2195 {
2196 #define M(a, b) (((a) == 0) || ((b) == 0) || ((a) == (b)))
2197 
2198         return (M(sq->kmc, sa->ipsa_kmc) && M(sq->kmp, sa->ipsa_kmp));
2199 
2200 #undef M
2201 }
2202 
2203 /*
2204  * Common function which extracts several PF_KEY extensions for ease of
2205  * SADB matching.
2206  *
2207  * XXX TODO: weed out ipsa_query_t fields not used during matching
2208  * or afterwards?
2209  */
2210 int
2211 sadb_form_query(keysock_in_t *ksi, uint32_t req, uint32_t match,
2212     ipsa_query_t *sq, int *diagnostic)
2213 {
2214         int i;
2215         ipsa_match_fn_t *mfpp = &(sq->matchers[0]);
2216 
2217         for (i = 0; i < IPSA_NMATCH; i++)
2218                 sq->matchers[i] = NULL;
2219 
2220         ASSERT((req & ~match) == 0);
2221 
2222         sq->req = req;
2223         sq->dstext = (sadb_address_t *)ksi->ks_in_extv[SADB_EXT_ADDRESS_DST];
2224         sq->srcext = (sadb_address_t *)ksi->ks_in_extv[SADB_EXT_ADDRESS_SRC];
2225         sq->assoc = (sadb_sa_t *)ksi->ks_in_extv[SADB_EXT_SA];
2226 
2227         if ((req & IPSA_Q_DST) && (sq->dstext == NULL)) {
2228                 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_DST;
2229                 return (EINVAL);
2230         }
2231         if ((req & IPSA_Q_SRC) && (sq->srcext == NULL)) {
2232                 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_SRC;
2233                 return (EINVAL);
2234         }
2235         if ((req & IPSA_Q_SA) && (sq->assoc == NULL)) {
2236                 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_SA;
2237                 return (EINVAL);
2238         }
2239 
2240         if (match & IPSA_Q_SA) {
2241                 *mfpp++ = sadb_match_spi;
2242                 sq->spi = sq->assoc->sadb_sa_spi;
2243         }
2244 
2245         if (sq->dstext != NULL)
2246                 sq->dst = (struct sockaddr_in *)(sq->dstext + 1);
2247         else {
2248                 sq->dst = NULL;
2249                 sq->dst6 = NULL;
2250                 sq->dstaddr = NULL;
2251         }
2252 
2253         if (sq->srcext != NULL)
2254                 sq->src = (struct sockaddr_in *)(sq->srcext + 1);
2255         else {
2256                 sq->src = NULL;
2257                 sq->src6 = NULL;
2258                 sq->srcaddr = NULL;
2259         }
2260 
2261         if (sq->dst != NULL)
2262                 sq->af = sq->dst->sin_family;
2263         else if (sq->src != NULL)
2264                 sq->af = sq->src->sin_family;
2265         else
2266                 sq->af = AF_INET;
2267 
2268         if (sq->af == AF_INET6) {
2269                 if ((match & IPSA_Q_DST) && (sq->dstext != NULL)) {
2270                         *mfpp++ = sadb_match_dst_v6;
2271                         sq->dst6 = (struct sockaddr_in6 *)sq->dst;
2272                         sq->dstaddr = (uint32_t *)&(sq->dst6->sin6_addr);
2273                 } else {
2274                         match &= ~IPSA_Q_DST;
2275                         sq->dstaddr = ALL_ZEROES_PTR;
2276                 }
2277 
2278                 if ((match & IPSA_Q_SRC) && (sq->srcext != NULL)) {
2279                         sq->src6 = (struct sockaddr_in6 *)(sq->srcext + 1);
2280                         sq->srcaddr = (uint32_t *)&sq->src6->sin6_addr;
2281                         if (sq->src6->sin6_family != AF_INET6) {
2282                                 *diagnostic = SADB_X_DIAGNOSTIC_AF_MISMATCH;
2283                                 return (EINVAL);
2284                         }
2285                         *mfpp++ = sadb_match_src_v6;
2286                 } else {
2287                         match &= ~IPSA_Q_SRC;
2288                         sq->srcaddr = ALL_ZEROES_PTR;
2289                 }
2290         } else {
2291                 sq->src6 = sq->dst6 = NULL;
2292                 if ((match & IPSA_Q_DST) && (sq->dstext != NULL)) {
2293                         *mfpp++ = sadb_match_dst_v4;
2294                         sq->dstaddr = (uint32_t *)&sq->dst->sin_addr;
2295                 } else {
2296                         match &= ~IPSA_Q_DST;
2297                         sq->dstaddr = ALL_ZEROES_PTR;
2298                 }
2299                 if ((match & IPSA_Q_SRC) && (sq->srcext != NULL)) {
2300                         sq->srcaddr = (uint32_t *)&sq->src->sin_addr;
2301                         if (sq->src->sin_family != AF_INET) {
2302                                 *diagnostic = SADB_X_DIAGNOSTIC_AF_MISMATCH;
2303                                 return (EINVAL);
2304                         }
2305                         *mfpp++ = sadb_match_src_v4;
2306                 } else {
2307                         match &= ~IPSA_Q_SRC;
2308                         sq->srcaddr = ALL_ZEROES_PTR;
2309                 }
2310         }
2311 
2312         sq->dstid = (sadb_ident_t *)ksi->ks_in_extv[SADB_EXT_IDENTITY_DST];
2313         if ((match & IPSA_Q_DSTID) && (sq->dstid != NULL)) {
2314                 sq->didstr = (char *)(sq->dstid + 1);
2315                 sq->didtype = sq->dstid->sadb_ident_type;
2316                 *mfpp++ = sadb_match_dstid;
2317         }
2318 
2319         sq->srcid = (sadb_ident_t *)ksi->ks_in_extv[SADB_EXT_IDENTITY_SRC];
2320 
2321         if ((match & IPSA_Q_SRCID) && (sq->srcid != NULL)) {
2322                 sq->sidstr = (char *)(sq->srcid + 1);
2323                 sq->sidtype = sq->srcid->sadb_ident_type;
2324                 *mfpp++ = sadb_match_srcid;
2325         }
2326 
2327         sq->kmcext = (sadb_x_kmc_t *)ksi->ks_in_extv[SADB_X_EXT_KM_COOKIE];
2328         sq->kmc = 0;
2329         sq->kmp = 0;
2330 
2331         if ((match & IPSA_Q_KMC) && (sq->kmcext)) {
2332                 sq->kmp = sq->kmcext->sadb_x_kmc_proto;
2333                 /* Be liberal in what we receive.  Special-case IKEv1. */
2334                 if (sq->kmp == SADB_X_KMP_IKE) {
2335                         /* Just in case in.iked is misbehaving... */
2336                         sq->kmcext->sadb_x_kmc_reserved = 0;
2337                 }
2338                 sq->kmc = sq->kmcext->sadb_x_kmc_cookie64;
2339                 *mfpp++ = sadb_match_kmc;
2340         }
2341 
2342         if (match & (IPSA_Q_INBOUND|IPSA_Q_OUTBOUND)) {
2343                 if (sq->af == AF_INET6)
2344                         sq->sp = &sq->spp->s_v6;
2345                 else
2346                         sq->sp = &sq->spp->s_v4;
2347         } else {
2348                 sq->sp = NULL;
2349         }
2350 
2351         if (match & IPSA_Q_INBOUND) {
2352                 sq->inhash = INBOUND_HASH(sq->sp, sq->assoc->sadb_sa_spi);
2353                 sq->inbound = &sq->sp->sdb_if[sq->inhash];
2354         } else {
2355                 sq->inhash = 0;
2356                 sq->inbound = NULL;
2357         }
2358 
2359         if (match & IPSA_Q_OUTBOUND) {
2360                 if (sq->af == AF_INET6) {
2361                         sq->outhash = OUTBOUND_HASH_V6(sq->sp, *(sq->dstaddr));
2362                 } else {
2363                         sq->outhash = OUTBOUND_HASH_V4(sq->sp, *(sq->dstaddr));
2364                 }
2365                 sq->outbound = &sq->sp->sdb_of[sq->outhash];
2366         } else {
2367                 sq->outhash = 0;
2368                 sq->outbound = NULL;
2369         }
2370         sq->match = match;
2371         return (0);
2372 }
2373 
2374 /*
2375  * Match an initialized query structure with a security association;
2376  * return B_TRUE on a match, B_FALSE on a miss.
2377  * Applies match functions set up by sadb_form_query() until one returns false.
2378  */
2379 boolean_t
2380 sadb_match_query(ipsa_query_t *sq, ipsa_t *sa)
2381 {
2382         ipsa_match_fn_t *mfpp = &(sq->matchers[0]);
2383         ipsa_match_fn_t mfp;
2384 
2385         for (mfp = *mfpp++; mfp != NULL; mfp = *mfpp++) {
2386                 if (!mfp(sq, sa))
2387                         return (B_FALSE);
2388         }
2389         return (B_TRUE);
2390 }
2391 
2392 /*
2393  * Walker callback function to delete sa's based on src/dst address.
2394  * Assumes that we're called with *head locked, no other locks held;
2395  * Conveniently, and not coincidentally, this is both what sadb_walker
2396  * gives us and also what sadb_unlinkassoc expects.
2397  */
2398 struct sadb_purge_state
2399 {
2400         ipsa_query_t sq;
2401         boolean_t inbnd;
2402         uint8_t sadb_sa_state;
2403 };
2404 
2405 static void
2406 sadb_purge_cb(isaf_t *head, ipsa_t *entry, void *cookie)
2407 {
2408         struct sadb_purge_state *ps = (struct sadb_purge_state *)cookie;
2409 
2410         ASSERT(MUTEX_HELD(&head->isaf_lock));
2411 
2412         mutex_enter(&entry->ipsa_lock);
2413 
2414         if (entry->ipsa_state == IPSA_STATE_LARVAL ||
2415             !sadb_match_query(&ps->sq, entry)) {
2416                 mutex_exit(&entry->ipsa_lock);
2417                 return;
2418         }
2419 
2420         if (ps->inbnd) {
2421                 sadb_delete_cluster(entry);
2422         }
2423         entry->ipsa_state = IPSA_STATE_DEAD;
2424         (void) sadb_torch_assoc(head, entry);
2425 }
2426 
2427 /*
2428  * Common code to purge an SA with a matching src or dst address.
2429  * Don't kill larval SA's in such a purge.
2430  */
2431 int
2432 sadb_purge_sa(mblk_t *mp, keysock_in_t *ksi, sadb_t *sp,
2433     int *diagnostic, queue_t *pfkey_q)
2434 {
2435         struct sadb_purge_state ps;
2436         int error = sadb_form_query(ksi, 0,
2437             IPSA_Q_SRC|IPSA_Q_DST|IPSA_Q_SRCID|IPSA_Q_DSTID|IPSA_Q_KMC,
2438             &ps.sq, diagnostic);
2439 
2440         if (error != 0)
2441                 return (error);
2442 
2443         /*
2444          * This is simple, crude, and effective.
2445          * Unimplemented optimizations (TBD):
2446          * - we can limit how many places we search based on where we
2447          * think the SA is filed.
2448          * - if we get a dst address, we can hash based on dst addr to find
2449          * the correct bucket in the outbound table.
2450          */
2451         ps.inbnd = B_TRUE;
2452         sadb_walker(sp->sdb_if, sp->sdb_hashsize, sadb_purge_cb, &ps);
2453         ps.inbnd = B_FALSE;
2454         sadb_walker(sp->sdb_of, sp->sdb_hashsize, sadb_purge_cb, &ps);
2455 
2456         ASSERT(mp->b_cont != NULL);
2457         sadb_pfkey_echo(pfkey_q, mp, (sadb_msg_t *)mp->b_cont->b_rptr, ksi,
2458             NULL);
2459         return (0);
2460 }
2461 
2462 static void
2463 sadb_delpair_state_one(isaf_t *head, ipsa_t *entry, void *cookie)
2464 {
2465         struct sadb_purge_state *ps = (struct sadb_purge_state *)cookie;
2466         isaf_t  *inbound_bucket;
2467         ipsa_t *peer_assoc;
2468         ipsa_query_t *sq = &ps->sq;
2469 
2470         ASSERT(MUTEX_HELD(&head->isaf_lock));
2471 
2472         mutex_enter(&entry->ipsa_lock);
2473 
2474         if ((entry->ipsa_state != ps->sadb_sa_state) ||
2475             ((sq->srcaddr != NULL) &&
2476             !IPSA_ARE_ADDR_EQUAL(entry->ipsa_srcaddr, sq->srcaddr, sq->af))) {
2477                 mutex_exit(&entry->ipsa_lock);
2478                 return;
2479         }
2480 
2481         /*
2482          * The isaf_t *, which is passed in , is always an outbound bucket,
2483          * and we are preserving the outbound-then-inbound hash-bucket lock
2484          * ordering. The sadb_walker() which triggers this function is called
2485          * only on the outbound fanout, and the corresponding inbound bucket
2486          * lock is safe to acquire here.
2487          */
2488 
2489         if (entry->ipsa_haspeer) {
2490                 inbound_bucket = INBOUND_BUCKET(sq->sp, entry->ipsa_spi);
2491                 mutex_enter(&inbound_bucket->isaf_lock);
2492                 peer_assoc = ipsec_getassocbyspi(inbound_bucket,
2493                     entry->ipsa_spi, entry->ipsa_srcaddr,
2494                     entry->ipsa_dstaddr, entry->ipsa_addrfam);
2495         } else {
2496                 inbound_bucket = INBOUND_BUCKET(sq->sp, entry->ipsa_otherspi);
2497                 mutex_enter(&inbound_bucket->isaf_lock);
2498                 peer_assoc = ipsec_getassocbyspi(inbound_bucket,
2499                     entry->ipsa_otherspi, entry->ipsa_dstaddr,
2500                     entry->ipsa_srcaddr, entry->ipsa_addrfam);
2501         }
2502 
2503         entry->ipsa_state = IPSA_STATE_DEAD;
2504         (void) sadb_torch_assoc(head, entry);
2505         if (peer_assoc != NULL) {
2506                 mutex_enter(&peer_assoc->ipsa_lock);
2507                 peer_assoc->ipsa_state = IPSA_STATE_DEAD;
2508                 (void) sadb_torch_assoc(inbound_bucket, peer_assoc);
2509         }
2510         mutex_exit(&inbound_bucket->isaf_lock);
2511 }
2512 
2513 static int
2514 sadb_delpair_state(mblk_t *mp, keysock_in_t *ksi, sadbp_t *spp,
2515     int *diagnostic, queue_t *pfkey_q)
2516 {
2517         sadb_sa_t *assoc = (sadb_sa_t *)ksi->ks_in_extv[SADB_EXT_SA];
2518         struct sadb_purge_state ps;
2519         int error;
2520 
2521         ps.sq.spp = spp;                /* XXX param */
2522 
2523         error = sadb_form_query(ksi, IPSA_Q_DST|IPSA_Q_SRC,
2524             IPSA_Q_SRC|IPSA_Q_DST|IPSA_Q_SRCID|IPSA_Q_DSTID|IPSA_Q_KMC,
2525             &ps.sq, diagnostic);
2526         if (error != 0)
2527                 return (error);
2528 
2529         ps.inbnd = B_FALSE;
2530         ps.sadb_sa_state = assoc->sadb_sa_state;
2531         sadb_walker(ps.sq.sp->sdb_of, ps.sq.sp->sdb_hashsize,
2532             sadb_delpair_state_one, &ps);
2533 
2534         ASSERT(mp->b_cont != NULL);
2535         sadb_pfkey_echo(pfkey_q, mp, (sadb_msg_t *)mp->b_cont->b_rptr,
2536             ksi, NULL);
2537         return (0);
2538 }
2539 
2540 /*
2541  * Common code to delete/get an SA.
2542  */
2543 int
2544 sadb_delget_sa(mblk_t *mp, keysock_in_t *ksi, sadbp_t *spp,
2545     int *diagnostic, queue_t *pfkey_q, uint8_t sadb_msg_type)
2546 {
2547         ipsa_query_t sq;
2548         ipsa_t *echo_target = NULL;
2549         ipsap_t ipsapp;
2550         uint_t  error = 0;
2551 
2552         if (sadb_msg_type == SADB_X_DELPAIR_STATE)
2553                 return (sadb_delpair_state(mp, ksi, spp, diagnostic, pfkey_q));
2554 
2555         sq.spp = spp;           /* XXX param */
2556         error = sadb_form_query(ksi, IPSA_Q_DST|IPSA_Q_SA,
2557             IPSA_Q_SRC|IPSA_Q_DST|IPSA_Q_SA|IPSA_Q_INBOUND|IPSA_Q_OUTBOUND,
2558             &sq, diagnostic);
2559         if (error != 0)
2560                 return (error);
2561 
2562         error = get_ipsa_pair(&sq, &ipsapp, diagnostic);
2563         if (error != 0) {
2564                 return (error);
2565         }
2566 
2567         echo_target = ipsapp.ipsap_sa_ptr;
2568         if (echo_target == NULL)
2569                 echo_target = ipsapp.ipsap_psa_ptr;
2570 
2571         if (sadb_msg_type == SADB_DELETE || sadb_msg_type == SADB_X_DELPAIR) {
2572                 /*
2573                  * Bucket locks will be required if SA is actually unlinked.
2574                  * get_ipsa_pair() returns valid hash bucket pointers even
2575                  * if it can't find a pair SA pointer. To prevent a potential
2576                  * deadlock, always lock the outbound bucket before the inbound.
2577                  */
2578                 if (ipsapp.in_inbound_table) {
2579                         mutex_enter(&ipsapp.ipsap_pbucket->isaf_lock);
2580                         mutex_enter(&ipsapp.ipsap_bucket->isaf_lock);
2581                 } else {
2582                         mutex_enter(&ipsapp.ipsap_bucket->isaf_lock);
2583                         mutex_enter(&ipsapp.ipsap_pbucket->isaf_lock);
2584                 }
2585 
2586                 if (ipsapp.ipsap_sa_ptr != NULL) {
2587                         mutex_enter(&ipsapp.ipsap_sa_ptr->ipsa_lock);
2588                         if (ipsapp.ipsap_sa_ptr->ipsa_flags & IPSA_F_INBOUND) {
2589                                 sadb_delete_cluster(ipsapp.ipsap_sa_ptr);
2590                         }
2591                         ipsapp.ipsap_sa_ptr->ipsa_state = IPSA_STATE_DEAD;
2592                         (void) sadb_torch_assoc(ipsapp.ipsap_bucket,
2593                             ipsapp.ipsap_sa_ptr);
2594                         /*
2595                          * sadb_torch_assoc() releases the ipsa_lock
2596                          * and calls sadb_unlinkassoc() which does a
2597                          * IPSA_REFRELE.
2598                          */
2599                 }
2600                 if (ipsapp.ipsap_psa_ptr != NULL) {
2601                         mutex_enter(&ipsapp.ipsap_psa_ptr->ipsa_lock);
2602                         if (sadb_msg_type == SADB_X_DELPAIR ||
2603                             ipsapp.ipsap_psa_ptr->ipsa_haspeer) {
2604                                 if (ipsapp.ipsap_psa_ptr->ipsa_flags &
2605                                     IPSA_F_INBOUND) {
2606                                         sadb_delete_cluster
2607                                             (ipsapp.ipsap_psa_ptr);
2608                                 }
2609                                 ipsapp.ipsap_psa_ptr->ipsa_state =
2610                                     IPSA_STATE_DEAD;
2611                                 (void) sadb_torch_assoc(ipsapp.ipsap_pbucket,
2612                                     ipsapp.ipsap_psa_ptr);
2613                         } else {
2614                                 /*
2615                                  * Only half of the "pair" has been deleted.
2616                                  * Update the remaining SA and remove references
2617                                  * to its pair SA, which is now gone.
2618                                  */
2619                                 ipsapp.ipsap_psa_ptr->ipsa_otherspi = 0;
2620                                 ipsapp.ipsap_psa_ptr->ipsa_flags &=
2621                                     ~IPSA_F_PAIRED;
2622                                 mutex_exit(&ipsapp.ipsap_psa_ptr->ipsa_lock);
2623                         }
2624                 } else if (sadb_msg_type == SADB_X_DELPAIR) {
2625                         *diagnostic = SADB_X_DIAGNOSTIC_PAIR_SA_NOTFOUND;
2626                         error = ESRCH;
2627                 }
2628                 mutex_exit(&ipsapp.ipsap_bucket->isaf_lock);
2629                 mutex_exit(&ipsapp.ipsap_pbucket->isaf_lock);
2630         }
2631 
2632         ASSERT(mp->b_cont != NULL);
2633 
2634         if (error == 0)
2635                 sadb_pfkey_echo(pfkey_q, mp, (sadb_msg_t *)
2636                     mp->b_cont->b_rptr, ksi, echo_target);
2637 
2638         destroy_ipsa_pair(&ipsapp);
2639 
2640         return (error);
2641 }
2642 
2643 /*
2644  * This function takes a sadb_sa_t and finds the ipsa_t structure
2645  * and the isaf_t (hash bucket) that its stored under. If the security
2646  * association has a peer, the ipsa_t structure and bucket for that security
2647  * association are also searched for. The "pair" of ipsa_t's and isaf_t's
2648  * are returned as a ipsap_t.
2649  *
2650  * The hash buckets are returned for convenience, if the calling function
2651  * needs to use the hash bucket locks, say to remove the SA's, it should
2652  * take care to observe the convention of locking outbound bucket then
2653  * inbound bucket. The flag in_inbound_table provides direction.
2654  *
2655  * Note that a "pair" is defined as one (but not both) of the following:
2656  *
2657  * A security association which has a soft reference to another security
2658  * association via its SPI.
2659  *
2660  * A security association that is not obviously "inbound" or "outbound" so
2661  * it appears in both hash tables, the "peer" being the same security
2662  * association in the other hash table.
2663  *
2664  * This function will return NULL if the ipsa_t can't be found in the
2665  * inbound or outbound  hash tables (not found). If only one ipsa_t is
2666  * found, the pair ipsa_t will be NULL. Both isaf_t values are valid
2667  * provided at least one ipsa_t is found.
2668  */
2669 static int
2670 get_ipsa_pair(ipsa_query_t *sq, ipsap_t *ipsapp, int *diagnostic)
2671 {
2672         uint32_t pair_srcaddr[IPSA_MAX_ADDRLEN];
2673         uint32_t pair_dstaddr[IPSA_MAX_ADDRLEN];
2674         uint32_t pair_spi;
2675 
2676         init_ipsa_pair(ipsapp);
2677 
2678         ipsapp->in_inbound_table = B_FALSE;
2679 
2680         /* Lock down both buckets. */
2681         mutex_enter(&sq->outbound->isaf_lock);
2682         mutex_enter(&sq->inbound->isaf_lock);
2683 
2684         if (sq->assoc->sadb_sa_flags & IPSA_F_INBOUND) {
2685                 ipsapp->ipsap_sa_ptr = ipsec_getassocbyspi(sq->inbound,
2686                     sq->assoc->sadb_sa_spi, sq->srcaddr, sq->dstaddr, sq->af);
2687                 if (ipsapp->ipsap_sa_ptr != NULL) {
2688                         ipsapp->ipsap_bucket = sq->inbound;
2689                         ipsapp->ipsap_pbucket = sq->outbound;
2690                         ipsapp->in_inbound_table = B_TRUE;
2691                 } else {
2692                         ipsapp->ipsap_sa_ptr = ipsec_getassocbyspi(sq->outbound,
2693                             sq->assoc->sadb_sa_spi, sq->srcaddr, sq->dstaddr,
2694                             sq->af);
2695                         ipsapp->ipsap_bucket = sq->outbound;
2696                         ipsapp->ipsap_pbucket = sq->inbound;
2697                 }
2698         } else {
2699                 /* IPSA_F_OUTBOUND is set *or* no directions flags set. */
2700                 ipsapp->ipsap_sa_ptr =
2701                     ipsec_getassocbyspi(sq->outbound,
2702                     sq->assoc->sadb_sa_spi, sq->srcaddr, sq->dstaddr, sq->af);
2703                 if (ipsapp->ipsap_sa_ptr != NULL) {
2704                         ipsapp->ipsap_bucket = sq->outbound;
2705                         ipsapp->ipsap_pbucket = sq->inbound;
2706                 } else {
2707                         ipsapp->ipsap_sa_ptr = ipsec_getassocbyspi(sq->inbound,
2708                             sq->assoc->sadb_sa_spi, sq->srcaddr, sq->dstaddr,
2709                             sq->af);
2710                         ipsapp->ipsap_bucket = sq->inbound;
2711                         ipsapp->ipsap_pbucket = sq->outbound;
2712                         if (ipsapp->ipsap_sa_ptr != NULL)
2713                                 ipsapp->in_inbound_table = B_TRUE;
2714                 }
2715         }
2716 
2717         if (ipsapp->ipsap_sa_ptr == NULL) {
2718                 mutex_exit(&sq->outbound->isaf_lock);
2719                 mutex_exit(&sq->inbound->isaf_lock);
2720                 *diagnostic = SADB_X_DIAGNOSTIC_SA_NOTFOUND;
2721                 return (ESRCH);
2722         }
2723 
2724         if ((ipsapp->ipsap_sa_ptr->ipsa_state == IPSA_STATE_LARVAL) &&
2725             ipsapp->in_inbound_table) {
2726                 mutex_exit(&sq->outbound->isaf_lock);
2727                 mutex_exit(&sq->inbound->isaf_lock);
2728                 return (0);
2729         }
2730 
2731         mutex_enter(&ipsapp->ipsap_sa_ptr->ipsa_lock);
2732         if (ipsapp->ipsap_sa_ptr->ipsa_haspeer) {
2733                 /*
2734                  * haspeer implies no sa_pairing, look for same spi
2735                  * in other hashtable.
2736                  */
2737                 ipsapp->ipsap_psa_ptr =
2738                     ipsec_getassocbyspi(ipsapp->ipsap_pbucket,
2739                     sq->assoc->sadb_sa_spi, sq->srcaddr, sq->dstaddr, sq->af);
2740                 mutex_exit(&ipsapp->ipsap_sa_ptr->ipsa_lock);
2741                 mutex_exit(&sq->outbound->isaf_lock);
2742                 mutex_exit(&sq->inbound->isaf_lock);
2743                 return (0);
2744         }
2745         pair_spi = ipsapp->ipsap_sa_ptr->ipsa_otherspi;
2746         IPSA_COPY_ADDR(&pair_srcaddr,
2747             ipsapp->ipsap_sa_ptr->ipsa_srcaddr, sq->af);
2748         IPSA_COPY_ADDR(&pair_dstaddr,
2749             ipsapp->ipsap_sa_ptr->ipsa_dstaddr, sq->af);
2750         mutex_exit(&ipsapp->ipsap_sa_ptr->ipsa_lock);
2751         mutex_exit(&sq->inbound->isaf_lock);
2752         mutex_exit(&sq->outbound->isaf_lock);
2753 
2754         if (pair_spi == 0) {
2755                 ASSERT(ipsapp->ipsap_bucket != NULL);
2756                 ASSERT(ipsapp->ipsap_pbucket != NULL);
2757                 return (0);
2758         }
2759 
2760         /* found sa in outbound sadb, peer should be inbound */
2761 
2762         if (ipsapp->in_inbound_table) {
2763                 /* Found SA in inbound table, pair will be in outbound. */
2764                 if (sq->af == AF_INET6) {
2765                         ipsapp->ipsap_pbucket = OUTBOUND_BUCKET_V6(sq->sp,
2766                             *(uint32_t *)pair_srcaddr);
2767                 } else {
2768                         ipsapp->ipsap_pbucket = OUTBOUND_BUCKET_V4(sq->sp,
2769                             *(uint32_t *)pair_srcaddr);
2770                 }
2771         } else {
2772                 ipsapp->ipsap_pbucket = INBOUND_BUCKET(sq->sp, pair_spi);
2773         }
2774         mutex_enter(&ipsapp->ipsap_pbucket->isaf_lock);
2775         ipsapp->ipsap_psa_ptr = ipsec_getassocbyspi(ipsapp->ipsap_pbucket,
2776             pair_spi, pair_dstaddr, pair_srcaddr, sq->af);
2777         mutex_exit(&ipsapp->ipsap_pbucket->isaf_lock);
2778         ASSERT(ipsapp->ipsap_bucket != NULL);
2779         ASSERT(ipsapp->ipsap_pbucket != NULL);
2780         return (0);
2781 }
2782 
2783 /*
2784  * Perform NAT-traversal cached checksum offset calculations here.
2785  */
2786 static void
2787 sadb_nat_calculations(ipsa_t *newbie, sadb_address_t *natt_loc_ext,
2788     sadb_address_t *natt_rem_ext, uint32_t *src_addr_ptr,
2789     uint32_t *dst_addr_ptr)
2790 {
2791         struct sockaddr_in *natt_loc, *natt_rem;
2792         uint32_t *natt_loc_ptr = NULL, *natt_rem_ptr = NULL;
2793         uint32_t running_sum = 0;
2794 
2795 #define DOWN_SUM(x) (x) = ((x) & 0xFFFF) +   ((x) >> 16)
2796 
2797         if (natt_rem_ext != NULL) {
2798                 uint32_t l_src;
2799                 uint32_t l_rem;
2800 
2801                 natt_rem = (struct sockaddr_in *)(natt_rem_ext + 1);
2802 
2803                 /* Ensured by sadb_addrfix(). */
2804                 ASSERT(natt_rem->sin_family == AF_INET);
2805 
2806                 natt_rem_ptr = (uint32_t *)(&natt_rem->sin_addr);
2807                 newbie->ipsa_remote_nat_port = natt_rem->sin_port;
2808                 l_src = *src_addr_ptr;
2809                 l_rem = *natt_rem_ptr;
2810 
2811                 /* Instead of IPSA_COPY_ADDR(), just copy first 32 bits. */
2812                 newbie->ipsa_natt_addr_rem = *natt_rem_ptr;
2813 
2814                 l_src = ntohl(l_src);
2815                 DOWN_SUM(l_src);
2816                 DOWN_SUM(l_src);
2817                 l_rem = ntohl(l_rem);
2818                 DOWN_SUM(l_rem);
2819                 DOWN_SUM(l_rem);
2820 
2821                 /*
2822                  * We're 1's complement for checksums, so check for wraparound
2823                  * here.
2824                  */
2825                 if (l_rem > l_src)
2826                         l_src--;
2827 
2828                 running_sum += l_src - l_rem;
2829 
2830                 DOWN_SUM(running_sum);
2831                 DOWN_SUM(running_sum);
2832         }
2833 
2834         if (natt_loc_ext != NULL) {
2835                 natt_loc = (struct sockaddr_in *)(natt_loc_ext + 1);
2836 
2837                 /* Ensured by sadb_addrfix(). */
2838                 ASSERT(natt_loc->sin_family == AF_INET);
2839 
2840                 natt_loc_ptr = (uint32_t *)(&natt_loc->sin_addr);
2841                 newbie->ipsa_local_nat_port = natt_loc->sin_port;
2842 
2843                 /* Instead of IPSA_COPY_ADDR(), just copy first 32 bits. */
2844                 newbie->ipsa_natt_addr_loc = *natt_loc_ptr;
2845 
2846                 /*
2847                  * NAT-T port agility means we may have natt_loc_ext, but
2848                  * only for a local-port change.
2849                  */
2850                 if (natt_loc->sin_addr.s_addr != INADDR_ANY) {
2851                         uint32_t l_dst = ntohl(*dst_addr_ptr);
2852                         uint32_t l_loc = ntohl(*natt_loc_ptr);
2853 
2854                         DOWN_SUM(l_loc);
2855                         DOWN_SUM(l_loc);
2856                         DOWN_SUM(l_dst);
2857                         DOWN_SUM(l_dst);
2858 
2859                         /*
2860                          * We're 1's complement for checksums, so check for
2861                          * wraparound here.
2862                          */
2863                         if (l_loc > l_dst)
2864                                 l_dst--;
2865 
2866                         running_sum += l_dst - l_loc;
2867                         DOWN_SUM(running_sum);
2868                         DOWN_SUM(running_sum);
2869                 }
2870         }
2871 
2872         newbie->ipsa_inbound_cksum = running_sum;
2873 #undef DOWN_SUM
2874 }
2875 
2876 /*
2877  * This function is called from consumers that need to insert a fully-grown
2878  * security association into its tables.  This function takes into account that
2879  * SAs can be "inbound", "outbound", or "both".  The "primary" and "secondary"
2880  * hash bucket parameters are set in order of what the SA will be most of the
2881  * time.  (For example, an SA with an unspecified source, and a multicast
2882  * destination will primarily be an outbound SA.  OTOH, if that destination
2883  * is unicast for this node, then the SA will primarily be inbound.)
2884  *
2885  * It takes a lot of parameters because even if clone is B_FALSE, this needs
2886  * to check both buckets for purposes of collision.
2887  *
2888  * Return 0 upon success.  Return various errnos (ENOMEM, EEXIST) for
2889  * various error conditions.  We may need to set samsg->sadb_x_msg_diagnostic
2890  * with additional diagnostic information because there is at least one EINVAL
2891  * case here.
