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                 /*
2334                  * Be liberal in what we receive.  Special-case the IKEv1
2335                  * cookie, which closed-source in.iked assumes is 32 bits.
2336                  * Now that we store all 64 bits, we should pre-zero the
2337                  * reserved field on behalf of closed-source in.iked.
2338                  */
2339                 if (sq->kmp == SADB_X_KMP_IKE) {
2340                         /* Just in case in.iked is misbehaving... */
2341                         sq->kmcext->sadb_x_kmc_reserved = 0;
2342                 }
2343                 sq->kmc = sq->kmcext->sadb_x_kmc_cookie64;
2344                 *mfpp++ = sadb_match_kmc;
2345         }
2346 
2347         if (match & (IPSA_Q_INBOUND|IPSA_Q_OUTBOUND)) {
2348                 if (sq->af == AF_INET6)
2349                         sq->sp = &sq->spp->s_v6;
2350                 else
2351                         sq->sp = &sq->spp->s_v4;
2352         } else {
2353                 sq->sp = NULL;
2354         }
2355 
2356         if (match & IPSA_Q_INBOUND) {
2357                 sq->inhash = INBOUND_HASH(sq->sp, sq->assoc->sadb_sa_spi);
2358                 sq->inbound = &sq->sp->sdb_if[sq->inhash];
2359         } else {
2360                 sq->inhash = 0;
2361                 sq->inbound = NULL;
2362         }
2363 
2364         if (match & IPSA_Q_OUTBOUND) {
2365                 if (sq->af == AF_INET6) {
2366                         sq->outhash = OUTBOUND_HASH_V6(sq->sp, *(sq->dstaddr));
2367                 } else {
2368                         sq->outhash = OUTBOUND_HASH_V4(sq->sp, *(sq->dstaddr));
2369                 }
2370                 sq->outbound = &sq->sp->sdb_of[sq->outhash];
2371         } else {
2372                 sq->outhash = 0;
2373                 sq->outbound = NULL;
2374         }
2375         sq->match = match;
2376         return (0);
2377 }
2378 
2379 /*
2380  * Match an initialized query structure with a security association;
2381  * return B_TRUE on a match, B_FALSE on a miss.
2382  * Applies match functions set up by sadb_form_query() until one returns false.
2383  */
2384 boolean_t
2385 sadb_match_query(ipsa_query_t *sq, ipsa_t *sa)
2386 {
2387         ipsa_match_fn_t *mfpp = &(sq->matchers[0]);
2388         ipsa_match_fn_t mfp;
2389 
2390         for (mfp = *mfpp++; mfp != NULL; mfp = *mfpp++) {
2391                 if (!mfp(sq, sa))
2392                         return (B_FALSE);
2393         }
2394         return (B_TRUE);
2395 }
2396 
2397 /*
2398  * Walker callback function to delete sa's based on src/dst address.
2399  * Assumes that we're called with *head locked, no other locks held;
2400  * Conveniently, and not coincidentally, this is both what sadb_walker
2401  * gives us and also what sadb_unlinkassoc expects.
2402  */
2403 struct sadb_purge_state
2404 {
2405         ipsa_query_t sq;
2406         boolean_t inbnd;
2407         uint8_t sadb_sa_state;
2408 };
2409 
2410 static void
2411 sadb_purge_cb(isaf_t *head, ipsa_t *entry, void *cookie)
2412 {
2413         struct sadb_purge_state *ps = (struct sadb_purge_state *)cookie;
2414 
2415         ASSERT(MUTEX_HELD(&head->isaf_lock));
2416 
2417         mutex_enter(&entry->ipsa_lock);
2418 
2419         if (entry->ipsa_state == IPSA_STATE_LARVAL ||
2420             !sadb_match_query(&ps->sq, entry)) {
2421                 mutex_exit(&entry->ipsa_lock);
2422                 return;
2423         }
2424 
2425         if (ps->inbnd) {
2426                 sadb_delete_cluster(entry);
2427         }
2428         entry->ipsa_state = IPSA_STATE_DEAD;
2429         (void) sadb_torch_assoc(head, entry);
2430 }
2431 
2432 /*
2433  * Common code to purge an SA with a matching src or dst address.
2434  * Don't kill larval SA's in such a purge.
2435  */
2436 int
2437 sadb_purge_sa(mblk_t *mp, keysock_in_t *ksi, sadb_t *sp,
2438     int *diagnostic, queue_t *pfkey_q)
2439 {
2440         struct sadb_purge_state ps;
2441         int error = sadb_form_query(ksi, 0,
2442             IPSA_Q_SRC|IPSA_Q_DST|IPSA_Q_SRCID|IPSA_Q_DSTID|IPSA_Q_KMC,
2443             &ps.sq, diagnostic);
2444 
2445         if (error != 0)
2446                 return (error);
2447 
2448         /*
2449          * This is simple, crude, and effective.
2450          * Unimplemented optimizations (TBD):
2451          * - we can limit how many places we search based on where we
2452          * think the SA is filed.
2453          * - if we get a dst address, we can hash based on dst addr to find
2454          * the correct bucket in the outbound table.
2455          */
2456         ps.inbnd = B_TRUE;
2457         sadb_walker(sp->sdb_if, sp->sdb_hashsize, sadb_purge_cb, &ps);
2458         ps.inbnd = B_FALSE;
2459         sadb_walker(sp->sdb_of, sp->sdb_hashsize, sadb_purge_cb, &ps);
2460 
2461         ASSERT(mp->b_cont != NULL);
2462         sadb_pfkey_echo(pfkey_q, mp, (sadb_msg_t *)mp->b_cont->b_rptr, ksi,
2463             NULL);
2464         return (0);
2465 }
2466 
2467 static void
2468 sadb_delpair_state_one(isaf_t *head, ipsa_t *entry, void *cookie)
2469 {
2470         struct sadb_purge_state *ps = (struct sadb_purge_state *)cookie;
2471         isaf_t  *inbound_bucket;
2472         ipsa_t *peer_assoc;
2473         ipsa_query_t *sq = &ps->sq;
2474 
2475         ASSERT(MUTEX_HELD(&head->isaf_lock));
2476 
2477         mutex_enter(&entry->ipsa_lock);
2478 
2479         if ((entry->ipsa_state != ps->sadb_sa_state) ||
2480             ((sq->srcaddr != NULL) &&
2481             !IPSA_ARE_ADDR_EQUAL(entry->ipsa_srcaddr, sq->srcaddr, sq->af))) {
2482                 mutex_exit(&entry->ipsa_lock);
2483                 return;
2484         }
2485 
2486         /*
2487          * The isaf_t *, which is passed in , is always an outbound bucket,
2488          * and we are preserving the outbound-then-inbound hash-bucket lock
2489          * ordering. The sadb_walker() which triggers this function is called
2490          * only on the outbound fanout, and the corresponding inbound bucket
2491          * lock is safe to acquire here.
2492          */
2493 
2494         if (entry->ipsa_haspeer) {
2495                 inbound_bucket = INBOUND_BUCKET(sq->sp, entry->ipsa_spi);
2496                 mutex_enter(&inbound_bucket->isaf_lock);
2497                 peer_assoc = ipsec_getassocbyspi(inbound_bucket,
2498                     entry->ipsa_spi, entry->ipsa_srcaddr,
2499                     entry->ipsa_dstaddr, entry->ipsa_addrfam);
2500         } else {
2501                 inbound_bucket = INBOUND_BUCKET(sq->sp, entry->ipsa_otherspi);
2502                 mutex_enter(&inbound_bucket->isaf_lock);
2503                 peer_assoc = ipsec_getassocbyspi(inbound_bucket,
2504                     entry->ipsa_otherspi, entry->ipsa_dstaddr,
2505                     entry->ipsa_srcaddr, entry->ipsa_addrfam);
2506         }
2507 
2508         entry->ipsa_state = IPSA_STATE_DEAD;
2509         (void) sadb_torch_assoc(head, entry);
2510         if (peer_assoc != NULL) {
2511                 mutex_enter(&peer_assoc->ipsa_lock);
2512                 peer_assoc->ipsa_state = IPSA_STATE_DEAD;
2513                 (void) sadb_torch_assoc(inbound_bucket, peer_assoc);
2514         }
2515         mutex_exit(&inbound_bucket->isaf_lock);
2516 }
2517 
2518 static int
2519 sadb_delpair_state(mblk_t *mp, keysock_in_t *ksi, sadbp_t *spp,
2520     int *diagnostic, queue_t *pfkey_q)
2521 {
2522         sadb_sa_t *assoc = (sadb_sa_t *)ksi->ks_in_extv[SADB_EXT_SA];
2523         struct sadb_purge_state ps;
2524         int error;
2525 
2526         ps.sq.spp = spp;                /* XXX param */
2527 
2528         error = sadb_form_query(ksi, IPSA_Q_DST|IPSA_Q_SRC,
2529             IPSA_Q_SRC|IPSA_Q_DST|IPSA_Q_SRCID|IPSA_Q_DSTID|IPSA_Q_KMC,
2530             &ps.sq, diagnostic);
2531         if (error != 0)
2532                 return (error);
2533 
2534         ps.inbnd = B_FALSE;
2535         ps.sadb_sa_state = assoc->sadb_sa_state;
2536         sadb_walker(ps.sq.sp->sdb_of, ps.sq.sp->sdb_hashsize,
2537             sadb_delpair_state_one, &ps);
2538 
2539         ASSERT(mp->b_cont != NULL);
2540         sadb_pfkey_echo(pfkey_q, mp, (sadb_msg_t *)mp->b_cont->b_rptr,
2541             ksi, NULL);
2542         return (0);
2543 }
2544 
2545 /*
2546  * Common code to delete/get an SA.
2547  */
2548 int
2549 sadb_delget_sa(mblk_t *mp, keysock_in_t *ksi, sadbp_t *spp,
2550     int *diagnostic, queue_t *pfkey_q, uint8_t sadb_msg_type)
2551 {
2552         ipsa_query_t sq;
2553         ipsa_t *echo_target = NULL;
2554         ipsap_t ipsapp;
2555         uint_t  error = 0;
2556 
2557         if (sadb_msg_type == SADB_X_DELPAIR_STATE)
2558                 return (sadb_delpair_state(mp, ksi, spp, diagnostic, pfkey_q));
2559 
2560         sq.spp = spp;           /* XXX param */
2561         error = sadb_form_query(ksi, IPSA_Q_DST|IPSA_Q_SA,
2562             IPSA_Q_SRC|IPSA_Q_DST|IPSA_Q_SA|IPSA_Q_INBOUND|IPSA_Q_OUTBOUND,
2563             &sq, diagnostic);
2564         if (error != 0)
2565                 return (error);
2566 
2567         error = get_ipsa_pair(&sq, &ipsapp, diagnostic);
2568         if (error != 0) {
2569                 return (error);
2570         }
2571 
2572         echo_target = ipsapp.ipsap_sa_ptr;
2573         if (echo_target == NULL)
2574                 echo_target = ipsapp.ipsap_psa_ptr;
2575 
2576         if (sadb_msg_type == SADB_DELETE || sadb_msg_type == SADB_X_DELPAIR) {
2577                 /*
2578                  * Bucket locks will be required if SA is actually unlinked.
2579                  * get_ipsa_pair() returns valid hash bucket pointers even
2580                  * if it can't find a pair SA pointer. To prevent a potential
2581                  * deadlock, always lock the outbound bucket before the inbound.
2582                  */
2583                 if (ipsapp.in_inbound_table) {
2584                         mutex_enter(&ipsapp.ipsap_pbucket->isaf_lock);
2585                         mutex_enter(&ipsapp.ipsap_bucket->isaf_lock);
2586                 } else {
2587                         mutex_enter(&ipsapp.ipsap_bucket->isaf_lock);
2588                         mutex_enter(&ipsapp.ipsap_pbucket->isaf_lock);
2589                 }
2590 
2591                 if (ipsapp.ipsap_sa_ptr != NULL) {
2592                         mutex_enter(&ipsapp.ipsap_sa_ptr->ipsa_lock);
2593                         if (ipsapp.ipsap_sa_ptr->ipsa_flags & IPSA_F_INBOUND) {
2594                                 sadb_delete_cluster(ipsapp.ipsap_sa_ptr);
2595                         }
2596                         ipsapp.ipsap_sa_ptr->ipsa_state = IPSA_STATE_DEAD;
2597                         (void) sadb_torch_assoc(ipsapp.ipsap_bucket,
2598                             ipsapp.ipsap_sa_ptr);
2599                         /*
2600                          * sadb_torch_assoc() releases the ipsa_lock
2601                          * and calls sadb_unlinkassoc() which does a
2602                          * IPSA_REFRELE.
2603                          */
2604                 }
2605                 if (ipsapp.ipsap_psa_ptr != NULL) {
2606                         mutex_enter(&ipsapp.ipsap_psa_ptr->ipsa_lock);
2607                         if (sadb_msg_type == SADB_X_DELPAIR ||
2608                             ipsapp.ipsap_psa_ptr->ipsa_haspeer) {
2609                                 if (ipsapp.ipsap_psa_ptr->ipsa_flags &
2610                                     IPSA_F_INBOUND) {
2611                                         sadb_delete_cluster
2612                                             (ipsapp.ipsap_psa_ptr);
2613                                 }
2614                                 ipsapp.ipsap_psa_ptr->ipsa_state =
2615                                     IPSA_STATE_DEAD;
2616                                 (void) sadb_torch_assoc(ipsapp.ipsap_pbucket,
2617                                     ipsapp.ipsap_psa_ptr);
2618                         } else {
2619                                 /*
2620                                  * Only half of the "pair" has been deleted.
2621                                  * Update the remaining SA and remove references
2622                                  * to its pair SA, which is now gone.
2623                                  */
2624                                 ipsapp.ipsap_psa_ptr->ipsa_otherspi = 0;
2625                                 ipsapp.ipsap_psa_ptr->ipsa_flags &=
2626                                     ~IPSA_F_PAIRED;
2627                                 mutex_exit(&ipsapp.ipsap_psa_ptr->ipsa_lock);
2628                         }
2629                 } else if (sadb_msg_type == SADB_X_DELPAIR) {
2630                         *diagnostic = SADB_X_DIAGNOSTIC_PAIR_SA_NOTFOUND;
2631                         error = ESRCH;
2632                 }
2633                 mutex_exit(&ipsapp.ipsap_bucket->isaf_lock);
2634                 mutex_exit(&ipsapp.ipsap_pbucket->isaf_lock);
2635         }
2636 
2637         ASSERT(mp->b_cont != NULL);
2638 
2639         if (error == 0)
2640                 sadb_pfkey_echo(pfkey_q, mp, (sadb_msg_t *)
2641                     mp->b_cont->b_rptr, ksi, echo_target);
2642 
2643         destroy_ipsa_pair(&ipsapp);
2644 
2645         return (error);
2646 }
2647 
2648 /*
2649  * This function takes a sadb_sa_t and finds the ipsa_t structure
2650  * and the isaf_t (hash bucket) that its stored under. If the security
2651  * association has a peer, the ipsa_t structure and bucket for that security
2652  * association are also searched for. The "pair" of ipsa_t's and isaf_t's
2653  * are returned as a ipsap_t.
2654  *
2655  * The hash buckets are returned for convenience, if the calling function
2656  * needs to use the hash bucket locks, say to remove the SA's, it should
2657  * take care to observe the convention of locking outbound bucket then
2658  * inbound bucket. The flag in_inbound_table provides direction.
2659  *
2660  * Note that a "pair" is defined as one (but not both) of the following:
2661  *
2662  * A security association which has a soft reference to another security
2663  * association via its SPI.
2664  *
2665  * A security association that is not obviously "inbound" or "outbound" so
2666  * it appears in both hash tables, the "peer" being the same security
2667  * association in the other hash table.
2668  *
2669  * This function will return NULL if the ipsa_t can't be found in the
2670  * inbound or outbound  hash tables (not found). If only one ipsa_t is
2671  * found, the pair ipsa_t will be NULL. Both isaf_t values are valid
2672  * provided at least one ipsa_t is found.
2673  */
2674 static int
2675 get_ipsa_pair(ipsa_query_t *sq, ipsap_t *ipsapp, int *diagnostic)
2676 {
2677         uint32_t pair_srcaddr[IPSA_MAX_ADDRLEN];
2678         uint32_t pair_dstaddr[IPSA_MAX_ADDRLEN];
2679         uint32_t pair_spi;
2680 
2681         init_ipsa_pair(ipsapp);
2682 
2683         ipsapp->in_inbound_table = B_FALSE;
2684 
2685         /* Lock down both buckets. */
2686         mutex_enter(&sq->outbound->isaf_lock);
2687         mutex_enter(&sq->inbound->isaf_lock);
2688 
2689         if (sq->assoc->sadb_sa_flags & IPSA_F_INBOUND) {
2690                 ipsapp->ipsap_sa_ptr = ipsec_getassocbyspi(sq->inbound,
2691                     sq->assoc->sadb_sa_spi, sq->srcaddr, sq->dstaddr, sq->af);
2692                 if (ipsapp->ipsap_sa_ptr != NULL) {
2693                         ipsapp->ipsap_bucket = sq->inbound;
2694                         ipsapp->ipsap_pbucket = sq->outbound;
2695                         ipsapp->in_inbound_table = B_TRUE;
2696                 } else {
2697                         ipsapp->ipsap_sa_ptr = ipsec_getassocbyspi(sq->outbound,
2698                             sq->assoc->sadb_sa_spi, sq->srcaddr, sq->dstaddr,
2699                             sq->af);
2700                         ipsapp->ipsap_bucket = sq->outbound;
2701                         ipsapp->ipsap_pbucket = sq->inbound;
2702                 }
2703         } else {
2704                 /* IPSA_F_OUTBOUND is set *or* no directions flags set. */
2705                 ipsapp->ipsap_sa_ptr =
2706                     ipsec_getassocbyspi(sq->outbound,
2707                     sq->assoc->sadb_sa_spi, sq->srcaddr, sq->dstaddr, sq->af);
2708                 if (ipsapp->ipsap_sa_ptr != NULL) {
2709                         ipsapp->ipsap_bucket = sq->outbound;
2710                         ipsapp->ipsap_pbucket = sq->inbound;
2711                 } else {
2712                         ipsapp->ipsap_sa_ptr = ipsec_getassocbyspi(sq->inbound,
2713                             sq->assoc->sadb_sa_spi, sq->srcaddr, sq->dstaddr,
2714                             sq->af);
2715                         ipsapp->ipsap_bucket = sq->inbound;
2716                         ipsapp->ipsap_pbucket = sq->outbound;
2717                         if (ipsapp->ipsap_sa_ptr != NULL)
2718                                 ipsapp->in_inbound_table = B_TRUE;
2719                 }
2720         }
2721 
2722         if (ipsapp->ipsap_sa_ptr == NULL) {
2723                 mutex_exit(&sq->outbound->isaf_lock);
2724                 mutex_exit(&sq->inbound->isaf_lock);
2725                 *diagnostic = SADB_X_DIAGNOSTIC_SA_NOTFOUND;
2726                 return (ESRCH);
2727         }
2728 
2729         if ((ipsapp->ipsap_sa_ptr->ipsa_state == IPSA_STATE_LARVAL) &&
2730             ipsapp->in_inbound_table) {
2731                 mutex_exit(&sq->outbound->isaf_lock);
2732                 mutex_exit(&sq->inbound->isaf_lock);
2733                 return (0);
2734         }
2735 
2736         mutex_enter(&ipsapp->ipsap_sa_ptr->ipsa_lock);
2737         if (ipsapp->ipsap_sa_ptr->ipsa_haspeer) {
2738                 /*
2739                  * haspeer implies no sa_pairing, look for same spi
2740                  * in other hashtable.
2741                  */
2742                 ipsapp->ipsap_psa_ptr =
2743                     ipsec_getassocbyspi(ipsapp->ipsap_pbucket,
2744                     sq->assoc->sadb_sa_spi, sq->srcaddr, sq->dstaddr, sq->af);
2745                 mutex_exit(&ipsapp->ipsap_sa_ptr->ipsa_lock);
2746                 mutex_exit(&sq->outbound->isaf_lock);
2747                 mutex_exit(&sq->inbound->isaf_lock);
2748                 return (0);
2749         }
2750         pair_spi = ipsapp->ipsap_sa_ptr->ipsa_otherspi;
2751         IPSA_COPY_ADDR(&pair_srcaddr,
2752             ipsapp->ipsap_sa_ptr->ipsa_srcaddr, sq->af);
2753         IPSA_COPY_ADDR(&pair_dstaddr,
2754             ipsapp->ipsap_sa_ptr->ipsa_dstaddr, sq->af);
2755         mutex_exit(&ipsapp->ipsap_sa_ptr->ipsa_lock);
2756         mutex_exit(&sq->inbound->isaf_lock);
2757         mutex_exit(&sq->outbound->isaf_lock);
2758 
2759         if (pair_spi == 0) {
2760                 ASSERT(ipsapp->ipsap_bucket != NULL);
2761                 ASSERT(ipsapp->ipsap_pbucket != NULL);
2762                 return (0);
2763         }
2764 
2765         /* found sa in outbound sadb, peer should be inbound */
2766 
2767         if (ipsapp->in_inbound_table) {
2768                 /* Found SA in inbound table, pair will be in outbound. */
2769                 if (sq->af == AF_INET6) {
2770                         ipsapp->ipsap_pbucket = OUTBOUND_BUCKET_V6(sq->sp,
2771                             *(uint32_t *)pair_srcaddr);
2772                 } else {
2773                         ipsapp->ipsap_pbucket = OUTBOUND_BUCKET_V4(sq->sp,
2774                             *(uint32_t *)pair_srcaddr);
2775                 }
2776         } else {
2777                 ipsapp->ipsap_pbucket = INBOUND_BUCKET(sq->sp, pair_spi);
2778         }
2779         mutex_enter(&ipsapp->ipsap_pbucket->isaf_lock);
2780         ipsapp->ipsap_psa_ptr = ipsec_getassocbyspi(ipsapp->ipsap_pbucket,
2781             pair_spi, pair_dstaddr, pair_srcaddr, sq->af);
2782         mutex_exit(&ipsapp->ipsap_pbucket->isaf_lock);
2783         ASSERT(ipsapp->ipsap_bucket != NULL);
2784         ASSERT(ipsapp->ipsap_pbucket != NULL);
2785         return (0);
2786 }
2787 
2788 /*
2789  * Perform NAT-traversal cached checksum offset calculations here.
2790  */
2791 static void
2792 sadb_nat_calculations(ipsa_t *newbie, sadb_address_t *natt_loc_ext,
2793     sadb_address_t *natt_rem_ext, uint32_t *src_addr_ptr,
2794     uint32_t *dst_addr_ptr)
2795 {
2796         struct sockaddr_in *natt_loc, *natt_rem;
2797         uint32_t *natt_loc_ptr = NULL, *natt_rem_ptr = NULL;
2798         uint32_t running_sum = 0;
2799 
2800 #define DOWN_SUM(x) (x) = ((x) & 0xFFFF) +   ((x) >> 16)
2801 
2802         if (natt_rem_ext != NULL) {
2803                 uint32_t l_src;
2804                 uint32_t l_rem;
2805 
2806                 natt_rem = (struct sockaddr_in *)(natt_rem_ext + 1);
2807 
2808                 /* Ensured by sadb_addrfix(). */
2809                 ASSERT(natt_rem->sin_family == AF_INET);
2810 
2811                 natt_rem_ptr = (uint32_t *)(&natt_rem->sin_addr);
2812                 newbie->ipsa_remote_nat_port = natt_rem->sin_port;
2813                 l_src = *src_addr_ptr;
2814                 l_rem = *natt_rem_ptr;
2815 
2816                 /* Instead of IPSA_COPY_ADDR(), just copy first 32 bits. */
2817                 newbie->ipsa_natt_addr_rem = *natt_rem_ptr;
2818 
2819                 l_src = ntohl(l_src);
2820                 DOWN_SUM(l_src);
2821                 DOWN_SUM(l_src);
2822                 l_rem = ntohl(l_rem);
2823                 DOWN_SUM(l_rem);
2824                 DOWN_SUM(l_rem);
2825 
2826                 /*
2827                  * We're 1's complement for checksums, so check for wraparound
2828                  * here.
2829                  */
2830                 if (l_rem > l_src)
2831                         l_src--;
2832 
2833                 running_sum += l_src - l_rem;
2834 
2835                 DOWN_SUM(running_sum);
2836                 DOWN_SUM(running_sum);
2837         }
2838 
2839         if (natt_loc_ext != NULL) {
2840                 natt_loc = (struct sockaddr_in *)(natt_loc_ext + 1);
2841 
2842                 /* Ensured by sadb_addrfix(). */
2843                 ASSERT(natt_loc->sin_family == AF_INET);
2844 
2845                 natt_loc_ptr = (uint32_t *)(&natt_loc->sin_addr);
2846                 newbie->ipsa_local_nat_port = natt_loc->sin_port;
2847 
2848                 /* Instead of IPSA_COPY_ADDR(), just copy first 32 bits. */
2849                 newbie->ipsa_natt_addr_loc = *natt_loc_ptr;
2850 
2851                 /*
2852                  * NAT-T port agility means we may have natt_loc_ext, but
2853                  * only for a local-port change.
2854                  */
2855                 if (natt_loc->sin_addr.s_addr != INADDR_ANY) {
2856                         uint32_t l_dst = ntohl(*dst_addr_ptr);
2857                         uint32_t l_loc = ntohl(*natt_loc_ptr);
2858 
2859                         DOWN_SUM(l_loc);
2860                         DOWN_SUM(l_loc);
2861                         DOWN_SUM(l_dst);
2862                         DOWN_SUM(l_dst);
2863 
2864                         /*
2865                          * We're 1's complement for checksums, so check for
2866                          * wraparound here.
2867                          */
2868                         if (l_loc > l_dst)
2869                                 l_dst--;
2870 
2871                         running_sum += l_dst - l_loc;
2872                         DOWN_SUM(running_sum);
2873                         DOWN_SUM(running_sum);
2874                 }
2875         }
2876 
2877         newbie->ipsa_inbound_cksum = running_sum;
2878 #undef DOWN_SUM
2879 }
2880 
2881 /*
2882  * This function is called from consumers that need to insert a fully-grown
2883  * security association into its tables.  This function takes into account that
2884  * SAs can be "inbound", "outbound", or "both".  The "primary" and "secondary"
2885  * hash bucket parameters are set in order of what the SA will be most of the
2886  * time.  (For example, an SA with an unspecified source, and a multicast
2887  * destination will primarily be an outbound SA.  OTOH, if that destination
2888  * is unicast for this node, then the SA will primarily be inbound.)
2889  *
2890  * It takes a lot of parameters because even if clone is B_FALSE, this needs
2891  * to check both buckets for purposes of collision.
2892  *
2893  * Return 0 upon success.  Return various errnos (ENOMEM, EEXIST) for
2894  * various error conditions.  We may need to set samsg->sadb_x_msg_diagnostic
2895  * with additional diagnostic information because there is at least one EINVAL
2896  * case here.
2897  */
2898 int
2899 sadb_common_add(queue_t *pfkey_q, mblk_t *mp, sadb_msg_t *samsg,
2900     keysock_in_t *ksi, isaf_t *primary, isaf_t *secondary,
2901     ipsa_t *newbie, boolean_t clone, boolean_t is_inbound, int *diagnostic,
2902     netstack_t *ns, sadbp_t *spp)
2903 {
2904         ipsa_t *newbie_clone = NULL, *scratch;
2905         ipsap_t ipsapp;
2906         sadb_sa_t *assoc = (sadb_sa_t *)ksi->ks_in_extv[SADB_EXT_SA];
2907         sadb_address_t *srcext =
2908             (sadb_address_t *)ksi->ks_in_extv[SADB_EXT_ADDRESS_SRC];
2909         sadb_address_t *dstext =
2910             (sadb_address_t *)ksi->ks_in_extv[SADB_EXT_ADDRESS_DST];
2911         sadb_address_t *isrcext =
2912             (sadb_address_t *)ksi->ks_in_extv[SADB_X_EXT_ADDRESS_INNER_SRC];
2913         sadb_address_t *idstext =
2914             (sadb_address_t *)ksi->ks_in_extv[SADB_X_EXT_ADDRESS_INNER_DST];
2915         sadb_x_kmc_t *kmcext =
2916             (sadb_x_kmc_t *)ksi->ks_in_extv[SADB_X_EXT_KM_COOKIE];
2917         sadb_key_t *akey = (sadb_key_t *)ksi->ks_in_extv[SADB_EXT_KEY_AUTH];
2918         sadb_key_t *ekey = (sadb_key_t *)ksi->ks_in_extv[SADB_EXT_KEY_ENCRYPT];
2919         sadb_sens_t *sens =
2920             (sadb_sens_t *)ksi->ks_in_extv[SADB_EXT_SENSITIVITY];
2921         sadb_sens_t *osens =
2922             (sadb_sens_t *)ksi->ks_in_extv[SADB_X_EXT_OUTER_SENS];
2923         sadb_x_pair_t *pair_ext =
2924             (sadb_x_pair_t *)ksi->ks_in_extv[SADB_X_EXT_PAIR];
2925         sadb_x_replay_ctr_t *replayext =
2926             (sadb_x_replay_ctr_t *)ksi->ks_in_extv[SADB_X_EXT_REPLAY_VALUE];
2927         uint8_t protocol =
2928             (samsg->sadb_msg_satype == SADB_SATYPE_AH) ? IPPROTO_AH:IPPROTO_ESP;
2929         int salt_offset;
2930         uint8_t *buf_ptr;
2931         struct sockaddr_in *src, *dst, *isrc, *idst;
2932         struct sockaddr_in6 *src6, *dst6, *isrc6, *idst6;
2933         sadb_lifetime_t *soft =
2934             (sadb_lifetime_t *)ksi->ks_in_extv[SADB_EXT_LIFETIME_SOFT];
2935         sadb_lifetime_t *hard =
2936             (sadb_lifetime_t *)ksi->ks_in_extv[SADB_EXT_LIFETIME_HARD];
2937         sadb_lifetime_t *idle =
2938             (sadb_lifetime_t *)ksi->ks_in_extv[SADB_X_EXT_LIFETIME_IDLE];
2939         sa_family_t af;
2940         int error = 0;
2941         boolean_t isupdate = (newbie != NULL);
2942         uint32_t *src_addr_ptr, *dst_addr_ptr, *isrc_addr_ptr, *idst_addr_ptr;
2943         ipsec_stack_t   *ipss = ns->netstack_ipsec;
2944         ip_stack_t      *ipst = ns->netstack_ip;
2945         ipsec_alginfo_t *alg;
2946         int             rcode;
2947         boolean_t       async = B_FALSE;
2948 
2949         init_ipsa_pair(&ipsapp);
2950 
2951         if (srcext == NULL) {
2952                 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_SRC;
2953                 return (EINVAL);
2954         }
2955         if (dstext == NULL) {
2956                 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_DST;
2957                 return (EINVAL);
2958         }
2959         if (assoc == NULL) {
2960                 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_SA;
2961                 return (EINVAL);
2962         }
2963 
2964         src = (struct sockaddr_in *)(srcext + 1);
2965         src6 = (struct sockaddr_in6 *)(srcext + 1);
2966         dst = (struct sockaddr_in *)(dstext + 1);
2967         dst6 = (struct sockaddr_in6 *)(dstext + 1);
2968         if (isrcext != NULL) {
2969                 isrc = (struct sockaddr_in *)(isrcext + 1);
2970                 isrc6 = (struct sockaddr_in6 *)(isrcext + 1);
2971                 ASSERT(idstext != NULL);
2972                 idst = (struct sockaddr_in *)(idstext + 1);
2973                 idst6 = (struct sockaddr_in6 *)(idstext + 1);
2974         } else {
2975                 isrc = NULL;
2976                 isrc6 = NULL;
2977         }
2978 
2979         af = src->sin_family;
2980 
2981         if (af == AF_INET) {
2982                 src_addr_ptr = (uint32_t *)&src->sin_addr;
2983                 dst_addr_ptr = (uint32_t *)&dst->sin_addr;
2984         } else {
2985                 ASSERT(af == AF_INET6);
2986                 src_addr_ptr = (uint32_t *)&src6->sin6_addr;
2987                 dst_addr_ptr = (uint32_t *)&dst6->sin6_addr;
2988         }
2989 
2990         if (!isupdate && (clone == B_TRUE || is_inbound == B_TRUE) &&
2991             cl_inet_checkspi &&
2992             (assoc->sadb_sa_state != SADB_X_SASTATE_ACTIVE_ELSEWHERE)) {
2993                 rcode = cl_inet_checkspi(ns->netstack_stackid, protocol,
2994                     assoc->sadb_sa_spi, NULL);
2995                 if (rcode == -1) {
2996                         return (EEXIST);
2997                 }
2998         }
2999 
3000         /*
3001          * Check to see if the new SA will be cloned AND paired. The
3002          * reason a SA will be cloned is the source or destination addresses
3003          * are not specific enough to determine if the SA goes in the outbound
3004          * or the inbound hash table, so its cloned and put in both. If
3005          * the SA is paired, it's soft linked to another SA for the other
3006          * direction. Keeping track and looking up SA's that are direction
3007          * unspecific and linked is too hard.
