1 /* $NetBSD: ipsec.c,v 1.170 2019/08/07 10:10:00 knakahara Exp $ */ 2 /* $FreeBSD: ipsec.c,v 1.2.2.2 2003/07/01 01:38:13 sam Exp $ */ 3 /* $KAME: ipsec.c,v 1.103 2001/05/24 07:14:18 sakane Exp $ */ 4 5 /* 6 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. 7 * All rights reserved. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 3. Neither the name of the project nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 */ 33 34 #include <sys/cdefs.h> 35 __KERNEL_RCSID(0, "$NetBSD: ipsec.c,v 1.170 2019/08/07 10:10:00 knakahara Exp $"); 36 37 /* 38 * IPsec controller part. 39 */ 40 41 #if defined(_KERNEL_OPT) 42 #include "opt_inet.h" 43 #include "opt_ipsec.h" 44 #endif 45 46 #include <sys/param.h> 47 #include <sys/systm.h> 48 #include <sys/mbuf.h> 49 #include <sys/domain.h> 50 #include <sys/protosw.h> 51 #include <sys/socket.h> 52 #include <sys/socketvar.h> 53 #include <sys/errno.h> 54 #include <sys/time.h> 55 #include <sys/kernel.h> 56 #include <sys/syslog.h> 57 #include <sys/sysctl.h> 58 #include <sys/proc.h> 59 #include <sys/kauth.h> 60 #include <sys/cpu.h> 61 #include <sys/kmem.h> 62 #include <sys/pserialize.h> 63 64 #include <net/if.h> 65 #include <net/route.h> 66 67 #include <netinet/in.h> 68 #include <netinet/in_systm.h> 69 #include <netinet/ip.h> 70 #include <netinet/ip_var.h> 71 #include <netinet/in_var.h> 72 #include <netinet/udp.h> 73 #include <netinet/udp_var.h> 74 #include <netinet/tcp.h> 75 #include <netinet/udp.h> 76 #include <netinet/ip_icmp.h> 77 #include <netinet/ip_private.h> 78 79 #include <netinet/ip6.h> 80 #ifdef INET6 81 #include <netinet6/ip6_var.h> 82 #endif 83 #include <netinet/in_pcb.h> 84 #include <netinet/in_offload.h> 85 #ifdef INET6 86 #include <netinet6/in6_pcb.h> 87 #include <netinet/icmp6.h> 88 #endif 89 90 #include <netipsec/ipsec.h> 91 #include <netipsec/ipsec_var.h> 92 #include <netipsec/ipsec_private.h> 93 #ifdef INET6 94 #include <netipsec/ipsec6.h> 95 #endif 96 #include <netipsec/ah_var.h> 97 #include <netipsec/esp_var.h> 98 #include <netipsec/ipcomp.h> /*XXX*/ 99 #include <netipsec/ipcomp_var.h> 100 101 #include <netipsec/key.h> 102 #include <netipsec/keydb.h> 103 #include <netipsec/key_debug.h> 104 105 #include <netipsec/xform.h> 106 107 int ipsec_used = 0; 108 int ipsec_enabled = 1; 109 110 #ifdef IPSEC_DEBUG 111 int ipsec_debug = 1; 112 113 /* 114 * When set to 1, IPsec will send packets with the same sequence number. 115 * This allows to verify if the other side has proper replay attacks detection. 116 */ 117 int ipsec_replay = 0; 118 119 /* 120 * When set 1, IPsec will send packets with corrupted HMAC. 121 * This allows to verify if the other side properly detects modified packets. 122 */ 123 int ipsec_integrity = 0; 124 #else 125 int ipsec_debug = 0; 126 #endif 127 128 percpu_t *ipsecstat_percpu; 129 130 int ip4_ah_offsetmask = 0; /* maybe IP_DF? */ 131 int ip4_ipsec_dfbit = 2; /* DF bit on encap. 