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 2008 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 #pragma ident "%Z%%M% %I% %E% SMI" 27 28 #include <sys/types.h> 29 #include <sys/stream.h> 30 #include <sys/stropts.h> 31 #include <sys/errno.h> 32 #include <sys/strlog.h> 33 #include <sys/tihdr.h> 34 #include <sys/socket.h> 35 #include <sys/ddi.h> 36 #include <sys/sunddi.h> 37 #include <sys/kmem.h> 38 #include <sys/zone.h> 39 #include <sys/sysmacros.h> 40 #include <sys/cmn_err.h> 41 #include <sys/vtrace.h> 42 #include <sys/debug.h> 43 #include <sys/atomic.h> 44 #include <sys/strsun.h> 45 #include <sys/random.h> 46 #include <netinet/in.h> 47 #include <net/if.h> 48 #include <netinet/ip6.h> 49 #include <net/pfkeyv2.h> 50 51 #include <inet/common.h> 52 #include <inet/mi.h> 53 #include <inet/nd.h> 54 #include <inet/ip.h> 55 #include <inet/ip_impl.h> 56 #include <inet/ip6.h> 57 #include <inet/sadb.h> 58 #include <inet/ipsec_info.h> 59 #include <inet/ipsec_impl.h> 60 #include <inet/ipsecesp.h> 61 #include <inet/ipdrop.h> 62 #include <inet/tcp.h> 63 #include <sys/kstat.h> 64 #include <sys/policy.h> 65 #include <sys/strsun.h> 66 #include <inet/udp_impl.h> 67 #include <sys/taskq.h> 68 #include <sys/note.h> 69 70 #include <sys/iphada.h> 71 72 /* 73 * Table of ND variables supported by ipsecesp. These are loaded into 74 * ipsecesp_g_nd in ipsecesp_init_nd. 75 * All of these are alterable, within the min/max values given, at run time. 76 */ 77 static ipsecespparam_t lcl_param_arr[] = { 78 /* min max value name */ 79 { 0, 3, 0, "ipsecesp_debug"}, 80 { 125, 32000, SADB_AGE_INTERVAL_DEFAULT, "ipsecesp_age_interval"}, 81 { 1, 10, 1, "ipsecesp_reap_delay"}, 82 { 1, SADB_MAX_REPLAY, 64, "ipsecesp_replay_size"}, 83 { 1, 300, 15, "ipsecesp_acquire_timeout"}, 84 { 1, 1800, 90, "ipsecesp_larval_timeout"}, 85 /* Default lifetime values for ACQUIRE messages. */ 86 { 0, 0xffffffffU, 0, "ipsecesp_default_soft_bytes"}, 87 { 0, 0xffffffffU, 0, "ipsecesp_default_hard_bytes"}, 88 { 0, 0xffffffffU, 24000, "ipsecesp_default_soft_addtime"}, 89 { 0, 0xffffffffU, 28800, "ipsecesp_default_hard_addtime"}, 90 { 0, 0xffffffffU, 0, "ipsecesp_default_soft_usetime"}, 91 { 0, 0xffffffffU, 0, "ipsecesp_default_hard_usetime"}, 92 { 0, 1, 0, "ipsecesp_log_unknown_spi"}, 93 { 0, 2, 1, "ipsecesp_padding_check"}, 94 { 0, 600, 20, "ipsecesp_nat_keepalive_interval"}, 95 }; 96 #define ipsecesp_debug ipsecesp_params[0].ipsecesp_param_value 97 #define ipsecesp_age_interval ipsecesp_params[1].ipsecesp_param_value 98 #define ipsecesp_age_int_max ipsecesp_params[1].ipsecesp_param_max 99 #define ipsecesp_reap_delay ipsecesp_params[2].ipsecesp_param_value 100 #define ipsecesp_replay_size ipsecesp_params[3].ipsecesp_param_value 101 #define ipsecesp_acquire_timeout \ 102 ipsecesp_params[4].ipsecesp_param_value 103 #define ipsecesp_larval_timeout \ 104 ipsecesp_params[5].ipsecesp_param_value 105 #define ipsecesp_default_soft_bytes \ 106 ipsecesp_params[6].ipsecesp_param_value 107 #define ipsecesp_default_hard_bytes \ 108 ipsecesp_params[7].ipsecesp_param_value 109 #define ipsecesp_default_soft_addtime \ 110 ipsecesp_params[8].ipsecesp_param_value 111 #define ipsecesp_default_hard_addtime \ 112 ipsecesp_params[9].ipsecesp_param_value 113 #define ipsecesp_default_soft_usetime \ 114 ipsecesp_params[10].ipsecesp_param_value 115 #define ipsecesp_default_hard_usetime \ 116 ipsecesp_params[11].ipsecesp_param_value 117 #define ipsecesp_log_unknown_spi \ 118 ipsecesp_params[12].ipsecesp_param_value 119 #define ipsecesp_padding_check \ 120 ipsecesp_params[13].ipsecesp_param_value 121 /* For ipsecesp_nat_keepalive_interval, see ipsecesp.h. */ 122 123 #define esp0dbg(a) printf a 124 /* NOTE: != 0 instead of > 0 so lint doesn't complain. */ 125 #define esp1dbg(espstack, a) if (espstack->ipsecesp_debug != 0) printf a 126 #define esp2dbg(espstack, a) if (espstack->ipsecesp_debug > 1) printf a 127 #define esp3dbg(espstack, a) if (espstack->ipsecesp_debug > 2) printf a 128 129 static int ipsecesp_open(queue_t *, dev_t *, int, int, cred_t *); 130 static int ipsecesp_close(queue_t *); 131 static void ipsecesp_rput(queue_t *, mblk_t *); 132 static void ipsecesp_wput(queue_t *, mblk_t *); 133 static void *ipsecesp_stack_init(netstackid_t stackid, netstack_t *ns); 134 static void ipsecesp_stack_fini(netstackid_t stackid, void *arg); 135 static void esp_send_acquire(ipsacq_t *, mblk_t *, netstack_t *); 136 137 static void esp_prepare_udp(netstack_t *, mblk_t *, ipha_t *); 138 static ipsec_status_t esp_outbound_accelerated(mblk_t *, uint_t); 139 static ipsec_status_t esp_inbound_accelerated(mblk_t *, mblk_t *, 140 boolean_t, ipsa_t *); 141 142 static boolean_t esp_register_out(uint32_t, uint32_t, uint_t, 143 ipsecesp_stack_t *); 144 static boolean_t esp_strip_header(mblk_t *, boolean_t, uint32_t, 145 kstat_named_t **, ipsecesp_stack_t *); 146 static ipsec_status_t esp_submit_req_inbound(mblk_t *, ipsa_t *, uint_t); 147 static ipsec_status_t esp_submit_req_outbound(mblk_t *, ipsa_t *, uchar_t *, 148 uint_t); 149 /* Setable in /etc/system */ 150 uint32_t esp_hash_size = IPSEC_DEFAULT_HASH_SIZE; 151 152 static struct module_info info = { 153 5137, "ipsecesp", 0, INFPSZ, 65536, 1024 154 }; 155 156 static struct qinit rinit = { 157 (pfi_t)ipsecesp_rput, NULL, ipsecesp_open, ipsecesp_close, NULL, &info, 158 NULL 159 }; 160 161 static struct qinit winit = { 162 (pfi_t)ipsecesp_wput, NULL, ipsecesp_open, ipsecesp_close, NULL, &info, 163 NULL 164 }; 165 166 struct streamtab ipsecespinfo = { 167 &rinit, &winit, NULL, NULL 168 }; 169 170 static taskq_t *esp_taskq; 171 172 /* 173 * OTOH, this one is set at open/close, and I'm D_MTQPAIR for now. 174 * 175 * Question: Do I need this, given that all instance's esps->esps_wq point 176 * to IP? 177 * 178 * Answer: Yes, because I need to know which queue is BOUND to 179 * IPPROTO_ESP 180 */ 181 182 /* 183 * Stats. This may eventually become a full-blown SNMP MIB once that spec 184 * stabilizes. 185 */ 186 187 typedef struct esp_kstats_s { 188 kstat_named_t esp_stat_num_aalgs; 189 kstat_named_t esp_stat_good_auth; 190 kstat_named_t esp_stat_bad_auth; 191 kstat_named_t esp_stat_bad_padding; 192 kstat_named_t esp_stat_replay_failures; 193 kstat_named_t esp_stat_replay_early_failures; 194 kstat_named_t esp_stat_keysock_in; 195 kstat_named_t esp_stat_out_requests; 196 kstat_named_t esp_stat_acquire_requests; 197 kstat_named_t esp_stat_bytes_expired; 198 kstat_named_t esp_stat_out_discards; 199 kstat_named_t esp_stat_in_accelerated; 200 kstat_named_t esp_stat_out_accelerated; 201 kstat_named_t esp_stat_noaccel; 202 kstat_named_t esp_stat_crypto_sync; 203 kstat_named_t esp_stat_crypto_async; 204 kstat_named_t esp_stat_crypto_failures; 205 kstat_named_t esp_stat_num_ealgs; 206 kstat_named_t esp_stat_bad_decrypt; 207 kstat_named_t esp_stat_sa_port_renumbers; 208 } esp_kstats_t; 209 210 /* 211 * espstack->esp_kstats is equal to espstack->esp_ksp->ks_data if 212 * kstat_create_netstack for espstack->esp_ksp succeeds, but when it 213 * fails, it will be NULL. Note this is done for all stack instances, 214 * so it *could* fail. hence a non-NULL checking is done for 215 * ESP_BUMP_STAT and ESP_DEBUMP_STAT 216 */ 217 #define ESP_BUMP_STAT(espstack, x) \ 218 do { \ 219 if (espstack->esp_kstats != NULL) \ 220 (espstack->esp_kstats->esp_stat_ ## x).value.ui64++; \ 221 _NOTE(CONSTCOND) \ 222 } while (0) 223 224 #define ESP_DEBUMP_STAT(espstack, x) \ 225 do { \ 226 if (espstack->esp_kstats != NULL) \ 227 (espstack->esp_kstats->esp_stat_ ## x).value.ui64--; \ 228 _NOTE(CONSTCOND) \ 229 } while (0) 230 231 static int esp_kstat_update(kstat_t *, int); 232 233 static boolean_t 234 esp_kstat_init(ipsecesp_stack_t *espstack, netstackid_t stackid) 235 { 236 espstack->esp_ksp = kstat_create_netstack("ipsecesp", 0, "esp_stat", 237 "net", KSTAT_TYPE_NAMED, 238 sizeof (esp_kstats_t) / sizeof (kstat_named_t), 239 KSTAT_FLAG_PERSISTENT, stackid); 240 241 if (espstack->esp_ksp == NULL || espstack->esp_ksp->ks_data == NULL) 242 return (B_FALSE); 243 244 espstack->esp_kstats = espstack->esp_ksp->ks_data; 245 246 espstack->esp_ksp->ks_update = esp_kstat_update; 247 espstack->esp_ksp->ks_private = (void *)(uintptr_t)stackid; 248 249 #define K64 KSTAT_DATA_UINT64 250 #define KI(x) kstat_named_init(&(espstack->esp_kstats->esp_stat_##x), #x, K64) 251 252 KI(num_aalgs); 253 KI(num_ealgs); 254 KI(good_auth); 255 KI(bad_auth); 256 KI(bad_padding); 257 KI(replay_failures); 258 KI(replay_early_failures); 259 KI(keysock_in); 260 KI(out_requests); 261 KI(acquire_requests); 262 KI(bytes_expired); 263 KI(out_discards); 264 KI(in_accelerated); 265 KI(out_accelerated); 266 KI(noaccel); 267 KI(crypto_sync); 268 KI(crypto_async); 269 KI(crypto_failures); 270 KI(bad_decrypt); 271 KI(sa_port_renumbers); 272 273 #undef KI 274 #undef K64 275 276 kstat_install(espstack->esp_ksp); 277 278 return (B_TRUE); 279 } 280 281 static int 282 esp_kstat_update(kstat_t *kp, int rw) 283 { 284 esp_kstats_t *ekp; 285 netstackid_t stackid = (zoneid_t)(uintptr_t)kp->ks_private; 286 netstack_t *ns; 287 ipsec_stack_t *ipss; 288 289 if ((kp == NULL) || (kp->ks_data == NULL)) 290 return (EIO); 291 292 if (rw == KSTAT_WRITE) 293 return (EACCES); 294 295 ns = netstack_find_by_stackid(stackid); 296 if (ns == NULL) 297 return (-1); 298 ipss = ns->netstack_ipsec; 299 if (ipss == NULL) { 300 netstack_rele(ns); 301 return (-1); 302 } 303 ekp = (esp_kstats_t *)kp->ks_data; 304 305 mutex_enter(&ipss->ipsec_alg_lock); 306 ekp->esp_stat_num_aalgs.value.ui64 = 307 ipss->ipsec_nalgs[IPSEC_ALG_AUTH]; 308 ekp->esp_stat_num_ealgs.value.ui64 = 309 ipss->ipsec_nalgs[IPSEC_ALG_ENCR]; 310 mutex_exit(&ipss->ipsec_alg_lock); 311 312 netstack_rele(ns); 313 return (0); 314 } 315 316 #ifdef DEBUG 317 /* 318 * Debug routine, useful to see pre-encryption data. 319 */ 320 static char * 321 dump_msg(mblk_t *mp) 322 { 323 char tmp_str[3], tmp_line[256]; 324 325 while (mp != NULL) { 326 unsigned char *ptr; 327 328 printf("mblk address 0x%p, length %ld, db_ref %d " 329 "type %d, base 0x%p, lim 0x%p\n", 330 (void *) mp, (long)(mp->b_wptr - mp->b_rptr), 331 mp->b_datap->db_ref, mp->b_datap->db_type, 332 (void *)mp->b_datap->db_base, (void *)mp->b_datap->db_lim); 333 ptr = mp->b_rptr; 334 335 tmp_line[0] = '\0'; 336 while (ptr < mp->b_wptr) { 337 uint_t diff; 338 339 diff = (ptr - mp->b_rptr); 340 if (!(diff & 0x1f)) { 341 if (strlen(tmp_line) > 0) { 342 printf("bytes: %s\n", tmp_line); 343 tmp_line[0] = '\0'; 344 } 345 } 346 if (!(diff & 0x3)) 347 (void) strcat(tmp_line, " "); 348 (void) sprintf(tmp_str, "%02x", *ptr); 349 (void) strcat(tmp_line, tmp_str); 350 ptr++; 351 } 352 if (strlen(tmp_line) > 0) 353 printf("bytes: %s\n", tmp_line); 354 355 mp = mp->b_cont; 356 } 357 358 return ("\n"); 359 } 360 361 #else /* DEBUG */ 362 static char * 363 dump_msg(mblk_t *mp) 364 { 365 printf("Find value of mp %p.\n", mp); 366 return ("\n"); 367 } 368 #endif /* DEBUG */ 369 370 /* 371 * Don't have to lock age_interval, as only one thread will access it at 372 * a time, because I control the one function that does with timeout(). 373 */ 374 static void 375 esp_ager(void *arg) 376 { 377 ipsecesp_stack_t *espstack = (ipsecesp_stack_t *)arg; 378 netstack_t *ns = espstack->ipsecesp_netstack; 379 hrtime_t begin = gethrtime(); 380 381 sadb_ager(&espstack->esp_sadb.s_v4, espstack->esp_pfkey_q, 382 espstack->esp_sadb.s_ip_q, espstack->ipsecesp_reap_delay, ns); 383 sadb_ager(&espstack->esp_sadb.s_v6, espstack->esp_pfkey_q, 384 espstack->esp_sadb.s_ip_q, espstack->ipsecesp_reap_delay, ns); 385 386 espstack->esp_event = sadb_retimeout(begin, espstack->esp_pfkey_q, 387 esp_ager, espstack, 388 &espstack->ipsecesp_age_interval, espstack->ipsecesp_age_int_max, 389 info.mi_idnum); 390 } 391 392 /* 393 * Get an ESP NDD parameter. 394 */ 395 /* ARGSUSED */ 396 static int 397 ipsecesp_param_get(q, mp, cp, cr) 398 queue_t *q; 399 mblk_t *mp; 400 caddr_t cp; 401 cred_t *cr; 402 { 403 ipsecespparam_t *ipsecesppa = (ipsecespparam_t *)cp; 404 uint_t value; 405 ipsecesp_stack_t *espstack = (ipsecesp_stack_t *)q->q_ptr; 406 407 mutex_enter(&espstack->ipsecesp_param_lock); 408 value = ipsecesppa->ipsecesp_param_value; 409 mutex_exit(&espstack->ipsecesp_param_lock); 410 411 (void) mi_mpprintf(mp, "%u", value); 412 return (0); 413 } 414 415 /* 416 * This routine sets an NDD variable in a ipsecespparam_t structure. 417 */ 418 /* ARGSUSED */ 419 static int 420 ipsecesp_param_set(q, mp, value, cp, cr) 421 queue_t *q; 422 mblk_t *mp; 423 char *value; 424 caddr_t cp; 425 cred_t *cr; 426 { 427 ulong_t new_value; 428 ipsecespparam_t *ipsecesppa = (ipsecespparam_t *)cp; 429 ipsecesp_stack_t *espstack = (ipsecesp_stack_t *)q->q_ptr; 430 431 /* 432 * Fail the request if the new value does not lie within the 433 * required bounds. 434 */ 435 if (ddi_strtoul(value, NULL, 10, &new_value) != 0 || 436 new_value < ipsecesppa->ipsecesp_param_min || 437 new_value > ipsecesppa->ipsecesp_param_max) { 438 return (EINVAL); 439 } 440 441 /* Set the new value */ 442 mutex_enter(&espstack->ipsecesp_param_lock); 443 ipsecesppa->ipsecesp_param_value = new_value; 444 mutex_exit(&espstack->ipsecesp_param_lock); 445 return (0); 446 } 447 448 /* 449 * Using lifetime NDD variables, fill in an extended combination's 450 * lifetime information. 451 */ 452 void 453 ipsecesp_fill_defs(sadb_x_ecomb_t *ecomb, netstack_t *ns) 454 { 455 ipsecesp_stack_t *espstack = ns->netstack_ipsecesp; 456 457 ecomb->sadb_x_ecomb_soft_bytes = espstack->ipsecesp_default_soft_bytes; 458 ecomb->sadb_x_ecomb_hard_bytes = espstack->ipsecesp_default_hard_bytes; 459 ecomb->sadb_x_ecomb_soft_addtime = 460 espstack->ipsecesp_default_soft_addtime; 461 ecomb->sadb_x_ecomb_hard_addtime = 462 espstack->ipsecesp_default_hard_addtime; 463 ecomb->sadb_x_ecomb_soft_usetime = 464 espstack->ipsecesp_default_soft_usetime; 465 ecomb->sadb_x_ecomb_hard_usetime = 466 espstack->ipsecesp_default_hard_usetime; 467 } 468 469 /* 470 * Initialize things for ESP at module load time. 471 */ 472 boolean_t 473 ipsecesp_ddi_init(void) 474 { 475 esp_taskq = taskq_create("esp_taskq", 1, minclsyspri, 476 IPSEC_TASKQ_MIN, IPSEC_TASKQ_MAX, 0); 477 478 /* 479 * We want to be informed each time a stack is created or 480 * destroyed in the kernel, so we can maintain the 481 * set of ipsecesp_stack_t's. 482 */ 483 netstack_register(NS_IPSECESP, ipsecesp_stack_init, NULL, 484 ipsecesp_stack_fini); 485 486 return (B_TRUE); 487 } 488 489 /* 490 * Walk through the param array specified registering each element with the 491 * named dispatch handler. 492 */ 493 static boolean_t 494 ipsecesp_param_register(IDP *ndp, ipsecespparam_t *espp, int cnt) 495 { 496 for (; cnt-- > 0; espp++) { 497 if (espp->ipsecesp_param_name != NULL && 498 espp->ipsecesp_param_name[0]) { 499 if (!nd_load(ndp, 500 espp->ipsecesp_param_name, 501 ipsecesp_param_get, ipsecesp_param_set, 502 (caddr_t)espp)) { 503 nd_free(ndp); 504 return (B_FALSE); 505 } 506 } 507 } 508 return (B_TRUE); 509 } 510 /* 511 * Initialize things for ESP for each stack instance 512 */ 513 static void * 514 ipsecesp_stack_init(netstackid_t stackid, netstack_t *ns) 515 { 516 ipsecesp_stack_t *espstack; 517 ipsecespparam_t *espp; 518 519 espstack = (ipsecesp_stack_t *)kmem_zalloc(sizeof (*espstack), 520 KM_SLEEP); 521 espstack->ipsecesp_netstack = ns; 522 523 espp = (ipsecespparam_t *)kmem_alloc(sizeof (lcl_param_arr), KM_SLEEP); 524 espstack->ipsecesp_params = espp; 525 bcopy(lcl_param_arr, espp, sizeof (lcl_param_arr)); 526 527 (void) ipsecesp_param_register(&espstack->ipsecesp_g_nd, espp, 528 A_CNT(lcl_param_arr)); 529 530 (void) esp_kstat_init(espstack, stackid); 531 532 espstack->esp_sadb.s_acquire_timeout = 533 &espstack->ipsecesp_acquire_timeout; 534 espstack->esp_sadb.s_acqfn = esp_send_acquire; 535 sadbp_init("ESP", &espstack->esp_sadb, SADB_SATYPE_ESP, esp_hash_size, 536 espstack->ipsecesp_netstack); 537 538 mutex_init(&espstack->ipsecesp_param_lock, NULL, MUTEX_DEFAULT, 0); 539 540 ip_drop_register(&espstack->esp_dropper, "IPsec ESP"); 541 return (espstack); 542 } 543 544 /* 545 * Destroy things for ESP at module unload time. 546 */ 547 void 548 ipsecesp_ddi_destroy(void) 549 { 550 netstack_unregister(NS_IPSECESP); 551 taskq_destroy(esp_taskq); 552 } 553 554 /* 555 * Destroy things for ESP for one stack instance 556 */ 557 static void 558 ipsecesp_stack_fini(netstackid_t stackid, void *arg) 559 { 560 ipsecesp_stack_t *espstack = (ipsecesp_stack_t *)arg; 561 562 if (espstack->esp_pfkey_q != NULL) { 563 (void) quntimeout(espstack->esp_pfkey_q, espstack->esp_event); 564 } 565 espstack->esp_sadb.s_acqfn = NULL; 566 espstack->esp_sadb.s_acquire_timeout = NULL; 567 sadbp_destroy(&espstack->esp_sadb, espstack->ipsecesp_netstack); 568 ip_drop_unregister(&espstack->esp_dropper); 569 mutex_destroy(&espstack->ipsecesp_param_lock); 570 nd_free(&espstack->ipsecesp_g_nd); 571 572 kmem_free(espstack->ipsecesp_params, sizeof (lcl_param_arr)); 573 espstack->ipsecesp_params = NULL; 574 kstat_delete_netstack(espstack->esp_ksp, stackid); 575 espstack->esp_ksp = NULL; 576 espstack->esp_kstats = NULL; 577 kmem_free(espstack, sizeof (*espstack)); 578 } 579 580 /* 581 * ESP module open routine. 