2892  */
2893 int
2894 sadb_common_add(queue_t *pfkey_q, mblk_t *mp, sadb_msg_t *samsg,
2895     keysock_in_t *ksi, isaf_t *primary, isaf_t *secondary,
2896     ipsa_t *newbie, boolean_t clone, boolean_t is_inbound, int *diagnostic,
2897     netstack_t *ns, sadbp_t *spp)
2898 {
2899         ipsa_t *newbie_clone = NULL, *scratch;
2900         ipsap_t ipsapp;
2901         sadb_sa_t *assoc = (sadb_sa_t *)ksi->ks_in_extv[SADB_EXT_SA];
2902         sadb_address_t *srcext =
2903             (sadb_address_t *)ksi->ks_in_extv[SADB_EXT_ADDRESS_SRC];
2904         sadb_address_t *dstext =
2905             (sadb_address_t *)ksi->ks_in_extv[SADB_EXT_ADDRESS_DST];
2906         sadb_address_t *isrcext =
2907             (sadb_address_t *)ksi->ks_in_extv[SADB_X_EXT_ADDRESS_INNER_SRC];
2908         sadb_address_t *idstext =
2909             (sadb_address_t *)ksi->ks_in_extv[SADB_X_EXT_ADDRESS_INNER_DST];
2910         sadb_x_kmc_t *kmcext =
2911             (sadb_x_kmc_t *)ksi->ks_in_extv[SADB_X_EXT_KM_COOKIE];
2912         sadb_key_t *akey = (sadb_key_t *)ksi->ks_in_extv[SADB_EXT_KEY_AUTH];
2913         sadb_key_t *ekey = (sadb_key_t *)ksi->ks_in_extv[SADB_EXT_KEY_ENCRYPT];
2914         sadb_sens_t *sens =
2915             (sadb_sens_t *)ksi->ks_in_extv[SADB_EXT_SENSITIVITY];
2916         sadb_sens_t *osens =
2917             (sadb_sens_t *)ksi->ks_in_extv[SADB_X_EXT_OUTER_SENS];
2918         sadb_x_pair_t *pair_ext =
2919             (sadb_x_pair_t *)ksi->ks_in_extv[SADB_X_EXT_PAIR];
2920         sadb_x_replay_ctr_t *replayext =
2921             (sadb_x_replay_ctr_t *)ksi->ks_in_extv[SADB_X_EXT_REPLAY_VALUE];
2922         uint8_t protocol =
2923             (samsg->sadb_msg_satype == SADB_SATYPE_AH) ? IPPROTO_AH:IPPROTO_ESP;
2924         int salt_offset;
2925         uint8_t *buf_ptr;
2926         struct sockaddr_in *src, *dst, *isrc, *idst;
2927         struct sockaddr_in6 *src6, *dst6, *isrc6, *idst6;
2928         sadb_lifetime_t *soft =
2929             (sadb_lifetime_t *)ksi->ks_in_extv[SADB_EXT_LIFETIME_SOFT];
2930         sadb_lifetime_t *hard =
2931             (sadb_lifetime_t *)ksi->ks_in_extv[SADB_EXT_LIFETIME_HARD];
2932         sadb_lifetime_t *idle =
2933             (sadb_lifetime_t *)ksi->ks_in_extv[SADB_X_EXT_LIFETIME_IDLE];
2934         sa_family_t af;
2935         int error = 0;
2936         boolean_t isupdate = (newbie != NULL);
2937         uint32_t *src_addr_ptr, *dst_addr_ptr, *isrc_addr_ptr, *idst_addr_ptr;
2938         ipsec_stack_t   *ipss = ns->netstack_ipsec;
2939         ip_stack_t      *ipst = ns->netstack_ip;
2940         ipsec_alginfo_t *alg;
2941         int             rcode;
2942         boolean_t       async = B_FALSE;
2943 
2944         init_ipsa_pair(&ipsapp);
2945 
2946         if (srcext == NULL) {
2947                 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_SRC;
2948                 return (EINVAL);
2949         }
2950         if (dstext == NULL) {
2951                 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_DST;
2952                 return (EINVAL);
2953         }
2954         if (assoc == NULL) {
2955                 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_SA;
2956                 return (EINVAL);
2957         }
2958 
2959         src = (struct sockaddr_in *)(srcext + 1);
2960         src6 = (struct sockaddr_in6 *)(srcext + 1);
2961         dst = (struct sockaddr_in *)(dstext + 1);
2962         dst6 = (struct sockaddr_in6 *)(dstext + 1);
2963         if (isrcext != NULL) {
2964                 isrc = (struct sockaddr_in *)(isrcext + 1);
2965                 isrc6 = (struct sockaddr_in6 *)(isrcext + 1);
2966                 ASSERT(idstext != NULL);
2967                 idst = (struct sockaddr_in *)(idstext + 1);
2968                 idst6 = (struct sockaddr_in6 *)(idstext + 1);
2969         } else {
2970                 isrc = NULL;
2971                 isrc6 = NULL;
2972         }
2973 
2974         af = src->sin_family;
2975 
2976         if (af == AF_INET) {
2977                 src_addr_ptr = (uint32_t *)&src->sin_addr;
2978                 dst_addr_ptr = (uint32_t *)&dst->sin_addr;
2979         } else {
2980                 ASSERT(af == AF_INET6);
2981                 src_addr_ptr = (uint32_t *)&src6->sin6_addr;
2982                 dst_addr_ptr = (uint32_t *)&dst6->sin6_addr;
2983         }
2984 
2985         if (!isupdate && (clone == B_TRUE || is_inbound == B_TRUE) &&
2986             cl_inet_checkspi &&
2987             (assoc->sadb_sa_state != SADB_X_SASTATE_ACTIVE_ELSEWHERE)) {
2988                 rcode = cl_inet_checkspi(ns->netstack_stackid, protocol,
2989                     assoc->sadb_sa_spi, NULL);
2990                 if (rcode == -1) {
2991                         return (EEXIST);
2992                 }
2993         }
2994 
2995         /*
2996          * Check to see if the new SA will be cloned AND paired. The
2997          * reason a SA will be cloned is the source or destination addresses
2998          * are not specific enough to determine if the SA goes in the outbound
2999          * or the inbound hash table, so its cloned and put in both. If
3000          * the SA is paired, it's soft linked to another SA for the other
3001          * direction. Keeping track and looking up SA's that are direction
3002          * unspecific and linked is too hard.
3003          */
3004         if (clone && (pair_ext != NULL)) {
3005                 *diagnostic = SADB_X_DIAGNOSTIC_PAIR_INAPPROPRIATE;
3006                 return (EINVAL);
3007         }
3008 
3009         if (!isupdate) {
3010                 newbie = sadb_makelarvalassoc(assoc->sadb_sa_spi,
3011                     src_addr_ptr, dst_addr_ptr, af, ns);
3012                 if (newbie == NULL)
3013                         return (ENOMEM);
3014         }
3015 
3016         mutex_enter(&newbie->ipsa_lock);
3017 
3018         if (isrc != NULL) {
3019                 if (isrc->sin_family == AF_INET) {
3020                         if (srcext->sadb_address_proto != IPPROTO_ENCAP) {
3021                                 if (srcext->sadb_address_proto != 0) {
3022                                         /*
3023                                          * Mismatched outer-packet protocol
3024                                          * and inner-packet address family.
3025                                          */
3026                                         mutex_exit(&newbie->ipsa_lock);
3027                                         error = EPROTOTYPE;
3028                                         *diagnostic =
3029                                             SADB_X_DIAGNOSTIC_INNER_AF_MISMATCH;
3030                                         goto error;
3031                                 } else {
3032                                         /* Fill in with explicit protocol. */
3033                                         srcext->sadb_address_proto =
3034                                             IPPROTO_ENCAP;
3035                                         dstext->sadb_address_proto =
3036                                             IPPROTO_ENCAP;
3037                                 }
3038                         }
3039                         isrc_addr_ptr = (uint32_t *)&isrc->sin_addr;
3040                         idst_addr_ptr = (uint32_t *)&idst->sin_addr;
3041                 } else {
3042                         ASSERT(isrc->sin_family == AF_INET6);
3043                         if (srcext->sadb_address_proto != IPPROTO_IPV6) {
3044                                 if (srcext->sadb_address_proto != 0) {
3045                                         /*
3046                                          * Mismatched outer-packet protocol
3047                                          * and inner-packet address family.
3048                                          */
3049                                         mutex_exit(&newbie->ipsa_lock);
3050                                         error = EPROTOTYPE;
3051                                         *diagnostic =
3052                                             SADB_X_DIAGNOSTIC_INNER_AF_MISMATCH;
3053                                         goto error;
3054                                 } else {
3055                                         /* Fill in with explicit protocol. */
3056                                         srcext->sadb_address_proto =
3057                                             IPPROTO_IPV6;
3058                                         dstext->sadb_address_proto =
3059                                             IPPROTO_IPV6;
3060                                 }
3061                         }
3062                         isrc_addr_ptr = (uint32_t *)&isrc6->sin6_addr;
3063                         idst_addr_ptr = (uint32_t *)&idst6->sin6_addr;
3064                 }
3065                 newbie->ipsa_innerfam = isrc->sin_family;
3066 
3067                 IPSA_COPY_ADDR(newbie->ipsa_innersrc, isrc_addr_ptr,
3068                     newbie->ipsa_innerfam);
3069                 IPSA_COPY_ADDR(newbie->ipsa_innerdst, idst_addr_ptr,
3070                     newbie->ipsa_innerfam);
3071                 newbie->ipsa_innersrcpfx = isrcext->sadb_address_prefixlen;
3072                 newbie->ipsa_innerdstpfx = idstext->sadb_address_prefixlen;
3073 
3074                 /* Unique value uses inner-ports for Tunnel Mode... */
3075                 newbie->ipsa_unique_id = SA_UNIQUE_ID(isrc->sin_port,
3076                     idst->sin_port, dstext->sadb_address_proto,
3077                     idstext->sadb_address_proto);
3078                 newbie->ipsa_unique_mask = SA_UNIQUE_MASK(isrc->sin_port,
3079                     idst->sin_port, dstext->sadb_address_proto,
3080                     idstext->sadb_address_proto);
3081         } else {
3082                 /* ... and outer-ports for Transport Mode. */
3083                 newbie->ipsa_unique_id = SA_UNIQUE_ID(src->sin_port,
3084                     dst->sin_port, dstext->sadb_address_proto, 0);
3085                 newbie->ipsa_unique_mask = SA_UNIQUE_MASK(src->sin_port,
3086                     dst->sin_port, dstext->sadb_address_proto, 0);
3087         }
3088         if (newbie->ipsa_unique_mask != (uint64_t)0)
3089                 newbie->ipsa_flags |= IPSA_F_UNIQUE;
3090 
3091         sadb_nat_calculations(newbie,
3092             (sadb_address_t *)ksi->ks_in_extv[SADB_X_EXT_ADDRESS_NATT_LOC],
3093             (sadb_address_t *)ksi->ks_in_extv[SADB_X_EXT_ADDRESS_NATT_REM],
3094             src_addr_ptr, dst_addr_ptr);
3095 
3096         newbie->ipsa_type = samsg->sadb_msg_satype;
3097 
3098         ASSERT((assoc->sadb_sa_state == SADB_SASTATE_MATURE) ||
3099             (assoc->sadb_sa_state == SADB_X_SASTATE_ACTIVE_ELSEWHERE));
3100         newbie->ipsa_auth_alg = assoc->sadb_sa_auth;
3101         newbie->ipsa_encr_alg = assoc->sadb_sa_encrypt;
3102 
3103         newbie->ipsa_flags |= assoc->sadb_sa_flags;
3104         if (newbie->ipsa_flags & SADB_X_SAFLAGS_NATT_LOC &&
3105             ksi->ks_in_extv[SADB_X_EXT_ADDRESS_NATT_LOC] == NULL) {
3106                 mutex_exit(&newbie->ipsa_lock);
3107                 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_NATT_LOC;
3108                 error = EINVAL;
3109                 goto error;
3110         }
3111         if (newbie->ipsa_flags & SADB_X_SAFLAGS_NATT_REM &&
3112             ksi->ks_in_extv[SADB_X_EXT_ADDRESS_NATT_REM] == NULL) {
3113                 mutex_exit(&newbie->ipsa_lock);
3114                 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_NATT_REM;
3115                 error = EINVAL;
3116                 goto error;
3117         }
3118         if (newbie->ipsa_flags & SADB_X_SAFLAGS_TUNNEL &&
3119             ksi->ks_in_extv[SADB_X_EXT_ADDRESS_INNER_SRC] == NULL) {
3120                 mutex_exit(&newbie->ipsa_lock);
3121                 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_INNER_SRC;
3122                 error = EINVAL;
3123                 goto error;
3124         }
3125         /*
3126          * If unspecified source address, force replay_wsize to 0.
3127          * This is because an SA that has multiple sources of secure
3128          * traffic cannot enforce a replay counter w/o synchronizing the
3129          * senders.
3130          */
3131         if (ksi->ks_in_srctype != KS_IN_ADDR_UNSPEC)
3132                 newbie->ipsa_replay_wsize = assoc->sadb_sa_replay;
3133         else
3134                 newbie->ipsa_replay_wsize = 0;
3135 
3136         newbie->ipsa_addtime = gethrestime_sec();
3137 
3138         if (kmcext != NULL) {
3139                 newbie->ipsa_kmp = kmcext->sadb_x_kmc_proto;
3140                 /* Be liberal in what we receive.  Special-case IKEv1. */
3141                 if (newbie->ipsa_kmp == SADB_X_KMP_IKE) {
3142                         /* Just in case in.iked is misbehaving... */
3143                         kmcext->sadb_x_kmc_reserved = 0;
3144                 }
3145                 newbie->ipsa_kmc = kmcext->sadb_x_kmc_cookie64;
3146         }
3147 
3148         /*
3149          * XXX CURRENT lifetime checks MAY BE needed for an UPDATE.
3150          * The spec says that one can update current lifetimes, but
3151          * that seems impractical, especially in the larval-to-mature
3152          * update that this function performs.
3153          */
3154         if (soft != NULL) {
3155                 newbie->ipsa_softaddlt = soft->sadb_lifetime_addtime;
3156                 newbie->ipsa_softuselt = soft->sadb_lifetime_usetime;
3157                 newbie->ipsa_softbyteslt = soft->sadb_lifetime_bytes;
3158                 newbie->ipsa_softalloc = soft->sadb_lifetime_allocations;
3159                 SET_EXPIRE(newbie, softaddlt, softexpiretime);
3160         }
3161         if (hard != NULL) {
3162                 newbie->ipsa_hardaddlt = hard->sadb_lifetime_addtime;
3163                 newbie->ipsa_harduselt = hard->sadb_lifetime_usetime;
3164                 newbie->ipsa_hardbyteslt = hard->sadb_lifetime_bytes;
3165                 newbie->ipsa_hardalloc = hard->sadb_lifetime_allocations;
3166                 SET_EXPIRE(newbie, hardaddlt, hardexpiretime);
3167         }
3168         if (idle != NULL) {
3169                 newbie->ipsa_idleaddlt = idle->sadb_lifetime_addtime;
3170                 newbie->ipsa_idleuselt = idle->sadb_lifetime_usetime;
3171                 newbie->ipsa_idleexpiretime = newbie->ipsa_addtime +
3172                     newbie->ipsa_idleaddlt;
3173                 newbie->ipsa_idletime = newbie->ipsa_idleaddlt;
3174         }
3175 
3176         newbie->ipsa_authtmpl = NULL;
3177         newbie->ipsa_encrtmpl = NULL;
3178 
3179 #ifdef IPSEC_LATENCY_TEST
3180         if (akey != NULL && newbie->ipsa_auth_alg != SADB_AALG_NONE) {
3181 #else
3182         if (akey != NULL) {
3183 #endif
3184                 async = (ipss->ipsec_algs_exec_mode[IPSEC_ALG_AUTH] ==
3185                     IPSEC_ALGS_EXEC_ASYNC);
3186 
3187                 newbie->ipsa_authkeybits = akey->sadb_key_bits;
3188                 newbie->ipsa_authkeylen = SADB_1TO8(akey->sadb_key_bits);
3189                 /* In case we have to round up to the next byte... */
3190                 if ((akey->sadb_key_bits & 0x7) != 0)
3191                         newbie->ipsa_authkeylen++;
3192                 newbie->ipsa_authkey = kmem_alloc(newbie->ipsa_authkeylen,
3193                     KM_NOSLEEP);
3194                 if (newbie->ipsa_authkey == NULL) {
3195                         error = ENOMEM;
3196                         mutex_exit(&newbie->ipsa_lock);
3197                         goto error;
3198                 }
3199                 bcopy(akey + 1, newbie->ipsa_authkey, newbie->ipsa_authkeylen);
3200                 bzero(akey + 1, newbie->ipsa_authkeylen);
3201 
3202                 /*
3203                  * Pre-initialize the kernel crypto framework key
3204                  * structure.
3205                  */
3206                 newbie->ipsa_kcfauthkey.ck_format = CRYPTO_KEY_RAW;
3207                 newbie->ipsa_kcfauthkey.ck_length = newbie->ipsa_authkeybits;
3208                 newbie->ipsa_kcfauthkey.ck_data = newbie->ipsa_authkey;
3209 
3210                 rw_enter(&ipss->ipsec_alg_lock, RW_READER);
3211                 alg = ipss->ipsec_alglists[IPSEC_ALG_AUTH]
3212                     [newbie->ipsa_auth_alg];
3213                 if (alg != NULL && ALG_VALID(alg)) {
3214                         newbie->ipsa_amech.cm_type = alg->alg_mech_type;
3215                         newbie->ipsa_amech.cm_param =
3216                             (char *)&newbie->ipsa_mac_len;
3217                         newbie->ipsa_amech.cm_param_len = sizeof (size_t);
3218                         newbie->ipsa_mac_len = (size_t)alg->alg_datalen;
3219                 } else {
3220                         newbie->ipsa_amech.cm_type = CRYPTO_MECHANISM_INVALID;
3221                 }
3222                 error = ipsec_create_ctx_tmpl(newbie, IPSEC_ALG_AUTH);
3223                 rw_exit(&ipss->ipsec_alg_lock);
3224                 if (error != 0) {
3225                         mutex_exit(&newbie->ipsa_lock);
3226                         /*
3227                          * An error here indicates that alg is the wrong type
3228                          * (IE: not authentication) or its not in the alg tables
3229                          * created by ipsecalgs(1m), or Kcf does not like the
3230                          * parameters passed in with this algorithm, which is
3231                          * probably a coding error!
3232                          */
3233                         *diagnostic = SADB_X_DIAGNOSTIC_BAD_CTX;
3234 
3235                         goto error;
3236                 }
3237         }
3238 
3239         if (ekey != NULL) {
3240                 rw_enter(&ipss->ipsec_alg_lock, RW_READER);
3241                 async = async || (ipss->ipsec_algs_exec_mode[IPSEC_ALG_ENCR] ==
3242                     IPSEC_ALGS_EXEC_ASYNC);
3243                 alg = ipss->ipsec_alglists[IPSEC_ALG_ENCR]
3244                     [newbie->ipsa_encr_alg];
3245 
3246                 if (alg != NULL && ALG_VALID(alg)) {
3247                         newbie->ipsa_emech.cm_type = alg->alg_mech_type;
3248                         newbie->ipsa_datalen = alg->alg_datalen;
3249                         if (alg->alg_flags & ALG_FLAG_COUNTERMODE)
3250                                 newbie->ipsa_flags |= IPSA_F_COUNTERMODE;
3251 
3252                         if (alg->alg_flags & ALG_FLAG_COMBINED) {
3253                                 newbie->ipsa_flags |= IPSA_F_COMBINED;
3254                                 newbie->ipsa_mac_len =  alg->alg_icvlen;
3255                         }
3256 
3257                         if (alg->alg_flags & ALG_FLAG_CCM)
3258                                 newbie->ipsa_noncefunc = ccm_params_init;
3259                         else if (alg->alg_flags & ALG_FLAG_GCM)
3260                                 newbie->ipsa_noncefunc = gcm_params_init;
3261                         else newbie->ipsa_noncefunc = cbc_params_init;
3262 
3263                         newbie->ipsa_saltlen = alg->alg_saltlen;
3264                         newbie->ipsa_saltbits = SADB_8TO1(newbie->ipsa_saltlen);
3265                         newbie->ipsa_iv_len = alg->alg_ivlen;
3266                         newbie->ipsa_nonce_len = newbie->ipsa_saltlen +
3267                             newbie->ipsa_iv_len;
3268                         newbie->ipsa_emech.cm_param = NULL;
3269                         newbie->ipsa_emech.cm_param_len = 0;
3270                 } else {
3271                         newbie->ipsa_emech.cm_type = CRYPTO_MECHANISM_INVALID;
3272                 }
3273                 rw_exit(&ipss->ipsec_alg_lock);
3274 
3275                 /*
3276                  * The byte stream following the sadb_key_t is made up of:
3277                  * key bytes, [salt bytes], [IV initial value]
3278                  * All of these have variable length. The IV is typically
3279                  * randomly generated by this function and not passed in.
3280                  * By supporting the injection of a known IV, the whole
3281                  * IPsec subsystem and the underlying crypto subsystem
3282                  * can be tested with known test vectors.
3283                  *
3284                  * The keying material has been checked by ext_check()
3285                  * and ipsec_valid_key_size(), after removing salt/IV
3286                  * bits, whats left is the encryption key. If this is too
3287                  * short, ipsec_create_ctx_tmpl() will fail and the SA
3288                  * won't get created.
3289                  *
3290                  * set ipsa_encrkeylen to length of key only.
3291                  */
3292                 newbie->ipsa_encrkeybits = ekey->sadb_key_bits;
3293                 newbie->ipsa_encrkeybits -= ekey->sadb_key_reserved;
3294                 newbie->ipsa_encrkeybits -= newbie->ipsa_saltbits;
3295                 newbie->ipsa_encrkeylen = SADB_1TO8(newbie->ipsa_encrkeybits);
3296 
3297                 /* In case we have to round up to the next byte... */
3298                 if ((ekey->sadb_key_bits & 0x7) != 0)
3299                         newbie->ipsa_encrkeylen++;
3300 
3301                 newbie->ipsa_encrkey = kmem_alloc(newbie->ipsa_encrkeylen,
3302                     KM_NOSLEEP);
3303                 if (newbie->ipsa_encrkey == NULL) {
3304                         error = ENOMEM;
3305                         mutex_exit(&newbie->ipsa_lock);
3306                         goto error;
3307                 }
3308 
3309                 buf_ptr = (uint8_t *)(ekey + 1);
3310                 bcopy(buf_ptr, newbie->ipsa_encrkey, newbie->ipsa_encrkeylen);
3311 
3312                 if (newbie->ipsa_flags & IPSA_F_COMBINED) {
3313                         /*
3314                          * Combined mode algs need a nonce. Copy the salt and
3315                          * IV into a buffer. The ipsa_nonce is a pointer into
3316                          * this buffer, some bytes at the start of the buffer
3317                          * may be unused, depends on the salt length. The IV
3318                          * is 64 bit aligned so it can be incremented as a
3319                          * uint64_t. Zero out key in samsg_t before freeing.
3320                          */
3321 
3322                         newbie->ipsa_nonce_buf = kmem_alloc(
3323                             sizeof (ipsec_nonce_t), KM_NOSLEEP);
3324                         if (newbie->ipsa_nonce_buf == NULL) {
3325                                 error = ENOMEM;
3326                                 mutex_exit(&newbie->ipsa_lock);
3327                                 goto error;
3328                         }
3329                         /*
3330                          * Initialize nonce and salt pointers to point
3331                          * to the nonce buffer. This is just in case we get
3332                          * bad data, the pointers will be valid, the data
3333                          * won't be.
3334                          *
3335                          * See sadb.h for layout of nonce.
3336                          */
3337                         newbie->ipsa_iv = &newbie->ipsa_nonce_buf->iv;
3338                         newbie->ipsa_salt = (uint8_t *)newbie->ipsa_nonce_buf;
3339                         newbie->ipsa_nonce = newbie->ipsa_salt;
3340                         if (newbie->ipsa_saltlen != 0) {
3341                                 salt_offset = MAXSALTSIZE -
3342                                     newbie->ipsa_saltlen;
3343                                 newbie->ipsa_salt = (uint8_t *)
3344                                     &newbie->ipsa_nonce_buf->salt[salt_offset];
3345                                 newbie->ipsa_nonce = newbie->ipsa_salt;
3346                                 buf_ptr += newbie->ipsa_encrkeylen;
3347                                 bcopy(buf_ptr, newbie->ipsa_salt,
3348                                     newbie->ipsa_saltlen);
3349                         }
3350                         /*
3351                          * The IV for CCM/GCM mode increments, it should not
3352                          * repeat. Get a random value for the IV, make a
3353                          * copy, the SA will expire when/if the IV ever
3354                          * wraps back to the initial value. If an Initial IV
3355                          * is passed in via PF_KEY, save this in the SA.
3356                          * Initialising IV for inbound is pointless as its
3357                          * taken from the inbound packet.
3358                          */
3359                         if (!is_inbound) {
3360                                 if (ekey->sadb_key_reserved != 0) {
3361                                         buf_ptr += newbie->ipsa_saltlen;
3362                                         bcopy(buf_ptr, (uint8_t *)newbie->
3363                                             ipsa_iv, SADB_1TO8(ekey->
3364                                             sadb_key_reserved));
3365                                 } else {
3366                                         (void) random_get_pseudo_bytes(
3367                                             (uint8_t *)newbie->ipsa_iv,
3368                                             newbie->ipsa_iv_len);
3369                                 }
3370                                 newbie->ipsa_iv_softexpire =
3371                                     (*newbie->ipsa_iv) << 9;
3372                                 newbie->ipsa_iv_hardexpire = *newbie->ipsa_iv;
3373                         }
3374                 }
3375                 bzero((ekey + 1), SADB_1TO8(ekey->sadb_key_bits));
3376 
3377                 /*
3378                  * Pre-initialize the kernel crypto framework key
3379                  * structure.
3380                  */
3381                 newbie->ipsa_kcfencrkey.ck_format = CRYPTO_KEY_RAW;
3382                 newbie->ipsa_kcfencrkey.ck_length = newbie->ipsa_encrkeybits;
3383                 newbie->ipsa_kcfencrkey.ck_data = newbie->ipsa_encrkey;
3384 
3385                 rw_enter(&ipss->ipsec_alg_lock, RW_READER);
3386                 error = ipsec_create_ctx_tmpl(newbie, IPSEC_ALG_ENCR);
3387                 rw_exit(&ipss->ipsec_alg_lock);
3388                 if (error != 0) {
3389                         mutex_exit(&newbie->ipsa_lock);
3390                         /* See above for error explanation. */
3391                         *diagnostic = SADB_X_DIAGNOSTIC_BAD_CTX;
3392                         goto error;
3393                 }
3394         }
3395 
3396         if (async)
3397                 newbie->ipsa_flags |= IPSA_F_ASYNC;
3398 
3399         /*
3400          * Ptrs to processing functions.
3401          */
3402         if (newbie->ipsa_type == SADB_SATYPE_ESP)
3403                 ipsecesp_init_funcs(newbie);
3404         else
3405                 ipsecah_init_funcs(newbie);
3406         ASSERT(newbie->ipsa_output_func != NULL &&
3407             newbie->ipsa_input_func != NULL);
3408 
3409         /*
3410          * Certificate ID stuff.
3411          */
3412         if (ksi->ks_in_extv[SADB_EXT_IDENTITY_SRC] != NULL) {
3413                 sadb_ident_t *id =
3414                     (sadb_ident_t *)ksi->ks_in_extv[SADB_EXT_IDENTITY_SRC];
3415 
3416                 /*
3417                  * Can assume strlen() will return okay because ext_check() in
3418                  * keysock.c prepares the string for us.
3419                  */
3420                 newbie->ipsa_src_cid = ipsid_lookup(id->sadb_ident_type,
3421                     (char *)(id+1), ns);
3422                 if (newbie->ipsa_src_cid == NULL) {
3423                         error = ENOMEM;
3424                         mutex_exit(&newbie->ipsa_lock);
3425                         goto error;
3426                 }
3427         }
3428 
3429         if (ksi->ks_in_extv[SADB_EXT_IDENTITY_DST] != NULL) {
3430                 sadb_ident_t *id =
3431                     (sadb_ident_t *)ksi->ks_in_extv[SADB_EXT_IDENTITY_DST];
3432 
3433                 /*
3434                  * Can assume strlen() will return okay because ext_check() in
3435                  * keysock.c prepares the string for us.
3436                  */
3437                 newbie->ipsa_dst_cid = ipsid_lookup(id->sadb_ident_type,
3438                     (char *)(id+1), ns);
3439                 if (newbie->ipsa_dst_cid == NULL) {
3440                         error = ENOMEM;
3441                         mutex_exit(&newbie->ipsa_lock);
3442                         goto error;
3443                 }
3444         }
3445 
3446         /*
3447          * sensitivity label handling code:
3448          * Convert sens + bitmap into cred_t, and associate it
3449          * with the new SA.
3450          */
3451         if (sens != NULL) {
3452                 uint64_t *bitmap = (uint64_t *)(sens + 1);
3453 
3454                 newbie->ipsa_tsl = sadb_label_from_sens(sens, bitmap);
3455         }
3456 
3457         /*
3458          * Likewise for outer sensitivity.
3459          */
3460         if (osens != NULL) {
3461                 uint64_t *bitmap = (uint64_t *)(osens + 1);
3462                 ts_label_t *tsl, *effective_tsl;
3463                 uint32_t *peer_addr_ptr;
3464                 zoneid_t zoneid = GLOBAL_ZONEID;
3465                 zone_t *zone;
3466 
3467                 peer_addr_ptr = is_inbound ? src_addr_ptr : dst_addr_ptr;
3468 
3469                 tsl = sadb_label_from_sens(osens, bitmap);
3470                 newbie->ipsa_mac_exempt = CONN_MAC_DEFAULT;
3471 
3472                 if (osens->sadb_x_sens_flags & SADB_X_SENS_IMPLICIT) {
3473                         newbie->ipsa_mac_exempt = CONN_MAC_IMPLICIT;
3474                 }
3475 
3476                 error = tsol_check_dest(tsl, peer_addr_ptr,
3477                     (af == AF_INET6)?IPV6_VERSION:IPV4_VERSION,
3478                     newbie->ipsa_mac_exempt, B_TRUE, &effective_tsl);
3479                 if (error != 0) {
3480                         label_rele(tsl);
3481                         mutex_exit(&newbie->ipsa_lock);
3482                         goto error;
3483                 }
3484 
3485                 if (effective_tsl != NULL) {
3486                         label_rele(tsl);
3487                         tsl = effective_tsl;
3488                 }
3489 
3490                 newbie->ipsa_otsl = tsl;
3491 
3492                 zone = zone_find_by_label(tsl);
3493                 if (zone != NULL) {
3494                         zoneid = zone->zone_id;
3495                         zone_rele(zone);
3496                 }
3497                 /*
3498                  * For exclusive stacks we set the zoneid to zero to operate
3499                  * as if in the global zone for tsol_compute_label_v4/v6
3500                  */
3501                 if (ipst->ips_netstack->netstack_stackid != GLOBAL_NETSTACKID)
3502                         zoneid = GLOBAL_ZONEID;
3503 
3504                 if (af == AF_INET6) {
3505                         error = tsol_compute_label_v6(tsl, zoneid,
3506                             (in6_addr_t *)peer_addr_ptr,
3507                             newbie->ipsa_opt_storage, ipst);
3508                 } else {
3509                         error = tsol_compute_label_v4(tsl, zoneid,
3510                             *peer_addr_ptr, newbie->ipsa_opt_storage, ipst);
3511                 }
3512                 if (error != 0) {
3513                         mutex_exit(&newbie->ipsa_lock);
3514                         goto error;
3515                 }
3516         }
3517 
3518 
3519         if (replayext != NULL) {
3520                 if ((replayext->sadb_x_rc_replay32 == 0) &&
3521                     (replayext->sadb_x_rc_replay64 != 0)) {
3522                         error = EOPNOTSUPP;
3523                         *diagnostic = SADB_X_DIAGNOSTIC_INVALID_REPLAY;
3524                         mutex_exit(&newbie->ipsa_lock);
3525                         goto error;
3526                 }
3527                 newbie->ipsa_replay = replayext->sadb_x_rc_replay32;
3528         }
3529 
3530         /* now that the SA has been updated, set its new state */
3531         newbie->ipsa_state = assoc->sadb_sa_state;
3532 
3533         if (clone) {
3534                 newbie->ipsa_haspeer = B_TRUE;
3535         } else {
3536                 if (!is_inbound) {
3537                         lifetime_fuzz(newbie);
3538                 }
3539         }
3540         /*
3541          * The less locks I hold when doing an insertion and possible cloning,
3542          * the better!
3543          */
3544         mutex_exit(&newbie->ipsa_lock);
3545 
3546         if (clone) {
3547                 newbie_clone = sadb_cloneassoc(newbie);
3548 
3549                 if (newbie_clone == NULL) {
3550                         error = ENOMEM;
3551                         goto error;
3552                 }
3553         }
3554 
3555         /*
3556          * Enter the bucket locks.  The order of entry is outbound,
3557          * inbound.  We map "primary" and "secondary" into outbound and inbound
3558          * based on the destination address type.  If the destination address
3559          * type is for a node that isn't mine (or potentially mine), the
3560          * "primary" bucket is the outbound one.