0: clear 1: set 2: copy */ 132 int ip4_esp_trans_deflev = IPSEC_LEVEL_USE; 133 int ip4_esp_net_deflev = IPSEC_LEVEL_USE; 134 int ip4_ah_trans_deflev = IPSEC_LEVEL_USE; 135 int ip4_ah_net_deflev = IPSEC_LEVEL_USE; 136 struct secpolicy ip4_def_policy; 137 int ip4_ipsec_ecn = 0; /* ECN ignore(-1)/forbidden(0)/allowed(1) */ 138 139 u_int ipsec_spdgen = 1; /* SPD generation # */ 140 141 static struct secpolicy ipsec_dummy_sp __read_mostly = { 142 .state = IPSEC_SPSTATE_ALIVE, 143 /* If ENTRUST, the dummy SP never be used. See ipsec_getpolicybysock. */ 144 .policy = IPSEC_POLICY_ENTRUST, 145 }; 146 147 static struct secpolicy *ipsec_checkpcbcache(struct mbuf *, 148 struct inpcbpolicy *, int); 149 static int ipsec_fillpcbcache(struct inpcbpolicy *, struct mbuf *, 150 struct secpolicy *, int); 151 static int ipsec_invalpcbcache(struct inpcbpolicy *, int); 152 153 /* 154 * Crypto support requirements: 155 * 156 * 1 require hardware support 157 * -1 require software support 158 * 0 take anything 159 */ 160 int crypto_support = 0; 161 162 static struct secpolicy *ipsec_getpolicybysock(struct mbuf *, u_int, 163 struct inpcb_hdr *, int *); 164 165 #ifdef INET6 166 int ip6_esp_trans_deflev = IPSEC_LEVEL_USE; 167 int ip6_esp_net_deflev = IPSEC_LEVEL_USE; 168 int ip6_ah_trans_deflev = IPSEC_LEVEL_USE; 169 int ip6_ah_net_deflev = IPSEC_LEVEL_USE; 170 struct secpolicy ip6_def_policy; 171 int ip6_ipsec_ecn = 0; /* ECN ignore(-1)/forbidden(0)/allowed(1) */ 172 #endif 173 174 static int ipsec_setspidx_inpcb(struct mbuf *, void *); 175 static int ipsec_setspidx(struct mbuf *, struct secpolicyindex *, int, int); 176 static void ipsec4_get_ulp(struct mbuf *m, struct secpolicyindex *, int); 177 static int ipsec4_setspidx_ipaddr(struct mbuf *, struct secpolicyindex *); 178 #ifdef INET6 179 static void ipsec6_get_ulp(struct mbuf *m, struct secpolicyindex *, int); 180 static int ipsec6_setspidx_ipaddr(struct mbuf *, struct secpolicyindex *); 181 #endif 182 static void ipsec_delpcbpolicy(struct inpcbpolicy *); 183 static void ipsec_destroy_policy(struct secpolicy *); 184 static int ipsec_sp_reject(const struct secpolicy *, const struct mbuf *); 185 static void vshiftl(unsigned char *, int, int); 186 static size_t ipsec_sp_hdrsiz(const struct secpolicy *, const struct mbuf *); 187 188 /* 189 * Try to validate and use cached policy on a PCB. 190 */ 191 static struct secpolicy * 192 ipsec_checkpcbcache(struct mbuf *m, struct inpcbpolicy *pcbsp, int dir) 193 { 194 struct secpolicyindex spidx; 195 struct secpolicy *sp = NULL; 196 int s; 197 198 KASSERT(IPSEC_DIR_IS_VALID(dir)); 199 KASSERT(pcbsp != NULL); 200 KASSERT(dir < __arraycount(pcbsp->sp_cache)); 201 KASSERT(inph_locked(pcbsp->sp_inph)); 202 203 /* 204 * Checking the generation and sp->state and taking a reference to an SP 205 * must be in a