582 */ 583 /* ARGSUSED */ 584 static int 585 ipsecesp_open(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp) 586 { 587 netstack_t *ns; 588 ipsecesp_stack_t *espstack; 589 590 if (secpolicy_ip_config(credp, B_FALSE) != 0) 591 return (EPERM); 592 593 if (q->q_ptr != NULL) 594 return (0); /* Re-open of an already open instance. */ 595 596 if (sflag != MODOPEN) 597 return (EINVAL); 598 599 ns = netstack_find_by_cred(credp); 600 ASSERT(ns != NULL); 601 espstack = ns->netstack_ipsecesp; 602 ASSERT(espstack != NULL); 603 604 /* 605 * ASSUMPTIONS (because I'm MT_OCEXCL): 606 * 607 * * I'm being pushed on top of IP for all my opens (incl. #1). 608 * * Only ipsecesp_open() can write into esp_sadb.s_ip_q. 609 * * Because of this, I can check lazily for esp_sadb.s_ip_q. 610 * 611 * If these assumptions are wrong, I'm in BIG trouble... 612 */ 613 614 q->q_ptr = espstack; 615 WR(q)->q_ptr = q->q_ptr; 616 617 if (espstack->esp_sadb.s_ip_q == NULL) { 618 struct T_unbind_req *tur; 619 620 espstack->esp_sadb.s_ip_q = WR(q); 621 /* Allocate an unbind... */ 622 espstack->esp_ip_unbind = allocb(sizeof (struct T_unbind_req), 623 BPRI_HI); 624 625 /* 626 * Send down T_BIND_REQ to bind IPPROTO_ESP. 627 * Handle the ACK here in ESP. 628 */ 629 qprocson(q); 630 if (espstack->esp_ip_unbind == NULL || 631 !sadb_t_bind_req(espstack->esp_sadb.s_ip_q, IPPROTO_ESP)) { 632 if (espstack->esp_ip_unbind != NULL) { 633 freeb(espstack->esp_ip_unbind); 634 espstack->esp_ip_unbind = NULL; 635 } 636 q->q_ptr = NULL; 637 netstack_rele(espstack->ipsecesp_netstack); 638 return (ENOMEM); 639 } 640 641 espstack->esp_ip_unbind->b_datap->db_type = M_PROTO; 642 tur = (struct T_unbind_req *)espstack->esp_ip_unbind->b_rptr; 643 tur->PRIM_type = T_UNBIND_REQ; 644 } else { 645 qprocson(q); 646 } 647 648 /* 649 * For now, there's not much I can do. I'll be getting a message 650 * passed down to me from keysock (in my wput), and a T_BIND_ACK 651 * up from IP (in my rput). 652 */ 653 654 return (0); 655 } 656 657 /* 658 * ESP module close routine. 659 */ 660 static int 661 ipsecesp_close(queue_t *q) 662 { 663 ipsecesp_stack_t *espstack = (ipsecesp_stack_t *)q->q_ptr; 664 665 /* 666 * If esp_sadb.s_ip_q is attached to this instance, send a 667 * T_UNBIND_REQ to IP for the instance before doing 668 * a qprocsoff(). 669 */ 670 if (WR(q) == espstack->esp_sadb.s_ip_q && 671 espstack->esp_ip_unbind != NULL) { 672 putnext(WR(q), espstack->esp_ip_unbind); 673 espstack->esp_ip_unbind = NULL; 674 } 675 676 /* 677 * Clean up q_ptr, if needed. 678 */ 679 qprocsoff(q); 680 681 /* Keysock queue check is safe, because of OCEXCL perimeter. */ 682 683 if (q == espstack->esp_pfkey_q) { 684 esp1dbg(espstack, 685 ("ipsecesp_close: Ummm... keysock is closing ESP.\n")); 686 espstack->esp_pfkey_q = NULL; 687 /* Detach qtimeouts. */ 688 (void) quntimeout(q, espstack->esp_event); 689 } 690 691 if (WR(q) == espstack->esp_sadb.s_ip_q) { 692 /* 693 * If the esp_sadb.s_ip_q is attached to this instance, find 694 * another. The OCEXCL outer perimeter helps us here. 695 */ 696 espstack->esp_sadb.s_ip_q = NULL; 697 698 /* 699 * Find a replacement queue for esp_sadb.s_ip_q. 700 */ 701 if (espstack->esp_pfkey_q != NULL && 702 espstack->esp_pfkey_q != RD(q)) { 703 /* 704 * See if we can use the pfkey_q. 705 */ 706 espstack->esp_sadb.s_ip_q = WR(espstack->esp_pfkey_q); 707 } 708 709 if (espstack->esp_sadb.s_ip_q == NULL || 710 !sadb_t_bind_req(espstack->esp_sadb.s_ip_q, IPPROTO_ESP)) { 711 esp1dbg(espstack, ("ipsecesp: Can't reassign ip_q.\n")); 712 espstack->esp_sadb.s_ip_q = NULL; 713 } else { 714 espstack->esp_ip_unbind = 715 allocb(sizeof (struct T_unbind_req), BPRI_HI); 716 717 if (espstack->esp_ip_unbind != NULL) { 718 struct T_unbind_req *tur; 719 720 espstack->esp_ip_unbind->b_datap->db_type = 721 M_PROTO; 722 tur = (struct T_unbind_req *) 723 espstack->esp_ip_unbind->b_rptr; 724 tur->PRIM_type = T_UNBIND_REQ; 725 } 726 /* If it's NULL, I can't do much here. */ 727 } 728 } 729 730 netstack_rele(espstack->ipsecesp_netstack); 731 return (0); 732 } 733 734 /* 735 * Add a number of bytes to what the SA has protected so far. Return 736 * B_TRUE if the SA can still protect that many bytes. 737 * 738 * Caller must REFRELE the passed-in assoc. This function must REFRELE 739 * any obtained peer SA. 740 */ 741 static boolean_t 742 esp_age_bytes(ipsa_t *assoc, uint64_t bytes, boolean_t inbound) 743 { 744 ipsa_t *inassoc, *outassoc; 745 isaf_t *bucket; 746 boolean_t inrc, outrc, isv6; 747 sadb_t *sp; 748 int outhash; 749 netstack_t *ns = assoc->ipsa_netstack; 750 ipsecesp_stack_t *espstack = ns->netstack_ipsecesp; 751 752 /* No peer? No problem! */ 753 if (!assoc->ipsa_haspeer) { 754 return (sadb_age_bytes(espstack->esp_pfkey_q, assoc, bytes, 755 B_TRUE)); 756 } 757 758 /* 759 * Otherwise, we want to grab both the original assoc and its peer. 760 * There might be a race for this, but if it's a real race, two 761 * expire messages may occur. We limit this by only sending the 762 * expire message on one of the peers, we'll pick the inbound 763 * arbitrarily. 764 * 765 * If we need tight synchronization on the peer SA, then we need to 766 * reconsider. 767 */ 768 769 /* Use address length to select IPv6/IPv4 */ 770 isv6 = (assoc->ipsa_addrfam == AF_INET6); 771 sp = isv6 ? &espstack->esp_sadb.s_v6 : &espstack->esp_sadb.s_v4; 772 773 if (inbound) { 774 inassoc = assoc; 775 if (isv6) { 776 outhash = OUTBOUND_HASH_V6(sp, *((in6_addr_t *) 777 &inassoc->ipsa_dstaddr)); 778 } else { 779 outhash = OUTBOUND_HASH_V4(sp, *((ipaddr_t *) 780 &inassoc->ipsa_dstaddr)); 781 } 782 bucket = &sp->sdb_of[outhash]; 783 mutex_enter(&bucket->isaf_lock); 784 outassoc = ipsec_getassocbyspi(bucket, inassoc->ipsa_spi, 785 inassoc->ipsa_srcaddr, inassoc->ipsa_dstaddr, 786 inassoc->ipsa_addrfam); 787 mutex_exit(&bucket->isaf_lock); 788 if (outassoc == NULL) { 789 /* Q: Do we wish to set haspeer == B_FALSE? */ 790 esp0dbg(("esp_age_bytes: " 791 "can't find peer for inbound.\n")); 792 return (sadb_age_bytes(espstack->esp_pfkey_q, inassoc, 793 bytes, B_TRUE)); 794 } 795 } else { 796 outassoc = assoc; 797 bucket = INBOUND_BUCKET(sp, outassoc->ipsa_spi); 798 mutex_enter(&bucket->isaf_lock); 799 inassoc = ipsec_getassocbyspi(bucket, outassoc->ipsa_spi, 800 outassoc->ipsa_srcaddr, outassoc->ipsa_dstaddr, 801 outassoc->ipsa_addrfam); 802 mutex_exit(&bucket->isaf_lock); 803 if (inassoc == NULL) { 804 /* Q: Do we wish to set haspeer == B_FALSE? */ 805 esp0dbg(("esp_age_bytes: " 806 "can't find peer for outbound.\n")); 807 return (sadb_age_bytes(espstack->esp_pfkey_q, outassoc, 808 bytes, B_TRUE)); 809 } 810 } 811 812 inrc = sadb_age_bytes(espstack->esp_pfkey_q, inassoc, bytes, B_TRUE); 813 outrc = sadb_age_bytes(espstack->esp_pfkey_q, outassoc, bytes, B_FALSE); 814 815 /* 816 * REFRELE any peer SA. 817 * 818 * Because of the multi-line macro nature of IPSA_REFRELE, keep 819 * them in { }. 820 */ 821 if (inbound) { 822 IPSA_REFRELE(outassoc); 823 } else { 824 IPSA_REFRELE(inassoc); 825 } 826 827 return (inrc && outrc); 828 } 829 830 /* 831 * Do incoming NAT-T manipulations for packet. 832 */ 833 static ipsec_status_t 834 esp_fix_natt_checksums(mblk_t *data_mp, ipsa_t *assoc) 835 { 836 ipha_t *ipha = (ipha_t *)data_mp->b_rptr; 837 tcpha_t *tcph; 838 udpha_t *udpha; 839 /* Initialize to our inbound cksum adjustment... */ 840 uint32_t sum = assoc->ipsa_inbound_cksum; 841 842 switch (ipha->ipha_protocol) { 843 case IPPROTO_TCP: 844 tcph = (tcpha_t *)(data_mp->b_rptr + 845 IPH_HDR_LENGTH(ipha)); 846 847 #define DOWN_SUM(x) (x) = ((x) & 0xFFFF) + ((x) >> 16) 848 sum += ~ntohs(tcph->tha_sum) & 0xFFFF; 849 DOWN_SUM(sum); 850 DOWN_SUM(sum); 851 tcph->tha_sum = ~htons(sum); 852 break; 853 case IPPROTO_UDP: 854 udpha = (udpha_t *)(data_mp->b_rptr + IPH_HDR_LENGTH(ipha)); 855 856 if (udpha->uha_checksum != 0) { 857 /* Adujst if the inbound one was not zero. */ 858 sum += ~ntohs(udpha->uha_checksum) & 0xFFFF; 859 DOWN_SUM(sum); 860 DOWN_SUM(sum); 861 udpha->uha_checksum = ~htons(sum); 862 if (udpha->uha_checksum == 0) 863 udpha->uha_checksum = 0xFFFF; 864 } 865 #undef DOWN_SUM 866 break; 867 case IPPROTO_IP: 868 /* 869 * This case is only an issue for self-encapsulated 870 * packets. So for now, fall through. 871 */ 872 break; 873 } 874 return (IPSEC_STATUS_SUCCESS); 875 } 876 877 878 /* 879 * Strip ESP header, check padding, and fix IP header. 880 * Returns B_TRUE on success, B_FALSE if an error occured. 881 */ 882 static boolean_t 883 esp_strip_header(mblk_t *data_mp, boolean_t isv4, uint32_t ivlen, 884 kstat_named_t **counter, ipsecesp_stack_t *espstack) 885 { 886 ipha_t *ipha; 887 ip6_t *ip6h; 888 uint_t divpoint; 889 mblk_t *scratch; 890 uint8_t nexthdr, padlen; 891 uint8_t lastpad; 892 ipsec_stack_t *ipss = espstack->ipsecesp_netstack->netstack_ipsec; 893 uint8_t *lastbyte; 894 895 /* 896 * Strip ESP data and fix IP header. 897 * 898 * XXX In case the beginning of esp_inbound() changes to not do a 899 * pullup, this part of the code can remain unchanged. 900 */ 901 if (isv4) { 902 ASSERT((data_mp->b_wptr - data_mp->b_rptr) >= sizeof (ipha_t)); 903 ipha = (ipha_t *)data_mp->b_rptr; 904 ASSERT((data_mp->b_wptr - data_mp->b_rptr) >= sizeof (esph_t) + 905 IPH_HDR_LENGTH(ipha)); 906 divpoint = IPH_HDR_LENGTH(ipha); 907 } else { 908 ASSERT((data_mp->b_wptr - data_mp->b_rptr) >= sizeof (ip6_t)); 909 ip6h = (ip6_t *)data_mp->b_rptr; 910 divpoint = ip_hdr_length_v6(data_mp, ip6h); 911 } 912 913 scratch = data_mp; 914 while (scratch->b_cont != NULL) 915 scratch = scratch->b_cont; 916 917 ASSERT((scratch->b_wptr - scratch->b_rptr) >= 3); 918 919 /* 920 * "Next header" and padding length are the last two bytes in the 921 * ESP-protected datagram, thus the explicit - 1 and - 2. 922 * lastpad is the last byte of the padding, which can be used for 923 * a quick check to see if the padding is correct. 924 */ 925 lastbyte = scratch->b_wptr - 1; 926 nexthdr = *lastbyte--; 927 padlen = *lastbyte--; 928 929 if (isv4) { 930 /* Fix part of the IP header. */ 931 ipha->ipha_protocol = nexthdr; 932 /* 933 * Reality check the padlen. The explicit - 2 is for the 934 * padding length and the next-header bytes. 935 */ 936 if (padlen >= ntohs(ipha->ipha_length) - sizeof (ipha_t) - 2 - 937 sizeof (esph_t) - ivlen) { 938 ESP_BUMP_STAT(espstack, bad_decrypt); 939 ipsec_rl_strlog(espstack->ipsecesp_netstack, 940 info.mi_idnum, 0, 0, 941 SL_ERROR | SL_WARN, 942 "Corrupt ESP packet (padlen too big).\n"); 943 esp1dbg(espstack, ("padlen (%d) is greater than:\n", 944 padlen)); 945 esp1dbg(espstack, ("pkt len(%d) - ip hdr - esp " 946 "hdr - ivlen(%d) = %d.\n", 947 ntohs(ipha->ipha_length), ivlen, 948 (int)(ntohs(ipha->ipha_length) - sizeof (ipha_t) - 949 2 - sizeof (esph_t) - ivlen))); 950 *counter = DROPPER(ipss, ipds_esp_bad_padlen); 951 return (B_FALSE); 952 } 953 954 /* 955 * Fix the rest of the header. The explicit - 2 is for the 956 * padding length and the next-header bytes. 957 */ 958 ipha->ipha_length = htons(ntohs(ipha->ipha_length) - padlen - 959 2 - sizeof (esph_t) - ivlen); 960 ipha->ipha_hdr_checksum = 0; 961 ipha->ipha_hdr_checksum = (uint16_t)ip_csum_hdr(ipha); 962 } else { 963 if (ip6h->ip6_nxt == IPPROTO_ESP) { 964 ip6h->ip6_nxt = nexthdr; 965 } else { 966 ip6_pkt_t ipp; 967 968 bzero(&ipp, sizeof (ipp)); 969 (void) ip_find_hdr_v6(data_mp, ip6h, &ipp, NULL); 970 if (ipp.ipp_dstopts != NULL) { 971 ipp.ipp_dstopts->ip6d_nxt = nexthdr; 972 } else if (ipp.ipp_rthdr != NULL) { 973 ipp.ipp_rthdr->ip6r_nxt = nexthdr; 974 } else if (ipp.ipp_hopopts != NULL) { 975 ipp.ipp_hopopts->ip6h_nxt = nexthdr; 976 } else { 977 /* Panic a DEBUG kernel. */ 978 ASSERT(ipp.ipp_hopopts != NULL); 979 /* Otherwise, pretend it's IP + ESP. */ 980 cmn_err(CE_WARN, "ESP IPv6 headers wrong.\n"); 981 ip6h->ip6_nxt = nexthdr; 982 } 983 } 984 985 if (padlen >= ntohs(ip6h->ip6_plen) - 2 - sizeof (esph_t) - 986 ivlen) { 987 ESP_BUMP_STAT(espstack, bad_decrypt); 988 ipsec_rl_strlog(espstack->ipsecesp_netstack, 989 info.mi_idnum, 0, 0, 990 SL_ERROR | SL_WARN, 991 "Corrupt ESP packet (v6 padlen too big).\n"); 992 esp1dbg(espstack, ("padlen (%d) is greater than:\n", 993 padlen)); 994 esp1dbg(espstack, 995 ("pkt len(%u) - ip hdr - esp hdr - ivlen(%d) = " 996 "%u.\n", (unsigned)(ntohs(ip6h->ip6_plen) 997 + sizeof (ip6_t)), ivlen, 998 (unsigned)(ntohs(ip6h->ip6_plen) - 2 - 999 sizeof (esph_t) - ivlen))); 1000 *counter = DROPPER(ipss, ipds_esp_bad_padlen); 1001 return (B_FALSE); 1002 } 1003 1004 1005 /* 1006 * Fix the rest of the header. The explicit - 2 is for the 1007 * padding length and the next-header bytes. IPv6 is nice, 1008 * because there's no hdr checksum! 1009 */ 1010 ip6h->ip6_plen = htons(ntohs(ip6h->ip6_plen) - padlen - 1011 2 - sizeof (esph_t) - ivlen); 1012 } 1013 1014 if (espstack->ipsecesp_padding_check > 0 && padlen > 0) { 1015 /* 1016 * Weak padding check: compare last-byte to length, they 1017 * should be equal. 1018 */ 1019 lastpad = *lastbyte--; 1020 1021 if (padlen != lastpad) { 1022 ipsec_rl_strlog(espstack->ipsecesp_netstack, 1023 info.mi_idnum, 0, 0, SL_ERROR | SL_WARN, 1024 "Corrupt ESP packet (lastpad != padlen).\n"); 1025 esp1dbg(espstack, 1026 ("lastpad (%d) not equal to padlen (%d):\n", 1027 lastpad, padlen)); 1028 ESP_BUMP_STAT(espstack, bad_padding); 1029 *counter = DROPPER(ipss, ipds_esp_bad_padding); 1030 return (B_FALSE); 1031 } 1032 1033 /* 1034 * Strong padding check: Check all pad bytes to see that 1035 * they're ascending. Go backwards using a descending counter 1036 * to verify. padlen == 1 is checked by previous block, so 1037 * only bother if we've more than 1 byte of padding. 1038 * Consequently, start the check one byte before the location 1039 * of "lastpad". 1040 */ 1041 if (espstack->ipsecesp_padding_check > 1) { 1042 /* 1043 * This assert may have to become an if and a pullup 1044 * if we start accepting multi-dblk mblks. For now, 1045 * though, any packet here will have been pulled up in 1046 * esp_inbound. 1047 */ 1048 ASSERT(MBLKL(scratch) >= lastpad + 3); 1049 1050 /* 1051 * Use "--lastpad" because we already checked the very 1052 * last pad byte previously. 1053 */ 1054 while (--lastpad != 0) { 1055 if (lastpad != *lastbyte) { 1056 ipsec_rl_strlog( 1057 espstack->ipsecesp_netstack, 1058 info.mi_idnum, 0, 0, 1059 SL_ERROR | SL_WARN, "Corrupt ESP " 1060 "packet (bad padding).\n"); 1061 esp1dbg(espstack, 1062 ("padding not in correct" 1063 " format:\n")); 1064 ESP_BUMP_STAT(espstack, bad_padding); 1065 *counter = DROPPER(ipss, 1066 ipds_esp_bad_padding); 1067 return (B_FALSE); 1068 } 1069 lastbyte--; 1070 } 1071 } 1072 } 1073 1074 /* Trim off the padding. */ 1075 ASSERT(data_mp->b_cont == NULL); 1076 data_mp->b_wptr -= (padlen + 2); 1077 1078 /* 1079 * Remove the ESP header. 1080 * 1081 * The above assertions about data_mp's size will make this work. 1082 * 1083 * XXX Question: If I send up and get back a contiguous mblk, 1084 * would it be quicker to bcopy over, or keep doing the dupb stuff? 1085 * I go with copying for now. 1086 */ 1087 1088 if (IS_P2ALIGNED(data_mp->b_rptr, sizeof (uint32_t)) && 1089 IS_P2ALIGNED(ivlen, sizeof (uint32_t))) { 1090 uint8_t *start = data_mp->b_rptr; 1091 uint32_t *src, *dst; 1092 1093 src = (uint32_t *)(start + divpoint); 1094 dst = (uint32_t *)(start + divpoint + sizeof (esph_t) + ivlen); 1095 1096 ASSERT(IS_P2ALIGNED(dst, sizeof (uint32_t)) && 1097 IS_P2ALIGNED(src, sizeof (uint32_t))); 1098 1099 do { 1100 src--; 1101 dst--; 1102 *dst = *src; 1103 } while (src != (uint32_t *)start); 1104 1105 data_mp->b_rptr = (uchar_t *)dst; 1106 } else { 1107 uint8_t *start = data_mp->b_rptr; 1108 uint8_t *src, *dst; 1109 1110 src = start + divpoint; 1111 dst = src + sizeof (esph_t) + ivlen; 1112 1113 do { 1114 src--; 1115 dst--; 1116 *dst = *src; 1117 } while (src != start); 1118 1119 data_mp->b_rptr = dst; 1120 } 1121 1122 esp2dbg(espstack, ("data_mp after inbound ESP adjustment:\n")); 1123 esp2dbg(espstack, (dump_msg(data_mp))); 1124 1125 return (B_TRUE); 1126 } 1127 1128 /* 1129 * Updating use times can be tricky business if the ipsa_haspeer flag is 1130 * set. This function is called once in an SA's lifetime. 