3561          */
3562         if (!is_inbound) {
3563                 /* primary == outbound */
3564                 mutex_enter(&primary->isaf_lock);
3565                 mutex_enter(&secondary->isaf_lock);
3566         } else {
3567                 /* primary == inbound */
3568                 mutex_enter(&secondary->isaf_lock);
3569                 mutex_enter(&primary->isaf_lock);
3570         }
3571 
3572         /*
3573          * sadb_insertassoc() doesn't increment the reference
3574          * count.  We therefore have to increment the
3575          * reference count one more time to reflect the
3576          * pointers of the table that reference this SA.
3577          */
3578         IPSA_REFHOLD(newbie);
3579 
3580         if (isupdate) {
3581                 /*
3582                  * Unlink from larval holding cell in the "inbound" fanout.
3583                  */
3584                 ASSERT(newbie->ipsa_linklock == &primary->isaf_lock ||
3585                     newbie->ipsa_linklock == &secondary->isaf_lock);
3586                 sadb_unlinkassoc(newbie);
3587         }
3588 
3589         mutex_enter(&newbie->ipsa_lock);
3590         error = sadb_insertassoc(newbie, primary);
3591         mutex_exit(&newbie->ipsa_lock);
3592 
3593         if (error != 0) {
3594                 /*
3595                  * Since sadb_insertassoc() failed, we must decrement the
3596                  * refcount again so the cleanup code will actually free
3597                  * the offending SA.
3598                  */
3599                 IPSA_REFRELE(newbie);
3600                 goto error_unlock;
3601         }
3602 
3603         if (newbie_clone != NULL) {
3604                 mutex_enter(&newbie_clone->ipsa_lock);
3605                 error = sadb_insertassoc(newbie_clone, secondary);
3606                 mutex_exit(&newbie_clone->ipsa_lock);
3607                 if (error != 0) {
3608                         /* Collision in secondary table. */
3609                         sadb_unlinkassoc(newbie);  /* This does REFRELE. */
3610                         goto error_unlock;
3611                 }
3612                 IPSA_REFHOLD(newbie_clone);
3613         } else {
3614                 ASSERT(primary != secondary);
3615                 scratch = ipsec_getassocbyspi(secondary, newbie->ipsa_spi,
3616                     ALL_ZEROES_PTR, newbie->ipsa_dstaddr, af);
3617                 if (scratch != NULL) {
3618                         /* Collision in secondary table. */
3619                         sadb_unlinkassoc(newbie);  /* This does REFRELE. */
3620                         /* Set the error, since ipsec_getassocbyspi() can't. */
3621                         error = EEXIST;
3622                         goto error_unlock;
3623                 }
3624         }
3625 
3626         /* OKAY!  So let's do some reality check assertions. */
3627 
3628         ASSERT(MUTEX_NOT_HELD(&newbie->ipsa_lock));
3629         ASSERT(newbie_clone == NULL ||
3630             (MUTEX_NOT_HELD(&newbie_clone->ipsa_lock)));
3631 
3632 error_unlock:
3633 
3634         /*
3635          * We can exit the locks in any order.  Only entrance needs to
3636          * follow any protocol.
3637          */
3638         mutex_exit(&secondary->isaf_lock);
3639         mutex_exit(&primary->isaf_lock);
3640 
3641         if (pair_ext != NULL && error == 0) {
3642                 /* update pair_spi if it exists. */
3643                 ipsa_query_t sq;
3644 
3645                 sq.spp = spp;           /* XXX param */
3646                 error = sadb_form_query(ksi, IPSA_Q_DST, IPSA_Q_SRC|IPSA_Q_DST|
3647                     IPSA_Q_SA|IPSA_Q_INBOUND|IPSA_Q_OUTBOUND, &sq, diagnostic);
3648                 if (error)
3649                         return (error);
3650 
3651                 error = get_ipsa_pair(&sq, &ipsapp, diagnostic);
3652 
3653                 if (error != 0)
3654                         goto error;
3655 
3656                 if (ipsapp.ipsap_psa_ptr != NULL) {
3657                         *diagnostic = SADB_X_DIAGNOSTIC_PAIR_ALREADY;
3658                         error = EINVAL;
3659                 } else {
3660                         /* update_pairing() sets diagnostic */
3661                         error = update_pairing(&ipsapp, &sq, ksi, diagnostic);
3662                 }
3663         }
3664         /* Common error point for this routine. */
3665 error:
3666         if (newbie != NULL) {
3667                 if (error != 0) {
3668                         /* This SA is broken, let the reaper clean up. */
3669                         mutex_enter(&newbie->ipsa_lock);
3670                         newbie->ipsa_state = IPSA_STATE_DEAD;
3671                         newbie->ipsa_hardexpiretime = 1;
3672                         mutex_exit(&newbie->ipsa_lock);
3673                 }
3674                 IPSA_REFRELE(newbie);
3675         }
3676         if (newbie_clone != NULL) {
3677                 IPSA_REFRELE(newbie_clone);
3678         }
3679 
3680         if (error == 0) {
3681                 /*
3682                  * Construct favorable PF_KEY return message and send to
3683                  * keysock. Update the flags in the original keysock message
3684                  * to reflect the actual flags in the new SA.
3685                  *  (Q:  Do I need to pass "newbie"?  If I do,
3686                  * make sure to REFHOLD, call, then REFRELE.)
3687                  */
3688                 assoc->sadb_sa_flags = newbie->ipsa_flags;
3689                 sadb_pfkey_echo(pfkey_q, mp, samsg, ksi, NULL);
3690         }
3691 
3692         destroy_ipsa_pair(&ipsapp);
3693         return (error);
3694 }
3695 
3696 /*
3697  * Set the time of first use for a security association.  Update any
3698  * expiration times as a result.
3699  */
3700 void
3701 sadb_set_usetime(ipsa_t *assoc)
3702 {
3703         time_t snapshot = gethrestime_sec();
3704 
3705         mutex_enter(&assoc->ipsa_lock);
3706         assoc->ipsa_lastuse = snapshot;
3707         assoc->ipsa_idleexpiretime = snapshot + assoc->ipsa_idletime;
3708 
3709         /*
3710          * Caller does check usetime before calling me usually, and
3711          * double-checking is better than a mutex_enter/exit hit.
3712          */
3713         if (assoc->ipsa_usetime == 0) {
3714                 /*
3715                  * This is redundant for outbound SA's, as
3716                  * ipsec_getassocbyconn() sets the IPSA_F_USED flag already.
3717                  * Inbound SAs, however, have no such protection.
3718                  */
3719                 assoc->ipsa_flags |= IPSA_F_USED;
3720                 assoc->ipsa_usetime = snapshot;
3721 
3722                 /*
3723                  * After setting the use time, see if we have a use lifetime
3724                  * that would cause the actual SA expiration time to shorten.
3725                  */
3726                 UPDATE_EXPIRE(assoc, softuselt, softexpiretime);
3727                 UPDATE_EXPIRE(assoc, harduselt, hardexpiretime);
3728         }
3729         mutex_exit(&assoc->ipsa_lock);
3730 }
3731 
3732 /*
3733  * Send up a PF_KEY expire message for this association.
3734  */
3735 static void
3736 sadb_expire_assoc(queue_t *pfkey_q, ipsa_t *assoc)
3737 {
3738         mblk_t *mp, *mp1;
3739         int alloclen, af;
3740         sadb_msg_t *samsg;
3741         sadb_lifetime_t *current, *expire;
3742         sadb_sa_t *saext;
3743         uint8_t *end;
3744         boolean_t tunnel_mode;
3745 
3746         ASSERT(MUTEX_HELD(&assoc->ipsa_lock));
3747 
3748         /* Don't bother sending if there's no queue. */
3749         if (pfkey_q == NULL)
3750                 return;
3751 
3752         mp = sadb_keysock_out(0);
3753         if (mp == NULL) {
3754                 /* cmn_err(CE_WARN, */
3755                 /*      "sadb_expire_assoc: Can't allocate KEYSOCK_OUT.\n"); */
3756                 return;
3757         }
3758 
3759         alloclen = sizeof (*samsg) + sizeof (*current) + sizeof (*expire) +
3760             2 * sizeof (sadb_address_t) + sizeof (*saext);
3761 
3762         af = assoc->ipsa_addrfam;
3763         switch (af) {
3764         case AF_INET:
3765                 alloclen += 2 * sizeof (struct sockaddr_in);
3766                 break;
3767         case AF_INET6:
3768                 alloclen += 2 * sizeof (struct sockaddr_in6);
3769                 break;
3770         default:
3771                 /* Won't happen unless there's a kernel bug. */
3772                 freeb(mp);
3773                 cmn_err(CE_WARN,
3774                     "sadb_expire_assoc: Unknown address length.\n");
3775                 return;
3776         }
3777 
3778         tunnel_mode = (assoc->ipsa_flags & IPSA_F_TUNNEL);
3779         if (tunnel_mode) {
3780                 alloclen += 2 * sizeof (sadb_address_t);
3781                 switch (assoc->ipsa_innerfam) {
3782                 case AF_INET:
3783                         alloclen += 2 * sizeof (struct sockaddr_in);
3784                         break;
3785                 case AF_INET6:
3786                         alloclen += 2 * sizeof (struct sockaddr_in6);
3787                         break;
3788                 default:
3789                         /* Won't happen unless there's a kernel bug. */
3790                         freeb(mp);
3791                         cmn_err(CE_WARN, "sadb_expire_assoc: "
3792                             "Unknown inner address length.\n");
3793                         return;
3794                 }
3795         }
3796 
3797         mp->b_cont = allocb(alloclen, BPRI_HI);
3798         if (mp->b_cont == NULL) {
3799                 freeb(mp);
3800                 /* cmn_err(CE_WARN, */
3801                 /*      "sadb_expire_assoc: Can't allocate message.\n"); */
3802                 return;
3803         }
3804 
3805         mp1 = mp;
3806         mp = mp->b_cont;
3807         end = mp->b_wptr + alloclen;
3808 
3809         samsg = (sadb_msg_t *)mp->b_wptr;
3810         mp->b_wptr += sizeof (*samsg);
3811         samsg->sadb_msg_version = PF_KEY_V2;
3812         samsg->sadb_msg_type = SADB_EXPIRE;
3813         samsg->sadb_msg_errno = 0;
3814         samsg->sadb_msg_satype = assoc->ipsa_type;
3815         samsg->sadb_msg_len = SADB_8TO64(alloclen);
3816         samsg->sadb_msg_reserved = 0;
3817         samsg->sadb_msg_seq = 0;
3818         samsg->sadb_msg_pid = 0;
3819 
3820         saext = (sadb_sa_t *)mp->b_wptr;
3821         mp->b_wptr += sizeof (*saext);
3822         saext->sadb_sa_len = SADB_8TO64(sizeof (*saext));
3823         saext->sadb_sa_exttype = SADB_EXT_SA;
3824         saext->sadb_sa_spi = assoc->ipsa_spi;
3825         saext->sadb_sa_replay = assoc->ipsa_replay_wsize;
3826         saext->sadb_sa_state = assoc->ipsa_state;
3827         saext->sadb_sa_auth = assoc->ipsa_auth_alg;
3828         saext->sadb_sa_encrypt = assoc->ipsa_encr_alg;
3829         saext->sadb_sa_flags = assoc->ipsa_flags;
3830 
3831         current = (sadb_lifetime_t *)mp->b_wptr;
3832         mp->b_wptr += sizeof (sadb_lifetime_t);
3833         current->sadb_lifetime_len = SADB_8TO64(sizeof (*current));
3834         current->sadb_lifetime_exttype = SADB_EXT_LIFETIME_CURRENT;
3835         /* We do not support the concept. */
3836         current->sadb_lifetime_allocations = 0;
3837         current->sadb_lifetime_bytes = assoc->ipsa_bytes;
3838         current->sadb_lifetime_addtime = assoc->ipsa_addtime;
3839         current->sadb_lifetime_usetime = assoc->ipsa_usetime;
3840 
3841         expire = (sadb_lifetime_t *)mp->b_wptr;
3842         mp->b_wptr += sizeof (*expire);
3843         expire->sadb_lifetime_len = SADB_8TO64(sizeof (*expire));
3844 
3845         if (assoc->ipsa_state == IPSA_STATE_DEAD) {
3846                 expire->sadb_lifetime_exttype = SADB_EXT_LIFETIME_HARD;
3847                 expire->sadb_lifetime_allocations = assoc->ipsa_hardalloc;
3848                 expire->sadb_lifetime_bytes = assoc->ipsa_hardbyteslt;
3849                 expire->sadb_lifetime_addtime = assoc->ipsa_hardaddlt;
3850                 expire->sadb_lifetime_usetime = assoc->ipsa_harduselt;
3851         } else if (assoc->ipsa_state == IPSA_STATE_DYING) {
3852                 expire->sadb_lifetime_exttype = SADB_EXT_LIFETIME_SOFT;
3853                 expire->sadb_lifetime_allocations = assoc->ipsa_softalloc;
3854                 expire->sadb_lifetime_bytes = assoc->ipsa_softbyteslt;
3855                 expire->sadb_lifetime_addtime = assoc->ipsa_softaddlt;
3856                 expire->sadb_lifetime_usetime = assoc->ipsa_softuselt;
3857         } else {
3858                 ASSERT(assoc->ipsa_state == IPSA_STATE_MATURE);
3859                 expire->sadb_lifetime_exttype = SADB_X_EXT_LIFETIME_IDLE;
3860                 expire->sadb_lifetime_allocations = 0;
3861                 expire->sadb_lifetime_bytes = 0;
3862                 expire->sadb_lifetime_addtime = assoc->ipsa_idleaddlt;
3863                 expire->sadb_lifetime_usetime = assoc->ipsa_idleuselt;
3864         }
3865 
3866         mp->b_wptr = sadb_make_addr_ext(mp->b_wptr, end, SADB_EXT_ADDRESS_SRC,
3867             af, assoc->ipsa_srcaddr, tunnel_mode ? 0 : SA_SRCPORT(assoc),
3868             SA_PROTO(assoc), 0);
3869         ASSERT(mp->b_wptr != NULL);
3870 
3871         mp->b_wptr = sadb_make_addr_ext(mp->b_wptr, end, SADB_EXT_ADDRESS_DST,
3872             af, assoc->ipsa_dstaddr, tunnel_mode ? 0 : SA_DSTPORT(assoc),
3873             SA_PROTO(assoc), 0);
3874         ASSERT(mp->b_wptr != NULL);
3875 
3876         if (tunnel_mode) {
3877                 mp->b_wptr = sadb_make_addr_ext(mp->b_wptr, end,
3878                     SADB_X_EXT_ADDRESS_INNER_SRC, assoc->ipsa_innerfam,
3879                     assoc->ipsa_innersrc, SA_SRCPORT(assoc), SA_IPROTO(assoc),
3880                     assoc->ipsa_innersrcpfx);
3881                 ASSERT(mp->b_wptr != NULL);
3882                 mp->b_wptr = sadb_make_addr_ext(mp->b_wptr, end,
3883                     SADB_X_EXT_ADDRESS_INNER_DST, assoc->ipsa_innerfam,
3884                     assoc->ipsa_innerdst, SA_DSTPORT(assoc), SA_IPROTO(assoc),
3885                     assoc->ipsa_innerdstpfx);
3886                 ASSERT(mp->b_wptr != NULL);
3887         }
3888 
3889         /* Can just putnext, we're ready to go! */
3890         putnext(pfkey_q, mp1);
3891 }
3892 
3893 /*
3894  * "Age" the SA with the number of bytes that was used to protect traffic.
3895  * Send an SADB_EXPIRE message if appropriate.  Return B_TRUE if there was
3896  * enough "charge" left in the SA to protect the data.  Return B_FALSE
3897  * otherwise.  (If B_FALSE is returned, the association either was, or became
3898  * DEAD.)
3899  */
3900 boolean_t
3901 sadb_age_bytes(queue_t *pfkey_q, ipsa_t *assoc, uint64_t bytes,
3902     boolean_t sendmsg)
3903 {
3904         boolean_t rc = B_TRUE;
3905         uint64_t newtotal;
3906 
3907         mutex_enter(&assoc->ipsa_lock);
3908         newtotal = assoc->ipsa_bytes + bytes;
3909         if (assoc->ipsa_hardbyteslt != 0 &&
3910             newtotal >= assoc->ipsa_hardbyteslt) {
3911                 if (assoc->ipsa_state != IPSA_STATE_DEAD) {
3912                         sadb_delete_cluster(assoc);
3913                         /*
3914                          * Send EXPIRE message to PF_KEY.  May wish to pawn
3915                          * this off on another non-interrupt thread.  Also
3916                          * unlink this SA immediately.
3917                          */
3918                         assoc->ipsa_state = IPSA_STATE_DEAD;
3919                         if (sendmsg)
3920                                 sadb_expire_assoc(pfkey_q, assoc);
3921                         /*
3922                          * Set non-zero expiration time so sadb_age_assoc()
3923                          * will work when reaping.
3924                          */
3925                         assoc->ipsa_hardexpiretime = (time_t)1;
3926                 } /* Else someone beat me to it! */
3927                 rc = B_FALSE;
3928         } else if (assoc->ipsa_softbyteslt != 0 &&
3929             (newtotal >= assoc->ipsa_softbyteslt)) {
3930                 if (assoc->ipsa_state < IPSA_STATE_DYING) {
3931                         /*
3932                          * Send EXPIRE message to PF_KEY.  May wish to pawn
3933                          * this off on another non-interrupt thread.
3934                          */
3935                         assoc->ipsa_state = IPSA_STATE_DYING;
3936                         assoc->ipsa_bytes = newtotal;
3937                         if (sendmsg)
3938                                 sadb_expire_assoc(pfkey_q, assoc);
3939                 } /* Else someone beat me to it! */
3940         }
3941         if (rc == B_TRUE)
3942                 assoc->ipsa_bytes = newtotal;
3943         mutex_exit(&assoc->ipsa_lock);
3944         return (rc);
3945 }
3946 
3947 /*
3948  * "Torch" an individual SA.  Returns NULL, so it can be tail-called from
3949  *     sadb_age_assoc().
3950  */
3951 static ipsa_t *
3952 sadb_torch_assoc(isaf_t *head, ipsa_t *sa)
3953 {
3954         ASSERT(MUTEX_HELD(&head->isaf_lock));
3955         ASSERT(MUTEX_HELD(&sa->ipsa_lock));
3956         ASSERT(sa->ipsa_state == IPSA_STATE_DEAD);
3957 
3958         /*
3959          * Force cached SAs to be revalidated..
3960          */
3961         head->isaf_gen++;
3962 
3963         mutex_exit(&sa->ipsa_lock);
3964         sadb_unlinkassoc(sa);
3965 
3966         return (NULL);
3967 }
3968 
3969 /*
3970  * Do various SA-is-idle activities depending on delta (the number of idle
3971  * seconds on the SA) and/or other properties of the SA.
3972  *
3973  * Return B_TRUE if I've sent a packet, because I have to drop the
3974  * association's mutex before sending a packet out the wire.
3975  */
3976 /* ARGSUSED */
3977 static boolean_t
3978 sadb_idle_activities(ipsa_t *assoc, time_t delta, boolean_t inbound)
3979 {
3980         ipsecesp_stack_t *espstack = assoc->ipsa_netstack->netstack_ipsecesp;
3981         int nat_t_interval = espstack->ipsecesp_nat_keepalive_interval;
3982 
3983         ASSERT(MUTEX_HELD(&assoc->ipsa_lock));
3984 
3985         if (!inbound && (assoc->ipsa_flags & IPSA_F_NATT_LOC) &&
3986             delta >= nat_t_interval &&
3987             gethrestime_sec() - assoc->ipsa_last_nat_t_ka >= nat_t_interval) {
3988                 ASSERT(assoc->ipsa_type == SADB_SATYPE_ESP);
3989                 assoc->ipsa_last_nat_t_ka = gethrestime_sec();
3990                 mutex_exit(&assoc->ipsa_lock);
3991                 ipsecesp_send_keepalive(assoc);
3992                 return (B_TRUE);
3993         }
3994         return (B_FALSE);
3995 }
3996 
3997 /*
3998  * Return "assoc" if haspeer is true and I send an expire.  This allows
3999  * the consumers' aging functions to tidy up an expired SA's peer.
4000  */
4001 static ipsa_t *
4002 sadb_age_assoc(isaf_t *head, queue_t *pfkey_q, ipsa_t *assoc,
4003     time_t current, int reap_delay, boolean_t inbound)
4004 {
4005         ipsa_t *retval = NULL;
4006         boolean_t dropped_mutex = B_FALSE;
4007 
4008         ASSERT(MUTEX_HELD(&head->isaf_lock));
4009 
4010         mutex_enter(&assoc->ipsa_lock);
4011 
4012         if (((assoc->ipsa_state == IPSA_STATE_LARVAL) ||
4013             ((assoc->ipsa_state == IPSA_STATE_IDLE) ||
4014             (assoc->ipsa_state == IPSA_STATE_ACTIVE_ELSEWHERE) &&
4015             (assoc->ipsa_hardexpiretime != 0))) &&
4016             (assoc->ipsa_hardexpiretime <= current)) {
4017                 assoc->ipsa_state = IPSA_STATE_DEAD;
4018                 return (sadb_torch_assoc(head, assoc));
4019         }
4020 
4021         /*
4022          * Check lifetimes.  Fortunately, SA setup is done
4023          * such that there are only two times to look at,
4024          * softexpiretime, and hardexpiretime.
4025          *
4026          * Check hard first.
4027          */
4028 
4029         if (assoc->ipsa_hardexpiretime != 0 &&
4030             assoc->ipsa_hardexpiretime <= current) {
4031                 if (assoc->ipsa_state == IPSA_STATE_DEAD)
4032                         return (sadb_torch_assoc(head, assoc));
4033 
4034                 if (inbound) {
4035                         sadb_delete_cluster(assoc);
4036                 }
4037 
4038                 /*
4039                  * Send SADB_EXPIRE with hard lifetime, delay for unlinking.
4040                  */
4041                 assoc->ipsa_state = IPSA_STATE_DEAD;
4042                 if (assoc->ipsa_haspeer || assoc->ipsa_otherspi != 0) {
4043                         /*
4044                          * If the SA is paired or peered with another, put
4045                          * a copy on a list which can be processed later, the
4046                          * pair/peer SA needs to be updated so the both die
4047                          * at the same time.
4048                          *
4049                          * If I return assoc, I have to bump up its reference
4050                          * count to keep with the ipsa_t reference count
4051                          * semantics.
4052                          */
4053                         IPSA_REFHOLD(assoc);
4054                         retval = assoc;
4055                 }
4056                 sadb_expire_assoc(pfkey_q, assoc);
4057                 assoc->ipsa_hardexpiretime = current + reap_delay;
4058         } else if (assoc->ipsa_softexpiretime != 0 &&
4059             assoc->ipsa_softexpiretime <= current &&
4060             assoc->ipsa_state < IPSA_STATE_DYING) {
4061                 /*
4062                  * Send EXPIRE message to PF_KEY.  May wish to pawn
4063                  * this off on another non-interrupt thread.
4064                  */
4065                 assoc->ipsa_state = IPSA_STATE_DYING;
4066                 if (assoc->ipsa_haspeer) {
4067                         /*
4068                          * If the SA has a peer, update the peer's state
4069                          * on SOFT_EXPIRE, this is mostly to prevent two
4070                          * expire messages from effectively the same SA.
4071                          *
4072                          * Don't care about paired SA's, then can (and should)
4073                          * be able to soft expire at different times.
4074                          *
4075                          * If I return assoc, I have to bump up its
4076                          * reference count to keep with the ipsa_t reference
4077                          * count semantics.
4078                          */
4079                         IPSA_REFHOLD(assoc);
4080                         retval = assoc;
4081                 }
4082                 sadb_expire_assoc(pfkey_q, assoc);
4083         } else if (assoc->ipsa_idletime != 0 &&
4084             assoc->ipsa_idleexpiretime <= current) {
4085                 if (assoc->ipsa_state == IPSA_STATE_ACTIVE_ELSEWHERE) {
4086                         assoc->ipsa_state = IPSA_STATE_IDLE;
4087                 }
4088 
4089                 /*
4090                  * Need to handle Mature case
4091                  */
4092                 if (assoc->ipsa_state == IPSA_STATE_MATURE) {
4093                         sadb_expire_assoc(pfkey_q, assoc);
4094                 }
4095         } else {
4096                 /* Check idle time activities. */
4097                 dropped_mutex = sadb_idle_activities(assoc,
4098                     current - assoc->ipsa_lastuse, inbound);
4099         }
4100 
4101         if (!dropped_mutex)
4102                 mutex_exit(&assoc->ipsa_lock);
4103         return (retval);
4104 }
4105 
4106 /*
4107  * Called by a consumer protocol to do ther dirty work of reaping dead
4108  * Security Associations.
4109  *
4110  * NOTE: sadb_age_assoc() marks expired SA's as DEAD but only removed
4111  * SA's that are already marked DEAD, so expired SA's are only reaped
4112  * the second time sadb_ager() runs.
4113  */
4114 void
4115 sadb_ager(sadb_t *sp, queue_t *pfkey_q, int reap_delay, netstack_t *ns)
4116 {
4117         int i;
4118         isaf_t *bucket;
4119         ipsa_t *assoc, *spare;
4120         iacqf_t *acqlist;
4121         ipsacq_t *acqrec, *spareacq;
4122         templist_t *haspeerlist, *newbie;
4123         /* Snapshot current time now. */
4124         time_t current = gethrestime_sec();
4125         haspeerlist = NULL;
4126 
4127         /*
4128          * Do my dirty work.  This includes aging real entries, aging
4129          * larvals, and aging outstanding ACQUIREs.
4130          *
4131          * I hope I don't tie up resources for too long.
4132          */
4133 
4134         /* Age acquires. */
4135 
4136         for (i = 0; i < sp->sdb_hashsize; i++) {
4137                 acqlist = &sp->sdb_acq[i];
4138                 mutex_enter(&acqlist->iacqf_lock);
4139                 for (acqrec = acqlist->iacqf_ipsacq; acqrec != NULL;
4140                     acqrec = spareacq) {
4141                         spareacq = acqrec->ipsacq_next;
4142                         if (current > acqrec->ipsacq_expire)
4143                                 sadb_destroy_acquire(acqrec, ns);
4144                 }
4145                 mutex_exit(&acqlist->iacqf_lock);
4146         }
4147 
4148         /* Age inbound associations. */
4149         for (i = 0; i < sp->sdb_hashsize; i++) {
4150                 bucket = &(sp->sdb_if[i]);
4151                 mutex_enter(&bucket->isaf_lock);
4152                 for (assoc = bucket->isaf_ipsa; assoc != NULL;
4153                     assoc = spare) {
4154                         spare = assoc->ipsa_next;
4155                         if (sadb_age_assoc(bucket, pfkey_q, assoc, current,
4156                             reap_delay, B_TRUE) != NULL) {
4157                                 /*
4158                                  * Put SA's which have a peer or SA's which
4159                                  * are paired on a list for processing after
4160                                  * all the hash tables have been walked.
4161                                  *
4162                                  * sadb_age_assoc() increments the refcnt,
4163                                  * effectively doing an IPSA_REFHOLD().
4164                                  */
4165                                 newbie = kmem_alloc(sizeof (*newbie),
4166                                     KM_NOSLEEP);
4167                                 if (newbie == NULL) {
4168                                         /*
4169                                          * Don't forget to REFRELE().
4170                                          */
4171                                         IPSA_REFRELE(assoc);
4172                                         continue;       /* for loop... */
4173                                 }
4174                                 newbie->next = haspeerlist;
4175                                 newbie->ipsa = assoc;
4176                                 haspeerlist = newbie;
4177                         }
4178                 }
4179                 mutex_exit(&bucket->isaf_lock);
4180         }
4181 
4182         age_pair_peer_list(haspeerlist, sp, B_FALSE);
4183         haspeerlist = NULL;
4184 
4185         /* Age outbound associations. */
4186         for (i = 0; i < sp->sdb_hashsize; i++) {
4187                 bucket = &(sp->sdb_of[i]);
4188                 mutex_enter(&bucket->isaf_lock);
4189                 for (assoc = bucket->isaf_ipsa; assoc != NULL;
4190                     assoc = spare) {
4191                         spare = assoc->ipsa_next;
4192                         if (sadb_age_assoc(bucket, pfkey_q, assoc, current,
4193                             reap_delay, B_FALSE) != NULL) {
4194                                 /*
4195                                  * sadb_age_assoc() increments the refcnt,
4196                                  * effectively doing an IPSA_REFHOLD().
4197                                  */
4198                                 newbie = kmem_alloc(sizeof (*newbie),
4199                                     KM_NOSLEEP);
4200                                 if (newbie == NULL) {
4201                                         /*
4202                                          * Don't forget to REFRELE().
4203                                          */
4204                                         IPSA_REFRELE(assoc);
4205                                         continue;       /* for loop... */
4206                                 }
4207                                 newbie->next = haspeerlist;
4208                                 newbie->ipsa = assoc;
4209                                 haspeerlist = newbie;
4210                         }
4211                 }
4212                 mutex_exit(&bucket->isaf_lock);
4213         }
4214 
4215         age_pair_peer_list(haspeerlist, sp, B_TRUE);
4216 
4217         /*
4218          * Run a GC pass to clean out dead identities.
4219          */
4220         ipsid_gc(ns);
4221 }
4222 
4223 /*
4224  * Figure out when to reschedule the ager.
4225  */
4226 timeout_id_t
4227 sadb_retimeout(hrtime_t begin, queue_t *pfkey_q, void (*ager)(void *),
4228     void *agerarg, uint_t *intp, uint_t intmax, short mid)
4229 {
4230         hrtime_t end = gethrtime();
4231         uint_t interval = *intp;        /* "interval" is in ms. */
4232 
4233         /*
4234          * See how long this took.  If it took too long, increase the
4235          * aging interval.
4236          */
4237         if ((end - begin) > MSEC2NSEC(interval)) {
4238                 if (interval >= intmax) {
4239                         /* XXX Rate limit this?  Or recommend flush? */
4240                         (void) strlog(mid, 0, 0, SL_ERROR | SL_WARN,
4241                             "Too many SA's to age out in %d msec.\n",
4242                             intmax);
4243                 } else {
4244                         /* Double by shifting by one bit. */
4245                         interval <<= 1;
4246                         interval = min(interval, intmax);
4247                 }
4248         } else if ((end - begin) <= (MSEC2NSEC(interval) / 2) &&
4249             interval > SADB_AGE_INTERVAL_DEFAULT) {
4250                 /*
4251                  * If I took less than half of the interval, then I should
4252                  * ratchet the interval back down.  Never automatically
4253                  * shift below the default aging interval.
4254                  *
4255                  * NOTE:This even overrides manual setting of the age
4256                  *      interval using NDD to lower the setting past the
4257                  *      default.  In other words, if you set the interval
4258                  *      lower than the default, and your SADB gets too big,
4259                  *      the interval will only self-lower back to the default.
4260                  */
4261                 /* Halve by shifting one bit. */
4262                 interval >>= 1;
4263                 interval = max(interval, SADB_AGE_INTERVAL_DEFAULT);
4264         }
4265         *intp = interval;
4266         return (qtimeout(pfkey_q, ager, agerarg,
4267             drv_usectohz(interval * (MICROSEC / MILLISEC))));
4268 }
4269 
4270 
4271 /*
4272  * Update the lifetime values of an SA.  This is the path an SADB_UPDATE
4273  * message takes when updating a MATURE or DYING SA.
4274  */
4275 static void
4276 sadb_update_lifetimes(ipsa_t *assoc, sadb_lifetime_t *hard,
4277     sadb_lifetime_t *soft, sadb_lifetime_t *idle, boolean_t outbound)
4278 {
4279         mutex_enter(&assoc->ipsa_lock);
4280 
4281         /*
4282          * XXX RFC 2367 mentions how an SADB_EXT_LIFETIME_CURRENT can be
4283          * passed in during an update message.  We currently don't handle
4284          * these.