critical section of pserialize. See key_unlink_sp. 206 */ 207 s = pserialize_read_enter(); 208 /* SPD table change invalidate all the caches. */ 209 if (ipsec_spdgen != pcbsp->sp_cache[dir].cachegen) { 210 ipsec_invalpcbcache(pcbsp, dir); 211 goto out; 212 } 213 sp = pcbsp->sp_cache[dir].cachesp; 214 if (sp == NULL) 215 goto out; 216 if (sp->state != IPSEC_SPSTATE_ALIVE) { 217 sp = NULL; 218 ipsec_invalpcbcache(pcbsp, dir); 219 goto out; 220 } 221 if ((pcbsp->sp_cacheflags & IPSEC_PCBSP_CONNECTED) == 0) { 222 /* NB: assume ipsec_setspidx never sleep */ 223 if (ipsec_setspidx(m, &spidx, dir, 1) != 0) { 224 sp = NULL; 225 goto out; 226 } 227 228 /* 229 * We have to make an exact match here since the cached rule 230 * might have lower priority than a rule that would otherwise 231 * have matched the packet. 232 */ 233 if (memcmp(&pcbsp->sp_cache[dir].cacheidx, &spidx, 234 sizeof(spidx))) { 235 sp = NULL; 236 goto out; 237 } 238 } else { 239 /* 240 * The pcb is connected, and the L4 code is sure that: 241 * - outgoing side uses inp_[lf]addr 242 * - incoming side looks up policy after inpcb lookup 243 * and address pair is know to be stable. We do not need 244 * to generate spidx again, nor check the address match again. 245 * 246 * For IPv4/v6 SOCK_STREAM sockets, this assumptions holds 247 * and there are calls to ipsec_pcbconn() from in_pcbconnect(). 248 */ 249 } 250 251 sp->lastused = time_second; 252 KEY_SP_REF(sp); 253 KEYDEBUG_PRINTF(KEYDEBUG_IPSEC_STAMP, 254 "DP cause refcnt++:%d SP:%p\n", 255 key_sp_refcnt(sp), pcbsp->sp_cache[dir].cachesp); 256 out: 257 pserialize_read_exit(s); 258 return sp; 259 } 260 261 static int 262 ipsec_fillpcbcache(struct inpcbpolicy *pcbsp, struct mbuf *m, 263 struct secpolicy *sp, int dir) 264 { 265 266 KASSERT(IPSEC_DIR_IS_INOROUT(dir)); 267 KASSERT(dir < __arraycount(pcbsp->sp_cache)); 268 KASSERT(inph_locked(pcbsp->sp_inph)); 269 270 pcbsp->sp_cache[dir].cachesp = NULL; 271 pcbsp->sp_cache[dir].cachehint = IPSEC_PCBHINT_UNKNOWN; 272 if (ipsec_setspidx(m, &pcbsp->sp_cache[dir].cacheidx, dir, 1) != 0) { 273 return EINVAL; 274 } 275 pcbsp->sp_cache[dir].cachesp = sp; 276 if (pcbsp->sp_cache[dir].cachesp) { 277 /* 278 * If the PCB is connected, we can remember a hint to 279 * possibly short-circuit IPsec processing in other places. 280 */ 281 if (pcbsp->sp_cacheflags & IPSEC_PCBSP_CONNECTED) { 282 switch (pcbsp->sp_cache[dir].cachesp->policy) { 283 case IPSEC_POLICY_NONE: 284 case IPSEC_POLICY_BYPASS: 285 pcbsp->sp_cache[dir].cachehint = 286 IPSEC_PCBHINT_NO; 287 break; 288 default: 289 pcbsp->sp_cache[dir].cachehint = 290 IPSEC_PCBHINT_YES; 291 } 292 } 293 } 294 pcbsp->sp_cache[dir].cachegen = ipsec_spdgen; 295 296 return 0; 297 } 298 299 static int 300 ipsec_invalpcbcache(struct inpcbpolicy *pcbsp, int dir) 301 { 302 int i; 303 304 KASSERT(inph_locked(pcbsp->sp_inph)); 305 306 for (i = IPSEC_DIR_INBOUND; i <= IPSEC_DIR_OUTBOUND; i++) { 307 if (dir != IPSEC_DIR_ANY && i != dir) 308 continue; 309 pcbsp->sp_cache[i].cachesp = NULL; 310 pcbsp->sp_cache[i].cachehint = IPSEC_PCBHINT_UNKNOWN; 311 pcbsp->sp_cache[i].cachegen = 0; 312 memset(&pcbsp->sp_cache[i].cacheidx, 0, 313 sizeof(pcbsp->sp_cache[i].cacheidx)); 314 } 315 return 0; 316 } 317 318 void 319 ipsec_pcbconn(struct inpcbpolicy *pcbsp) 320 { 321 322 KASSERT(inph_locked(pcbsp->sp_inph)); 323 324 pcbsp->sp_cacheflags |= IPSEC_PCBSP_CONNECTED; 325 ipsec_invalpcbcache(pcbsp, IPSEC_DIR_ANY); 326 } 327 328 void 329 ipsec_pcbdisconn(struct inpcbpolicy *pcbsp) 330 { 331 332 KASSERT(inph_locked(pcbsp->sp_inph)); 333 334 pcbsp->sp_cacheflags &= ~IPSEC_PCBSP_CONNECTED; 335 ipsec_invalpcbcache(pcbsp, IPSEC_DIR_ANY); 336 } 337 338 void 339 ipsec_invalpcbcacheall(void) 340 { 341 342 if (ipsec_spdgen == UINT_MAX) 343 ipsec_spdgen = 1; 344 else 345 ipsec_spdgen++; 346 } 347 348 /* 349 * Return a held reference to the default SP. 350 */ 351 static struct secpolicy * 352 key_get_default_sp(int af, const char *where, int tag) 353 { 354 struct secpolicy *sp; 355 356 KEYDEBUG_PRINTF(KEYDEBUG_IPSEC_STAMP, "DP from %s:%u\n", where, tag); 357 358 switch(af) { 359 case AF_INET: 360 sp = &ip4_def_policy; 361 break; 362 #ifdef INET6 363 case AF_INET6: 364 sp = &ip6_def_policy; 365 break; 366 #endif 367 default: 368 KEYDEBUG_PRINTF(KEYDEBUG_IPSEC_STAMP, 369 "unexpected protocol family %u\n", af); 370 return NULL; 371 } 372 373 if (sp->policy != IPSEC_POLICY_DISCARD && 374 sp->policy != IPSEC_POLICY_NONE) { 375 IPSECLOG(LOG_INFO, "fixed system default policy: %d->%d\n", 376 sp->policy, IPSEC_POLICY_NONE); 377 sp->policy = IPSEC_POLICY_NONE; 378 } 379 KEY_SP_REF(sp); 380 381 KEYDEBUG_PRINTF(KEYDEBUG_IPSEC_STAMP, "DP returns SP:%p (%u)\n", 382 sp, key_sp_refcnt(sp)); 383 return sp; 384 } 385 386 #define KEY_GET_DEFAULT_SP(af) \ 387 key_get_default_sp((af), __func__, __LINE__) 388 389 /* 390 * For OUTBOUND packet having a socket. Searching SPD for packet, 391 * and return a pointer to SP. 392 * OUT: NULL: no appropriate SP found, the following value is set to error. 393 * 0 : bypass 394 * EACCES : discard packet. 395 * ENOENT : ipsec_acquire() in progress, maybe. 396 * others : error occurred. 397 * others: a pointer to SP 398 * 399 * NOTE: IPv6 mapped address concern is implemented here. 400 */ 401 static struct secpolicy * 402 ipsec_getpolicybysock(struct mbuf *m, u_int dir, struct inpcb_hdr *inph, 403 int *error) 404 { 405 struct inpcbpolicy *pcbsp = NULL; 406 struct secpolicy *currsp = NULL; /* policy on socket */ 407 struct secpolicy *sp; 408 int af; 409 410 KASSERT(m != NULL); 411 KASSERT(inph != NULL); 412 KASSERT(error != NULL); 413 KASSERTMSG(IPSEC_DIR_IS_INOROUT(dir), "invalid direction %u", dir); 414 415 KASSERT(inph->inph_socket != NULL); 416 