1131 * 1132 * Caller has to REFRELE "assoc" which is passed in. This function has 1133 * to REFRELE any peer SA that is obtained. 1134 */ 1135 static void 1136 esp_set_usetime(ipsa_t *assoc, boolean_t inbound) 1137 { 1138 ipsa_t *inassoc, *outassoc; 1139 isaf_t *bucket; 1140 sadb_t *sp; 1141 int outhash; 1142 boolean_t isv6; 1143 netstack_t *ns = assoc->ipsa_netstack; 1144 ipsecesp_stack_t *espstack = ns->netstack_ipsecesp; 1145 1146 /* No peer? No problem! */ 1147 if (!assoc->ipsa_haspeer) { 1148 sadb_set_usetime(assoc); 1149 return; 1150 } 1151 1152 /* 1153 * Otherwise, we want to grab both the original assoc and its peer. 1154 * There might be a race for this, but if it's a real race, the times 1155 * will be out-of-synch by at most a second, and since our time 1156 * granularity is a second, this won't be a problem. 1157 * 1158 * If we need tight synchronization on the peer SA, then we need to 1159 * reconsider. 1160 */ 1161 1162 /* Use address length to select IPv6/IPv4 */ 1163 isv6 = (assoc->ipsa_addrfam == AF_INET6); 1164 sp = isv6 ? &espstack->esp_sadb.s_v6 : &espstack->esp_sadb.s_v4; 1165 1166 if (inbound) { 1167 inassoc = assoc; 1168 if (isv6) { 1169 outhash = OUTBOUND_HASH_V6(sp, *((in6_addr_t *) 1170 &inassoc->ipsa_dstaddr)); 1171 } else { 1172 outhash = OUTBOUND_HASH_V4(sp, *((ipaddr_t *) 1173 &inassoc->ipsa_dstaddr)); 1174 } 1175 bucket = &sp->sdb_of[outhash]; 1176 mutex_enter(&bucket->isaf_lock); 1177 outassoc = ipsec_getassocbyspi(bucket, inassoc->ipsa_spi, 1178 inassoc->ipsa_srcaddr, inassoc->ipsa_dstaddr, 1179 inassoc->ipsa_addrfam); 1180 mutex_exit(&bucket->isaf_lock); 1181 if (outassoc == NULL) { 1182 /* Q: Do we wish to set haspeer == B_FALSE? */ 1183 esp0dbg(("esp_set_usetime: " 1184 "can't find peer for inbound.\n")); 1185 sadb_set_usetime(inassoc); 1186 return; 1187 } 1188 } else { 1189 outassoc = assoc; 1190 bucket = INBOUND_BUCKET(sp, outassoc->ipsa_spi); 1191 mutex_enter(&bucket->isaf_lock); 1192 inassoc = ipsec_getassocbyspi(bucket, outassoc->ipsa_spi, 1193 outassoc->ipsa_srcaddr, outassoc->ipsa_dstaddr, 1194 outassoc->ipsa_addrfam); 1195 mutex_exit(&bucket->isaf_lock); 1196 if (inassoc == NULL) { 1197 /* Q: Do we wish to set haspeer == B_FALSE? */ 1198 esp0dbg(("esp_set_usetime: " 1199 "can't find peer for outbound.\n")); 1200 sadb_set_usetime(outassoc); 1201 return; 1202 } 1203 } 1204 1205 /* Update usetime on both. */ 1206 sadb_set_usetime(inassoc); 1207 sadb_set_usetime(outassoc); 1208 1209 /* 1210 * REFRELE any peer SA. 1211 * 1212 * Because of the multi-line macro nature of IPSA_REFRELE, keep 1213 * them in { }. 1214 */ 1215 if (inbound) { 1216 IPSA_REFRELE(outassoc); 1217 } else { 1218 IPSA_REFRELE(inassoc); 1219 } 1220 } 1221 1222 /* 1223 * Handle ESP inbound data for IPv4 and IPv6. 1224 * On success returns B_TRUE, on failure returns B_FALSE and frees the 1225 * mblk chain ipsec_in_mp. 1226 */ 1227 ipsec_status_t 1228 esp_inbound(mblk_t *ipsec_in_mp, void *arg) 1229 { 1230 mblk_t *data_mp = ipsec_in_mp->b_cont; 1231 ipsec_in_t *ii = (ipsec_in_t *)ipsec_in_mp->b_rptr; 1232 esph_t *esph = (esph_t *)arg; 1233 ipsa_t *ipsa = ii->ipsec_in_esp_sa; 1234 netstack_t *ns = ii->ipsec_in_ns; 1235 ipsecesp_stack_t *espstack = ns->netstack_ipsecesp; 1236 ipsec_stack_t *ipss = ns->netstack_ipsec; 1237 1238 /* 1239 * We may wish to check replay in-range-only here as an optimization. 1240 * Include the reality check of ipsa->ipsa_replay > 1241 * ipsa->ipsa_replay_wsize for times when it's the first N packets, 1242 * where N == ipsa->ipsa_replay_wsize. 1243 * 1244 * Another check that may come here later is the "collision" check. 1245 * If legitimate packets flow quickly enough, this won't be a problem, 1246 * but collisions may cause authentication algorithm crunching to 1247 * take place when it doesn't need to. 1248 */ 1249 if (!sadb_replay_peek(ipsa, esph->esph_replay)) { 1250 ESP_BUMP_STAT(espstack, replay_early_failures); 1251 IP_ESP_BUMP_STAT(ipss, in_discards); 1252 /* 1253 * TODO: Extract inbound interface from the IPSEC_IN 1254 * message's ii->ipsec_in_rill_index. 1255 */ 1256 ip_drop_packet(ipsec_in_mp, B_TRUE, NULL, NULL, 1257 DROPPER(ipss, ipds_esp_early_replay), 1258 &espstack->esp_dropper); 1259 return (IPSEC_STATUS_FAILED); 1260 } 1261 1262 /* 1263 * Has this packet already been processed by a hardware 1264 * IPsec accelerator? 1265 */ 1266 if (ii->ipsec_in_accelerated) { 1267 ipsec_status_t rv; 1268 esp3dbg(espstack, 1269 ("esp_inbound: pkt processed by ill=%d isv6=%d\n", 1270 ii->ipsec_in_ill_index, !ii->ipsec_in_v4)); 1271 rv = esp_inbound_accelerated(ipsec_in_mp, 1272 data_mp, ii->ipsec_in_v4, ipsa); 1273 return (rv); 1274 } 1275 ESP_BUMP_STAT(espstack, noaccel); 1276 1277 /* 1278 * Adjust the IP header's payload length to reflect the removal 1279 * of the ICV. 1280 */ 1281 if (!ii->ipsec_in_v4) { 1282 ip6_t *ip6h = (ip6_t *)data_mp->b_rptr; 1283 ip6h->ip6_plen = htons(ntohs(ip6h->ip6_plen) - 1284 ipsa->ipsa_mac_len); 1285 } else { 1286 ipha_t *ipha = (ipha_t *)data_mp->b_rptr; 1287 ipha->ipha_length = htons(ntohs(ipha->ipha_length) - 1288 ipsa->ipsa_mac_len); 1289 } 1290 1291 /* submit the request to the crypto framework */ 1292 return (esp_submit_req_inbound(ipsec_in_mp, ipsa, 1293 (uint8_t *)esph - data_mp->b_rptr)); 1294 } 1295 1296 /* 1297 * Perform the really difficult work of inserting the proposed situation. 1298 * Called while holding the algorithm lock. 1299 */ 1300 static void 1301 esp_insert_prop(sadb_prop_t *prop, ipsacq_t *acqrec, uint_t combs) 1302 { 1303 sadb_comb_t *comb = (sadb_comb_t *)(prop + 1); 1304 ipsec_out_t *io; 1305 ipsec_action_t *ap; 1306 ipsec_prot_t *prot; 1307 netstack_t *ns; 1308 ipsecesp_stack_t *espstack; 1309 ipsec_stack_t *ipss; 1310 1311 io = (ipsec_out_t *)acqrec->ipsacq_mp->b_rptr; 1312 ASSERT(io->ipsec_out_type == IPSEC_OUT); 1313 ns = io->ipsec_out_ns; 1314 espstack = ns->netstack_ipsecesp; 1315 ipss = ns->netstack_ipsec; 1316 ASSERT(MUTEX_HELD(&ipss->ipsec_alg_lock)); 1317 1318 prop->sadb_prop_exttype = SADB_EXT_PROPOSAL; 1319 prop->sadb_prop_len = SADB_8TO64(sizeof (sadb_prop_t)); 1320 *(uint32_t *)(&prop->sadb_prop_replay) = 0; /* Quick zero-out! */ 1321 1322 prop->sadb_prop_replay = espstack->ipsecesp_replay_size; 1323 1324 /* 1325 * Based upon algorithm properties, and what-not, prioritize 1326 * a proposal. If the IPSEC_OUT message has an algorithm specified, 1327 * use it first and foremost. 1328 * 1329 * For each action in policy list 1330 * Add combination. If I've hit limit, return. 1331 */ 1332 1333 for (ap = acqrec->ipsacq_act; ap != NULL; 1334 ap = ap->ipa_next) { 1335 ipsec_alginfo_t *ealg = NULL; 1336 ipsec_alginfo_t *aalg = NULL; 1337 1338 if (ap->ipa_act.ipa_type != IPSEC_POLICY_APPLY) 1339 continue; 1340 1341 prot = &ap->ipa_act.ipa_apply; 1342 1343 if (!(prot->ipp_use_esp)) 1344 continue; 1345 1346 if (prot->ipp_esp_auth_alg != 0) { 1347 aalg = ipss->ipsec_alglists[IPSEC_ALG_AUTH] 1348 [prot->ipp_esp_auth_alg]; 1349 if (aalg == NULL || !ALG_VALID(aalg)) 1350 continue; 1351 } 1352 1353 ASSERT(prot->ipp_encr_alg > 0); 1354 ealg = ipss->ipsec_alglists[IPSEC_ALG_ENCR] 1355 [prot->ipp_encr_alg]; 1356 if (ealg == NULL || !ALG_VALID(ealg)) 1357 continue; 1358 1359 comb->sadb_comb_flags = 0; 1360 comb->sadb_comb_reserved = 0; 1361 comb->sadb_comb_encrypt = ealg->alg_id; 1362 comb->sadb_comb_encrypt_minbits = 1363 MAX(prot->ipp_espe_minbits, ealg->alg_ef_minbits); 1364 comb->sadb_comb_encrypt_maxbits = 1365 MIN(prot->ipp_espe_maxbits, ealg->alg_ef_maxbits); 1366 if (aalg == NULL) { 1367 comb->sadb_comb_auth = 0; 1368 comb->sadb_comb_auth_minbits = 0; 1369 comb->sadb_comb_auth_maxbits = 0; 1370 } else { 1371 comb->sadb_comb_auth = aalg->alg_id; 1372 comb->sadb_comb_auth_minbits = 1373 MAX(prot->ipp_espa_minbits, aalg->alg_ef_minbits); 1374 comb->sadb_comb_auth_maxbits = 1375 MIN(prot->ipp_espa_maxbits, aalg->alg_ef_maxbits); 1376 } 1377 1378 /* 1379 * The following may be based on algorithm 1380 * properties, but in the meantime, we just pick 1381 * some good, sensible numbers. Key mgmt. can 1382 * (and perhaps should) be the place to finalize 1383 * such decisions. 1384 */ 1385 1386 /* 1387 * No limits on allocations, since we really don't 1388 * support that concept currently. 1389 */ 1390 comb->sadb_comb_soft_allocations = 0; 1391 comb->sadb_comb_hard_allocations = 0; 1392 1393 /* 1394 * These may want to come from policy rule.. 1395 */ 1396 comb->sadb_comb_soft_bytes = 1397 espstack->ipsecesp_default_soft_bytes; 1398 comb->sadb_comb_hard_bytes = 1399 espstack->ipsecesp_default_hard_bytes; 1400 comb->sadb_comb_soft_addtime = 1401 espstack->ipsecesp_default_soft_addtime; 1402 comb->sadb_comb_hard_addtime = 1403 espstack->ipsecesp_default_hard_addtime; 1404 comb->sadb_comb_soft_usetime = 1405 espstack->ipsecesp_default_soft_usetime; 1406 comb->sadb_comb_hard_usetime = 1407 espstack->ipsecesp_default_hard_usetime; 1408 1409 prop->sadb_prop_len += SADB_8TO64(sizeof (*comb)); 1410 if (--combs == 0) 1411 break; /* out of space.. */ 1412 comb++; 1413 } 1414 } 1415 1416 /* 1417 * Prepare and actually send the SADB_ACQUIRE message to PF_KEY. 1418 */ 1419 static void 1420 esp_send_acquire(ipsacq_t *acqrec, mblk_t *extended, netstack_t *ns) 1421 { 1422 uint_t combs; 1423 sadb_msg_t *samsg; 1424 sadb_prop_t *prop; 1425 mblk_t *pfkeymp, *msgmp; 1426 ipsecesp_stack_t *espstack = ns->netstack_ipsecesp; 1427 ipsec_stack_t *ipss = ns->netstack_ipsec; 1428 1429 ESP_BUMP_STAT(espstack, acquire_requests); 1430 1431 if (espstack->esp_pfkey_q == NULL) { 1432 mutex_exit(&acqrec->ipsacq_lock); 1433 return; 1434 } 1435 1436 /* Set up ACQUIRE. */ 1437 pfkeymp = sadb_setup_acquire(acqrec, SADB_SATYPE_ESP, 1438 ns->netstack_ipsec); 1439 if (pfkeymp == NULL) { 1440 esp0dbg(("sadb_setup_acquire failed.\n")); 1441 mutex_exit(&acqrec->ipsacq_lock); 1442 return; 1443 } 1444 ASSERT(MUTEX_HELD(&ipss->ipsec_alg_lock)); 1445 combs = ipss->ipsec_nalgs[IPSEC_ALG_AUTH] * 1446 ipss->ipsec_nalgs[IPSEC_ALG_ENCR]; 1447 msgmp = pfkeymp->b_cont; 1448 samsg = (sadb_msg_t *)(msgmp->b_rptr); 1449 1450 /* Insert proposal here. */ 1451 1452 prop = (sadb_prop_t *)(((uint64_t *)samsg) + samsg->sadb_msg_len); 1453 esp_insert_prop(prop, acqrec, combs); 1454 samsg->sadb_msg_len += prop->sadb_prop_len; 1455 msgmp->b_wptr += SADB_64TO8(samsg->sadb_msg_len); 1456 1457 mutex_exit(&ipss->ipsec_alg_lock); 1458 1459 /* 1460 * Must mutex_exit() before sending PF_KEY message up, in 1461 * order to avoid recursive mutex_enter() if there are no registered 1462 * listeners. 1463 * 1464 * Once I've sent the message, I'm cool anyway. 1465 */ 1466 mutex_exit(&acqrec->ipsacq_lock); 1467 if (extended != NULL) { 1468 putnext(espstack->esp_pfkey_q, extended); 1469 } 1470 putnext(espstack->esp_pfkey_q, pfkeymp); 1471 } 1472 1473 /* 1474 * Handle the SADB_GETSPI message. Create a larval SA. 1475 */ 1476 static void 1477 esp_getspi(mblk_t *mp, keysock_in_t *ksi, ipsecesp_stack_t *espstack) 1478 { 1479 ipsa_t *newbie, *target; 1480 isaf_t *outbound, *inbound; 1481 int rc, diagnostic; 1482 sadb_sa_t *assoc; 1483 keysock_out_t *kso; 1484 uint32_t newspi; 1485 1486 /* 1487 * Randomly generate a proposed SPI value 1488 */ 1489 (void) random_get_pseudo_bytes((uint8_t *)&newspi, sizeof (uint32_t)); 1490 newbie = sadb_getspi(ksi, newspi, &diagnostic, 1491 espstack->ipsecesp_netstack); 1492 1493 if (newbie == NULL) { 1494 sadb_pfkey_error(espstack->esp_pfkey_q, mp, ENOMEM, diagnostic, 1495 ksi->ks_in_serial); 1496 return; 1497 } else if (newbie == (ipsa_t *)-1) { 1498 sadb_pfkey_error(espstack->esp_pfkey_q, mp, EINVAL, diagnostic, 1499 ksi->ks_in_serial); 1500 return; 1501 } 1502 1503 /* 1504 * XXX - We may randomly collide. We really should recover from this. 1505 * Unfortunately, that could require spending way-too-much-time 1506 * in here. For now, let the user retry. 1507 */ 1508 1509 if (newbie->ipsa_addrfam == AF_INET6) { 1510 outbound = OUTBOUND_BUCKET_V6(&espstack->esp_sadb.s_v6, 1511 *(uint32_t *)(newbie->ipsa_dstaddr)); 1512 inbound = INBOUND_BUCKET(&espstack->esp_sadb.s_v6, 1513 newbie->ipsa_spi); 1514 } else { 1515 ASSERT(newbie->ipsa_addrfam == AF_INET); 1516 outbound = OUTBOUND_BUCKET_V4(&espstack->esp_sadb.s_v4, 1517 *(uint32_t *)(newbie->ipsa_dstaddr)); 1518 inbound = INBOUND_BUCKET(&espstack->esp_sadb.s_v4, 1519 newbie->ipsa_spi); 1520 } 1521 1522 mutex_enter(&outbound->isaf_lock); 1523 mutex_enter(&inbound->isaf_lock); 1524 1525 /* 1526 * Check for collisions (i.e. did sadb_getspi() return with something 1527 * that already exists?). 1528 * 1529 * Try outbound first. Even though SADB_GETSPI is traditionally 1530 * for inbound SAs, you never know what a user might do. 1531 */ 1532 target = ipsec_getassocbyspi(outbound, newbie->ipsa_spi, 1533 newbie->ipsa_srcaddr, newbie->ipsa_dstaddr, newbie->ipsa_addrfam); 1534 if (target == NULL) { 1535 target = ipsec_getassocbyspi(inbound, newbie->ipsa_spi, 1536 newbie->ipsa_srcaddr, newbie->ipsa_dstaddr, 1537 newbie->ipsa_addrfam); 1538 } 1539 1540 /* 1541 * I don't have collisions elsewhere! 1542 * (Nor will I because I'm still holding inbound/outbound locks.) 1543 */ 1544 1545 if (target != NULL) { 1546 rc = EEXIST; 1547 IPSA_REFRELE(target); 1548 } else { 1549 /* 1550 * sadb_insertassoc() also checks for collisions, so 1551 * if there's a colliding entry, rc will be set 1552 * to EEXIST. 1553 */ 1554 rc = sadb_insertassoc(newbie, inbound); 1555 newbie->ipsa_hardexpiretime = gethrestime_sec(); 1556 newbie->ipsa_hardexpiretime += 1557 espstack->ipsecesp_larval_timeout; 1558 } 1559 1560 /* 1561 * Can exit outbound mutex. Hold inbound until we're done 1562 * with newbie. 1563 */ 1564 mutex_exit(&outbound->isaf_lock); 1565 1566 if (rc != 0) { 1567 mutex_exit(&inbound->isaf_lock); 1568 IPSA_REFRELE(newbie); 1569 sadb_pfkey_error(espstack->esp_pfkey_q, mp, rc, 1570 SADB_X_DIAGNOSTIC_NONE, ksi->ks_in_serial); 1571 return; 1572 } 1573 1574 1575 /* Can write here because I'm still holding the bucket lock. */ 1576 newbie->ipsa_type = SADB_SATYPE_ESP; 1577 1578 /* 1579 * Construct successful return message. We have one thing going 1580 * for us in PF_KEY v2. That's the fact that 1581 * sizeof (sadb_spirange_t) == sizeof (sadb_sa_t) 1582 */ 1583 assoc = (sadb_sa_t *)ksi->ks_in_extv[SADB_EXT_SPIRANGE]; 1584 assoc->sadb_sa_exttype = SADB_EXT_SA; 1585 assoc->sadb_sa_spi = newbie->ipsa_spi; 1586 *((uint64_t *)(&assoc->sadb_sa_replay)) = 0; 1587 mutex_exit(&inbound->isaf_lock); 1588 1589 /* Convert KEYSOCK_IN to KEYSOCK_OUT. */ 1590 kso = (keysock_out_t *)ksi; 1591 kso->ks_out_len = sizeof (*kso); 1592 kso->ks_out_serial = ksi->ks_in_serial; 1593 kso->ks_out_type = KEYSOCK_OUT; 1594 1595 /* 1596 * Can safely putnext() to esp_pfkey_q, because this is a turnaround 1597 * from the esp_pfkey_q. 1598 */ 1599 putnext(espstack->esp_pfkey_q, mp); 1600 } 1601 1602 /* 1603 * Insert the ESP header into a packet. Duplicate an mblk, and insert a newly 1604 * allocated mblk with the ESP header in between the two. 1605 */ 1606 static boolean_t 1607 esp_insert_esp(mblk_t *mp, mblk_t *esp_mp, uint_t divpoint, 1608 ipsecesp_stack_t *espstack) 1609 { 1610 mblk_t *split_mp = mp; 1611 uint_t wheretodiv = divpoint; 1612 1613 while ((split_mp->b_wptr - split_mp->b_rptr) < wheretodiv) { 1614 wheretodiv -= (split_mp->b_wptr - split_mp->b_rptr); 1615 split_mp = split_mp->b_cont; 1616 ASSERT(split_mp != NULL); 1617 } 1618 1619 if (split_mp->b_wptr - split_mp->b_rptr != wheretodiv) { 1620 mblk_t *scratch; 1621 1622 /* "scratch" is the 2nd half, split_mp is the first. */ 1623 scratch = dupb(split_mp); 1624 if (scratch == NULL) { 1625 esp1dbg(espstack, 1626 ("esp_insert_esp: can't allocate scratch.\n")); 1627 return (B_FALSE); 1628 } 1629 /* NOTE: dupb() doesn't set b_cont appropriately. */ 1630 scratch->b_cont = split_mp->b_cont; 1631 scratch->b_rptr += wheretodiv; 1632 split_mp->b_wptr = split_mp->b_rptr + wheretodiv; 1633 split_mp->b_cont = scratch; 1634 } 1635 /* 1636 * At this point, split_mp is exactly "wheretodiv" bytes long, and 1637 * holds the end of the pre-ESP part of the datagram. 