4285          */
4286 
4287         if (hard != NULL) {
4288                 if (hard->sadb_lifetime_bytes != 0)
4289                         assoc->ipsa_hardbyteslt = hard->sadb_lifetime_bytes;
4290                 if (hard->sadb_lifetime_usetime != 0)
4291                         assoc->ipsa_harduselt = hard->sadb_lifetime_usetime;
4292                 if (hard->sadb_lifetime_addtime != 0)
4293                         assoc->ipsa_hardaddlt = hard->sadb_lifetime_addtime;
4294                 if (assoc->ipsa_hardaddlt != 0) {
4295                         assoc->ipsa_hardexpiretime =
4296                             assoc->ipsa_addtime + assoc->ipsa_hardaddlt;
4297                 }
4298                 if (assoc->ipsa_harduselt != 0 &&
4299                     assoc->ipsa_flags & IPSA_F_USED) {
4300                         UPDATE_EXPIRE(assoc, harduselt, hardexpiretime);
4301                 }
4302                 if (hard->sadb_lifetime_allocations != 0)
4303                         assoc->ipsa_hardalloc = hard->sadb_lifetime_allocations;
4304         }
4305 
4306         if (soft != NULL) {
4307                 if (soft->sadb_lifetime_bytes != 0) {
4308                         if (soft->sadb_lifetime_bytes >
4309                             assoc->ipsa_hardbyteslt) {
4310                                 assoc->ipsa_softbyteslt =
4311                                     assoc->ipsa_hardbyteslt;
4312                         } else {
4313                                 assoc->ipsa_softbyteslt =
4314                                     soft->sadb_lifetime_bytes;
4315                         }
4316                 }
4317                 if (soft->sadb_lifetime_usetime != 0) {
4318                         if (soft->sadb_lifetime_usetime >
4319                             assoc->ipsa_harduselt) {
4320                                 assoc->ipsa_softuselt =
4321                                     assoc->ipsa_harduselt;
4322                         } else {
4323                                 assoc->ipsa_softuselt =
4324                                     soft->sadb_lifetime_usetime;
4325                         }
4326                 }
4327                 if (soft->sadb_lifetime_addtime != 0) {
4328                         if (soft->sadb_lifetime_addtime >
4329                             assoc->ipsa_hardexpiretime) {
4330                                 assoc->ipsa_softexpiretime =
4331                                     assoc->ipsa_hardexpiretime;
4332                         } else {
4333                                 assoc->ipsa_softaddlt =
4334                                     soft->sadb_lifetime_addtime;
4335                         }
4336                 }
4337                 if (assoc->ipsa_softaddlt != 0) {
4338                         assoc->ipsa_softexpiretime =
4339                             assoc->ipsa_addtime + assoc->ipsa_softaddlt;
4340                 }
4341                 if (assoc->ipsa_softuselt != 0 &&
4342                     assoc->ipsa_flags & IPSA_F_USED) {
4343                         UPDATE_EXPIRE(assoc, softuselt, softexpiretime);
4344                 }
4345                 if (outbound && assoc->ipsa_softexpiretime != 0) {
4346                         if (assoc->ipsa_state == IPSA_STATE_MATURE)
4347                                 lifetime_fuzz(assoc);
4348                 }
4349 
4350                 if (soft->sadb_lifetime_allocations != 0)
4351                         assoc->ipsa_softalloc = soft->sadb_lifetime_allocations;
4352         }
4353 
4354         if (idle != NULL) {
4355                 time_t current = gethrestime_sec();
4356                 if ((assoc->ipsa_idleexpiretime <= current) &&
4357                     (assoc->ipsa_idleaddlt == idle->sadb_lifetime_addtime)) {
4358                         assoc->ipsa_idleexpiretime =
4359                             current + assoc->ipsa_idleaddlt;
4360                 }
4361                 if (idle->sadb_lifetime_addtime != 0)
4362                         assoc->ipsa_idleaddlt = idle->sadb_lifetime_addtime;
4363                 if (idle->sadb_lifetime_usetime != 0)
4364                         assoc->ipsa_idleuselt = idle->sadb_lifetime_usetime;
4365                 if (assoc->ipsa_idleaddlt != 0) {
4366                         assoc->ipsa_idleexpiretime =
4367                             current + idle->sadb_lifetime_addtime;
4368                         assoc->ipsa_idletime = idle->sadb_lifetime_addtime;
4369                 }
4370                 if (assoc->ipsa_idleuselt != 0) {
4371                         if (assoc->ipsa_idletime != 0) {
4372                                 assoc->ipsa_idletime = min(assoc->ipsa_idletime,
4373                                     assoc->ipsa_idleuselt);
4374                         assoc->ipsa_idleexpiretime =
4375                             current + assoc->ipsa_idletime;
4376                         } else {
4377                                 assoc->ipsa_idleexpiretime =
4378                                     current + assoc->ipsa_idleuselt;
4379                                 assoc->ipsa_idletime = assoc->ipsa_idleuselt;
4380                         }
4381                 }
4382         }
4383         mutex_exit(&assoc->ipsa_lock);
4384 }
4385 
4386 static int
4387 sadb_update_state(ipsa_t *assoc, uint_t new_state, mblk_t **ipkt_lst)
4388 {
4389         int rcode = 0;
4390         time_t current = gethrestime_sec();
4391 
4392         mutex_enter(&assoc->ipsa_lock);
4393 
4394         switch (new_state) {
4395         case SADB_X_SASTATE_ACTIVE_ELSEWHERE:
4396                 if (assoc->ipsa_state == SADB_X_SASTATE_IDLE) {
4397                         assoc->ipsa_state = IPSA_STATE_ACTIVE_ELSEWHERE;
4398                         assoc->ipsa_idleexpiretime =
4399                             current + assoc->ipsa_idletime;
4400                 }
4401                 break;
4402         case SADB_X_SASTATE_IDLE:
4403                 if (assoc->ipsa_state == SADB_X_SASTATE_ACTIVE_ELSEWHERE) {
4404                         assoc->ipsa_state = IPSA_STATE_IDLE;
4405                         assoc->ipsa_idleexpiretime =
4406                             current + assoc->ipsa_idletime;
4407                 } else {
4408                         rcode = EINVAL;
4409                 }
4410                 break;
4411 
4412         case SADB_X_SASTATE_ACTIVE:
4413                 if (assoc->ipsa_state != SADB_X_SASTATE_IDLE) {
4414                         rcode = EINVAL;
4415                         break;
4416                 }
4417                 assoc->ipsa_state = IPSA_STATE_MATURE;
4418                 assoc->ipsa_idleexpiretime = current + assoc->ipsa_idletime;
4419 
4420                 if (ipkt_lst == NULL) {
4421                         break;
4422                 }
4423 
4424                 if (assoc->ipsa_bpkt_head != NULL) {
4425                         *ipkt_lst = assoc->ipsa_bpkt_head;
4426                         assoc->ipsa_bpkt_head = assoc->ipsa_bpkt_tail = NULL;
4427                         assoc->ipsa_mblkcnt = 0;
4428                 } else {
4429                         *ipkt_lst = NULL;
4430                 }
4431                 break;
4432         default:
4433                 rcode = EINVAL;
4434                 break;
4435         }
4436 
4437         mutex_exit(&assoc->ipsa_lock);
4438         return (rcode);
4439 }
4440 
4441 /*
4442  * Check a proposed KMC update for sanity.
4443  */
4444 static int
4445 sadb_check_kmc(ipsa_query_t *sq, ipsa_t *sa, int *diagnostic)
4446 {
4447         uint32_t kmp = sq->kmp;
4448         uint64_t kmc = sq->kmc;
4449 
4450         if (sa == NULL)
4451                 return (0);
4452 
4453         if (sa->ipsa_state == IPSA_STATE_DEAD)
4454                 return (ESRCH); /* DEAD == Not there, in this case. */
4455 
4456         if ((kmp != 0) && (sa->ipsa_kmp != 0) && (sa->ipsa_kmp != kmp)) {
4457                 *diagnostic = SADB_X_DIAGNOSTIC_DUPLICATE_KMP;
4458                 return (EINVAL);
4459         }
4460 
4461         if ((kmc != 0) && (sa->ipsa_kmc != 0) && (sa->ipsa_kmc != kmc)) {
4462                 *diagnostic = SADB_X_DIAGNOSTIC_DUPLICATE_KMC;
4463                 return (EINVAL);
4464         }
4465 
4466         return (0);
4467 }
4468 
4469 /*
4470  * Actually update the KMC info.
4471  */
4472 static void
4473 sadb_update_kmc(ipsa_query_t *sq, ipsa_t *sa)
4474 {
4475         uint32_t kmp = sq->kmp;
4476         uint64_t kmc = sq->kmc;
4477 
4478         if (kmp != 0)
4479                 sa->ipsa_kmp = kmp;
4480         if (kmc != 0)
4481                 sa->ipsa_kmc = kmc;
4482 }
4483 
4484 /*
4485  * Common code to update an SA.
4486  */
4487 
4488 int
4489 sadb_update_sa(mblk_t *mp, keysock_in_t *ksi, mblk_t **ipkt_lst,
4490     sadbp_t *spp, int *diagnostic, queue_t *pfkey_q,
4491     int (*add_sa_func)(mblk_t *, keysock_in_t *, int *, netstack_t *),
4492     netstack_t *ns, uint8_t sadb_msg_type)
4493 {
4494         sadb_key_t *akey = (sadb_key_t *)ksi->ks_in_extv[SADB_EXT_KEY_AUTH];
4495         sadb_key_t *ekey = (sadb_key_t *)ksi->ks_in_extv[SADB_EXT_KEY_ENCRYPT];
4496         sadb_x_replay_ctr_t *replext =
4497             (sadb_x_replay_ctr_t *)ksi->ks_in_extv[SADB_X_EXT_REPLAY_VALUE];
4498         sadb_lifetime_t *soft =
4499             (sadb_lifetime_t *)ksi->ks_in_extv[SADB_EXT_LIFETIME_SOFT];
4500         sadb_lifetime_t *hard =
4501             (sadb_lifetime_t *)ksi->ks_in_extv[SADB_EXT_LIFETIME_HARD];
4502         sadb_lifetime_t *idle =
4503             (sadb_lifetime_t *)ksi->ks_in_extv[SADB_X_EXT_LIFETIME_IDLE];
4504         sadb_x_pair_t *pair_ext =
4505             (sadb_x_pair_t *)ksi->ks_in_extv[SADB_X_EXT_PAIR];
4506         ipsa_t *echo_target = NULL;
4507         ipsap_t ipsapp;
4508         ipsa_query_t sq;
4509         time_t current = gethrestime_sec();
4510 
4511         sq.spp = spp;           /* XXX param */
4512         int error = sadb_form_query(ksi, IPSA_Q_SRC|IPSA_Q_DST|IPSA_Q_SA,
4513             IPSA_Q_SRC|IPSA_Q_DST|IPSA_Q_SA|IPSA_Q_INBOUND|IPSA_Q_OUTBOUND|
4514             IPSA_Q_KMC,
4515             &sq, diagnostic);
4516 
4517         if (error != 0)
4518                 return (error);
4519 
4520         error = get_ipsa_pair(&sq, &ipsapp, diagnostic);
4521         if (error != 0)
4522                 return (error);
4523 
4524         if (ipsapp.ipsap_psa_ptr == NULL && ipsapp.ipsap_sa_ptr != NULL) {
4525                 if (ipsapp.ipsap_sa_ptr->ipsa_state == IPSA_STATE_LARVAL) {
4526                         /*
4527                          * REFRELE the target and let the add_sa_func()
4528                          * deal with updating a larval SA.
4529                          */
4530                         destroy_ipsa_pair(&ipsapp);
4531                         return (add_sa_func(mp, ksi, diagnostic, ns));
4532                 }
4533         }
4534 
4535         /*
4536          * At this point we have an UPDATE to a MATURE SA. There should
4537          * not be any keying material present.
4538          */
4539         if (akey != NULL) {
4540                 *diagnostic = SADB_X_DIAGNOSTIC_AKEY_PRESENT;
4541                 error = EINVAL;
4542                 goto bail;
4543         }
4544         if (ekey != NULL) {
4545                 *diagnostic = SADB_X_DIAGNOSTIC_EKEY_PRESENT;
4546                 error = EINVAL;
4547                 goto bail;
4548         }
4549 
4550         if (sq.assoc->sadb_sa_state == SADB_X_SASTATE_ACTIVE_ELSEWHERE) {
4551                 if (ipsapp.ipsap_sa_ptr != NULL &&
4552                     ipsapp.ipsap_sa_ptr->ipsa_state == IPSA_STATE_IDLE) {
4553                         if ((error = sadb_update_state(ipsapp.ipsap_sa_ptr,
4554                             sq.assoc->sadb_sa_state, NULL)) != 0) {
4555                                 *diagnostic = SADB_X_DIAGNOSTIC_BAD_SASTATE;
4556                                 goto bail;
4557                         }
4558                 }
4559                 if (ipsapp.ipsap_psa_ptr != NULL &&
4560                     ipsapp.ipsap_psa_ptr->ipsa_state == IPSA_STATE_IDLE) {
4561                         if ((error = sadb_update_state(ipsapp.ipsap_psa_ptr,
4562                             sq.assoc->sadb_sa_state, NULL)) != 0) {
4563                                 *diagnostic = SADB_X_DIAGNOSTIC_BAD_SASTATE;
4564                                 goto bail;
4565                         }
4566                 }
4567         }
4568         if (sq.assoc->sadb_sa_state == SADB_X_SASTATE_ACTIVE) {
4569                 if (ipsapp.ipsap_sa_ptr != NULL) {
4570                         error = sadb_update_state(ipsapp.ipsap_sa_ptr,
4571                             sq.assoc->sadb_sa_state,
4572                             (ipsapp.ipsap_sa_ptr->ipsa_flags &
4573                             IPSA_F_INBOUND) ? ipkt_lst : NULL);
4574                         if (error) {
4575                                 *diagnostic = SADB_X_DIAGNOSTIC_BAD_SASTATE;
4576                                 goto bail;
4577                         }
4578                 }
4579                 if (ipsapp.ipsap_psa_ptr != NULL) {
4580                         error = sadb_update_state(ipsapp.ipsap_psa_ptr,
4581                             sq.assoc->sadb_sa_state,
4582                             (ipsapp.ipsap_psa_ptr->ipsa_flags &
4583                             IPSA_F_INBOUND) ? ipkt_lst : NULL);
4584                         if (error) {
4585                                 *diagnostic = SADB_X_DIAGNOSTIC_BAD_SASTATE;
4586                                 goto bail;
4587                         }
4588                 }
4589                 sadb_pfkey_echo(pfkey_q, mp, (sadb_msg_t *)mp->b_cont->b_rptr,
4590                     ksi, echo_target);
4591                 goto bail;
4592         }
4593 
4594         /*
4595          * Reality checks for updates of active associations.
4596          * Sundry first-pass UPDATE-specific reality checks.
4597          * Have to do the checks here, because it's after the add_sa code.
4598          * XXX STATS : logging/stats here?
4599          */
4600 
4601         if (!((sq.assoc->sadb_sa_state == SADB_SASTATE_MATURE) ||
4602             (sq.assoc->sadb_sa_state == SADB_X_SASTATE_ACTIVE_ELSEWHERE))) {
4603                 *diagnostic = SADB_X_DIAGNOSTIC_BAD_SASTATE;
4604                 error = EINVAL;
4605                 goto bail;
4606         }
4607         if (sq.assoc->sadb_sa_flags & ~spp->s_updateflags) {
4608                 *diagnostic = SADB_X_DIAGNOSTIC_BAD_SAFLAGS;
4609                 error = EINVAL;
4610                 goto bail;
4611         }
4612         if (ksi->ks_in_extv[SADB_EXT_LIFETIME_CURRENT] != NULL) {
4613                 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_LIFETIME;
4614                 error = EOPNOTSUPP;
4615                 goto bail;
4616         }
4617 
4618         if ((*diagnostic = sadb_hardsoftchk(hard, soft, idle)) != 0) {
4619                 error = EINVAL;
4620                 goto bail;
4621         }
4622 
4623         if ((*diagnostic = sadb_labelchk(ksi)) != 0)
4624                 return (EINVAL);
4625 
4626         error = sadb_check_kmc(&sq, ipsapp.ipsap_sa_ptr, diagnostic);
4627         if (error != 0)
4628                 goto bail;
4629 
4630         error = sadb_check_kmc(&sq, ipsapp.ipsap_psa_ptr, diagnostic);
4631         if (error != 0)
4632                 goto bail;
4633 
4634 
4635         if (ipsapp.ipsap_sa_ptr != NULL) {
4636                 /*
4637                  * Do not allow replay value change for MATURE or LARVAL SA.
4638                  */
4639 
4640                 if ((replext != NULL) &&
4641                     ((ipsapp.ipsap_sa_ptr->ipsa_state == IPSA_STATE_LARVAL) ||
4642                     (ipsapp.ipsap_sa_ptr->ipsa_state == IPSA_STATE_MATURE))) {
4643                         *diagnostic = SADB_X_DIAGNOSTIC_BAD_SASTATE;
4644                         error = EINVAL;
4645                         goto bail;
4646                 }
4647         }
4648 
4649 
4650         if (ipsapp.ipsap_sa_ptr != NULL) {
4651                 sadb_update_lifetimes(ipsapp.ipsap_sa_ptr, hard, soft,
4652                     idle, B_TRUE);
4653                 sadb_update_kmc(&sq, ipsapp.ipsap_sa_ptr);
4654                 if ((replext != NULL) &&
4655                     (ipsapp.ipsap_sa_ptr->ipsa_replay_wsize != 0)) {
4656                         /*
4657                          * If an inbound SA, update the replay counter
4658                          * and check off all the other sequence number
4659                          */
4660                         if (ksi->ks_in_dsttype == KS_IN_ADDR_ME) {
4661                                 if (!sadb_replay_check(ipsapp.ipsap_sa_ptr,
4662                                     replext->sadb_x_rc_replay32)) {
4663                                         *diagnostic =
4664                                             SADB_X_DIAGNOSTIC_INVALID_REPLAY;
4665                                         error = EINVAL;
4666                                         goto bail;
4667                                 }
4668                                 mutex_enter(&ipsapp.ipsap_sa_ptr->ipsa_lock);
4669                                 ipsapp.ipsap_sa_ptr->ipsa_idleexpiretime =
4670                                     current +
4671                                     ipsapp.ipsap_sa_ptr->ipsa_idletime;
4672                                 mutex_exit(&ipsapp.ipsap_sa_ptr->ipsa_lock);
4673                         } else {
4674                                 mutex_enter(&ipsapp.ipsap_sa_ptr->ipsa_lock);
4675                                 ipsapp.ipsap_sa_ptr->ipsa_replay =
4676                                     replext->sadb_x_rc_replay32;
4677                                 ipsapp.ipsap_sa_ptr->ipsa_idleexpiretime =
4678                                     current +
4679                                     ipsapp.ipsap_sa_ptr->ipsa_idletime;
4680                                 mutex_exit(&ipsapp.ipsap_sa_ptr->ipsa_lock);
4681                         }
4682                 }
4683         }
4684 
4685         if (sadb_msg_type == SADB_X_UPDATEPAIR) {
4686                 if (ipsapp.ipsap_psa_ptr != NULL) {
4687                         sadb_update_lifetimes(ipsapp.ipsap_psa_ptr, hard, soft,
4688                             idle, B_FALSE);
4689                         sadb_update_kmc(&sq, ipsapp.ipsap_psa_ptr);
4690                 } else {
4691                         *diagnostic = SADB_X_DIAGNOSTIC_PAIR_SA_NOTFOUND;
4692                         error = ESRCH;
4693                         goto bail;
4694                 }
4695         }
4696 
4697         if (pair_ext != NULL)
4698                 error = update_pairing(&ipsapp, &sq, ksi, diagnostic);
4699 
4700         if (error == 0)
4701                 sadb_pfkey_echo(pfkey_q, mp, (sadb_msg_t *)mp->b_cont->b_rptr,
4702                     ksi, echo_target);
4703 bail:
4704 
4705         destroy_ipsa_pair(&ipsapp);
4706 
4707         return (error);
4708 }
4709 
4710 
4711 static int
4712 update_pairing(ipsap_t *ipsapp, ipsa_query_t *sq, keysock_in_t *ksi,
4713     int *diagnostic)
4714 {
4715         sadb_sa_t *assoc = (sadb_sa_t *)ksi->ks_in_extv[SADB_EXT_SA];
4716         sadb_x_pair_t *pair_ext =
4717             (sadb_x_pair_t *)ksi->ks_in_extv[SADB_X_EXT_PAIR];
4718         int error = 0;
4719         ipsap_t oipsapp;
4720         boolean_t undo_pair = B_FALSE;
4721         uint32_t ipsa_flags;
4722 
4723         if (pair_ext->sadb_x_pair_spi == 0 || pair_ext->sadb_x_pair_spi ==
4724             assoc->sadb_sa_spi) {
4725                 *diagnostic = SADB_X_DIAGNOSTIC_PAIR_INAPPROPRIATE;
4726                 return (EINVAL);
4727         }
4728 
4729         /*
4730          * Assume for now that the spi value provided in the SADB_UPDATE
4731          * message was valid, update the SA with its pair spi value.
4732          * If the spi turns out to be bogus or the SA no longer exists
4733          * then this will be detected when the reverse update is made
4734          * below.
4735          */
4736         mutex_enter(&ipsapp->ipsap_sa_ptr->ipsa_lock);
4737         ipsapp->ipsap_sa_ptr->ipsa_flags |= IPSA_F_PAIRED;
4738         ipsapp->ipsap_sa_ptr->ipsa_otherspi = pair_ext->sadb_x_pair_spi;
4739         mutex_exit(&ipsapp->ipsap_sa_ptr->ipsa_lock);
4740 
4741         /*
4742          * After updating the ipsa_otherspi element of the SA, get_ipsa_pair()
4743          * should now return pointers to the SA *AND* its pair, if this is not
4744          * the case, the "otherspi" either did not exist or was deleted. Also
4745          * check that "otherspi" is not already paired. If everything looks
4746          * good, complete the update. IPSA_REFRELE the first pair_pointer
4747          * after this update to ensure its not deleted until we are done.
4748          */
4749         error = get_ipsa_pair(sq, &oipsapp, diagnostic);
4750         if (error != 0) {
4751                 /*
4752                  * This should never happen, calling function still has
4753                  * IPSA_REFHELD on the SA we just updated.
4754                  */
4755                 return (error); /* XXX EINVAL instead of ESRCH? */
4756         }
4757 
4758         if (oipsapp.ipsap_psa_ptr == NULL) {
4759                 *diagnostic = SADB_X_DIAGNOSTIC_PAIR_INAPPROPRIATE;
4760                 error = EINVAL;
4761                 undo_pair = B_TRUE;
4762         } else {
4763                 ipsa_flags = oipsapp.ipsap_psa_ptr->ipsa_flags;
4764                 if ((oipsapp.ipsap_psa_ptr->ipsa_state == IPSA_STATE_DEAD) ||
4765                     (oipsapp.ipsap_psa_ptr->ipsa_state == IPSA_STATE_DYING)) {
4766                         /* Its dead Jim! */
4767                         *diagnostic = SADB_X_DIAGNOSTIC_PAIR_INAPPROPRIATE;
4768                         undo_pair = B_TRUE;
4769                 } else if ((ipsa_flags & (IPSA_F_OUTBOUND | IPSA_F_INBOUND)) ==
4770                     (IPSA_F_OUTBOUND | IPSA_F_INBOUND)) {
4771                         /* This SA is in both hashtables. */
4772                         *diagnostic = SADB_X_DIAGNOSTIC_PAIR_INAPPROPRIATE;
4773                         undo_pair = B_TRUE;
4774                 } else if (ipsa_flags & IPSA_F_PAIRED) {
4775                         /* This SA is already paired with another. */
4776                         *diagnostic = SADB_X_DIAGNOSTIC_PAIR_ALREADY;
4777                         undo_pair = B_TRUE;
4778                 }
4779         }
4780 
4781         if (undo_pair) {
4782                 /* The pair SA does not exist. */
4783                 mutex_enter(&ipsapp->ipsap_sa_ptr->ipsa_lock);
4784                 ipsapp->ipsap_sa_ptr->ipsa_flags &= ~IPSA_F_PAIRED;
4785                 ipsapp->ipsap_sa_ptr->ipsa_otherspi = 0;
4786                 mutex_exit(&ipsapp->ipsap_sa_ptr->ipsa_lock);
4787         } else {
4788                 mutex_enter(&oipsapp.ipsap_psa_ptr->ipsa_lock);
4789                 oipsapp.ipsap_psa_ptr->ipsa_otherspi = assoc->sadb_sa_spi;
4790                 oipsapp.ipsap_psa_ptr->ipsa_flags |= IPSA_F_PAIRED;
4791                 mutex_exit(&oipsapp.ipsap_psa_ptr->ipsa_lock);
4792         }
4793 
4794         destroy_ipsa_pair(&oipsapp);
4795         return (error);
4796 }
4797 
4798 /*
4799  * The following functions deal with ACQUIRE LISTS.  An ACQUIRE list is
4800  * a list of outstanding SADB_ACQUIRE messages.  If ipsec_getassocbyconn() fails
4801  * for an outbound datagram, that datagram is queued up on an ACQUIRE record,
4802  * and an SADB_ACQUIRE message is sent up.  Presumably, a user-space key
4803  * management daemon will process the ACQUIRE, use a SADB_GETSPI to reserve
4804  * an SPI value and a larval SA, then SADB_UPDATE the larval SA, and ADD the
4805  * other direction's SA.
4806  */
4807 
4808 /*
4809  * Check the ACQUIRE lists.  If there's an existing ACQUIRE record,
4810  * grab it, lock it, and return it.  Otherwise return NULL.
4811  *
4812  * XXX MLS number of arguments getting unwieldy here
4813  */
4814 static ipsacq_t *
4815 sadb_checkacquire(iacqf_t *bucket, ipsec_action_t *ap, ipsec_policy_t *pp,
4816     uint32_t *src, uint32_t *dst, uint32_t *isrc, uint32_t *idst,
4817     uint64_t unique_id, ts_label_t *tsl)
4818 {
4819         ipsacq_t *walker;
4820         sa_family_t fam;
4821         uint32_t blank_address[4] = {0, 0, 0, 0};
4822 
4823         if (isrc == NULL) {
4824                 ASSERT(idst == NULL);
4825                 isrc = idst = blank_address;
4826         }
4827 
4828         /*
4829          * Scan list for duplicates.  Check for UNIQUE, src/dest, policy.
4830          *
4831          * XXX May need search for duplicates based on other things too!
4832          */
4833         for (walker = bucket->iacqf_ipsacq; walker != NULL;
4834             walker = walker->ipsacq_next) {
4835                 mutex_enter(&walker->ipsacq_lock);
4836                 fam = walker->ipsacq_addrfam;
4837                 if (IPSA_ARE_ADDR_EQUAL(dst, walker->ipsacq_dstaddr, fam) &&
4838                     IPSA_ARE_ADDR_EQUAL(src, walker->ipsacq_srcaddr, fam) &&
4839                     ip_addr_match((uint8_t *)isrc, walker->ipsacq_innersrcpfx,
4840                     (in6_addr_t *)walker->ipsacq_innersrc) &&
4841                     ip_addr_match((uint8_t *)idst, walker->ipsacq_innerdstpfx,
4842                     (in6_addr_t *)walker->ipsacq_innerdst) &&
4843                     (ap == walker->ipsacq_act) &&
4844                     (pp == walker->ipsacq_policy) &&
4845                     /* XXX do deep compares of ap/pp? */
4846                     (unique_id == walker->ipsacq_unique_id) &&
4847                     (ipsec_label_match(tsl, walker->ipsacq_tsl)))
4848                         break;                  /* everything matched */
4849                 mutex_exit(&walker->ipsacq_lock);
4850         }
4851 
4852         return (walker);
4853 }
4854 
4855 /*
4856  * Generate an SADB_ACQUIRE base message mblk, including KEYSOCK_OUT metadata.
4857  * In other words, this will return, upon success, a two-mblk chain.
4858  */
4859 static inline mblk_t *
4860 sadb_acquire_msg_base(minor_t serial, uint8_t satype, uint32_t seq, pid_t pid)
4861 {
4862         mblk_t *mp;
4863         sadb_msg_t *samsg;
4864 
4865         mp = sadb_keysock_out(serial);
4866         if (mp == NULL)
4867                 return (NULL);
4868         mp->b_cont = allocb(sizeof (sadb_msg_t), BPRI_HI);
4869         if (mp->b_cont == NULL) {
4870                 freeb(mp);
4871                 return (NULL);
4872         }
4873 
4874         samsg = (sadb_msg_t *)mp->b_cont->b_rptr;
4875         mp->b_cont->b_wptr += sizeof (*samsg);
4876         samsg->sadb_msg_version = PF_KEY_V2;
4877         samsg->sadb_msg_type = SADB_ACQUIRE;
4878         samsg->sadb_msg_errno = 0;
4879         samsg->sadb_msg_reserved = 0;
4880         samsg->sadb_msg_satype = satype;
4881         samsg->sadb_msg_seq = seq;
4882         samsg->sadb_msg_pid = pid;
4883 
4884         return (mp);
4885 }
4886 
4887 /*
4888  * Generate address and TX/MLS sensitivity label PF_KEY extensions that are
4889  * common to both regular and extended ACQUIREs.
4890  */
4891 static mblk_t *
4892 sadb_acquire_msg_common(ipsec_selector_t *sel, ipsec_policy_t *pp,
4893     ipsec_action_t *ap, boolean_t tunnel_mode, ts_label_t *tsl,
4894     sadb_sens_t *sens)
4895 {
4896         size_t len;
4897         mblk_t *mp;
4898         uint8_t *start, *cur, *end;
4899         uint32_t *saddrptr, *daddrptr;
4900         sa_family_t af;
4901         ipsec_action_t *oldap;
4902         ipsec_selkey_t *ipsl;
4903         uint8_t proto, pfxlen;
4904         uint16_t lport, rport;
4905         int senslen = 0;
4906 
4907         /*
4908          * Get action pointer set if it isn't already.
4909          */
4910         oldap = ap;
4911         if (pp != NULL) {
4912                 ap = pp->ipsp_act;
4913                 if (ap == NULL)
4914                         ap = oldap;
4915         }
4916 
4917         /*
4918          * Biggest-case scenario:
4919          * 4x (sadb_address_t + struct sockaddr_in6)
4920          *      (src, dst, isrc, idst)
4921          *      (COMING SOON, 6x, because of triggering-packet contents.)
4922          * sadb_x_kmc_t
4923          * sadb_sens_t
4924          * And wiggle room for label bitvectors.  Luckily there are
4925          * programmatic ways to find it.
4926          */
4927         len = 4 * (sizeof (sadb_address_t) + sizeof (struct sockaddr_in6));
4928 
4929         /* Figure out full and proper length of sensitivity labels. */
4930         if (sens != NULL) {
4931                 ASSERT(tsl == NULL);
4932                 senslen = SADB_64TO8(sens->sadb_sens_len);
4933         } else if (tsl != NULL) {
4934                 senslen = sadb_sens_len_from_label(tsl);
4935         }
4936 #ifdef DEBUG
4937         else {
4938                 ASSERT(senslen == 0);
4939         }
4940 #endif /* DEBUG */
4941         len += senslen;
4942 
4943         mp = allocb(len, BPRI_HI);
4944         if (mp == NULL)
4945                 return (NULL);
4946 
4947         start = mp->b_rptr;
4948         end = start + len;
4949         cur = start;
4950 
4951         /*
4952          * Address extensions first, from most-recently-defined to least.
4953          * (This should immediately trigger surprise or verify robustness on
4954          * older apps, like in.iked.)
4955          */
4956         if (tunnel_mode) {
4957                 /*
4958                  * Form inner address extensions based NOT on the inner
4959                  * selectors (i.e. the packet data), but on the policy's
4960                  * selector key (i.e. the policy's selector information).
4961                  *
4962                  * NOTE:  The position of IPv4 and IPv6 addresses is the
4963                  * same in ipsec_selkey_t (unless the compiler does very
4964                  * strange things with unions, consult your local C language
4965                  * lawyer for details).
4966                  */
4967                 ASSERT(pp != NULL);
4968 
4969                 ipsl = &(pp->ipsp_sel->ipsl_key);
4970                 if (ipsl->ipsl_valid & IPSL_IPV4) {
4971                         af = AF_INET;
4972                         ASSERT(sel->ips_protocol == IPPROTO_ENCAP);
4973                         ASSERT(!(ipsl->ipsl_valid & IPSL_IPV6));
4974                 } else {
4975                         af = AF_INET6;
4976                         ASSERT(sel->ips_protocol == IPPROTO_IPV6);
4977                         ASSERT(ipsl->ipsl_valid & IPSL_IPV6);
4978                 }
4979 
4980                 if (ipsl->ipsl_valid & IPSL_LOCAL_ADDR) {
4981                         saddrptr = (uint32_t *)(&ipsl->ipsl_local);
4982                         pfxlen = ipsl->ipsl_local_pfxlen;
4983                 } else {
4984                         saddrptr = (uint32_t *)(&ipv6_all_zeros);
4985                         pfxlen = 0;
4986                 }
4987                 /* XXX What about ICMP type/code? */
4988                 lport = (ipsl->ipsl_valid & IPSL_LOCAL_PORT) ?
4989                     ipsl->ipsl_lport : 0;
4990                 proto = (ipsl->ipsl_valid & IPSL_PROTOCOL) ?
4991                     ipsl->ipsl_proto : 0;
4992 
4993                 cur = sadb_make_addr_ext(cur, end, SADB_X_EXT_ADDRESS_INNER_SRC,
4994                     af, saddrptr, lport, proto, pfxlen);
4995                 if (cur == NULL) {
4996                         freeb(mp);
4997                         return (NULL);
4998                 }
4999 
5000                 if (ipsl->ipsl_valid & IPSL_REMOTE_ADDR) {
5001                         daddrptr = (uint32_t *)(&ipsl->ipsl_remote);
5002                         pfxlen = ipsl->ipsl_remote_pfxlen;
5003                 } else {
5004                         daddrptr = (uint32_t *)(&ipv6_all_zeros);
5005                         pfxlen = 0;
5006                 }
5007                 /* XXX What about ICMP type/code? */
5008                 rport = (ipsl->ipsl_valid & IPSL_REMOTE_PORT) ?
5009                     ipsl->ipsl_rport : 0;
5010 
5011                 cur = sadb_make_addr_ext(cur, end, SADB_X_EXT_ADDRESS_INNER_DST,
5012                     af, daddrptr, rport, proto, pfxlen);
5013                 if (cur == NULL) {
5014                         freeb(mp);
5015                         return (NULL);
5016                 }
5017                 /*
5018                  * TODO  - if we go to 3884's dream of transport mode IP-in-IP
5019                  * _with_ inner-packet address selectors, we'll need to further
5020                  * distinguish tunnel mode here.  For now, having inner
5021                  * addresses and/or ports is sufficient.