3008          */
3009         if (clone && (pair_ext != NULL)) {
3010                 *diagnostic = SADB_X_DIAGNOSTIC_PAIR_INAPPROPRIATE;
3011                 return (EINVAL);
3012         }
3013 
3014         if (!isupdate) {
3015                 newbie = sadb_makelarvalassoc(assoc->sadb_sa_spi,
3016                     src_addr_ptr, dst_addr_ptr, af, ns);
3017                 if (newbie == NULL)
3018                         return (ENOMEM);
3019         }
3020 
3021         mutex_enter(&newbie->ipsa_lock);
3022 
3023         if (isrc != NULL) {
3024                 if (isrc->sin_family == AF_INET) {
3025                         if (srcext->sadb_address_proto != IPPROTO_ENCAP) {
3026                                 if (srcext->sadb_address_proto != 0) {
3027                                         /*
3028                                          * Mismatched outer-packet protocol
3029                                          * and inner-packet address family.
3030                                          */
3031                                         mutex_exit(&newbie->ipsa_lock);
3032                                         error = EPROTOTYPE;
3033                                         *diagnostic =
3034                                             SADB_X_DIAGNOSTIC_INNER_AF_MISMATCH;
3035                                         goto error;
3036                                 } else {
3037                                         /* Fill in with explicit protocol. */
3038                                         srcext->sadb_address_proto =
3039                                             IPPROTO_ENCAP;
3040                                         dstext->sadb_address_proto =
3041                                             IPPROTO_ENCAP;
3042                                 }
3043                         }
3044                         isrc_addr_ptr = (uint32_t *)&isrc->sin_addr;
3045                         idst_addr_ptr = (uint32_t *)&idst->sin_addr;
3046                 } else {
3047                         ASSERT(isrc->sin_family == AF_INET6);
3048                         if (srcext->sadb_address_proto != IPPROTO_IPV6) {
3049                                 if (srcext->sadb_address_proto != 0) {
3050                                         /*
3051                                          * Mismatched outer-packet protocol
3052                                          * and inner-packet address family.
3053                                          */
3054                                         mutex_exit(&newbie->ipsa_lock);
3055                                         error = EPROTOTYPE;
3056                                         *diagnostic =
3057                                             SADB_X_DIAGNOSTIC_INNER_AF_MISMATCH;
3058                                         goto error;
3059                                 } else {
3060                                         /* Fill in with explicit protocol. */
3061                                         srcext->sadb_address_proto =
3062                                             IPPROTO_IPV6;
3063                                         dstext->sadb_address_proto =
3064                                             IPPROTO_IPV6;
3065                                 }
3066                         }
3067                         isrc_addr_ptr = (uint32_t *)&isrc6->sin6_addr;
3068                         idst_addr_ptr = (uint32_t *)&idst6->sin6_addr;
3069                 }
3070                 newbie->ipsa_innerfam = isrc->sin_family;
3071 
3072                 IPSA_COPY_ADDR(newbie->ipsa_innersrc, isrc_addr_ptr,
3073                     newbie->ipsa_innerfam);
3074                 IPSA_COPY_ADDR(newbie->ipsa_innerdst, idst_addr_ptr,
3075                     newbie->ipsa_innerfam);
3076                 newbie->ipsa_innersrcpfx = isrcext->sadb_address_prefixlen;
3077                 newbie->ipsa_innerdstpfx = idstext->sadb_address_prefixlen;
3078 
3079                 /* Unique value uses inner-ports for Tunnel Mode... */
3080                 newbie->ipsa_unique_id = SA_UNIQUE_ID(isrc->sin_port,
3081                     idst->sin_port, dstext->sadb_address_proto,
3082                     idstext->sadb_address_proto);
3083                 newbie->ipsa_unique_mask = SA_UNIQUE_MASK(isrc->sin_port,
3084                     idst->sin_port, dstext->sadb_address_proto,
3085                     idstext->sadb_address_proto);
3086         } else {
3087                 /* ... and outer-ports for Transport Mode. */
3088                 newbie->ipsa_unique_id = SA_UNIQUE_ID(src->sin_port,
3089                     dst->sin_port, dstext->sadb_address_proto, 0);
3090                 newbie->ipsa_unique_mask = SA_UNIQUE_MASK(src->sin_port,
3091                     dst->sin_port, dstext->sadb_address_proto, 0);
3092         }
3093         if (newbie->ipsa_unique_mask != (uint64_t)0)
3094                 newbie->ipsa_flags |= IPSA_F_UNIQUE;
3095 
3096         sadb_nat_calculations(newbie,
3097             (sadb_address_t *)ksi->ks_in_extv[SADB_X_EXT_ADDRESS_NATT_LOC],
3098             (sadb_address_t *)ksi->ks_in_extv[SADB_X_EXT_ADDRESS_NATT_REM],
3099             src_addr_ptr, dst_addr_ptr);
3100 
3101         newbie->ipsa_type = samsg->sadb_msg_satype;
3102 
3103         ASSERT((assoc->sadb_sa_state == SADB_SASTATE_MATURE) ||
3104             (assoc->sadb_sa_state == SADB_X_SASTATE_ACTIVE_ELSEWHERE));
3105         newbie->ipsa_auth_alg = assoc->sadb_sa_auth;
3106         newbie->ipsa_encr_alg = assoc->sadb_sa_encrypt;
3107 
3108         newbie->ipsa_flags |= assoc->sadb_sa_flags;
3109         if (newbie->ipsa_flags & SADB_X_SAFLAGS_NATT_LOC &&
3110             ksi->ks_in_extv[SADB_X_EXT_ADDRESS_NATT_LOC] == NULL) {
3111                 mutex_exit(&newbie->ipsa_lock);
3112                 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_NATT_LOC;
3113                 error = EINVAL;
3114                 goto error;
3115         }
3116         if (newbie->ipsa_flags & SADB_X_SAFLAGS_NATT_REM &&
3117             ksi->ks_in_extv[SADB_X_EXT_ADDRESS_NATT_REM] == NULL) {
3118                 mutex_exit(&newbie->ipsa_lock);
3119                 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_NATT_REM;
3120                 error = EINVAL;
3121                 goto error;
3122         }
3123         if (newbie->ipsa_flags & SADB_X_SAFLAGS_TUNNEL &&
3124             ksi->ks_in_extv[SADB_X_EXT_ADDRESS_INNER_SRC] == NULL) {
3125                 mutex_exit(&newbie->ipsa_lock);
3126                 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_INNER_SRC;
3127                 error = EINVAL;
3128                 goto error;
3129         }
3130         /*
3131          * If unspecified source address, force replay_wsize to 0.
3132          * This is because an SA that has multiple sources of secure
3133          * traffic cannot enforce a replay counter w/o synchronizing the
3134          * senders.
3135          */
3136         if (ksi->ks_in_srctype != KS_IN_ADDR_UNSPEC)
3137                 newbie->ipsa_replay_wsize = assoc->sadb_sa_replay;
3138         else
3139                 newbie->ipsa_replay_wsize = 0;
3140 
3141         newbie->ipsa_addtime = gethrestime_sec();
3142 
3143         if (kmcext != NULL) {
3144                 newbie->ipsa_kmp = kmcext->sadb_x_kmc_proto;
3145                 /*
3146                  * Be liberal in what we receive.  Special-case the IKEv1
3147                  * cookie, which closed-source in.iked assumes is 32 bits.
3148                  * Now that we store all 64 bits, we should pre-zero the
3149                  * reserved field on behalf of closed-source in.iked.
3150                  */
3151                 if (newbie->ipsa_kmp == SADB_X_KMP_IKE) {
3152                         /* Just in case in.iked is misbehaving... */
3153                         kmcext->sadb_x_kmc_reserved = 0;
3154                 }
3155                 newbie->ipsa_kmc = kmcext->sadb_x_kmc_cookie64;
3156         }
3157 
3158         /*
3159          * XXX CURRENT lifetime checks MAY BE needed for an UPDATE.
3160          * The spec says that one can update current lifetimes, but
3161          * that seems impractical, especially in the larval-to-mature
3162          * update that this function performs.
3163          */
3164         if (soft != NULL) {
3165                 newbie->ipsa_softaddlt = soft->sadb_lifetime_addtime;
3166                 newbie->ipsa_softuselt = soft->sadb_lifetime_usetime;
3167                 newbie->ipsa_softbyteslt = soft->sadb_lifetime_bytes;
3168                 newbie->ipsa_softalloc = soft->sadb_lifetime_allocations;
3169                 SET_EXPIRE(newbie, softaddlt, softexpiretime);
3170         }
3171         if (hard != NULL) {
3172                 newbie->ipsa_hardaddlt = hard->sadb_lifetime_addtime;
3173                 newbie->ipsa_harduselt = hard->sadb_lifetime_usetime;
3174                 newbie->ipsa_hardbyteslt = hard->sadb_lifetime_bytes;
3175                 newbie->ipsa_hardalloc = hard->sadb_lifetime_allocations;
3176                 SET_EXPIRE(newbie, hardaddlt, hardexpiretime);
3177         }
3178         if (idle != NULL) {
3179                 newbie->ipsa_idleaddlt = idle->sadb_lifetime_addtime;
3180                 newbie->ipsa_idleuselt = idle->sadb_lifetime_usetime;
3181                 newbie->ipsa_idleexpiretime = newbie->ipsa_addtime +
3182                     newbie->ipsa_idleaddlt;
3183                 newbie->ipsa_idletime = newbie->ipsa_idleaddlt;
3184         }
3185 
3186         newbie->ipsa_authtmpl = NULL;
3187         newbie->ipsa_encrtmpl = NULL;
3188 
3189 #ifdef IPSEC_LATENCY_TEST
3190         if (akey != NULL && newbie->ipsa_auth_alg != SADB_AALG_NONE) {
3191 #else
3192         if (akey != NULL) {
3193 #endif
3194                 async = (ipss->ipsec_algs_exec_mode[IPSEC_ALG_AUTH] ==
3195                     IPSEC_ALGS_EXEC_ASYNC);
3196 
3197                 newbie->ipsa_authkeybits = akey->sadb_key_bits;
3198                 newbie->ipsa_authkeylen = SADB_1TO8(akey->sadb_key_bits);
3199                 /* In case we have to round up to the next byte... */
3200                 if ((akey->sadb_key_bits & 0x7) != 0)
3201                         newbie->ipsa_authkeylen++;
3202                 newbie->ipsa_authkey = kmem_alloc(newbie->ipsa_authkeylen,
3203                     KM_NOSLEEP);
3204                 if (newbie->ipsa_authkey == NULL) {
3205                         error = ENOMEM;
3206                         mutex_exit(&newbie->ipsa_lock);
3207                         goto error;
3208                 }
3209                 bcopy(akey + 1, newbie->ipsa_authkey, newbie->ipsa_authkeylen);
3210                 bzero(akey + 1, newbie->ipsa_authkeylen);
3211 
3212                 /*
3213                  * Pre-initialize the kernel crypto framework key
3214                  * structure.
3215                  */
3216                 newbie->ipsa_kcfauthkey.ck_format = CRYPTO_KEY_RAW;
3217                 newbie->ipsa_kcfauthkey.ck_length = newbie->ipsa_authkeybits;
3218                 newbie->ipsa_kcfauthkey.ck_data = newbie->ipsa_authkey;
3219 
3220                 rw_enter(&ipss->ipsec_alg_lock, RW_READER);
3221                 alg = ipss->ipsec_alglists[IPSEC_ALG_AUTH]
3222                     [newbie->ipsa_auth_alg];
3223                 if (alg != NULL && ALG_VALID(alg)) {
3224                         newbie->ipsa_amech.cm_type = alg->alg_mech_type;
3225                         newbie->ipsa_amech.cm_param =
3226                             (char *)&newbie->ipsa_mac_len;
3227                         newbie->ipsa_amech.cm_param_len = sizeof (size_t);
3228                         newbie->ipsa_mac_len = (size_t)alg->alg_datalen;
3229                 } else {
3230                         newbie->ipsa_amech.cm_type = CRYPTO_MECHANISM_INVALID;
3231                 }
3232                 error = ipsec_create_ctx_tmpl(newbie, IPSEC_ALG_AUTH);
3233                 rw_exit(&ipss->ipsec_alg_lock);
3234                 if (error != 0) {
3235                         mutex_exit(&newbie->ipsa_lock);
3236                         /*
3237                          * An error here indicates that alg is the wrong type
3238                          * (IE: not authentication) or its not in the alg tables
3239                          * created by ipsecalgs(1m), or Kcf does not like the
3240                          * parameters passed in with this algorithm, which is
3241                          * probably a coding error!
3242                          */
3243                         *diagnostic = SADB_X_DIAGNOSTIC_BAD_CTX;
3244 
3245                         goto error;
3246                 }
3247         }
3248 
3249         if (ekey != NULL) {
3250                 rw_enter(&ipss->ipsec_alg_lock, RW_READER);
3251                 async = async || (ipss->ipsec_algs_exec_mode[IPSEC_ALG_ENCR] ==
3252                     IPSEC_ALGS_EXEC_ASYNC);
3253                 alg = ipss->ipsec_alglists[IPSEC_ALG_ENCR]
3254                     [newbie->ipsa_encr_alg];
3255 
3256                 if (alg != NULL && ALG_VALID(alg)) {
3257                         newbie->ipsa_emech.cm_type = alg->alg_mech_type;
3258                         newbie->ipsa_datalen = alg->alg_datalen;
3259                         if (alg->alg_flags & ALG_FLAG_COUNTERMODE)
3260                                 newbie->ipsa_flags |= IPSA_F_COUNTERMODE;
3261 
3262                         if (alg->alg_flags & ALG_FLAG_COMBINED) {
3263                                 newbie->ipsa_flags |= IPSA_F_COMBINED;
3264                                 newbie->ipsa_mac_len =  alg->alg_icvlen;
3265                         }
3266 
3267                         if (alg->alg_flags & ALG_FLAG_CCM)
3268                                 newbie->ipsa_noncefunc = ccm_params_init;
3269                         else if (alg->alg_flags & ALG_FLAG_GCM)
3270                                 newbie->ipsa_noncefunc = gcm_params_init;
3271                         else newbie->ipsa_noncefunc = cbc_params_init;
3272 
3273                         newbie->ipsa_saltlen = alg->alg_saltlen;
3274                         newbie->ipsa_saltbits = SADB_8TO1(newbie->ipsa_saltlen);
3275                         newbie->ipsa_iv_len = alg->alg_ivlen;
3276                         newbie->ipsa_nonce_len = newbie->ipsa_saltlen +
3277                             newbie->ipsa_iv_len;
3278                         newbie->ipsa_emech.cm_param = NULL;
3279                         newbie->ipsa_emech.cm_param_len = 0;
3280                 } else {
3281                         newbie->ipsa_emech.cm_type = CRYPTO_MECHANISM_INVALID;
3282                 }
3283                 rw_exit(&ipss->ipsec_alg_lock);
3284 
3285                 /*
3286                  * The byte stream following the sadb_key_t is made up of:
3287                  * key bytes, [salt bytes], [IV initial value]
3288                  * All of these have variable length. The IV is typically
3289                  * randomly generated by this function and not passed in.
3290                  * By supporting the injection of a known IV, the whole
3291                  * IPsec subsystem and the underlying crypto subsystem
3292                  * can be tested with known test vectors.
3293                  *
3294                  * The keying material has been checked by ext_check()
3295                  * and ipsec_valid_key_size(), after removing salt/IV
3296                  * bits, whats left is the encryption key. If this is too
3297                  * short, ipsec_create_ctx_tmpl() will fail and the SA
3298                  * won't get created.
3299                  *
3300                  * set ipsa_encrkeylen to length of key only.
3301                  */
3302                 newbie->ipsa_encrkeybits = ekey->sadb_key_bits;
3303                 newbie->ipsa_encrkeybits -= ekey->sadb_key_reserved;
3304                 newbie->ipsa_encrkeybits -= newbie->ipsa_saltbits;
3305                 newbie->ipsa_encrkeylen = SADB_1TO8(newbie->ipsa_encrkeybits);
3306 
3307                 /* In case we have to round up to the next byte... */
3308                 if ((ekey->sadb_key_bits & 0x7) != 0)
3309                         newbie->ipsa_encrkeylen++;
3310 
3311                 newbie->ipsa_encrkey = kmem_alloc(newbie->ipsa_encrkeylen,
3312                     KM_NOSLEEP);
3313                 if (newbie->ipsa_encrkey == NULL) {
3314                         error = ENOMEM;
3315                         mutex_exit(&newbie->ipsa_lock);
3316                         goto error;
3317                 }
3318 
3319                 buf_ptr = (uint8_t *)(ekey + 1);
3320                 bcopy(buf_ptr, newbie->ipsa_encrkey, newbie->ipsa_encrkeylen);
3321 
3322                 if (newbie->ipsa_flags & IPSA_F_COMBINED) {
3323                         /*
3324                          * Combined mode algs need a nonce. Copy the salt and
3325                          * IV into a buffer. The ipsa_nonce is a pointer into
3326                          * this buffer, some bytes at the start of the buffer
3327                          * may be unused, depends on the salt length. The IV
3328                          * is 64 bit aligned so it can be incremented as a
3329                          * uint64_t. Zero out key in samsg_t before freeing.
3330                          */
3331 
3332                         newbie->ipsa_nonce_buf = kmem_alloc(
3333                             sizeof (ipsec_nonce_t), KM_NOSLEEP);
3334                         if (newbie->ipsa_nonce_buf == NULL) {
3335                                 error = ENOMEM;
3336                                 mutex_exit(&newbie->ipsa_lock);
3337                                 goto error;
3338                         }
3339                         /*
3340                          * Initialize nonce and salt pointers to point
3341                          * to the nonce buffer. This is just in case we get
3342                          * bad data, the pointers will be valid, the data
3343                          * won't be.
3344                          *
3345                          * See sadb.h for layout of nonce.
3346                          */
3347                         newbie->ipsa_iv = &newbie->ipsa_nonce_buf->iv;
3348                         newbie->ipsa_salt = (uint8_t *)newbie->ipsa_nonce_buf;
3349                         newbie->ipsa_nonce = newbie->ipsa_salt;
3350                         if (newbie->ipsa_saltlen != 0) {
3351                                 salt_offset = MAXSALTSIZE -
3352                                     newbie->ipsa_saltlen;
3353                                 newbie->ipsa_salt = (uint8_t *)
3354                                     &newbie->ipsa_nonce_buf->salt[salt_offset];
3355                                 newbie->ipsa_nonce = newbie->ipsa_salt;
3356                                 buf_ptr += newbie->ipsa_encrkeylen;
3357                                 bcopy(buf_ptr, newbie->ipsa_salt,
3358                                     newbie->ipsa_saltlen);
3359                         }
3360                         /*
3361                          * The IV for CCM/GCM mode increments, it should not
3362                          * repeat. Get a random value for the IV, make a
3363                          * copy, the SA will expire when/if the IV ever
3364                          * wraps back to the initial value. If an Initial IV
3365                          * is passed in via PF_KEY, save this in the SA.
3366                          * Initialising IV for inbound is pointless as its
3367                          * taken from the inbound packet.
3368                          */
3369                         if (!is_inbound) {
3370                                 if (ekey->sadb_key_reserved != 0) {
3371                                         buf_ptr += newbie->ipsa_saltlen;
3372                                         bcopy(buf_ptr, (uint8_t *)newbie->
3373                                             ipsa_iv, SADB_1TO8(ekey->
3374                                             sadb_key_reserved));
3375                                 } else {
3376                                         (void) random_get_pseudo_bytes(
3377                                             (uint8_t *)newbie->ipsa_iv,
3378                                             newbie->ipsa_iv_len);
3379                                 }
3380                                 newbie->ipsa_iv_softexpire =
3381                                     (*newbie->ipsa_iv) << 9;
3382                                 newbie->ipsa_iv_hardexpire = *newbie->ipsa_iv;
3383                         }
3384                 }
3385                 bzero((ekey + 1), SADB_1TO8(ekey->sadb_key_bits));
3386 
3387                 /*
3388                  * Pre-initialize the kernel crypto framework key
3389                  * structure.
3390                  */
3391                 newbie->ipsa_kcfencrkey.ck_format = CRYPTO_KEY_RAW;
3392                 newbie->ipsa_kcfencrkey.ck_length = newbie->ipsa_encrkeybits;
3393                 newbie->ipsa_kcfencrkey.ck_data = newbie->ipsa_encrkey;
3394 
3395                 rw_enter(&ipss->ipsec_alg_lock, RW_READER);
3396                 error = ipsec_create_ctx_tmpl(newbie, IPSEC_ALG_ENCR);
3397                 rw_exit(&ipss->ipsec_alg_lock);
3398                 if (error != 0) {
3399                         mutex_exit(&newbie->ipsa_lock);
3400                         /* See above for error explanation. */
3401                         *diagnostic = SADB_X_DIAGNOSTIC_BAD_CTX;
3402                         goto error;
3403                 }
3404         }
3405 
3406         if (async)
3407                 newbie->ipsa_flags |= IPSA_F_ASYNC;
3408 
3409         /*
3410          * Ptrs to processing functions.
3411          */
3412         if (newbie->ipsa_type == SADB_SATYPE_ESP)
3413                 ipsecesp_init_funcs(newbie);
3414         else
3415                 ipsecah_init_funcs(newbie);
3416         ASSERT(newbie->ipsa_output_func != NULL &&
3417             newbie->ipsa_input_func != NULL);
3418 
3419         /*
3420          * Certificate ID stuff.
3421          */
3422         if (ksi->ks_in_extv[SADB_EXT_IDENTITY_SRC] != NULL) {
3423                 sadb_ident_t *id =
3424                     (sadb_ident_t *)ksi->ks_in_extv[SADB_EXT_IDENTITY_SRC];
3425 
3426                 /*
3427                  * Can assume strlen() will return okay because ext_check() in
3428                  * keysock.c prepares the string for us.
3429                  */
3430                 newbie->ipsa_src_cid = ipsid_lookup(id->sadb_ident_type,
3431                     (char *)(id+1), ns);
3432                 if (newbie->ipsa_src_cid == NULL) {
3433                         error = ENOMEM;
3434                         mutex_exit(&newbie->ipsa_lock);
3435                         goto error;
3436                 }
3437         }
3438 
3439         if (ksi->ks_in_extv[SADB_EXT_IDENTITY_DST] != NULL) {
3440                 sadb_ident_t *id =
3441                     (sadb_ident_t *)ksi->ks_in_extv[SADB_EXT_IDENTITY_DST];
3442 
3443                 /*
3444                  * Can assume strlen() will return okay because ext_check() in
3445                  * keysock.c prepares the string for us.
3446                  */
3447                 newbie->ipsa_dst_cid = ipsid_lookup(id->sadb_ident_type,
3448                     (char *)(id+1), ns);
3449                 if (newbie->ipsa_dst_cid == NULL) {
3450                         error = ENOMEM;
3451                         mutex_exit(&newbie->ipsa_lock);
3452                         goto error;
3453                 }
3454         }
3455 
3456         /*
3457          * sensitivity label handling code:
3458          * Convert sens + bitmap into cred_t, and associate it
3459          * with the new SA.
3460          */
3461         if (sens != NULL) {
3462                 uint64_t *bitmap = (uint64_t *)(sens + 1);
3463 
3464                 newbie->ipsa_tsl = sadb_label_from_sens(sens, bitmap);
3465         }
3466 
3467         /*
3468          * Likewise for outer sensitivity.
3469          */
3470         if (osens != NULL) {
3471                 uint64_t *bitmap = (uint64_t *)(osens + 1);
3472                 ts_label_t *tsl, *effective_tsl;
3473                 uint32_t *peer_addr_ptr;
3474                 zoneid_t zoneid = GLOBAL_ZONEID;
3475                 zone_t *zone;
3476 
3477                 peer_addr_ptr = is_inbound ? src_addr_ptr : dst_addr_ptr;
3478 
3479                 tsl = sadb_label_from_sens(osens, bitmap);
3480                 newbie->ipsa_mac_exempt = CONN_MAC_DEFAULT;
3481 
3482                 if (osens->sadb_x_sens_flags & SADB_X_SENS_IMPLICIT) {
3483                         newbie->ipsa_mac_exempt = CONN_MAC_IMPLICIT;
3484                 }
3485 
3486                 error = tsol_check_dest(tsl, peer_addr_ptr,
3487                     (af == AF_INET6)?IPV6_VERSION:IPV4_VERSION,
3488                     newbie->ipsa_mac_exempt, B_TRUE, &effective_tsl);
3489                 if (error != 0) {
3490                         label_rele(tsl);
3491                         mutex_exit(&newbie->ipsa_lock);
3492                         goto error;
3493                 }
3494 
3495                 if (effective_tsl != NULL) {
3496                         label_rele(tsl);
3497                         tsl = effective_tsl;
3498                 }
3499 
3500                 newbie->ipsa_otsl = tsl;
3501 
3502                 zone = zone_find_by_label(tsl);
3503                 if (zone != NULL) {
3504                         zoneid = zone->zone_id;
3505                         zone_rele(zone);
3506                 }
3507                 /*
3508                  * For exclusive stacks we set the zoneid to zero to operate
3509                  * as if in the global zone for tsol_compute_label_v4/v6
3510                  */
3511                 if (ipst->ips_netstack->netstack_stackid != GLOBAL_NETSTACKID)
3512                         zoneid = GLOBAL_ZONEID;
3513 
3514                 if (af == AF_INET6) {
3515                         error = tsol_compute_label_v6(tsl, zoneid,
3516                             (in6_addr_t *)peer_addr_ptr,
3517                             newbie->ipsa_opt_storage, ipst);
3518                 } else {
3519                         error = tsol_compute_label_v4(tsl, zoneid,
3520                             *peer_addr_ptr, newbie->ipsa_opt_storage, ipst);
3521                 }
3522                 if (error != 0) {
3523                         mutex_exit(&newbie->ipsa_lock);
3524                         goto error;
3525                 }
3526         }
3527 
3528 
3529         if (replayext != NULL) {
3530                 if ((replayext->sadb_x_rc_replay32 == 0) &&
3531                     (replayext->sadb_x_rc_replay64 != 0)) {
3532                         error = EOPNOTSUPP;
3533                         *diagnostic = SADB_X_DIAGNOSTIC_INVALID_REPLAY;
3534                         mutex_exit(&newbie->ipsa_lock);
3535                         goto error;
3536                 }
3537                 newbie->ipsa_replay = replayext->sadb_x_rc_replay32;
3538         }
3539 
3540         /* now that the SA has been updated, set its new state */
3541         newbie->ipsa_state = assoc->sadb_sa_state;
3542 
3543         if (clone) {
3544                 newbie->ipsa_haspeer = B_TRUE;
3545         } else {
3546                 if (!is_inbound) {
3547                         lifetime_fuzz(newbie);
3548                 }
3549         }
3550         /*
3551          * The less locks I hold when doing an insertion and possible cloning,
3552          * the better!
3553          */
3554         mutex_exit(&newbie->ipsa_lock);
3555 
3556         if (clone) {
3557                 newbie_clone = sadb_cloneassoc(newbie);
3558 
3559                 if (newbie_clone == NULL) {
3560                         error = ENOMEM;
3561                         goto error;
3562                 }
3563         }
3564 
3565         /*
3566          * Enter the bucket locks.  The order of entry is outbound,
3567          * inbound.  We map "primary" and "secondary" into outbound and inbound
3568          * based on the destination address type.  If the destination address
3569          * type is for a node that isn't mine (or potentially mine), the
3570          * "primary" bucket is the outbound one.
3571          */
3572         if (!is_inbound) {
3573                 /* primary == outbound */
3574                 mutex_enter(&primary->isaf_lock);
3575                 mutex_enter(&secondary->isaf_lock);
3576         } else {
3577                 /* primary == inbound */
3578                 mutex_enter(&secondary->isaf_lock);
3579                 mutex_enter(&primary->isaf_lock);
3580         }
3581 
3582         /*
3583          * sadb_insertassoc() doesn't increment the reference
3584          * count.  We therefore have to increment the
3585          * reference count one more time to reflect the
3586          * pointers of the table that reference this SA.
3587          */
3588         IPSA_REFHOLD(newbie);
3589 
3590         if (isupdate) {
3591                 /*
3592                  * Unlink from larval holding cell in the "inbound" fanout.
3593                  */
3594                 ASSERT(newbie->ipsa_linklock == &primary->isaf_lock ||
3595                     newbie->ipsa_linklock == &secondary->isaf_lock);
3596                 sadb_unlinkassoc(newbie);
3597         }
3598 
3599         mutex_enter(&newbie->ipsa_lock);
3600         error = sadb_insertassoc(newbie, primary);
3601         mutex_exit(&newbie->ipsa_lock);
3602 
3603         if (error != 0) {
3604                 /*
3605                  * Since sadb_insertassoc() failed, we must decrement the
3606                  * refcount again so the cleanup code will actually free
3607                  * the offending SA.
3608                  */
3609                 IPSA_REFRELE(newbie);
3610                 goto error_unlock;
3611         }
3612 
3613         if (newbie_clone != NULL) {
3614                 mutex_enter(&newbie_clone->ipsa_lock);
3615                 error = sadb_insertassoc(newbie_clone, secondary);
3616                 mutex_exit(&newbie_clone->ipsa_lock);
3617                 if (error != 0) {
3618                         /* Collision in secondary table. */
3619                         sadb_unlinkassoc(newbie);  /* This does REFRELE. */
3620                         goto error_unlock;
3621                 }
3622                 IPSA_REFHOLD(newbie_clone);
3623         } else {
3624                 ASSERT(primary != secondary);
3625                 scratch = ipsec_getassocbyspi(secondary, newbie->ipsa_spi,
3626                     ALL_ZEROES_PTR, newbie->ipsa_dstaddr, af);
3627                 if (scratch != NULL) {
3628                         /* Collision in secondary table. */
3629                         sadb_unlinkassoc(newbie);  /* This does REFRELE. */
3630                         /* Set the error, since ipsec_getassocbyspi() can't. */
3631                         error = EEXIST;
3632                         goto error_unlock;
3633                 }
3634         }
3635 
3636         /* OKAY!  So let's do some reality check assertions. */
3637 
3638         ASSERT(MUTEX_NOT_HELD(&newbie->ipsa_lock));
3639         ASSERT(newbie_clone == NULL ||
3640             (MUTEX_NOT_HELD(&newbie_clone->ipsa_lock)));
3641 
3642 error_unlock:
3643 
3644         /*
3645          * We can exit the locks in any order.  Only entrance needs to
3646          * follow any protocol.
3647          */
3648         mutex_exit(&secondary->isaf_lock);
3649         mutex_exit(&primary->isaf_lock);
3650 
3651         if (pair_ext != NULL && error == 0) {
3652                 /* update pair_spi if it exists. */
3653                 ipsa_query_t sq;
3654 
3655                 sq.spp = spp;           /* XXX param */
3656                 error = sadb_form_query(ksi, IPSA_Q_DST, IPSA_Q_SRC|IPSA_Q_DST|
3657                     IPSA_Q_SA|IPSA_Q_INBOUND|IPSA_Q_OUTBOUND, &sq, diagnostic);
3658                 if (error)
3659                         return (error);
3660 
3661                 error = get_ipsa_pair(&sq, &ipsapp, diagnostic);
3662 
3663                 if (error != 0)
3664                         goto error;
3665 
3666                 if (ipsapp.ipsap_psa_ptr != NULL) {
3667                         *diagnostic = SADB_X_DIAGNOSTIC_PAIR_ALREADY;
3668                         error = EINVAL;
3669                 } else {
3670                         /* update_pairing() sets diagnostic */
3671                         error = update_pairing(&ipsapp, &sq, ksi, diagnostic);
3672                 }
3673         }
3674         /* Common error point for this routine. */
3675 error:
3676         if (newbie != NULL) {
3677                 if (error != 0) {
3678                         /* This SA is broken, let the reaper clean up. */
3679                         mutex_enter(&newbie->ipsa_lock);
3680                         newbie->ipsa_state = IPSA_STATE_DEAD;
3681                         newbie->ipsa_hardexpiretime = 1;
3682                         mutex_exit(&newbie->ipsa_lock);
3683                 }
3684                 IPSA_REFRELE(newbie);
3685         }
3686         if (newbie_clone != NULL) {
3687                 IPSA_REFRELE(newbie_clone);
3688         }
3689 
3690         if (error == 0) {
3691                 /*
3692                  * Construct favorable PF_KEY return message and send to
3693                  * keysock. Update the flags in the original keysock message
3694                  * to reflect the actual flags in the new SA.