1638 */ 1639 esp_mp->b_cont = split_mp->b_cont; 1640 split_mp->b_cont = esp_mp; 1641 1642 return (B_TRUE); 1643 } 1644 1645 /* 1646 * Section 7 of RFC 3947 says: 1647 * 1648 * 7. Recovering from the Expiring NAT Mappings 1649 * 1650 * There are cases where NAT box decides to remove mappings that are still 1651 * alive (for example, when the keepalive interval is too long, or when the 1652 * NAT box is rebooted). To recover from this, ends that are NOT behind 1653 * NAT SHOULD use the last valid UDP encapsulated IKE or IPsec packet from 1654 * the other end to determine which IP and port addresses should be used. 1655 * The host behind dynamic NAT MUST NOT do this, as otherwise it opens a 1656 * DoS attack possibility because the IP address or port of the other host 1657 * will not change (it is not behind NAT). 1658 * 1659 * Keepalives cannot be used for these purposes, as they are not 1660 * authenticated, but any IKE authenticated IKE packet or ESP packet can be 1661 * used to detect whether the IP address or the port has changed. 1662 * 1663 * The following function will check an SA and its explicitly-set pair to see 1664 * if the NAT-T remote port matches the received packet (which must have 1665 * passed ESP authentication, see esp_in_done() for the caller context). If 1666 * there is a mismatch, the SAs are updated. It is not important if we race 1667 * with a transmitting thread, as if there is a transmitting thread, it will 1668 * merely emit a packet that will most-likely be dropped. 1669 * 1670 * "ports" are ordered src,dst, and assoc is an inbound SA, where src should 1671 * match ipsa_remote_nat_port and dst should match ipsa_local_nat_port. 1672 */ 1673 #ifdef _LITTLE_ENDIAN 1674 #define FIRST_16(x) ((x) & 0xFFFF) 1675 #define NEXT_16(x) (((x) >> 16) & 0xFFFF) 1676 #else 1677 #define FIRST_16(x) (((x) >> 16) & 0xFFFF) 1678 #define NEXT_16(x) ((x) & 0xFFFF) 1679 #endif 1680 static void 1681 esp_port_freshness(uint32_t ports, ipsa_t *assoc) 1682 { 1683 uint16_t remote = FIRST_16(ports); 1684 uint16_t local = NEXT_16(ports); 1685 ipsa_t *outbound_peer; 1686 isaf_t *bucket; 1687 ipsecesp_stack_t *espstack = assoc->ipsa_netstack->netstack_ipsecesp; 1688 1689 /* We found a conn_t, therefore local != 0. */ 1690 ASSERT(local != 0); 1691 /* Assume an IPv4 SA. */ 1692 ASSERT(assoc->ipsa_addrfam == AF_INET); 1693 1694 /* 1695 * On-the-wire rport == 0 means something's very wrong. 1696 * An unpaired SA is also useless to us. 1697 * If we are behind the NAT, don't bother. 1698 * A zero local NAT port defaults to 4500, so check that too. 1699 * And, of course, if the ports already match, we don't need to 1700 * bother. 1701 */ 1702 if (remote == 0 || assoc->ipsa_otherspi == 0 || 1703 (assoc->ipsa_flags & IPSA_F_BEHIND_NAT) || 1704 (assoc->ipsa_remote_nat_port == 0 && 1705 remote == htons(IPPORT_IKE_NATT)) || 1706 remote == assoc->ipsa_remote_nat_port) 1707 return; 1708 1709 /* Try and snag the peer. NOTE: Assume IPv4 for now. */ 1710 bucket = OUTBOUND_BUCKET_V4(&(espstack->esp_sadb.s_v4), 1711 assoc->ipsa_srcaddr[0]); 1712 mutex_enter(&bucket->isaf_lock); 1713 outbound_peer = ipsec_getassocbyspi(bucket, assoc->ipsa_otherspi, 1714 assoc->ipsa_dstaddr, assoc->ipsa_srcaddr, AF_INET); 1715 mutex_exit(&bucket->isaf_lock); 1716 1717 /* We probably lost a race to a deleting or expiring thread. */ 1718 if (outbound_peer == NULL) 1719 return; 1720 1721 /* 1722 * Hold the mutexes for both SAs so we don't race another inbound 1723 * thread. A lock-entry order shouldn't matter, since all other 1724 * per-ipsa locks are individually held-then-released. 1725 * 1726 * Luckily, this has nothing to do with the remote-NAT address, 1727 * so we don't have to re-scribble the cached-checksum differential. 1728 */ 1729 mutex_enter(&outbound_peer->ipsa_lock); 1730 mutex_enter(&assoc->ipsa_lock); 1731 outbound_peer->ipsa_remote_nat_port = assoc->ipsa_remote_nat_port = 1732 remote; 1733 mutex_exit(&assoc->ipsa_lock); 1734 mutex_exit(&outbound_peer->ipsa_lock); 1735 IPSA_REFRELE(outbound_peer); 1736 ESP_BUMP_STAT(espstack, sa_port_renumbers); 1737 } 1738 1739 /* 1740 * Finish processing of an inbound ESP packet after processing by the 1741 * crypto framework. 1742 * - Remove the ESP header. 1743 * - Send packet back to IP. 1744 * If authentication was performed on the packet, this function is called 1745 * only if the authentication succeeded. 1746 * On success returns B_TRUE, on failure returns B_FALSE and frees the 1747 * mblk chain ipsec_in_mp. 1748 */ 1749 static ipsec_status_t 1750 esp_in_done(mblk_t *ipsec_in_mp) 1751 { 1752 ipsec_in_t *ii = (ipsec_in_t *)ipsec_in_mp->b_rptr; 1753 mblk_t *data_mp; 1754 ipsa_t *assoc; 1755 uint_t espstart; 1756 uint32_t ivlen = 0; 1757 uint_t processed_len; 1758 esph_t *esph; 1759 kstat_named_t *counter; 1760 boolean_t is_natt; 1761 netstack_t *ns = ii->ipsec_in_ns; 1762 ipsecesp_stack_t *espstack = ns->netstack_ipsecesp; 1763 ipsec_stack_t *ipss = ns->netstack_ipsec; 1764 1765 assoc = ii->ipsec_in_esp_sa; 1766 ASSERT(assoc != NULL); 1767 1768 is_natt = ((assoc->ipsa_flags & IPSA_F_NATT) != 0); 1769 1770 /* get the pointer to the ESP header */ 1771 if (assoc->ipsa_encr_alg == SADB_EALG_NULL) { 1772 /* authentication-only ESP */ 1773 espstart = ii->ipsec_in_crypto_data.cd_offset; 1774 processed_len = ii->ipsec_in_crypto_data.cd_length; 1775 } else { 1776 /* encryption present */ 1777 ivlen = assoc->ipsa_iv_len; 1778 if (assoc->ipsa_auth_alg == SADB_AALG_NONE) { 1779 /* encryption-only ESP */ 1780 espstart = ii->ipsec_in_crypto_data.cd_offset - 1781 sizeof (esph_t) - assoc->ipsa_iv_len; 1782 processed_len = ii->ipsec_in_crypto_data.cd_length + 1783 ivlen; 1784 } else { 1785 /* encryption with authentication */ 1786 espstart = ii->ipsec_in_crypto_dual_data.dd_offset1; 1787 processed_len = ii->ipsec_in_crypto_dual_data.dd_len2 + 1788 ivlen; 1789 } 1790 } 1791 1792 data_mp = ipsec_in_mp->b_cont; 1793 esph = (esph_t *)(data_mp->b_rptr + espstart); 1794 1795 if (assoc->ipsa_auth_alg != IPSA_AALG_NONE) { 1796 /* authentication passed if we reach this point */ 1797 ESP_BUMP_STAT(espstack, good_auth); 1798 data_mp->b_wptr -= assoc->ipsa_mac_len; 1799 1800 /* 1801 * Check replay window here! 1802 * For right now, assume keysock will set the replay window 1803 * size to zero for SAs that have an unspecified sender. 1804 * This may change... 1805 */ 1806 1807 if (!sadb_replay_check(assoc, esph->esph_replay)) { 1808 /* 1809 * Log the event. As of now we print out an event. 1810 * Do not print the replay failure number, or else 1811 * syslog cannot collate the error messages. Printing 1812 * the replay number that failed opens a denial-of- 1813 * service attack. 1814 */ 1815 ipsec_assocfailure(info.mi_idnum, 0, 0, 1816 SL_ERROR | SL_WARN, 1817 "Replay failed for ESP spi 0x%x, dst %s.\n", 1818 assoc->ipsa_spi, assoc->ipsa_dstaddr, 1819 assoc->ipsa_addrfam, espstack->ipsecesp_netstack); 1820 ESP_BUMP_STAT(espstack, replay_failures); 1821 counter = DROPPER(ipss, ipds_esp_replay); 1822 goto drop_and_bail; 1823 } 1824 1825 if (is_natt) 1826 esp_port_freshness(ii->ipsec_in_esp_udp_ports, assoc); 1827 } 1828 1829 esp_set_usetime(assoc, B_TRUE); 1830 1831 if (!esp_age_bytes(assoc, processed_len, B_TRUE)) { 1832 /* The ipsa has hit hard expiration, LOG and AUDIT. */ 1833 ipsec_assocfailure(info.mi_idnum, 0, 0, 1834 SL_ERROR | SL_WARN, 1835 "ESP association 0x%x, dst %s had bytes expire.\n", 1836 assoc->ipsa_spi, assoc->ipsa_dstaddr, assoc->ipsa_addrfam, 1837 espstack->ipsecesp_netstack); 1838 ESP_BUMP_STAT(espstack, bytes_expired); 1839 counter = DROPPER(ipss, ipds_esp_bytes_expire); 1840 goto drop_and_bail; 1841 } 1842 1843 /* 1844 * Remove ESP header and padding from packet. I hope the compiler 1845 * spews "branch, predict taken" code for this. 1846 */ 1847 1848 if (esp_strip_header(data_mp, ii->ipsec_in_v4, ivlen, &counter, 1849 espstack)) { 1850 if (is_natt) 1851 return (esp_fix_natt_checksums(data_mp, assoc)); 1852 return (IPSEC_STATUS_SUCCESS); 1853 } 1854 1855 esp1dbg(espstack, ("esp_in_done: esp_strip_header() failed\n")); 1856 drop_and_bail: 1857 IP_ESP_BUMP_STAT(ipss, in_discards); 1858 /* 1859 * TODO: Extract inbound interface from the IPSEC_IN message's 1860 * ii->ipsec_in_rill_index. 1861 */ 1862 ip_drop_packet(ipsec_in_mp, B_TRUE, NULL, NULL, counter, 1863 &espstack->esp_dropper); 1864 return (IPSEC_STATUS_FAILED); 1865 } 1866 1867 /* 1868 * Called upon failing the inbound ICV check. The message passed as 1869 * argument is freed. 1870 */ 1871 static void 1872 esp_log_bad_auth(mblk_t *ipsec_in) 1873 { 1874 ipsec_in_t *ii = (ipsec_in_t *)ipsec_in->b_rptr; 1875 ipsa_t *assoc = ii->ipsec_in_esp_sa; 1876 netstack_t *ns = ii->ipsec_in_ns; 1877 ipsecesp_stack_t *espstack = ns->netstack_ipsecesp; 1878 ipsec_stack_t *ipss = ns->netstack_ipsec; 1879 1880 /* 1881 * Log the event. Don't print to the console, block 1882 * potential denial-of-service attack. 1883 */ 1884 ESP_BUMP_STAT(espstack, bad_auth); 1885 1886 ipsec_assocfailure(info.mi_idnum, 0, 0, SL_ERROR | SL_WARN, 1887 "ESP Authentication failed for spi 0x%x, dst %s.\n", 1888 assoc->ipsa_spi, assoc->ipsa_dstaddr, assoc->ipsa_addrfam, 1889 espstack->ipsecesp_netstack); 1890 1891 IP_ESP_BUMP_STAT(ipss, in_discards); 1892 /* 1893 * TODO: Extract inbound interface from the IPSEC_IN 1894 * message's ii->ipsec_in_rill_index. 1895 */ 1896 ip_drop_packet(ipsec_in, B_TRUE, NULL, NULL, 1897 DROPPER(ipss, ipds_esp_bad_auth), 1898 &espstack->esp_dropper); 1899 } 1900 1901 1902 /* 1903 * Invoked for outbound packets after ESP processing. If the packet 1904 * also requires AH, performs the AH SA selection and AH processing. 1905 * Returns B_TRUE if the AH processing was not needed or if it was 1906 * performed successfully. Returns B_FALSE and consumes the passed mblk 1907 * if AH processing was required but could not be performed. 1908 */ 1909 static boolean_t 1910 esp_do_outbound_ah(mblk_t *ipsec_mp) 1911 { 1912 ipsec_out_t *io = (ipsec_out_t *)ipsec_mp->b_rptr; 1913 ipsec_status_t ipsec_rc; 1914 ipsec_action_t *ap; 1915 1916 ap = io->ipsec_out_act; 1917 if (ap == NULL) { 1918 ipsec_policy_t *pp = io->ipsec_out_policy; 1919 ap = pp->ipsp_act; 1920 } 1921 1922 if (!ap->ipa_want_ah) 1923 return (B_TRUE); 1924 1925 ASSERT(io->ipsec_out_ah_done == B_FALSE); 1926 1927 if (io->ipsec_out_ah_sa == NULL) { 1928 if (!ipsec_outbound_sa(ipsec_mp, IPPROTO_AH)) { 1929 sadb_acquire(ipsec_mp, io, B_TRUE, B_FALSE); 1930 return (B_FALSE); 1931 } 1932 } 1933 ASSERT(io->ipsec_out_ah_sa != NULL); 1934 1935 io->ipsec_out_ah_done = B_TRUE; 1936 ipsec_rc = io->ipsec_out_ah_sa->ipsa_output_func(ipsec_mp); 1937 return (ipsec_rc == IPSEC_STATUS_SUCCESS); 1938 } 1939 1940 1941 /* 1942 * Kernel crypto framework callback invoked after completion of async 1943 * crypto requests. 1944 */ 1945 static void 1946 esp_kcf_callback(void *arg, int status) 1947 { 1948 mblk_t *ipsec_mp = (mblk_t *)arg; 1949 ipsec_in_t *ii = (ipsec_in_t *)ipsec_mp->b_rptr; 1950 ipsec_out_t *io = (ipsec_out_t *)ipsec_mp->b_rptr; 1951 boolean_t is_inbound = (ii->ipsec_in_type == IPSEC_IN); 1952 netstackid_t stackid; 1953 netstack_t *ns, *ns_arg; 1954 ipsecesp_stack_t *espstack; 1955 ipsec_stack_t *ipss; 1956 1957 ASSERT(ipsec_mp->b_cont != NULL); 1958 1959 if (is_inbound) { 1960 stackid = ii->ipsec_in_stackid; 1961 ns_arg = ii->ipsec_in_ns; 1962 } else { 1963 stackid = io->ipsec_out_stackid; 1964 ns_arg = io->ipsec_out_ns; 1965 } 1966 1967 /* 1968 * Verify that the netstack is still around; could have vanished 1969 * while kEf was doing its work. 1970 */ 1971 ns = netstack_find_by_stackid(stackid); 1972 if (ns == NULL || ns != ns_arg) { 1973 /* Disappeared on us */ 1974 if (ns != NULL) 1975 netstack_rele(ns); 1976 freemsg(ipsec_mp); 1977 return; 1978 } 1979 1980 espstack = ns->netstack_ipsecesp; 1981 ipss = ns->netstack_ipsec; 1982 1983 if (status == CRYPTO_SUCCESS) { 1984 if (is_inbound) { 1985 if (esp_in_done(ipsec_mp) != IPSEC_STATUS_SUCCESS) { 1986 netstack_rele(ns); 1987 return; 1988 } 1989 /* finish IPsec processing */ 1990 ip_fanout_proto_again(ipsec_mp, NULL, NULL, NULL); 1991 } else { 1992 /* 1993 * If a ICV was computed, it was stored by the 1994 * crypto framework at the end of the packet. 1995 */ 1996 ipha_t *ipha = (ipha_t *)ipsec_mp->b_cont->b_rptr; 1997 1998 esp_set_usetime(io->ipsec_out_esp_sa, B_FALSE); 1999 /* NAT-T packet. */ 2000 if (ipha->ipha_protocol == IPPROTO_UDP) 2001 esp_prepare_udp(ns, ipsec_mp->b_cont, ipha); 2002 2003 /* do AH processing if needed */ 2004 if (!esp_do_outbound_ah(ipsec_mp)) { 2005 netstack_rele(ns); 2006 return; 2007 } 2008 /* finish IPsec processing */ 2009 if (IPH_HDR_VERSION(ipha) == IP_VERSION) { 2010 ip_wput_ipsec_out(NULL, ipsec_mp, ipha, NULL, 2011 NULL); 2012 } else { 2013 ip6_t *ip6h = (ip6_t *)ipha; 2014 ip_wput_ipsec_out_v6(NULL, ipsec_mp, ip6h, 2015 NULL, NULL); 2016 } 2017 } 2018 2019 } else if (status == CRYPTO_INVALID_MAC) { 2020 esp_log_bad_auth(ipsec_mp); 2021 2022 } else { 2023 esp1dbg(espstack, 2024 ("esp_kcf_callback: crypto failed with 0x%x\n", 2025 status)); 2026 ESP_BUMP_STAT(espstack, crypto_failures); 2027 if (is_inbound) 2028 IP_ESP_BUMP_STAT(ipss, in_discards); 2029 else 2030 ESP_BUMP_STAT(espstack, out_discards); 2031 ip_drop_packet(ipsec_mp, is_inbound, NULL, NULL, 2032 DROPPER(ipss, ipds_esp_crypto_failed), 2033 &espstack->esp_dropper); 2034 } 2035 netstack_rele(ns); 2036 } 2037 2038 /* 2039 * Invoked on crypto framework failure during inbound and outbound processing. 2040 */ 2041 static void 2042 esp_crypto_failed(mblk_t *mp, boolean_t is_inbound, int kef_rc, 2043 ipsecesp_stack_t *espstack) 2044 { 2045 ipsec_stack_t *ipss = espstack->ipsecesp_netstack->netstack_ipsec; 2046 2047 esp1dbg(espstack, ("crypto failed for %s ESP with 0x%x\n", 2048 is_inbound ? "inbound" : "outbound", kef_rc)); 2049 ip_drop_packet(mp, is_inbound, NULL, NULL, 2050 DROPPER(ipss, ipds_esp_crypto_failed), 2051 &espstack->esp_dropper); 2052 ESP_BUMP_STAT(espstack, crypto_failures); 2053 if (is_inbound) 2054 IP_ESP_BUMP_STAT(ipss, in_discards); 2055 else 2056 ESP_BUMP_STAT(espstack, out_discards); 2057 } 2058 2059 #define ESP_INIT_CALLREQ(_cr) { \ 2060 (_cr)->cr_flag = CRYPTO_SKIP_REQID|CRYPTO_RESTRICTED; \ 2061 (_cr)->cr_callback_arg = ipsec_mp; \ 2062 (_cr)->cr_callback_func = esp_kcf_callback; \ 2063 } 2064 2065 #define ESP_INIT_CRYPTO_MAC(mac, icvlen, icvbuf) { \ 2066 (mac)->cd_format = CRYPTO_DATA_RAW; \ 2067 (mac)->cd_offset = 0; \ 2068 (mac)->cd_length = icvlen; \ 2069 (mac)->cd_raw.iov_base = (char *)icvbuf; \ 2070 (mac)->cd_raw.iov_len = icvlen; \ 2071 } 2072 2073 #define ESP_INIT_CRYPTO_DATA(data, mp, off, len) { \ 2074 if (MBLKL(mp) >= (len) + (off)) { \ 2075 (data)->cd_format = CRYPTO_DATA_RAW; \ 2076 (data)->cd_raw.iov_base = (char *)(mp)->b_rptr; \ 2077 (data)->cd_raw.iov_len = MBLKL(mp); \ 2078 (data)->cd_offset = off; \ 2079 } else { \ 2080 (data)->cd_format = CRYPTO_DATA_MBLK; \ 2081 (data)->cd_mp = mp; \ 2082 (data)->cd_offset = off; \ 2083 } \ 2084 (data)->cd_length = len; \ 2085 } 2086 2087 #define ESP_INIT_CRYPTO_DUAL_DATA(data, mp, off1, len1, off2, len2) { \ 2088 (data)->dd_format = CRYPTO_DATA_MBLK; \ 2089 (data)->dd_mp = mp; \ 2090 (data)->dd_len1 = len1; \ 2091 (data)->dd_offset1 = off1; \ 2092 (data)->dd_len2 = len2; \ 2093 (data)->dd_offset2 = off2; \ 2094 } 2095 2096 static ipsec_status_t 2097 esp_submit_req_inbound(mblk_t *ipsec_mp, ipsa_t *assoc, uint_t esph_offset) 2098 { 2099 ipsec_in_t *ii = (ipsec_in_t *)ipsec_mp->b_rptr; 2100 boolean_t do_auth; 2101 uint_t auth_offset, msg_len, auth_len; 2102 crypto_call_req_t call_req; 2103 mblk_t *esp_mp; 2104 int kef_rc = CRYPTO_FAILED; 2105 uint_t icv_len = assoc->ipsa_mac_len; 2106 crypto_ctx_template_t auth_ctx_tmpl; 2107 boolean_t do_encr; 2108 uint_t encr_offset, encr_len; 2109 uint_t iv_len = assoc->ipsa_iv_len; 2110 crypto_ctx_template_t encr_ctx_tmpl; 2111 netstack_t *ns = ii->ipsec_in_ns; 2112 ipsecesp_stack_t *espstack = ns->netstack_ipsecesp; 2113 ipsec_stack_t *ipss = ns->netstack_ipsec; 2114 2115 ASSERT(ii->ipsec_in_type == IPSEC_IN); 2116 2117 /* 2118 * In case kEF queues and calls back, keep netstackid_t for 2119 * verification that the IP instance is still around in 2120 * esp_kcf_callback(). 