5022                  *
5023                  * Meanwhile, whack proto/ports to reflect IP-in-IP for the
5024                  * outer addresses.
5025                  */
5026                 proto = sel->ips_protocol;   /* Either _ENCAP or _IPV6 */
5027                 lport = rport = 0;
5028         } else if ((ap != NULL) && (!ap->ipa_want_unique)) {
5029                 /*
5030                  * For cases when the policy calls out specific ports (or not).
5031                  */
5032                 proto = 0;
5033                 lport = 0;
5034                 rport = 0;
5035                 if (pp != NULL) {
5036                         ipsl = &(pp->ipsp_sel->ipsl_key);
5037                         if (ipsl->ipsl_valid & IPSL_PROTOCOL)
5038                                 proto = ipsl->ipsl_proto;
5039                         if (ipsl->ipsl_valid & IPSL_REMOTE_PORT)
5040                                 rport = ipsl->ipsl_rport;
5041                         if (ipsl->ipsl_valid & IPSL_LOCAL_PORT)
5042                                 lport = ipsl->ipsl_lport;
5043                 }
5044         } else {
5045                 /*
5046                  * For require-unique-SA policies.
5047                  */
5048                 proto = sel->ips_protocol;
5049                 lport = sel->ips_local_port;
5050                 rport = sel->ips_remote_port;
5051         }
5052 
5053         /*
5054          * Regular addresses.  These are outer-packet ones for tunnel mode.
5055          * Or for transport mode, the regulard address & port information.
5056          */
5057         af = sel->ips_isv4 ? AF_INET : AF_INET6;
5058 
5059         /*
5060          * NOTE:  The position of IPv4 and IPv6 addresses is the same in
5061          * ipsec_selector_t.
5062          */
5063         cur = sadb_make_addr_ext(cur, end, SADB_EXT_ADDRESS_SRC, af,
5064             (uint32_t *)(&sel->ips_local_addr_v6), lport, proto, 0);
5065         if (cur == NULL) {
5066                 freeb(mp);
5067                 return (NULL);
5068         }
5069 
5070         cur = sadb_make_addr_ext(cur, end, SADB_EXT_ADDRESS_DST, af,
5071             (uint32_t *)(&sel->ips_remote_addr_v6), rport, proto, 0);
5072         if (cur == NULL) {
5073                 freeb(mp);
5074                 return (NULL);
5075         }
5076 
5077         /*
5078          * If present, generate a sensitivity label.
5079          */
5080         if (cur + senslen > end) {
5081                 freeb(mp);
5082                 return (NULL);
5083         }
5084         if (sens != NULL) {
5085                 /* Explicit sadb_sens_t, usually from inverse-ACQUIRE. */
5086                 bcopy(sens, cur, senslen);
5087         } else if (tsl != NULL) {
5088                 /* Generate sadb_sens_t from ACQUIRE source. */
5089                 sadb_sens_from_label((sadb_sens_t *)cur, SADB_EXT_SENSITIVITY,
5090                     tsl, senslen);
5091         }
5092 #ifdef DEBUG
5093         else {
5094                 ASSERT(senslen == 0);
5095         }
5096 #endif /* DEBUG */
5097         cur += senslen;
5098         mp->b_wptr = cur;
5099 
5100         return (mp);
5101 }
5102 
5103 /*
5104  * Generate a regular ACQUIRE's proposal extension and KMC information..
5105  */
5106 static mblk_t *
5107 sadb_acquire_prop(ipsec_action_t *ap, netstack_t *ns, boolean_t do_esp)
5108 {
5109         ipsec_stack_t *ipss = ns->netstack_ipsec;
5110         ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
5111         ipsecah_stack_t *ahstack = ns->netstack_ipsecah;
5112         mblk_t *mp = NULL;
5113         sadb_prop_t *prop;
5114         sadb_comb_t *comb;
5115         ipsec_action_t *walker;
5116         int ncombs, allocsize, ealgid, aalgid, aminbits, amaxbits, eminbits,
5117             emaxbits, replay;
5118         uint64_t softbytes, hardbytes, softaddtime, hardaddtime, softusetime,
5119             hardusetime;
5120         uint64_t kmc = 0;
5121         uint32_t kmp = 0;
5122 
5123         /*
5124          * Since it's an rwlock read, AND writing to the IPsec algorithms is
5125          * rare, just acquire it once up top, and drop it upon return.
5126          */
5127         rw_enter(&ipss->ipsec_alg_lock, RW_READER);
5128         if (do_esp) {
5129                 uint64_t num_aalgs, num_ealgs;
5130 
5131                 if (espstack->esp_kstats == NULL)
5132                         goto bail;
5133 
5134                 num_aalgs = ipss->ipsec_nalgs[IPSEC_ALG_AUTH];
5135                 num_ealgs = ipss->ipsec_nalgs[IPSEC_ALG_ENCR];
5136                 if (num_ealgs == 0)
5137                         goto bail;      /* IPsec not loaded yet, apparently. */
5138                 num_aalgs++;    /* No-auth or self-auth-crypto ESP. */
5139 
5140                 /* Use netstack's maximum loaded algorithms... */
5141                 ncombs = num_ealgs * num_aalgs;
5142                 replay =  espstack->ipsecesp_replay_size;
5143         } else {
5144                 if (ahstack->ah_kstats == NULL)
5145                         goto bail;
5146 
5147                 ncombs = ipss->ipsec_nalgs[IPSEC_ALG_AUTH];
5148 
5149                 if (ncombs == 0)
5150                         goto bail;      /* IPsec not loaded yet, apparently. */
5151                 replay =  ahstack->ipsecah_replay_size;
5152         }
5153 
5154         allocsize = sizeof (*prop) + ncombs * sizeof (*comb) +
5155             sizeof (sadb_x_kmc_t);
5156         mp = allocb(allocsize, BPRI_HI);
5157         if (mp == NULL)
5158                 goto bail;
5159         prop = (sadb_prop_t *)mp->b_rptr;
5160         mp->b_wptr += sizeof (*prop);
5161         comb = (sadb_comb_t *)mp->b_wptr;
5162         /* Decrement allocsize, if it goes to or below 0, stop. */
5163         allocsize -= sizeof (*prop);
5164         prop->sadb_prop_exttype = SADB_EXT_PROPOSAL;
5165         prop->sadb_prop_len = SADB_8TO64(sizeof (*prop));
5166         *(uint32_t *)(&prop->sadb_prop_replay) = 0;      /* Quick zero-out! */
5167         prop->sadb_prop_replay = replay;
5168 
5169         /*
5170          * Based upon algorithm properties, and what-not, prioritize a
5171          * proposal, based on the ordering of the ESP algorithms in the
5172          * alternatives in the policy rule or socket that was placed
5173          * in the acquire record.
5174          *
5175          * For each action in policy list
5176          *   Add combination.
5177          *   I should not hit it, but if I've hit limit, return.
5178          */
5179 
5180         for (walker = ap; walker != NULL; walker = walker->ipa_next) {
5181                 ipsec_alginfo_t *ealg, *aalg;
5182                 ipsec_prot_t *prot;
5183 
5184                 if (walker->ipa_act.ipa_type != IPSEC_POLICY_APPLY)
5185                         continue;
5186 
5187                 prot = &walker->ipa_act.ipa_apply;
5188                 if (walker->ipa_act.ipa_apply.ipp_km_proto != 0)
5189                         kmp = walker->ipa_act.ipa_apply.ipp_km_proto;
5190                 if (walker->ipa_act.ipa_apply.ipp_km_cookie != 0)
5191                         kmc = walker->ipa_act.ipa_apply.ipp_km_cookie;
5192                 if (walker->ipa_act.ipa_apply.ipp_replay_depth) {
5193                         prop->sadb_prop_replay =
5194                             walker->ipa_act.ipa_apply.ipp_replay_depth;
5195                 }
5196 
5197                 if (do_esp) {
5198                         if (!prot->ipp_use_esp)
5199                                 continue;
5200 
5201                         if (prot->ipp_esp_auth_alg != 0) {
5202                                 aalg = ipss->ipsec_alglists[IPSEC_ALG_AUTH]
5203                                     [prot->ipp_esp_auth_alg];
5204                                 if (aalg == NULL || !ALG_VALID(aalg))
5205                                         continue;
5206                         } else
5207                                 aalg = NULL;
5208 
5209                         ASSERT(prot->ipp_encr_alg > 0);
5210                         ealg = ipss->ipsec_alglists[IPSEC_ALG_ENCR]
5211                             [prot->ipp_encr_alg];
5212                         if (ealg == NULL || !ALG_VALID(ealg))
5213                                 continue;
5214 
5215                         /*
5216                          * These may want to come from policy rule..
5217                          */
5218                         softbytes = espstack->ipsecesp_default_soft_bytes;
5219                         hardbytes = espstack->ipsecesp_default_hard_bytes;
5220                         softaddtime = espstack->ipsecesp_default_soft_addtime;
5221                         hardaddtime = espstack->ipsecesp_default_hard_addtime;
5222                         softusetime = espstack->ipsecesp_default_soft_usetime;
5223                         hardusetime = espstack->ipsecesp_default_hard_usetime;
5224                 } else {
5225                         if (!prot->ipp_use_ah)
5226                                 continue;
5227                         ealg = NULL;
5228                         aalg = ipss->ipsec_alglists[IPSEC_ALG_AUTH]
5229                             [prot->ipp_auth_alg];
5230                         if (aalg == NULL || !ALG_VALID(aalg))
5231                                 continue;
5232 
5233                         /*
5234                          * These may want to come from policy rule..
5235                          */
5236                         softbytes = ahstack->ipsecah_default_soft_bytes;
5237                         hardbytes = ahstack->ipsecah_default_hard_bytes;
5238                         softaddtime = ahstack->ipsecah_default_soft_addtime;
5239                         hardaddtime = ahstack->ipsecah_default_hard_addtime;
5240                         softusetime = ahstack->ipsecah_default_soft_usetime;
5241                         hardusetime = ahstack->ipsecah_default_hard_usetime;
5242                 }
5243 
5244                 if (ealg == NULL) {
5245                         ealgid = eminbits = emaxbits = 0;
5246                 } else {
5247                         ealgid = ealg->alg_id;
5248                         eminbits =
5249                             MAX(prot->ipp_espe_minbits, ealg->alg_ef_minbits);
5250                         emaxbits =
5251                             MIN(prot->ipp_espe_maxbits, ealg->alg_ef_maxbits);
5252                 }
5253 
5254                 if (aalg == NULL) {
5255                         aalgid = aminbits = amaxbits = 0;
5256                 } else {
5257                         aalgid = aalg->alg_id;
5258                         aminbits = MAX(prot->ipp_espa_minbits,
5259                             aalg->alg_ef_minbits);
5260                         amaxbits = MIN(prot->ipp_espa_maxbits,
5261                             aalg->alg_ef_maxbits);
5262                 }
5263 
5264                 comb->sadb_comb_flags = 0;
5265                 comb->sadb_comb_reserved = 0;
5266                 comb->sadb_comb_encrypt = ealgid;
5267                 comb->sadb_comb_encrypt_minbits = eminbits;
5268                 comb->sadb_comb_encrypt_maxbits = emaxbits;
5269                 comb->sadb_comb_auth = aalgid;
5270                 comb->sadb_comb_auth_minbits = aminbits;
5271                 comb->sadb_comb_auth_maxbits = amaxbits;
5272                 comb->sadb_comb_soft_allocations = 0;
5273                 comb->sadb_comb_hard_allocations = 0;
5274                 comb->sadb_comb_soft_bytes = softbytes;
5275                 comb->sadb_comb_hard_bytes = hardbytes;
5276                 comb->sadb_comb_soft_addtime = softaddtime;
5277                 comb->sadb_comb_hard_addtime = hardaddtime;
5278                 comb->sadb_comb_soft_usetime = softusetime;
5279                 comb->sadb_comb_hard_usetime = hardusetime;
5280 
5281                 prop->sadb_prop_len += SADB_8TO64(sizeof (*comb));
5282                 mp->b_wptr += sizeof (*comb);
5283                 allocsize -= sizeof (*comb);
5284                 /* Should never dip BELOW sizeof (KM cookie extension). */
5285                 ASSERT3S(allocsize, >=, sizeof (sadb_x_kmc_t));
5286                 if (allocsize <= sizeof (sadb_x_kmc_t))
5287                         break;  /* out of space.. */
5288                 comb++;
5289         }
5290 
5291         /* Don't include KMC extension if there's no room. */
5292         if (((kmp != 0) || (kmc != 0)) && allocsize >= sizeof (sadb_x_kmc_t)) {
5293                 if (sadb_make_kmc_ext(mp->b_wptr,
5294                     mp->b_wptr + sizeof (sadb_x_kmc_t), kmp, kmc) == NULL) {
5295                         freeb(mp);
5296                         mp = NULL;
5297                         goto bail;
5298                 }
5299                 mp->b_wptr += sizeof (sadb_x_kmc_t);
5300                 prop->sadb_prop_len += SADB_8TO64(sizeof (sadb_x_kmc_t));
5301         }
5302 
5303 bail:
5304         rw_exit(&ipss->ipsec_alg_lock);
5305         return (mp);
5306 }
5307 
5308 /*
5309  * Generate an extended ACQUIRE's extended-proposal extension.
5310  */
5311 static mblk_t *
5312 sadb_acquire_extended_prop(ipsec_action_t *ap, netstack_t *ns)
5313 {
5314         sadb_prop_t *eprop;
5315         uint8_t *cur, *end;
5316         mblk_t *mp;
5317         int allocsize, numecombs = 0, numalgdescs = 0;
5318         uint32_t kmp = 0, replay = 0;
5319         uint64_t kmc = 0;
5320         ipsec_action_t *walker;
5321 
5322         allocsize = sizeof (*eprop);
5323 
5324         /*
5325          * Going to walk through the action list twice.  Once for allocation
5326          * measurement, and once for actual construction.
5327          */
5328         for (walker = ap; walker != NULL; walker = walker->ipa_next) {
5329                 ipsec_prot_t *ipp;
5330 
5331                 /*
5332                  * Skip non-IPsec policies
5333                  */
5334                 if (walker->ipa_act.ipa_type != IPSEC_ACT_APPLY)
5335                         continue;
5336 
5337                 ipp = &walker->ipa_act.ipa_apply;
5338 
5339                 if (walker->ipa_act.ipa_apply.ipp_km_proto)
5340                         kmp = ipp->ipp_km_proto;
5341                 if (walker->ipa_act.ipa_apply.ipp_km_cookie)
5342                         kmc = ipp->ipp_km_cookie;
5343                 if (walker->ipa_act.ipa_apply.ipp_replay_depth)
5344                         replay = ipp->ipp_replay_depth;
5345 
5346                 if (ipp->ipp_use_ah)
5347                         numalgdescs++;
5348                 if (ipp->ipp_use_esp) {
5349                         numalgdescs++;
5350                         if (ipp->ipp_use_espa)
5351                                 numalgdescs++;
5352                 }
5353 
5354                 numecombs++;
5355         }
5356         ASSERT(numecombs > 0);
5357 
5358         allocsize += numecombs * sizeof (sadb_x_ecomb_t) +
5359             numalgdescs * sizeof (sadb_x_algdesc_t) + sizeof (sadb_x_kmc_t);
5360         mp = allocb(allocsize, BPRI_HI);
5361         if (mp == NULL)
5362                 return (NULL);
5363         eprop = (sadb_prop_t *)mp->b_rptr;
5364         end = mp->b_rptr + allocsize;
5365         cur = mp->b_rptr + sizeof (*eprop);
5366 
5367         eprop->sadb_prop_exttype = SADB_X_EXT_EPROP;
5368         eprop->sadb_x_prop_ereserved = 0;
5369         eprop->sadb_x_prop_numecombs = 0;
5370         *(uint32_t *)(&eprop->sadb_prop_replay) = 0;     /* Quick zero-out! */
5371         /* Pick ESP's replay default if need be. */
5372         eprop->sadb_prop_replay = (replay == 0) ?
5373             ns->netstack_ipsecesp->ipsecesp_replay_size : replay;
5374 
5375         /* This time, walk through and actually allocate. */
5376         for (walker = ap; walker != NULL; walker = walker->ipa_next) {
5377                 /*
5378                  * Skip non-IPsec policies
5379                  */
5380                 if (walker->ipa_act.ipa_type != IPSEC_ACT_APPLY)
5381                         continue;
5382                 cur = sadb_action_to_ecomb(cur, end, walker, ns);
5383                 if (cur == NULL) {
5384                         /* NOTE: inverse-ACQUIRE should note this as ENOMEM. */
5385                         freeb(mp);
5386                         return (NULL);
5387                 }
5388                 eprop->sadb_x_prop_numecombs++;
5389         }
5390 
5391         ASSERT(end - cur >= sizeof (sadb_x_kmc_t));
5392         if ((kmp != 0) || (kmc != 0)) {
5393                 cur = sadb_make_kmc_ext(cur, end, kmp, kmc);
5394                 if (cur == NULL) {
5395                         freeb(mp);
5396                         return (NULL);
5397                 }
5398         }
5399         mp->b_wptr = cur;
5400         eprop->sadb_prop_len = SADB_8TO64(cur - mp->b_rptr);
5401 
5402         return (mp);
5403 }
5404 
5405 /*
5406  * For this mblk, insert a new acquire record.  Assume bucket contains addrs
5407  * of all of the same length.  Give up (and drop) if memory
5408  * cannot be allocated for a new one; otherwise, invoke callback to
5409  * send the acquire up..
5410  *
5411  * In cases where we need both AH and ESP, add the SA to the ESP ACQUIRE
5412  * list.  The ah_add_sa_finish() routines can look at the packet's attached
5413  * attributes and handle this case specially.
5414  */
5415 void
5416 sadb_acquire(mblk_t *datamp, ip_xmit_attr_t *ixa, boolean_t need_ah,
5417     boolean_t need_esp)
5418 {
5419         mblk_t  *asyncmp, *regular, *extended, *common, *prop, *eprop;
5420         sadbp_t *spp;
5421         sadb_t *sp;
5422         ipsacq_t *newbie;
5423         iacqf_t *bucket;
5424         ipha_t *ipha = (ipha_t *)datamp->b_rptr;
5425         ip6_t *ip6h = (ip6_t *)datamp->b_rptr;
5426         uint32_t *src, *dst, *isrc, *idst;
5427         ipsec_policy_t *pp = ixa->ixa_ipsec_policy;
5428         ipsec_action_t *ap = ixa->ixa_ipsec_action;
5429         sa_family_t af;
5430         int hashoffset;
5431         uint32_t seq;
5432         uint64_t unique_id = 0;
5433         boolean_t tunnel_mode = (ixa->ixa_flags & IXAF_IPSEC_TUNNEL) != 0;
5434         ts_label_t      *tsl;
5435         netstack_t      *ns = ixa->ixa_ipst->ips_netstack;
5436         ipsec_stack_t   *ipss = ns->netstack_ipsec;
5437         ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
5438         ipsecah_stack_t *ahstack = ns->netstack_ipsecah;
5439         ipsec_selector_t sel;
5440         queue_t *q;
5441 
5442         ASSERT((pp != NULL) || (ap != NULL));
5443 
5444         ASSERT(need_ah || need_esp);
5445 
5446         /* Assign sadb pointers */
5447         if (need_esp) {
5448                 /*
5449                  * ESP happens first if we need both AH and ESP.
5450                  */
5451                 spp = &espstack->esp_sadb;
5452         } else {
5453                 spp = &ahstack->ah_sadb;
5454         }
5455         sp = (ixa->ixa_flags & IXAF_IS_IPV4) ? &spp->s_v4 : &spp->s_v6;
5456 
5457         if (is_system_labeled())
5458                 tsl = ixa->ixa_tsl;
5459         else
5460                 tsl = NULL;
5461 
5462         if (ap == NULL)
5463                 ap = pp->ipsp_act;
5464         ASSERT(ap != NULL);
5465 
5466         if (ap->ipa_act.ipa_apply.ipp_use_unique || tunnel_mode)
5467                 unique_id = SA_FORM_UNIQUE_ID(ixa);
5468 
5469         /*
5470          * Set up an ACQUIRE record.
5471          *
5472          * Immediately, make sure the ACQUIRE sequence number doesn't slip
5473          * below the lowest point allowed in the kernel.  (In other words,
5474          * make sure the high bit on the sequence number is set.)
5475          */
5476 
5477         seq = keysock_next_seq(ns) | IACQF_LOWEST_SEQ;
5478 
5479         if (IPH_HDR_VERSION(ipha) == IP_VERSION) {
5480                 src = (uint32_t *)&ipha->ipha_src;
5481                 dst = (uint32_t *)&ipha->ipha_dst;
5482                 af = AF_INET;
5483                 hashoffset = OUTBOUND_HASH_V4(sp, ipha->ipha_dst);
5484                 ASSERT(ixa->ixa_flags & IXAF_IS_IPV4);
5485         } else {
5486                 ASSERT(IPH_HDR_VERSION(ipha) == IPV6_VERSION);
5487                 src = (uint32_t *)&ip6h->ip6_src;
5488                 dst = (uint32_t *)&ip6h->ip6_dst;
5489                 af = AF_INET6;
5490                 hashoffset = OUTBOUND_HASH_V6(sp, ip6h->ip6_dst);
5491                 ASSERT(!(ixa->ixa_flags & IXAF_IS_IPV4));
5492         }
5493 
5494         if (tunnel_mode) {
5495                 if (pp == NULL) {
5496                         /*
5497                          * Tunnel mode with no policy pointer means this is a
5498                          * reflected ICMP (like a ECHO REQUEST) that came in
5499                          * with self-encapsulated protection.  Until we better
5500                          * support this, drop the packet.
5501                          */
5502                         ip_drop_packet(datamp, B_FALSE, NULL,
5503                             DROPPER(ipss, ipds_spd_got_selfencap),
5504                             &ipss->ipsec_spd_dropper);
5505                         return;
5506                 }
5507                 /* Snag inner addresses. */
5508                 isrc = ixa->ixa_ipsec_insrc;
5509                 idst = ixa->ixa_ipsec_indst;
5510         } else {
5511                 isrc = idst = NULL;
5512         }
5513 
5514         /*
5515          * Check buckets to see if there is an existing entry.  If so,
5516          * grab it.  sadb_checkacquire locks newbie if found.
5517          */
5518         bucket = &(sp->sdb_acq[hashoffset]);
5519         mutex_enter(&bucket->iacqf_lock);
5520         newbie = sadb_checkacquire(bucket, ap, pp, src, dst, isrc, idst,
5521             unique_id, tsl);
5522 
5523         if (newbie == NULL) {
5524                 /*
5525                  * Otherwise, allocate a new one.
5526                  */
5527                 newbie = kmem_zalloc(sizeof (*newbie), KM_NOSLEEP);
5528                 if (newbie == NULL) {
5529                         mutex_exit(&bucket->iacqf_lock);
5530                         ip_drop_packet(datamp, B_FALSE, NULL,
5531                             DROPPER(ipss, ipds_sadb_acquire_nomem),
5532                             &ipss->ipsec_sadb_dropper);
5533                         return;
5534                 }
5535                 newbie->ipsacq_policy = pp;
5536                 if (pp != NULL) {
5537                         IPPOL_REFHOLD(pp);
5538                 }
5539                 IPACT_REFHOLD(ap);
5540                 newbie->ipsacq_act = ap;
5541                 newbie->ipsacq_linklock = &bucket->iacqf_lock;
5542                 newbie->ipsacq_next = bucket->iacqf_ipsacq;
5543                 newbie->ipsacq_ptpn = &bucket->iacqf_ipsacq;
5544                 if (newbie->ipsacq_next != NULL)
5545                         newbie->ipsacq_next->ipsacq_ptpn = &newbie->ipsacq_next;
5546 
5547                 bucket->iacqf_ipsacq = newbie;
5548                 mutex_init(&newbie->ipsacq_lock, NULL, MUTEX_DEFAULT, NULL);
5549                 mutex_enter(&newbie->ipsacq_lock);
5550         }
5551 
5552         /*
5553          * XXX MLS does it actually help us to drop the bucket lock here?
5554          * we have inserted a half-built, locked acquire record into the
5555          * bucket.  any competing thread will now be able to lock the bucket
5556          * to scan it, but will immediately pile up on the new acquire
5557          * record's lock; I don't think we gain anything here other than to
5558          * disperse blame for lock contention.
5559          *
5560          * we might be able to dispense with acquire record locks entirely..
5561          * just use the bucket locks..
5562          */
5563 
5564         mutex_exit(&bucket->iacqf_lock);
5565 
5566         /*
5567          * This assert looks silly for now, but we may need to enter newbie's
5568          * mutex during a search.
5569          */
5570         ASSERT(MUTEX_HELD(&newbie->ipsacq_lock));
5571 
5572         /*
5573          * Make the ip_xmit_attr_t into something we can queue.
5574          * If no memory it frees datamp.
5575          */
5576         asyncmp = ip_xmit_attr_to_mblk(ixa);
5577         if (asyncmp != NULL)
5578                 linkb(asyncmp, datamp);
5579 
5580         /* Queue up packet.  Use b_next. */
5581 
5582         if (asyncmp == NULL) {
5583                 /* Statistics for allocation failure */
5584                 if (ixa->ixa_flags & IXAF_IS_IPV4) {
5585                         BUMP_MIB(&ixa->ixa_ipst->ips_ip_mib,
5586                             ipIfStatsOutDiscards);
5587                 } else {
5588                         BUMP_MIB(&ixa->ixa_ipst->ips_ip6_mib,
5589                             ipIfStatsOutDiscards);
5590                 }
5591                 ip_drop_output("No memory for asyncmp", datamp, NULL);
5592                 freemsg(datamp);
5593                 /*
5594                  * The acquire record will be freed quickly if it's new
5595                  * (ipsacq_expire == 0), and will proceed as if no packet
5596                  * showed up if not.
5597                  */
5598                 mutex_exit(&newbie->ipsacq_lock);
5599                 return;
5600         } else if (newbie->ipsacq_numpackets == 0) {
5601                 /* First one. */
5602                 newbie->ipsacq_mp = asyncmp;
5603                 newbie->ipsacq_numpackets = 1;
5604                 newbie->ipsacq_expire = gethrestime_sec();
5605                 /*
5606                  * Extended ACQUIRE with both AH+ESP will use ESP's timeout
5607                  * value.
5608                  */
5609                 newbie->ipsacq_expire += *spp->s_acquire_timeout;
5610                 newbie->ipsacq_seq = seq;
5611                 newbie->ipsacq_addrfam = af;
5612 
5613                 newbie->ipsacq_srcport = ixa->ixa_ipsec_src_port;
5614                 newbie->ipsacq_dstport = ixa->ixa_ipsec_dst_port;
5615                 newbie->ipsacq_icmp_type = ixa->ixa_ipsec_icmp_type;
5616                 newbie->ipsacq_icmp_code = ixa->ixa_ipsec_icmp_code;
5617                 if (tunnel_mode) {
5618                         newbie->ipsacq_inneraddrfam = ixa->ixa_ipsec_inaf;
5619                         newbie->ipsacq_proto = ixa->ixa_ipsec_inaf == AF_INET6 ?
5620                             IPPROTO_IPV6 : IPPROTO_ENCAP;
5621                         newbie->ipsacq_innersrcpfx = ixa->ixa_ipsec_insrcpfx;
5622                         newbie->ipsacq_innerdstpfx = ixa->ixa_ipsec_indstpfx;
5623                         IPSA_COPY_ADDR(newbie->ipsacq_innersrc,
5624                             ixa->ixa_ipsec_insrc, ixa->ixa_ipsec_inaf);
5625                         IPSA_COPY_ADDR(newbie->ipsacq_innerdst,
5626                             ixa->ixa_ipsec_indst, ixa->ixa_ipsec_inaf);
5627                 } else {
5628                         newbie->ipsacq_proto = ixa->ixa_ipsec_proto;
5629                 }
5630                 newbie->ipsacq_unique_id = unique_id;
5631 
5632                 if (tsl != NULL) {
5633                         label_hold(tsl);
5634                         newbie->ipsacq_tsl = tsl;
5635                 }
5636         } else {
5637                 /* Scan to the end of the list & insert. */
5638                 mblk_t *lastone = newbie->ipsacq_mp;
5639 
5640                 while (lastone->b_next != NULL)
5641                         lastone = lastone->b_next;
5642                 lastone->b_next = asyncmp;
5643                 if (newbie->ipsacq_numpackets++ == ipsacq_maxpackets) {
5644                         newbie->ipsacq_numpackets = ipsacq_maxpackets;
5645                         lastone = newbie->ipsacq_mp;
5646                         newbie->ipsacq_mp = lastone->b_next;
5647                         lastone->b_next = NULL;
5648 
5649                         /* Freeing the async message */
5650                         lastone = ip_xmit_attr_free_mblk(lastone);
5651                         ip_drop_packet(lastone, B_FALSE, NULL,
5652                             DROPPER(ipss, ipds_sadb_acquire_toofull),
5653                             &ipss->ipsec_sadb_dropper);
5654                 } else {
5655                         IP_ACQUIRE_STAT(ipss, qhiwater,
5656                             newbie->ipsacq_numpackets);
5657                 }
5658         }
5659 
5660         /*
5661          * Reset addresses.  Set them to the most recently added mblk chain,
5662          * so that the address pointers in the acquire record will point
5663          * at an mblk still attached to the acquire list.
5664          */
5665 
5666         newbie->ipsacq_srcaddr = src;
5667         newbie->ipsacq_dstaddr = dst;
5668 
5669         /*
5670          * If the acquire record has more than one queued packet, we've
5671          * already sent an ACQUIRE, and don't need to repeat ourself.
5672          */
5673         if (newbie->ipsacq_seq != seq || newbie->ipsacq_numpackets > 1) {
5674                 /* I have an acquire outstanding already! */
5675                 mutex_exit(&newbie->ipsacq_lock);
5676                 return;
5677         }
5678 
5679         if (need_esp) {
5680                 ESP_BUMP_STAT(espstack, acquire_requests);
5681                 q = espstack->esp_pfkey_q;
5682         } else {
5683                 /*
5684                  * Two cases get us here:
5685                  * 1.) AH-only policy.
5686                  *
5687                  * 2.) A continuation of an AH+ESP policy, and this is the
5688                  * post-ESP, AH-needs-to-send-a-regular-ACQUIRE case.
5689                  * (i.e. called from esp_do_outbound_ah().)
5690                  */
5691                 AH_BUMP_STAT(ahstack, acquire_requests);
5692                 q = ahstack->ah_pfkey_q;
5693         }
5694 
5695         /*
5696          * Get selectors and other policy-expression bits needed for an
5697          * ACQUIRE.
5698          */
5699         bzero(&sel, sizeof (sel));
5700         sel.ips_isv4 = (ixa->ixa_flags & IXAF_IS_IPV4) != 0;
5701         if (tunnel_mode) {
5702                 sel.ips_protocol = (ixa->ixa_ipsec_inaf == AF_INET) ?
5703                     IPPROTO_ENCAP : IPPROTO_IPV6;
5704         } else {
5705                 sel.ips_protocol = ixa->ixa_ipsec_proto;
5706                 sel.ips_local_port = ixa->ixa_ipsec_src_port;
5707                 sel.ips_remote_port = ixa->ixa_ipsec_dst_port;
5708         }
5709         sel.ips_icmp_type = ixa->ixa_ipsec_icmp_type;
5710         sel.ips_icmp_code = ixa->ixa_ipsec_icmp_code;
5711         sel.ips_is_icmp_inv_acq = 0;
5712         if (af == AF_INET) {
5713                 sel.ips_local_addr_v4 = ipha->ipha_src;
5714                 sel.ips_remote_addr_v4 = ipha->ipha_dst;
5715         } else {
5716                 sel.ips_local_addr_v6 = ip6h->ip6_src;
5717                 sel.ips_remote_addr_v6 = ip6h->ip6_dst;
5718         }
5719 
5720 
5721         /*
5722          * 1. Generate addresses, kmc, and sensitivity.  These are "common"
5723          * and should be an mblk pointed to by common. TBD -- eventually it
5724          * will include triggering packet contents as more address extensions.
5725          *
5726          * 2. Generate ACQUIRE & KEYSOCK_OUT and single-protocol proposal.
5727          * These are "regular" and "prop".  String regular->b_cont->b_cont =
5728          * common, common->b_cont = prop.
5729          *
5730          * 3. If extended register got turned on, generate EXT_ACQUIRE &
5731          * KEYSOCK_OUT and multi-protocol eprop. These are "extended" and
5732          * "eprop".  String extended->b_cont->b_cont = dupb(common) and
5733          * extended->b_cont->b_cont->b_cont = prop.
5734          *
5735          * 4. Deliver:  putnext(q, regular) and if there, putnext(q, extended).
5736          */
5737 
5738         regular = extended = prop = eprop = NULL;
5739 
5740         common = sadb_acquire_msg_common(&sel, pp, ap, tunnel_mode, tsl, NULL);
5741         if (common == NULL)
5742                 goto bail;
5743 
5744         regular = sadb_acquire_msg_base(0, (need_esp ?
5745             SADB_SATYPE_ESP : SADB_SATYPE_AH), newbie->ipsacq_seq, 0);
5746         if (regular == NULL)
5747                 goto bail;
5748 
5749         /*
5750          * Pardon the boolean cleverness. At least one of need_* must be true.
5751          * If they are equal, it's an AH & ESP policy and ESP needs to go
5752          * first.  If they aren't, just check the contents of need_esp.