3695                  *  (Q:  Do I need to pass "newbie"?  If I do,
3696                  * make sure to REFHOLD, call, then REFRELE.)
3697                  */
3698                 assoc->sadb_sa_flags = newbie->ipsa_flags;
3699                 sadb_pfkey_echo(pfkey_q, mp, samsg, ksi, NULL);
3700         }
3701 
3702         destroy_ipsa_pair(&ipsapp);
3703         return (error);
3704 }
3705 
3706 /*
3707  * Set the time of first use for a security association.  Update any
3708  * expiration times as a result.
3709  */
3710 void
3711 sadb_set_usetime(ipsa_t *assoc)
3712 {
3713         time_t snapshot = gethrestime_sec();
3714 
3715         mutex_enter(&assoc->ipsa_lock);
3716         assoc->ipsa_lastuse = snapshot;
3717         assoc->ipsa_idleexpiretime = snapshot + assoc->ipsa_idletime;
3718 
3719         /*
3720          * Caller does check usetime before calling me usually, and
3721          * double-checking is better than a mutex_enter/exit hit.
3722          */
3723         if (assoc->ipsa_usetime == 0) {
3724                 /*
3725                  * This is redundant for outbound SA's, as
3726                  * ipsec_getassocbyconn() sets the IPSA_F_USED flag already.
3727                  * Inbound SAs, however, have no such protection.
3728                  */
3729                 assoc->ipsa_flags |= IPSA_F_USED;
3730                 assoc->ipsa_usetime = snapshot;
3731 
3732                 /*
3733                  * After setting the use time, see if we have a use lifetime
3734                  * that would cause the actual SA expiration time to shorten.
3735                  */
3736                 UPDATE_EXPIRE(assoc, softuselt, softexpiretime);
3737                 UPDATE_EXPIRE(assoc, harduselt, hardexpiretime);
3738         }
3739         mutex_exit(&assoc->ipsa_lock);
3740 }
3741 
3742 /*
3743  * Send up a PF_KEY expire message for this association.
3744  */
3745 static void
3746 sadb_expire_assoc(queue_t *pfkey_q, ipsa_t *assoc)
3747 {
3748         mblk_t *mp, *mp1;
3749         int alloclen, af;
3750         sadb_msg_t *samsg;
3751         sadb_lifetime_t *current, *expire;
3752         sadb_sa_t *saext;
3753         uint8_t *end;
3754         boolean_t tunnel_mode;
3755 
3756         ASSERT(MUTEX_HELD(&assoc->ipsa_lock));
3757 
3758         /* Don't bother sending if there's no queue. */
3759         if (pfkey_q == NULL)
3760                 return;
3761 
3762         mp = sadb_keysock_out(0);
3763         if (mp == NULL) {
3764                 /* cmn_err(CE_WARN, */
3765                 /*      "sadb_expire_assoc: Can't allocate KEYSOCK_OUT.\n"); */
3766                 return;
3767         }
3768 
3769         alloclen = sizeof (*samsg) + sizeof (*current) + sizeof (*expire) +
3770             2 * sizeof (sadb_address_t) + sizeof (*saext);
3771 
3772         af = assoc->ipsa_addrfam;
3773         switch (af) {
3774         case AF_INET:
3775                 alloclen += 2 * sizeof (struct sockaddr_in);
3776                 break;
3777         case AF_INET6:
3778                 alloclen += 2 * sizeof (struct sockaddr_in6);
3779                 break;
3780         default:
3781                 /* Won't happen unless there's a kernel bug. */
3782                 freeb(mp);
3783                 cmn_err(CE_WARN,
3784                     "sadb_expire_assoc: Unknown address length.\n");
3785                 return;
3786         }
3787 
3788         tunnel_mode = (assoc->ipsa_flags & IPSA_F_TUNNEL);
3789         if (tunnel_mode) {
3790                 alloclen += 2 * sizeof (sadb_address_t);
3791                 switch (assoc->ipsa_innerfam) {
3792                 case AF_INET:
3793                         alloclen += 2 * sizeof (struct sockaddr_in);
3794                         break;
3795                 case AF_INET6:
3796                         alloclen += 2 * sizeof (struct sockaddr_in6);
3797                         break;
3798                 default:
3799                         /* Won't happen unless there's a kernel bug. */
3800                         freeb(mp);
3801                         cmn_err(CE_WARN, "sadb_expire_assoc: "
3802                             "Unknown inner address length.\n");
3803                         return;
3804                 }
3805         }
3806 
3807         mp->b_cont = allocb(alloclen, BPRI_HI);
3808         if (mp->b_cont == NULL) {
3809                 freeb(mp);
3810                 /* cmn_err(CE_WARN, */
3811                 /*      "sadb_expire_assoc: Can't allocate message.\n"); */
3812                 return;
3813         }
3814 
3815         mp1 = mp;
3816         mp = mp->b_cont;
3817         end = mp->b_wptr + alloclen;
3818 
3819         samsg = (sadb_msg_t *)mp->b_wptr;
3820         mp->b_wptr += sizeof (*samsg);
3821         samsg->sadb_msg_version = PF_KEY_V2;
3822         samsg->sadb_msg_type = SADB_EXPIRE;
3823         samsg->sadb_msg_errno = 0;
3824         samsg->sadb_msg_satype = assoc->ipsa_type;
3825         samsg->sadb_msg_len = SADB_8TO64(alloclen);
3826         samsg->sadb_msg_reserved = 0;
3827         samsg->sadb_msg_seq = 0;
3828         samsg->sadb_msg_pid = 0;
3829 
3830         saext = (sadb_sa_t *)mp->b_wptr;
3831         mp->b_wptr += sizeof (*saext);
3832         saext->sadb_sa_len = SADB_8TO64(sizeof (*saext));
3833         saext->sadb_sa_exttype = SADB_EXT_SA;
3834         saext->sadb_sa_spi = assoc->ipsa_spi;
3835         saext->sadb_sa_replay = assoc->ipsa_replay_wsize;
3836         saext->sadb_sa_state = assoc->ipsa_state;
3837         saext->sadb_sa_auth = assoc->ipsa_auth_alg;
3838         saext->sadb_sa_encrypt = assoc->ipsa_encr_alg;
3839         saext->sadb_sa_flags = assoc->ipsa_flags;
3840 
3841         current = (sadb_lifetime_t *)mp->b_wptr;
3842         mp->b_wptr += sizeof (sadb_lifetime_t);
3843         current->sadb_lifetime_len = SADB_8TO64(sizeof (*current));
3844         current->sadb_lifetime_exttype = SADB_EXT_LIFETIME_CURRENT;
3845         /* We do not support the concept. */
3846         current->sadb_lifetime_allocations = 0;
3847         current->sadb_lifetime_bytes = assoc->ipsa_bytes;
3848         current->sadb_lifetime_addtime = assoc->ipsa_addtime;
3849         current->sadb_lifetime_usetime = assoc->ipsa_usetime;
3850 
3851         expire = (sadb_lifetime_t *)mp->b_wptr;
3852         mp->b_wptr += sizeof (*expire);
3853         expire->sadb_lifetime_len = SADB_8TO64(sizeof (*expire));
3854 
3855         if (assoc->ipsa_state == IPSA_STATE_DEAD) {
3856                 expire->sadb_lifetime_exttype = SADB_EXT_LIFETIME_HARD;
3857                 expire->sadb_lifetime_allocations = assoc->ipsa_hardalloc;
3858                 expire->sadb_lifetime_bytes = assoc->ipsa_hardbyteslt;
3859                 expire->sadb_lifetime_addtime = assoc->ipsa_hardaddlt;
3860                 expire->sadb_lifetime_usetime = assoc->ipsa_harduselt;
3861         } else if (assoc->ipsa_state == IPSA_STATE_DYING) {
3862                 expire->sadb_lifetime_exttype = SADB_EXT_LIFETIME_SOFT;
3863                 expire->sadb_lifetime_allocations = assoc->ipsa_softalloc;
3864                 expire->sadb_lifetime_bytes = assoc->ipsa_softbyteslt;
3865                 expire->sadb_lifetime_addtime = assoc->ipsa_softaddlt;
3866                 expire->sadb_lifetime_usetime = assoc->ipsa_softuselt;
3867         } else {
3868                 ASSERT(assoc->ipsa_state == IPSA_STATE_MATURE);
3869                 expire->sadb_lifetime_exttype = SADB_X_EXT_LIFETIME_IDLE;
3870                 expire->sadb_lifetime_allocations = 0;
3871                 expire->sadb_lifetime_bytes = 0;
3872                 expire->sadb_lifetime_addtime = assoc->ipsa_idleaddlt;
3873                 expire->sadb_lifetime_usetime = assoc->ipsa_idleuselt;
3874         }
3875 
3876         mp->b_wptr = sadb_make_addr_ext(mp->b_wptr, end, SADB_EXT_ADDRESS_SRC,
3877             af, assoc->ipsa_srcaddr, tunnel_mode ? 0 : SA_SRCPORT(assoc),
3878             SA_PROTO(assoc), 0);
3879         ASSERT(mp->b_wptr != NULL);
3880 
3881         mp->b_wptr = sadb_make_addr_ext(mp->b_wptr, end, SADB_EXT_ADDRESS_DST,
3882             af, assoc->ipsa_dstaddr, tunnel_mode ? 0 : SA_DSTPORT(assoc),
3883             SA_PROTO(assoc), 0);
3884         ASSERT(mp->b_wptr != NULL);
3885 
3886         if (tunnel_mode) {
3887                 mp->b_wptr = sadb_make_addr_ext(mp->b_wptr, end,
3888                     SADB_X_EXT_ADDRESS_INNER_SRC, assoc->ipsa_innerfam,
3889                     assoc->ipsa_innersrc, SA_SRCPORT(assoc), SA_IPROTO(assoc),
3890                     assoc->ipsa_innersrcpfx);
3891                 ASSERT(mp->b_wptr != NULL);
3892                 mp->b_wptr = sadb_make_addr_ext(mp->b_wptr, end,
3893                     SADB_X_EXT_ADDRESS_INNER_DST, assoc->ipsa_innerfam,
3894                     assoc->ipsa_innerdst, SA_DSTPORT(assoc), SA_IPROTO(assoc),
3895                     assoc->ipsa_innerdstpfx);
3896                 ASSERT(mp->b_wptr != NULL);
3897         }
3898 
3899         /* Can just putnext, we're ready to go! */
3900         putnext(pfkey_q, mp1);
3901 }
3902 
3903 /*
3904  * "Age" the SA with the number of bytes that was used to protect traffic.
3905  * Send an SADB_EXPIRE message if appropriate.  Return B_TRUE if there was
3906  * enough "charge" left in the SA to protect the data.  Return B_FALSE
3907  * otherwise.  (If B_FALSE is returned, the association either was, or became
3908  * DEAD.)
3909  */
3910 boolean_t
3911 sadb_age_bytes(queue_t *pfkey_q, ipsa_t *assoc, uint64_t bytes,
3912     boolean_t sendmsg)
3913 {
3914         boolean_t rc = B_TRUE;
3915         uint64_t newtotal;
3916 
3917         mutex_enter(&assoc->ipsa_lock);
3918         newtotal = assoc->ipsa_bytes + bytes;
3919         if (assoc->ipsa_hardbyteslt != 0 &&
3920             newtotal >= assoc->ipsa_hardbyteslt) {
3921                 if (assoc->ipsa_state != IPSA_STATE_DEAD) {
3922                         sadb_delete_cluster(assoc);
3923                         /*
3924                          * Send EXPIRE message to PF_KEY.  May wish to pawn
3925                          * this off on another non-interrupt thread.  Also
3926                          * unlink this SA immediately.
3927                          */
3928                         assoc->ipsa_state = IPSA_STATE_DEAD;
3929                         if (sendmsg)
3930                                 sadb_expire_assoc(pfkey_q, assoc);
3931                         /*
3932                          * Set non-zero expiration time so sadb_age_assoc()
3933                          * will work when reaping.
3934                          */
3935                         assoc->ipsa_hardexpiretime = (time_t)1;
3936                 } /* Else someone beat me to it! */
3937                 rc = B_FALSE;
3938         } else if (assoc->ipsa_softbyteslt != 0 &&
3939             (newtotal >= assoc->ipsa_softbyteslt)) {
3940                 if (assoc->ipsa_state < IPSA_STATE_DYING) {
3941                         /*
3942                          * Send EXPIRE message to PF_KEY.  May wish to pawn
3943                          * this off on another non-interrupt thread.
3944                          */
3945                         assoc->ipsa_state = IPSA_STATE_DYING;
3946                         assoc->ipsa_bytes = newtotal;
3947                         if (sendmsg)
3948                                 sadb_expire_assoc(pfkey_q, assoc);
3949                 } /* Else someone beat me to it! */
3950         }
3951         if (rc == B_TRUE)
3952                 assoc->ipsa_bytes = newtotal;
3953         mutex_exit(&assoc->ipsa_lock);
3954         return (rc);
3955 }
3956 
3957 /*
3958  * "Torch" an individual SA.  Returns NULL, so it can be tail-called from
3959  *     sadb_age_assoc().
3960  */
3961 static ipsa_t *
3962 sadb_torch_assoc(isaf_t *head, ipsa_t *sa)
3963 {
3964         ASSERT(MUTEX_HELD(&head->isaf_lock));
3965         ASSERT(MUTEX_HELD(&sa->ipsa_lock));
3966         ASSERT(sa->ipsa_state == IPSA_STATE_DEAD);
3967 
3968         /*
3969          * Force cached SAs to be revalidated..
3970          */
3971         head->isaf_gen++;
3972 
3973         mutex_exit(&sa->ipsa_lock);
3974         sadb_unlinkassoc(sa);
3975 
3976         return (NULL);
3977 }
3978 
3979 /*
3980  * Do various SA-is-idle activities depending on delta (the number of idle
3981  * seconds on the SA) and/or other properties of the SA.
3982  *
3983  * Return B_TRUE if I've sent a packet, because I have to drop the
3984  * association's mutex before sending a packet out the wire.
3985  */
3986 /* ARGSUSED */
3987 static boolean_t
3988 sadb_idle_activities(ipsa_t *assoc, time_t delta, boolean_t inbound)
3989 {
3990         ipsecesp_stack_t *espstack = assoc->ipsa_netstack->netstack_ipsecesp;
3991         int nat_t_interval = espstack->ipsecesp_nat_keepalive_interval;
3992 
3993         ASSERT(MUTEX_HELD(&assoc->ipsa_lock));
3994 
3995         if (!inbound && (assoc->ipsa_flags & IPSA_F_NATT_LOC) &&
3996             delta >= nat_t_interval &&
3997             gethrestime_sec() - assoc->ipsa_last_nat_t_ka >= nat_t_interval) {
3998                 ASSERT(assoc->ipsa_type == SADB_SATYPE_ESP);
3999                 assoc->ipsa_last_nat_t_ka = gethrestime_sec();
4000                 mutex_exit(&assoc->ipsa_lock);
4001                 ipsecesp_send_keepalive(assoc);
4002                 return (B_TRUE);
4003         }
4004         return (B_FALSE);
4005 }
4006 
4007 /*
4008  * Return "assoc" if haspeer is true and I send an expire.  This allows
4009  * the consumers' aging functions to tidy up an expired SA's peer.
4010  */
4011 static ipsa_t *
4012 sadb_age_assoc(isaf_t *head, queue_t *pfkey_q, ipsa_t *assoc,
4013     time_t current, int reap_delay, boolean_t inbound)
4014 {
4015         ipsa_t *retval = NULL;
4016         boolean_t dropped_mutex = B_FALSE;
4017 
4018         ASSERT(MUTEX_HELD(&head->isaf_lock));
4019 
4020         mutex_enter(&assoc->ipsa_lock);
4021 
4022         if (((assoc->ipsa_state == IPSA_STATE_LARVAL) ||
4023             ((assoc->ipsa_state == IPSA_STATE_IDLE) ||
4024             (assoc->ipsa_state == IPSA_STATE_ACTIVE_ELSEWHERE) &&
4025             (assoc->ipsa_hardexpiretime != 0))) &&
4026             (assoc->ipsa_hardexpiretime <= current)) {
4027                 assoc->ipsa_state = IPSA_STATE_DEAD;
4028                 return (sadb_torch_assoc(head, assoc));
4029         }
4030 
4031         /*
4032          * Check lifetimes.  Fortunately, SA setup is done
4033          * such that there are only two times to look at,
4034          * softexpiretime, and hardexpiretime.
4035          *
4036          * Check hard first.
4037          */
4038 
4039         if (assoc->ipsa_hardexpiretime != 0 &&
4040             assoc->ipsa_hardexpiretime <= current) {
4041                 if (assoc->ipsa_state == IPSA_STATE_DEAD)
4042                         return (sadb_torch_assoc(head, assoc));
4043 
4044                 if (inbound) {
4045                         sadb_delete_cluster(assoc);
4046                 }
4047 
4048                 /*
4049                  * Send SADB_EXPIRE with hard lifetime, delay for unlinking.
4050                  */
4051                 assoc->ipsa_state = IPSA_STATE_DEAD;
4052                 if (assoc->ipsa_haspeer || assoc->ipsa_otherspi != 0) {
4053                         /*
4054                          * If the SA is paired or peered with another, put
4055                          * a copy on a list which can be processed later, the
4056                          * pair/peer SA needs to be updated so the both die
4057                          * at the same time.
4058                          *
4059                          * If I return assoc, I have to bump up its reference
4060                          * count to keep with the ipsa_t reference count
4061                          * semantics.
4062                          */
4063                         IPSA_REFHOLD(assoc);
4064                         retval = assoc;
4065                 }
4066                 sadb_expire_assoc(pfkey_q, assoc);
4067                 assoc->ipsa_hardexpiretime = current + reap_delay;
4068         } else if (assoc->ipsa_softexpiretime != 0 &&
4069             assoc->ipsa_softexpiretime <= current &&
4070             assoc->ipsa_state < IPSA_STATE_DYING) {
4071                 /*
4072                  * Send EXPIRE message to PF_KEY.  May wish to pawn
4073                  * this off on another non-interrupt thread.
4074                  */
4075                 assoc->ipsa_state = IPSA_STATE_DYING;
4076                 if (assoc->ipsa_haspeer) {
4077                         /*
4078                          * If the SA has a peer, update the peer's state
4079                          * on SOFT_EXPIRE, this is mostly to prevent two
4080                          * expire messages from effectively the same SA.
4081                          *
4082                          * Don't care about paired SA's, then can (and should)
4083                          * be able to soft expire at different times.
4084                          *
4085                          * If I return assoc, I have to bump up its
4086                          * reference count to keep with the ipsa_t reference
4087                          * count semantics.
4088                          */
4089                         IPSA_REFHOLD(assoc);
4090                         retval = assoc;
4091                 }
4092                 sadb_expire_assoc(pfkey_q, assoc);
4093         } else if (assoc->ipsa_idletime != 0 &&
4094             assoc->ipsa_idleexpiretime <= current) {
4095                 if (assoc->ipsa_state == IPSA_STATE_ACTIVE_ELSEWHERE) {
4096                         assoc->ipsa_state = IPSA_STATE_IDLE;
4097                 }
4098 
4099                 /*
4100                  * Need to handle Mature case
4101                  */
4102                 if (assoc->ipsa_state == IPSA_STATE_MATURE) {
4103                         sadb_expire_assoc(pfkey_q, assoc);
4104                 }
4105         } else {
4106                 /* Check idle time activities. */
4107                 dropped_mutex = sadb_idle_activities(assoc,
4108                     current - assoc->ipsa_lastuse, inbound);
4109         }
4110 
4111         if (!dropped_mutex)
4112                 mutex_exit(&assoc->ipsa_lock);
4113         return (retval);
4114 }
4115 
4116 /*
4117  * Called by a consumer protocol to do ther dirty work of reaping dead
4118  * Security Associations.
4119  *
4120  * NOTE: sadb_age_assoc() marks expired SA's as DEAD but only removed
4121  * SA's that are already marked DEAD, so expired SA's are only reaped
4122  * the second time sadb_ager() runs.
4123  */
4124 void
4125 sadb_ager(sadb_t *sp, queue_t *pfkey_q, int reap_delay, netstack_t *ns)
4126 {
4127         int i;
4128         isaf_t *bucket;
4129         ipsa_t *assoc, *spare;
4130         iacqf_t *acqlist;
4131         ipsacq_t *acqrec, *spareacq;
4132         templist_t *haspeerlist, *newbie;
4133         /* Snapshot current time now. */
4134         time_t current = gethrestime_sec();
4135         haspeerlist = NULL;
4136 
4137         /*
4138          * Do my dirty work.  This includes aging real entries, aging
4139          * larvals, and aging outstanding ACQUIREs.
4140          *
4141          * I hope I don't tie up resources for too long.
4142          */
4143 
4144         /* Age acquires. */
4145 
4146         for (i = 0; i < sp->sdb_hashsize; i++) {
4147                 acqlist = &sp->sdb_acq[i];
4148                 mutex_enter(&acqlist->iacqf_lock);
4149                 for (acqrec = acqlist->iacqf_ipsacq; acqrec != NULL;
4150                     acqrec = spareacq) {
4151                         spareacq = acqrec->ipsacq_next;
4152                         if (current > acqrec->ipsacq_expire)
4153                                 sadb_destroy_acquire(acqrec, ns);
4154                 }
4155                 mutex_exit(&acqlist->iacqf_lock);
4156         }
4157 
4158         /* Age inbound associations. */
4159         for (i = 0; i < sp->sdb_hashsize; i++) {
4160                 bucket = &(sp->sdb_if[i]);
4161                 mutex_enter(&bucket->isaf_lock);
4162                 for (assoc = bucket->isaf_ipsa; assoc != NULL;
4163                     assoc = spare) {
4164                         spare = assoc->ipsa_next;
4165                         if (sadb_age_assoc(bucket, pfkey_q, assoc, current,
4166                             reap_delay, B_TRUE) != NULL) {
4167                                 /*
4168                                  * Put SA's which have a peer or SA's which
4169                                  * are paired on a list for processing after
4170                                  * all the hash tables have been walked.
4171                                  *
4172                                  * sadb_age_assoc() increments the refcnt,
4173                                  * effectively doing an IPSA_REFHOLD().
4174                                  */
4175                                 newbie = kmem_alloc(sizeof (*newbie),
4176                                     KM_NOSLEEP);
4177                                 if (newbie == NULL) {
4178                                         /*
4179                                          * Don't forget to REFRELE().
4180                                          */
4181                                         IPSA_REFRELE(assoc);
4182                                         continue;       /* for loop... */
4183                                 }
4184                                 newbie->next = haspeerlist;
4185                                 newbie->ipsa = assoc;
4186                                 haspeerlist = newbie;
4187                         }
4188                 }
4189                 mutex_exit(&bucket->isaf_lock);
4190         }
4191 
4192         age_pair_peer_list(haspeerlist, sp, B_FALSE);
4193         haspeerlist = NULL;
4194 
4195         /* Age outbound associations. */
4196         for (i = 0; i < sp->sdb_hashsize; i++) {
4197                 bucket = &(sp->sdb_of[i]);
4198                 mutex_enter(&bucket->isaf_lock);
4199                 for (assoc = bucket->isaf_ipsa; assoc != NULL;
4200                     assoc = spare) {
4201                         spare = assoc->ipsa_next;
4202                         if (sadb_age_assoc(bucket, pfkey_q, assoc, current,
4203                             reap_delay, B_FALSE) != NULL) {
4204                                 /*
4205                                  * sadb_age_assoc() increments the refcnt,
4206                                  * effectively doing an IPSA_REFHOLD().
4207                                  */
4208                                 newbie = kmem_alloc(sizeof (*newbie),
4209                                     KM_NOSLEEP);
4210                                 if (newbie == NULL) {
4211                                         /*
4212                                          * Don't forget to REFRELE().
4213                                          */
4214                                         IPSA_REFRELE(assoc);
4215                                         continue;       /* for loop... */
4216                                 }
4217                                 newbie->next = haspeerlist;
4218                                 newbie->ipsa = assoc;
4219                                 haspeerlist = newbie;
4220                         }
4221                 }
4222                 mutex_exit(&bucket->isaf_lock);
4223         }
4224 
4225         age_pair_peer_list(haspeerlist, sp, B_TRUE);
4226 
4227         /*
4228          * Run a GC pass to clean out dead identities.
4229          */
4230         ipsid_gc(ns);
4231 }
4232 
4233 /*
4234  * Figure out when to reschedule the ager.
4235  */
4236 timeout_id_t
4237 sadb_retimeout(hrtime_t begin, queue_t *pfkey_q, void (*ager)(void *),
4238     void *agerarg, uint_t *intp, uint_t intmax, short mid)
4239 {
4240         hrtime_t end = gethrtime();
4241         uint_t interval = *intp;        /* "interval" is in ms. */
4242 
4243         /*
4244          * See how long this took.  If it took too long, increase the
4245          * aging interval.
4246          */
4247         if ((end - begin) > MSEC2NSEC(interval)) {
4248                 if (interval >= intmax) {
4249                         /* XXX Rate limit this?  Or recommend flush? */
4250                         (void) strlog(mid, 0, 0, SL_ERROR | SL_WARN,
4251                             "Too many SA's to age out in %d msec.\n",
4252                             intmax);
4253                 } else {
4254                         /* Double by shifting by one bit. */
4255                         interval <<= 1;
4256                         interval = min(interval, intmax);
4257                 }
4258         } else if ((end - begin) <= (MSEC2NSEC(interval) / 2) &&
4259             interval > SADB_AGE_INTERVAL_DEFAULT) {
4260                 /*
4261                  * If I took less than half of the interval, then I should
4262                  * ratchet the interval back down.  Never automatically
4263                  * shift below the default aging interval.
4264                  *
4265                  * NOTE:This even overrides manual setting of the age
4266                  *      interval using NDD to lower the setting past the
4267                  *      default.  In other words, if you set the interval
4268                  *      lower than the default, and your SADB gets too big,
4269                  *      the interval will only self-lower back to the default.
4270                  */
4271                 /* Halve by shifting one bit. */
4272                 interval >>= 1;
4273                 interval = max(interval, SADB_AGE_INTERVAL_DEFAULT);
4274         }
4275         *intp = interval;
4276         return (qtimeout(pfkey_q, ager, agerarg,
4277             drv_usectohz(interval * (MICROSEC / MILLISEC))));
4278 }
4279 
4280 
4281 /*
4282  * Update the lifetime values of an SA.  This is the path an SADB_UPDATE
4283  * message takes when updating a MATURE or DYING SA.
4284  */
4285 static void
4286 sadb_update_lifetimes(ipsa_t *assoc, sadb_lifetime_t *hard,
4287     sadb_lifetime_t *soft, sadb_lifetime_t *idle, boolean_t outbound)
4288 {
4289         mutex_enter(&assoc->ipsa_lock);
4290 
4291         /*
4292          * XXX RFC 2367 mentions how an SADB_EXT_LIFETIME_CURRENT can be
4293          * passed in during an update message.  We currently don't handle
4294          * these.
4295          */
4296 
4297         if (hard != NULL) {
4298                 if (hard->sadb_lifetime_bytes != 0)
4299                         assoc->ipsa_hardbyteslt = hard->sadb_lifetime_bytes;
4300                 if (hard->sadb_lifetime_usetime != 0)
4301                         assoc->ipsa_harduselt = hard->sadb_lifetime_usetime;
4302                 if (hard->sadb_lifetime_addtime != 0)
4303                         assoc->ipsa_hardaddlt = hard->sadb_lifetime_addtime;
4304                 if (assoc->ipsa_hardaddlt != 0) {
4305                         assoc->ipsa_hardexpiretime =
4306                             assoc->ipsa_addtime + assoc->ipsa_hardaddlt;
4307                 }
4308                 if (assoc->ipsa_harduselt != 0 &&
4309                     assoc->ipsa_flags & IPSA_F_USED) {
4310                         UPDATE_EXPIRE(assoc, harduselt, hardexpiretime);
4311                 }
4312                 if (hard->sadb_lifetime_allocations != 0)
4313                         assoc->ipsa_hardalloc = hard->sadb_lifetime_allocations;
4314         }
4315 
4316         if (soft != NULL) {
4317                 if (soft->sadb_lifetime_bytes != 0) {
4318                         if (soft->sadb_lifetime_bytes >
4319                             assoc->ipsa_hardbyteslt) {
4320                                 assoc->ipsa_softbyteslt =
4321                                     assoc->ipsa_hardbyteslt;
4322                         } else {
4323                                 assoc->ipsa_softbyteslt =
4324                                     soft->sadb_lifetime_bytes;
4325                         }
4326                 }
4327                 if (soft->sadb_lifetime_usetime != 0) {
4328                         if (soft->sadb_lifetime_usetime >
4329                             assoc->ipsa_harduselt) {
4330                                 assoc->ipsa_softuselt =
4331                                     assoc->ipsa_harduselt;
4332                         } else {
4333                                 assoc->ipsa_softuselt =
4334                                     soft->sadb_lifetime_usetime;
4335                         }
4336                 }
4337                 if (soft->sadb_lifetime_addtime != 0) {
4338                         if (soft->sadb_lifetime_addtime >
4339                             assoc->ipsa_hardexpiretime) {
4340                                 assoc->ipsa_softexpiretime =
4341                                     assoc->ipsa_hardexpiretime;
4342                         } else {
4343                                 assoc->ipsa_softaddlt =
4344                                     soft->sadb_lifetime_addtime;
4345                         }
4346                 }
4347                 if (assoc->ipsa_softaddlt != 0) {
4348                         assoc->ipsa_softexpiretime =
4349                             assoc->ipsa_addtime + assoc->ipsa_softaddlt;
4350                 }
4351                 if (assoc->ipsa_softuselt != 0 &&
4352                     assoc->ipsa_flags & IPSA_F_USED) {
4353                         UPDATE_EXPIRE(assoc, softuselt, softexpiretime);
4354                 }
4355                 if (outbound && assoc->ipsa_softexpiretime != 0) {
4356                         if (assoc->ipsa_state == IPSA_STATE_MATURE)
4357                                 lifetime_fuzz(assoc);
4358                 }
4359 
4360                 if (soft->sadb_lifetime_allocations != 0)
4361                         assoc->ipsa_softalloc = soft->sadb_lifetime_allocations;
4362         }
4363 
4364         if (idle != NULL) {
4365                 time_t current = gethrestime_sec();
4366                 if ((assoc->ipsa_idleexpiretime <= current) &&
4367                     (assoc->ipsa_idleaddlt == idle->sadb_lifetime_addtime)) {
4368                         assoc->ipsa_idleexpiretime =
4369                             current + assoc->ipsa_idleaddlt;
4370                 }
4371                 if (idle->sadb_lifetime_addtime != 0)
4372                         assoc->ipsa_idleaddlt = idle->sadb_lifetime_addtime;
4373                 if (idle->sadb_lifetime_usetime != 0)
4374                         assoc->ipsa_idleuselt = idle->sadb_lifetime_usetime;
4375                 if (assoc->ipsa_idleaddlt != 0) {
4376                         assoc->ipsa_idleexpiretime =
4377                             current + idle->sadb_lifetime_addtime;
4378                         assoc->ipsa_idletime = idle->sadb_lifetime_addtime;
4379                 }
4380                 if (assoc->ipsa_idleuselt != 0) {
4381                         if (assoc->ipsa_idletime != 0) {
4382                                 assoc->ipsa_idletime = min(assoc->ipsa_idletime,
4383                                     assoc->ipsa_idleuselt);
4384                         assoc->ipsa_idleexpiretime =
4385                             current + assoc->ipsa_idletime;
4386                         } else {
4387                                 assoc->ipsa_idleexpiretime =
4388                                     current + assoc->ipsa_idleuselt;
4389                                 assoc->ipsa_idletime = assoc->ipsa_idleuselt;
4390                         }
4391                 }
4392         }
4393         mutex_exit(&assoc->ipsa_lock);
4394 }
4395 
4396 static int
4397 sadb_update_state(ipsa_t *assoc, uint_t new_state, mblk_t **ipkt_lst)
4398 {
4399         int rcode = 0;
4400         time_t current = gethrestime_sec();
4401 
4402         mutex_enter(&assoc->ipsa_lock);
4403 
4404         switch (new_state) {
4405         case SADB_X_SASTATE_ACTIVE_ELSEWHERE:
4406                 if (assoc->ipsa_state == SADB_X_SASTATE_IDLE) {
4407                         assoc->ipsa_state = IPSA_STATE_ACTIVE_ELSEWHERE;
4408                         assoc->ipsa_idleexpiretime =
4409                             current + assoc->ipsa_idletime;
4410                 }
4411                 break;
4412         case SADB_X_SASTATE_IDLE:
4413                 if (assoc->ipsa_state == SADB_X_SASTATE_ACTIVE_ELSEWHERE) {
4414                         assoc->ipsa_state = IPSA_STATE_IDLE;
4415                         assoc->ipsa_idleexpiretime =
4416                             current + assoc->ipsa_idletime;
4417                 } else {
4418                         rcode = EINVAL;
4419                 }
4420                 break;
4421 
4422         case SADB_X_SASTATE_ACTIVE:
4423                 if (assoc->ipsa_state != SADB_X_SASTATE_IDLE) {
4424                         rcode = EINVAL;
4425                         break;
4426                 }
4427                 assoc->ipsa_state = IPSA_STATE_MATURE;
4428                 assoc->ipsa_idleexpiretime = current + assoc->ipsa_idletime;
4429 
4430                 if (ipkt_lst == NULL) {
4431                         break;
4432                 }
4433 
4434                 if (assoc->ipsa_bpkt_head != NULL) {
4435                         *ipkt_lst = assoc->ipsa_bpkt_head;
4436                         assoc->ipsa_bpkt_head = assoc->ipsa_bpkt_tail = NULL;
4437                         assoc->ipsa_mblkcnt = 0;
4438                 } else {
4439                         *ipkt_lst = NULL;
4440                 }
4441                 break;
4442         default:
4443                 rcode = EINVAL;
4444                 break;
4445         }
4446 
4447         mutex_exit(&assoc->ipsa_lock);
4448         return (rcode);
4449 }
4450 
4451 /*
4452  * Check a proposed KMC update for sanity.