2121 */ 2122 ii->ipsec_in_stackid = ns->netstack_stackid; 2123 2124 do_auth = assoc->ipsa_auth_alg != SADB_AALG_NONE; 2125 do_encr = assoc->ipsa_encr_alg != SADB_EALG_NULL; 2126 2127 /* 2128 * An inbound packet is of the form: 2129 * IPSEC_IN -> [IP,options,ESP,IV,data,ICV,pad] 2130 */ 2131 esp_mp = ipsec_mp->b_cont; 2132 msg_len = MBLKL(esp_mp); 2133 2134 ESP_INIT_CALLREQ(&call_req); 2135 2136 if (do_auth) { 2137 /* force asynchronous processing? */ 2138 if (ipss->ipsec_algs_exec_mode[IPSEC_ALG_AUTH] == 2139 IPSEC_ALGS_EXEC_ASYNC) 2140 call_req.cr_flag |= CRYPTO_ALWAYS_QUEUE; 2141 2142 /* authentication context template */ 2143 IPSEC_CTX_TMPL(assoc, ipsa_authtmpl, IPSEC_ALG_AUTH, 2144 auth_ctx_tmpl); 2145 2146 /* ICV to be verified */ 2147 ESP_INIT_CRYPTO_MAC(&ii->ipsec_in_crypto_mac, 2148 icv_len, esp_mp->b_wptr - icv_len); 2149 2150 /* authentication starts at the ESP header */ 2151 auth_offset = esph_offset; 2152 auth_len = msg_len - auth_offset - icv_len; 2153 if (!do_encr) { 2154 /* authentication only */ 2155 /* initialize input data argument */ 2156 ESP_INIT_CRYPTO_DATA(&ii->ipsec_in_crypto_data, 2157 esp_mp, auth_offset, auth_len); 2158 2159 /* call the crypto framework */ 2160 kef_rc = crypto_mac_verify(&assoc->ipsa_amech, 2161 &ii->ipsec_in_crypto_data, 2162 &assoc->ipsa_kcfauthkey, auth_ctx_tmpl, 2163 &ii->ipsec_in_crypto_mac, &call_req); 2164 } 2165 } 2166 2167 if (do_encr) { 2168 /* force asynchronous processing? */ 2169 if (ipss->ipsec_algs_exec_mode[IPSEC_ALG_ENCR] == 2170 IPSEC_ALGS_EXEC_ASYNC) 2171 call_req.cr_flag |= CRYPTO_ALWAYS_QUEUE; 2172 2173 /* encryption template */ 2174 IPSEC_CTX_TMPL(assoc, ipsa_encrtmpl, IPSEC_ALG_ENCR, 2175 encr_ctx_tmpl); 2176 2177 /* skip IV, since it is passed separately */ 2178 encr_offset = esph_offset + sizeof (esph_t) + iv_len; 2179 encr_len = msg_len - encr_offset; 2180 2181 if (!do_auth) { 2182 /* decryption only */ 2183 /* initialize input data argument */ 2184 ESP_INIT_CRYPTO_DATA(&ii->ipsec_in_crypto_data, 2185 esp_mp, encr_offset, encr_len); 2186 2187 /* specify IV */ 2188 ii->ipsec_in_crypto_data.cd_miscdata = 2189 (char *)esp_mp->b_rptr + sizeof (esph_t) + 2190 esph_offset; 2191 2192 /* call the crypto framework */ 2193 kef_rc = crypto_decrypt(&assoc->ipsa_emech, 2194 &ii->ipsec_in_crypto_data, 2195 &assoc->ipsa_kcfencrkey, encr_ctx_tmpl, 2196 NULL, &call_req); 2197 } 2198 } 2199 2200 if (do_auth && do_encr) { 2201 /* dual operation */ 2202 /* initialize input data argument */ 2203 ESP_INIT_CRYPTO_DUAL_DATA(&ii->ipsec_in_crypto_dual_data, 2204 esp_mp, auth_offset, auth_len, 2205 encr_offset, encr_len - icv_len); 2206 2207 /* specify IV */ 2208 ii->ipsec_in_crypto_dual_data.dd_miscdata = 2209 (char *)esp_mp->b_rptr + sizeof (esph_t) + esph_offset; 2210 2211 /* call the framework */ 2212 kef_rc = crypto_mac_verify_decrypt(&assoc->ipsa_amech, 2213 &assoc->ipsa_emech, &ii->ipsec_in_crypto_dual_data, 2214 &assoc->ipsa_kcfauthkey, &assoc->ipsa_kcfencrkey, 2215 auth_ctx_tmpl, encr_ctx_tmpl, &ii->ipsec_in_crypto_mac, 2216 NULL, &call_req); 2217 } 2218 2219 switch (kef_rc) { 2220 case CRYPTO_SUCCESS: 2221 ESP_BUMP_STAT(espstack, crypto_sync); 2222 return (esp_in_done(ipsec_mp)); 2223 case CRYPTO_QUEUED: 2224 /* esp_kcf_callback() will be invoked on completion */ 2225 ESP_BUMP_STAT(espstack, crypto_async); 2226 return (IPSEC_STATUS_PENDING); 2227 case CRYPTO_INVALID_MAC: 2228 ESP_BUMP_STAT(espstack, crypto_sync); 2229 esp_log_bad_auth(ipsec_mp); 2230 return (IPSEC_STATUS_FAILED); 2231 } 2232 2233 esp_crypto_failed(ipsec_mp, B_TRUE, kef_rc, espstack); 2234 return (IPSEC_STATUS_FAILED); 2235 } 2236 2237 /* 2238 * Compute the IP and UDP checksums -- common code for both keepalives and 2239 * actual ESP-in-UDP packets. Be flexible with multiple mblks because ESP 2240 * uses mblk-insertion to insert the UDP header. 2241 * TODO - If there is an easy way to prep a packet for HW checksums, make 2242 * it happen here. 2243 */ 2244 static void 2245 esp_prepare_udp(netstack_t *ns, mblk_t *mp, ipha_t *ipha) 2246 { 2247 int offset; 2248 uint32_t cksum; 2249 uint16_t *arr; 2250 mblk_t *udpmp = mp; 2251 uint_t hlen = IPH_HDR_LENGTH(ipha); 2252 2253 ASSERT(MBLKL(mp) >= sizeof (ipha_t)); 2254 2255 ipha->ipha_hdr_checksum = 0; 2256 ipha->ipha_hdr_checksum = ip_csum_hdr(ipha); 2257 2258 if (ns->netstack_udp->us_do_checksum) { 2259 ASSERT(MBLKL(udpmp) >= sizeof (udpha_t)); 2260 /* arr points to the IP header. */ 2261 arr = (uint16_t *)ipha; 2262 IP_STAT(ns->netstack_ip, ip_out_sw_cksum); 2263 IP_STAT_UPDATE(ns->netstack_ip, ip_udp_out_sw_cksum_bytes, 2264 ntohs(htons(ipha->ipha_length) - hlen)); 2265 /* arr[6-9] are the IP addresses. */ 2266 cksum = IP_UDP_CSUM_COMP + arr[6] + arr[7] + arr[8] + arr[9] + 2267 ntohs(htons(ipha->ipha_length) - hlen); 2268 cksum = IP_CSUM(mp, hlen, cksum); 2269 offset = hlen + UDP_CHECKSUM_OFFSET; 2270 while (offset >= MBLKL(udpmp)) { 2271 offset -= MBLKL(udpmp); 2272 udpmp = udpmp->b_cont; 2273 } 2274 /* arr points to the UDP header's checksum field. */ 2275 arr = (uint16_t *)(udpmp->b_rptr + offset); 2276 *arr = cksum; 2277 } 2278 } 2279 2280 /* 2281 * Send a one-byte UDP NAT-T keepalive. Construct an IPSEC_OUT too that'll 2282 * get fed into esp_send_udp/ip_wput_ipsec_out. 2283 */ 2284 void 2285 ipsecesp_send_keepalive(ipsa_t *assoc) 2286 { 2287 mblk_t *mp = NULL, *ipsec_mp = NULL; 2288 ipha_t *ipha; 2289 udpha_t *udpha; 2290 ipsec_out_t *io; 2291 2292 ASSERT(!MUTEX_HELD(&assoc->ipsa_lock)); 2293 2294 mp = allocb(sizeof (ipha_t) + sizeof (udpha_t) + 1, BPRI_HI); 2295 if (mp == NULL) 2296 return; 2297 ipha = (ipha_t *)mp->b_rptr; 2298 ipha->ipha_version_and_hdr_length = IP_SIMPLE_HDR_VERSION; 2299 ipha->ipha_type_of_service = 0; 2300 ipha->ipha_length = htons(sizeof (ipha_t) + sizeof (udpha_t) + 1); 2301 /* Use the low-16 of the SPI so we have some clue where it came from. */ 2302 ipha->ipha_ident = *(((uint16_t *)(&assoc->ipsa_spi)) + 1); 2303 ipha->ipha_fragment_offset_and_flags = 0; /* Too small to fragment! */ 2304 ipha->ipha_ttl = 0xFF; 2305 ipha->ipha_protocol = IPPROTO_UDP; 2306 ipha->ipha_hdr_checksum = 0; 2307 ipha->ipha_src = assoc->ipsa_srcaddr[0]; 2308 ipha->ipha_dst = assoc->ipsa_dstaddr[0]; 2309 udpha = (udpha_t *)(ipha + 1); 2310 udpha->uha_src_port = (assoc->ipsa_local_nat_port != 0) ? 2311 assoc->ipsa_local_nat_port : htons(IPPORT_IKE_NATT); 2312 udpha->uha_dst_port = (assoc->ipsa_remote_nat_port != 0) ? 2313 assoc->ipsa_remote_nat_port : htons(IPPORT_IKE_NATT); 2314 udpha->uha_length = htons(sizeof (udpha_t) + 1); 2315 udpha->uha_checksum = 0; 2316 mp->b_wptr = (uint8_t *)(udpha + 1); 2317 *(mp->b_wptr++) = 0xFF; 2318 2319 ipsec_mp = ipsec_alloc_ipsec_out(assoc->ipsa_netstack); 2320 if (ipsec_mp == NULL) { 2321 freeb(mp); 2322 return; 2323 } 2324 ipsec_mp->b_cont = mp; 2325 io = (ipsec_out_t *)ipsec_mp->b_rptr; 2326 io->ipsec_out_zoneid = 2327 netstackid_to_zoneid(assoc->ipsa_netstack->netstack_stackid); 2328 2329 esp_prepare_udp(assoc->ipsa_netstack, mp, ipha); 2330 ip_wput_ipsec_out(NULL, ipsec_mp, ipha, NULL, NULL); 2331 } 2332 2333 static ipsec_status_t 2334 esp_submit_req_outbound(mblk_t *ipsec_mp, ipsa_t *assoc, uchar_t *icv_buf, 2335 uint_t payload_len) 2336 { 2337 ipsec_out_t *io = (ipsec_out_t *)ipsec_mp->b_rptr; 2338 uint_t auth_len; 2339 crypto_call_req_t call_req; 2340 mblk_t *esp_mp; 2341 int kef_rc = CRYPTO_FAILED; 2342 uint_t icv_len = assoc->ipsa_mac_len; 2343 crypto_ctx_template_t auth_ctx_tmpl; 2344 boolean_t do_auth; 2345 boolean_t do_encr; 2346 uint_t iv_len = assoc->ipsa_iv_len; 2347 crypto_ctx_template_t encr_ctx_tmpl; 2348 boolean_t is_natt = ((assoc->ipsa_flags & IPSA_F_NATT) != 0); 2349 size_t esph_offset = (is_natt ? UDPH_SIZE : 0); 2350 netstack_t *ns = io->ipsec_out_ns; 2351 ipsecesp_stack_t *espstack = ns->netstack_ipsecesp; 2352 ipsec_stack_t *ipss = ns->netstack_ipsec; 2353 2354 esp3dbg(espstack, ("esp_submit_req_outbound:%s", 2355 is_natt ? "natt" : "not natt")); 2356 2357 ASSERT(io->ipsec_out_type == IPSEC_OUT); 2358 2359 /* 2360 * In case kEF queues and calls back, keep netstackid_t for 2361 * verification that the IP instance is still around in 2362 * esp_kcf_callback(). 2363 */ 2364 io->ipsec_out_stackid = ns->netstack_stackid; 2365 2366 do_encr = assoc->ipsa_encr_alg != SADB_EALG_NULL; 2367 do_auth = assoc->ipsa_auth_alg != SADB_AALG_NONE; 2368 2369 /* 2370 * Outbound IPsec packets are of the form: 2371 * IPSEC_OUT -> [IP,options] -> [ESP,IV] -> [data] -> [pad,ICV] 2372 * unless it's NATT, then it's 2373 * IPSEC_OUT -> [IP,options] -> [udp][ESP,IV] -> [data] -> [pad,ICV] 2374 * Get a pointer to the mblk containing the ESP header. 2375 */ 2376 ASSERT(ipsec_mp->b_cont != NULL && ipsec_mp->b_cont->b_cont != NULL); 2377 esp_mp = ipsec_mp->b_cont->b_cont; 2378 2379 ESP_INIT_CALLREQ(&call_req); 2380 2381 if (do_auth) { 2382 /* force asynchronous processing? */ 2383 if (ipss->ipsec_algs_exec_mode[IPSEC_ALG_AUTH] == 2384 IPSEC_ALGS_EXEC_ASYNC) 2385 call_req.cr_flag |= CRYPTO_ALWAYS_QUEUE; 2386 2387 /* authentication context template */ 2388 IPSEC_CTX_TMPL(assoc, ipsa_authtmpl, IPSEC_ALG_AUTH, 2389 auth_ctx_tmpl); 2390 2391 /* where to store the computed mac */ 2392 ESP_INIT_CRYPTO_MAC(&io->ipsec_out_crypto_mac, 2393 icv_len, icv_buf); 2394 2395 /* authentication starts at the ESP header */ 2396 auth_len = payload_len + iv_len + sizeof (esph_t); 2397 if (!do_encr) { 2398 /* authentication only */ 2399 /* initialize input data argument */ 2400 ESP_INIT_CRYPTO_DATA(&io->ipsec_out_crypto_data, 2401 esp_mp, esph_offset, auth_len); 2402 2403 /* call the crypto framework */ 2404 kef_rc = crypto_mac(&assoc->ipsa_amech, 2405 &io->ipsec_out_crypto_data, 2406 &assoc->ipsa_kcfauthkey, auth_ctx_tmpl, 2407 &io->ipsec_out_crypto_mac, &call_req); 2408 } 2409 } 2410 2411 if (do_encr) { 2412 /* force asynchronous processing? */ 2413 if (ipss->ipsec_algs_exec_mode[IPSEC_ALG_ENCR] == 2414 IPSEC_ALGS_EXEC_ASYNC) 2415 call_req.cr_flag |= CRYPTO_ALWAYS_QUEUE; 2416 2417 /* encryption context template */ 2418 IPSEC_CTX_TMPL(assoc, ipsa_encrtmpl, IPSEC_ALG_ENCR, 2419 encr_ctx_tmpl); 2420 2421 if (!do_auth) { 2422 /* encryption only, skip mblk that contains ESP hdr */ 2423 /* initialize input data argument */ 2424 ESP_INIT_CRYPTO_DATA(&io->ipsec_out_crypto_data, 2425 esp_mp->b_cont, 0, payload_len); 2426 2427 /* specify IV */ 2428 io->ipsec_out_crypto_data.cd_miscdata = 2429 (char *)esp_mp->b_rptr + sizeof (esph_t) + 2430 esph_offset; 2431 2432 /* call the crypto framework */ 2433 kef_rc = crypto_encrypt(&assoc->ipsa_emech, 2434 &io->ipsec_out_crypto_data, 2435 &assoc->ipsa_kcfencrkey, encr_ctx_tmpl, 2436 NULL, &call_req); 2437 } 2438 } 2439 2440 if (do_auth && do_encr) { 2441 /* 2442 * Encryption and authentication: 2443 * Pass the pointer to the mblk chain starting at the ESP 2444 * header to the framework. Skip the ESP header mblk 2445 * for encryption, which is reflected by an encryption 2446 * offset equal to the length of that mblk. Start 2447 * the authentication at the ESP header, i.e. use an 2448 * authentication offset of zero. 2449 */ 2450 ESP_INIT_CRYPTO_DUAL_DATA(&io->ipsec_out_crypto_dual_data, 2451 esp_mp, MBLKL(esp_mp), payload_len, esph_offset, auth_len); 2452 2453 /* specify IV */ 2454 io->ipsec_out_crypto_dual_data.dd_miscdata = 2455 (char *)esp_mp->b_rptr + sizeof (esph_t) + esph_offset; 2456 2457 /* call the framework */ 2458 kef_rc = crypto_encrypt_mac(&assoc->ipsa_emech, 2459 &assoc->ipsa_amech, NULL, 2460 &assoc->ipsa_kcfencrkey, &assoc->ipsa_kcfauthkey, 2461 encr_ctx_tmpl, auth_ctx_tmpl, 2462 &io->ipsec_out_crypto_dual_data, 2463 &io->ipsec_out_crypto_mac, &call_req); 2464 } 2465 2466 switch (kef_rc) { 2467 case CRYPTO_SUCCESS: 2468 ESP_BUMP_STAT(espstack, crypto_sync); 2469 esp_set_usetime(assoc, B_FALSE); 2470 if (is_natt) 2471 esp_prepare_udp(ns, ipsec_mp->b_cont, 2472 (ipha_t *)ipsec_mp->b_cont->b_rptr); 2473 return (IPSEC_STATUS_SUCCESS); 2474 case CRYPTO_QUEUED: 2475 /* esp_kcf_callback() will be invoked on completion */ 2476 ESP_BUMP_STAT(espstack, crypto_async); 2477 return (IPSEC_STATUS_PENDING); 2478 } 2479 2480 esp_crypto_failed(ipsec_mp, B_TRUE, kef_rc, espstack); 2481 return (IPSEC_STATUS_FAILED); 2482 } 2483 2484 /* 2485 * Handle outbound IPsec processing for IPv4 and IPv6 2486 * On success returns B_TRUE, on failure returns B_FALSE and frees the 2487 * mblk chain ipsec_in_mp. 2488 */ 2489 static ipsec_status_t 2490 esp_outbound(mblk_t *mp) 2491 { 2492 mblk_t *ipsec_out_mp, *data_mp, *espmp, *tailmp; 2493 ipsec_out_t *io; 2494 ipha_t *ipha; 2495 ip6_t *ip6h; 2496 esph_t *esph; 2497 uint_t af; 2498 uint8_t *nhp; 2499 uintptr_t divpoint, datalen, adj, padlen, i, alloclen; 2500 uintptr_t esplen = sizeof (esph_t); 2501 uint8_t protocol; 2502 ipsa_t *assoc; 2503 uint_t iv_len, mac_len = 0; 2504 uchar_t *icv_buf; 2505 udpha_t *udpha; 2506 boolean_t is_natt = B_FALSE; 2507 netstack_t *ns; 2508 ipsecesp_stack_t *espstack; 2509 ipsec_stack_t *ipss; 2510 2511 ipsec_out_mp = mp; 2512 data_mp = ipsec_out_mp->b_cont; 2513 2514 io = (ipsec_out_t *)ipsec_out_mp->b_rptr; 2515 ns = io->ipsec_out_ns; 2516 espstack = ns->netstack_ipsecesp; 2517 ipss = ns->netstack_ipsec; 2518 2519 ESP_BUMP_STAT(espstack, out_requests); 2520 2521 /* 2522 * <sigh> We have to copy the message here, because TCP (for example) 2523 * keeps a dupb() of the message lying around for retransmission. 2524 * Since ESP changes the whole of the datagram, we have to create our 2525 * own copy lest we clobber TCP's data. Since we have to copy anyway, 2526 * we might as well make use of msgpullup() and get the mblk into one 2527 * contiguous piece! 2528 */ 2529 ipsec_out_mp->b_cont = msgpullup(data_mp, -1); 2530 if (ipsec_out_mp->b_cont == NULL) { 2531 esp0dbg(("esp_outbound: msgpullup() failed, " 2532 "dropping packet.\n")); 2533 ipsec_out_mp->b_cont = data_mp; 2534 /* 2535 * TODO: Find the outbound IRE for this packet and 2536 * pass it to ip_drop_packet(). 2537 */ 2538 ip_drop_packet(ipsec_out_mp, B_FALSE, NULL, NULL, 2539 DROPPER(ipss, ipds_esp_nomem), 2540 &espstack->esp_dropper); 2541 return (IPSEC_STATUS_FAILED); 2542 } else { 2543 freemsg(data_mp); 2544 data_mp = ipsec_out_mp->b_cont; 2545 } 2546 2547 /* 2548 * Reality check.... 2549 */ 2550 2551 ipha = (ipha_t *)data_mp->b_rptr; /* So we can call esp_acquire(). */ 2552 2553 if (io->ipsec_out_v4) { 2554 af = AF_INET; 2555 divpoint = IPH_HDR_LENGTH(ipha); 2556 datalen = ntohs(ipha->ipha_length) - divpoint; 2557 nhp = (uint8_t *)&ipha->ipha_protocol; 2558 } else { 2559 ip6_pkt_t ipp; 2560 2561 af = AF_INET6; 2562 ip6h = (ip6_t *)ipha; 2563 bzero(&ipp, sizeof (ipp)); 2564 divpoint = ip_find_hdr_v6(data_mp, ip6h, &ipp, NULL); 2565 if (ipp.ipp_dstopts != NULL && 2566 ipp.ipp_dstopts->ip6d_nxt != IPPROTO_ROUTING) { 2567 /* 2568 * Destination options are tricky. If we get in here, 2569 * then we have a terminal header following the 2570 * destination options. We need to adjust backwards 2571 * so we insert ESP BEFORE the destination options 2572 * bag. (So that the dstopts get encrypted!) 2573 * 2574 * Since this is for outbound packets only, we know 2575 * that non-terminal destination options only precede 2576 * routing headers. 2577 */ 2578 divpoint -= ipp.ipp_dstoptslen; 2579 } 2580 datalen = ntohs(ip6h->ip6_plen) + sizeof (ip6_t) - divpoint; 2581 2582 if (ipp.ipp_rthdr != NULL) { 2583 nhp = &ipp.ipp_rthdr->ip6r_nxt; 2584 } else if (ipp.ipp_hopopts != NULL) { 2585 nhp = &ipp.ipp_hopopts->ip6h_nxt; 2586 } else { 2587 ASSERT(divpoint == sizeof (ip6_t)); 2588 /* It's probably IP + ESP. */ 2589 nhp = &ip6h->ip6_nxt; 2590 } 2591 } 2592 assoc = io->ipsec_out_esp_sa; 2593 ASSERT(assoc != NULL); 2594 2595 if (assoc->ipsa_auth_alg != SADB_AALG_NONE) 2596 mac_len = assoc->ipsa_mac_len; 2597 2598 if (assoc->ipsa_flags & IPSA_F_NATT) { 2599 /* wedge in fake UDP */ 2600 is_natt = B_TRUE; 2601 esplen += UDPH_SIZE; 2602 } 2603 2604 /* 2605 * Set up ESP header and encryption padding for ENCR PI request. 2606 */ 2607 2608 /* Determine the padding length. Pad to 4-bytes for no-encryption. */ 2609 if (assoc->ipsa_encr_alg != SADB_EALG_NULL) { 2610 iv_len = assoc->ipsa_iv_len; 2611 2612 /* 2613 * Include the two additional bytes (hence the - 2) for the 2614 * padding length and the next header. Take this into account 2615 * when calculating the actual length of the padding. 2616 */ 2617 ASSERT(ISP2(iv_len)); 2618 padlen = ((unsigned)(iv_len - datalen - 2)) & (iv_len - 1); 2619 } else { 2620 iv_len = 0; 2621 padlen = ((unsigned)(sizeof (uint32_t) - datalen - 2)) & 2622 (sizeof (uint32_t) - 1); 2623 } 2624 2625 /* Allocate ESP header and IV. */ 2626 esplen += iv_len; 2627 2628 /* 2629 * Update association byte-count lifetimes. Don't forget to take 2630 * into account the padding length and next-header (hence the + 2). 