5753          */
5754         prop = sadb_acquire_prop(ap, ns, need_esp);
5755         if (prop == NULL)
5756                 goto bail;
5757 
5758         /* Link the parts together. */
5759         regular->b_cont->b_cont = common;
5760         common->b_cont = prop;
5761         /*
5762          * Prop is now linked, so don't freemsg() it if the extended
5763          * construction goes off the rails.
5764          */
5765         prop = NULL;
5766 
5767         ((sadb_msg_t *)(regular->b_cont->b_rptr))->sadb_msg_len =
5768             SADB_8TO64(msgsize(regular->b_cont));
5769 
5770         /*
5771          * If we need an extended ACQUIRE, build it here.
5772          */
5773         if (keysock_extended_reg(ns)) {
5774                 /* NOTE: "common" still points to what we need. */
5775                 extended = sadb_acquire_msg_base(0, 0, newbie->ipsacq_seq, 0);
5776                 if (extended == NULL) {
5777                         common = NULL;
5778                         goto bail;
5779                 }
5780 
5781                 extended->b_cont->b_cont = dupb(common);
5782                 common = NULL;
5783                 if (extended->b_cont->b_cont == NULL)
5784                         goto bail;
5785 
5786                 eprop = sadb_acquire_extended_prop(ap, ns);
5787                 if (eprop == NULL)
5788                         goto bail;
5789                 extended->b_cont->b_cont->b_cont = eprop;
5790 
5791                 ((sadb_msg_t *)(extended->b_cont->b_rptr))->sadb_msg_len =
5792                     SADB_8TO64(msgsize(extended->b_cont));
5793         }
5794 
5795         /* So we don't hold a lock across putnext()... */
5796         mutex_exit(&newbie->ipsacq_lock);
5797 
5798         if (extended != NULL)
5799                 putnext(q, extended);
5800         ASSERT(regular != NULL);
5801         putnext(q, regular);
5802         return;
5803 
5804 bail:
5805         /* Make this acquire record go away quickly... */
5806         newbie->ipsacq_expire = 0;
5807         /* Exploit freemsg(NULL) being legal for fun & profit. */
5808         freemsg(common);
5809         freemsg(prop);
5810         freemsg(extended);
5811         freemsg(regular);
5812         mutex_exit(&newbie->ipsacq_lock);
5813 }
5814 
5815 /*
5816  * Unlink and free an acquire record.
5817  */
5818 void
5819 sadb_destroy_acquire(ipsacq_t *acqrec, netstack_t *ns)
5820 {
5821         mblk_t          *mp;
5822         ipsec_stack_t   *ipss = ns->netstack_ipsec;
5823 
5824         ASSERT(MUTEX_HELD(acqrec->ipsacq_linklock));
5825 
5826         if (acqrec->ipsacq_policy != NULL) {
5827                 IPPOL_REFRELE(acqrec->ipsacq_policy);
5828         }
5829         if (acqrec->ipsacq_act != NULL) {
5830                 IPACT_REFRELE(acqrec->ipsacq_act);
5831         }
5832 
5833         /* Unlink */
5834         *(acqrec->ipsacq_ptpn) = acqrec->ipsacq_next;
5835         if (acqrec->ipsacq_next != NULL)
5836                 acqrec->ipsacq_next->ipsacq_ptpn = acqrec->ipsacq_ptpn;
5837 
5838         if (acqrec->ipsacq_tsl != NULL) {
5839                 label_rele(acqrec->ipsacq_tsl);
5840                 acqrec->ipsacq_tsl = NULL;
5841         }
5842 
5843         /*
5844          * Free hanging mp's.
5845          *
5846          * XXX Instead of freemsg(), perhaps use IPSEC_REQ_FAILED.
5847          */
5848 
5849         mutex_enter(&acqrec->ipsacq_lock);
5850         while (acqrec->ipsacq_mp != NULL) {
5851                 mp = acqrec->ipsacq_mp;
5852                 acqrec->ipsacq_mp = mp->b_next;
5853                 mp->b_next = NULL;
5854                 /* Freeing the async message */
5855                 mp = ip_xmit_attr_free_mblk(mp);
5856                 ip_drop_packet(mp, B_FALSE, NULL,
5857                     DROPPER(ipss, ipds_sadb_acquire_timeout),
5858                     &ipss->ipsec_sadb_dropper);
5859         }
5860         mutex_exit(&acqrec->ipsacq_lock);
5861 
5862         /* Free */
5863         mutex_destroy(&acqrec->ipsacq_lock);
5864         kmem_free(acqrec, sizeof (*acqrec));
5865 }
5866 
5867 /*
5868  * Destroy an acquire list fanout.
5869  */
5870 static void
5871 sadb_destroy_acqlist(iacqf_t **listp, uint_t numentries, boolean_t forever,
5872     netstack_t *ns)
5873 {
5874         int i;
5875         iacqf_t *list = *listp;
5876 
5877         if (list == NULL)
5878                 return;
5879 
5880         for (i = 0; i < numentries; i++) {
5881                 mutex_enter(&(list[i].iacqf_lock));
5882                 while (list[i].iacqf_ipsacq != NULL)
5883                         sadb_destroy_acquire(list[i].iacqf_ipsacq, ns);
5884                 mutex_exit(&(list[i].iacqf_lock));
5885                 if (forever)
5886                         mutex_destroy(&(list[i].iacqf_lock));
5887         }
5888 
5889         if (forever) {
5890                 *listp = NULL;
5891                 kmem_free(list, numentries * sizeof (*list));
5892         }
5893 }
5894 
5895 /*
5896  * Create an algorithm descriptor for an extended ACQUIRE.  Filter crypto
5897  * framework's view of reality vs. IPsec's.  EF's wins, BTW.
5898  */
5899 static uint8_t *
5900 sadb_new_algdesc(uint8_t *start, uint8_t *limit,
5901     sadb_x_ecomb_t *ecomb, uint8_t satype, uint8_t algtype,
5902     uint8_t alg, uint16_t minbits, uint16_t maxbits, ipsec_stack_t *ipss)
5903 {
5904         uint8_t *cur = start;
5905         ipsec_alginfo_t *algp;
5906         sadb_x_algdesc_t *algdesc = (sadb_x_algdesc_t *)cur;
5907 
5908         cur += sizeof (*algdesc);
5909         if (cur >= limit)
5910                 return (NULL);
5911 
5912         ecomb->sadb_x_ecomb_numalgs++;
5913 
5914         /*
5915          * Normalize vs. crypto framework's limits.  This way, you can specify
5916          * a stronger policy, and when the framework loads a stronger version,
5917          * you can just keep plowing w/o rewhacking your SPD.
5918          */
5919         rw_enter(&ipss->ipsec_alg_lock, RW_READER);
5920         algp = ipss->ipsec_alglists[(algtype == SADB_X_ALGTYPE_AUTH) ?
5921             IPSEC_ALG_AUTH : IPSEC_ALG_ENCR][alg];
5922         if (algp == NULL) {
5923                 rw_exit(&ipss->ipsec_alg_lock);
5924                 return (NULL);  /* Algorithm doesn't exist.  Fail gracefully. */
5925         }
5926         if (minbits < algp->alg_ef_minbits)
5927                 minbits = algp->alg_ef_minbits;
5928         if (maxbits > algp->alg_ef_maxbits)
5929                 maxbits = algp->alg_ef_maxbits;
5930         rw_exit(&ipss->ipsec_alg_lock);
5931 
5932         algdesc->sadb_x_algdesc_reserved = SADB_8TO1(algp->alg_saltlen);
5933         algdesc->sadb_x_algdesc_satype = satype;
5934         algdesc->sadb_x_algdesc_algtype = algtype;
5935         algdesc->sadb_x_algdesc_alg = alg;
5936         algdesc->sadb_x_algdesc_minbits = minbits;
5937         algdesc->sadb_x_algdesc_maxbits = maxbits;
5938 
5939         return (cur);
5940 }
5941 
5942 /*
5943  * Convert the given ipsec_action_t into an ecomb starting at *ecomb
5944  * which must fit before *limit
5945  *
5946  * return NULL if we ran out of room or a pointer to the end of the ecomb.
5947  */
5948 static uint8_t *
5949 sadb_action_to_ecomb(uint8_t *start, uint8_t *limit, ipsec_action_t *act,
5950     netstack_t *ns)
5951 {
5952         uint8_t *cur = start;
5953         sadb_x_ecomb_t *ecomb = (sadb_x_ecomb_t *)cur;
5954         ipsec_prot_t *ipp;
5955         ipsec_stack_t *ipss = ns->netstack_ipsec;
5956 
5957         cur += sizeof (*ecomb);
5958         if (cur >= limit)
5959                 return (NULL);
5960 
5961         ASSERT(act->ipa_act.ipa_type == IPSEC_ACT_APPLY);
5962 
5963         ipp = &act->ipa_act.ipa_apply;
5964 
5965         ecomb->sadb_x_ecomb_numalgs = 0;
5966         ecomb->sadb_x_ecomb_reserved = 0;
5967         ecomb->sadb_x_ecomb_reserved2 = 0;
5968         /*
5969          * No limits on allocations, since we really don't support that
5970          * concept currently.
5971          */
5972         ecomb->sadb_x_ecomb_soft_allocations = 0;
5973         ecomb->sadb_x_ecomb_hard_allocations = 0;
5974 
5975         /*
5976          * XXX TBD: Policy or global parameters will eventually be
5977          * able to fill in some of these.
5978          */
5979         ecomb->sadb_x_ecomb_flags = 0;
5980         ecomb->sadb_x_ecomb_soft_bytes = 0;
5981         ecomb->sadb_x_ecomb_hard_bytes = 0;
5982         ecomb->sadb_x_ecomb_soft_addtime = 0;
5983         ecomb->sadb_x_ecomb_hard_addtime = 0;
5984         ecomb->sadb_x_ecomb_soft_usetime = 0;
5985         ecomb->sadb_x_ecomb_hard_usetime = 0;
5986 
5987         if (ipp->ipp_use_ah) {
5988                 cur = sadb_new_algdesc(cur, limit, ecomb,
5989                     SADB_SATYPE_AH, SADB_X_ALGTYPE_AUTH, ipp->ipp_auth_alg,
5990                     ipp->ipp_ah_minbits, ipp->ipp_ah_maxbits, ipss);
5991                 if (cur == NULL)
5992                         return (NULL);
5993                 ipsecah_fill_defs(ecomb, ns);
5994         }
5995 
5996         if (ipp->ipp_use_esp) {
5997                 if (ipp->ipp_use_espa) {
5998                         cur = sadb_new_algdesc(cur, limit, ecomb,
5999                             SADB_SATYPE_ESP, SADB_X_ALGTYPE_AUTH,
6000                             ipp->ipp_esp_auth_alg,
6001                             ipp->ipp_espa_minbits,
6002                             ipp->ipp_espa_maxbits, ipss);
6003                         if (cur == NULL)
6004                                 return (NULL);
6005                 }
6006 
6007                 cur = sadb_new_algdesc(cur, limit, ecomb,
6008                     SADB_SATYPE_ESP, SADB_X_ALGTYPE_CRYPT,
6009                     ipp->ipp_encr_alg,
6010                     ipp->ipp_espe_minbits,
6011                     ipp->ipp_espe_maxbits, ipss);
6012                 if (cur == NULL)
6013                         return (NULL);
6014                 /* Fill in lifetimes if and only if AH didn't already... */
6015                 if (!ipp->ipp_use_ah)
6016                         ipsecesp_fill_defs(ecomb, ns);
6017         }
6018 
6019         return (cur);
6020 }
6021 
6022 #include <sys/tsol/label_macro.h> /* XXX should not need this */
6023 
6024 /*
6025  * From a cred_t, construct a sensitivity label extension
6026  *
6027  * We send up a fixed-size sensitivity label bitmap, and are perhaps
6028  * overly chummy with the underlying data structures here.
6029  */
6030 
6031 /* ARGSUSED */
6032 int
6033 sadb_sens_len_from_label(ts_label_t *tsl)
6034 {
6035         int baselen = sizeof (sadb_sens_t) + _C_LEN * 4;
6036         return (roundup(baselen, sizeof (uint64_t)));
6037 }
6038 
6039 void
6040 sadb_sens_from_label(sadb_sens_t *sens, int exttype, ts_label_t *tsl,
6041     int senslen)
6042 {
6043         uint8_t *bitmap;
6044         bslabel_t *sl;
6045 
6046         /* LINTED */
6047         ASSERT((_C_LEN & 1) == 0);
6048         ASSERT((senslen & 7) == 0);
6049 
6050         sl = label2bslabel(tsl);
6051 
6052         sens->sadb_sens_exttype = exttype;
6053         sens->sadb_sens_len = SADB_8TO64(senslen);
6054 
6055         sens->sadb_sens_dpd = tsl->tsl_doi;
6056         sens->sadb_sens_sens_level = LCLASS(sl);
6057         sens->sadb_sens_integ_level = 0; /* TBD */
6058         sens->sadb_sens_sens_len = _C_LEN >> 1;
6059         sens->sadb_sens_integ_len = 0; /* TBD */
6060         sens->sadb_x_sens_flags = 0;
6061 
6062         bitmap = (uint8_t *)(sens + 1);
6063         bcopy(&(((_bslabel_impl_t *)sl)->compartments), bitmap, _C_LEN * 4);
6064 }
6065 
6066 /*
6067  * Okay, how do we report errors/invalid labels from this?
6068  * With a special designated "not a label" cred_t ?
6069  */
6070 /* ARGSUSED */
6071 ts_label_t *
6072 sadb_label_from_sens(sadb_sens_t *sens, uint64_t *bitmap)
6073 {
6074         int bitmap_len = SADB_64TO8(sens->sadb_sens_sens_len);
6075         bslabel_t sl;
6076         ts_label_t *tsl;
6077 
6078         if (sens->sadb_sens_integ_level != 0)
6079                 return (NULL);
6080         if (sens->sadb_sens_integ_len != 0)
6081                 return (NULL);
6082         if (bitmap_len > _C_LEN * 4)
6083                 return (NULL);
6084 
6085         bsllow(&sl);
6086         LCLASS_SET((_bslabel_impl_t *)&sl, sens->sadb_sens_sens_level);
6087         bcopy(bitmap, &((_bslabel_impl_t *)&sl)->compartments,
6088             bitmap_len);
6089 
6090         tsl = labelalloc(&sl, sens->sadb_sens_dpd, KM_NOSLEEP);
6091         if (tsl == NULL)
6092                 return (NULL);
6093 
6094         if (sens->sadb_x_sens_flags & SADB_X_SENS_UNLABELED)
6095                 tsl->tsl_flags |= TSLF_UNLABELED;
6096         return (tsl);
6097 }
6098 
6099 /* End XXX label-library-leakage */
6100 
6101 /*
6102  * Given an SADB_GETSPI message, find an appropriately ranged SA and
6103  * allocate an SA.  If there are message improprieties, return (ipsa_t *)-1.
6104  * If there was a memory allocation error, return NULL.  (Assume NULL !=
6105  * (ipsa_t *)-1).
6106  *
6107  * master_spi is passed in host order.
6108  */
6109 ipsa_t *
6110 sadb_getspi(keysock_in_t *ksi, uint32_t master_spi, int *diagnostic,
6111     netstack_t *ns, uint_t sa_type)
6112 {
6113         sadb_address_t *src =
6114             (sadb_address_t *)ksi->ks_in_extv[SADB_EXT_ADDRESS_SRC],
6115             *dst = (sadb_address_t *)ksi->ks_in_extv[SADB_EXT_ADDRESS_DST];
6116         sadb_spirange_t *range =
6117             (sadb_spirange_t *)ksi->ks_in_extv[SADB_EXT_SPIRANGE];
6118         struct sockaddr_in *ssa, *dsa;
6119         struct sockaddr_in6 *ssa6, *dsa6;
6120         uint32_t *srcaddr, *dstaddr;
6121         sa_family_t af;
6122         uint32_t add, min, max;
6123         uint8_t protocol =
6124             (sa_type == SADB_SATYPE_AH) ? IPPROTO_AH : IPPROTO_ESP;
6125 
6126         if (src == NULL) {
6127                 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_SRC;
6128                 return ((ipsa_t *)-1);
6129         }
6130         if (dst == NULL) {
6131                 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_DST;
6132                 return ((ipsa_t *)-1);
6133         }
6134         if (range == NULL) {
6135                 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_RANGE;
6136                 return ((ipsa_t *)-1);
6137         }
6138 
6139         min = ntohl(range->sadb_spirange_min);
6140         max = ntohl(range->sadb_spirange_max);
6141         dsa = (struct sockaddr_in *)(dst + 1);
6142         dsa6 = (struct sockaddr_in6 *)dsa;
6143 
6144         ssa = (struct sockaddr_in *)(src + 1);
6145         ssa6 = (struct sockaddr_in6 *)ssa;
6146         ASSERT(dsa->sin_family == ssa->sin_family);
6147 
6148         srcaddr = ALL_ZEROES_PTR;
6149         af = dsa->sin_family;
6150         switch (af) {
6151         case AF_INET:
6152                 if (src != NULL)
6153                         srcaddr = (uint32_t *)(&ssa->sin_addr);
6154                 dstaddr = (uint32_t *)(&dsa->sin_addr);
6155                 break;
6156         case AF_INET6:
6157                 if (src != NULL)
6158                         srcaddr = (uint32_t *)(&ssa6->sin6_addr);
6159                 dstaddr = (uint32_t *)(&dsa6->sin6_addr);
6160                 break;
6161         default:
6162                 *diagnostic = SADB_X_DIAGNOSTIC_BAD_DST_AF;
6163                 return ((ipsa_t *)-1);
6164         }
6165 
6166         if (master_spi < min || master_spi > max) {
6167                 /* Return a random value in the range. */
6168                 if (cl_inet_getspi) {
6169                         cl_inet_getspi(ns->netstack_stackid, protocol,
6170                             (uint8_t *)&add, sizeof (add), NULL);
6171                 } else {
6172                         (void) random_get_pseudo_bytes((uint8_t *)&add,
6173                             sizeof (add));
6174                 }
6175                 master_spi = min + (add % (max - min + 1));
6176         }
6177 
6178         /*
6179          * Since master_spi is passed in host order, we need to htonl() it
6180          * for the purposes of creating a new SA.
6181          */
6182         return (sadb_makelarvalassoc(htonl(master_spi), srcaddr, dstaddr, af,
6183             ns));
6184 }
6185 
6186 /*
6187  *
6188  * Locate an ACQUIRE and nuke it.  If I have an samsg that's larger than the
6189  * base header, just ignore it.  Otherwise, lock down the whole ACQUIRE list
6190  * and scan for the sequence number in question.  I may wish to accept an
6191  * address pair with it, for easier searching.
6192  *
6193  * Caller frees the message, so we don't have to here.
6194  *
6195  * NOTE:        The pfkey_q parameter may be used in the future for ACQUIRE
6196  *              failures.
6197  */
6198 /* ARGSUSED */
6199 void
6200 sadb_in_acquire(sadb_msg_t *samsg, sadbp_t *sp, queue_t *pfkey_q,
6201     netstack_t *ns)
6202 {
6203         int i;
6204         ipsacq_t *acqrec;
6205         iacqf_t *bucket;
6206 
6207         /*
6208          * I only accept the base header for this!
6209          * Though to be honest, requiring the dst address would help
6210          * immensely.
6211          *
6212          * XXX  There are already cases where I can get the dst address.
6213          */
6214         if (samsg->sadb_msg_len > SADB_8TO64(sizeof (*samsg)))
6215                 return;
6216 
6217         /*
6218          * Using the samsg->sadb_msg_seq, find the ACQUIRE record, delete it,
6219          * (and in the future send a message to IP with the appropriate error
6220          * number).
6221          *
6222          * Q: Do I want to reject if pid != 0?
6223          */
6224 
6225         for (i = 0; i < sp->s_v4.sdb_hashsize; i++) {
6226                 bucket = &sp->s_v4.sdb_acq[i];
6227                 mutex_enter(&bucket->iacqf_lock);
6228                 for (acqrec = bucket->iacqf_ipsacq; acqrec != NULL;
6229                     acqrec = acqrec->ipsacq_next) {
6230                         if (samsg->sadb_msg_seq == acqrec->ipsacq_seq)
6231                                 break;  /* for acqrec... loop. */
6232                 }
6233                 if (acqrec != NULL)
6234                         break;  /* for i = 0... loop. */
6235 
6236                 mutex_exit(&bucket->iacqf_lock);
6237         }
6238 
6239         if (acqrec == NULL) {
6240                 for (i = 0; i < sp->s_v6.sdb_hashsize; i++) {
6241                         bucket = &sp->s_v6.sdb_acq[i];
6242                         mutex_enter(&bucket->iacqf_lock);
6243                         for (acqrec = bucket->iacqf_ipsacq; acqrec != NULL;
6244                             acqrec = acqrec->ipsacq_next) {
6245                                 if (samsg->sadb_msg_seq == acqrec->ipsacq_seq)
6246                                         break;  /* for acqrec... loop. */
6247                         }
6248                         if (acqrec != NULL)
6249                                 break;  /* for i = 0... loop. */
6250 
6251                         mutex_exit(&bucket->iacqf_lock);
6252                 }
6253         }
6254 
6255 
6256         if (acqrec == NULL)
6257                 return;
6258 
6259         /*
6260          * What do I do with the errno and IP?  I may need mp's services a
6261          * little more.  See sadb_destroy_acquire() for future directions
6262          * beyond free the mblk chain on the acquire record.
6263          */
6264 
6265         ASSERT(&bucket->iacqf_lock == acqrec->ipsacq_linklock);
6266         sadb_destroy_acquire(acqrec, ns);
6267         /* Have to exit mutex here, because of breaking out of for loop. */
6268         mutex_exit(&bucket->iacqf_lock);
6269 }
6270 
6271 /*
6272  * The following functions work with the replay windows of an SA.  They assume
6273  * the ipsa->ipsa_replay_arr is an array of uint64_t, and that the bit vector
6274  * represents the highest sequence number packet received, and back
6275  * (ipsa->ipsa_replay_wsize) packets.
6276  */
6277 
6278 /*
6279  * Is the replay bit set?
6280  */
6281 static boolean_t
6282 ipsa_is_replay_set(ipsa_t *ipsa, uint32_t offset)
6283 {
6284         uint64_t bit = (uint64_t)1 << (uint64_t)(offset & 63);
6285 
6286         return ((bit & ipsa->ipsa_replay_arr[offset >> 6]) ? B_TRUE : B_FALSE);
6287 }
6288 
6289 /*
6290  * Shift the bits of the replay window over.
6291  */
6292 static void
6293 ipsa_shift_replay(ipsa_t *ipsa, uint32_t shift)
6294 {
6295         int i;
6296         int jump = ((shift - 1) >> 6) + 1;
6297 
6298         if (shift == 0)
6299                 return;
6300 
6301         for (i = (ipsa->ipsa_replay_wsize - 1) >> 6; i >= 0; i--) {
6302                 if (i + jump <= (ipsa->ipsa_replay_wsize - 1) >> 6) {
6303                         ipsa->ipsa_replay_arr[i + jump] |=
6304                             ipsa->ipsa_replay_arr[i] >> (64 - (shift & 63));
6305                 }
6306                 ipsa->ipsa_replay_arr[i] <<= shift;
6307         }
6308 }
6309 
6310 /*
6311  * Set a bit in the bit vector.
6312  */
6313 static void
6314 ipsa_set_replay(ipsa_t *ipsa, uint32_t offset)
6315 {
6316         uint64_t bit = (uint64_t)1 << (uint64_t)(offset & 63);
6317 
6318         ipsa->ipsa_replay_arr[offset >> 6] |= bit;
6319 }
6320 
6321 #define SADB_MAX_REPLAY_VALUE 0xffffffff
6322 
6323 /*
6324  * Assume caller has NOT done ntohl() already on seq.  Check to see
6325  * if replay sequence number "seq" has been seen already.
6326  */
6327 boolean_t
6328 sadb_replay_check(ipsa_t *ipsa, uint32_t seq)
6329 {
6330         boolean_t rc;
6331         uint32_t diff;
6332 
6333         if (ipsa->ipsa_replay_wsize == 0)
6334                 return (B_TRUE);
6335 
6336         /*
6337          * NOTE:  I've already checked for 0 on the wire in sadb_replay_peek().
6338          */
6339 
6340         /* Convert sequence number into host order before holding the mutex. */
6341         seq = ntohl(seq);
6342 
6343         mutex_enter(&ipsa->ipsa_lock);
6344 
6345         /* Initialize inbound SA's ipsa_replay field to last one received. */
6346         if (ipsa->ipsa_replay == 0)
6347                 ipsa->ipsa_replay = 1;
6348 
6349         if (seq > ipsa->ipsa_replay) {
6350                 /*
6351                  * I have received a new "highest value received".  Shift
6352                  * the replay window over.
6353                  */
6354                 diff = seq - ipsa->ipsa_replay;
6355                 if (diff < ipsa->ipsa_replay_wsize) {
6356                         /* In replay window, shift bits over. */
6357                         ipsa_shift_replay(ipsa, diff);
6358                 } else {
6359                         /* WAY FAR AHEAD, clear bits and start again. */
6360                         bzero(ipsa->ipsa_replay_arr,
6361                             sizeof (ipsa->ipsa_replay_arr));
6362                 }
6363                 ipsa_set_replay(ipsa, 0);
6364                 ipsa->ipsa_replay = seq;
6365                 rc = B_TRUE;
6366                 goto done;
6367         }
6368         diff = ipsa->ipsa_replay - seq;
6369         if (diff >= ipsa->ipsa_replay_wsize || ipsa_is_replay_set(ipsa, diff)) {
6370                 rc = B_FALSE;
6371                 goto done;
6372         }
6373         /* Set this packet as seen. */
6374         ipsa_set_replay(ipsa, diff);
6375 
6376         rc = B_TRUE;
6377 done:
6378         mutex_exit(&ipsa->ipsa_lock);
6379         return (rc);
6380 }
6381 
6382 /*
6383  * "Peek" and see if we should even bother going through the effort of
6384  * running an authentication check on the sequence number passed in.
6385  * this takes into account packets that are below the replay window,
6386  * and collisions with already replayed packets.  Return B_TRUE if it
6387  * is okay to proceed, B_FALSE if this packet should be dropped immediately.
6388  * Assume same byte-ordering as sadb_replay_check.
6389  */
6390 boolean_t
6391 sadb_replay_peek(ipsa_t *ipsa, uint32_t seq)
6392 {
6393         boolean_t rc = B_FALSE;
6394         uint32_t diff;
6395 
6396         if (ipsa->ipsa_replay_wsize == 0)
6397                 return (B_TRUE);
6398 
6399         /*
6400          * 0 is 0, regardless of byte order... :)
6401          *
6402          * If I get 0 on the wire (and there is a replay window) then the
6403          * sender most likely wrapped.  This ipsa may need to be marked or
6404          * something.
6405          */
6406         if (seq == 0)
6407                 return (B_FALSE);
6408 
6409         seq = ntohl(seq);
6410         mutex_enter(&ipsa->ipsa_lock);
6411         if (seq < ipsa->ipsa_replay - ipsa->ipsa_replay_wsize &&
6412             ipsa->ipsa_replay >= ipsa->ipsa_replay_wsize)
6413                 goto done;
6414 
6415         /*
6416          * If I've hit 0xffffffff, then quite honestly, I don't need to
6417          * bother with formalities.  I'm not accepting any more packets
6418          * on this SA.
6419          */
6420         if (ipsa->ipsa_replay == SADB_MAX_REPLAY_VALUE) {
6421                 /*
6422                  * Since we're already holding the lock, update the
6423                  * expire time ala. sadb_replay_delete() and return.
6424                  */
6425                 ipsa->ipsa_hardexpiretime = (time_t)1;
6426                 goto done;
6427         }
6428 
6429         if (seq <= ipsa->ipsa_replay) {
6430                 /*
6431                  * This seq is in the replay window.  I'm not below it,
6432                  * because I already checked for that above!
6433                  */
6434                 diff = ipsa->ipsa_replay - seq;
6435                 if (ipsa_is_replay_set(ipsa, diff))
6436                         goto done;
6437         }
6438         /* Else return B_TRUE, I'm going to advance the window. */
6439 
6440         rc = B_TRUE;
6441 done:
6442         mutex_exit(&ipsa->ipsa_lock);
6443         return (rc);
6444 }
6445 
6446 /*
6447  * Delete a single SA.
6448  *
6449  * For now, use the quick-and-dirty trick of making the association's
6450  * hard-expire lifetime (time_t)1, ensuring deletion by the *_ager().
6451  */
6452 void
6453 sadb_replay_delete(ipsa_t *assoc)
6454 {
6455         mutex_enter(&assoc->ipsa_lock);
6456         assoc->ipsa_hardexpiretime = (time_t)1;
6457         mutex_exit(&assoc->ipsa_lock);
6458 }
6459 
6460 /*
6461  * Special front-end to ipsec_rl_strlog() dealing with SA failure.
6462  * this is designed to take only a format string with "* %x * %s *", so
6463  * that "spi" is printed first, then "addr" is converted using inet_pton().
6464  *
6465  * This is abstracted out to save the stack space for only when inet_pton()
6466  * is called.  Make sure "spi" is in network order; it usually is when this
6467  * would get called.
6468  */
6469 void
6470 ipsec_assocfailure(short mid, short sid, char level, ushort_t sl, char *fmt,
6471     uint32_t spi, void *addr, int af, netstack_t *ns)
6472 {
6473         char buf[INET6_ADDRSTRLEN];
6474 
6475         ASSERT(af == AF_INET6 || af == AF_INET);
6476 
6477         ipsec_rl_strlog(ns, mid, sid, level, sl, fmt, ntohl(spi),
6478             inet_ntop(af, addr, buf, sizeof (buf)));
6479 }
6480 
6481 /*
6482  * Fills in a reference to the policy, if any, from the conn, in *ppp
6483  */
6484 static void
6485 ipsec_conn_pol(ipsec_selector_t *sel, conn_t *connp, ipsec_policy_t **ppp)
6486 {
6487         ipsec_policy_t  *pp;
6488         ipsec_latch_t   *ipl = connp->conn_latch;
6489 
6490         if ((ipl != NULL) && (connp->conn_ixa->ixa_ipsec_policy != NULL)) {
6491                 pp = connp->conn_ixa->ixa_ipsec_policy;
6492                 IPPOL_REFHOLD(pp);
6493         } else {
6494                 pp = ipsec_find_policy(IPSEC_TYPE_OUTBOUND, connp, sel,
6495                     connp->conn_netstack);
6496         }
6497         *ppp = pp;
6498 }
6499 
6500 /*
6501  * The following functions scan through active conn_t structures
6502  * and return a reference to the best-matching policy it can find.
6503  * Caller must release the reference.
6504  */
6505 static void
6506 ipsec_udp_pol(ipsec_selector_t *sel, ipsec_policy_t **ppp, ip_stack_t *ipst)
6507 {
6508         connf_t *connfp;
6509         conn_t *connp = NULL;
6510         ipsec_selector_t portonly;
6511 
6512         bzero((void *)&portonly, sizeof (portonly));
6513 
6514         if (sel->ips_local_port == 0)
6515                 return;
6516 
6517         connfp = &ipst->ips_ipcl_udp_fanout[IPCL_UDP_HASH(sel->ips_local_port,
6518             ipst)];
6519         mutex_enter(&connfp->connf_lock);
6520 
6521         if (sel->ips_isv4) {
6522                 connp = connfp->connf_head;
6523                 while (connp != NULL) {
6524                         if (IPCL_UDP_MATCH(connp, sel->ips_local_port,
6525                             sel->ips_local_addr_v4, sel->ips_remote_port,
6526                             sel->ips_remote_addr_v4))
6527                                 break;
6528                         connp = connp->conn_next;
6529                 }
6530 
6531                 if (connp == NULL) {
6532                         /* Try port-only match in IPv6. */
6533                         portonly.ips_local_port = sel->ips_local_port;
6534                         sel = &portonly;
6535                 }
6536         }
6537 
6538         if (connp == NULL) {
6539                 connp = connfp->connf_head;
6540                 while (connp != NULL) {
6541                         if (IPCL_UDP_MATCH_V6(connp, sel->ips_local_port,
6542                             sel->ips_local_addr_v6, sel->ips_remote_port,
6543                             sel->ips_remote_addr_v6))
6544                                 break;
6545                         connp = connp->conn_next;
6546                 }
6547 
6548                 if (connp == NULL) {
6549                         mutex_exit(&connfp->connf_lock);
6550                         return;
6551                 }
6552         }
6553 
6554         CONN_INC_REF(connp);
6555         mutex_exit(&connfp->connf_lock);
6556 
6557         ipsec_conn_pol(sel, connp, ppp);
6558         CONN_DEC_REF(connp);
6559 }
6560 
6561 static conn_t *
6562 ipsec_find_listen_conn(uint16_t *pptr, ipsec_selector_t *sel, ip_stack_t *ipst)
6563 {
6564         connf_t *connfp;
6565         conn_t *connp = NULL;
6566         const in6_addr_t *v6addrmatch = &sel->ips_local_addr_v6;
6567 
6568         if (sel->ips_local_port == 0)
6569                 return (NULL);
6570 
6571         connfp = &ipst->ips_ipcl_bind_fanout[
6572             IPCL_BIND_HASH(sel->ips_local_port, ipst)];
6573         mutex_enter(&connfp->connf_lock);
6574 
6575         if (sel->ips_isv4) {
6576                 connp = connfp->connf_head;
6577                 while (connp != NULL) {
6578                         if (IPCL_BIND_MATCH(connp, IPPROTO_TCP,
6579                             sel->ips_local_addr_v4, pptr[1]))
6580                                 break;
6581                         connp = connp->conn_next;
6582                 }
6583 
6584                 if (connp == NULL) {
6585                         /* Match to all-zeroes. */
6586                         v6addrmatch = &ipv6_all_zeros;
6587                 }
6588         }
6589 
6590         if (connp == NULL) {
6591                 connp = connfp->connf_head;
6592                 while (connp != NULL) {
6593                         if (IPCL_BIND_MATCH_V6(connp, IPPROTO_TCP,
6594                             *v6addrmatch, pptr[1]))
6595                                 break;
6596                         connp = connp->conn_next;
6597                 }
6598 
6599                 if (connp == NULL) {
6600                         mutex_exit(&connfp->connf_lock);
6601                         return (NULL);
6602                 }
6603         }
6604 
6605         CONN_INC_REF(connp);
6606         mutex_exit(&connfp->connf_lock);
6607         return (connp);
6608 }
6609 
6610 static void
6611 ipsec_tcp_pol(ipsec_selector_t *sel, ipsec_policy_t **ppp, ip_stack_t *ipst)
6612 {
6613         connf_t         *connfp;
6614         conn_t          *connp;
6615         uint32_t        ports;
6616         uint16_t        *pptr = (uint16_t *)&ports;
6617 
6618         /*
6619          * Find TCP state in the following order:
6620          * 1.) Connected conns.