4453  */
4454 static int
4455 sadb_check_kmc(ipsa_query_t *sq, ipsa_t *sa, int *diagnostic)
4456 {
4457         uint32_t kmp = sq->kmp;
4458         uint64_t kmc = sq->kmc;
4459 
4460         if (sa == NULL)
4461                 return (0);
4462 
4463         if (sa->ipsa_state == IPSA_STATE_DEAD)
4464                 return (ESRCH); /* DEAD == Not there, in this case. */
4465 
4466         if ((kmp != 0) && (sa->ipsa_kmp != 0) && (sa->ipsa_kmp != kmp)) {
4467                 *diagnostic = SADB_X_DIAGNOSTIC_DUPLICATE_KMP;
4468                 return (EINVAL);
4469         }
4470 
4471         if ((kmc != 0) && (sa->ipsa_kmc != 0) && (sa->ipsa_kmc != kmc)) {
4472                 *diagnostic = SADB_X_DIAGNOSTIC_DUPLICATE_KMC;
4473                 return (EINVAL);
4474         }
4475 
4476         return (0);
4477 }
4478 
4479 /*
4480  * Actually update the KMC info.
4481  */
4482 static void
4483 sadb_update_kmc(ipsa_query_t *sq, ipsa_t *sa)
4484 {
4485         uint32_t kmp = sq->kmp;
4486         uint64_t kmc = sq->kmc;
4487 
4488         if (kmp != 0)
4489                 sa->ipsa_kmp = kmp;
4490         if (kmc != 0)
4491                 sa->ipsa_kmc = kmc;
4492 }
4493 
4494 /*
4495  * Common code to update an SA.
4496  */
4497 
4498 int
4499 sadb_update_sa(mblk_t *mp, keysock_in_t *ksi, mblk_t **ipkt_lst,
4500     sadbp_t *spp, int *diagnostic, queue_t *pfkey_q,
4501     int (*add_sa_func)(mblk_t *, keysock_in_t *, int *, netstack_t *),
4502     netstack_t *ns, uint8_t sadb_msg_type)
4503 {
4504         sadb_key_t *akey = (sadb_key_t *)ksi->ks_in_extv[SADB_EXT_KEY_AUTH];
4505         sadb_key_t *ekey = (sadb_key_t *)ksi->ks_in_extv[SADB_EXT_KEY_ENCRYPT];
4506         sadb_x_replay_ctr_t *replext =
4507             (sadb_x_replay_ctr_t *)ksi->ks_in_extv[SADB_X_EXT_REPLAY_VALUE];
4508         sadb_lifetime_t *soft =
4509             (sadb_lifetime_t *)ksi->ks_in_extv[SADB_EXT_LIFETIME_SOFT];
4510         sadb_lifetime_t *hard =
4511             (sadb_lifetime_t *)ksi->ks_in_extv[SADB_EXT_LIFETIME_HARD];
4512         sadb_lifetime_t *idle =
4513             (sadb_lifetime_t *)ksi->ks_in_extv[SADB_X_EXT_LIFETIME_IDLE];
4514         sadb_x_pair_t *pair_ext =
4515             (sadb_x_pair_t *)ksi->ks_in_extv[SADB_X_EXT_PAIR];
4516         ipsa_t *echo_target = NULL;
4517         ipsap_t ipsapp;
4518         ipsa_query_t sq;
4519         time_t current = gethrestime_sec();
4520 
4521         sq.spp = spp;           /* XXX param */
4522         int error = sadb_form_query(ksi, IPSA_Q_SRC|IPSA_Q_DST|IPSA_Q_SA,
4523             IPSA_Q_SRC|IPSA_Q_DST|IPSA_Q_SA|IPSA_Q_INBOUND|IPSA_Q_OUTBOUND|
4524             IPSA_Q_KMC,
4525             &sq, diagnostic);
4526 
4527         if (error != 0)
4528                 return (error);
4529 
4530         error = get_ipsa_pair(&sq, &ipsapp, diagnostic);
4531         if (error != 0)
4532                 return (error);
4533 
4534         if (ipsapp.ipsap_psa_ptr == NULL && ipsapp.ipsap_sa_ptr != NULL) {
4535                 if (ipsapp.ipsap_sa_ptr->ipsa_state == IPSA_STATE_LARVAL) {
4536                         /*
4537                          * REFRELE the target and let the add_sa_func()
4538                          * deal with updating a larval SA.
4539                          */
4540                         destroy_ipsa_pair(&ipsapp);
4541                         return (add_sa_func(mp, ksi, diagnostic, ns));
4542                 }
4543         }
4544 
4545         /*
4546          * At this point we have an UPDATE to a MATURE SA. There should
4547          * not be any keying material present.
4548          */
4549         if (akey != NULL) {
4550                 *diagnostic = SADB_X_DIAGNOSTIC_AKEY_PRESENT;
4551                 error = EINVAL;
4552                 goto bail;
4553         }
4554         if (ekey != NULL) {
4555                 *diagnostic = SADB_X_DIAGNOSTIC_EKEY_PRESENT;
4556                 error = EINVAL;
4557                 goto bail;
4558         }
4559 
4560         if (sq.assoc->sadb_sa_state == SADB_X_SASTATE_ACTIVE_ELSEWHERE) {
4561                 if (ipsapp.ipsap_sa_ptr != NULL &&
4562                     ipsapp.ipsap_sa_ptr->ipsa_state == IPSA_STATE_IDLE) {
4563                         if ((error = sadb_update_state(ipsapp.ipsap_sa_ptr,
4564                             sq.assoc->sadb_sa_state, NULL)) != 0) {
4565                                 *diagnostic = SADB_X_DIAGNOSTIC_BAD_SASTATE;
4566                                 goto bail;
4567                         }
4568                 }
4569                 if (ipsapp.ipsap_psa_ptr != NULL &&
4570                     ipsapp.ipsap_psa_ptr->ipsa_state == IPSA_STATE_IDLE) {
4571                         if ((error = sadb_update_state(ipsapp.ipsap_psa_ptr,
4572                             sq.assoc->sadb_sa_state, NULL)) != 0) {
4573                                 *diagnostic = SADB_X_DIAGNOSTIC_BAD_SASTATE;
4574                                 goto bail;
4575                         }
4576                 }
4577         }
4578         if (sq.assoc->sadb_sa_state == SADB_X_SASTATE_ACTIVE) {
4579                 if (ipsapp.ipsap_sa_ptr != NULL) {
4580                         error = sadb_update_state(ipsapp.ipsap_sa_ptr,
4581                             sq.assoc->sadb_sa_state,
4582                             (ipsapp.ipsap_sa_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                 if (ipsapp.ipsap_psa_ptr != NULL) {
4590                         error = sadb_update_state(ipsapp.ipsap_psa_ptr,
4591                             sq.assoc->sadb_sa_state,
4592                             (ipsapp.ipsap_psa_ptr->ipsa_flags &
4593                             IPSA_F_INBOUND) ? ipkt_lst : NULL);
4594                         if (error) {
4595                                 *diagnostic = SADB_X_DIAGNOSTIC_BAD_SASTATE;
4596                                 goto bail;
4597                         }
4598                 }
4599                 sadb_pfkey_echo(pfkey_q, mp, (sadb_msg_t *)mp->b_cont->b_rptr,
4600                     ksi, echo_target);
4601                 goto bail;
4602         }
4603 
4604         /*
4605          * Reality checks for updates of active associations.
4606          * Sundry first-pass UPDATE-specific reality checks.
4607          * Have to do the checks here, because it's after the add_sa code.
4608          * XXX STATS : logging/stats here?
4609          */
4610 
4611         if (!((sq.assoc->sadb_sa_state == SADB_SASTATE_MATURE) ||
4612             (sq.assoc->sadb_sa_state == SADB_X_SASTATE_ACTIVE_ELSEWHERE))) {
4613                 *diagnostic = SADB_X_DIAGNOSTIC_BAD_SASTATE;
4614                 error = EINVAL;
4615                 goto bail;
4616         }
4617         if (sq.assoc->sadb_sa_flags & ~spp->s_updateflags) {
4618                 *diagnostic = SADB_X_DIAGNOSTIC_BAD_SAFLAGS;
4619                 error = EINVAL;
4620                 goto bail;
4621         }
4622         if (ksi->ks_in_extv[SADB_EXT_LIFETIME_CURRENT] != NULL) {
4623                 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_LIFETIME;
4624                 error = EOPNOTSUPP;
4625                 goto bail;
4626         }
4627 
4628         if ((*diagnostic = sadb_hardsoftchk(hard, soft, idle)) != 0) {
4629                 error = EINVAL;
4630                 goto bail;
4631         }
4632 
4633         if ((*diagnostic = sadb_labelchk(ksi)) != 0)
4634                 return (EINVAL);
4635 
4636         error = sadb_check_kmc(&sq, ipsapp.ipsap_sa_ptr, diagnostic);
4637         if (error != 0)
4638                 goto bail;
4639 
4640         error = sadb_check_kmc(&sq, ipsapp.ipsap_psa_ptr, diagnostic);
4641         if (error != 0)
4642                 goto bail;
4643 
4644 
4645         if (ipsapp.ipsap_sa_ptr != NULL) {
4646                 /*
4647                  * Do not allow replay value change for MATURE or LARVAL SA.
4648                  */
4649 
4650                 if ((replext != NULL) &&
4651                     ((ipsapp.ipsap_sa_ptr->ipsa_state == IPSA_STATE_LARVAL) ||
4652                     (ipsapp.ipsap_sa_ptr->ipsa_state == IPSA_STATE_MATURE))) {
4653                         *diagnostic = SADB_X_DIAGNOSTIC_BAD_SASTATE;
4654                         error = EINVAL;
4655                         goto bail;
4656                 }
4657         }
4658 
4659 
4660         if (ipsapp.ipsap_sa_ptr != NULL) {
4661                 sadb_update_lifetimes(ipsapp.ipsap_sa_ptr, hard, soft,
4662                     idle, B_TRUE);
4663                 sadb_update_kmc(&sq, ipsapp.ipsap_sa_ptr);
4664                 if ((replext != NULL) &&
4665                     (ipsapp.ipsap_sa_ptr->ipsa_replay_wsize != 0)) {
4666                         /*
4667                          * If an inbound SA, update the replay counter
4668                          * and check off all the other sequence number
4669                          */
4670                         if (ksi->ks_in_dsttype == KS_IN_ADDR_ME) {
4671                                 if (!sadb_replay_check(ipsapp.ipsap_sa_ptr,
4672                                     replext->sadb_x_rc_replay32)) {
4673                                         *diagnostic =
4674                                             SADB_X_DIAGNOSTIC_INVALID_REPLAY;
4675                                         error = EINVAL;
4676                                         goto bail;
4677                                 }
4678                                 mutex_enter(&ipsapp.ipsap_sa_ptr->ipsa_lock);
4679                                 ipsapp.ipsap_sa_ptr->ipsa_idleexpiretime =
4680                                     current +
4681                                     ipsapp.ipsap_sa_ptr->ipsa_idletime;
4682                                 mutex_exit(&ipsapp.ipsap_sa_ptr->ipsa_lock);
4683                         } else {
4684                                 mutex_enter(&ipsapp.ipsap_sa_ptr->ipsa_lock);
4685                                 ipsapp.ipsap_sa_ptr->ipsa_replay =
4686                                     replext->sadb_x_rc_replay32;
4687                                 ipsapp.ipsap_sa_ptr->ipsa_idleexpiretime =
4688                                     current +
4689                                     ipsapp.ipsap_sa_ptr->ipsa_idletime;
4690                                 mutex_exit(&ipsapp.ipsap_sa_ptr->ipsa_lock);
4691                         }
4692                 }
4693         }
4694 
4695         if (sadb_msg_type == SADB_X_UPDATEPAIR) {
4696                 if (ipsapp.ipsap_psa_ptr != NULL) {
4697                         sadb_update_lifetimes(ipsapp.ipsap_psa_ptr, hard, soft,
4698                             idle, B_FALSE);
4699                         sadb_update_kmc(&sq, ipsapp.ipsap_psa_ptr);
4700                 } else {
4701                         *diagnostic = SADB_X_DIAGNOSTIC_PAIR_SA_NOTFOUND;
4702                         error = ESRCH;
4703                         goto bail;
4704                 }
4705         }
4706 
4707         if (pair_ext != NULL)
4708                 error = update_pairing(&ipsapp, &sq, ksi, diagnostic);
4709 
4710         if (error == 0)
4711                 sadb_pfkey_echo(pfkey_q, mp, (sadb_msg_t *)mp->b_cont->b_rptr,
4712                     ksi, echo_target);
4713 bail:
4714 
4715         destroy_ipsa_pair(&ipsapp);
4716 
4717         return (error);
4718 }
4719 
4720 
4721 static int
4722 update_pairing(ipsap_t *ipsapp, ipsa_query_t *sq, keysock_in_t *ksi,
4723     int *diagnostic)
4724 {
4725         sadb_sa_t *assoc = (sadb_sa_t *)ksi->ks_in_extv[SADB_EXT_SA];
4726         sadb_x_pair_t *pair_ext =
4727             (sadb_x_pair_t *)ksi->ks_in_extv[SADB_X_EXT_PAIR];
4728         int error = 0;
4729         ipsap_t oipsapp;
4730         boolean_t undo_pair = B_FALSE;
4731         uint32_t ipsa_flags;
4732 
4733         if (pair_ext->sadb_x_pair_spi == 0 || pair_ext->sadb_x_pair_spi ==
4734             assoc->sadb_sa_spi) {
4735                 *diagnostic = SADB_X_DIAGNOSTIC_PAIR_INAPPROPRIATE;
4736                 return (EINVAL);
4737         }
4738 
4739         /*
4740          * Assume for now that the spi value provided in the SADB_UPDATE
4741          * message was valid, update the SA with its pair spi value.
4742          * If the spi turns out to be bogus or the SA no longer exists
4743          * then this will be detected when the reverse update is made
4744          * below.
4745          */
4746         mutex_enter(&ipsapp->ipsap_sa_ptr->ipsa_lock);
4747         ipsapp->ipsap_sa_ptr->ipsa_flags |= IPSA_F_PAIRED;
4748         ipsapp->ipsap_sa_ptr->ipsa_otherspi = pair_ext->sadb_x_pair_spi;
4749         mutex_exit(&ipsapp->ipsap_sa_ptr->ipsa_lock);
4750 
4751         /*
4752          * After updating the ipsa_otherspi element of the SA, get_ipsa_pair()
4753          * should now return pointers to the SA *AND* its pair, if this is not
4754          * the case, the "otherspi" either did not exist or was deleted. Also
4755          * check that "otherspi" is not already paired. If everything looks
4756          * good, complete the update. IPSA_REFRELE the first pair_pointer
4757          * after this update to ensure its not deleted until we are done.
4758          */
4759         error = get_ipsa_pair(sq, &oipsapp, diagnostic);
4760         if (error != 0) {
4761                 /*
4762                  * This should never happen, calling function still has
4763                  * IPSA_REFHELD on the SA we just updated.
4764                  */
4765                 return (error); /* XXX EINVAL instead of ESRCH? */
4766         }
4767 
4768         if (oipsapp.ipsap_psa_ptr == NULL) {
4769                 *diagnostic = SADB_X_DIAGNOSTIC_PAIR_INAPPROPRIATE;
4770                 error = EINVAL;
4771                 undo_pair = B_TRUE;
4772         } else {
4773                 ipsa_flags = oipsapp.ipsap_psa_ptr->ipsa_flags;
4774                 if ((oipsapp.ipsap_psa_ptr->ipsa_state == IPSA_STATE_DEAD) ||
4775                     (oipsapp.ipsap_psa_ptr->ipsa_state == IPSA_STATE_DYING)) {
4776                         /* Its dead Jim! */
4777                         *diagnostic = SADB_X_DIAGNOSTIC_PAIR_INAPPROPRIATE;
4778                         undo_pair = B_TRUE;
4779                 } else if ((ipsa_flags & (IPSA_F_OUTBOUND | IPSA_F_INBOUND)) ==
4780                     (IPSA_F_OUTBOUND | IPSA_F_INBOUND)) {
4781                         /* This SA is in both hashtables. */
4782                         *diagnostic = SADB_X_DIAGNOSTIC_PAIR_INAPPROPRIATE;
4783                         undo_pair = B_TRUE;
4784                 } else if (ipsa_flags & IPSA_F_PAIRED) {
4785                         /* This SA is already paired with another. */
4786                         *diagnostic = SADB_X_DIAGNOSTIC_PAIR_ALREADY;
4787                         undo_pair = B_TRUE;
4788                 }
4789         }
4790 
4791         if (undo_pair) {
4792                 /* The pair SA does not exist. */
4793                 mutex_enter(&ipsapp->ipsap_sa_ptr->ipsa_lock);
4794                 ipsapp->ipsap_sa_ptr->ipsa_flags &= ~IPSA_F_PAIRED;
4795                 ipsapp->ipsap_sa_ptr->ipsa_otherspi = 0;
4796                 mutex_exit(&ipsapp->ipsap_sa_ptr->ipsa_lock);
4797         } else {
4798                 mutex_enter(&oipsapp.ipsap_psa_ptr->ipsa_lock);
4799                 oipsapp.ipsap_psa_ptr->ipsa_otherspi = assoc->sadb_sa_spi;
4800                 oipsapp.ipsap_psa_ptr->ipsa_flags |= IPSA_F_PAIRED;
4801                 mutex_exit(&oipsapp.ipsap_psa_ptr->ipsa_lock);
4802         }
4803 
4804         destroy_ipsa_pair(&oipsapp);
4805         return (error);
4806 }
4807 
4808 /*
4809  * The following functions deal with ACQUIRE LISTS.  An ACQUIRE list is
4810  * a list of outstanding SADB_ACQUIRE messages.  If ipsec_getassocbyconn() fails
4811  * for an outbound datagram, that datagram is queued up on an ACQUIRE record,
4812  * and an SADB_ACQUIRE message is sent up.  Presumably, a user-space key
4813  * management daemon will process the ACQUIRE, use a SADB_GETSPI to reserve
4814  * an SPI value and a larval SA, then SADB_UPDATE the larval SA, and ADD the
4815  * other direction's SA.
4816  */
4817 
4818 /*
4819  * Check the ACQUIRE lists.  If there's an existing ACQUIRE record,
4820  * grab it, lock it, and return it.  Otherwise return NULL.
4821  *
4822  * XXX MLS number of arguments getting unwieldy here
4823  */
4824 static ipsacq_t *
4825 sadb_checkacquire(iacqf_t *bucket, ipsec_action_t *ap, ipsec_policy_t *pp,
4826     uint32_t *src, uint32_t *dst, uint32_t *isrc, uint32_t *idst,
4827     uint64_t unique_id, ts_label_t *tsl)
4828 {
4829         ipsacq_t *walker;
4830         sa_family_t fam;
4831         uint32_t blank_address[4] = {0, 0, 0, 0};
4832 
4833         if (isrc == NULL) {
4834                 ASSERT(idst == NULL);
4835                 isrc = idst = blank_address;
4836         }
4837 
4838         /*
4839          * Scan list for duplicates.  Check for UNIQUE, src/dest, policy.
4840          *
4841          * XXX May need search for duplicates based on other things too!
4842          */
4843         for (walker = bucket->iacqf_ipsacq; walker != NULL;
4844             walker = walker->ipsacq_next) {
4845                 mutex_enter(&walker->ipsacq_lock);
4846                 fam = walker->ipsacq_addrfam;
4847                 if (IPSA_ARE_ADDR_EQUAL(dst, walker->ipsacq_dstaddr, fam) &&
4848                     IPSA_ARE_ADDR_EQUAL(src, walker->ipsacq_srcaddr, fam) &&
4849                     ip_addr_match((uint8_t *)isrc, walker->ipsacq_innersrcpfx,
4850                     (in6_addr_t *)walker->ipsacq_innersrc) &&
4851                     ip_addr_match((uint8_t *)idst, walker->ipsacq_innerdstpfx,
4852                     (in6_addr_t *)walker->ipsacq_innerdst) &&
4853                     (ap == walker->ipsacq_act) &&
4854                     (pp == walker->ipsacq_policy) &&
4855                     /* XXX do deep compares of ap/pp? */
4856                     (unique_id == walker->ipsacq_unique_id) &&
4857                     (ipsec_label_match(tsl, walker->ipsacq_tsl)))
4858                         break;                  /* everything matched */
4859                 mutex_exit(&walker->ipsacq_lock);
4860         }
4861 
4862         return (walker);
4863 }
4864 
4865 /*
4866  * Generate an SADB_ACQUIRE base message mblk, including KEYSOCK_OUT metadata.
4867  * In other words, this will return, upon success, a two-mblk chain.
4868  */
4869 static inline mblk_t *
4870 sadb_acquire_msg_base(minor_t serial, uint8_t satype, uint32_t seq, pid_t pid)
4871 {
4872         mblk_t *mp;
4873         sadb_msg_t *samsg;
4874 
4875         mp = sadb_keysock_out(serial);
4876         if (mp == NULL)
4877                 return (NULL);
4878         mp->b_cont = allocb(sizeof (sadb_msg_t), BPRI_HI);
4879         if (mp->b_cont == NULL) {
4880                 freeb(mp);
4881                 return (NULL);
4882         }
4883 
4884         samsg = (sadb_msg_t *)mp->b_cont->b_rptr;
4885         mp->b_cont->b_wptr += sizeof (*samsg);
4886         samsg->sadb_msg_version = PF_KEY_V2;
4887         samsg->sadb_msg_type = SADB_ACQUIRE;
4888         samsg->sadb_msg_errno = 0;
4889         samsg->sadb_msg_reserved = 0;
4890         samsg->sadb_msg_satype = satype;
4891         samsg->sadb_msg_seq = seq;
4892         samsg->sadb_msg_pid = pid;
4893 
4894         return (mp);
4895 }
4896 
4897 /*
4898  * Generate address and TX/MLS sensitivity label PF_KEY extensions that are
4899  * common to both regular and extended ACQUIREs.
4900  */
4901 static mblk_t *
4902 sadb_acquire_msg_common(ipsec_selector_t *sel, ipsec_policy_t *pp,
4903     ipsec_action_t *ap, boolean_t tunnel_mode, ts_label_t *tsl,
4904     sadb_sens_t *sens)
4905 {
4906         size_t len;
4907         mblk_t *mp;
4908         uint8_t *start, *cur, *end;
4909         uint32_t *saddrptr, *daddrptr;
4910         sa_family_t af;
4911         ipsec_action_t *oldap;
4912         ipsec_selkey_t *ipsl;
4913         uint8_t proto, pfxlen;
4914         uint16_t lport, rport;
4915         int senslen = 0;
4916 
4917         /*
4918          * Get action pointer set if it isn't already.
4919          */
4920         oldap = ap;
4921         if (pp != NULL) {
4922                 ap = pp->ipsp_act;
4923                 if (ap == NULL)
4924                         ap = oldap;
4925         }
4926 
4927         /*
4928          * Biggest-case scenario:
4929          * 4x (sadb_address_t + struct sockaddr_in6)
4930          *      (src, dst, isrc, idst)
4931          *      (COMING SOON, 6x, because of triggering-packet contents.)
4932          * sadb_x_kmc_t
4933          * sadb_sens_t
4934          * And wiggle room for label bitvectors.  Luckily there are
4935          * programmatic ways to find it.
4936          */
4937         len = 4 * (sizeof (sadb_address_t) + sizeof (struct sockaddr_in6));
4938 
4939         /* Figure out full and proper length of sensitivity labels. */
4940         if (sens != NULL) {
4941                 ASSERT(tsl == NULL);
4942                 senslen = SADB_64TO8(sens->sadb_sens_len);
4943         } else if (tsl != NULL) {
4944                 senslen = sadb_sens_len_from_label(tsl);
4945         }
4946 #ifdef DEBUG
4947         else {
4948                 ASSERT(senslen == 0);
4949         }
4950 #endif /* DEBUG */
4951         len += senslen;
4952 
4953         mp = allocb(len, BPRI_HI);
4954         if (mp == NULL)
4955                 return (NULL);
4956 
4957         start = mp->b_rptr;
4958         end = start + len;
4959         cur = start;
4960 
4961         /*
4962          * Address extensions first, from most-recently-defined to least.
4963          * (This should immediately trigger surprise or verify robustness on
4964          * older apps, like in.iked.)
4965          */
4966         if (tunnel_mode) {
4967                 /*
4968                  * Form inner address extensions based NOT on the inner
4969                  * selectors (i.e. the packet data), but on the policy's
4970                  * selector key (i.e. the policy's selector information).
4971                  *
4972                  * NOTE:  The position of IPv4 and IPv6 addresses is the
4973                  * same in ipsec_selkey_t (unless the compiler does very
4974                  * strange things with unions, consult your local C language
4975                  * lawyer for details).
4976                  */
4977                 ASSERT(pp != NULL);
4978 
4979                 ipsl = &(pp->ipsp_sel->ipsl_key);
4980                 if (ipsl->ipsl_valid & IPSL_IPV4) {
4981                         af = AF_INET;
4982                         ASSERT(sel->ips_protocol == IPPROTO_ENCAP);
4983                         ASSERT(!(ipsl->ipsl_valid & IPSL_IPV6));
4984                 } else {
4985                         af = AF_INET6;
4986                         ASSERT(sel->ips_protocol == IPPROTO_IPV6);
4987                         ASSERT(ipsl->ipsl_valid & IPSL_IPV6);
4988                 }
4989 
4990                 if (ipsl->ipsl_valid & IPSL_LOCAL_ADDR) {
4991                         saddrptr = (uint32_t *)(&ipsl->ipsl_local);
4992                         pfxlen = ipsl->ipsl_local_pfxlen;
4993                 } else {
4994                         saddrptr = (uint32_t *)(&ipv6_all_zeros);
4995                         pfxlen = 0;
4996                 }
4997                 /* XXX What about ICMP type/code? */
4998                 lport = (ipsl->ipsl_valid & IPSL_LOCAL_PORT) ?
4999                     ipsl->ipsl_lport : 0;
5000                 proto = (ipsl->ipsl_valid & IPSL_PROTOCOL) ?
5001                     ipsl->ipsl_proto : 0;
5002 
5003                 cur = sadb_make_addr_ext(cur, end, SADB_X_EXT_ADDRESS_INNER_SRC,
5004                     af, saddrptr, lport, proto, pfxlen);
5005                 if (cur == NULL) {
5006                         freeb(mp);
5007                         return (NULL);
5008                 }
5009 
5010                 if (ipsl->ipsl_valid & IPSL_REMOTE_ADDR) {
5011                         daddrptr = (uint32_t *)(&ipsl->ipsl_remote);
5012                         pfxlen = ipsl->ipsl_remote_pfxlen;
5013                 } else {
5014                         daddrptr = (uint32_t *)(&ipv6_all_zeros);
5015                         pfxlen = 0;
5016                 }
5017                 /* XXX What about ICMP type/code? */
5018                 rport = (ipsl->ipsl_valid & IPSL_REMOTE_PORT) ?
5019                     ipsl->ipsl_rport : 0;
5020 
5021                 cur = sadb_make_addr_ext(cur, end, SADB_X_EXT_ADDRESS_INNER_DST,
5022                     af, daddrptr, rport, proto, pfxlen);
5023                 if (cur == NULL) {
5024                         freeb(mp);
5025                         return (NULL);
5026                 }
5027                 /*
5028                  * TODO  - if we go to 3884's dream of transport mode IP-in-IP
5029                  * _with_ inner-packet address selectors, we'll need to further
5030                  * distinguish tunnel mode here.  For now, having inner
5031                  * addresses and/or ports is sufficient.
5032                  *
5033                  * Meanwhile, whack proto/ports to reflect IP-in-IP for the
5034                  * outer addresses.
5035                  */
5036                 proto = sel->ips_protocol;   /* Either _ENCAP or _IPV6 */
5037                 lport = rport = 0;
5038         } else if ((ap != NULL) && (!ap->ipa_want_unique)) {
5039                 /*
5040                  * For cases when the policy calls out specific ports (or not).
5041                  */
5042                 proto = 0;
5043                 lport = 0;
5044                 rport = 0;
5045                 if (pp != NULL) {
5046                         ipsl = &(pp->ipsp_sel->ipsl_key);
5047                         if (ipsl->ipsl_valid & IPSL_PROTOCOL)
5048                                 proto = ipsl->ipsl_proto;
5049                         if (ipsl->ipsl_valid & IPSL_REMOTE_PORT)
5050                                 rport = ipsl->ipsl_rport;
5051                         if (ipsl->ipsl_valid & IPSL_LOCAL_PORT)
5052                                 lport = ipsl->ipsl_lport;
5053                 }
5054         } else {
5055                 /*
5056                  * For require-unique-SA policies.
5057                  */
5058                 proto = sel->ips_protocol;
5059                 lport = sel->ips_local_port;
5060                 rport = sel->ips_remote_port;
5061         }
5062 
5063         /*
5064          * Regular addresses.  These are outer-packet ones for tunnel mode.
5065          * Or for transport mode, the regulard address & port information.
5066          */
5067         af = sel->ips_isv4 ? AF_INET : AF_INET6;
5068 
5069         /*
5070          * NOTE:  The position of IPv4 and IPv6 addresses is the same in
5071          * ipsec_selector_t.
5072          */
5073         cur = sadb_make_addr_ext(cur, end, SADB_EXT_ADDRESS_SRC, af,
5074             (uint32_t *)(&sel->ips_local_addr_v6), lport, proto, 0);
5075         if (cur == NULL) {
5076                 freeb(mp);
5077                 return (NULL);
5078         }
5079 
5080         cur = sadb_make_addr_ext(cur, end, SADB_EXT_ADDRESS_DST, af,
5081             (uint32_t *)(&sel->ips_remote_addr_v6), rport, proto, 0);
5082         if (cur == NULL) {
5083                 freeb(mp);
5084                 return (NULL);
5085         }
5086 
5087         /*
5088          * If present, generate a sensitivity label.
5089          */
5090         if (cur + senslen > end) {
5091                 freeb(mp);
5092                 return (NULL);
5093         }
5094         if (sens != NULL) {
5095                 /* Explicit sadb_sens_t, usually from inverse-ACQUIRE. */
5096                 bcopy(sens, cur, senslen);
5097         } else if (tsl != NULL) {
5098                 /* Generate sadb_sens_t from ACQUIRE source. */
5099                 sadb_sens_from_label((sadb_sens_t *)cur, SADB_EXT_SENSITIVITY,
5100                     tsl, senslen);
5101         }
5102 #ifdef DEBUG
5103         else {
5104                 ASSERT(senslen == 0);
5105         }
5106 #endif /* DEBUG */
5107         cur += senslen;
5108         mp->b_wptr = cur;
5109 
5110         return (mp);
5111 }
5112 
5113 /*
5114  * Generate a regular ACQUIRE's proposal extension and KMC information..