2631 * 2632 * Use the amount of data fed into the "encryption algorithm". This 2633 * is the IV, the data length, the padding length, and the final two 2634 * bytes (padlen, and next-header). 2635 * 2636 */ 2637 2638 if (!esp_age_bytes(assoc, datalen + padlen + iv_len + 2, B_FALSE)) { 2639 /* 2640 * TODO: Find the outbound IRE for this packet and 2641 * pass it to ip_drop_packet(). 2642 */ 2643 ip_drop_packet(mp, B_FALSE, NULL, NULL, 2644 DROPPER(ipss, ipds_esp_bytes_expire), 2645 &espstack->esp_dropper); 2646 return (IPSEC_STATUS_FAILED); 2647 } 2648 2649 espmp = allocb(esplen, BPRI_HI); 2650 if (espmp == NULL) { 2651 ESP_BUMP_STAT(espstack, out_discards); 2652 esp1dbg(espstack, ("esp_outbound: can't allocate espmp.\n")); 2653 /* 2654 * TODO: Find the outbound IRE for this packet and 2655 * pass it to ip_drop_packet(). 2656 */ 2657 ip_drop_packet(mp, B_FALSE, NULL, NULL, 2658 DROPPER(ipss, ipds_esp_nomem), 2659 &espstack->esp_dropper); 2660 return (IPSEC_STATUS_FAILED); 2661 } 2662 espmp->b_wptr += esplen; 2663 esph = (esph_t *)espmp->b_rptr; 2664 2665 if (is_natt) { 2666 esp3dbg(espstack, ("esp_outbound: NATT")); 2667 2668 udpha = (udpha_t *)espmp->b_rptr; 2669 udpha->uha_src_port = (assoc->ipsa_local_nat_port != 0) ? 2670 assoc->ipsa_local_nat_port : htons(IPPORT_IKE_NATT); 2671 udpha->uha_dst_port = (assoc->ipsa_remote_nat_port != 0) ? 2672 assoc->ipsa_remote_nat_port : htons(IPPORT_IKE_NATT); 2673 /* 2674 * Set the checksum to 0, so that the esp_prepare_udp() call 2675 * can do the right thing. 2676 */ 2677 udpha->uha_checksum = 0; 2678 esph = (esph_t *)(udpha + 1); 2679 } 2680 2681 esph->esph_spi = assoc->ipsa_spi; 2682 2683 esph->esph_replay = htonl(atomic_add_32_nv(&assoc->ipsa_replay, 1)); 2684 if (esph->esph_replay == 0 && assoc->ipsa_replay_wsize != 0) { 2685 /* 2686 * XXX We have replay counter wrapping. 2687 * We probably want to nuke this SA (and its peer). 2688 */ 2689 ipsec_assocfailure(info.mi_idnum, 0, 0, 2690 SL_ERROR | SL_CONSOLE | SL_WARN, 2691 "Outbound ESP SA (0x%x, %s) has wrapped sequence.\n", 2692 esph->esph_spi, assoc->ipsa_dstaddr, af, 2693 espstack->ipsecesp_netstack); 2694 2695 ESP_BUMP_STAT(espstack, out_discards); 2696 sadb_replay_delete(assoc); 2697 /* 2698 * TODO: Find the outbound IRE for this packet and 2699 * pass it to ip_drop_packet(). 2700 */ 2701 ip_drop_packet(mp, B_FALSE, NULL, NULL, 2702 DROPPER(ipss, ipds_esp_replay), 2703 &espstack->esp_dropper); 2704 return (IPSEC_STATUS_FAILED); 2705 } 2706 2707 /* 2708 * Set the IV to a random quantity. We do not require the 2709 * highest quality random bits, but for best security with CBC 2710 * mode ciphers, the value must be unlikely to repeat and also 2711 * must not be known in advance to an adversary capable of 2712 * influencing the plaintext. 2713 */ 2714 (void) random_get_pseudo_bytes((uint8_t *)(esph + 1), iv_len); 2715 2716 /* Fix the IP header. */ 2717 alloclen = padlen + 2 + mac_len; 2718 adj = alloclen + (espmp->b_wptr - espmp->b_rptr); 2719 2720 protocol = *nhp; 2721 2722 if (io->ipsec_out_v4) { 2723 ipha->ipha_length = htons(ntohs(ipha->ipha_length) + adj); 2724 if (is_natt) { 2725 *nhp = IPPROTO_UDP; 2726 udpha->uha_length = htons(ntohs(ipha->ipha_length) - 2727 IPH_HDR_LENGTH(ipha)); 2728 } else { 2729 *nhp = IPPROTO_ESP; 2730 } 2731 ipha->ipha_hdr_checksum = 0; 2732 ipha->ipha_hdr_checksum = (uint16_t)ip_csum_hdr(ipha); 2733 } else { 2734 ip6h->ip6_plen = htons(ntohs(ip6h->ip6_plen) + adj); 2735 *nhp = IPPROTO_ESP; 2736 } 2737 2738 /* I've got the two ESP mblks, now insert them. */ 2739 2740 esp2dbg(espstack, ("data_mp before outbound ESP adjustment:\n")); 2741 esp2dbg(espstack, (dump_msg(data_mp))); 2742 2743 if (!esp_insert_esp(data_mp, espmp, divpoint, espstack)) { 2744 ESP_BUMP_STAT(espstack, out_discards); 2745 /* NOTE: esp_insert_esp() only fails if there's no memory. */ 2746 /* 2747 * TODO: Find the outbound IRE for this packet and 2748 * pass it to ip_drop_packet(). 2749 */ 2750 ip_drop_packet(mp, B_FALSE, NULL, NULL, 2751 DROPPER(ipss, ipds_esp_nomem), 2752 &espstack->esp_dropper); 2753 freeb(espmp); 2754 return (IPSEC_STATUS_FAILED); 2755 } 2756 2757 /* Append padding (and leave room for ICV). */ 2758 for (tailmp = data_mp; tailmp->b_cont != NULL; tailmp = tailmp->b_cont) 2759 ; 2760 if (tailmp->b_wptr + alloclen > tailmp->b_datap->db_lim) { 2761 tailmp->b_cont = allocb(alloclen, BPRI_HI); 2762 if (tailmp->b_cont == NULL) { 2763 ESP_BUMP_STAT(espstack, out_discards); 2764 esp0dbg(("esp_outbound: Can't allocate tailmp.\n")); 2765 /* 2766 * TODO: Find the outbound IRE for this packet and 2767 * pass it to ip_drop_packet(). 2768 */ 2769 ip_drop_packet(mp, B_FALSE, NULL, NULL, 2770 DROPPER(ipss, ipds_esp_nomem), 2771 &espstack->esp_dropper); 2772 return (IPSEC_STATUS_FAILED); 2773 } 2774 tailmp = tailmp->b_cont; 2775 } 2776 2777 /* 2778 * If there's padding, N bytes of padding must be of the form 0x1, 2779 * 0x2, 0x3... 0xN. 2780 */ 2781 for (i = 0; i < padlen; ) { 2782 i++; 2783 *tailmp->b_wptr++ = i; 2784 } 2785 *tailmp->b_wptr++ = i; 2786 *tailmp->b_wptr++ = protocol; 2787 2788 esp2dbg(espstack, ("data_Mp before encryption:\n")); 2789 esp2dbg(espstack, (dump_msg(data_mp))); 2790 2791 /* 2792 * The packet is eligible for hardware acceleration if the 2793 * following conditions are satisfied: 2794 * 2795 * 1. the packet will not be fragmented 2796 * 2. the provider supports the algorithms specified by SA 2797 * 3. there is no pending control message being exchanged 2798 * 4. snoop is not attached 2799 * 5. the destination address is not a multicast address 2800 * 2801 * All five of these conditions are checked by IP prior to 2802 * sending the packet to ESP. 2803 * 2804 * But We, and We Alone, can, nay MUST check if the packet 2805 * is over NATT, and then disqualify it from hardware 2806 * acceleration. 2807 */ 2808 2809 if (io->ipsec_out_is_capab_ill && !(assoc->ipsa_flags & IPSA_F_NATT)) { 2810 return (esp_outbound_accelerated(ipsec_out_mp, mac_len)); 2811 } 2812 ESP_BUMP_STAT(espstack, noaccel); 2813 2814 /* 2815 * Okay. I've set up the pre-encryption ESP. Let's do it! 2816 */ 2817 2818 if (mac_len > 0) { 2819 ASSERT(tailmp->b_wptr + mac_len <= tailmp->b_datap->db_lim); 2820 icv_buf = tailmp->b_wptr; 2821 tailmp->b_wptr += mac_len; 2822 } else { 2823 icv_buf = NULL; 2824 } 2825 2826 return (esp_submit_req_outbound(ipsec_out_mp, assoc, icv_buf, 2827 datalen + padlen + 2)); 2828 } 2829 2830 /* 2831 * IP calls this to validate the ICMP errors that 2832 * we got from the network. 2833 */ 2834 ipsec_status_t 2835 ipsecesp_icmp_error(mblk_t *ipsec_mp) 2836 { 2837 ipsec_in_t *ii = (ipsec_in_t *)ipsec_mp->b_rptr; 2838 boolean_t is_inbound = (ii->ipsec_in_type == IPSEC_IN); 2839 netstack_t *ns; 2840 ipsecesp_stack_t *espstack; 2841 ipsec_stack_t *ipss; 2842 2843 if (is_inbound) { 2844 ns = ii->ipsec_in_ns; 2845 } else { 2846 ipsec_out_t *io = (ipsec_out_t *)ipsec_mp->b_rptr; 2847 2848 ns = io->ipsec_out_ns; 2849 } 2850 espstack = ns->netstack_ipsecesp; 2851 ipss = ns->netstack_ipsec; 2852 2853 /* 2854 * Unless we get an entire packet back, this function is useless. 2855 * Why? 2856 * 2857 * 1.) Partial packets are useless, because the "next header" 2858 * is at the end of the decrypted ESP packet. Without the 2859 * whole packet, this is useless. 2860 * 2861 * 2.) If we every use a stateful cipher, such as a stream or a 2862 * one-time pad, we can't do anything. 2863 * 2864 * Since the chances of us getting an entire packet back are very 2865 * very small, we discard here. 2866 */ 2867 IP_ESP_BUMP_STAT(ipss, in_discards); 2868 ip_drop_packet(ipsec_mp, B_TRUE, NULL, NULL, 2869 DROPPER(ipss, ipds_esp_icmp), 2870 &espstack->esp_dropper); 2871 return (IPSEC_STATUS_FAILED); 2872 } 2873 2874 /* 2875 * ESP module read put routine. 2876 */ 2877 /* ARGSUSED */ 2878 static void 2879 ipsecesp_rput(queue_t *q, mblk_t *mp) 2880 { 2881 ipsecesp_stack_t *espstack = (ipsecesp_stack_t *)q->q_ptr; 2882 2883 ASSERT(mp->b_datap->db_type != M_CTL); /* No more IRE_DB_REQ. */ 2884 2885 switch (mp->b_datap->db_type) { 2886 case M_PROTO: 2887 case M_PCPROTO: 2888 /* TPI message of some sort. */ 2889 switch (*((t_scalar_t *)mp->b_rptr)) { 2890 case T_BIND_ACK: 2891 esp3dbg(espstack, 2892 ("Thank you IP from ESP for T_BIND_ACK\n")); 2893 break; 2894 case T_ERROR_ACK: 2895 cmn_err(CE_WARN, 2896 "ipsecesp: ESP received T_ERROR_ACK from IP."); 2897 /* 2898 * Make esp_sadb.s_ip_q NULL, and in the 2899 * future, perhaps try again. 2900 */ 2901 espstack->esp_sadb.s_ip_q = NULL; 2902 break; 2903 case T_OK_ACK: 2904 /* Probably from a (rarely sent) T_UNBIND_REQ. */ 2905 break; 2906 default: 2907 esp0dbg(("Unknown M_{,PC}PROTO message.\n")); 2908 } 2909 freemsg(mp); 2910 break; 2911 default: 2912 /* For now, passthru message. */ 2913 esp2dbg(espstack, ("ESP got unknown mblk type %d.\n", 2914 mp->b_datap->db_type)); 2915 putnext(q, mp); 2916 } 2917 } 2918 2919 /* 2920 * Construct an SADB_REGISTER message with the current algorithms. 2921 */ 2922 static boolean_t 2923 esp_register_out(uint32_t sequence, uint32_t pid, uint_t serial, 2924 ipsecesp_stack_t *espstack) 2925 { 2926 mblk_t *pfkey_msg_mp, *keysock_out_mp; 2927 sadb_msg_t *samsg; 2928 sadb_supported_t *sasupp_auth = NULL; 2929 sadb_supported_t *sasupp_encr = NULL; 2930 sadb_alg_t *saalg; 2931 uint_t allocsize = sizeof (*samsg); 2932 uint_t i, numalgs_snap; 2933 int current_aalgs; 2934 ipsec_alginfo_t **authalgs; 2935 uint_t num_aalgs; 2936 int current_ealgs; 2937 ipsec_alginfo_t **encralgs; 2938 uint_t num_ealgs; 2939 ipsec_stack_t *ipss = espstack->ipsecesp_netstack->netstack_ipsec; 2940 2941 /* Allocate the KEYSOCK_OUT. */ 2942 keysock_out_mp = sadb_keysock_out(serial); 2943 if (keysock_out_mp == NULL) { 2944 esp0dbg(("esp_register_out: couldn't allocate mblk.\n")); 2945 return (B_FALSE); 2946 } 2947 2948 /* 2949 * Allocate the PF_KEY message that follows KEYSOCK_OUT. 2950 */ 2951 2952 mutex_enter(&ipss->ipsec_alg_lock); 2953 2954 /* 2955 * Fill SADB_REGISTER message's algorithm descriptors. Hold 2956 * down the lock while filling it. 2957 * 2958 * Return only valid algorithms, so the number of algorithms 2959 * to send up may be less than the number of algorithm entries 2960 * in the table. 2961 */ 2962 authalgs = ipss->ipsec_alglists[IPSEC_ALG_AUTH]; 2963 for (num_aalgs = 0, i = 0; i < IPSEC_MAX_ALGS; i++) 2964 if (authalgs[i] != NULL && ALG_VALID(authalgs[i])) 2965 num_aalgs++; 2966 2967 if (num_aalgs != 0) { 2968 allocsize += (num_aalgs * sizeof (*saalg)); 2969 allocsize += sizeof (*sasupp_auth); 2970 } 2971 encralgs = ipss->ipsec_alglists[IPSEC_ALG_ENCR]; 2972 for (num_ealgs = 0, i = 0; i < IPSEC_MAX_ALGS; i++) 2973 if (encralgs[i] != NULL && ALG_VALID(encralgs[i])) 2974 num_ealgs++; 2975 2976 if (num_ealgs != 0) { 2977 allocsize += (num_ealgs * sizeof (*saalg)); 2978 allocsize += sizeof (*sasupp_encr); 2979 } 2980 keysock_out_mp->b_cont = allocb(allocsize, BPRI_HI); 2981 if (keysock_out_mp->b_cont == NULL) { 2982 mutex_exit(&ipss->ipsec_alg_lock); 2983 freemsg(keysock_out_mp); 2984 return (B_FALSE); 2985 } 2986 2987 pfkey_msg_mp = keysock_out_mp->b_cont; 2988 pfkey_msg_mp->b_wptr += allocsize; 2989 if (num_aalgs != 0) { 2990 sasupp_auth = (sadb_supported_t *) 2991 (pfkey_msg_mp->b_rptr + sizeof (*samsg)); 2992 saalg = (sadb_alg_t *)(sasupp_auth + 1); 2993 2994 ASSERT(((ulong_t)saalg & 0x7) == 0); 2995 2996 numalgs_snap = 0; 2997 for (i = 0; 2998 ((i < IPSEC_MAX_ALGS) && (numalgs_snap < num_aalgs)); 2999 i++) { 3000 if (authalgs[i] == NULL || !ALG_VALID(authalgs[i])) 3001 continue; 3002 3003 saalg->sadb_alg_id = authalgs[i]->alg_id; 3004 saalg->sadb_alg_ivlen = 0; 3005 saalg->sadb_alg_minbits = authalgs[i]->alg_ef_minbits; 3006 saalg->sadb_alg_maxbits = authalgs[i]->alg_ef_maxbits; 3007 saalg->sadb_x_alg_defincr = authalgs[i]->alg_ef_default; 3008 saalg->sadb_x_alg_increment = 3009 authalgs[i]->alg_increment; 3010 numalgs_snap++; 3011 saalg++; 3012 } 3013 ASSERT(numalgs_snap == num_aalgs); 3014 #ifdef DEBUG 3015 /* 3016 * Reality check to make sure I snagged all of the 3017 * algorithms. 3018 */ 3019 for (; i < IPSEC_MAX_ALGS; i++) { 3020 if (authalgs[i] != NULL && ALG_VALID(authalgs[i])) { 3021 cmn_err(CE_PANIC, "esp_register_out()! " 3022 "Missed aalg #%d.\n", i); 3023 } 3024 } 3025 #endif /* DEBUG */ 3026 } else { 3027 saalg = (sadb_alg_t *)(pfkey_msg_mp->b_rptr + sizeof (*samsg)); 3028 } 3029 3030 if (num_ealgs != 0) { 3031 sasupp_encr = (sadb_supported_t *)saalg; 3032 saalg = (sadb_alg_t *)(sasupp_encr + 1); 3033 3034 numalgs_snap = 0; 3035 for (i = 0; 3036 ((i < IPSEC_MAX_ALGS) && (numalgs_snap < num_ealgs)); i++) { 3037 if (encralgs[i] == NULL || !ALG_VALID(encralgs[i])) 3038 continue; 3039 saalg->sadb_alg_id = encralgs[i]->alg_id; 3040 saalg->sadb_alg_ivlen = encralgs[i]->alg_datalen; 3041 saalg->sadb_alg_minbits = encralgs[i]->alg_ef_minbits; 3042 saalg->sadb_alg_maxbits = encralgs[i]->alg_ef_maxbits; 3043 saalg->sadb_x_alg_defincr = encralgs[i]->alg_ef_default; 3044 saalg->sadb_x_alg_increment = 3045 encralgs[i]->alg_increment; 3046 numalgs_snap++; 3047 saalg++; 3048 } 3049 ASSERT(numalgs_snap == num_ealgs); 3050 #ifdef DEBUG 3051 /* 3052 * Reality check to make sure I snagged all of the 3053 * algorithms. 3054 */ 3055 for (; i < IPSEC_MAX_ALGS; i++) { 3056 if (encralgs[i] != NULL && ALG_VALID(encralgs[i])) { 3057 cmn_err(CE_PANIC, "esp_register_out()! " 3058 "Missed ealg #%d.\n", i); 3059 } 3060 } 3061 #endif /* DEBUG */ 3062 } 3063 3064 current_aalgs = num_aalgs; 3065 current_ealgs = num_ealgs; 3066 3067 mutex_exit(&ipss->ipsec_alg_lock); 3068 3069 /* Now fill the rest of the SADB_REGISTER message. */ 3070 3071 samsg = (sadb_msg_t *)pfkey_msg_mp->b_rptr; 3072 samsg->sadb_msg_version = PF_KEY_V2; 3073 samsg->sadb_msg_type = SADB_REGISTER; 3074 samsg->sadb_msg_errno = 0; 3075 samsg->sadb_msg_satype = SADB_SATYPE_ESP; 3076 samsg->sadb_msg_len = SADB_8TO64(allocsize); 3077 samsg->sadb_msg_reserved = 0; 3078 /* 3079 * Assume caller has sufficient sequence/pid number info. If it's one 3080 * from me over a new alg., I could give two hoots about sequence. 3081 */ 3082 samsg->sadb_msg_seq = sequence; 3083 samsg->sadb_msg_pid = pid; 3084 3085 if (sasupp_auth != NULL) { 3086 sasupp_auth->sadb_supported_len = SADB_8TO64( 3087 sizeof (*sasupp_auth) + sizeof (*saalg) * current_aalgs); 3088 sasupp_auth->sadb_supported_exttype = SADB_EXT_SUPPORTED_AUTH; 3089 sasupp_auth->sadb_supported_reserved = 0; 3090 } 3091 3092 if (sasupp_encr != NULL) { 3093 sasupp_encr->sadb_supported_len = SADB_8TO64( 3094 sizeof (*sasupp_encr) + sizeof (*saalg) * current_ealgs); 3095 sasupp_encr->sadb_supported_exttype = 3096 SADB_EXT_SUPPORTED_ENCRYPT; 3097 sasupp_encr->sadb_supported_reserved = 0; 3098 } 3099 3100 if (espstack->esp_pfkey_q != NULL) 3101 putnext(espstack->esp_pfkey_q, keysock_out_mp); 3102 else { 3103 freemsg(keysock_out_mp); 3104 return (B_FALSE); 3105 } 3106 3107 return (B_TRUE); 3108 } 3109 3110 /* 3111 * Invoked when the algorithm table changes. Causes SADB_REGISTER 3112 * messages continaining the current list of algorithms to be 3113 * sent up to the ESP listeners. 3114 */ 3115 void 3116 ipsecesp_algs_changed(netstack_t *ns) 3117 { 3118 ipsecesp_stack_t *espstack = ns->netstack_ipsecesp; 3119 3120 /* 3121 * Time to send a PF_KEY SADB_REGISTER message to ESP listeners 3122 * everywhere. (The function itself checks for NULL esp_pfkey_q.) 3123 */ 3124 (void) esp_register_out(0, 0, 0, espstack); 3125 } 3126 3127 /* 3128 * taskq_dispatch handler. 3129 */ 3130 static void 3131 inbound_task(void *arg) 3132 { 3133 esph_t *esph; 3134 mblk_t *mp = (mblk_t *)arg; 3135 ipsec_in_t *ii = (ipsec_in_t *)mp->b_rptr; 3136 netstack_t *ns = ii->ipsec_in_ns; 3137 ipsecesp_stack_t *espstack = ns->netstack_ipsecesp; 3138 int ipsec_rc; 3139 3140 esp2dbg(espstack, ("in ESP inbound_task")); 3141 ASSERT(espstack != NULL); 3142 3143 esph = ipsec_inbound_esp_sa(mp, ns); 3144 if (esph == NULL) 3145 return; 3146 ASSERT(ii->ipsec_in_esp_sa != NULL); 3147 ipsec_rc = ii->ipsec_in_esp_sa->ipsa_input_func(mp, esph); 3148 if (ipsec_rc != IPSEC_STATUS_SUCCESS) 3149 return; 3150 ip_fanout_proto_again(mp, NULL, NULL, NULL); 3151 } 3152 3153 /* 3154 * Now that weak-key passed, actually ADD the security association, and 3155 * send back a reply ADD message. 