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