5115  */
5116 static mblk_t *
5117 sadb_acquire_prop(ipsec_action_t *ap, netstack_t *ns, boolean_t do_esp)
5118 {
5119         ipsec_stack_t *ipss = ns->netstack_ipsec;
5120         ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
5121         ipsecah_stack_t *ahstack = ns->netstack_ipsecah;
5122         mblk_t *mp = NULL;
5123         sadb_prop_t *prop;
5124         sadb_comb_t *comb;
5125         ipsec_action_t *walker;
5126         int ncombs, allocsize, ealgid, aalgid, aminbits, amaxbits, eminbits,
5127             emaxbits, replay;
5128         uint64_t softbytes, hardbytes, softaddtime, hardaddtime, softusetime,
5129             hardusetime;
5130         uint64_t kmc = 0;
5131         uint32_t kmp = 0;
5132 
5133         /*
5134          * Since it's an rwlock read, AND writing to the IPsec algorithms is
5135          * rare, just acquire it once up top, and drop it upon return.
5136          */
5137         rw_enter(&ipss->ipsec_alg_lock, RW_READER);
5138         if (do_esp) {
5139                 uint64_t num_aalgs, num_ealgs;
5140 
5141                 if (espstack->esp_kstats == NULL)
5142                         goto bail;
5143 
5144                 num_aalgs = ipss->ipsec_nalgs[IPSEC_ALG_AUTH];
5145                 num_ealgs = ipss->ipsec_nalgs[IPSEC_ALG_ENCR];
5146                 if (num_ealgs == 0)
5147                         goto bail;      /* IPsec not loaded yet, apparently. */
5148                 num_aalgs++;    /* No-auth or self-auth-crypto ESP. */
5149 
5150                 /* Use netstack's maximum loaded algorithms... */
5151                 ncombs = num_ealgs * num_aalgs;
5152                 replay =  espstack->ipsecesp_replay_size;
5153         } else {
5154                 if (ahstack->ah_kstats == NULL)
5155                         goto bail;
5156 
5157                 ncombs = ipss->ipsec_nalgs[IPSEC_ALG_AUTH];
5158 
5159                 if (ncombs == 0)
5160                         goto bail;      /* IPsec not loaded yet, apparently. */
5161                 replay =  ahstack->ipsecah_replay_size;
5162         }
5163 
5164         allocsize = sizeof (*prop) + ncombs * sizeof (*comb) +
5165             sizeof (sadb_x_kmc_t);
5166         mp = allocb(allocsize, BPRI_HI);
5167         if (mp == NULL)
5168                 goto bail;
5169         prop = (sadb_prop_t *)mp->b_rptr;
5170         mp->b_wptr += sizeof (*prop);
5171         comb = (sadb_comb_t *)mp->b_wptr;
5172         /* Decrement allocsize, if it goes to or below 0, stop. */
5173         allocsize -= sizeof (*prop);
5174         prop->sadb_prop_exttype = SADB_EXT_PROPOSAL;
5175         prop->sadb_prop_len = SADB_8TO64(sizeof (*prop));
5176         *(uint32_t *)(&prop->sadb_prop_replay) = 0;      /* Quick zero-out! */
5177         prop->sadb_prop_replay = replay;
5178 
5179         /*
5180          * Based upon algorithm properties, and what-not, prioritize a
5181          * proposal, based on the ordering of the ESP algorithms in the
5182          * alternatives in the policy rule or socket that was placed
5183          * in the acquire record.
5184          *
5185          * For each action in policy list
5186          *   Add combination.
5187          *   I should not hit it, but if I've hit limit, return.
5188          */
5189 
5190         for (walker = ap; walker != NULL; walker = walker->ipa_next) {
5191                 ipsec_alginfo_t *ealg, *aalg;
5192                 ipsec_prot_t *prot;
5193 
5194                 if (walker->ipa_act.ipa_type != IPSEC_POLICY_APPLY)
5195                         continue;
5196 
5197                 prot = &walker->ipa_act.ipa_apply;
5198                 if (walker->ipa_act.ipa_apply.ipp_km_proto != 0)
5199                         kmp = walker->ipa_act.ipa_apply.ipp_km_proto;
5200                 if (walker->ipa_act.ipa_apply.ipp_km_cookie != 0)
5201                         kmc = walker->ipa_act.ipa_apply.ipp_km_cookie;
5202                 if (walker->ipa_act.ipa_apply.ipp_replay_depth) {
5203                         prop->sadb_prop_replay =
5204                             walker->ipa_act.ipa_apply.ipp_replay_depth;
5205                 }
5206 
5207                 if (do_esp) {
5208                         if (!prot->ipp_use_esp)
5209                                 continue;
5210 
5211                         if (prot->ipp_esp_auth_alg != 0) {
5212                                 aalg = ipss->ipsec_alglists[IPSEC_ALG_AUTH]
5213                                     [prot->ipp_esp_auth_alg];
5214                                 if (aalg == NULL || !ALG_VALID(aalg))
5215                                         continue;
5216                         } else
5217                                 aalg = NULL;
5218 
5219                         ASSERT(prot->ipp_encr_alg > 0);
5220                         ealg = ipss->ipsec_alglists[IPSEC_ALG_ENCR]
5221                             [prot->ipp_encr_alg];
5222                         if (ealg == NULL || !ALG_VALID(ealg))
5223                                 continue;
5224 
5225                         /*
5226                          * These may want to come from policy rule..
5227                          */
5228                         softbytes = espstack->ipsecesp_default_soft_bytes;
5229                         hardbytes = espstack->ipsecesp_default_hard_bytes;
5230                         softaddtime = espstack->ipsecesp_default_soft_addtime;
5231                         hardaddtime = espstack->ipsecesp_default_hard_addtime;
5232                         softusetime = espstack->ipsecesp_default_soft_usetime;
5233                         hardusetime = espstack->ipsecesp_default_hard_usetime;
5234                 } else {
5235                         if (!prot->ipp_use_ah)
5236                                 continue;
5237                         ealg = NULL;
5238                         aalg = ipss->ipsec_alglists[IPSEC_ALG_AUTH]
5239                             [prot->ipp_auth_alg];
5240                         if (aalg == NULL || !ALG_VALID(aalg))
5241                                 continue;
5242 
5243                         /*
5244                          * These may want to come from policy rule..
5245                          */
5246                         softbytes = ahstack->ipsecah_default_soft_bytes;
5247                         hardbytes = ahstack->ipsecah_default_hard_bytes;
5248                         softaddtime = ahstack->ipsecah_default_soft_addtime;
5249                         hardaddtime = ahstack->ipsecah_default_hard_addtime;
5250                         softusetime = ahstack->ipsecah_default_soft_usetime;
5251                         hardusetime = ahstack->ipsecah_default_hard_usetime;
5252                 }
5253 
5254                 if (ealg == NULL) {
5255                         ealgid = eminbits = emaxbits = 0;
5256                 } else {
5257                         ealgid = ealg->alg_id;
5258                         eminbits =
5259                             MAX(prot->ipp_espe_minbits, ealg->alg_ef_minbits);
5260                         emaxbits =
5261                             MIN(prot->ipp_espe_maxbits, ealg->alg_ef_maxbits);
5262                 }
5263 
5264                 if (aalg == NULL) {
5265                         aalgid = aminbits = amaxbits = 0;
5266                 } else {
5267                         aalgid = aalg->alg_id;
5268                         aminbits = MAX(prot->ipp_espa_minbits,
5269                             aalg->alg_ef_minbits);
5270                         amaxbits = MIN(prot->ipp_espa_maxbits,
5271                             aalg->alg_ef_maxbits);
5272                 }
5273 
5274                 comb->sadb_comb_flags = 0;
5275                 comb->sadb_comb_reserved = 0;
5276                 comb->sadb_comb_encrypt = ealgid;
5277                 comb->sadb_comb_encrypt_minbits = eminbits;
5278                 comb->sadb_comb_encrypt_maxbits = emaxbits;
5279                 comb->sadb_comb_auth = aalgid;
5280                 comb->sadb_comb_auth_minbits = aminbits;
5281                 comb->sadb_comb_auth_maxbits = amaxbits;
5282                 comb->sadb_comb_soft_allocations = 0;
5283                 comb->sadb_comb_hard_allocations = 0;
5284                 comb->sadb_comb_soft_bytes = softbytes;
5285                 comb->sadb_comb_hard_bytes = hardbytes;
5286                 comb->sadb_comb_soft_addtime = softaddtime;
5287                 comb->sadb_comb_hard_addtime = hardaddtime;
5288                 comb->sadb_comb_soft_usetime = softusetime;
5289                 comb->sadb_comb_hard_usetime = hardusetime;
5290 
5291                 prop->sadb_prop_len += SADB_8TO64(sizeof (*comb));
5292                 mp->b_wptr += sizeof (*comb);
5293                 allocsize -= sizeof (*comb);
5294                 /* Should never dip BELOW sizeof (KM cookie extension). */
5295                 ASSERT3S(allocsize, >=, sizeof (sadb_x_kmc_t));
5296                 if (allocsize <= sizeof (sadb_x_kmc_t))
5297                         break;  /* out of space.. */
5298                 comb++;
5299         }
5300 
5301         /* Don't include KMC extension if there's no room. */
5302         if (((kmp != 0) || (kmc != 0)) && allocsize >= sizeof (sadb_x_kmc_t)) {
5303                 if (sadb_make_kmc_ext(mp->b_wptr,
5304                     mp->b_wptr + sizeof (sadb_x_kmc_t), kmp, kmc) == NULL) {
5305                         freeb(mp);
5306                         mp = NULL;
5307                         goto bail;
5308                 }
5309                 mp->b_wptr += sizeof (sadb_x_kmc_t);
5310                 prop->sadb_prop_len += SADB_8TO64(sizeof (sadb_x_kmc_t));
5311         }
5312 
5313 bail:
5314         rw_exit(&ipss->ipsec_alg_lock);
5315         return (mp);
5316 }
5317 
5318 /*
5319  * Generate an extended ACQUIRE's extended-proposal extension.
5320  */
5321 static mblk_t *
5322 sadb_acquire_extended_prop(ipsec_action_t *ap, netstack_t *ns)
5323 {
5324         sadb_prop_t *eprop;
5325         uint8_t *cur, *end;
5326         mblk_t *mp;
5327         int allocsize, numecombs = 0, numalgdescs = 0;
5328         uint32_t kmp = 0, replay = 0;
5329         uint64_t kmc = 0;
5330         ipsec_action_t *walker;
5331 
5332         allocsize = sizeof (*eprop);
5333 
5334         /*
5335          * Going to walk through the action list twice.  Once for allocation
5336          * measurement, and once for actual construction.
5337          */
5338         for (walker = ap; walker != NULL; walker = walker->ipa_next) {
5339                 ipsec_prot_t *ipp;
5340 
5341                 /*
5342                  * Skip non-IPsec policies
5343                  */
5344                 if (walker->ipa_act.ipa_type != IPSEC_ACT_APPLY)
5345                         continue;
5346 
5347                 ipp = &walker->ipa_act.ipa_apply;
5348 
5349                 if (walker->ipa_act.ipa_apply.ipp_km_proto)
5350                         kmp = ipp->ipp_km_proto;
5351                 if (walker->ipa_act.ipa_apply.ipp_km_cookie)
5352                         kmc = ipp->ipp_km_cookie;
5353                 if (walker->ipa_act.ipa_apply.ipp_replay_depth)
5354                         replay = ipp->ipp_replay_depth;
5355 
5356                 if (ipp->ipp_use_ah)
5357                         numalgdescs++;
5358                 if (ipp->ipp_use_esp) {
5359                         numalgdescs++;
5360                         if (ipp->ipp_use_espa)
5361                                 numalgdescs++;
5362                 }
5363 
5364                 numecombs++;
5365         }
5366         ASSERT(numecombs > 0);
5367 
5368         allocsize += numecombs * sizeof (sadb_x_ecomb_t) +
5369             numalgdescs * sizeof (sadb_x_algdesc_t) + sizeof (sadb_x_kmc_t);
5370         mp = allocb(allocsize, BPRI_HI);
5371         if (mp == NULL)
5372                 return (NULL);
5373         eprop = (sadb_prop_t *)mp->b_rptr;
5374         end = mp->b_rptr + allocsize;
5375         cur = mp->b_rptr + sizeof (*eprop);
5376 
5377         eprop->sadb_prop_exttype = SADB_X_EXT_EPROP;
5378         eprop->sadb_x_prop_ereserved = 0;
5379         eprop->sadb_x_prop_numecombs = 0;
5380         *(uint32_t *)(&eprop->sadb_prop_replay) = 0;     /* Quick zero-out! */
5381         /* Pick ESP's replay default if need be. */
5382         eprop->sadb_prop_replay = (replay == 0) ?
5383             ns->netstack_ipsecesp->ipsecesp_replay_size : replay;
5384 
5385         /* This time, walk through and actually allocate. */
5386         for (walker = ap; walker != NULL; walker = walker->ipa_next) {
5387                 /*
5388                  * Skip non-IPsec policies
5389                  */
5390                 if (walker->ipa_act.ipa_type != IPSEC_ACT_APPLY)
5391                         continue;
5392                 cur = sadb_action_to_ecomb(cur, end, walker, ns);
5393                 if (cur == NULL) {
5394                         /* NOTE: inverse-ACQUIRE should note this as ENOMEM. */
5395                         freeb(mp);
5396                         return (NULL);
5397                 }
5398                 eprop->sadb_x_prop_numecombs++;
5399         }
5400 
5401         ASSERT(end - cur >= sizeof (sadb_x_kmc_t));
5402         if ((kmp != 0) || (kmc != 0)) {
5403                 cur = sadb_make_kmc_ext(cur, end, kmp, kmc);
5404                 if (cur == NULL) {
5405                         freeb(mp);
5406                         return (NULL);
5407                 }
5408         }
5409         mp->b_wptr = cur;
5410         eprop->sadb_prop_len = SADB_8TO64(cur - mp->b_rptr);
5411 
5412         return (mp);
5413 }
5414 
5415 /*
5416  * For this mblk, insert a new acquire record.  Assume bucket contains addrs
5417  * of all of the same length.  Give up (and drop) if memory
5418  * cannot be allocated for a new one; otherwise, invoke callback to
5419  * send the acquire up..
5420  *
5421  * In cases where we need both AH and ESP, add the SA to the ESP ACQUIRE
5422  * list.  The ah_add_sa_finish() routines can look at the packet's attached
5423  * attributes and handle this case specially.
5424  */
5425 void
5426 sadb_acquire(mblk_t *datamp, ip_xmit_attr_t *ixa, boolean_t need_ah,
5427     boolean_t need_esp)
5428 {
5429         mblk_t  *asyncmp, *regular, *extended, *common, *prop, *eprop;
5430         sadbp_t *spp;
5431         sadb_t *sp;
5432         ipsacq_t *newbie;
5433         iacqf_t *bucket;
5434         ipha_t *ipha = (ipha_t *)datamp->b_rptr;
5435         ip6_t *ip6h = (ip6_t *)datamp->b_rptr;
5436         uint32_t *src, *dst, *isrc, *idst;
5437         ipsec_policy_t *pp = ixa->ixa_ipsec_policy;
5438         ipsec_action_t *ap = ixa->ixa_ipsec_action;
5439         sa_family_t af;
5440         int hashoffset;
5441         uint32_t seq;
5442         uint64_t unique_id = 0;
5443         boolean_t tunnel_mode = (ixa->ixa_flags & IXAF_IPSEC_TUNNEL) != 0;
5444         ts_label_t      *tsl;
5445         netstack_t      *ns = ixa->ixa_ipst->ips_netstack;
5446         ipsec_stack_t   *ipss = ns->netstack_ipsec;
5447         ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
5448         ipsecah_stack_t *ahstack = ns->netstack_ipsecah;
5449         ipsec_selector_t sel;
5450         queue_t *q;
5451 
5452         ASSERT((pp != NULL) || (ap != NULL));
5453 
5454         ASSERT(need_ah || need_esp);
5455 
5456         /* Assign sadb pointers */
5457         if (need_esp) {
5458                 /*
5459                  * ESP happens first if we need both AH and ESP.
5460                  */
5461                 spp = &espstack->esp_sadb;
5462         } else {
5463                 spp = &ahstack->ah_sadb;
5464         }
5465         sp = (ixa->ixa_flags & IXAF_IS_IPV4) ? &spp->s_v4 : &spp->s_v6;
5466 
5467         if (is_system_labeled())
5468                 tsl = ixa->ixa_tsl;
5469         else
5470                 tsl = NULL;
5471 
5472         if (ap == NULL)
5473                 ap = pp->ipsp_act;
5474         ASSERT(ap != NULL);
5475 
5476         if (ap->ipa_act.ipa_apply.ipp_use_unique || tunnel_mode)
5477                 unique_id = SA_FORM_UNIQUE_ID(ixa);
5478 
5479         /*
5480          * Set up an ACQUIRE record.
5481          *
5482          * Immediately, make sure the ACQUIRE sequence number doesn't slip
5483          * below the lowest point allowed in the kernel.  (In other words,
5484          * make sure the high bit on the sequence number is set.)
5485          */
5486 
5487         seq = keysock_next_seq(ns) | IACQF_LOWEST_SEQ;
5488 
5489         if (IPH_HDR_VERSION(ipha) == IP_VERSION) {
5490                 src = (uint32_t *)&ipha->ipha_src;
5491                 dst = (uint32_t *)&ipha->ipha_dst;
5492                 af = AF_INET;
5493                 hashoffset = OUTBOUND_HASH_V4(sp, ipha->ipha_dst);
5494                 ASSERT(ixa->ixa_flags & IXAF_IS_IPV4);
5495         } else {
5496                 ASSERT(IPH_HDR_VERSION(ipha) == IPV6_VERSION);
5497                 src = (uint32_t *)&ip6h->ip6_src;
5498                 dst = (uint32_t *)&ip6h->ip6_dst;
5499                 af = AF_INET6;
5500                 hashoffset = OUTBOUND_HASH_V6(sp, ip6h->ip6_dst);
5501                 ASSERT(!(ixa->ixa_flags & IXAF_IS_IPV4));
5502         }
5503 
5504         if (tunnel_mode) {
5505                 if (pp == NULL) {
5506                         /*
5507                          * Tunnel mode with no policy pointer means this is a
5508                          * reflected ICMP (like a ECHO REQUEST) that came in
5509                          * with self-encapsulated protection.  Until we better
5510                          * support this, drop the packet.
5511                          */
5512                         ip_drop_packet(datamp, B_FALSE, NULL,
5513                             DROPPER(ipss, ipds_spd_got_selfencap),
5514                             &ipss->ipsec_spd_dropper);
5515                         return;
5516                 }
5517                 /* Snag inner addresses. */
5518                 isrc = ixa->ixa_ipsec_insrc;
5519                 idst = ixa->ixa_ipsec_indst;
5520         } else {
5521                 isrc = idst = NULL;
5522         }
5523 
5524         /*
5525          * Check buckets to see if there is an existing entry.  If so,
5526          * grab it.  sadb_checkacquire locks newbie if found.
5527          */
5528         bucket = &(sp->sdb_acq[hashoffset]);
5529         mutex_enter(&bucket->iacqf_lock);
5530         newbie = sadb_checkacquire(bucket, ap, pp, src, dst, isrc, idst,
5531             unique_id, tsl);
5532 
5533         if (newbie == NULL) {
5534                 /*
5535                  * Otherwise, allocate a new one.
5536                  */
5537                 newbie = kmem_zalloc(sizeof (*newbie), KM_NOSLEEP);
5538                 if (newbie == NULL) {
5539                         mutex_exit(&bucket->iacqf_lock);
5540                         ip_drop_packet(datamp, B_FALSE, NULL,
5541                             DROPPER(ipss, ipds_sadb_acquire_nomem),
5542                             &ipss->ipsec_sadb_dropper);
5543                         return;
5544                 }
5545                 newbie->ipsacq_policy = pp;
5546                 if (pp != NULL) {
5547                         IPPOL_REFHOLD(pp);
5548                 }
5549                 IPACT_REFHOLD(ap);
5550                 newbie->ipsacq_act = ap;
5551                 newbie->ipsacq_linklock = &bucket->iacqf_lock;
5552                 newbie->ipsacq_next = bucket->iacqf_ipsacq;
5553                 newbie->ipsacq_ptpn = &bucket->iacqf_ipsacq;
5554                 if (newbie->ipsacq_next != NULL)
5555                         newbie->ipsacq_next->ipsacq_ptpn = &newbie->ipsacq_next;
5556 
5557                 bucket->iacqf_ipsacq = newbie;
5558                 mutex_init(&newbie->ipsacq_lock, NULL, MUTEX_DEFAULT, NULL);
5559                 mutex_enter(&newbie->ipsacq_lock);
5560         }
5561 
5562         /*
5563          * XXX MLS does it actually help us to drop the bucket lock here?
5564          * we have inserted a half-built, locked acquire record into the
5565          * bucket.  any competing thread will now be able to lock the bucket
5566          * to scan it, but will immediately pile up on the new acquire
5567          * record's lock; I don't think we gain anything here other than to
5568          * disperse blame for lock contention.
5569          *
5570          * we might be able to dispense with acquire record locks entirely..
5571          * just use the bucket locks..
5572          */
5573 
5574         mutex_exit(&bucket->iacqf_lock);
5575 
5576         /*
5577          * This assert looks silly for now, but we may need to enter newbie's
5578          * mutex during a search.
5579          */
5580         ASSERT(MUTEX_HELD(&newbie->ipsacq_lock));
5581 
5582         /*
5583          * Make the ip_xmit_attr_t into something we can queue.
5584          * If no memory it frees datamp.
5585          */
5586         asyncmp = ip_xmit_attr_to_mblk(ixa);
5587         if (asyncmp != NULL)
5588                 linkb(asyncmp, datamp);
5589 
5590         /* Queue up packet.  Use b_next. */
5591 
5592         if (asyncmp == NULL) {
5593                 /* Statistics for allocation failure */
5594                 if (ixa->ixa_flags & IXAF_IS_IPV4) {
5595                         BUMP_MIB(&ixa->ixa_ipst->ips_ip_mib,
5596                             ipIfStatsOutDiscards);
5597                 } else {
5598                         BUMP_MIB(&ixa->ixa_ipst->ips_ip6_mib,
5599                             ipIfStatsOutDiscards);
5600                 }
5601                 ip_drop_output("No memory for asyncmp", datamp, NULL);
5602                 freemsg(datamp);
5603                 /*
5604                  * The acquire record will be freed quickly if it's new
5605                  * (ipsacq_expire == 0), and will proceed as if no packet
5606                  * showed up if not.
5607                  */
5608                 mutex_exit(&newbie->ipsacq_lock);
5609                 return;
5610         } else if (newbie->ipsacq_numpackets == 0) {
5611                 /* First one. */
5612                 newbie->ipsacq_mp = asyncmp;
5613                 newbie->ipsacq_numpackets = 1;
5614                 newbie->ipsacq_expire = gethrestime_sec();
5615                 /*
5616                  * Extended ACQUIRE with both AH+ESP will use ESP's timeout
5617                  * value.
5618                  */
5619                 newbie->ipsacq_expire += *spp->s_acquire_timeout;
5620                 newbie->ipsacq_seq = seq;
5621                 newbie->ipsacq_addrfam = af;
5622 
5623                 newbie->ipsacq_srcport = ixa->ixa_ipsec_src_port;
5624                 newbie->ipsacq_dstport = ixa->ixa_ipsec_dst_port;
5625                 newbie->ipsacq_icmp_type = ixa->ixa_ipsec_icmp_type;
5626                 newbie->ipsacq_icmp_code = ixa->ixa_ipsec_icmp_code;
5627                 if (tunnel_mode) {
5628                         newbie->ipsacq_inneraddrfam = ixa->ixa_ipsec_inaf;
5629                         newbie->ipsacq_proto = ixa->ixa_ipsec_inaf == AF_INET6 ?
5630                             IPPROTO_IPV6 : IPPROTO_ENCAP;
5631                         newbie->ipsacq_innersrcpfx = ixa->ixa_ipsec_insrcpfx;
5632                         newbie->ipsacq_innerdstpfx = ixa->ixa_ipsec_indstpfx;
5633                         IPSA_COPY_ADDR(newbie->ipsacq_innersrc,
5634                             ixa->ixa_ipsec_insrc, ixa->ixa_ipsec_inaf);
5635                         IPSA_COPY_ADDR(newbie->ipsacq_innerdst,
5636                             ixa->ixa_ipsec_indst, ixa->ixa_ipsec_inaf);
5637                 } else {
5638                         newbie->ipsacq_proto = ixa->ixa_ipsec_proto;
5639                 }
5640                 newbie->ipsacq_unique_id = unique_id;
5641 
5642                 if (tsl != NULL) {
5643                         label_hold(tsl);
5644                         newbie->ipsacq_tsl = tsl;
5645                 }
5646         } else {
5647                 /* Scan to the end of the list & insert. */
5648                 mblk_t *lastone = newbie->ipsacq_mp;
5649 
5650                 while (lastone->b_next != NULL)
5651                         lastone = lastone->b_next;
5652                 lastone->b_next = asyncmp;
5653                 if (newbie->ipsacq_numpackets++ == ipsacq_maxpackets) {
5654                         newbie->ipsacq_numpackets = ipsacq_maxpackets;
5655                         lastone = newbie->ipsacq_mp;
5656                         newbie->ipsacq_mp = lastone->b_next;
5657                         lastone->b_next = NULL;
5658 
5659                         /* Freeing the async message */
5660                         lastone = ip_xmit_attr_free_mblk(lastone);
5661                         ip_drop_packet(lastone, B_FALSE, NULL,
5662                             DROPPER(ipss, ipds_sadb_acquire_toofull),
5663                             &ipss->ipsec_sadb_dropper);
5664                 } else {
5665                         IP_ACQUIRE_STAT(ipss, qhiwater,
5666                             newbie->ipsacq_numpackets);
5667                 }
5668         }
5669 
5670         /*
5671          * Reset addresses.  Set them to the most recently added mblk chain,
5672          * so that the address pointers in the acquire record will point
5673          * at an mblk still attached to the acquire list.
5674          */
5675 
5676         newbie->ipsacq_srcaddr = src;
5677         newbie->ipsacq_dstaddr = dst;
5678 
5679         /*
5680          * If the acquire record has more than one queued packet, we've
5681          * already sent an ACQUIRE, and don't need to repeat ourself.
5682          */
5683         if (newbie->ipsacq_seq != seq || newbie->ipsacq_numpackets > 1) {
5684                 /* I have an acquire outstanding already! */
5685                 mutex_exit(&newbie->ipsacq_lock);
5686                 return;
5687         }
5688 
5689         if (need_esp) {
5690                 ESP_BUMP_STAT(espstack, acquire_requests);
5691                 q = espstack->esp_pfkey_q;
5692         } else {
5693                 /*
5694                  * Two cases get us here:
5695                  * 1.) AH-only policy.
5696                  *
5697                  * 2.) A continuation of an AH+ESP policy, and this is the
5698                  * post-ESP, AH-needs-to-send-a-regular-ACQUIRE case.
5699                  * (i.e. called from esp_do_outbound_ah().)
5700                  */
5701                 AH_BUMP_STAT(ahstack, acquire_requests);
5702                 q = ahstack->ah_pfkey_q;
5703         }
5704 
5705         /*
5706          * Get selectors and other policy-expression bits needed for an
5707          * ACQUIRE.
5708          */
5709         bzero(&sel, sizeof (sel));
5710         sel.ips_isv4 = (ixa->ixa_flags & IXAF_IS_IPV4) != 0;
5711         if (tunnel_mode) {
5712                 sel.ips_protocol = (ixa->ixa_ipsec_inaf == AF_INET) ?
5713                     IPPROTO_ENCAP : IPPROTO_IPV6;
5714         } else {
5715                 sel.ips_protocol = ixa->ixa_ipsec_proto;
5716                 sel.ips_local_port = ixa->ixa_ipsec_src_port;
5717                 sel.ips_remote_port = ixa->ixa_ipsec_dst_port;
5718         }
5719         sel.ips_icmp_type = ixa->ixa_ipsec_icmp_type;
5720         sel.ips_icmp_code = ixa->ixa_ipsec_icmp_code;
5721         sel.ips_is_icmp_inv_acq = 0;
5722         if (af == AF_INET) {
5723                 sel.ips_local_addr_v4 = ipha->ipha_src;
5724                 sel.ips_remote_addr_v4 = ipha->ipha_dst;
5725         } else {
5726                 sel.ips_local_addr_v6 = ip6h->ip6_src;
5727                 sel.ips_remote_addr_v6 = ip6h->ip6_dst;
5728         }
5729 
5730 
5731         /*
5732          * 1. Generate addresses, kmc, and sensitivity.  These are "common"
5733          * and should be an mblk pointed to by common. TBD -- eventually it
5734          * will include triggering packet contents as more address extensions.
5735          *
5736          * 2. Generate ACQUIRE & KEYSOCK_OUT and single-protocol proposal.
5737          * These are "regular" and "prop".  String regular->b_cont->b_cont =
5738          * common, common->b_cont = prop.
5739          *
5740          * 3. If extended register got turned on, generate EXT_ACQUIRE &
5741          * KEYSOCK_OUT and multi-protocol eprop. These are "extended" and
5742          * "eprop".  String extended->b_cont->b_cont = dupb(common) and
5743          * extended->b_cont->b_cont->b_cont = prop.
5744          *
5745          * 4. Deliver:  putnext(q, regular) and if there, putnext(q, extended).
5746          */
5747 
5748         regular = extended = prop = eprop = NULL;
5749 
5750         common = sadb_acquire_msg_common(&sel, pp, ap, tunnel_mode, tsl, NULL);
5751         if (common == NULL)
5752                 goto bail;
5753 
5754         regular = sadb_acquire_msg_base(0, (need_esp ?
5755             SADB_SATYPE_ESP : SADB_SATYPE_AH), newbie->ipsacq_seq, 0);
5756         if (regular == NULL)
5757                 goto bail;
5758 
5759         /*
5760          * Pardon the boolean cleverness. At least one of need_* must be true.
5761          * If they are equal, it's an AH & ESP policy and ESP needs to go
5762          * first.  If they aren't, just check the contents of need_esp.
5763          */
5764         prop = sadb_acquire_prop(ap, ns, need_esp);
5765         if (prop == NULL)
5766                 goto bail;
5767 
5768         /* Link the parts together. */
5769         regular->b_cont->b_cont = common;
5770         common->b_cont = prop;
5771         /*
5772          * Prop is now linked, so don't freemsg() it if the extended
5773          * construction goes off the rails.
5774          */
5775         prop = NULL;
5776 
5777         ((sadb_msg_t *)(regular->b_cont->b_rptr))->sadb_msg_len =
5778             SADB_8TO64(msgsize(regular->b_cont));
5779 
5780         /*
5781          * If we need an extended ACQUIRE, build it here.
5782          */
5783         if (keysock_extended_reg(ns)) {
5784                 /* NOTE: "common" still points to what we need. */
5785                 extended = sadb_acquire_msg_base(0, 0, newbie->ipsacq_seq, 0);
5786                 if (extended == NULL) {
5787                         common = NULL;
5788                         goto bail;
5789                 }
5790 
5791                 extended->b_cont->b_cont = dupb(common);
5792                 common = NULL;
5793                 if (extended->b_cont->b_cont == NULL)
5794                         goto bail;
5795 
5796                 eprop = sadb_acquire_extended_prop(ap, ns);
5797                 if (eprop == NULL)
5798                         goto bail;
5799                 extended->b_cont->b_cont->b_cont = eprop;
5800 
5801                 ((sadb_msg_t *)(extended->b_cont->b_rptr))->sadb_msg_len =
5802                     SADB_8TO64(msgsize(extended->b_cont));
5803         }
5804 
5805         /* So we don't hold a lock across putnext()... */
5806         mutex_exit(&newbie->ipsacq_lock);
5807 
5808         if (extended != NULL)
5809                 putnext(q, extended);
5810         ASSERT(regular != NULL);
5811         putnext(q, regular);
5812         return;
5813 
5814 bail:
5815         /* Make this acquire record go away quickly... */
5816         newbie->ipsacq_expire = 0;
5817         /* Exploit freemsg(NULL) being legal for fun & profit. */
5818         freemsg(common);
5819         freemsg(prop);
5820         freemsg(extended);
5821         freemsg(regular);
5822         mutex_exit(&newbie->ipsacq_lock);
5823 }
5824 
5825 /*
5826  * Unlink and free an acquire record.