3156 */ 3157 static int 3158 esp_add_sa_finish(mblk_t *mp, sadb_msg_t *samsg, keysock_in_t *ksi, 3159 int *diagnostic, ipsecesp_stack_t *espstack) 3160 { 3161 isaf_t *primary = NULL, *secondary, *inbound, *outbound; 3162 sadb_sa_t *assoc = (sadb_sa_t *)ksi->ks_in_extv[SADB_EXT_SA]; 3163 sadb_address_t *dstext = 3164 (sadb_address_t *)ksi->ks_in_extv[SADB_EXT_ADDRESS_DST]; 3165 struct sockaddr_in *dst; 3166 struct sockaddr_in6 *dst6; 3167 boolean_t is_ipv4, clone = B_FALSE, is_inbound = B_FALSE; 3168 uint32_t *dstaddr; 3169 ipsa_t *larval = NULL; 3170 ipsacq_t *acqrec; 3171 iacqf_t *acq_bucket; 3172 mblk_t *acq_msgs = NULL; 3173 int rc; 3174 sadb_t *sp; 3175 int outhash; 3176 mblk_t *lpkt; 3177 ipsec_stack_t *ipss = espstack->ipsecesp_netstack->netstack_ipsec; 3178 3179 /* 3180 * Locate the appropriate table(s). 3181 */ 3182 3183 dst = (struct sockaddr_in *)(dstext + 1); 3184 dst6 = (struct sockaddr_in6 *)dst; 3185 is_ipv4 = (dst->sin_family == AF_INET); 3186 if (is_ipv4) { 3187 sp = &espstack->esp_sadb.s_v4; 3188 dstaddr = (uint32_t *)(&dst->sin_addr); 3189 outhash = OUTBOUND_HASH_V4(sp, *(ipaddr_t *)dstaddr); 3190 } else { 3191 sp = &espstack->esp_sadb.s_v6; 3192 dstaddr = (uint32_t *)(&dst6->sin6_addr); 3193 outhash = OUTBOUND_HASH_V6(sp, *(in6_addr_t *)dstaddr); 3194 } 3195 3196 inbound = INBOUND_BUCKET(sp, assoc->sadb_sa_spi); 3197 outbound = &sp->sdb_of[outhash]; 3198 3199 /* 3200 * Use the direction flags provided by the KMD to determine 3201 * if the inbound or outbound table should be the primary 3202 * for this SA. If these flags were absent then make this 3203 * decision based on the addresses. 3204 */ 3205 if (assoc->sadb_sa_flags & IPSA_F_INBOUND) { 3206 primary = inbound; 3207 secondary = outbound; 3208 is_inbound = B_TRUE; 3209 if (assoc->sadb_sa_flags & IPSA_F_OUTBOUND) 3210 clone = B_TRUE; 3211 } else { 3212 if (assoc->sadb_sa_flags & IPSA_F_OUTBOUND) { 3213 primary = outbound; 3214 secondary = inbound; 3215 } 3216 } 3217 3218 if (primary == NULL) { 3219 /* 3220 * The KMD did not set a direction flag, determine which 3221 * table to insert the SA into based on addresses. 3222 */ 3223 switch (ksi->ks_in_dsttype) { 3224 case KS_IN_ADDR_MBCAST: 3225 clone = B_TRUE; /* All mcast SAs can be bidirectional */ 3226 assoc->sadb_sa_flags |= IPSA_F_OUTBOUND; 3227 /* FALLTHRU */ 3228 /* 3229 * If the source address is either one of mine, or unspecified 3230 * (which is best summed up by saying "not 'not mine'"), 3231 * then the association is potentially bi-directional, 3232 * in that it can be used for inbound traffic and outbound 3233 * traffic. The best example of such an SA is a multicast 3234 * SA (which allows me to receive the outbound traffic). 3235 */ 3236 case KS_IN_ADDR_ME: 3237 assoc->sadb_sa_flags |= IPSA_F_INBOUND; 3238 primary = inbound; 3239 secondary = outbound; 3240 if (ksi->ks_in_srctype != KS_IN_ADDR_NOTME) 3241 clone = B_TRUE; 3242 is_inbound = B_TRUE; 3243 break; 3244 /* 3245 * If the source address literally not mine (either 3246 * unspecified or not mine), then this SA may have an 3247 * address that WILL be mine after some configuration. 3248 * We pay the price for this by making it a bi-directional 3249 * SA. 3250 */ 3251 case KS_IN_ADDR_NOTME: 3252 assoc->sadb_sa_flags |= IPSA_F_OUTBOUND; 3253 primary = outbound; 3254 secondary = inbound; 3255 if (ksi->ks_in_srctype != KS_IN_ADDR_ME) { 3256 assoc->sadb_sa_flags |= IPSA_F_INBOUND; 3257 clone = B_TRUE; 3258 } 3259 break; 3260 default: 3261 *diagnostic = SADB_X_DIAGNOSTIC_BAD_DST; 3262 return (EINVAL); 3263 } 3264 } 3265 3266 /* 3267 * Find a ACQUIRE list entry if possible. If we've added an SA that 3268 * suits the needs of an ACQUIRE list entry, we can eliminate the 3269 * ACQUIRE list entry and transmit the enqueued packets. Use the 3270 * high-bit of the sequence number to queue it. Key off destination 3271 * addr, and change acqrec's state. 3272 */ 3273 3274 if (samsg->sadb_msg_seq & IACQF_LOWEST_SEQ) { 3275 acq_bucket = &sp->sdb_acq[outhash]; 3276 mutex_enter(&acq_bucket->iacqf_lock); 3277 for (acqrec = acq_bucket->iacqf_ipsacq; acqrec != NULL; 3278 acqrec = acqrec->ipsacq_next) { 3279 mutex_enter(&acqrec->ipsacq_lock); 3280 /* 3281 * Q: I only check sequence. Should I check dst? 3282 * A: Yes, check dest because those are the packets 3283 * that are queued up. 3284 */ 3285 if (acqrec->ipsacq_seq == samsg->sadb_msg_seq && 3286 IPSA_ARE_ADDR_EQUAL(dstaddr, 3287 acqrec->ipsacq_dstaddr, acqrec->ipsacq_addrfam)) 3288 break; 3289 mutex_exit(&acqrec->ipsacq_lock); 3290 } 3291 if (acqrec != NULL) { 3292 /* 3293 * AHA! I found an ACQUIRE record for this SA. 3294 * Grab the msg list, and free the acquire record. 3295 * I already am holding the lock for this record, 3296 * so all I have to do is free it. 3297 */ 3298 acq_msgs = acqrec->ipsacq_mp; 3299 acqrec->ipsacq_mp = NULL; 3300 mutex_exit(&acqrec->ipsacq_lock); 3301 sadb_destroy_acquire(acqrec, 3302 espstack->ipsecesp_netstack); 3303 } 3304 mutex_exit(&acq_bucket->iacqf_lock); 3305 } 3306 3307 /* 3308 * Find PF_KEY message, and see if I'm an update. If so, find entry 3309 * in larval list (if there). 3310 */ 3311 3312 if (samsg->sadb_msg_type == SADB_UPDATE) { 3313 mutex_enter(&inbound->isaf_lock); 3314 larval = ipsec_getassocbyspi(inbound, assoc->sadb_sa_spi, 3315 ALL_ZEROES_PTR, dstaddr, dst->sin_family); 3316 mutex_exit(&inbound->isaf_lock); 3317 3318 if ((larval == NULL) || 3319 (larval->ipsa_state != IPSA_STATE_LARVAL)) { 3320 *diagnostic = SADB_X_DIAGNOSTIC_SA_NOTFOUND; 3321 if (larval != NULL) { 3322 IPSA_REFRELE(larval); 3323 } 3324 esp0dbg(("Larval update, but larval disappeared.\n")); 3325 return (ESRCH); 3326 } /* Else sadb_common_add unlinks it for me! */ 3327 } 3328 3329 lpkt = NULL; 3330 if (larval != NULL) 3331 lpkt = sadb_clear_lpkt(larval); 3332 3333 rc = sadb_common_add(espstack->esp_sadb.s_ip_q, espstack->esp_pfkey_q, 3334 mp, samsg, ksi, primary, secondary, larval, clone, is_inbound, 3335 diagnostic, espstack->ipsecesp_netstack, &espstack->esp_sadb); 3336 3337 if (rc == 0 && lpkt != NULL) { 3338 rc = !taskq_dispatch(esp_taskq, inbound_task, 3339 (void *) lpkt, TQ_NOSLEEP); 3340 } 3341 3342 if (rc != 0) { 3343 ip_drop_packet(lpkt, B_TRUE, NULL, NULL, 3344 DROPPER(ipss, ipds_sadb_inlarval_timeout), 3345 &espstack->esp_dropper); 3346 } 3347 3348 /* 3349 * How much more stack will I create with all of these 3350 * esp_outbound() calls? 3351 */ 3352 3353 while (acq_msgs != NULL) { 3354 mblk_t *mp = acq_msgs; 3355 3356 acq_msgs = acq_msgs->b_next; 3357 mp->b_next = NULL; 3358 if (rc == 0) { 3359 if (ipsec_outbound_sa(mp, IPPROTO_ESP)) { 3360 ((ipsec_out_t *)(mp->b_rptr))-> 3361 ipsec_out_esp_done = B_TRUE; 3362 if (esp_outbound(mp) == IPSEC_STATUS_SUCCESS) { 3363 ipha_t *ipha; 3364 3365 /* do AH processing if needed */ 3366 if (!esp_do_outbound_ah(mp)) 3367 continue; 3368 3369 ipha = (ipha_t *)mp->b_cont->b_rptr; 3370 3371 /* finish IPsec processing */ 3372 if (is_ipv4) { 3373 ip_wput_ipsec_out(NULL, mp, 3374 ipha, NULL, NULL); 3375 } else { 3376 ip6_t *ip6h = (ip6_t *)ipha; 3377 ip_wput_ipsec_out_v6(NULL, 3378 mp, ip6h, NULL, NULL); 3379 } 3380 } 3381 continue; 3382 } 3383 } 3384 ESP_BUMP_STAT(espstack, out_discards); 3385 ip_drop_packet(mp, B_FALSE, NULL, NULL, 3386 DROPPER(ipss, ipds_sadb_acquire_timeout), 3387 &espstack->esp_dropper); 3388 } 3389 3390 return (rc); 3391 } 3392 3393 /* 3394 * Add new ESP security association. This may become a generic AH/ESP 3395 * routine eventually. 3396 */ 3397 static int 3398 esp_add_sa(mblk_t *mp, keysock_in_t *ksi, int *diagnostic, netstack_t *ns) 3399 { 3400 sadb_sa_t *assoc = (sadb_sa_t *)ksi->ks_in_extv[SADB_EXT_SA]; 3401 sadb_address_t *srcext = 3402 (sadb_address_t *)ksi->ks_in_extv[SADB_EXT_ADDRESS_SRC]; 3403 sadb_address_t *dstext = 3404 (sadb_address_t *)ksi->ks_in_extv[SADB_EXT_ADDRESS_DST]; 3405 sadb_address_t *isrcext = 3406 (sadb_address_t *)ksi->ks_in_extv[SADB_X_EXT_ADDRESS_INNER_SRC]; 3407 sadb_address_t *idstext = 3408 (sadb_address_t *)ksi->ks_in_extv[SADB_X_EXT_ADDRESS_INNER_DST]; 3409 sadb_address_t *nttext_loc = 3410 (sadb_address_t *)ksi->ks_in_extv[SADB_X_EXT_ADDRESS_NATT_LOC]; 3411 sadb_address_t *nttext_rem = 3412 (sadb_address_t *)ksi->ks_in_extv[SADB_X_EXT_ADDRESS_NATT_REM]; 3413 sadb_key_t *akey = (sadb_key_t *)ksi->ks_in_extv[SADB_EXT_KEY_AUTH]; 3414 sadb_key_t *ekey = (sadb_key_t *)ksi->ks_in_extv[SADB_EXT_KEY_ENCRYPT]; 3415 struct sockaddr_in *src, *dst; 3416 struct sockaddr_in *natt_loc, *natt_rem; 3417 struct sockaddr_in6 *natt_loc6, *natt_rem6; 3418 sadb_lifetime_t *soft = 3419 (sadb_lifetime_t *)ksi->ks_in_extv[SADB_EXT_LIFETIME_SOFT]; 3420 sadb_lifetime_t *hard = 3421 (sadb_lifetime_t *)ksi->ks_in_extv[SADB_EXT_LIFETIME_HARD]; 3422 ipsecesp_stack_t *espstack = ns->netstack_ipsecesp; 3423 ipsec_stack_t *ipss = ns->netstack_ipsec; 3424 3425 /* I need certain extensions present for an ADD message. */ 3426 if (srcext == NULL) { 3427 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_SRC; 3428 return (EINVAL); 3429 } 3430 if (dstext == NULL) { 3431 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_DST; 3432 return (EINVAL); 3433 } 3434 if (isrcext == NULL && idstext != NULL) { 3435 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_INNER_SRC; 3436 return (EINVAL); 3437 } 3438 if (isrcext != NULL && idstext == NULL) { 3439 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_INNER_DST; 3440 return (EINVAL); 3441 } 3442 if (assoc == NULL) { 3443 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_SA; 3444 return (EINVAL); 3445 } 3446 if (ekey == NULL && assoc->sadb_sa_encrypt != SADB_EALG_NULL) { 3447 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_EKEY; 3448 return (EINVAL); 3449 } 3450 3451 src = (struct sockaddr_in *)(srcext + 1); 3452 dst = (struct sockaddr_in *)(dstext + 1); 3453 natt_loc = (struct sockaddr_in *)(nttext_loc + 1); 3454 natt_loc6 = (struct sockaddr_in6 *)(nttext_loc + 1); 3455 natt_rem = (struct sockaddr_in *)(nttext_rem + 1); 3456 natt_rem6 = (struct sockaddr_in6 *)(nttext_rem + 1); 3457 3458 /* Sundry ADD-specific reality checks. */ 3459 /* XXX STATS : Logging/stats here? */ 3460 if (assoc->sadb_sa_state != SADB_SASTATE_MATURE) { 3461 *diagnostic = SADB_X_DIAGNOSTIC_BAD_SASTATE; 3462 return (EINVAL); 3463 } 3464 if (assoc->sadb_sa_encrypt == SADB_EALG_NONE) { 3465 *diagnostic = SADB_X_DIAGNOSTIC_BAD_EALG; 3466 return (EINVAL); 3467 } 3468 3469 if (assoc->sadb_sa_encrypt == SADB_EALG_NULL && 3470 assoc->sadb_sa_auth == SADB_AALG_NONE) { 3471 *diagnostic = SADB_X_DIAGNOSTIC_BAD_AALG; 3472 return (EINVAL); 3473 } 3474 3475 if (assoc->sadb_sa_flags & ~espstack->esp_sadb.s_addflags) { 3476 *diagnostic = SADB_X_DIAGNOSTIC_BAD_SAFLAGS; 3477 return (EINVAL); 3478 } 3479 3480 if ((*diagnostic = sadb_hardsoftchk(hard, soft)) != 0) { 3481 return (EINVAL); 3482 } 3483 ASSERT(src->sin_family == dst->sin_family); 3484 3485 if (assoc->sadb_sa_flags & SADB_X_SAFLAGS_NATT_LOC) { 3486 if (nttext_loc == NULL) { 3487 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_NATT_LOC; 3488 return (EINVAL); 3489 } 3490 3491 if (natt_loc->sin_family == AF_INET6 && 3492 !IN6_IS_ADDR_V4MAPPED(&natt_loc6->sin6_addr)) { 3493 *diagnostic = SADB_X_DIAGNOSTIC_MALFORMED_NATT_LOC; 3494 return (EINVAL); 3495 } 3496 } 3497 3498 if (assoc->sadb_sa_flags & SADB_X_SAFLAGS_NATT_REM) { 3499 if (nttext_rem == NULL) { 3500 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_NATT_REM; 3501 return (EINVAL); 3502 } 3503 if (natt_rem->sin_family == AF_INET6 && 3504 !IN6_IS_ADDR_V4MAPPED(&natt_rem6->sin6_addr)) { 3505 *diagnostic = SADB_X_DIAGNOSTIC_MALFORMED_NATT_REM; 3506 return (EINVAL); 3507 } 3508 } 3509 3510 3511 /* Stuff I don't support, for now. XXX Diagnostic? */ 3512 if (ksi->ks_in_extv[SADB_EXT_LIFETIME_CURRENT] != NULL || 3513 ksi->ks_in_extv[SADB_EXT_SENSITIVITY] != NULL) 3514 return (EOPNOTSUPP); 3515 3516 /* 3517 * XXX Policy : I'm not checking identities or sensitivity 3518 * labels at this time, but if I did, I'd do them here, before I sent 3519 * the weak key check up to the algorithm. 3520 */ 3521 3522 mutex_enter(&ipss->ipsec_alg_lock); 3523 3524 /* 3525 * First locate the authentication algorithm. 3526 */ 3527 if (akey != NULL) { 3528 ipsec_alginfo_t *aalg; 3529 3530 aalg = ipss->ipsec_alglists[IPSEC_ALG_AUTH] 3531 [assoc->sadb_sa_auth]; 3532 if (aalg == NULL || !ALG_VALID(aalg)) { 3533 mutex_exit(&ipss->ipsec_alg_lock); 3534 esp1dbg(espstack, ("Couldn't find auth alg #%d.\n", 3535 assoc->sadb_sa_auth)); 3536 *diagnostic = SADB_X_DIAGNOSTIC_BAD_AALG; 3537 return (EINVAL); 3538 } 3539 3540 /* 3541 * Sanity check key sizes. 3542 * Note: It's not possible to use SADB_AALG_NONE because 3543 * this auth_alg is not defined with ALG_FLAG_VALID. If this 3544 * ever changes, the same check for SADB_AALG_NONE and 3545 * a auth_key != NULL should be made here ( see below). 3546 */ 3547 if (!ipsec_valid_key_size(akey->sadb_key_bits, aalg)) { 3548 mutex_exit(&ipss->ipsec_alg_lock); 3549 *diagnostic = SADB_X_DIAGNOSTIC_BAD_AKEYBITS; 3550 return (EINVAL); 3551 } 3552 ASSERT(aalg->alg_mech_type != CRYPTO_MECHANISM_INVALID); 3553 3554 /* check key and fix parity if needed */ 3555 if (ipsec_check_key(aalg->alg_mech_type, akey, B_TRUE, 3556 diagnostic) != 0) { 3557 mutex_exit(&ipss->ipsec_alg_lock); 3558 return (EINVAL); 3559 } 3560 } 3561 3562 /* 3563 * Then locate the encryption algorithm. 3564 */ 3565 if (ekey != NULL) { 3566 ipsec_alginfo_t *ealg; 3567 3568 ealg = ipss->ipsec_alglists[IPSEC_ALG_ENCR] 3569 [assoc->sadb_sa_encrypt]; 3570 if (ealg == NULL || !ALG_VALID(ealg)) { 3571 mutex_exit(&ipss->ipsec_alg_lock); 3572 esp1dbg(espstack, ("Couldn't find encr alg #%d.\n", 3573 assoc->sadb_sa_encrypt)); 3574 *diagnostic = SADB_X_DIAGNOSTIC_BAD_EALG; 3575 return (EINVAL); 3576 } 3577 3578 /* 3579 * Sanity check key sizes. If the encryption algorithm is 3580 * SADB_EALG_NULL but the encryption key is NOT 3581 * NULL then complain. 3582 */ 3583 if ((assoc->sadb_sa_encrypt == SADB_EALG_NULL) || 3584 (!ipsec_valid_key_size(ekey->sadb_key_bits, ealg))) { 3585 mutex_exit(&ipss->ipsec_alg_lock); 3586 *diagnostic = SADB_X_DIAGNOSTIC_BAD_EKEYBITS; 3587 return (EINVAL); 3588 } 3589 ASSERT(ealg->alg_mech_type != CRYPTO_MECHANISM_INVALID); 3590 3591 /* check key */ 3592 if (ipsec_check_key(ealg->alg_mech_type, ekey, B_FALSE, 3593 diagnostic) != 0) { 3594 mutex_exit(&ipss->ipsec_alg_lock); 3595 return (EINVAL); 3596 } 3597 } 3598 mutex_exit(&ipss->ipsec_alg_lock); 3599 3600 return (esp_add_sa_finish(mp, (sadb_msg_t *)mp->b_cont->b_rptr, ksi, 3601 diagnostic, espstack)); 3602 } 3603 3604 /* 3605 * Update a security association. Updates come in two varieties. The first 3606 * is an update of lifetimes on a non-larval SA. The second is an update of 3607 * a larval SA, which ends up looking a lot more like an add. 3608 */ 3609 static int 3610 esp_update_sa(mblk_t *mp, keysock_in_t *ksi, int *diagnostic, 3611 ipsecesp_stack_t *espstack, uint8_t sadb_msg_type) 3612 { 3613 sadb_address_t *dstext = 3614 (sadb_address_t *)ksi->ks_in_extv[SADB_EXT_ADDRESS_DST]; 3615 3616 if (dstext == NULL) { 3617 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_DST; 3618 return (EINVAL); 3619 } 3620 3621 return (sadb_update_sa(mp, ksi, &espstack->esp_sadb, 3622 diagnostic, espstack->esp_pfkey_q, 3623 esp_add_sa, espstack->ipsecesp_netstack, sadb_msg_type)); 3624 } 3625 3626 /* 3627 * Delete a security association. This is REALLY likely to be code common to 3628 * both AH and ESP. Find the association, then unlink it. 3629 */ 3630 static int 3631 esp_del_sa(mblk_t *mp, keysock_in_t *ksi, int *diagnostic, 3632 ipsecesp_stack_t *espstack, uint8_t sadb_msg_type) 3633 { 3634 sadb_sa_t *assoc = (sadb_sa_t *)ksi->ks_in_extv[SADB_EXT_SA]; 3635 sadb_address_t *dstext = 3636 (sadb_address_t *)ksi->ks_in_extv[SADB_EXT_ADDRESS_DST]; 3637 sadb_address_t *srcext = 3638 (sadb_address_t *)ksi->ks_in_extv[SADB_EXT_ADDRESS_SRC]; 3639 struct sockaddr_in *sin; 3640 3641 if (assoc == NULL) { 3642 if (dstext != NULL) { 3643 sin = (struct sockaddr_in *)(dstext + 1); 3644 } else if (srcext != NULL) { 3645 sin = (struct sockaddr_in *)(srcext + 1); 3646 } else { 3647 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_SA; 3648 return (EINVAL); 3649 } 3650 return (sadb_purge_sa(mp, ksi, 3651 (sin->sin_family == AF_INET6) ? &espstack->esp_sadb.s_v6 : 3652 &espstack->esp_sadb.s_v4, espstack->esp_pfkey_q, 3653 espstack->esp_sadb.s_ip_q)); 3654 } 3655 3656 return (sadb_delget_sa(mp, ksi, &espstack->esp_sadb, diagnostic, 3657 espstack->esp_pfkey_q, sadb_msg_type)); 3658 } 3659 3660 /* 3661 * Convert the entire contents of all of ESP's SA tables into PF_KEY SADB_DUMP 3662 * messages. 