6621          * 2.) Listeners.
6622          *
6623          * Even though #2 will be the common case for inbound traffic, only
6624          * following this order insures correctness.
6625          */
6626 
6627         if (sel->ips_local_port == 0)
6628                 return;
6629 
6630         /*
6631          * 0 should be fport, 1 should be lport.  SRC is the local one here.
6632          * See ipsec_construct_inverse_acquire() for details.
6633          */
6634         pptr[0] = sel->ips_remote_port;
6635         pptr[1] = sel->ips_local_port;
6636 
6637         connfp = &ipst->ips_ipcl_conn_fanout[
6638             IPCL_CONN_HASH(sel->ips_remote_addr_v4, ports, ipst)];
6639         mutex_enter(&connfp->connf_lock);
6640         connp = connfp->connf_head;
6641 
6642         if (sel->ips_isv4) {
6643                 while (connp != NULL) {
6644                         if (IPCL_CONN_MATCH(connp, IPPROTO_TCP,
6645                             sel->ips_remote_addr_v4, sel->ips_local_addr_v4,
6646                             ports))
6647                                 break;
6648                         connp = connp->conn_next;
6649                 }
6650         } else {
6651                 while (connp != NULL) {
6652                         if (IPCL_CONN_MATCH_V6(connp, IPPROTO_TCP,
6653                             sel->ips_remote_addr_v6, sel->ips_local_addr_v6,
6654                             ports))
6655                                 break;
6656                         connp = connp->conn_next;
6657                 }
6658         }
6659 
6660         if (connp != NULL) {
6661                 CONN_INC_REF(connp);
6662                 mutex_exit(&connfp->connf_lock);
6663         } else {
6664                 mutex_exit(&connfp->connf_lock);
6665 
6666                 /* Try the listen hash. */
6667                 if ((connp = ipsec_find_listen_conn(pptr, sel, ipst)) == NULL)
6668                         return;
6669         }
6670 
6671         ipsec_conn_pol(sel, connp, ppp);
6672         CONN_DEC_REF(connp);
6673 }
6674 
6675 static void
6676 ipsec_sctp_pol(ipsec_selector_t *sel, ipsec_policy_t **ppp,
6677     ip_stack_t *ipst)
6678 {
6679         conn_t          *connp;
6680         uint32_t        ports;
6681         uint16_t        *pptr = (uint16_t *)&ports;
6682 
6683         /*
6684          * Find SCP state in the following order:
6685          * 1.) Connected conns.
6686          * 2.) Listeners.
6687          *
6688          * Even though #2 will be the common case for inbound traffic, only
6689          * following this order insures correctness.
6690          */
6691 
6692         if (sel->ips_local_port == 0)
6693                 return;
6694 
6695         /*
6696          * 0 should be fport, 1 should be lport.  SRC is the local one here.
6697          * See ipsec_construct_inverse_acquire() for details.
6698          */
6699         pptr[0] = sel->ips_remote_port;
6700         pptr[1] = sel->ips_local_port;
6701 
6702         /*
6703          * For labeled systems, there's no need to check the
6704          * label here.  It's known to be good as we checked
6705          * before allowing the connection to become bound.
6706          */
6707         if (sel->ips_isv4) {
6708                 in6_addr_t      src, dst;
6709 
6710                 IN6_IPADDR_TO_V4MAPPED(sel->ips_remote_addr_v4, &dst);
6711                 IN6_IPADDR_TO_V4MAPPED(sel->ips_local_addr_v4, &src);
6712                 connp = sctp_find_conn(&dst, &src, ports, ALL_ZONES,
6713                     0, ipst->ips_netstack->netstack_sctp);
6714         } else {
6715                 connp = sctp_find_conn(&sel->ips_remote_addr_v6,
6716                     &sel->ips_local_addr_v6, ports, ALL_ZONES,
6717                     0, ipst->ips_netstack->netstack_sctp);
6718         }
6719         if (connp == NULL)
6720                 return;
6721         ipsec_conn_pol(sel, connp, ppp);
6722         CONN_DEC_REF(connp);
6723 }
6724 
6725 /*
6726  * Fill in a query for the SPD (in "sel") using two PF_KEY address extensions.
6727  * Returns 0 or errno, and always sets *diagnostic to something appropriate
6728  * to PF_KEY.
6729  *
6730  * NOTE:  For right now, this function (and ipsec_selector_t for that matter),
6731  * ignore prefix lengths in the address extension.  Since we match on first-
6732  * entered policies, this shouldn't matter.  Also, since we normalize prefix-
6733  * set addresses to mask out the lower bits, we should get a suitable search
6734  * key for the SPD anyway.  This is the function to change if the assumption
6735  * about suitable search keys is wrong.
6736  */
6737 static int
6738 ipsec_get_inverse_acquire_sel(ipsec_selector_t *sel, sadb_address_t *srcext,
6739     sadb_address_t *dstext, int *diagnostic)
6740 {
6741         struct sockaddr_in *src, *dst;
6742         struct sockaddr_in6 *src6, *dst6;
6743 
6744         *diagnostic = 0;
6745 
6746         bzero(sel, sizeof (*sel));
6747         sel->ips_protocol = srcext->sadb_address_proto;
6748         dst = (struct sockaddr_in *)(dstext + 1);
6749         if (dst->sin_family == AF_INET6) {
6750                 dst6 = (struct sockaddr_in6 *)dst;
6751                 src6 = (struct sockaddr_in6 *)(srcext + 1);
6752                 if (src6->sin6_family != AF_INET6) {
6753                         *diagnostic = SADB_X_DIAGNOSTIC_AF_MISMATCH;
6754                         return (EINVAL);
6755                 }
6756                 sel->ips_remote_addr_v6 = dst6->sin6_addr;
6757                 sel->ips_local_addr_v6 = src6->sin6_addr;
6758                 if (sel->ips_protocol == IPPROTO_ICMPV6) {
6759                         sel->ips_is_icmp_inv_acq = 1;
6760                 } else {
6761                         sel->ips_remote_port = dst6->sin6_port;
6762                         sel->ips_local_port = src6->sin6_port;
6763                 }
6764                 sel->ips_isv4 = B_FALSE;
6765         } else {
6766                 src = (struct sockaddr_in *)(srcext + 1);
6767                 if (src->sin_family != AF_INET) {
6768                         *diagnostic = SADB_X_DIAGNOSTIC_AF_MISMATCH;
6769                         return (EINVAL);
6770                 }
6771                 sel->ips_remote_addr_v4 = dst->sin_addr.s_addr;
6772                 sel->ips_local_addr_v4 = src->sin_addr.s_addr;
6773                 if (sel->ips_protocol == IPPROTO_ICMP) {
6774                         sel->ips_is_icmp_inv_acq = 1;
6775                 } else {
6776                         sel->ips_remote_port = dst->sin_port;
6777                         sel->ips_local_port = src->sin_port;
6778                 }
6779                 sel->ips_isv4 = B_TRUE;
6780         }
6781         return (0);
6782 }
6783 
6784 /*
6785  * We have encapsulation.
6786  * - Lookup tun_t by address and look for an associated
6787  *   tunnel policy
6788  * - If there are inner selectors
6789  *   - check ITPF_P_TUNNEL and ITPF_P_ACTIVE
6790  *   - Look up tunnel policy based on selectors
6791  * - Else
6792  *   - Sanity check the negotation
6793  *   - If appropriate, fall through to global policy
6794  */
6795 static int
6796 ipsec_tun_pol(ipsec_selector_t *sel, ipsec_policy_t **ppp,
6797     sadb_address_t *innsrcext, sadb_address_t *inndstext, ipsec_tun_pol_t *itp,
6798     int *diagnostic)
6799 {
6800         int err;
6801         ipsec_policy_head_t *polhead;
6802 
6803         *diagnostic = 0;
6804 
6805         /* Check for inner selectors and act appropriately */
6806 
6807         if (innsrcext != NULL) {
6808                 /* Inner selectors present */
6809                 ASSERT(inndstext != NULL);
6810                 if ((itp == NULL) ||
6811                     (itp->itp_flags & (ITPF_P_ACTIVE | ITPF_P_TUNNEL)) !=
6812                     (ITPF_P_ACTIVE | ITPF_P_TUNNEL)) {
6813                         /*
6814                          * If inner packet selectors, we must have negotiate
6815                          * tunnel and active policy.  If the tunnel has
6816                          * transport-mode policy set on it, or has no policy,
6817                          * fail.
6818                          */
6819                         return (ENOENT);
6820                 } else {
6821                         /*
6822                          * Reset "sel" to indicate inner selectors.  Pass
6823                          * inner PF_KEY address extensions for this to happen.
6824                          */
6825                         if ((err = ipsec_get_inverse_acquire_sel(sel,
6826                             innsrcext, inndstext, diagnostic)) != 0)
6827                                 return (err);
6828                         /*
6829                          * Now look for a tunnel policy based on those inner
6830                          * selectors.  (Common code is below.)
6831                          */
6832                 }
6833         } else {
6834                 /* No inner selectors present */
6835                 if ((itp == NULL) || !(itp->itp_flags & ITPF_P_ACTIVE)) {
6836                         /*
6837                          * Transport mode negotiation with no tunnel policy
6838                          * configured - return to indicate a global policy
6839                          * check is needed.
6840                          */
6841                         return (0);
6842                 } else if (itp->itp_flags & ITPF_P_TUNNEL) {
6843                         /* Tunnel mode set with no inner selectors. */
6844                         return (ENOENT);
6845                 }
6846                 /*
6847                  * Else, this is a tunnel policy configured with ifconfig(1m)
6848                  * or "negotiate transport" with ipsecconf(1m).  We have an
6849                  * itp with policy set based on any match, so don't bother
6850                  * changing fields in "sel".
6851                  */
6852         }
6853 
6854         ASSERT(itp != NULL);
6855         polhead = itp->itp_policy;
6856         ASSERT(polhead != NULL);
6857         rw_enter(&polhead->iph_lock, RW_READER);
6858         *ppp = ipsec_find_policy_head(NULL, polhead, IPSEC_TYPE_INBOUND, sel);
6859         rw_exit(&polhead->iph_lock);
6860 
6861         /*
6862          * Don't default to global if we didn't find a matching policy entry.
6863          * Instead, send ENOENT, just like if we hit a transport-mode tunnel.
6864          */
6865         if (*ppp == NULL)
6866                 return (ENOENT);
6867 
6868         return (0);
6869 }
6870 
6871 /*
6872  * For sctp conn_faddr is the primary address, hence this is of limited
6873  * use for sctp.
6874  */
6875 static void
6876 ipsec_oth_pol(ipsec_selector_t *sel, ipsec_policy_t **ppp,
6877     ip_stack_t *ipst)
6878 {
6879         boolean_t       isv4 = sel->ips_isv4;
6880         connf_t         *connfp;
6881         conn_t          *connp;
6882 
6883         if (isv4) {
6884                 connfp = &ipst->ips_ipcl_proto_fanout_v4[sel->ips_protocol];
6885         } else {
6886                 connfp = &ipst->ips_ipcl_proto_fanout_v6[sel->ips_protocol];
6887         }
6888 
6889         mutex_enter(&connfp->connf_lock);
6890         for (connp = connfp->connf_head; connp != NULL;
6891             connp = connp->conn_next) {
6892                 if (isv4) {
6893                         if ((connp->conn_laddr_v4 == INADDR_ANY ||
6894                             connp->conn_laddr_v4 == sel->ips_local_addr_v4) &&
6895                             (connp->conn_faddr_v4 == INADDR_ANY ||
6896                             connp->conn_faddr_v4 == sel->ips_remote_addr_v4))
6897                                 break;
6898                 } else {
6899                         if ((IN6_IS_ADDR_UNSPECIFIED(&connp->conn_laddr_v6) ||
6900                             IN6_ARE_ADDR_EQUAL(&connp->conn_laddr_v6,
6901                             &sel->ips_local_addr_v6)) &&
6902                             (IN6_IS_ADDR_UNSPECIFIED(&connp->conn_faddr_v6) ||
6903                             IN6_ARE_ADDR_EQUAL(&connp->conn_faddr_v6,
6904                             &sel->ips_remote_addr_v6)))
6905                                 break;
6906                 }
6907         }
6908         if (connp == NULL) {
6909                 mutex_exit(&connfp->connf_lock);
6910                 return;
6911         }
6912 
6913         CONN_INC_REF(connp);
6914         mutex_exit(&connfp->connf_lock);
6915 
6916         ipsec_conn_pol(sel, connp, ppp);
6917         CONN_DEC_REF(connp);
6918 }
6919 
6920 /*
6921  * Construct an inverse ACQUIRE reply based on:
6922  *
6923  * 1.) Current global policy.
6924  * 2.) An conn_t match depending on what all was passed in the extv[].
6925  * 3.) A tunnel's policy head.
6926  * ...
6927  * N.) Other stuff TBD (e.g. identities)
6928  *
6929  * If there is an error, set sadb_msg_errno and sadb_x_msg_diagnostic
6930  * in this function so the caller can extract them where appropriately.
6931  *
6932  * The SRC address is the local one - just like an outbound ACQUIRE message.
6933  *
6934  * XXX MLS: key management supplies a label which we just reflect back up
6935  * again.  clearly we need to involve the label in the rest of the checks.
6936  */
6937 mblk_t *
6938 ipsec_construct_inverse_acquire(sadb_msg_t *samsg, sadb_ext_t *extv[],
6939     netstack_t *ns)
6940 {
6941         int err;
6942         int diagnostic;
6943         sadb_address_t *srcext = (sadb_address_t *)extv[SADB_EXT_ADDRESS_SRC],
6944             *dstext = (sadb_address_t *)extv[SADB_EXT_ADDRESS_DST],
6945             *innsrcext = (sadb_address_t *)extv[SADB_X_EXT_ADDRESS_INNER_SRC],
6946             *inndstext = (sadb_address_t *)extv[SADB_X_EXT_ADDRESS_INNER_DST];
6947         sadb_sens_t *sens = (sadb_sens_t *)extv[SADB_EXT_SENSITIVITY];
6948         struct sockaddr_in6 *src, *dst;
6949         struct sockaddr_in6 *isrc, *idst;
6950         ipsec_tun_pol_t *itp = NULL;
6951         ipsec_policy_t *pp = NULL;
6952         ipsec_selector_t sel, isel;
6953         mblk_t *retmp = NULL;
6954         ip_stack_t      *ipst = ns->netstack_ip;
6955 
6956 
6957         /* Normalize addresses */
6958         if (sadb_addrcheck(NULL, (mblk_t *)samsg, (sadb_ext_t *)srcext, 0, ns)
6959             == KS_IN_ADDR_UNKNOWN) {
6960                 err = EINVAL;
6961                 diagnostic = SADB_X_DIAGNOSTIC_BAD_SRC;
6962                 goto bail;
6963         }
6964         src = (struct sockaddr_in6 *)(srcext + 1);
6965         if (sadb_addrcheck(NULL, (mblk_t *)samsg, (sadb_ext_t *)dstext, 0, ns)
6966             == KS_IN_ADDR_UNKNOWN) {
6967                 err = EINVAL;
6968                 diagnostic = SADB_X_DIAGNOSTIC_BAD_DST;
6969                 goto bail;
6970         }
6971         dst = (struct sockaddr_in6 *)(dstext + 1);
6972         if (src->sin6_family != dst->sin6_family) {
6973                 err = EINVAL;
6974                 diagnostic = SADB_X_DIAGNOSTIC_AF_MISMATCH;
6975                 goto bail;
6976         }
6977 
6978         /* Check for tunnel mode and act appropriately */
6979         if (innsrcext != NULL) {
6980                 if (inndstext == NULL) {
6981                         err = EINVAL;
6982                         diagnostic = SADB_X_DIAGNOSTIC_MISSING_INNER_DST;
6983                         goto bail;
6984                 }
6985                 if (sadb_addrcheck(NULL, (mblk_t *)samsg,
6986                     (sadb_ext_t *)innsrcext, 0, ns) == KS_IN_ADDR_UNKNOWN) {
6987                         err = EINVAL;
6988                         diagnostic = SADB_X_DIAGNOSTIC_MALFORMED_INNER_SRC;
6989                         goto bail;
6990                 }
6991                 isrc = (struct sockaddr_in6 *)(innsrcext + 1);
6992                 if (sadb_addrcheck(NULL, (mblk_t *)samsg,
6993                     (sadb_ext_t *)inndstext, 0, ns) == KS_IN_ADDR_UNKNOWN) {
6994                         err = EINVAL;
6995                         diagnostic = SADB_X_DIAGNOSTIC_MALFORMED_INNER_DST;
6996                         goto bail;
6997                 }
6998                 idst = (struct sockaddr_in6 *)(inndstext + 1);
6999                 if (isrc->sin6_family != idst->sin6_family) {
7000                         err = EINVAL;
7001                         diagnostic = SADB_X_DIAGNOSTIC_INNER_AF_MISMATCH;
7002                         goto bail;
7003                 }
7004                 if (isrc->sin6_family != AF_INET &&
7005                     isrc->sin6_family != AF_INET6) {
7006                         err = EINVAL;
7007                         diagnostic = SADB_X_DIAGNOSTIC_BAD_INNER_SRC_AF;
7008                         goto bail;
7009                 }
7010         } else if (inndstext != NULL) {
7011                 err = EINVAL;
7012                 diagnostic = SADB_X_DIAGNOSTIC_MISSING_INNER_SRC;
7013                 goto bail;
7014         }
7015 
7016         /* Get selectors first, based on outer addresses */
7017         err = ipsec_get_inverse_acquire_sel(&sel, srcext, dstext, &diagnostic);
7018         if (err != 0)
7019                 goto bail;
7020 
7021         /* Check for tunnel mode mismatches. */
7022         if (innsrcext != NULL &&
7023             ((isrc->sin6_family == AF_INET &&
7024             sel.ips_protocol != IPPROTO_ENCAP && sel.ips_protocol != 0) ||
7025             (isrc->sin6_family == AF_INET6 &&
7026             sel.ips_protocol != IPPROTO_IPV6 && sel.ips_protocol != 0))) {
7027                 err = EPROTOTYPE;
7028                 goto bail;
7029         }
7030 
7031         /*
7032          * Okay, we have the addresses and other selector information.
7033          * Let's first find a conn...
7034          */
7035         pp = NULL;
7036         switch (sel.ips_protocol) {
7037         case IPPROTO_TCP:
7038                 ipsec_tcp_pol(&sel, &pp, ipst);
7039                 break;
7040         case IPPROTO_UDP:
7041                 ipsec_udp_pol(&sel, &pp, ipst);
7042                 break;
7043         case IPPROTO_SCTP:
7044                 ipsec_sctp_pol(&sel, &pp, ipst);
7045                 break;
7046         case IPPROTO_ENCAP:
7047         case IPPROTO_IPV6:
7048                 /*
7049                  * Assume sel.ips_remote_addr_* has the right address at
7050                  * that exact position.
7051                  */
7052                 itp = itp_get_byaddr((uint32_t *)(&sel.ips_local_addr_v6),
7053                     (uint32_t *)(&sel.ips_remote_addr_v6), src->sin6_family,
7054                     ipst);
7055 
7056                 if (innsrcext == NULL) {
7057                         /*
7058                          * Transport-mode tunnel, make sure we fake out isel
7059                          * to contain something based on the outer protocol.
7060                          */
7061                         bzero(&isel, sizeof (isel));
7062                         isel.ips_isv4 = (sel.ips_protocol == IPPROTO_ENCAP);
7063                 } /* Else isel is initialized by ipsec_tun_pol(). */
7064                 err = ipsec_tun_pol(&isel, &pp, innsrcext, inndstext, itp,
7065                     &diagnostic);
7066                 /*
7067                  * NOTE:  isel isn't used for now, but in RFC 430x IPsec, it
7068                  * may be.
7069                  */
7070                 if (err != 0)
7071                         goto bail;
7072                 break;
7073         default:
7074                 ipsec_oth_pol(&sel, &pp, ipst);
7075                 break;
7076         }
7077 
7078         /*
7079          * If we didn't find a matching conn_t or other policy head, take a
7080          * look in the global policy.
7081          */
7082         if (pp == NULL) {
7083                 pp = ipsec_find_policy(IPSEC_TYPE_OUTBOUND, NULL, &sel, ns);
7084                 if (pp == NULL) {
7085                         /* There's no global policy. */
7086                         err = ENOENT;
7087                         diagnostic = 0;
7088                         goto bail;
7089                 }
7090         }
7091 
7092         ASSERT(pp != NULL);
7093         retmp = sadb_acquire_msg_base(0, 0, samsg->sadb_msg_seq,
7094             samsg->sadb_msg_pid);
7095         if (retmp != NULL) {
7096                 /* Remove KEYSOCK_OUT, because caller constructs it instead. */
7097                 mblk_t *kso = retmp;
7098 
7099                 retmp = retmp->b_cont;
7100                 freeb(kso);
7101                 /* Append addresses... */
7102                 retmp->b_cont = sadb_acquire_msg_common(&sel, pp, NULL,
7103                     (itp != NULL && (itp->itp_flags & ITPF_P_TUNNEL)), NULL,
7104                     sens);
7105                 if (retmp->b_cont == NULL) {
7106                         freemsg(retmp);
7107                         retmp = NULL;
7108                 }
7109                 /* And the policy result. */
7110                 retmp->b_cont->b_cont =
7111                     sadb_acquire_extended_prop(pp->ipsp_act, ns);
7112                 if (retmp->b_cont->b_cont == NULL) {
7113                         freemsg(retmp);
7114                         retmp = NULL;
7115                 }
7116                 ((sadb_msg_t *)retmp->b_rptr)->sadb_msg_len =
7117                     SADB_8TO64(msgsize(retmp));
7118         }
7119 
7120         if (pp != NULL) {
7121                 IPPOL_REFRELE(pp);
7122         }
7123         ASSERT(err == 0 && diagnostic == 0);
7124         if (retmp == NULL)
7125                 err = ENOMEM;
7126 bail:
7127         if (itp != NULL) {
7128                 ITP_REFRELE(itp, ns);
7129         }
7130         samsg->sadb_msg_errno = (uint8_t)err;
7131         samsg->sadb_x_msg_diagnostic = (uint16_t)diagnostic;
7132         return (retmp);
7133 }
7134 
7135 /*
7136  * ipsa_lpkt is a one-element queue, only manipulated by the next two
7137  * functions.  They have to hold the ipsa_lock because of potential races
7138  * between key management using SADB_UPDATE, and inbound packets that may
7139  * queue up on the larval SA (hence the 'l' in "lpkt").
7140  */
7141 
7142 /*
7143  * sadb_set_lpkt:
7144  *
7145  * Returns the passed-in packet if the SA is no longer larval.
7146  *
7147  * Returns NULL if the SA is larval, and needs to be swapped into the SA for
7148  * processing after an SADB_UPDATE.
7149  */
7150 mblk_t *
7151 sadb_set_lpkt(ipsa_t *ipsa, mblk_t *npkt, ip_recv_attr_t *ira)
7152 {
7153         mblk_t          *opkt;
7154 
7155         mutex_enter(&ipsa->ipsa_lock);
7156         opkt = ipsa->ipsa_lpkt;
7157         if (ipsa->ipsa_state == IPSA_STATE_LARVAL) {
7158                 /*
7159                  * Consume npkt and place it in the LARVAL SA's inbound
7160                  * packet slot.
7161                  */
7162                 mblk_t  *attrmp;
7163 
7164                 attrmp = ip_recv_attr_to_mblk(ira);
7165                 if (attrmp == NULL) {
7166                         ill_t *ill = ira->ira_ill;
7167 
7168                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
7169                         ip_drop_input("ipIfStatsInDiscards", npkt, ill);
7170                         freemsg(npkt);
7171                         opkt = NULL;
7172                 } else {
7173                         ASSERT(attrmp->b_cont == NULL);
7174                         attrmp->b_cont = npkt;
7175                         ipsa->ipsa_lpkt = attrmp;
7176                 }
7177                 npkt = NULL;
7178         } else {
7179                 /*
7180                  * If not larval, we lost the race.  NOTE: ipsa_lpkt may still
7181                  * have been non-NULL in the non-larval case, because of
7182                  * inbound packets arriving prior to sadb_common_add()
7183                  * transferring the SA completely out of larval state, but
7184                  * after lpkt was grabbed by the AH/ESP-specific add routines.
7185                  * We should clear the old ipsa_lpkt in this case to make sure
7186                  * that it doesn't linger on the now-MATURE IPsec SA, or get
7187                  * picked up as an out-of-order packet.
7188                  */
7189                 ipsa->ipsa_lpkt = NULL;
7190         }
7191         mutex_exit(&ipsa->ipsa_lock);
7192 
7193         if (opkt != NULL) {
7194                 ipsec_stack_t   *ipss;
7195 
7196                 ipss = ira->ira_ill->ill_ipst->ips_netstack->netstack_ipsec;
7197                 opkt = ip_recv_attr_free_mblk(opkt);
7198                 ip_drop_packet(opkt, B_TRUE, ira->ira_ill,
7199                     DROPPER(ipss, ipds_sadb_inlarval_replace),
7200                     &ipss->ipsec_sadb_dropper);
7201         }
7202         return (npkt);
7203 }
7204 
7205 /*
7206  * sadb_clear_lpkt: Atomically clear ipsa->ipsa_lpkt and return the
7207  * previous value.
7208  */
7209 mblk_t *
7210 sadb_clear_lpkt(ipsa_t *ipsa)
7211 {
7212         mblk_t *opkt;
7213 
7214         mutex_enter(&ipsa->ipsa_lock);
7215         opkt = ipsa->ipsa_lpkt;
7216         ipsa->ipsa_lpkt = NULL;
7217         mutex_exit(&ipsa->ipsa_lock);
7218         return (opkt);
7219 }
7220 
7221 /*
7222  * Buffer a packet that's in IDLE state as set by Solaris Clustering.
7223  */
7224 void
7225 sadb_buf_pkt(ipsa_t *ipsa, mblk_t *bpkt, ip_recv_attr_t *ira)
7226 {
7227         netstack_t      *ns = ira->ira_ill->ill_ipst->ips_netstack;
7228         ipsec_stack_t   *ipss = ns->netstack_ipsec;
7229         in6_addr_t *srcaddr = (in6_addr_t *)(&ipsa->ipsa_srcaddr);
7230         in6_addr_t *dstaddr = (in6_addr_t *)(&ipsa->ipsa_dstaddr);
7231         mblk_t          *mp;
7232 
7233         ASSERT(ipsa->ipsa_state == IPSA_STATE_IDLE);
7234 
7235         if (cl_inet_idlesa == NULL) {
7236                 ip_drop_packet(bpkt, B_TRUE, ira->ira_ill,
7237                     DROPPER(ipss, ipds_sadb_inidle_overflow),
7238                     &ipss->ipsec_sadb_dropper);
7239                 return;
7240         }
7241 
7242         cl_inet_idlesa(ns->netstack_stackid,
7243             (ipsa->ipsa_type == SADB_SATYPE_AH) ? IPPROTO_AH : IPPROTO_ESP,
7244             ipsa->ipsa_spi, ipsa->ipsa_addrfam, *srcaddr, *dstaddr, NULL);
7245 
7246         mp = ip_recv_attr_to_mblk(ira);
7247         if (mp == NULL) {
7248                 ip_drop_packet(bpkt, B_TRUE, ira->ira_ill,
7249                     DROPPER(ipss, ipds_sadb_inidle_overflow),
7250                     &ipss->ipsec_sadb_dropper);
7251                 return;
7252         }
7253         linkb(mp, bpkt);
7254 
7255         mutex_enter(&ipsa->ipsa_lock);
7256         ipsa->ipsa_mblkcnt++;
7257         if (ipsa->ipsa_bpkt_head == NULL) {
7258                 ipsa->ipsa_bpkt_head = ipsa->ipsa_bpkt_tail = bpkt;
7259         } else {
7260                 ipsa->ipsa_bpkt_tail->b_next = bpkt;
7261                 ipsa->ipsa_bpkt_tail = bpkt;
7262                 if (ipsa->ipsa_mblkcnt > SADB_MAX_IDLEPKTS) {
7263                         mblk_t *tmp;
7264 
7265                         tmp = ipsa->ipsa_bpkt_head;
7266                         ipsa->ipsa_bpkt_head = ipsa->ipsa_bpkt_head->b_next;
7267                         tmp = ip_recv_attr_free_mblk(tmp);
7268                         ip_drop_packet(tmp, B_TRUE, NULL,
7269                             DROPPER(ipss, ipds_sadb_inidle_overflow),
7270                             &ipss->ipsec_sadb_dropper);
7271                         ipsa->ipsa_mblkcnt --;
7272                 }
7273         }
7274         mutex_exit(&ipsa->ipsa_lock);
7275 }
7276 
7277 /*
7278  * Stub function that taskq_dispatch() invokes to take the mblk (in arg)
7279  * and put into STREAMS again.
7280  */
7281 void
7282 sadb_clear_buf_pkt(void *ipkt)
7283 {
7284         mblk_t  *tmp, *buf_pkt;
7285         ip_recv_attr_t  iras;
7286 
7287         buf_pkt = (mblk_t *)ipkt;
7288 
7289         while (buf_pkt != NULL) {
7290                 mblk_t *data_mp;
7291 
7292                 tmp = buf_pkt->b_next;
7293                 buf_pkt->b_next = NULL;
7294 
7295                 data_mp = buf_pkt->b_cont;
7296                 buf_pkt->b_cont = NULL;
7297                 if (!ip_recv_attr_from_mblk(buf_pkt, &iras)) {
7298                         /* The ill or ip_stack_t disappeared on us. */
7299                         ip_drop_input("ip_recv_attr_from_mblk", data_mp, NULL);
7300                         freemsg(data_mp);
7301                 } else {
7302                         ip_input_post_ipsec(data_mp, &iras);
7303                 }
7304                 ira_cleanup(&iras, B_TRUE);
7305                 buf_pkt = tmp;
7306         }
7307 }
7308 /*
7309  * Walker callback used by sadb_alg_update() to free/create crypto
7310  * context template when a crypto software provider is removed or
7311  * added.
7312  */
7313 
7314 struct sadb_update_alg_state {
7315         ipsec_algtype_t alg_type;
7316         uint8_t alg_id;
7317         boolean_t is_added;
7318         boolean_t async_auth;
7319         boolean_t async_encr;
7320 };
7321 
7322 static void
7323 sadb_alg_update_cb(isaf_t *head, ipsa_t *entry, void *cookie)
7324 {
7325         struct sadb_update_alg_state *update_state =
7326             (struct sadb_update_alg_state *)cookie;
7327         crypto_ctx_template_t *ctx_tmpl = NULL;
7328 
7329         ASSERT(MUTEX_HELD(&head->isaf_lock));
7330 
7331         if (entry->ipsa_state == IPSA_STATE_LARVAL)
7332                 return;
7333 
7334         mutex_enter(&entry->ipsa_lock);
7335 
7336         if ((entry->ipsa_encr_alg != SADB_EALG_NONE && entry->ipsa_encr_alg !=
7337             SADB_EALG_NULL && update_state->async_encr) ||
7338             (entry->ipsa_auth_alg != SADB_AALG_NONE &&
7339             update_state->async_auth)) {
7340                 entry->ipsa_flags |= IPSA_F_ASYNC;
7341         } else {
7342                 entry->ipsa_flags &= ~IPSA_F_ASYNC;
7343         }
7344 
7345         switch (update_state->alg_type) {
7346         case IPSEC_ALG_AUTH:
7347                 if (entry->ipsa_auth_alg == update_state->alg_id)
7348                         ctx_tmpl = &entry->ipsa_authtmpl;
7349                 break;
7350         case IPSEC_ALG_ENCR:
7351                 if (entry->ipsa_encr_alg == update_state->alg_id)
7352                         ctx_tmpl = &entry->ipsa_encrtmpl;
7353                 break;
7354         default:
7355                 ctx_tmpl = NULL;
7356         }
7357 
7358         if (ctx_tmpl == NULL) {
7359                 mutex_exit(&entry->ipsa_lock);
7360                 return;
7361         }
7362 
7363         /*
7364          * The context template of the SA may be affected by the change
7365          * of crypto provider.