5827  */
5828 void
5829 sadb_destroy_acquire(ipsacq_t *acqrec, netstack_t *ns)
5830 {
5831         mblk_t          *mp;
5832         ipsec_stack_t   *ipss = ns->netstack_ipsec;
5833 
5834         ASSERT(MUTEX_HELD(acqrec->ipsacq_linklock));
5835 
5836         if (acqrec->ipsacq_policy != NULL) {
5837                 IPPOL_REFRELE(acqrec->ipsacq_policy);
5838         }
5839         if (acqrec->ipsacq_act != NULL) {
5840                 IPACT_REFRELE(acqrec->ipsacq_act);
5841         }
5842 
5843         /* Unlink */
5844         *(acqrec->ipsacq_ptpn) = acqrec->ipsacq_next;
5845         if (acqrec->ipsacq_next != NULL)
5846                 acqrec->ipsacq_next->ipsacq_ptpn = acqrec->ipsacq_ptpn;
5847 
5848         if (acqrec->ipsacq_tsl != NULL) {
5849                 label_rele(acqrec->ipsacq_tsl);
5850                 acqrec->ipsacq_tsl = NULL;
5851         }
5852 
5853         /*
5854          * Free hanging mp's.
5855          *
5856          * XXX Instead of freemsg(), perhaps use IPSEC_REQ_FAILED.
5857          */
5858 
5859         mutex_enter(&acqrec->ipsacq_lock);
5860         while (acqrec->ipsacq_mp != NULL) {
5861                 mp = acqrec->ipsacq_mp;
5862                 acqrec->ipsacq_mp = mp->b_next;
5863                 mp->b_next = NULL;
5864                 /* Freeing the async message */
5865                 mp = ip_xmit_attr_free_mblk(mp);
5866                 ip_drop_packet(mp, B_FALSE, NULL,
5867                     DROPPER(ipss, ipds_sadb_acquire_timeout),
5868                     &ipss->ipsec_sadb_dropper);
5869         }
5870         mutex_exit(&acqrec->ipsacq_lock);
5871 
5872         /* Free */
5873         mutex_destroy(&acqrec->ipsacq_lock);
5874         kmem_free(acqrec, sizeof (*acqrec));
5875 }
5876 
5877 /*
5878  * Destroy an acquire list fanout.
5879  */
5880 static void
5881 sadb_destroy_acqlist(iacqf_t **listp, uint_t numentries, boolean_t forever,
5882     netstack_t *ns)
5883 {
5884         int i;
5885         iacqf_t *list = *listp;
5886 
5887         if (list == NULL)
5888                 return;
5889 
5890         for (i = 0; i < numentries; i++) {
5891                 mutex_enter(&(list[i].iacqf_lock));
5892                 while (list[i].iacqf_ipsacq != NULL)
5893                         sadb_destroy_acquire(list[i].iacqf_ipsacq, ns);
5894                 mutex_exit(&(list[i].iacqf_lock));
5895                 if (forever)
5896                         mutex_destroy(&(list[i].iacqf_lock));
5897         }
5898 
5899         if (forever) {
5900                 *listp = NULL;
5901                 kmem_free(list, numentries * sizeof (*list));
5902         }
5903 }
5904 
5905 /*
5906  * Create an algorithm descriptor for an extended ACQUIRE.  Filter crypto
5907  * framework's view of reality vs. IPsec's.  EF's wins, BTW.
5908  */
5909 static uint8_t *
5910 sadb_new_algdesc(uint8_t *start, uint8_t *limit,
5911     sadb_x_ecomb_t *ecomb, uint8_t satype, uint8_t algtype,
5912     uint8_t alg, uint16_t minbits, uint16_t maxbits, ipsec_stack_t *ipss)
5913 {
5914         uint8_t *cur = start;
5915         ipsec_alginfo_t *algp;
5916         sadb_x_algdesc_t *algdesc = (sadb_x_algdesc_t *)cur;
5917 
5918         cur += sizeof (*algdesc);
5919         if (cur >= limit)
5920                 return (NULL);
5921 
5922         ecomb->sadb_x_ecomb_numalgs++;
5923 
5924         /*
5925          * Normalize vs. crypto framework's limits.  This way, you can specify
5926          * a stronger policy, and when the framework loads a stronger version,
5927          * you can just keep plowing w/o rewhacking your SPD.
5928          */
5929         rw_enter(&ipss->ipsec_alg_lock, RW_READER);
5930         algp = ipss->ipsec_alglists[(algtype == SADB_X_ALGTYPE_AUTH) ?
5931             IPSEC_ALG_AUTH : IPSEC_ALG_ENCR][alg];
5932         if (algp == NULL) {
5933                 rw_exit(&ipss->ipsec_alg_lock);
5934                 return (NULL);  /* Algorithm doesn't exist.  Fail gracefully. */
5935         }
5936         if (minbits < algp->alg_ef_minbits)
5937                 minbits = algp->alg_ef_minbits;
5938         if (maxbits > algp->alg_ef_maxbits)
5939                 maxbits = algp->alg_ef_maxbits;
5940         rw_exit(&ipss->ipsec_alg_lock);
5941 
5942         algdesc->sadb_x_algdesc_reserved = SADB_8TO1(algp->alg_saltlen);
5943         algdesc->sadb_x_algdesc_satype = satype;
5944         algdesc->sadb_x_algdesc_algtype = algtype;
5945         algdesc->sadb_x_algdesc_alg = alg;
5946         algdesc->sadb_x_algdesc_minbits = minbits;
5947         algdesc->sadb_x_algdesc_maxbits = maxbits;
5948 
5949         return (cur);
5950 }
5951 
5952 /*
5953  * Convert the given ipsec_action_t into an ecomb starting at *ecomb
5954  * which must fit before *limit
5955  *
5956  * return NULL if we ran out of room or a pointer to the end of the ecomb.
5957  */
5958 static uint8_t *
5959 sadb_action_to_ecomb(uint8_t *start, uint8_t *limit, ipsec_action_t *act,
5960     netstack_t *ns)
5961 {
5962         uint8_t *cur = start;
5963         sadb_x_ecomb_t *ecomb = (sadb_x_ecomb_t *)cur;
5964         ipsec_prot_t *ipp;
5965         ipsec_stack_t *ipss = ns->netstack_ipsec;
5966 
5967         cur += sizeof (*ecomb);
5968         if (cur >= limit)
5969                 return (NULL);
5970 
5971         ASSERT(act->ipa_act.ipa_type == IPSEC_ACT_APPLY);
5972 
5973         ipp = &act->ipa_act.ipa_apply;
5974 
5975         ecomb->sadb_x_ecomb_numalgs = 0;
5976         ecomb->sadb_x_ecomb_reserved = 0;
5977         ecomb->sadb_x_ecomb_reserved2 = 0;
5978         /*
5979          * No limits on allocations, since we really don't support that
5980          * concept currently.
5981          */
5982         ecomb->sadb_x_ecomb_soft_allocations = 0;
5983         ecomb->sadb_x_ecomb_hard_allocations = 0;
5984 
5985         /*
5986          * XXX TBD: Policy or global parameters will eventually be
5987          * able to fill in some of these.
5988          */
5989         ecomb->sadb_x_ecomb_flags = 0;
5990         ecomb->sadb_x_ecomb_soft_bytes = 0;
5991         ecomb->sadb_x_ecomb_hard_bytes = 0;
5992         ecomb->sadb_x_ecomb_soft_addtime = 0;
5993         ecomb->sadb_x_ecomb_hard_addtime = 0;
5994         ecomb->sadb_x_ecomb_soft_usetime = 0;
5995         ecomb->sadb_x_ecomb_hard_usetime = 0;
5996 
5997         if (ipp->ipp_use_ah) {
5998                 cur = sadb_new_algdesc(cur, limit, ecomb,
5999                     SADB_SATYPE_AH, SADB_X_ALGTYPE_AUTH, ipp->ipp_auth_alg,
6000                     ipp->ipp_ah_minbits, ipp->ipp_ah_maxbits, ipss);
6001                 if (cur == NULL)
6002                         return (NULL);
6003                 ipsecah_fill_defs(ecomb, ns);
6004         }
6005 
6006         if (ipp->ipp_use_esp) {
6007                 if (ipp->ipp_use_espa) {
6008                         cur = sadb_new_algdesc(cur, limit, ecomb,
6009                             SADB_SATYPE_ESP, SADB_X_ALGTYPE_AUTH,
6010                             ipp->ipp_esp_auth_alg,
6011                             ipp->ipp_espa_minbits,
6012                             ipp->ipp_espa_maxbits, ipss);
6013                         if (cur == NULL)
6014                                 return (NULL);
6015                 }
6016 
6017                 cur = sadb_new_algdesc(cur, limit, ecomb,
6018                     SADB_SATYPE_ESP, SADB_X_ALGTYPE_CRYPT,
6019                     ipp->ipp_encr_alg,
6020                     ipp->ipp_espe_minbits,
6021                     ipp->ipp_espe_maxbits, ipss);
6022                 if (cur == NULL)
6023                         return (NULL);
6024                 /* Fill in lifetimes if and only if AH didn't already... */
6025                 if (!ipp->ipp_use_ah)
6026                         ipsecesp_fill_defs(ecomb, ns);
6027         }
6028 
6029         return (cur);
6030 }
6031 
6032 #include <sys/tsol/label_macro.h> /* XXX should not need this */
6033 
6034 /*
6035  * From a cred_t, construct a sensitivity label extension
6036  *
6037  * We send up a fixed-size sensitivity label bitmap, and are perhaps
6038  * overly chummy with the underlying data structures here.
6039  */
6040 
6041 /* ARGSUSED */
6042 int
6043 sadb_sens_len_from_label(ts_label_t *tsl)
6044 {
6045         int baselen = sizeof (sadb_sens_t) + _C_LEN * 4;
6046         return (roundup(baselen, sizeof (uint64_t)));
6047 }
6048 
6049 void
6050 sadb_sens_from_label(sadb_sens_t *sens, int exttype, ts_label_t *tsl,
6051     int senslen)
6052 {
6053         uint8_t *bitmap;
6054         bslabel_t *sl;
6055 
6056         /* LINTED */
6057         ASSERT((_C_LEN & 1) == 0);
6058         ASSERT((senslen & 7) == 0);
6059 
6060         sl = label2bslabel(tsl);
6061 
6062         sens->sadb_sens_exttype = exttype;
6063         sens->sadb_sens_len = SADB_8TO64(senslen);
6064 
6065         sens->sadb_sens_dpd = tsl->tsl_doi;
6066         sens->sadb_sens_sens_level = LCLASS(sl);
6067         sens->sadb_sens_integ_level = 0; /* TBD */
6068         sens->sadb_sens_sens_len = _C_LEN >> 1;
6069         sens->sadb_sens_integ_len = 0; /* TBD */
6070         sens->sadb_x_sens_flags = 0;
6071 
6072         bitmap = (uint8_t *)(sens + 1);
6073         bcopy(&(((_bslabel_impl_t *)sl)->compartments), bitmap, _C_LEN * 4);
6074 }
6075 
6076 /*
6077  * Okay, how do we report errors/invalid labels from this?
6078  * With a special designated "not a label" cred_t ?
6079  */
6080 /* ARGSUSED */
6081 ts_label_t *
6082 sadb_label_from_sens(sadb_sens_t *sens, uint64_t *bitmap)
6083 {
6084         int bitmap_len = SADB_64TO8(sens->sadb_sens_sens_len);
6085         bslabel_t sl;
6086         ts_label_t *tsl;
6087 
6088         if (sens->sadb_sens_integ_level != 0)
6089                 return (NULL);
6090         if (sens->sadb_sens_integ_len != 0)
6091                 return (NULL);
6092         if (bitmap_len > _C_LEN * 4)
6093                 return (NULL);
6094 
6095         bsllow(&sl);
6096         LCLASS_SET((_bslabel_impl_t *)&sl, sens->sadb_sens_sens_level);
6097         bcopy(bitmap, &((_bslabel_impl_t *)&sl)->compartments,
6098             bitmap_len);
6099 
6100         tsl = labelalloc(&sl, sens->sadb_sens_dpd, KM_NOSLEEP);
6101         if (tsl == NULL)
6102                 return (NULL);
6103 
6104         if (sens->sadb_x_sens_flags & SADB_X_SENS_UNLABELED)
6105                 tsl->tsl_flags |= TSLF_UNLABELED;
6106         return (tsl);
6107 }
6108 
6109 /* End XXX label-library-leakage */
6110 
6111 /*
6112  * Given an SADB_GETSPI message, find an appropriately ranged SA and
6113  * allocate an SA.  If there are message improprieties, return (ipsa_t *)-1.
6114  * If there was a memory allocation error, return NULL.  (Assume NULL !=
6115  * (ipsa_t *)-1).
6116  *
6117  * master_spi is passed in host order.
6118  */
6119 ipsa_t *
6120 sadb_getspi(keysock_in_t *ksi, uint32_t master_spi, int *diagnostic,
6121     netstack_t *ns, uint_t sa_type)
6122 {
6123         sadb_address_t *src =
6124             (sadb_address_t *)ksi->ks_in_extv[SADB_EXT_ADDRESS_SRC],
6125             *dst = (sadb_address_t *)ksi->ks_in_extv[SADB_EXT_ADDRESS_DST];
6126         sadb_spirange_t *range =
6127             (sadb_spirange_t *)ksi->ks_in_extv[SADB_EXT_SPIRANGE];
6128         struct sockaddr_in *ssa, *dsa;
6129         struct sockaddr_in6 *ssa6, *dsa6;
6130         uint32_t *srcaddr, *dstaddr;
6131         sa_family_t af;
6132         uint32_t add, min, max;
6133         uint8_t protocol =
6134             (sa_type == SADB_SATYPE_AH) ? IPPROTO_AH : IPPROTO_ESP;
6135 
6136         if (src == NULL) {
6137                 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_SRC;
6138                 return ((ipsa_t *)-1);
6139         }
6140         if (dst == NULL) {
6141                 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_DST;
6142                 return ((ipsa_t *)-1);
6143         }
6144         if (range == NULL) {
6145                 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_RANGE;
6146                 return ((ipsa_t *)-1);
6147         }
6148 
6149         min = ntohl(range->sadb_spirange_min);
6150         max = ntohl(range->sadb_spirange_max);
6151         dsa = (struct sockaddr_in *)(dst + 1);
6152         dsa6 = (struct sockaddr_in6 *)dsa;
6153 
6154         ssa = (struct sockaddr_in *)(src + 1);
6155         ssa6 = (struct sockaddr_in6 *)ssa;
6156         ASSERT(dsa->sin_family == ssa->sin_family);
6157 
6158         srcaddr = ALL_ZEROES_PTR;
6159         af = dsa->sin_family;
6160         switch (af) {
6161         case AF_INET:
6162                 if (src != NULL)
6163                         srcaddr = (uint32_t *)(&ssa->sin_addr);
6164                 dstaddr = (uint32_t *)(&dsa->sin_addr);
6165                 break;
6166         case AF_INET6:
6167                 if (src != NULL)
6168                         srcaddr = (uint32_t *)(&ssa6->sin6_addr);
6169                 dstaddr = (uint32_t *)(&dsa6->sin6_addr);
6170                 break;
6171         default:
6172                 *diagnostic = SADB_X_DIAGNOSTIC_BAD_DST_AF;
6173                 return ((ipsa_t *)-1);
6174         }
6175 
6176         if (master_spi < min || master_spi > max) {
6177                 /* Return a random value in the range. */
6178                 if (cl_inet_getspi) {
6179                         cl_inet_getspi(ns->netstack_stackid, protocol,
6180                             (uint8_t *)&add, sizeof (add), NULL);
6181                 } else {
6182                         (void) random_get_pseudo_bytes((uint8_t *)&add,
6183                             sizeof (add));
6184                 }
6185                 master_spi = min + (add % (max - min + 1));
6186         }
6187 
6188         /*
6189          * Since master_spi is passed in host order, we need to htonl() it
6190          * for the purposes of creating a new SA.
6191          */
6192         return (sadb_makelarvalassoc(htonl(master_spi), srcaddr, dstaddr, af,
6193             ns));
6194 }
6195 
6196 /*
6197  *
6198  * Locate an ACQUIRE and nuke it.  If I have an samsg that's larger than the
6199  * base header, just ignore it.  Otherwise, lock down the whole ACQUIRE list
6200  * and scan for the sequence number in question.  I may wish to accept an
6201  * address pair with it, for easier searching.
6202  *
6203  * Caller frees the message, so we don't have to here.
6204  *
6205  * NOTE:        The pfkey_q parameter may be used in the future for ACQUIRE
6206  *              failures.
6207  */
6208 /* ARGSUSED */
6209 void
6210 sadb_in_acquire(sadb_msg_t *samsg, sadbp_t *sp, queue_t *pfkey_q,
6211     netstack_t *ns)
6212 {
6213         int i;
6214         ipsacq_t *acqrec;
6215         iacqf_t *bucket;
6216 
6217         /*
6218          * I only accept the base header for this!
6219          * Though to be honest, requiring the dst address would help
6220          * immensely.
6221          *
6222          * XXX  There are already cases where I can get the dst address.
6223          */
6224         if (samsg->sadb_msg_len > SADB_8TO64(sizeof (*samsg)))
6225                 return;
6226 
6227         /*
6228          * Using the samsg->sadb_msg_seq, find the ACQUIRE record, delete it,
6229          * (and in the future send a message to IP with the appropriate error
6230          * number).
6231          *
6232          * Q: Do I want to reject if pid != 0?
6233          */
6234 
6235         for (i = 0; i < sp->s_v4.sdb_hashsize; i++) {
6236                 bucket = &sp->s_v4.sdb_acq[i];
6237                 mutex_enter(&bucket->iacqf_lock);
6238                 for (acqrec = bucket->iacqf_ipsacq; acqrec != NULL;
6239                     acqrec = acqrec->ipsacq_next) {
6240                         if (samsg->sadb_msg_seq == acqrec->ipsacq_seq)
6241                                 break;  /* for acqrec... loop. */
6242                 }
6243                 if (acqrec != NULL)
6244                         break;  /* for i = 0... loop. */
6245 
6246                 mutex_exit(&bucket->iacqf_lock);
6247         }
6248 
6249         if (acqrec == NULL) {
6250                 for (i = 0; i < sp->s_v6.sdb_hashsize; i++) {
6251                         bucket = &sp->s_v6.sdb_acq[i];
6252                         mutex_enter(&bucket->iacqf_lock);
6253                         for (acqrec = bucket->iacqf_ipsacq; acqrec != NULL;
6254                             acqrec = acqrec->ipsacq_next) {
6255                                 if (samsg->sadb_msg_seq == acqrec->ipsacq_seq)
6256                                         break;  /* for acqrec... loop. */
6257                         }
6258                         if (acqrec != NULL)
6259                                 break;  /* for i = 0... loop. */
6260 
6261                         mutex_exit(&bucket->iacqf_lock);
6262                 }
6263         }
6264 
6265 
6266         if (acqrec == NULL)
6267                 return;
6268 
6269         /*
6270          * What do I do with the errno and IP?  I may need mp's services a
6271          * little more.  See sadb_destroy_acquire() for future directions
6272          * beyond free the mblk chain on the acquire record.
6273          */
6274 
6275         ASSERT(&bucket->iacqf_lock == acqrec->ipsacq_linklock);
6276         sadb_destroy_acquire(acqrec, ns);
6277         /* Have to exit mutex here, because of breaking out of for loop. */
6278         mutex_exit(&bucket->iacqf_lock);
6279 }
6280 
6281 /*
6282  * The following functions work with the replay windows of an SA.  They assume
6283  * the ipsa->ipsa_replay_arr is an array of uint64_t, and that the bit vector
6284  * represents the highest sequence number packet received, and back
6285  * (ipsa->ipsa_replay_wsize) packets.
6286  */
6287 
6288 /*
6289  * Is the replay bit set?
6290  */
6291 static boolean_t
6292 ipsa_is_replay_set(ipsa_t *ipsa, uint32_t offset)
6293 {
6294         uint64_t bit = (uint64_t)1 << (uint64_t)(offset & 63);
6295 
6296         return ((bit & ipsa->ipsa_replay_arr[offset >> 6]) ? B_TRUE : B_FALSE);
6297 }
6298 
6299 /*
6300  * Shift the bits of the replay window over.
6301  */
6302 static void
6303 ipsa_shift_replay(ipsa_t *ipsa, uint32_t shift)
6304 {
6305         int i;
6306         int jump = ((shift - 1) >> 6) + 1;
6307 
6308         if (shift == 0)
6309                 return;
6310 
6311         for (i = (ipsa->ipsa_replay_wsize - 1) >> 6; i >= 0; i--) {
6312                 if (i + jump <= (ipsa->ipsa_replay_wsize - 1) >> 6) {
6313                         ipsa->ipsa_replay_arr[i + jump] |=
6314                             ipsa->ipsa_replay_arr[i] >> (64 - (shift & 63));
6315                 }
6316                 ipsa->ipsa_replay_arr[i] <<= shift;
6317         }
6318 }
6319 
6320 /*
6321  * Set a bit in the bit vector.
6322  */
6323 static void
6324 ipsa_set_replay(ipsa_t *ipsa, uint32_t offset)
6325 {
6326         uint64_t bit = (uint64_t)1 << (uint64_t)(offset & 63);
6327 
6328         ipsa->ipsa_replay_arr[offset >> 6] |= bit;
6329 }
6330 
6331 #define SADB_MAX_REPLAY_VALUE 0xffffffff
6332 
6333 /*
6334  * Assume caller has NOT done ntohl() already on seq.  Check to see
6335  * if replay sequence number "seq" has been seen already.
6336  */
6337 boolean_t
6338 sadb_replay_check(ipsa_t *ipsa, uint32_t seq)
6339 {
6340         boolean_t rc;
6341         uint32_t diff;
6342 
6343         if (ipsa->ipsa_replay_wsize == 0)
6344                 return (B_TRUE);
6345 
6346         /*
6347          * NOTE:  I've already checked for 0 on the wire in sadb_replay_peek().
6348          */
6349 
6350         /* Convert sequence number into host order before holding the mutex. */
6351         seq = ntohl(seq);
6352 
6353         mutex_enter(&ipsa->ipsa_lock);
6354 
6355         /* Initialize inbound SA's ipsa_replay field to last one received. */
6356         if (ipsa->ipsa_replay == 0)
6357                 ipsa->ipsa_replay = 1;
6358 
6359         if (seq > ipsa->ipsa_replay) {
6360                 /*
6361                  * I have received a new "highest value received".  Shift
6362                  * the replay window over.
6363                  */
6364                 diff = seq - ipsa->ipsa_replay;
6365                 if (diff < ipsa->ipsa_replay_wsize) {
6366                         /* In replay window, shift bits over. */
6367                         ipsa_shift_replay(ipsa, diff);
6368                 } else {
6369                         /* WAY FAR AHEAD, clear bits and start again. */
6370                         bzero(ipsa->ipsa_replay_arr,
6371                             sizeof (ipsa->ipsa_replay_arr));
6372                 }
6373                 ipsa_set_replay(ipsa, 0);
6374                 ipsa->ipsa_replay = seq;
6375                 rc = B_TRUE;
6376                 goto done;
6377         }
6378         diff = ipsa->ipsa_replay - seq;
6379         if (diff >= ipsa->ipsa_replay_wsize || ipsa_is_replay_set(ipsa, diff)) {
6380                 rc = B_FALSE;
6381                 goto done;
6382         }
6383         /* Set this packet as seen. */
6384         ipsa_set_replay(ipsa, diff);
6385 
6386         rc = B_TRUE;
6387 done:
6388         mutex_exit(&ipsa->ipsa_lock);
6389         return (rc);
6390 }
6391 
6392 /*
6393  * "Peek" and see if we should even bother going through the effort of
6394  * running an authentication check on the sequence number passed in.
6395  * this takes into account packets that are below the replay window,
6396  * and collisions with already replayed packets.  Return B_TRUE if it
6397  * is okay to proceed, B_FALSE if this packet should be dropped immediately.
6398  * Assume same byte-ordering as sadb_replay_check.
6399  */
6400 boolean_t
6401 sadb_replay_peek(ipsa_t *ipsa, uint32_t seq)
6402 {
6403         boolean_t rc = B_FALSE;
6404         uint32_t diff;
6405 
6406         if (ipsa->ipsa_replay_wsize == 0)
6407                 return (B_TRUE);
6408 
6409         /*
6410          * 0 is 0, regardless of byte order... :)
6411          *
6412          * If I get 0 on the wire (and there is a replay window) then the
6413          * sender most likely wrapped.  This ipsa may need to be marked or
6414          * something.
6415          */
6416         if (seq == 0)
6417                 return (B_FALSE);
6418 
6419         seq = ntohl(seq);
6420         mutex_enter(&ipsa->ipsa_lock);
6421         if (seq < ipsa->ipsa_replay - ipsa->ipsa_replay_wsize &&
6422             ipsa->ipsa_replay >= ipsa->ipsa_replay_wsize)
6423                 goto done;
6424 
6425         /*
6426          * If I've hit 0xffffffff, then quite honestly, I don't need to
6427          * bother with formalities.  I'm not accepting any more packets
6428          * on this SA.
6429          */
6430         if (ipsa->ipsa_replay == SADB_MAX_REPLAY_VALUE) {
6431                 /*
6432                  * Since we're already holding the lock, update the
6433                  * expire time ala. sadb_replay_delete() and return.
6434                  */
6435                 ipsa->ipsa_hardexpiretime = (time_t)1;
6436                 goto done;
6437         }
6438 
6439         if (seq <= ipsa->ipsa_replay) {
6440                 /*
6441                  * This seq is in the replay window.  I'm not below it,
6442                  * because I already checked for that above!
6443                  */
6444                 diff = ipsa->ipsa_replay - seq;
6445                 if (ipsa_is_replay_set(ipsa, diff))
6446                         goto done;
6447         }
6448         /* Else return B_TRUE, I'm going to advance the window. */
6449 
6450         rc = B_TRUE;
6451 done:
6452         mutex_exit(&ipsa->ipsa_lock);
6453         return (rc);
6454 }
6455 
6456 /*
6457  * Delete a single SA.
6458  *
6459  * For now, use the quick-and-dirty trick of making the association's
6460  * hard-expire lifetime (time_t)1, ensuring deletion by the *_ager().
6461  */
6462 void
6463 sadb_replay_delete(ipsa_t *assoc)
6464 {
6465         mutex_enter(&assoc->ipsa_lock);
6466         assoc->ipsa_hardexpiretime = (time_t)1;
6467         mutex_exit(&assoc->ipsa_lock);
6468 }
6469 
6470 /*
6471  * Special front-end to ipsec_rl_strlog() dealing with SA failure.
6472  * this is designed to take only a format string with "* %x * %s *", so
6473  * that "spi" is printed first, then "addr" is converted using inet_pton().
6474  *
6475  * This is abstracted out to save the stack space for only when inet_pton()
6476  * is called.  Make sure "spi" is in network order; it usually is when this
6477  * would get called.
6478  */
6479 void
6480 ipsec_assocfailure(short mid, short sid, char level, ushort_t sl, char *fmt,
6481     uint32_t spi, void *addr, int af, netstack_t *ns)
6482 {
6483         char buf[INET6_ADDRSTRLEN];
6484 
6485         ASSERT(af == AF_INET6 || af == AF_INET);
6486 
6487         ipsec_rl_strlog(ns, mid, sid, level, sl, fmt, ntohl(spi),
6488             inet_ntop(af, addr, buf, sizeof (buf)));
6489 }
6490 
6491 /*
6492  * Fills in a reference to the policy, if any, from the conn, in *ppp
6493  */
6494 static void
6495 ipsec_conn_pol(ipsec_selector_t *sel, conn_t *connp, ipsec_policy_t **ppp)
6496 {
6497         ipsec_policy_t  *pp;
6498         ipsec_latch_t   *ipl = connp->conn_latch;
6499 
6500         if ((ipl != NULL) && (connp->conn_ixa->ixa_ipsec_policy != NULL)) {
6501                 pp = connp->conn_ixa->ixa_ipsec_policy;
6502                 IPPOL_REFHOLD(pp);
6503         } else {
6504                 pp = ipsec_find_policy(IPSEC_TYPE_OUTBOUND, connp, sel,
6505                     connp->conn_netstack);
6506         }
6507         *ppp = pp;
6508 }
6509 
6510 /*
6511  * The following functions scan through active conn_t structures
6512  * and return a reference to the best-matching policy it can find.
6513  * Caller must release the reference.
6514  */
6515 static void
6516 ipsec_udp_pol(ipsec_selector_t *sel, ipsec_policy_t **ppp, ip_stack_t *ipst)
6517 {
6518         connf_t *connfp;
6519         conn_t *connp = NULL;
6520         ipsec_selector_t portonly;
6521 
6522         bzero((void *)&portonly, sizeof (portonly));
6523 
6524         if (sel->ips_local_port == 0)
6525                 return;
6526 
6527         connfp = &ipst->ips_ipcl_udp_fanout[IPCL_UDP_HASH(sel->ips_local_port,
6528             ipst)];
6529         mutex_enter(&connfp->connf_lock);
6530 
6531         if (sel->ips_isv4) {
6532                 connp = connfp->connf_head;
6533                 while (connp != NULL) {
6534                         if (IPCL_UDP_MATCH(connp, sel->ips_local_port,
6535                             sel->ips_local_addr_v4, sel->ips_remote_port,
6536                             sel->ips_remote_addr_v4))
6537                                 break;
6538                         connp = connp->conn_next;
6539                 }
6540 
6541                 if (connp == NULL) {
6542                         /* Try port-only match in IPv6. */
6543                         portonly.ips_local_port = sel->ips_local_port;
6544                         sel = &portonly;
6545                 }
6546         }
6547 
6548         if (connp == NULL) {
6549                 connp = connfp->connf_head;
6550                 while (connp != NULL) {
6551                         if (IPCL_UDP_MATCH_V6(connp, sel->ips_local_port,
6552                             sel->ips_local_addr_v6, sel->ips_remote_port,
6553                             sel->ips_remote_addr_v6))
6554                                 break;
6555                         connp = connp->conn_next;
6556                 }
6557 
6558                 if (connp == NULL) {
6559                         mutex_exit(&connfp->connf_lock);
6560                         return;
6561                 }
6562         }
6563 
6564         CONN_INC_REF(connp);
6565         mutex_exit(&connfp->connf_lock);
6566 
6567         ipsec_conn_pol(sel, connp, ppp);
6568         CONN_DEC_REF(connp);
6569 }
6570 
6571 static conn_t *
6572 ipsec_find_listen_conn(uint16_t *pptr, ipsec_selector_t *sel, ip_stack_t *ipst)
6573 {
6574         connf_t *connfp;
6575         conn_t *connp = NULL;
6576         const in6_addr_t *v6addrmatch = &sel->ips_local_addr_v6;
6577 
6578         if (sel->ips_local_port == 0)
6579                 return (NULL);
6580 
6581         connfp = &ipst->ips_ipcl_bind_fanout[
6582             IPCL_BIND_HASH(sel->ips_local_port, ipst)];
6583         mutex_enter(&connfp->connf_lock);
6584 
6585         if (sel->ips_isv4) {
6586                 connp = connfp->connf_head;
6587                 while (connp != NULL) {
6588                         if (IPCL_BIND_MATCH(connp, IPPROTO_TCP,
6589                             sel->ips_local_addr_v4, pptr[1]))
6590                                 break;
6591                         connp = connp->conn_next;
6592                 }
6593 
6594                 if (connp == NULL) {
6595                         /* Match to all-zeroes. */
6596                         v6addrmatch = &ipv6_all_zeros;
6597                 }
6598         }
6599 
6600         if (connp == NULL) {
6601                 connp = connfp->connf_head;
6602                 while (connp != NULL) {
6603                         if (IPCL_BIND_MATCH_V6(connp, IPPROTO_TCP,
6604                             *v6addrmatch, pptr[1]))
6605                                 break;
6606                         connp = connp->conn_next;
6607                 }
6608 
6609                 if (connp == NULL) {
6610                         mutex_exit(&connfp->connf_lock);
6611                         return (NULL);
6612                 }
6613         }
6614 
6615         CONN_INC_REF(connp);
6616         mutex_exit(&connfp->connf_lock);
6617         return (connp);
6618 }
6619 
6620 static void
6621 ipsec_tcp_pol(ipsec_selector_t *sel, ipsec_policy_t **ppp, ip_stack_t *ipst)
6622 {
6623         connf_t         *connfp;
6624         conn_t          *connp;
6625         uint32_t        ports;
6626         uint16_t        *pptr = (uint16_t *)&ports;
6627 
6628         /*
6629          * Find TCP state in the following order:
6630          * 1.) Connected conns.