3663 */ 3664 static void 3665 esp_dump(mblk_t *mp, keysock_in_t *ksi, ipsecesp_stack_t *espstack) 3666 { 3667 int error; 3668 sadb_msg_t *samsg; 3669 3670 /* 3671 * Dump each fanout, bailing if error is non-zero. 3672 */ 3673 3674 error = sadb_dump(espstack->esp_pfkey_q, mp, ksi->ks_in_serial, 3675 &espstack->esp_sadb.s_v4); 3676 if (error != 0) 3677 goto bail; 3678 3679 error = sadb_dump(espstack->esp_pfkey_q, mp, ksi->ks_in_serial, 3680 &espstack->esp_sadb.s_v6); 3681 bail: 3682 ASSERT(mp->b_cont != NULL); 3683 samsg = (sadb_msg_t *)mp->b_cont->b_rptr; 3684 samsg->sadb_msg_errno = (uint8_t)error; 3685 sadb_pfkey_echo(espstack->esp_pfkey_q, mp, 3686 (sadb_msg_t *)mp->b_cont->b_rptr, ksi, NULL); 3687 } 3688 3689 /* 3690 * First-cut reality check for an inbound PF_KEY message. 3691 */ 3692 static boolean_t 3693 esp_pfkey_reality_failures(mblk_t *mp, keysock_in_t *ksi, 3694 ipsecesp_stack_t *espstack) 3695 { 3696 int diagnostic; 3697 3698 if (ksi->ks_in_extv[SADB_EXT_PROPOSAL] != NULL) { 3699 diagnostic = SADB_X_DIAGNOSTIC_PROP_PRESENT; 3700 goto badmsg; 3701 } 3702 if (ksi->ks_in_extv[SADB_EXT_SUPPORTED_AUTH] != NULL || 3703 ksi->ks_in_extv[SADB_EXT_SUPPORTED_ENCRYPT] != NULL) { 3704 diagnostic = SADB_X_DIAGNOSTIC_SUPP_PRESENT; 3705 goto badmsg; 3706 } 3707 return (B_FALSE); /* False ==> no failures */ 3708 3709 badmsg: 3710 sadb_pfkey_error(espstack->esp_pfkey_q, mp, EINVAL, diagnostic, 3711 ksi->ks_in_serial); 3712 return (B_TRUE); /* True ==> failures */ 3713 } 3714 3715 /* 3716 * ESP parsing of PF_KEY messages. Keysock did most of the really silly 3717 * error cases. What I receive is a fully-formed, syntactically legal 3718 * PF_KEY message. I then need to check semantics... 3719 * 3720 * This code may become common to AH and ESP. Stay tuned. 3721 * 3722 * I also make the assumption that db_ref's are cool. If this assumption 3723 * is wrong, this means that someone other than keysock or me has been 3724 * mucking with PF_KEY messages. 3725 */ 3726 static void 3727 esp_parse_pfkey(mblk_t *mp, ipsecesp_stack_t *espstack) 3728 { 3729 mblk_t *msg = mp->b_cont; 3730 sadb_msg_t *samsg; 3731 keysock_in_t *ksi; 3732 int error; 3733 int diagnostic = SADB_X_DIAGNOSTIC_NONE; 3734 3735 ASSERT(msg != NULL); 3736 3737 samsg = (sadb_msg_t *)msg->b_rptr; 3738 ksi = (keysock_in_t *)mp->b_rptr; 3739 3740 /* 3741 * If applicable, convert unspecified AF_INET6 to unspecified 3742 * AF_INET. And do other address reality checks. 3743 */ 3744 if (!sadb_addrfix(ksi, espstack->esp_pfkey_q, mp, 3745 espstack->ipsecesp_netstack) || 3746 esp_pfkey_reality_failures(mp, ksi, espstack)) { 3747 return; 3748 } 3749 3750 switch (samsg->sadb_msg_type) { 3751 case SADB_ADD: 3752 error = esp_add_sa(mp, ksi, &diagnostic, 3753 espstack->ipsecesp_netstack); 3754 if (error != 0) { 3755 sadb_pfkey_error(espstack->esp_pfkey_q, mp, error, 3756 diagnostic, ksi->ks_in_serial); 3757 } 3758 /* else esp_add_sa() took care of things. */ 3759 break; 3760 case SADB_DELETE: 3761 case SADB_X_DELPAIR: 3762 error = esp_del_sa(mp, ksi, &diagnostic, espstack, 3763 samsg->sadb_msg_type); 3764 if (error != 0) { 3765 sadb_pfkey_error(espstack->esp_pfkey_q, mp, error, 3766 diagnostic, ksi->ks_in_serial); 3767 } 3768 /* Else esp_del_sa() took care of things. */ 3769 break; 3770 case SADB_GET: 3771 error = sadb_delget_sa(mp, ksi, &espstack->esp_sadb, 3772 &diagnostic, espstack->esp_pfkey_q, samsg->sadb_msg_type); 3773 if (error != 0) { 3774 sadb_pfkey_error(espstack->esp_pfkey_q, mp, error, 3775 diagnostic, ksi->ks_in_serial); 3776 } 3777 /* Else sadb_get_sa() took care of things. */ 3778 break; 3779 case SADB_FLUSH: 3780 sadbp_flush(&espstack->esp_sadb, espstack->ipsecesp_netstack); 3781 sadb_pfkey_echo(espstack->esp_pfkey_q, mp, samsg, ksi, NULL); 3782 break; 3783 case SADB_REGISTER: 3784 /* 3785 * Hmmm, let's do it! Check for extensions (there should 3786 * be none), extract the fields, call esp_register_out(), 3787 * then either free or report an error. 3788 * 3789 * Keysock takes care of the PF_KEY bookkeeping for this. 3790 */ 3791 if (esp_register_out(samsg->sadb_msg_seq, samsg->sadb_msg_pid, 3792 ksi->ks_in_serial, espstack)) { 3793 freemsg(mp); 3794 } else { 3795 /* 3796 * Only way this path hits is if there is a memory 3797 * failure. It will not return B_FALSE because of 3798 * lack of esp_pfkey_q if I am in wput(). 3799 */ 3800 sadb_pfkey_error(espstack->esp_pfkey_q, mp, ENOMEM, 3801 diagnostic, ksi->ks_in_serial); 3802 } 3803 break; 3804 case SADB_UPDATE: 3805 case SADB_X_UPDATEPAIR: 3806 /* 3807 * Find a larval, if not there, find a full one and get 3808 * strict. 3809 */ 3810 error = esp_update_sa(mp, ksi, &diagnostic, espstack, 3811 samsg->sadb_msg_type); 3812 if (error != 0) { 3813 sadb_pfkey_error(espstack->esp_pfkey_q, mp, error, 3814 diagnostic, ksi->ks_in_serial); 3815 } 3816 /* else esp_update_sa() took care of things. */ 3817 break; 3818 case SADB_GETSPI: 3819 /* 3820 * Reserve a new larval entry. 3821 */ 3822 esp_getspi(mp, ksi, espstack); 3823 break; 3824 case SADB_ACQUIRE: 3825 /* 3826 * Find larval and/or ACQUIRE record and kill it (them), I'm 3827 * most likely an error. Inbound ACQUIRE messages should only 3828 * have the base header. 3829 */ 3830 sadb_in_acquire(samsg, &espstack->esp_sadb, 3831 espstack->esp_pfkey_q, espstack->ipsecesp_netstack); 3832 freemsg(mp); 3833 break; 3834 case SADB_DUMP: 3835 /* 3836 * Dump all entries. 3837 */ 3838 esp_dump(mp, ksi, espstack); 3839 /* esp_dump will take care of the return message, etc. */ 3840 break; 3841 case SADB_EXPIRE: 3842 /* Should never reach me. */ 3843 sadb_pfkey_error(espstack->esp_pfkey_q, mp, EOPNOTSUPP, 3844 diagnostic, ksi->ks_in_serial); 3845 break; 3846 default: 3847 sadb_pfkey_error(espstack->esp_pfkey_q, mp, EINVAL, 3848 SADB_X_DIAGNOSTIC_UNKNOWN_MSG, ksi->ks_in_serial); 3849 break; 3850 } 3851 } 3852 3853 /* 3854 * Handle case where PF_KEY says it can't find a keysock for one of my 3855 * ACQUIRE messages. 3856 */ 3857 static void 3858 esp_keysock_no_socket(mblk_t *mp, ipsecesp_stack_t *espstack) 3859 { 3860 sadb_msg_t *samsg; 3861 keysock_out_err_t *kse = (keysock_out_err_t *)mp->b_rptr; 3862 3863 if (mp->b_cont == NULL) { 3864 freemsg(mp); 3865 return; 3866 } 3867 samsg = (sadb_msg_t *)mp->b_cont->b_rptr; 3868 3869 /* 3870 * If keysock can't find any registered, delete the acquire record 3871 * immediately, and handle errors. 3872 */ 3873 if (samsg->sadb_msg_type == SADB_ACQUIRE) { 3874 samsg->sadb_msg_errno = kse->ks_err_errno; 3875 samsg->sadb_msg_len = SADB_8TO64(sizeof (*samsg)); 3876 /* 3877 * Use the write-side of the esp_pfkey_q, in case there is 3878 * no esp_sadb.s_ip_q. 3879 */ 3880 sadb_in_acquire(samsg, &espstack->esp_sadb, 3881 WR(espstack->esp_pfkey_q), espstack->ipsecesp_netstack); 3882 } 3883 3884 freemsg(mp); 3885 } 3886 3887 /* 3888 * ESP module write put routine. 3889 */ 3890 static void 3891 ipsecesp_wput(queue_t *q, mblk_t *mp) 3892 { 3893 ipsec_info_t *ii; 3894 struct iocblk *iocp; 3895 ipsecesp_stack_t *espstack = (ipsecesp_stack_t *)q->q_ptr; 3896 3897 esp3dbg(espstack, ("In esp_wput().\n")); 3898 3899 /* NOTE: Each case must take care of freeing or passing mp. */ 3900 switch (mp->b_datap->db_type) { 3901 case M_CTL: 3902 if ((mp->b_wptr - mp->b_rptr) < sizeof (ipsec_info_t)) { 3903 /* Not big enough message. */ 3904 freemsg(mp); 3905 break; 3906 } 3907 ii = (ipsec_info_t *)mp->b_rptr; 3908 3909 switch (ii->ipsec_info_type) { 3910 case KEYSOCK_OUT_ERR: 3911 esp1dbg(espstack, ("Got KEYSOCK_OUT_ERR message.\n")); 3912 esp_keysock_no_socket(mp, espstack); 3913 break; 3914 case KEYSOCK_IN: 3915 ESP_BUMP_STAT(espstack, keysock_in); 3916 esp3dbg(espstack, ("Got KEYSOCK_IN message.\n")); 3917 3918 /* Parse the message. */ 3919 esp_parse_pfkey(mp, espstack); 3920 break; 3921 case KEYSOCK_HELLO: 3922 sadb_keysock_hello(&espstack->esp_pfkey_q, q, mp, 3923 esp_ager, (void *)espstack, &espstack->esp_event, 3924 SADB_SATYPE_ESP); 3925 break; 3926 default: 3927 esp2dbg(espstack, ("Got M_CTL from above of 0x%x.\n", 3928 ii->ipsec_info_type)); 3929 freemsg(mp); 3930 break; 3931 } 3932 break; 3933 case M_IOCTL: 3934 iocp = (struct iocblk *)mp->b_rptr; 3935 switch (iocp->ioc_cmd) { 3936 case ND_SET: 3937 case ND_GET: 3938 if (nd_getset(q, espstack->ipsecesp_g_nd, mp)) { 3939 qreply(q, mp); 3940 return; 3941 } else { 3942 iocp->ioc_error = ENOENT; 3943 } 3944 /* FALLTHRU */ 3945 default: 3946 /* We really don't support any other ioctls, do we? */ 3947 3948 /* Return EINVAL */ 3949 if (iocp->ioc_error != ENOENT) 3950 iocp->ioc_error = EINVAL; 3951 iocp->ioc_count = 0; 3952 mp->b_datap->db_type = M_IOCACK; 3953 qreply(q, mp); 3954 return; 3955 } 3956 default: 3957 esp3dbg(espstack, 3958 ("Got default message, type %d, passing to IP.\n", 3959 mp->b_datap->db_type)); 3960 putnext(q, mp); 3961 } 3962 } 3963 3964 /* 3965 * Process an outbound ESP packet that can be accelerated by a IPsec 3966 * hardware acceleration capable Provider. 3967 * The caller already inserted and initialized the ESP header. 3968 * This function allocates a tagging M_CTL, and adds room at the end 3969 * of the packet to hold the ICV if authentication is needed. 3970 * 3971 * On success returns B_TRUE, on failure returns B_FALSE and frees the 3972 * mblk chain ipsec_out. 3973 */ 3974 static ipsec_status_t 3975 esp_outbound_accelerated(mblk_t *ipsec_out, uint_t icv_len) 3976 { 3977 ipsec_out_t *io; 3978 mblk_t *lastmp; 3979 netstack_t *ns; 3980 ipsecesp_stack_t *espstack; 3981 ipsec_stack_t *ipss; 3982 3983 io = (ipsec_out_t *)ipsec_out->b_rptr; 3984 ns = io->ipsec_out_ns; 3985 espstack = ns->netstack_ipsecesp; 3986 ipss = ns->netstack_ipsec; 3987 3988 ESP_BUMP_STAT(espstack, out_accelerated); 3989 3990 /* mark packet as being accelerated in IPSEC_OUT */ 3991 ASSERT(io->ipsec_out_accelerated == B_FALSE); 3992 io->ipsec_out_accelerated = B_TRUE; 3993 3994 /* 3995 * add room at the end of the packet for the ICV if needed 3996 */ 3997 if (icv_len > 0) { 3998 /* go to last mblk */ 3999 lastmp = ipsec_out; /* For following while loop. */ 4000 do { 4001 lastmp = lastmp->b_cont; 4002 } while (lastmp->b_cont != NULL); 4003 4004 /* if not enough available room, allocate new mblk */ 4005 if ((lastmp->b_wptr + icv_len) > lastmp->b_datap->db_lim) { 4006 lastmp->b_cont = allocb(icv_len, BPRI_HI); 4007 if (lastmp->b_cont == NULL) { 4008 ESP_BUMP_STAT(espstack, out_discards); 4009 ip_drop_packet(ipsec_out, B_FALSE, NULL, NULL, 4010 DROPPER(ipss, ipds_esp_nomem), 4011 &espstack->esp_dropper); 4012 return (IPSEC_STATUS_FAILED); 4013 } 4014 lastmp = lastmp->b_cont; 4015 } 4016 lastmp->b_wptr += icv_len; 4017 } 4018 4019 return (IPSEC_STATUS_SUCCESS); 4020 } 4021 4022 /* 4023 * Process an inbound accelerated ESP packet. 4024 * On success returns B_TRUE, on failure returns B_FALSE and frees the 4025 * mblk chain ipsec_in. 4026 */ 4027 static ipsec_status_t 4028 esp_inbound_accelerated(mblk_t *ipsec_in, mblk_t *data_mp, boolean_t isv4, 4029 ipsa_t *assoc) 4030 { 4031 ipsec_in_t *ii = (ipsec_in_t *)ipsec_in->b_rptr; 4032 mblk_t *hada_mp; 4033 uint32_t icv_len = 0; 4034 da_ipsec_t *hada; 4035 ipha_t *ipha; 4036 ip6_t *ip6h; 4037 kstat_named_t *counter; 4038 netstack_t *ns = ii->ipsec_in_ns; 4039 ipsecesp_stack_t *espstack = ns->netstack_ipsecesp; 4040 ipsec_stack_t *ipss = ns->netstack_ipsec; 4041 4042 ESP_BUMP_STAT(espstack, in_accelerated); 4043 4044 hada_mp = ii->ipsec_in_da; 4045 ASSERT(hada_mp != NULL); 4046 hada = (da_ipsec_t *)hada_mp->b_rptr; 4047 4048 /* 4049 * We only support one level of decapsulation in hardware, so 4050 * nuke the pointer. 4051 */ 4052 ii->ipsec_in_da = NULL; 4053 ii->ipsec_in_accelerated = B_FALSE; 4054 4055 if (assoc->ipsa_auth_alg != IPSA_AALG_NONE) { 4056 /* 4057 * ESP with authentication. We expect the Provider to have 4058 * computed the ICV and placed it in the hardware acceleration 4059 * data attributes. 4060 * 4061 * Extract ICV length from attributes M_CTL and sanity check 4062 * its value. We allow the mblk to be smaller than da_ipsec_t 4063 * for a small ICV, as long as the entire ICV fits within the 4064 * mblk. 4065 * 4066 * Also ensures that the ICV length computed by Provider 4067 * corresponds to the ICV length of the agorithm specified by 4068 * the SA. 4069 */ 4070 icv_len = hada->da_icv_len; 4071 if ((icv_len != assoc->ipsa_mac_len) || 4072 (icv_len > DA_ICV_MAX_LEN) || (MBLKL(hada_mp) < 4073 (sizeof (da_ipsec_t) - DA_ICV_MAX_LEN + icv_len))) { 4074 esp0dbg(("esp_inbound_accelerated: " 4075 "ICV len (%u) incorrect or mblk too small (%u)\n", 4076 icv_len, (uint32_t)(MBLKL(hada_mp)))); 4077 counter = DROPPER(ipss, ipds_esp_bad_auth); 4078 goto esp_in_discard; 4079 } 4080 } 4081 4082 /* get pointers to IP header */ 4083 if (isv4) { 4084 ipha = (ipha_t *)data_mp->b_rptr; 4085 } else { 4086 ip6h = (ip6_t *)data_mp->b_rptr; 4087 } 4088 4089 /* 4090 * Compare ICV in ESP packet vs ICV computed by adapter. 4091 * We also remove the ICV from the end of the packet since 4092 * it will no longer be needed. 4093 * 4094 * Assume that esp_inbound() already ensured that the pkt 4095 * was in one mblk. 4096 */ 4097 ASSERT(data_mp->b_cont == NULL); 4098 data_mp->b_wptr -= icv_len; 4099 /* adjust IP header */ 4100 if (isv4) 4101 ipha->ipha_length = htons(ntohs(ipha->ipha_length) - icv_len); 4102 else 4103 ip6h->ip6_plen = htons(ntohs(ip6h->ip6_plen) - icv_len); 4104 if (icv_len && bcmp(hada->da_icv, data_mp->b_wptr, icv_len)) { 4105 int af; 4106 void *addr; 4107 4108 if (isv4) { 4109 addr = &ipha->ipha_dst; 4110 af = AF_INET; 4111 } else { 4112 addr = &ip6h->ip6_dst; 4113 af = AF_INET6; 4114 } 4115 4116 /* 4117 * Log the event. Don't print to the console, block 4118 * potential denial-of-service attack. 4119 */ 4120 ESP_BUMP_STAT(espstack, bad_auth); 4121 ipsec_assocfailure(info.mi_idnum, 0, 0, SL_ERROR | SL_WARN, 4122 "ESP Authentication failed spi %x, dst_addr %s", 4123 assoc->ipsa_spi, addr, af, espstack->ipsecesp_netstack); 4124 counter = DROPPER(ipss, ipds_esp_bad_auth); 4125 goto esp_in_discard; 4126 } 4127 4128 esp3dbg(espstack, ("esp_inbound_accelerated: ESP authentication " 4129 "succeeded, checking replay\n")); 4130 4131 ipsec_in->b_cont = data_mp; 4132 4133 /* 4134 * Remove ESP header and padding from packet. 4135 */ 4136 if (!esp_strip_header(data_mp, ii->ipsec_in_v4, assoc->ipsa_iv_len, 4137 &counter, espstack)) { 4138 esp1dbg(espstack, ("esp_inbound_accelerated: " 4139 "esp_strip_header() failed\n")); 4140 goto esp_in_discard; 4141 } 4142 4143 freeb(hada_mp); 4144 4145 /* 4146 * Account for usage.. 4147 */ 4148 if (!esp_age_bytes(assoc, msgdsize(data_mp), B_TRUE)) { 4149 /* The ipsa has hit hard expiration, LOG and AUDIT. */ 4150 ESP_BUMP_STAT(espstack, bytes_expired); 4151 IP_ESP_BUMP_STAT(ipss, in_discards); 4152 ipsec_assocfailure(info.mi_idnum, 0, 0, SL_ERROR | SL_WARN, 4153 "ESP association 0x%x, dst %s had bytes expire.\n", 4154 assoc->ipsa_spi, assoc->ipsa_dstaddr, assoc->ipsa_addrfam, 4155 espstack->ipsecesp_netstack); 4156 ip_drop_packet(ipsec_in, B_TRUE, NULL, NULL, 4157 DROPPER(ipss, ipds_esp_bytes_expire), 4158 &espstack->esp_dropper); 4159 return (IPSEC_STATUS_FAILED); 4160 } 4161 4162 /* done processing the packet */ 4163 return (IPSEC_STATUS_SUCCESS); 4164 4165 esp_in_discard: 4166 IP_ESP_BUMP_STAT(ipss, in_discards); 4167 freeb(hada_mp); 4168 4169 ipsec_in->b_cont = data_mp; /* For ip_drop_packet()'s sake... */ 4170 ip_drop_packet(ipsec_in, B_TRUE, NULL, NULL, counter, 4171 &espstack->esp_dropper); 4172 4173 return (IPSEC_STATUS_FAILED); 4174 } 4175 4176 /* 4177 * Wrapper to allow IP to trigger an ESP association failure message 4178 * during inbound SA selection. 4179 */ 4180 void 4181 ipsecesp_in_assocfailure(mblk_t *mp, char level, ushort_t sl, char *fmt, 4182 uint32_t spi, void *addr, int af, ipsecesp_stack_t *espstack) 4183 { 4184 ipsec_stack_t *ipss = espstack->ipsecesp_netstack->netstack_ipsec; 4185 4186 if (espstack->ipsecesp_log_unknown_spi) { 4187 ipsec_assocfailure(info.mi_idnum, 0, level, sl, fmt, spi, 4188 addr, af, espstack->ipsecesp_netstack); 4189 } 4190 4191 ip_drop_packet(mp, B_TRUE, NULL, NULL, 4192 DROPPER(ipss, ipds_esp_no_sa), 4193 &espstack->esp_dropper); 4194 } 4195 4196 /* 4197 * Initialize the ESP input and output processing functions. 4198 */ 4199 void 4200 ipsecesp_init_funcs(ipsa_t *sa) 4201 { 4202 if (sa->ipsa_output_func == NULL) 4203 sa->ipsa_output_func = esp_outbound; 4204 if (sa->ipsa_input_func == NULL) 4205 sa->ipsa_input_func = esp_inbound; 4206 } 4207