7366          */
7367         if (update_state->is_added) {
7368                 /* create the context template if not already done */
7369                 if (*ctx_tmpl == NULL) {
7370                         (void) ipsec_create_ctx_tmpl(entry,
7371                             update_state->alg_type);
7372                 }
7373         } else {
7374                 /*
7375                  * The crypto provider was removed. If the context template
7376                  * exists but it is no longer valid, free it.
7377                  */
7378                 if (*ctx_tmpl != NULL)
7379                         ipsec_destroy_ctx_tmpl(entry, update_state->alg_type);
7380         }
7381 
7382         mutex_exit(&entry->ipsa_lock);
7383 }
7384 
7385 /*
7386  * Invoked by IP when an software crypto provider has been updated, or if
7387  * the crypto synchrony changes.  The type and id of the corresponding
7388  * algorithm is passed as argument.  The type is set to ALL in the case of
7389  * a synchrony change.
7390  *
7391  * is_added is B_TRUE if the provider was added, B_FALSE if it was
7392  * removed. The function updates the SADB and free/creates the
7393  * context templates associated with SAs if needed.
7394  */
7395 
7396 #define SADB_ALG_UPDATE_WALK(sadb, table) \
7397     sadb_walker((sadb).table, (sadb).sdb_hashsize, sadb_alg_update_cb, \
7398         &update_state)
7399 
7400 void
7401 sadb_alg_update(ipsec_algtype_t alg_type, uint8_t alg_id, boolean_t is_added,
7402     netstack_t *ns)
7403 {
7404         struct sadb_update_alg_state update_state;
7405         ipsecah_stack_t *ahstack = ns->netstack_ipsecah;
7406         ipsecesp_stack_t        *espstack = ns->netstack_ipsecesp;
7407         ipsec_stack_t *ipss = ns->netstack_ipsec;
7408 
7409         update_state.alg_type = alg_type;
7410         update_state.alg_id = alg_id;
7411         update_state.is_added = is_added;
7412         update_state.async_auth = ipss->ipsec_algs_exec_mode[IPSEC_ALG_AUTH] ==
7413             IPSEC_ALGS_EXEC_ASYNC;
7414         update_state.async_encr = ipss->ipsec_algs_exec_mode[IPSEC_ALG_ENCR] ==
7415             IPSEC_ALGS_EXEC_ASYNC;
7416 
7417         if (alg_type == IPSEC_ALG_AUTH || alg_type == IPSEC_ALG_ALL) {
7418                 /* walk the AH tables only for auth. algorithm changes */
7419                 SADB_ALG_UPDATE_WALK(ahstack->ah_sadb.s_v4, sdb_of);
7420                 SADB_ALG_UPDATE_WALK(ahstack->ah_sadb.s_v4, sdb_if);
7421                 SADB_ALG_UPDATE_WALK(ahstack->ah_sadb.s_v6, sdb_of);
7422                 SADB_ALG_UPDATE_WALK(ahstack->ah_sadb.s_v6, sdb_if);
7423         }
7424 
7425         /* walk the ESP tables */
7426         SADB_ALG_UPDATE_WALK(espstack->esp_sadb.s_v4, sdb_of);
7427         SADB_ALG_UPDATE_WALK(espstack->esp_sadb.s_v4, sdb_if);
7428         SADB_ALG_UPDATE_WALK(espstack->esp_sadb.s_v6, sdb_of);
7429         SADB_ALG_UPDATE_WALK(espstack->esp_sadb.s_v6, sdb_if);
7430 }
7431 
7432 /*
7433  * Creates a context template for the specified SA. This function
7434  * is called when an SA is created and when a context template needs
7435  * to be created due to a change of software provider.
7436  */
7437 int
7438 ipsec_create_ctx_tmpl(ipsa_t *sa, ipsec_algtype_t alg_type)
7439 {
7440         ipsec_alginfo_t *alg;
7441         crypto_mechanism_t mech;
7442         crypto_key_t *key;
7443         crypto_ctx_template_t *sa_tmpl;
7444         int rv;
7445         ipsec_stack_t   *ipss = sa->ipsa_netstack->netstack_ipsec;
7446 
7447         ASSERT(RW_READ_HELD(&ipss->ipsec_alg_lock));
7448         ASSERT(MUTEX_HELD(&sa->ipsa_lock));
7449 
7450         /* get pointers to the algorithm info, context template, and key */
7451         switch (alg_type) {
7452         case IPSEC_ALG_AUTH:
7453                 key = &sa->ipsa_kcfauthkey;
7454                 sa_tmpl = &sa->ipsa_authtmpl;
7455                 alg = ipss->ipsec_alglists[alg_type][sa->ipsa_auth_alg];
7456                 break;
7457         case IPSEC_ALG_ENCR:
7458                 key = &sa->ipsa_kcfencrkey;
7459                 sa_tmpl = &sa->ipsa_encrtmpl;
7460                 alg = ipss->ipsec_alglists[alg_type][sa->ipsa_encr_alg];
7461                 break;
7462         default:
7463                 alg = NULL;
7464         }
7465 
7466         if (alg == NULL || !ALG_VALID(alg))
7467                 return (EINVAL);
7468 
7469         /* initialize the mech info structure for the framework */
7470         ASSERT(alg->alg_mech_type != CRYPTO_MECHANISM_INVALID);
7471         mech.cm_type = alg->alg_mech_type;
7472         mech.cm_param = NULL;
7473         mech.cm_param_len = 0;
7474 
7475         /* create a new context template */
7476         rv = crypto_create_ctx_template(&mech, key, sa_tmpl, KM_NOSLEEP);
7477 
7478         /*
7479          * CRYPTO_MECH_NOT_SUPPORTED can be returned if only hardware
7480          * providers are available for that mechanism. In that case
7481          * we don't fail, and will generate the context template from
7482          * the framework callback when a software provider for that
7483          * mechanism registers.
7484          *
7485          * The context template is assigned the special value
7486          * IPSEC_CTX_TMPL_ALLOC if the allocation failed due to a
7487          * lack of memory. No attempt will be made to use
7488          * the context template if it is set to this value.
7489          */
7490         if (rv == CRYPTO_HOST_MEMORY) {
7491                 *sa_tmpl = IPSEC_CTX_TMPL_ALLOC;
7492         } else if (rv != CRYPTO_SUCCESS) {
7493                 *sa_tmpl = NULL;
7494                 if (rv != CRYPTO_MECH_NOT_SUPPORTED)
7495                         return (EINVAL);
7496         }
7497 
7498         return (0);
7499 }
7500 
7501 /*
7502  * Destroy the context template of the specified algorithm type
7503  * of the specified SA. Must be called while holding the SA lock.
7504  */
7505 void
7506 ipsec_destroy_ctx_tmpl(ipsa_t *sa, ipsec_algtype_t alg_type)
7507 {
7508         ASSERT(MUTEX_HELD(&sa->ipsa_lock));
7509 
7510         if (alg_type == IPSEC_ALG_AUTH) {
7511                 if (sa->ipsa_authtmpl == IPSEC_CTX_TMPL_ALLOC)
7512                         sa->ipsa_authtmpl = NULL;
7513                 else if (sa->ipsa_authtmpl != NULL) {
7514                         crypto_destroy_ctx_template(sa->ipsa_authtmpl);
7515                         sa->ipsa_authtmpl = NULL;
7516                 }
7517         } else {
7518                 ASSERT(alg_type == IPSEC_ALG_ENCR);
7519                 if (sa->ipsa_encrtmpl == IPSEC_CTX_TMPL_ALLOC)
7520                         sa->ipsa_encrtmpl = NULL;
7521                 else if (sa->ipsa_encrtmpl != NULL) {
7522                         crypto_destroy_ctx_template(sa->ipsa_encrtmpl);
7523                         sa->ipsa_encrtmpl = NULL;
7524                 }
7525         }
7526 }
7527 
7528 /*
7529  * Use the kernel crypto framework to check the validity of a key received
7530  * via keysock. Returns 0 if the key is OK, -1 otherwise.
7531  */
7532 int
7533 ipsec_check_key(crypto_mech_type_t mech_type, sadb_key_t *sadb_key,
7534     boolean_t is_auth, int *diag)
7535 {
7536         crypto_mechanism_t mech;
7537         crypto_key_t crypto_key;
7538         int crypto_rc;
7539 
7540         mech.cm_type = mech_type;
7541         mech.cm_param = NULL;
7542         mech.cm_param_len = 0;
7543 
7544         crypto_key.ck_format = CRYPTO_KEY_RAW;
7545         crypto_key.ck_data = sadb_key + 1;
7546         crypto_key.ck_length = sadb_key->sadb_key_bits;
7547 
7548         crypto_rc = crypto_key_check(&mech, &crypto_key);
7549 
7550         switch (crypto_rc) {
7551         case CRYPTO_SUCCESS:
7552                 return (0);
7553         case CRYPTO_MECHANISM_INVALID:
7554         case CRYPTO_MECH_NOT_SUPPORTED:
7555                 *diag = is_auth ? SADB_X_DIAGNOSTIC_BAD_AALG :
7556                     SADB_X_DIAGNOSTIC_BAD_EALG;
7557                 break;
7558         case CRYPTO_KEY_SIZE_RANGE:
7559                 *diag = is_auth ? SADB_X_DIAGNOSTIC_BAD_AKEYBITS :
7560                     SADB_X_DIAGNOSTIC_BAD_EKEYBITS;
7561                 break;
7562         case CRYPTO_WEAK_KEY:
7563                 *diag = is_auth ? SADB_X_DIAGNOSTIC_WEAK_AKEY :
7564                     SADB_X_DIAGNOSTIC_WEAK_EKEY;
7565                 break;
7566         }
7567 
7568         return (-1);
7569 }
7570 
7571 /*
7572  * Whack options in the outer IP header when ipsec changes the outer label
7573  *
7574  * This is inelegant and really could use refactoring.
7575  */
7576 mblk_t *
7577 sadb_whack_label_v4(mblk_t *mp, ipsa_t *assoc, kstat_named_t *counter,
7578     ipdropper_t *dropper)
7579 {
7580         int delta;
7581         int plen;
7582         dblk_t *db;
7583         int hlen;
7584         uint8_t *opt_storage = assoc->ipsa_opt_storage;
7585         ipha_t *ipha = (ipha_t *)mp->b_rptr;
7586 
7587         plen = ntohs(ipha->ipha_length);
7588 
7589         delta = tsol_remove_secopt(ipha, MBLKL(mp));
7590         mp->b_wptr += delta;
7591         plen += delta;
7592 
7593         /* XXX XXX code copied from tsol_check_label */
7594 
7595         /* Make sure we have room for the worst-case addition */
7596         hlen = IPH_HDR_LENGTH(ipha) + opt_storage[IPOPT_OLEN];
7597         hlen = (hlen + 3) & ~3;
7598         if (hlen > IP_MAX_HDR_LENGTH)
7599                 hlen = IP_MAX_HDR_LENGTH;
7600         hlen -= IPH_HDR_LENGTH(ipha);
7601 
7602         db = mp->b_datap;
7603         if ((db->db_ref != 1) || (mp->b_wptr + hlen > db->db_lim)) {
7604                 int copylen;
7605                 mblk_t *new_mp;
7606 
7607                 /* allocate enough to be meaningful, but not *too* much */
7608                 copylen = MBLKL(mp);
7609                 if (copylen > 256)
7610                         copylen = 256;
7611                 new_mp = allocb_tmpl(hlen + copylen +
7612                     (mp->b_rptr - mp->b_datap->db_base), mp);
7613 
7614                 if (new_mp == NULL) {
7615                         ip_drop_packet(mp, B_FALSE, NULL, counter,  dropper);
7616                         return (NULL);
7617                 }
7618 
7619                 /* keep the bias */
7620                 new_mp->b_rptr += mp->b_rptr - mp->b_datap->db_base;
7621                 new_mp->b_wptr = new_mp->b_rptr + copylen;
7622                 bcopy(mp->b_rptr, new_mp->b_rptr, copylen);
7623                 new_mp->b_cont = mp;
7624                 if ((mp->b_rptr += copylen) >= mp->b_wptr) {
7625                         new_mp->b_cont = mp->b_cont;
7626                         freeb(mp);
7627                 }
7628                 mp = new_mp;
7629                 ipha = (ipha_t *)mp->b_rptr;
7630         }
7631 
7632         delta = tsol_prepend_option(assoc->ipsa_opt_storage, ipha, MBLKL(mp));
7633 
7634         ASSERT(delta != -1);
7635 
7636         plen += delta;
7637         mp->b_wptr += delta;
7638 
7639         /*
7640          * Paranoia
7641          */
7642         db = mp->b_datap;
7643 
7644         ASSERT3P(mp->b_wptr, <=, db->db_lim);
7645         ASSERT3P(mp->b_rptr, <=, db->db_lim);
7646 
7647         ASSERT3P(mp->b_wptr, >=, db->db_base);
7648         ASSERT3P(mp->b_rptr, >=, db->db_base);
7649         /* End paranoia */
7650 
7651         ipha->ipha_length = htons(plen);
7652 
7653         return (mp);
7654 }
7655 
7656 mblk_t *
7657 sadb_whack_label_v6(mblk_t *mp, ipsa_t *assoc, kstat_named_t *counter,
7658     ipdropper_t *dropper)
7659 {
7660         int delta;
7661         int plen;
7662         dblk_t *db;
7663         int hlen;
7664         uint8_t *opt_storage = assoc->ipsa_opt_storage;
7665         uint_t sec_opt_len; /* label option length not including type, len */
7666         ip6_t *ip6h = (ip6_t *)mp->b_rptr;
7667 
7668         plen = ntohs(ip6h->ip6_plen);
7669 
7670         delta = tsol_remove_secopt_v6(ip6h, MBLKL(mp));
7671         mp->b_wptr += delta;
7672         plen += delta;
7673 
7674         /* XXX XXX code copied from tsol_check_label_v6 */
7675         /*
7676          * Make sure we have room for the worst-case addition. Add 2 bytes for
7677          * the hop-by-hop ext header's next header and length fields. Add
7678          * another 2 bytes for the label option type, len and then round
7679          * up to the next 8-byte multiple.
7680          */
7681         sec_opt_len = opt_storage[1];
7682 
7683         db = mp->b_datap;
7684         hlen = (4 + sec_opt_len + 7) & ~7;
7685 
7686         if ((db->db_ref != 1) || (mp->b_wptr + hlen > db->db_lim)) {
7687                 int copylen;
7688                 mblk_t *new_mp;
7689                 uint16_t hdr_len;
7690 
7691                 hdr_len = ip_hdr_length_v6(mp, ip6h);
7692                 /*
7693                  * Allocate enough to be meaningful, but not *too* much.
7694                  * Also all the IPv6 extension headers must be in the same mblk
7695                  */
7696                 copylen = MBLKL(mp);
7697                 if (copylen > 256)
7698                         copylen = 256;
7699                 if (copylen < hdr_len)
7700                         copylen = hdr_len;
7701                 new_mp = allocb_tmpl(hlen + copylen +
7702                     (mp->b_rptr - mp->b_datap->db_base), mp);
7703                 if (new_mp == NULL) {
7704                         ip_drop_packet(mp, B_FALSE, NULL, counter,  dropper);
7705                         return (NULL);
7706                 }
7707 
7708                 /* keep the bias */
7709                 new_mp->b_rptr += mp->b_rptr - mp->b_datap->db_base;
7710                 new_mp->b_wptr = new_mp->b_rptr + copylen;
7711                 bcopy(mp->b_rptr, new_mp->b_rptr, copylen);
7712                 new_mp->b_cont = mp;
7713                 if ((mp->b_rptr += copylen) >= mp->b_wptr) {
7714                         new_mp->b_cont = mp->b_cont;
7715                         freeb(mp);
7716                 }
7717                 mp = new_mp;
7718                 ip6h = (ip6_t *)mp->b_rptr;
7719         }
7720 
7721         delta = tsol_prepend_option_v6(assoc->ipsa_opt_storage,
7722             ip6h, MBLKL(mp));
7723 
7724         ASSERT(delta != -1);
7725 
7726         plen += delta;
7727         mp->b_wptr += delta;
7728 
7729         /*
7730          * Paranoia
7731          */
7732         db = mp->b_datap;
7733 
7734         ASSERT3P(mp->b_wptr, <=, db->db_lim);
7735         ASSERT3P(mp->b_rptr, <=, db->db_lim);
7736 
7737         ASSERT3P(mp->b_wptr, >=, db->db_base);
7738         ASSERT3P(mp->b_rptr, >=, db->db_base);
7739         /* End paranoia */
7740 
7741         ip6h->ip6_plen = htons(plen);
7742 
7743         return (mp);
7744 }
7745 
7746 /* Whack the labels and update ip_xmit_attr_t as needed */
7747 mblk_t *
7748 sadb_whack_label(mblk_t *mp, ipsa_t *assoc, ip_xmit_attr_t *ixa,
7749     kstat_named_t *counter, ipdropper_t *dropper)
7750 {
7751         int adjust;
7752         int iplen;
7753 
7754         if (ixa->ixa_flags & IXAF_IS_IPV4) {
7755                 ipha_t          *ipha = (ipha_t *)mp->b_rptr;
7756 
7757                 ASSERT(IPH_HDR_VERSION(ipha) == IPV4_VERSION);
7758                 iplen = ntohs(ipha->ipha_length);
7759                 mp = sadb_whack_label_v4(mp, assoc, counter, dropper);
7760                 if (mp == NULL)
7761                         return (NULL);
7762 
7763                 ipha = (ipha_t *)mp->b_rptr;
7764                 ASSERT(IPH_HDR_VERSION(ipha) == IPV4_VERSION);
7765                 adjust = (int)ntohs(ipha->ipha_length) - iplen;
7766         } else {
7767                 ip6_t           *ip6h = (ip6_t *)mp->b_rptr;
7768 
7769                 ASSERT(IPH_HDR_VERSION(ip6h) == IPV6_VERSION);
7770                 iplen = ntohs(ip6h->ip6_plen);
7771                 mp = sadb_whack_label_v6(mp, assoc, counter, dropper);
7772                 if (mp == NULL)
7773                         return (NULL);
7774 
7775                 ip6h = (ip6_t *)mp->b_rptr;
7776                 ASSERT(IPH_HDR_VERSION(ip6h) == IPV6_VERSION);
7777                 adjust = (int)ntohs(ip6h->ip6_plen) - iplen;
7778         }
7779         ixa->ixa_pktlen += adjust;
7780         ixa->ixa_ip_hdr_length += adjust;
7781         return (mp);
7782 }
7783 
7784 /*
7785  * If this is an outgoing SA then add some fuzz to the
7786  * SOFT EXPIRE time. The reason for this is to stop
7787  * peers trying to renegotiate SOFT expiring SA's at
7788  * the same time. The amount of fuzz needs to be at
7789  * least 8 seconds which is the typical interval
7790  * sadb_ager(), although this is only a guide as it
7791  * selftunes.
7792  */
7793 static void
7794 lifetime_fuzz(ipsa_t *assoc)
7795 {
7796         uint8_t rnd;
7797 
7798         if (assoc->ipsa_softaddlt == 0)
7799                 return;
7800 
7801         (void) random_get_pseudo_bytes(&rnd, sizeof (rnd));
7802         rnd = (rnd & 0xF) + 8;
7803         assoc->ipsa_softexpiretime -= rnd;
7804         assoc->ipsa_softaddlt -= rnd;
7805 }
7806 
7807 static void
7808 destroy_ipsa_pair(ipsap_t *ipsapp)
7809 {
7810         /*
7811          * Because of the multi-line macro nature of IPSA_REFRELE, keep
7812          * them in { }.
7813          */
7814         if (ipsapp->ipsap_sa_ptr != NULL) {
7815                 IPSA_REFRELE(ipsapp->ipsap_sa_ptr);
7816         }
7817         if (ipsapp->ipsap_psa_ptr != NULL) {
7818                 IPSA_REFRELE(ipsapp->ipsap_psa_ptr);
7819         }
7820         init_ipsa_pair(ipsapp);
7821 }
7822 
7823 static void
7824 init_ipsa_pair(ipsap_t *ipsapp)
7825 {
7826         ipsapp->ipsap_bucket = NULL;
7827         ipsapp->ipsap_sa_ptr = NULL;
7828         ipsapp->ipsap_pbucket = NULL;
7829         ipsapp->ipsap_psa_ptr = NULL;
7830 }
7831 
7832 /*
7833  * The sadb_ager() function walks through the hash tables of SA's and ages
7834  * them, if the SA expires as a result, its marked as DEAD and will be reaped
7835  * the next time sadb_ager() runs. SA's which are paired or have a peer (same
7836  * SA appears in both the inbound and outbound tables because its not possible
7837  * to determine its direction) are placed on a list when they expire. This is
7838  * to ensure that pair/peer SA's are reaped at the same time, even if they
7839  * expire at different times.
7840  *
7841  * This function is called twice by sadb_ager(), one after processing the
7842  * inbound table, then again after processing the outbound table.
7843  */
7844 void
7845 age_pair_peer_list(templist_t *haspeerlist, sadb_t *sp, boolean_t outbound)
7846 {
7847         templist_t *listptr;
7848         int outhash;
7849         isaf_t *bucket;
7850         boolean_t haspeer;
7851         ipsa_t *peer_assoc, *dying;
7852         /*
7853          * Haspeer cases will contain both IPv4 and IPv6.  This code
7854          * is address independent.
7855          */
7856         while (haspeerlist != NULL) {
7857                 /* "dying" contains the SA that has a peer. */
7858                 dying = haspeerlist->ipsa;
7859                 haspeer = (dying->ipsa_haspeer);
7860                 listptr = haspeerlist;
7861                 haspeerlist = listptr->next;
7862                 kmem_free(listptr, sizeof (*listptr));
7863                 /*
7864                  * Pick peer bucket based on addrfam.
7865                  */
7866                 if (outbound) {
7867                         if (haspeer)
7868                                 bucket = INBOUND_BUCKET(sp, dying->ipsa_spi);
7869                         else
7870                                 bucket = INBOUND_BUCKET(sp,
7871                                     dying->ipsa_otherspi);
7872                 } else { /* inbound */
7873                         if (haspeer) {
7874                                 if (dying->ipsa_addrfam == AF_INET6) {
7875                                         outhash = OUTBOUND_HASH_V6(sp,
7876                                             *((in6_addr_t *)&dying->
7877                                             ipsa_dstaddr));
7878                                 } else {
7879                                         outhash = OUTBOUND_HASH_V4(sp,
7880                                             *((ipaddr_t *)&dying->
7881                                             ipsa_dstaddr));
7882                                 }
7883                         } else if (dying->ipsa_addrfam == AF_INET6) {
7884                                 outhash = OUTBOUND_HASH_V6(sp,
7885                                     *((in6_addr_t *)&dying->
7886                                     ipsa_srcaddr));
7887                         } else {
7888                                 outhash = OUTBOUND_HASH_V4(sp,
7889                                     *((ipaddr_t *)&dying->
7890                                     ipsa_srcaddr));
7891                         }
7892                         bucket = &(sp->sdb_of[outhash]);
7893                 }
7894 
7895                 mutex_enter(&bucket->isaf_lock);
7896                 /*
7897                  * "haspeer" SA's have the same src/dst address ordering,
7898                  * "paired" SA's have the src/dst addresses reversed.
7899                  */
7900                 if (haspeer) {
7901                         peer_assoc = ipsec_getassocbyspi(bucket,
7902                             dying->ipsa_spi, dying->ipsa_srcaddr,
7903                             dying->ipsa_dstaddr, dying->ipsa_addrfam);
7904                 } else {
7905                         peer_assoc = ipsec_getassocbyspi(bucket,
7906                             dying->ipsa_otherspi, dying->ipsa_dstaddr,
7907                             dying->ipsa_srcaddr, dying->ipsa_addrfam);
7908                 }
7909 
7910                 mutex_exit(&bucket->isaf_lock);
7911                 if (peer_assoc != NULL) {
7912                         mutex_enter(&peer_assoc->ipsa_lock);
7913                         mutex_enter(&dying->ipsa_lock);
7914                         if (!haspeer) {
7915                                 /*
7916                                  * Only SA's which have a "peer" or are
7917                                  * "paired" end up on this list, so this
7918                                  * must be a "paired" SA, update the flags
7919                                  * to break the pair.
7920                                  */
7921                                 peer_assoc->ipsa_otherspi = 0;
7922                                 peer_assoc->ipsa_flags &= ~IPSA_F_PAIRED;
7923                                 dying->ipsa_otherspi = 0;
7924                                 dying->ipsa_flags &= ~IPSA_F_PAIRED;
7925                         }
7926                         if (haspeer || outbound) {
7927                                 /*
7928                                  * Update the state of the "inbound" SA when
7929                                  * the "outbound" SA has expired. Don't update
7930                                  * the "outbound" SA when the "inbound" SA
7931                                  * SA expires because setting the hard_addtime
7932                                  * below will cause this to happen.
7933                                  */
7934                                 peer_assoc->ipsa_state = dying->ipsa_state;
7935                         }
7936                         if (dying->ipsa_state == IPSA_STATE_DEAD)
7937                                 peer_assoc->ipsa_hardexpiretime = 1;
7938 
7939                         mutex_exit(&dying->ipsa_lock);
7940                         mutex_exit(&peer_assoc->ipsa_lock);
7941                         IPSA_REFRELE(peer_assoc);
7942                 }
7943                 IPSA_REFRELE(dying);
7944         }
7945 }
7946 
7947 /*
7948  * Ensure that the IV used for CCM mode never repeats. The IV should
7949  * only be updated by this function. Also check to see if the IV
7950  * is about to wrap and generate a SOFT Expire. This function is only
7951  * called for outgoing packets, the IV for incomming packets is taken
7952  * from the wire. If the outgoing SA needs to be expired, update
7953  * the matching incomming SA.
7954  */
7955 boolean_t
7956 update_iv(uint8_t *iv_ptr, queue_t *pfkey_q, ipsa_t *assoc,
7957     ipsecesp_stack_t *espstack)
7958 {
7959         boolean_t rc = B_TRUE;
7960         isaf_t *inbound_bucket;
7961         sadb_t *sp;
7962         ipsa_t *pair_sa = NULL;
7963         int sa_new_state = 0;
7964 
7965         /* For non counter modes, the IV is random data. */
7966         if (!(assoc->ipsa_flags & IPSA_F_COUNTERMODE)) {
7967                 (void) random_get_pseudo_bytes(iv_ptr, assoc->ipsa_iv_len);
7968                 return (rc);
7969         }
7970 
7971         mutex_enter(&assoc->ipsa_lock);
7972 
7973         (*assoc->ipsa_iv)++;
7974 
7975         if (*assoc->ipsa_iv == assoc->ipsa_iv_hardexpire) {
7976                 sa_new_state = IPSA_STATE_DEAD;
7977                 rc = B_FALSE;
7978         } else if (*assoc->ipsa_iv == assoc->ipsa_iv_softexpire) {
7979                 if (assoc->ipsa_state != IPSA_STATE_DYING) {
7980                         /*
7981                          * This SA may have already been expired when its
7982                          * PAIR_SA expired.
7983                          */
7984                         sa_new_state = IPSA_STATE_DYING;
7985                 }
7986         }
7987         if (sa_new_state) {
7988                 /*
7989                  * If there is a state change, we need to update this SA
7990                  * and its "pair", we can find the bucket for the "pair" SA
7991                  * while holding the ipsa_t mutex, but we won't actually
7992                  * update anything untill the ipsa_t mutex has been released
7993                  * for _this_ SA.
7994                  */
7995                 assoc->ipsa_state = sa_new_state;
7996                 if (assoc->ipsa_addrfam == AF_INET6) {
7997                         sp = &espstack->esp_sadb.s_v6;
7998                 } else {
7999                         sp = &espstack->esp_sadb.s_v4;
8000                 }
8001                 inbound_bucket = INBOUND_BUCKET(sp, assoc->ipsa_otherspi);
8002                 sadb_expire_assoc(pfkey_q, assoc);
8003         }
8004         if (rc == B_TRUE)
8005                 bcopy(assoc->ipsa_iv, iv_ptr, assoc->ipsa_iv_len);
8006 
8007         mutex_exit(&assoc->ipsa_lock);
8008 
8009         if (sa_new_state) {
8010                 /* Find the inbound SA, need to lock hash bucket. */
8011                 mutex_enter(&inbound_bucket->isaf_lock);
8012                 pair_sa = ipsec_getassocbyspi(inbound_bucket,
8013                     assoc->ipsa_otherspi, assoc->ipsa_dstaddr,
8014                     assoc->ipsa_srcaddr, assoc->ipsa_addrfam);
8015                 mutex_exit(&inbound_bucket->isaf_lock);
8016                 if (pair_sa != NULL) {
8017                         mutex_enter(&pair_sa->ipsa_lock);
8018                         pair_sa->ipsa_state = sa_new_state;
8019                         mutex_exit(&pair_sa->ipsa_lock);
8020                         IPSA_REFRELE(pair_sa);
8021                 }
8022         }
8023 
8024         return (rc);
8025 }
8026 
8027 void
8028 ccm_params_init(ipsa_t *assoc, uchar_t *esph, uint_t data_len, uchar_t *iv_ptr,
8029     ipsa_cm_mech_t *cm_mech, crypto_data_t *crypto_data)
8030 {
8031         uchar_t *nonce;
8032         crypto_mechanism_t *combined_mech;
8033         CK_AES_CCM_PARAMS *params;
8034 
8035         combined_mech = (crypto_mechanism_t *)cm_mech;
8036         params = (CK_AES_CCM_PARAMS *)(combined_mech + 1);
8037         nonce = (uchar_t *)(params + 1);
8038         params->ulMACSize = assoc->ipsa_mac_len;
8039         params->ulNonceSize = assoc->ipsa_nonce_len;
8040         params->ulAuthDataSize = sizeof (esph_t);
8041         params->ulDataSize = data_len;
8042         params->nonce = nonce;
8043         params->authData = esph;
8044 
8045         cm_mech->combined_mech.cm_type = assoc->ipsa_emech.cm_type;
8046         cm_mech->combined_mech.cm_param_len = sizeof (CK_AES_CCM_PARAMS);
8047         cm_mech->combined_mech.cm_param = (caddr_t)params;
8048         /* See gcm_params_init() for comments. */
8049         bcopy(assoc->ipsa_nonce, nonce, assoc->ipsa_saltlen);
8050         nonce += assoc->ipsa_saltlen;
8051         bcopy(iv_ptr, nonce, assoc->ipsa_iv_len);
8052         crypto_data->cd_miscdata = NULL;
8053 }
8054 
8055 /* ARGSUSED */
8056 void
8057 cbc_params_init(ipsa_t *assoc, uchar_t *esph, uint_t data_len, uchar_t *iv_ptr,
8058     ipsa_cm_mech_t *cm_mech, crypto_data_t *crypto_data)
8059 {
8060         cm_mech->combined_mech.cm_type = assoc->ipsa_emech.cm_type;
8061         cm_mech->combined_mech.cm_param_len = 0;
8062         cm_mech->combined_mech.cm_param = NULL;
8063         crypto_data->cd_miscdata = (char *)iv_ptr;
8064 }
8065 
8066 /* ARGSUSED */
8067 void
8068 gcm_params_init(ipsa_t *assoc, uchar_t *esph, uint_t data_len, uchar_t *iv_ptr,
8069     ipsa_cm_mech_t *cm_mech, crypto_data_t *crypto_data)
8070 {
8071         uchar_t *nonce;
8072         crypto_mechanism_t *combined_mech;
8073         CK_AES_GCM_PARAMS *params;
8074 
8075         combined_mech = (crypto_mechanism_t *)cm_mech;
8076         params = (CK_AES_GCM_PARAMS *)(combined_mech + 1);
8077         nonce = (uchar_t *)(params + 1);
8078 
8079         params->pIv = nonce;
8080         params->ulIvLen = assoc->ipsa_nonce_len;
8081         params->ulIvBits = SADB_8TO1(assoc->ipsa_nonce_len);
8082         params->pAAD = esph;
8083         params->ulAADLen = sizeof (esph_t);
8084         params->ulTagBits = SADB_8TO1(assoc->ipsa_mac_len);
8085 
8086         cm_mech->combined_mech.cm_type = assoc->ipsa_emech.cm_type;
8087         cm_mech->combined_mech.cm_param_len = sizeof (CK_AES_GCM_PARAMS);
8088         cm_mech->combined_mech.cm_param = (caddr_t)params;
8089         /*
8090          * Create the nonce, which is made up of the salt and the IV.
8091          * Copy the salt from the SA and the IV from the packet.
8092          * For inbound packets we copy the IV from the packet because it
8093          * was set by the sending system, for outbound packets we copy the IV
8094          * from the packet because the IV in the SA may be changed by another
8095          * thread, the IV in the packet was created while holding a mutex.
8096          */
8097         bcopy(assoc->ipsa_nonce, nonce, assoc->ipsa_saltlen);
8098         nonce += assoc->ipsa_saltlen;
8099         bcopy(iv_ptr, nonce, assoc->ipsa_iv_len);
8100         crypto_data->cd_miscdata = NULL;
8101 }