6631          * 2.) Listeners.
6632          *
6633          * Even though #2 will be the common case for inbound traffic, only
6634          * following this order insures correctness.
6635          */
6636 
6637         if (sel->ips_local_port == 0)
6638                 return;
6639 
6640         /*
6641          * 0 should be fport, 1 should be lport.  SRC is the local one here.
6642          * See ipsec_construct_inverse_acquire() for details.
6643          */
6644         pptr[0] = sel->ips_remote_port;
6645         pptr[1] = sel->ips_local_port;
6646 
6647         connfp = &ipst->ips_ipcl_conn_fanout[
6648             IPCL_CONN_HASH(sel->ips_remote_addr_v4, ports, ipst)];
6649         mutex_enter(&connfp->connf_lock);
6650         connp = connfp->connf_head;
6651 
6652         if (sel->ips_isv4) {
6653                 while (connp != NULL) {
6654                         if (IPCL_CONN_MATCH(connp, IPPROTO_TCP,
6655                             sel->ips_remote_addr_v4, sel->ips_local_addr_v4,
6656                             ports))
6657                                 break;
6658                         connp = connp->conn_next;
6659                 }
6660         } else {
6661                 while (connp != NULL) {
6662                         if (IPCL_CONN_MATCH_V6(connp, IPPROTO_TCP,
6663                             sel->ips_remote_addr_v6, sel->ips_local_addr_v6,
6664                             ports))
6665                                 break;
6666                         connp = connp->conn_next;
6667                 }
6668         }
6669 
6670         if (connp != NULL) {
6671                 CONN_INC_REF(connp);
6672                 mutex_exit(&connfp->connf_lock);
6673         } else {
6674                 mutex_exit(&connfp->connf_lock);
6675 
6676                 /* Try the listen hash. */
6677                 if ((connp = ipsec_find_listen_conn(pptr, sel, ipst)) == NULL)
6678                         return;
6679         }
6680 
6681         ipsec_conn_pol(sel, connp, ppp);
6682         CONN_DEC_REF(connp);
6683 }
6684 
6685 static void
6686 ipsec_sctp_pol(ipsec_selector_t *sel, ipsec_policy_t **ppp,
6687     ip_stack_t *ipst)
6688 {
6689         conn_t          *connp;
6690         uint32_t        ports;
6691         uint16_t        *pptr = (uint16_t *)&ports;
6692 
6693         /*
6694          * Find SCP state in the following order:
6695          * 1.) Connected conns.
6696          * 2.) Listeners.
6697          *
6698          * Even though #2 will be the common case for inbound traffic, only
6699          * following this order insures correctness.
6700          */
6701 
6702         if (sel->ips_local_port == 0)
6703                 return;
6704 
6705         /*
6706          * 0 should be fport, 1 should be lport.  SRC is the local one here.
6707          * See ipsec_construct_inverse_acquire() for details.
6708          */
6709         pptr[0] = sel->ips_remote_port;
6710         pptr[1] = sel->ips_local_port;
6711 
6712         /*
6713          * For labeled systems, there's no need to check the
6714          * label here.  It's known to be good as we checked
6715          * before allowing the connection to become bound.
6716          */
6717         if (sel->ips_isv4) {
6718                 in6_addr_t      src, dst;
6719 
6720                 IN6_IPADDR_TO_V4MAPPED(sel->ips_remote_addr_v4, &dst);
6721                 IN6_IPADDR_TO_V4MAPPED(sel->ips_local_addr_v4, &src);
6722                 connp = sctp_find_conn(&dst, &src, ports, ALL_ZONES,
6723                     0, ipst->ips_netstack->netstack_sctp);
6724         } else {
6725                 connp = sctp_find_conn(&sel->ips_remote_addr_v6,
6726                     &sel->ips_local_addr_v6, ports, ALL_ZONES,
6727                     0, ipst->ips_netstack->netstack_sctp);
6728         }
6729         if (connp == NULL)
6730                 return;
6731         ipsec_conn_pol(sel, connp, ppp);
6732         CONN_DEC_REF(connp);
6733 }
6734 
6735 /*
6736  * Fill in a query for the SPD (in "sel") using two PF_KEY address extensions.
6737  * Returns 0 or errno, and always sets *diagnostic to something appropriate
6738  * to PF_KEY.
6739  *
6740  * NOTE:  For right now, this function (and ipsec_selector_t for that matter),
6741  * ignore prefix lengths in the address extension.  Since we match on first-
6742  * entered policies, this shouldn't matter.  Also, since we normalize prefix-
6743  * set addresses to mask out the lower bits, we should get a suitable search
6744  * key for the SPD anyway.  This is the function to change if the assumption
6745  * about suitable search keys is wrong.
6746  */
6747 static int
6748 ipsec_get_inverse_acquire_sel(ipsec_selector_t *sel, sadb_address_t *srcext,
6749     sadb_address_t *dstext, int *diagnostic)
6750 {
6751         struct sockaddr_in *src, *dst;
6752         struct sockaddr_in6 *src6, *dst6;
6753 
6754         *diagnostic = 0;
6755 
6756         bzero(sel, sizeof (*sel));
6757         sel->ips_protocol = srcext->sadb_address_proto;
6758         dst = (struct sockaddr_in *)(dstext + 1);
6759         if (dst->sin_family == AF_INET6) {
6760                 dst6 = (struct sockaddr_in6 *)dst;
6761                 src6 = (struct sockaddr_in6 *)(srcext + 1);
6762                 if (src6->sin6_family != AF_INET6) {
6763                         *diagnostic = SADB_X_DIAGNOSTIC_AF_MISMATCH;
6764                         return (EINVAL);
6765                 }
6766                 sel->ips_remote_addr_v6 = dst6->sin6_addr;
6767                 sel->ips_local_addr_v6 = src6->sin6_addr;
6768                 if (sel->ips_protocol == IPPROTO_ICMPV6) {
6769                         sel->ips_is_icmp_inv_acq = 1;
6770                 } else {
6771                         sel->ips_remote_port = dst6->sin6_port;
6772                         sel->ips_local_port = src6->sin6_port;
6773                 }
6774                 sel->ips_isv4 = B_FALSE;
6775         } else {
6776                 src = (struct sockaddr_in *)(srcext + 1);
6777                 if (src->sin_family != AF_INET) {
6778                         *diagnostic = SADB_X_DIAGNOSTIC_AF_MISMATCH;
6779                         return (EINVAL);
6780                 }
6781                 sel->ips_remote_addr_v4 = dst->sin_addr.s_addr;
6782                 sel->ips_local_addr_v4 = src->sin_addr.s_addr;
6783                 if (sel->ips_protocol == IPPROTO_ICMP) {
6784                         sel->ips_is_icmp_inv_acq = 1;
6785                 } else {
6786                         sel->ips_remote_port = dst->sin_port;
6787                         sel->ips_local_port = src->sin_port;
6788                 }
6789                 sel->ips_isv4 = B_TRUE;
6790         }
6791         return (0);
6792 }
6793 
6794 /*
6795  * We have encapsulation.
6796  * - Lookup tun_t by address and look for an associated
6797  *   tunnel policy
6798  * - If there are inner selectors
6799  *   - check ITPF_P_TUNNEL and ITPF_P_ACTIVE
6800  *   - Look up tunnel policy based on selectors
6801  * - Else
6802  *   - Sanity check the negotation
6803  *   - If appropriate, fall through to global policy
6804  */
6805 static int
6806 ipsec_tun_pol(ipsec_selector_t *sel, ipsec_policy_t **ppp,
6807     sadb_address_t *innsrcext, sadb_address_t *inndstext, ipsec_tun_pol_t *itp,
6808     int *diagnostic)
6809 {
6810         int err;
6811         ipsec_policy_head_t *polhead;
6812 
6813         *diagnostic = 0;
6814 
6815         /* Check for inner selectors and act appropriately */
6816 
6817         if (innsrcext != NULL) {
6818                 /* Inner selectors present */
6819                 ASSERT(inndstext != NULL);
6820                 if ((itp == NULL) ||
6821                     (itp->itp_flags & (ITPF_P_ACTIVE | ITPF_P_TUNNEL)) !=
6822                     (ITPF_P_ACTIVE | ITPF_P_TUNNEL)) {
6823                         /*
6824                          * If inner packet selectors, we must have negotiate
6825                          * tunnel and active policy.  If the tunnel has
6826                          * transport-mode policy set on it, or has no policy,
6827                          * fail.
6828                          */
6829                         return (ENOENT);
6830                 } else {
6831                         /*
6832                          * Reset "sel" to indicate inner selectors.  Pass
6833                          * inner PF_KEY address extensions for this to happen.
6834                          */
6835                         if ((err = ipsec_get_inverse_acquire_sel(sel,
6836                             innsrcext, inndstext, diagnostic)) != 0)
6837                                 return (err);
6838                         /*
6839                          * Now look for a tunnel policy based on those inner
6840                          * selectors.  (Common code is below.)
6841                          */
6842                 }
6843         } else {
6844                 /* No inner selectors present */
6845                 if ((itp == NULL) || !(itp->itp_flags & ITPF_P_ACTIVE)) {
6846                         /*
6847                          * Transport mode negotiation with no tunnel policy
6848                          * configured - return to indicate a global policy
6849                          * check is needed.
6850                          */
6851                         return (0);
6852                 } else if (itp->itp_flags & ITPF_P_TUNNEL) {
6853                         /* Tunnel mode set with no inner selectors. */
6854                         return (ENOENT);
6855                 }
6856                 /*
6857                  * Else, this is a tunnel policy configured with ifconfig(1m)
6858                  * or "negotiate transport" with ipsecconf(1m).  We have an
6859                  * itp with policy set based on any match, so don't bother
6860                  * changing fields in "sel".
6861                  */
6862         }
6863 
6864         ASSERT(itp != NULL);
6865         polhead = itp->itp_policy;
6866         ASSERT(polhead != NULL);
6867         rw_enter(&polhead->iph_lock, RW_READER);
6868         *ppp = ipsec_find_policy_head(NULL, polhead, IPSEC_TYPE_INBOUND, sel);
6869         rw_exit(&polhead->iph_lock);
6870 
6871         /*
6872          * Don't default to global if we didn't find a matching policy entry.
6873          * Instead, send ENOENT, just like if we hit a transport-mode tunnel.
6874          */
6875         if (*ppp == NULL)
6876                 return (ENOENT);
6877 
6878         return (0);
6879 }
6880 
6881 /*
6882  * For sctp conn_faddr is the primary address, hence this is of limited
6883  * use for sctp.
6884  */
6885 static void
6886 ipsec_oth_pol(ipsec_selector_t *sel, ipsec_policy_t **ppp,
6887     ip_stack_t *ipst)
6888 {
6889         boolean_t       isv4 = sel->ips_isv4;
6890         connf_t         *connfp;
6891         conn_t          *connp;
6892 
6893         if (isv4) {
6894                 connfp = &ipst->ips_ipcl_proto_fanout_v4[sel->ips_protocol];
6895         } else {
6896                 connfp = &ipst->ips_ipcl_proto_fanout_v6[sel->ips_protocol];
6897         }
6898 
6899         mutex_enter(&connfp->connf_lock);
6900         for (connp = connfp->connf_head; connp != NULL;
6901             connp = connp->conn_next) {
6902                 if (isv4) {
6903                         if ((connp->conn_laddr_v4 == INADDR_ANY ||
6904                             connp->conn_laddr_v4 == sel->ips_local_addr_v4) &&
6905                             (connp->conn_faddr_v4 == INADDR_ANY ||
6906                             connp->conn_faddr_v4 == sel->ips_remote_addr_v4))
6907                                 break;
6908                 } else {
6909                         if ((IN6_IS_ADDR_UNSPECIFIED(&connp->conn_laddr_v6) ||
6910                             IN6_ARE_ADDR_EQUAL(&connp->conn_laddr_v6,
6911                             &sel->ips_local_addr_v6)) &&
6912                             (IN6_IS_ADDR_UNSPECIFIED(&connp->conn_faddr_v6) ||
6913                             IN6_ARE_ADDR_EQUAL(&connp->conn_faddr_v6,
6914                             &sel->ips_remote_addr_v6)))
6915                                 break;
6916                 }
6917         }
6918         if (connp == NULL) {
6919                 mutex_exit(&connfp->connf_lock);
6920                 return;
6921         }
6922 
6923         CONN_INC_REF(connp);
6924         mutex_exit(&connfp->connf_lock);
6925 
6926         ipsec_conn_pol(sel, connp, ppp);
6927         CONN_DEC_REF(connp);
6928 }
6929 
6930 /*
6931  * Construct an inverse ACQUIRE reply based on:
6932  *
6933  * 1.) Current global policy.
6934  * 2.) An conn_t match depending on what all was passed in the extv[].
6935  * 3.) A tunnel's policy head.
6936  * ...
6937  * N.) Other stuff TBD (e.g. identities)
6938  *
6939  * If there is an error, set sadb_msg_errno and sadb_x_msg_diagnostic
6940  * in this function so the caller can extract them where appropriately.
6941  *
6942  * The SRC address is the local one - just like an outbound ACQUIRE message.
6943  *
6944  * XXX MLS: key management supplies a label which we just reflect back up
6945  * again.  clearly we need to involve the label in the rest of the checks.
6946  */
6947 mblk_t *
6948 ipsec_construct_inverse_acquire(sadb_msg_t *samsg, sadb_ext_t *extv[],
6949     netstack_t *ns)
6950 {
6951         int err;
6952         int diagnostic;
6953         sadb_address_t *srcext = (sadb_address_t *)extv[SADB_EXT_ADDRESS_SRC],
6954             *dstext = (sadb_address_t *)extv[SADB_EXT_ADDRESS_DST],
6955             *innsrcext = (sadb_address_t *)extv[SADB_X_EXT_ADDRESS_INNER_SRC],
6956             *inndstext = (sadb_address_t *)extv[SADB_X_EXT_ADDRESS_INNER_DST];
6957         sadb_sens_t *sens = (sadb_sens_t *)extv[SADB_EXT_SENSITIVITY];
6958         struct sockaddr_in6 *src, *dst;
6959         struct sockaddr_in6 *isrc, *idst;
6960         ipsec_tun_pol_t *itp = NULL;
6961         ipsec_policy_t *pp = NULL;
6962         ipsec_selector_t sel, isel;
6963         mblk_t *retmp = NULL;
6964         ip_stack_t      *ipst = ns->netstack_ip;
6965 
6966 
6967         /* Normalize addresses */
6968         if (sadb_addrcheck(NULL, (mblk_t *)samsg, (sadb_ext_t *)srcext, 0, ns)
6969             == KS_IN_ADDR_UNKNOWN) {
6970                 err = EINVAL;
6971                 diagnostic = SADB_X_DIAGNOSTIC_BAD_SRC;
6972                 goto bail;
6973         }
6974         src = (struct sockaddr_in6 *)(srcext + 1);
6975         if (sadb_addrcheck(NULL, (mblk_t *)samsg, (sadb_ext_t *)dstext, 0, ns)
6976             == KS_IN_ADDR_UNKNOWN) {
6977                 err = EINVAL;
6978                 diagnostic = SADB_X_DIAGNOSTIC_BAD_DST;
6979                 goto bail;
6980         }
6981         dst = (struct sockaddr_in6 *)(dstext + 1);
6982         if (src->sin6_family != dst->sin6_family) {
6983                 err = EINVAL;
6984                 diagnostic = SADB_X_DIAGNOSTIC_AF_MISMATCH;
6985                 goto bail;
6986         }
6987 
6988         /* Check for tunnel mode and act appropriately */
6989         if (innsrcext != NULL) {
6990                 if (inndstext == NULL) {
6991                         err = EINVAL;
6992                         diagnostic = SADB_X_DIAGNOSTIC_MISSING_INNER_DST;
6993                         goto bail;
6994                 }
6995                 if (sadb_addrcheck(NULL, (mblk_t *)samsg,
6996                     (sadb_ext_t *)innsrcext, 0, ns) == KS_IN_ADDR_UNKNOWN) {
6997                         err = EINVAL;
6998                         diagnostic = SADB_X_DIAGNOSTIC_MALFORMED_INNER_SRC;
6999                         goto bail;
7000                 }
7001                 isrc = (struct sockaddr_in6 *)(innsrcext + 1);
7002                 if (sadb_addrcheck(NULL, (mblk_t *)samsg,
7003                     (sadb_ext_t *)inndstext, 0, ns) == KS_IN_ADDR_UNKNOWN) {
7004                         err = EINVAL;
7005                         diagnostic = SADB_X_DIAGNOSTIC_MALFORMED_INNER_DST;
7006                         goto bail;
7007                 }
7008                 idst = (struct sockaddr_in6 *)(inndstext + 1);
7009                 if (isrc->sin6_family != idst->sin6_family) {
7010                         err = EINVAL;
7011                         diagnostic = SADB_X_DIAGNOSTIC_INNER_AF_MISMATCH;
7012                         goto bail;
7013                 }
7014                 if (isrc->sin6_family != AF_INET &&
7015                     isrc->sin6_family != AF_INET6) {
7016                         err = EINVAL;
7017                         diagnostic = SADB_X_DIAGNOSTIC_BAD_INNER_SRC_AF;
7018                         goto bail;
7019                 }
7020         } else if (inndstext != NULL) {
7021                 err = EINVAL;
7022                 diagnostic = SADB_X_DIAGNOSTIC_MISSING_INNER_SRC;
7023                 goto bail;
7024         }
7025 
7026         /* Get selectors first, based on outer addresses */
7027         err = ipsec_get_inverse_acquire_sel(&sel, srcext, dstext, &diagnostic);
7028         if (err != 0)
7029                 goto bail;
7030 
7031         /* Check for tunnel mode mismatches. */
7032         if (innsrcext != NULL &&
7033             ((isrc->sin6_family == AF_INET &&
7034             sel.ips_protocol != IPPROTO_ENCAP && sel.ips_protocol != 0) ||
7035             (isrc->sin6_family == AF_INET6 &&
7036             sel.ips_protocol != IPPROTO_IPV6 && sel.ips_protocol != 0))) {
7037                 err = EPROTOTYPE;
7038                 goto bail;
7039         }
7040 
7041         /*
7042          * Okay, we have the addresses and other selector information.
7043          * Let's first find a conn...
7044          */
7045         pp = NULL;
7046         switch (sel.ips_protocol) {
7047         case IPPROTO_TCP:
7048                 ipsec_tcp_pol(&sel, &pp, ipst);
7049                 break;
7050         case IPPROTO_UDP:
7051                 ipsec_udp_pol(&sel, &pp, ipst);
7052                 break;
7053         case IPPROTO_SCTP:
7054                 ipsec_sctp_pol(&sel, &pp, ipst);
7055                 break;
7056         case IPPROTO_ENCAP:
7057         case IPPROTO_IPV6:
7058                 /*
7059                  * Assume sel.ips_remote_addr_* has the right address at
7060                  * that exact position.
7061                  */
7062                 itp = itp_get_byaddr((uint32_t *)(&sel.ips_local_addr_v6),
7063                     (uint32_t *)(&sel.ips_remote_addr_v6), src->sin6_family,
7064                     ipst);
7065 
7066                 if (innsrcext == NULL) {
7067                         /*
7068                          * Transport-mode tunnel, make sure we fake out isel
7069                          * to contain something based on the outer protocol.
7070                          */
7071                         bzero(&isel, sizeof (isel));
7072                         isel.ips_isv4 = (sel.ips_protocol == IPPROTO_ENCAP);
7073                 } /* Else isel is initialized by ipsec_tun_pol(). */
7074                 err = ipsec_tun_pol(&isel, &pp, innsrcext, inndstext, itp,
7075                     &diagnostic);
7076                 /*
7077                  * NOTE:  isel isn't used for now, but in RFC 430x IPsec, it
7078                  * may be.
7079                  */
7080                 if (err != 0)
7081                         goto bail;
7082                 break;
7083         default:
7084                 ipsec_oth_pol(&sel, &pp, ipst);
7085                 break;
7086         }
7087 
7088         /*
7089          * If we didn't find a matching conn_t or other policy head, take a
7090          * look in the global policy.
7091          */
7092         if (pp == NULL) {
7093                 pp = ipsec_find_policy(IPSEC_TYPE_OUTBOUND, NULL, &sel, ns);
7094                 if (pp == NULL) {
7095                         /* There's no global policy. */
7096                         err = ENOENT;
7097                         diagnostic = 0;
7098                         goto bail;
7099                 }
7100         }
7101 
7102         ASSERT(pp != NULL);
7103         retmp = sadb_acquire_msg_base(0, 0, samsg->sadb_msg_seq,
7104             samsg->sadb_msg_pid);
7105         if (retmp != NULL) {
7106                 /* Remove KEYSOCK_OUT, because caller constructs it instead. */
7107                 mblk_t *kso = retmp;
7108 
7109                 retmp = retmp->b_cont;
7110                 freeb(kso);
7111                 /* Append addresses... */
7112                 retmp->b_cont = sadb_acquire_msg_common(&sel, pp, NULL,
7113                     (itp != NULL && (itp->itp_flags & ITPF_P_TUNNEL)), NULL,
7114                     sens);
7115                 if (retmp->b_cont == NULL) {
7116                         freemsg(retmp);
7117                         retmp = NULL;
7118                 }
7119                 /* And the policy result. */
7120                 retmp->b_cont->b_cont =
7121                     sadb_acquire_extended_prop(pp->ipsp_act, ns);
7122                 if (retmp->b_cont->b_cont == NULL) {
7123                         freemsg(retmp);
7124                         retmp = NULL;
7125                 }
7126                 ((sadb_msg_t *)retmp->b_rptr)->sadb_msg_len =
7127                     SADB_8TO64(msgsize(retmp));
7128         }
7129 
7130         if (pp != NULL) {
7131                 IPPOL_REFRELE(pp);
7132         }
7133         ASSERT(err == 0 && diagnostic == 0);
7134         if (retmp == NULL)
7135                 err = ENOMEM;
7136 bail:
7137         if (itp != NULL) {
7138                 ITP_REFRELE(itp, ns);
7139         }
7140         samsg->sadb_msg_errno = (uint8_t)err;
7141         samsg->sadb_x_msg_diagnostic = (uint16_t)diagnostic;
7142         return (retmp);
7143 }
7144 
7145 /*
7146  * ipsa_lpkt is a one-element queue, only manipulated by the next two
7147  * functions.  They have to hold the ipsa_lock because of potential races
7148  * between key management using SADB_UPDATE, and inbound packets that may
7149  * queue up on the larval SA (hence the 'l' in "lpkt").
7150  */
7151 
7152 /*
7153  * sadb_set_lpkt:
7154  *
7155  * Returns the passed-in packet if the SA is no longer larval.
7156  *
7157  * Returns NULL if the SA is larval, and needs to be swapped into the SA for
7158  * processing after an SADB_UPDATE.
7159  */
7160 mblk_t *
7161 sadb_set_lpkt(ipsa_t *ipsa, mblk_t *npkt, ip_recv_attr_t *ira)
7162 {
7163         mblk_t          *opkt;
7164 
7165         mutex_enter(&ipsa->ipsa_lock);
7166         opkt = ipsa->ipsa_lpkt;
7167         if (ipsa->ipsa_state == IPSA_STATE_LARVAL) {
7168                 /*
7169                  * Consume npkt and place it in the LARVAL SA's inbound
7170                  * packet slot.
7171                  */
7172                 mblk_t  *attrmp;
7173 
7174                 attrmp = ip_recv_attr_to_mblk(ira);
7175                 if (attrmp == NULL) {
7176                         ill_t *ill = ira->ira_ill;
7177 
7178                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
7179                         ip_drop_input("ipIfStatsInDiscards", npkt, ill);
7180                         freemsg(npkt);
7181                         opkt = NULL;
7182                 } else {
7183                         ASSERT(attrmp->b_cont == NULL);
7184                         attrmp->b_cont = npkt;
7185                         ipsa->ipsa_lpkt = attrmp;
7186                 }
7187                 npkt = NULL;
7188         } else {
7189                 /*
7190                  * If not larval, we lost the race.  NOTE: ipsa_lpkt may still
7191                  * have been non-NULL in the non-larval case, because of
7192                  * inbound packets arriving prior to sadb_common_add()
7193                  * transferring the SA completely out of larval state, but
7194                  * after lpkt was grabbed by the AH/ESP-specific add routines.
7195                  * We should clear the old ipsa_lpkt in this case to make sure
7196                  * that it doesn't linger on the now-MATURE IPsec SA, or get
7197                  * picked up as an out-of-order packet.
7198                  */
7199                 ipsa->ipsa_lpkt = NULL;
7200         }
7201         mutex_exit(&ipsa->ipsa_lock);
7202 
7203         if (opkt != NULL) {
7204                 ipsec_stack_t   *ipss;
7205 
7206                 ipss = ira->ira_ill->ill_ipst->ips_netstack->netstack_ipsec;
7207                 opkt = ip_recv_attr_free_mblk(opkt);
7208                 ip_drop_packet(opkt, B_TRUE, ira->ira_ill,
7209                     DROPPER(ipss, ipds_sadb_inlarval_replace),
7210                     &ipss->ipsec_sadb_dropper);
7211         }
7212         return (npkt);
7213 }
7214 
7215 /*
7216  * sadb_clear_lpkt: Atomically clear ipsa->ipsa_lpkt and return the
7217  * previous value.
7218  */
7219 mblk_t *
7220 sadb_clear_lpkt(ipsa_t *ipsa)
7221 {
7222         mblk_t *opkt;
7223 
7224         mutex_enter(&ipsa->ipsa_lock);
7225         opkt = ipsa->ipsa_lpkt;
7226         ipsa->ipsa_lpkt = NULL;
7227         mutex_exit(&ipsa->ipsa_lock);
7228         return (opkt);
7229 }
7230 
7231 /*
7232  * Buffer a packet that's in IDLE state as set by Solaris Clustering.
7233  */
7234 void
7235 sadb_buf_pkt(ipsa_t *ipsa, mblk_t *bpkt, ip_recv_attr_t *ira)
7236 {
7237         netstack_t      *ns = ira->ira_ill->ill_ipst->ips_netstack;
7238         ipsec_stack_t   *ipss = ns->netstack_ipsec;
7239         in6_addr_t *srcaddr = (in6_addr_t *)(&ipsa->ipsa_srcaddr);
7240         in6_addr_t *dstaddr = (in6_addr_t *)(&ipsa->ipsa_dstaddr);
7241         mblk_t          *mp;
7242 
7243         ASSERT(ipsa->ipsa_state == IPSA_STATE_IDLE);
7244 
7245         if (cl_inet_idlesa == NULL) {
7246                 ip_drop_packet(bpkt, B_TRUE, ira->ira_ill,
7247                     DROPPER(ipss, ipds_sadb_inidle_overflow),
7248                     &ipss->ipsec_sadb_dropper);
7249                 return;
7250         }
7251 
7252         cl_inet_idlesa(ns->netstack_stackid,
7253             (ipsa->ipsa_type == SADB_SATYPE_AH) ? IPPROTO_AH : IPPROTO_ESP,
7254             ipsa->ipsa_spi, ipsa->ipsa_addrfam, *srcaddr, *dstaddr, NULL);
7255 
7256         mp = ip_recv_attr_to_mblk(ira);
7257         if (mp == NULL) {
7258                 ip_drop_packet(bpkt, B_TRUE, ira->ira_ill,
7259                     DROPPER(ipss, ipds_sadb_inidle_overflow),
7260                     &ipss->ipsec_sadb_dropper);
7261                 return;
7262         }
7263         linkb(mp, bpkt);
7264 
7265         mutex_enter(&ipsa->ipsa_lock);
7266         ipsa->ipsa_mblkcnt++;
7267         if (ipsa->ipsa_bpkt_head == NULL) {
7268                 ipsa->ipsa_bpkt_head = ipsa->ipsa_bpkt_tail = bpkt;
7269         } else {
7270                 ipsa->ipsa_bpkt_tail->b_next = bpkt;
7271                 ipsa->ipsa_bpkt_tail = bpkt;
7272                 if (ipsa->ipsa_mblkcnt > SADB_MAX_IDLEPKTS) {
7273                         mblk_t *tmp;
7274 
7275                         tmp = ipsa->ipsa_bpkt_head;
7276                         ipsa->ipsa_bpkt_head = ipsa->ipsa_bpkt_head->b_next;
7277                         tmp = ip_recv_attr_free_mblk(tmp);
7278                         ip_drop_packet(tmp, B_TRUE, NULL,
7279                             DROPPER(ipss, ipds_sadb_inidle_overflow),
7280                             &ipss->ipsec_sadb_dropper);
7281                         ipsa->ipsa_mblkcnt --;
7282                 }
7283         }
7284         mutex_exit(&ipsa->ipsa_lock);
7285 }
7286 
7287 /*
7288  * Stub function that taskq_dispatch() invokes to take the mblk (in arg)
7289  * and put into STREAMS again.
7290  */
7291 void
7292 sadb_clear_buf_pkt(void *ipkt)
7293 {
7294         mblk_t  *tmp, *buf_pkt;
7295         ip_recv_attr_t  iras;
7296 
7297         buf_pkt = (mblk_t *)ipkt;
7298 
7299         while (buf_pkt != NULL) {
7300                 mblk_t *data_mp;
7301 
7302                 tmp = buf_pkt->b_next;
7303                 buf_pkt->b_next = NULL;
7304 
7305                 data_mp = buf_pkt->b_cont;
7306                 buf_pkt->b_cont = NULL;
7307                 if (!ip_recv_attr_from_mblk(buf_pkt, &iras)) {
7308                         /* The ill or ip_stack_t disappeared on us. */
7309                         ip_drop_input("ip_recv_attr_from_mblk", data_mp, NULL);
7310                         freemsg(data_mp);
7311                 } else {
7312                         ip_input_post_ipsec(data_mp, &iras);
7313                 }
7314                 ira_cleanup(&iras, B_TRUE);
7315                 buf_pkt = tmp;
7316         }
7317 }
7318 /*
7319  * Walker callback used by sadb_alg_update() to free/create crypto
7320  * context template when a crypto software provider is removed or
7321  * added.
7322  */
7323 
7324 struct sadb_update_alg_state {
7325         ipsec_algtype_t alg_type;
7326         uint8_t alg_id;
7327         boolean_t is_added;
7328         boolean_t async_auth;
7329         boolean_t async_encr;
7330 };
7331 
7332 static void
7333 sadb_alg_update_cb(isaf_t *head, ipsa_t *entry, void *cookie)
7334 {
7335         struct sadb_update_alg_state *update_state =
7336             (struct sadb_update_alg_state *)cookie;
7337         crypto_ctx_template_t *ctx_tmpl = NULL;
7338 
7339         ASSERT(MUTEX_HELD(&head->isaf_lock));
7340 
7341         if (entry->ipsa_state == IPSA_STATE_LARVAL)
7342                 return;
7343 
7344         mutex_enter(&entry->ipsa_lock);
7345 
7346         if ((entry->ipsa_encr_alg != SADB_EALG_NONE && entry->ipsa_encr_alg !=
7347             SADB_EALG_NULL && update_state->async_encr) ||
7348             (entry->ipsa_auth_alg != SADB_AALG_NONE &&
7349             update_state->async_auth)) {
7350                 entry->ipsa_flags |= IPSA_F_ASYNC;
7351         } else {
7352                 entry->ipsa_flags &= ~IPSA_F_ASYNC;
7353         }
7354 
7355         switch (update_state->alg_type) {
7356         case IPSEC_ALG_AUTH:
7357                 if (entry->ipsa_auth_alg == update_state->alg_id)
7358                         ctx_tmpl = &entry->ipsa_authtmpl;
7359                 break;
7360         case IPSEC_ALG_ENCR:
7361                 if (entry->ipsa_encr_alg == update_state->alg_id)
7362                         ctx_tmpl = &entry->ipsa_encrtmpl;
7363                 break;
7364         default:
7365                 ctx_tmpl = NULL;
7366         }
7367 
7368         if (ctx_tmpl == NULL) {
7369                 mutex_exit(&entry->ipsa_lock);
7370                 return;
7371         }
7372 
7373         /*
7374          * The context template of the SA may be affected by the change
7375          * of crypto provider.
7376          */
7377         if (update_state->is_added) {
7378                 /* create the context template if not already done */
7379                 if (*ctx_tmpl == NULL) {
7380                         (void) ipsec_create_ctx_tmpl(entry,
7381                             update_state->alg_type);
7382                 }
7383         } else {
7384                 /*
7385                  * The crypto provider was removed. If the context template
7386                  * exists but it is no longer valid, free it.
7387                  */
7388                 if (*ctx_tmpl != NULL)
7389                         ipsec_destroy_ctx_tmpl(entry, update_state->alg_type);
7390         }
7391 
7392         mutex_exit(&entry->ipsa_lock);
7393 }
7394 
7395 /*
7396  * Invoked by IP when an software crypto provider has been updated, or if
7397  * the crypto synchrony changes.  The type and id of the corresponding
7398  * algorithm is passed as argument.  The type is set to ALL in the case of
7399  * a synchrony change.
7400  *
7401  * is_added is B_TRUE if the provider was added, B_FALSE if it was
7402  * removed. The function updates the SADB and free/creates the
7403  * context templates associated with SAs if needed.
7404  */
7405 
7406 #define SADB_ALG_UPDATE_WALK(sadb, table) \
7407     sadb_walker((sadb).table, (sadb).sdb_hashsize, sadb_alg_update_cb, \
7408         &update_state)
7409 
7410 void
7411 sadb_alg_update(ipsec_algtype_t alg_type, uint8_t alg_id, boolean_t is_added,
7412     netstack_t *ns)
7413 {
7414         struct sadb_update_alg_state update_state;
7415         ipsecah_stack_t *ahstack = ns->netstack_ipsecah;
7416         ipsecesp_stack_t        *espstack = ns->netstack_ipsecesp;
7417         ipsec_stack_t *ipss = ns->netstack_ipsec;
7418 
7419         update_state.alg_type = alg_type;
7420         update_state.alg_id = alg_id;
7421         update_state.is_added = is_added;
7422         update_state.async_auth = ipss->ipsec_algs_exec_mode[IPSEC_ALG_AUTH] ==
7423             IPSEC_ALGS_EXEC_ASYNC;
7424         update_state.async_encr = ipss->ipsec_algs_exec_mode[IPSEC_ALG_ENCR] ==
7425             IPSEC_ALGS_EXEC_ASYNC;
7426 
7427         if (alg_type == IPSEC_ALG_AUTH || alg_type == IPSEC_ALG_ALL) {
7428                 /* walk the AH tables only for auth. algorithm changes */
7429                 SADB_ALG_UPDATE_WALK(ahstack->ah_sadb.s_v4, sdb_of);
7430                 SADB_ALG_UPDATE_WALK(ahstack->ah_sadb.s_v4, sdb_if);
7431                 SADB_ALG_UPDATE_WALK(ahstack->ah_sadb.s_v6, sdb_of);
7432                 SADB_ALG_UPDATE_WALK(ahstack->ah_sadb.s_v6, sdb_if);
7433         }
7434 
7435         /* walk the ESP tables */
7436         SADB_ALG_UPDATE_WALK(espstack->esp_sadb.s_v4, sdb_of);
7437         SADB_ALG_UPDATE_WALK(espstack->esp_sadb.s_v4, sdb_if);
7438         SADB_ALG_UPDATE_WALK(espstack->esp_sadb.s_v6, sdb_of);
7439         SADB_ALG_UPDATE_WALK(espstack->esp_sadb.s_v6, sdb_if);
7440 }
7441 
7442 /*
7443  * Creates a context template for the specified SA. This function
7444  * is called when an SA is created and when a context template needs
7445  * to be created due to a change of software provider.
7446  */
7447 int
7448 ipsec_create_ctx_tmpl(ipsa_t *sa, ipsec_algtype_t alg_type)
7449 {
7450         ipsec_alginfo_t *alg;
7451         crypto_mechanism_t mech;
7452         crypto_key_t *key;
7453         crypto_ctx_template_t *sa_tmpl;
7454         int rv;
7455         ipsec_stack_t   *ipss = sa->ipsa_netstack->netstack_ipsec;
7456 
7457         ASSERT(RW_READ_HELD(&ipss->ipsec_alg_lock));
7458         ASSERT(MUTEX_HELD(&sa->ipsa_lock));
7459 
7460         /* get pointers to the algorithm info, context template, and key */
7461         switch (alg_type) {
7462         case IPSEC_ALG_AUTH:
7463                 key = &sa->ipsa_kcfauthkey;
7464                 sa_tmpl = &sa->ipsa_authtmpl;
7465                 alg = ipss->ipsec_alglists[alg_type][sa->ipsa_auth_alg];
7466                 break;
7467         case IPSEC_ALG_ENCR:
7468                 key = &sa->ipsa_kcfencrkey;
7469                 sa_tmpl = &sa->ipsa_encrtmpl;
7470                 alg = ipss->ipsec_alglists[alg_type][sa->ipsa_encr_alg];
7471                 break;
7472         default:
7473                 alg = NULL;
7474         }
7475 
7476         if (alg == NULL || !ALG_VALID(alg))
7477                 return (EINVAL);
7478 
7479         /* initialize the mech info structure for the framework */
7480         ASSERT(alg->alg_mech_type != CRYPTO_MECHANISM_INVALID);
7481         mech.cm_type = alg->alg_mech_type;
7482         mech.cm_param = NULL;
7483         mech.cm_param_len = 0;
7484 
7485         /* create a new context template */
7486         rv = crypto_create_ctx_template(&mech, key, sa_tmpl, KM_NOSLEEP);
7487 
7488         /*
7489          * CRYPTO_MECH_NOT_SUPPORTED can be returned if only hardware
7490          * providers are available for that mechanism. In that case
7491          * we don't fail, and will generate the context template from
7492          * the framework callback when a software provider for that
7493          * mechanism registers.
7494          *
7495          * The context template is assigned the special value
7496          * IPSEC_CTX_TMPL_ALLOC if the allocation failed due to a
7497          * lack of memory. No attempt will be made to use
7498          * the context template if it is set to this value.
7499          */
7500         if (rv == CRYPTO_HOST_MEMORY) {
7501                 *sa_tmpl = IPSEC_CTX_TMPL_ALLOC;
7502         } else if (rv != CRYPTO_SUCCESS) {
7503                 *sa_tmpl = NULL;
7504                 if (rv != CRYPTO_MECH_NOT_SUPPORTED)
7505                         return (EINVAL);
7506         }
7507 
7508         return (0);
7509 }
7510 
7511 /*
7512  * Destroy the context template of the specified algorithm type
7513  * of the specified SA. Must be called while holding the SA lock.
7514  */
7515 void
7516 ipsec_destroy_ctx_tmpl(ipsa_t *sa, ipsec_algtype_t alg_type)
7517 {
7518         ASSERT(MUTEX_HELD(&sa->ipsa_lock));
7519 
7520         if (alg_type == IPSEC_ALG_AUTH) {
7521                 if (sa->ipsa_authtmpl == IPSEC_CTX_TMPL_ALLOC)
7522                         sa->ipsa_authtmpl = NULL;
7523                 else if (sa->ipsa_authtmpl != NULL) {
7524                         crypto_destroy_ctx_template(sa->ipsa_authtmpl);
7525                         sa->ipsa_authtmpl = NULL;
7526                 }
7527         } else {
7528                 ASSERT(alg_type == IPSEC_ALG_ENCR);
7529                 if (sa->ipsa_encrtmpl == IPSEC_CTX_TMPL_ALLOC)
7530                         sa->ipsa_encrtmpl = NULL;
7531                 else if (sa->ipsa_encrtmpl != NULL) {
7532                         crypto_destroy_ctx_template(sa->ipsa_encrtmpl);
7533                         sa->ipsa_encrtmpl = NULL;
7534                 }
7535         }
7536 }
7537 
7538 /*
7539  * Use the kernel crypto framework to check the validity of a key received
7540  * via keysock. Returns 0 if the key is OK, -1 otherwise.
7541  */
7542 int
7543 ipsec_check_key(crypto_mech_type_t mech_type, sadb_key_t *sadb_key,
7544     boolean_t is_auth, int *diag)
7545 {
7546         crypto_mechanism_t mech;
7547         crypto_key_t crypto_key;
7548         int crypto_rc;
7549 
7550         mech.cm_type = mech_type;
7551         mech.cm_param = NULL;
7552         mech.cm_param_len = 0;
7553 
7554         crypto_key.ck_format = CRYPTO_KEY_RAW;
7555         crypto_key.ck_data = sadb_key + 1;
7556         crypto_key.ck_length = sadb_key->sadb_key_bits;
7557 
7558         crypto_rc = crypto_key_check(&mech, &crypto_key);
7559 
7560         switch (crypto_rc) {
7561         case CRYPTO_SUCCESS:
7562                 return (0);
7563         case CRYPTO_MECHANISM_INVALID:
7564         case CRYPTO_MECH_NOT_SUPPORTED:
7565                 *diag = is_auth ? SADB_X_DIAGNOSTIC_BAD_AALG :
7566                     SADB_X_DIAGNOSTIC_BAD_EALG;
7567                 break;
7568         case CRYPTO_KEY_SIZE_RANGE:
7569                 *diag = is_auth ? SADB_X_DIAGNOSTIC_BAD_AKEYBITS :
7570                     SADB_X_DIAGNOSTIC_BAD_EKEYBITS;
7571                 break;
7572         case CRYPTO_WEAK_KEY:
7573                 *diag = is_auth ? SADB_X_DIAGNOSTIC_WEAK_AKEY :
7574                     SADB_X_DIAGNOSTIC_WEAK_EKEY;
7575                 break;
7576         }
7577 
7578         return (-1);
7579 }
7580 
7581 /*
7582  * Whack options in the outer IP header when ipsec changes the outer label
7583  *
7584  * This is inelegant and really could use refactoring.
7585  */
7586 mblk_t *
7587 sadb_whack_label_v4(mblk_t *mp, ipsa_t *assoc, kstat_named_t *counter,
7588     ipdropper_t *dropper)
7589 {
7590         int delta;
7591         int plen;
7592         dblk_t *db;
7593         int hlen;
7594         uint8_t *opt_storage = assoc->ipsa_opt_storage;
7595         ipha_t *ipha = (ipha_t *)mp->b_rptr;
7596 
7597         plen = ntohs(ipha->ipha_length);
7598 
7599         delta = tsol_remove_secopt(ipha, MBLKL(mp));
7600         mp->b_wptr += delta;
7601         plen += delta;
7602 
7603         /* XXX XXX code copied from tsol_check_label */
7604 
7605         /* Make sure we have room for the worst-case addition */
7606         hlen = IPH_HDR_LENGTH(ipha) + opt_storage[IPOPT_OLEN];
7607         hlen = (hlen + 3) & ~3;
7608         if (hlen > IP_MAX_HDR_LENGTH)
7609                 hlen = IP_MAX_HDR_LENGTH;
7610         hlen -= IPH_HDR_LENGTH(ipha);
7611 
7612         db = mp->b_datap;
7613         if ((db->db_ref != 1) || (mp->b_wptr + hlen > db->db_lim)) {
7614                 int copylen;
7615                 mblk_t *new_mp;
7616 
7617                 /* allocate enough to be meaningful, but not *too* much */
7618                 copylen = MBLKL(mp);
7619                 if (copylen > 256)
7620                         copylen = 256;
7621                 new_mp = allocb_tmpl(hlen + copylen +
7622                     (mp->b_rptr - mp->b_datap->db_base), mp);
7623 
7624                 if (new_mp == NULL) {
7625                         ip_drop_packet(mp, B_FALSE, NULL, counter,  dropper);
7626                         return (NULL);
7627                 }
7628 
7629                 /* keep the bias */
7630                 new_mp->b_rptr += mp->b_rptr - mp->b_datap->db_base;
7631                 new_mp->b_wptr = new_mp->b_rptr + copylen;
7632                 bcopy(mp->b_rptr, new_mp->b_rptr, copylen);
7633                 new_mp->b_cont = mp;
7634                 if ((mp->b_rptr += copylen) >= mp->b_wptr) {
7635                         new_mp->b_cont = mp->b_cont;
7636                         freeb(mp);
7637                 }
7638                 mp = new_mp;
7639                 ipha = (ipha_t *)mp->b_rptr;
7640         }
7641 
7642         delta = tsol_prepend_option(assoc->ipsa_opt_storage, ipha, MBLKL(mp));
7643 
7644         ASSERT(delta != -1);
7645 
7646         plen += delta;
7647         mp->b_wptr += delta;
7648 
7649         /*
7650          * Paranoia
7651          */
7652         db = mp->b_datap;
7653 
7654         ASSERT3P(mp->b_wptr, <=, db->db_lim);
7655         ASSERT3P(mp->b_rptr, <=, db->db_lim);
7656 
7657         ASSERT3P(mp->b_wptr, >=, db->db_base);
7658         ASSERT3P(mp->b_rptr, >=, db->db_base);
7659         /* End paranoia */
7660 
7661         ipha->ipha_length = htons(plen);
7662 
7663         return (mp);
7664 }
7665 
7666 mblk_t *
7667 sadb_whack_label_v6(mblk_t *mp, ipsa_t *assoc, kstat_named_t *counter,
7668     ipdropper_t *dropper)
7669 {
7670         int delta;
7671         int plen;
7672         dblk_t *db;
7673         int hlen;
7674         uint8_t *opt_storage = assoc->ipsa_opt_storage;
7675         uint_t sec_opt_len; /* label option length not including type, len */
7676         ip6_t *ip6h = (ip6_t *)mp->b_rptr;
7677 
7678         plen = ntohs(ip6h->ip6_plen);
7679 
7680         delta = tsol_remove_secopt_v6(ip6h, MBLKL(mp));
7681         mp->b_wptr += delta;
7682         plen += delta;
7683 
7684         /* XXX XXX code copied from tsol_check_label_v6 */
7685         /*
7686          * Make sure we have room for the worst-case addition. Add 2 bytes for
7687          * the hop-by-hop ext header's next header and length fields. Add
7688          * another 2 bytes for the label option type, len and then round
7689          * up to the next 8-byte multiple.
7690          */
7691         sec_opt_len = opt_storage[1];
7692 
7693         db = mp->b_datap;
7694         hlen = (4 + sec_opt_len + 7) & ~7;
7695 
7696         if ((db->db_ref != 1) || (mp->b_wptr + hlen > db->db_lim)) {
7697                 int copylen;
7698                 mblk_t *new_mp;
7699                 uint16_t hdr_len;
7700 
7701                 hdr_len = ip_hdr_length_v6(mp, ip6h);
7702                 /*
7703                  * Allocate enough to be meaningful, but not *too* much.
7704                  * Also all the IPv6 extension headers must be in the same mblk
7705                  */
7706                 copylen = MBLKL(mp);
7707                 if (copylen > 256)
7708                         copylen = 256;
7709                 if (copylen < hdr_len)
7710                         copylen = hdr_len;
7711                 new_mp = allocb_tmpl(hlen + copylen +
7712                     (mp->b_rptr - mp->b_datap->db_base), mp);
7713                 if (new_mp == NULL) {
7714                         ip_drop_packet(mp, B_FALSE, NULL, counter,  dropper);
7715                         return (NULL);
7716                 }
7717 
7718                 /* keep the bias */
7719                 new_mp->b_rptr += mp->b_rptr - mp->b_datap->db_base;
7720                 new_mp->b_wptr = new_mp->b_rptr + copylen;
7721                 bcopy(mp->b_rptr, new_mp->b_rptr, copylen);
7722                 new_mp->b_cont = mp;
7723                 if ((mp->b_rptr += copylen) >= mp->b_wptr) {
7724                         new_mp->b_cont = mp->b_cont;
7725                         freeb(mp);
7726                 }
7727                 mp = new_mp;
7728                 ip6h = (ip6_t *)mp->b_rptr;
7729         }
7730 
7731         delta = tsol_prepend_option_v6(assoc->ipsa_opt_storage,
7732             ip6h, MBLKL(mp));
7733 
7734         ASSERT(delta != -1);
7735 
7736         plen += delta;
7737         mp->b_wptr += delta;
7738 
7739         /*
7740          * Paranoia
7741          */
7742         db = mp->b_datap;
7743 
7744         ASSERT3P(mp->b_wptr, <=, db->db_lim);
7745         ASSERT3P(mp->b_rptr, <=, db->db_lim);
7746 
7747         ASSERT3P(mp->b_wptr, >=, db->db_base);
7748         ASSERT3P(mp->b_rptr, >=, db->db_base);
7749         /* End paranoia */
7750 
7751         ip6h->ip6_plen = htons(plen);
7752 
7753         return (mp);
7754 }
7755 
7756 /* Whack the labels and update ip_xmit_attr_t as needed */
7757 mblk_t *
7758 sadb_whack_label(mblk_t *mp, ipsa_t *assoc, ip_xmit_attr_t *ixa,
7759     kstat_named_t *counter, ipdropper_t *dropper)
7760 {
7761         int adjust;
7762         int iplen;
7763 
7764         if (ixa->ixa_flags & IXAF_IS_IPV4) {
7765                 ipha_t          *ipha = (ipha_t *)mp->b_rptr;
7766 
7767                 ASSERT(IPH_HDR_VERSION(ipha) == IPV4_VERSION);
7768                 iplen = ntohs(ipha->ipha_length);
7769                 mp = sadb_whack_label_v4(mp, assoc, counter, dropper);
7770                 if (mp == NULL)
7771                         return (NULL);
7772 
7773                 ipha = (ipha_t *)mp->b_rptr;
7774                 ASSERT(IPH_HDR_VERSION(ipha) == IPV4_VERSION);
7775                 adjust = (int)ntohs(ipha->ipha_length) - iplen;
7776         } else {
7777                 ip6_t           *ip6h = (ip6_t *)mp->b_rptr;
7778 
7779                 ASSERT(IPH_HDR_VERSION(ip6h) == IPV6_VERSION);
7780                 iplen = ntohs(ip6h->ip6_plen);
7781                 mp = sadb_whack_label_v6(mp, assoc, counter, dropper);
7782                 if (mp == NULL)
7783                         return (NULL);
7784 
7785                 ip6h = (ip6_t *)mp->b_rptr;
7786                 ASSERT(IPH_HDR_VERSION(ip6h) == IPV6_VERSION);
7787                 adjust = (int)ntohs(ip6h->ip6_plen) - iplen;
7788         }
7789         ixa->ixa_pktlen += adjust;
7790         ixa->ixa_ip_hdr_length += adjust;
7791         return (mp);
7792 }
7793 
7794 /*
7795  * If this is an outgoing SA then add some fuzz to the
7796  * SOFT EXPIRE time. The reason for this is to stop
7797  * peers trying to renegotiate SOFT expiring SA's at
7798  * the same time. The amount of fuzz needs to be at
7799  * least 8 seconds which is the typical interval
7800  * sadb_ager(), although this is only a guide as it
7801  * selftunes.
7802  */
7803 static void
7804 lifetime_fuzz(ipsa_t *assoc)
7805 {
7806         uint8_t rnd;
7807 
7808         if (assoc->ipsa_softaddlt == 0)
7809                 return;
7810 
7811         (void) random_get_pseudo_bytes(&rnd, sizeof (rnd));
7812         rnd = (rnd & 0xF) + 8;
7813         assoc->ipsa_softexpiretime -= rnd;
7814         assoc->ipsa_softaddlt -= rnd;
7815 }
7816 
7817 static void
7818 destroy_ipsa_pair(ipsap_t *ipsapp)
7819 {
7820         /*
7821          * Because of the multi-line macro nature of IPSA_REFRELE, keep
7822          * them in { }.
7823          */
7824         if (ipsapp->ipsap_sa_ptr != NULL) {
7825                 IPSA_REFRELE(ipsapp->ipsap_sa_ptr);
7826         }
7827         if (ipsapp->ipsap_psa_ptr != NULL) {
7828                 IPSA_REFRELE(ipsapp->ipsap_psa_ptr);
7829         }
7830         init_ipsa_pair(ipsapp);
7831 }
7832 
7833 static void
7834 init_ipsa_pair(ipsap_t *ipsapp)
7835 {
7836         ipsapp->ipsap_bucket = NULL;
7837         ipsapp->ipsap_sa_ptr = NULL;
7838         ipsapp->ipsap_pbucket = NULL;
7839         ipsapp->ipsap_psa_ptr = NULL;
7840 }
7841 
7842 /*
7843  * The sadb_ager() function walks through the hash tables of SA's and ages
7844  * them, if the SA expires as a result, its marked as DEAD and will be reaped
7845  * the next time sadb_ager() runs. SA's which are paired or have a peer (same
7846  * SA appears in both the inbound and outbound tables because its not possible
7847  * to determine its direction) are placed on a list when they expire. This is
7848  * to ensure that pair/peer SA's are reaped at the same time, even if they
7849  * expire at different times.
7850  *
7851  * This function is called twice by sadb_ager(), one after processing the
7852  * inbound table, then again after processing the outbound table.
7853  */
7854 void
7855 age_pair_peer_list(templist_t *haspeerlist, sadb_t *sp, boolean_t outbound)
7856 {
7857         templist_t *listptr;
7858         int outhash;
7859         isaf_t *bucket;
7860         boolean_t haspeer;
7861         ipsa_t *peer_assoc, *dying;
7862         /*
7863          * Haspeer cases will contain both IPv4 and IPv6.  This code
7864          * is address independent.
7865          */
7866         while (haspeerlist != NULL) {
7867                 /* "dying" contains the SA that has a peer. */
7868                 dying = haspeerlist->ipsa;
7869                 haspeer = (dying->ipsa_haspeer);
7870                 listptr = haspeerlist;
7871                 haspeerlist = listptr->next;
7872                 kmem_free(listptr, sizeof (*listptr));
7873                 /*
7874                  * Pick peer bucket based on addrfam.
7875                  */
7876                 if (outbound) {
7877                         if (haspeer)
7878                                 bucket = INBOUND_BUCKET(sp, dying->ipsa_spi);
7879                         else
7880                                 bucket = INBOUND_BUCKET(sp,
7881                                     dying->ipsa_otherspi);
7882                 } else { /* inbound */
7883                         if (haspeer) {
7884                                 if (dying->ipsa_addrfam == AF_INET6) {
7885                                         outhash = OUTBOUND_HASH_V6(sp,
7886                                             *((in6_addr_t *)&dying->
7887                                             ipsa_dstaddr));
7888                                 } else {
7889                                         outhash = OUTBOUND_HASH_V4(sp,
7890                                             *((ipaddr_t *)&dying->
7891                                             ipsa_dstaddr));
7892                                 }
7893                         } else if (dying->ipsa_addrfam == AF_INET6) {
7894                                 outhash = OUTBOUND_HASH_V6(sp,
7895                                     *((in6_addr_t *)&dying->
7896                                     ipsa_srcaddr));
7897                         } else {
7898                                 outhash = OUTBOUND_HASH_V4(sp,
7899                                     *((ipaddr_t *)&dying->
7900                                     ipsa_srcaddr));
7901                         }
7902                         bucket = &(sp->sdb_of[outhash]);
7903                 }
7904 
7905                 mutex_enter(&bucket->isaf_lock);
7906                 /*
7907                  * "haspeer" SA's have the same src/dst address ordering,
7908                  * "paired" SA's have the src/dst addresses reversed.
7909                  */
7910                 if (haspeer) {
7911                         peer_assoc = ipsec_getassocbyspi(bucket,
7912                             dying->ipsa_spi, dying->ipsa_srcaddr,
7913                             dying->ipsa_dstaddr, dying->ipsa_addrfam);
7914                 } else {
7915                         peer_assoc = ipsec_getassocbyspi(bucket,
7916                             dying->ipsa_otherspi, dying->ipsa_dstaddr,
7917                             dying->ipsa_srcaddr, dying->ipsa_addrfam);
7918                 }
7919 
7920                 mutex_exit(&bucket->isaf_lock);
7921                 if (peer_assoc != NULL) {
7922                         mutex_enter(&peer_assoc->ipsa_lock);
7923                         mutex_enter(&dying->ipsa_lock);
7924                         if (!haspeer) {
7925                                 /*
7926                                  * Only SA's which have a "peer" or are
7927                                  * "paired" end up on this list, so this
7928                                  * must be a "paired" SA, update the flags
7929                                  * to break the pair.
7930                                  */
7931                                 peer_assoc->ipsa_otherspi = 0;
7932                                 peer_assoc->ipsa_flags &= ~IPSA_F_PAIRED;
7933                                 dying->ipsa_otherspi = 0;
7934                                 dying->ipsa_flags &= ~IPSA_F_PAIRED;
7935                         }
7936                         if (haspeer || outbound) {
7937                                 /*
7938                                  * Update the state of the "inbound" SA when
7939                                  * the "outbound" SA has expired. Don't update
7940                                  * the "outbound" SA when the "inbound" SA
7941                                  * SA expires because setting the hard_addtime
7942                                  * below will cause this to happen.
7943                                  */
7944                                 peer_assoc->ipsa_state = dying->ipsa_state;
7945                         }
7946                         if (dying->ipsa_state == IPSA_STATE_DEAD)
7947                                 peer_assoc->ipsa_hardexpiretime = 1;
7948 
7949                         mutex_exit(&dying->ipsa_lock);
7950                         mutex_exit(&peer_assoc->ipsa_lock);
7951                         IPSA_REFRELE(peer_assoc);
7952                 }
7953                 IPSA_REFRELE(dying);
7954         }
7955 }
7956 
7957 /*
7958  * Ensure that the IV used for CCM mode never repeats. The IV should
7959  * only be updated by this function. Also check to see if the IV
7960  * is about to wrap and generate a SOFT Expire. This function is only
7961  * called for outgoing packets, the IV for incomming packets is taken
7962  * from the wire. If the outgoing SA needs to be expired, update
7963  * the matching incomming SA.
7964  */
7965 boolean_t
7966 update_iv(uint8_t *iv_ptr, queue_t *pfkey_q, ipsa_t *assoc,
7967     ipsecesp_stack_t *espstack)
7968 {
7969         boolean_t rc = B_TRUE;
7970         isaf_t *inbound_bucket;
7971         sadb_t *sp;
7972         ipsa_t *pair_sa = NULL;
7973         int sa_new_state = 0;
7974 
7975         /* For non counter modes, the IV is random data. */
7976         if (!(assoc->ipsa_flags & IPSA_F_COUNTERMODE)) {
7977                 (void) random_get_pseudo_bytes(iv_ptr, assoc->ipsa_iv_len);
7978                 return (rc);
7979         }
7980 
7981         mutex_enter(&assoc->ipsa_lock);
7982 
7983         (*assoc->ipsa_iv)++;
7984 
7985         if (*assoc->ipsa_iv == assoc->ipsa_iv_hardexpire) {
7986                 sa_new_state = IPSA_STATE_DEAD;
7987                 rc = B_FALSE;
7988         } else if (*assoc->ipsa_iv == assoc->ipsa_iv_softexpire) {
7989                 if (assoc->ipsa_state != IPSA_STATE_DYING) {
7990                         /*
7991                          * This SA may have already been expired when its
7992                          * PAIR_SA expired.
7993                          */
7994                         sa_new_state = IPSA_STATE_DYING;
7995                 }
7996         }
7997         if (sa_new_state) {
7998                 /*
7999                  * If there is a state change, we need to update this SA
8000                  * and its "pair", we can find the bucket for the "pair" SA
8001                  * while holding the ipsa_t mutex, but we won't actually
8002                  * update anything untill the ipsa_t mutex has been released
8003                  * for _this_ SA.
8004                  */
8005                 assoc->ipsa_state = sa_new_state;
8006                 if (assoc->ipsa_addrfam == AF_INET6) {
8007                         sp = &espstack->esp_sadb.s_v6;
8008                 } else {
8009                         sp = &espstack->esp_sadb.s_v4;
8010                 }
8011                 inbound_bucket = INBOUND_BUCKET(sp, assoc->ipsa_otherspi);
8012                 sadb_expire_assoc(pfkey_q, assoc);
8013         }
8014         if (rc == B_TRUE)
8015                 bcopy(assoc->ipsa_iv, iv_ptr, assoc->ipsa_iv_len);
8016 
8017         mutex_exit(&assoc->ipsa_lock);
8018 
8019         if (sa_new_state) {
8020                 /* Find the inbound SA, need to lock hash bucket. */
8021                 mutex_enter(&inbound_bucket->isaf_lock);
8022                 pair_sa = ipsec_getassocbyspi(inbound_bucket,
8023                     assoc->ipsa_otherspi, assoc->ipsa_dstaddr,
8024                     assoc->ipsa_srcaddr, assoc->ipsa_addrfam);
8025                 mutex_exit(&inbound_bucket->isaf_lock);
8026                 if (pair_sa != NULL) {
8027                         mutex_enter(&pair_sa->ipsa_lock);
8028                         pair_sa->ipsa_state = sa_new_state;
8029                         mutex_exit(&pair_sa->ipsa_lock);
8030                         IPSA_REFRELE(pair_sa);
8031                 }
8032         }
8033 
8034         return (rc);
8035 }
8036 
8037 void
8038 ccm_params_init(ipsa_t *assoc, uchar_t *esph, uint_t data_len, uchar_t *iv_ptr,
8039     ipsa_cm_mech_t *cm_mech, crypto_data_t *crypto_data)
8040 {
8041         uchar_t *nonce;
8042         crypto_mechanism_t *combined_mech;
8043         CK_AES_CCM_PARAMS *params;
8044 
8045         combined_mech = (crypto_mechanism_t *)cm_mech;
8046         params = (CK_AES_CCM_PARAMS *)(combined_mech + 1);
8047         nonce = (uchar_t *)(params + 1);
8048         params->ulMACSize = assoc->ipsa_mac_len;
8049         params->ulNonceSize = assoc->ipsa_nonce_len;
8050         params->ulAuthDataSize = sizeof (esph_t);
8051         params->ulDataSize = data_len;
8052         params->nonce = nonce;
8053         params->authData = esph;
8054 
8055         cm_mech->combined_mech.cm_type = assoc->ipsa_emech.cm_type;
8056         cm_mech->combined_mech.cm_param_len = sizeof (CK_AES_CCM_PARAMS);
8057         cm_mech->combined_mech.cm_param = (caddr_t)params;
8058         /* See gcm_params_init() for comments. */
8059         bcopy(assoc->ipsa_nonce, nonce, assoc->ipsa_saltlen);
8060         nonce += assoc->ipsa_saltlen;
8061         bcopy(iv_ptr, nonce, assoc->ipsa_iv_len);
8062         crypto_data->cd_miscdata = NULL;
8063 }
8064 
8065 /* ARGSUSED */
8066 void
8067 cbc_params_init(ipsa_t *assoc, uchar_t *esph, uint_t data_len, uchar_t *iv_ptr,
8068     ipsa_cm_mech_t *cm_mech, crypto_data_t *crypto_data)
8069 {
8070         cm_mech->combined_mech.cm_type = assoc->ipsa_emech.cm_type;
8071         cm_mech->combined_mech.cm_param_len = 0;
8072         cm_mech->combined_mech.cm_param = NULL;
8073         crypto_data->cd_miscdata = (char *)iv_ptr;
8074 }
8075 
8076 /* ARGSUSED */
8077 void
8078 gcm_params_init(ipsa_t *assoc, uchar_t *esph, uint_t data_len, uchar_t *iv_ptr,
8079     ipsa_cm_mech_t *cm_mech, crypto_data_t *crypto_data)
8080 {
8081         uchar_t *nonce;
8082         crypto_mechanism_t *combined_mech;
8083         CK_AES_GCM_PARAMS *params;
8084 
8085         combined_mech = (crypto_mechanism_t *)cm_mech;
8086         params = (CK_AES_GCM_PARAMS *)(combined_mech + 1);
8087         nonce = (uchar_t *)(params + 1);
8088 
8089         params->pIv = nonce;
8090         params->ulIvLen = assoc->ipsa_nonce_len;
8091         params->ulIvBits = SADB_8TO1(assoc->ipsa_nonce_len);
8092         params->pAAD = esph;
8093         params->ulAADLen = sizeof (esph_t);
8094         params->ulTagBits = SADB_8TO1(assoc->ipsa_mac_len);
8095 
8096         cm_mech->combined_mech.cm_type = assoc->ipsa_emech.cm_type;
8097         cm_mech->combined_mech.cm_param_len = sizeof (CK_AES_GCM_PARAMS);
8098         cm_mech->combined_mech.cm_param = (caddr_t)params;
8099         /*
8100          * Create the nonce, which is made up of the salt and the IV.
8101          * Copy the salt from the SA and the IV from the packet.
8102          * For inbound packets we copy the IV from the packet because it
8103          * was set by the sending system, for outbound packets we copy the IV
8104          * from the packet because the IV in the SA may be changed by another
8105          * thread, the IV in the packet was created while holding a mutex.
8106          */
8107         bcopy(assoc->ipsa_nonce, nonce, assoc->ipsa_saltlen);
8108         nonce += assoc->ipsa_saltlen;
8109         bcopy(iv_ptr, nonce, assoc->ipsa_iv_len);
8110         crypto_data->cd_miscdata = NULL;
8111 }