KASSERT(inph_locked(inph)); 417 418 /* XXX FIXME inpcb/in6pcb vs socket*/ 419 af = inph->inph_af; 420 KASSERTMSG(af == AF_INET || af == AF_INET6, 421 "unexpected protocol family %u", af); 422 423 KASSERT(inph->inph_sp != NULL); 424 /* If we have a cached entry, and if it is still valid, use it. */ 425 IPSEC_STATINC(IPSEC_STAT_SPDCACHELOOKUP); 426 currsp = ipsec_checkpcbcache(m, inph->inph_sp, dir); 427 if (currsp) { 428 *error = 0; 429 return currsp; 430 } 431 IPSEC_STATINC(IPSEC_STAT_SPDCACHEMISS); 432 433 switch (af) { 434 case AF_INET: 435 #if defined(INET6) 436 case AF_INET6: 437 #endif 438 *error = ipsec_setspidx_inpcb(m, inph); 439 pcbsp = inph->inph_sp; 440 break; 441 default: 442 *error = EPFNOSUPPORT; 443 break; 444 } 445 if (*error) 446 return NULL; 447 448 KASSERT(pcbsp != NULL); 449 switch (dir) { 450 case IPSEC_DIR_INBOUND: 451 currsp = pcbsp->sp_in; 452 break; 453 case IPSEC_DIR_OUTBOUND: 454 currsp = pcbsp->sp_out; 455 break; 456 } 457 KASSERT(currsp != NULL); 458 459 if (pcbsp->priv) { /* when privileged socket */ 460 switch (currsp->policy) { 461 case IPSEC_POLICY_BYPASS: 462 case IPSEC_POLICY_IPSEC: 463 KEY_SP_REF(currsp); 464 sp = currsp; 465 break; 466 467 case IPSEC_POLICY_ENTRUST: 468 /* look for a policy in SPD */ 469 if (key_havesp(dir)) 470 sp = KEY_LOOKUP_SP_BYSPIDX(&currsp->spidx, dir); 471 else 472 sp = NULL; 473 if (sp == NULL) /* no SP found */ 474 sp = KEY_GET_DEFAULT_SP(af); 475 break; 476 477 default: 478 IPSECLOG(LOG_ERR, "Invalid policy for PCB %d\n", 479 currsp->policy); 480 *error = EINVAL; 481 return NULL; 482 } 483 } else { /* unpriv, SPD has policy */ 484 if (key_havesp(dir)) 485 sp = KEY_LOOKUP_SP_BYSPIDX(&currsp->spidx, dir); 486 else 487 sp = NULL; 488 if (sp == NULL) { /* no SP found */ 489 switch (currsp->policy) { 490 case IPSEC_POLICY_BYPASS: 491 IPSECLOG(LOG_ERR, "Illegal policy for " 492 "non-priviliged defined %d\n", 493 currsp->policy); 494 *error = EINVAL; 495 return NULL; 496 497 case IPSEC_POLICY_ENTRUST: 498 sp = KEY_GET_DEFAULT_SP(af); 499 break; 500 501 case IPSEC_POLICY_IPSEC: 502 KEY_SP_REF(currsp); 503 sp = currsp; 504 break; 505 506 default: 507 IPSECLOG(LOG_ERR, "Invalid policy for " 508 "PCB %d\n", currsp->policy); 509 *error = EINVAL; 510 return NULL; 511 } 512 } 513 } 514 KASSERTMSG(sp != NULL, "null SP (priv %u policy %u", pcbsp->priv, 515 currsp->policy); 516 KEYDEBUG_PRINTF(KEYDEBUG_IPSEC_STAMP, 517 "DP (priv %u policy %u) allocates SP:%p (refcnt %u)\n", 518 pcbsp->priv, currsp->policy, sp, key_sp_refcnt(sp)); 519 ipsec_fillpcbcache(pcbsp, m, sp, dir); 520 return sp; 521 } 522 523 /* 524 * For FORWARDING packet or OUTBOUND without a socket. Searching SPD for packet, 525 * and return a pointer to SP. 526 * OUT: positive: a pointer to the entry for security policy leaf matched. 527 * NULL: no appropriate SP found, the following value is set to error. 528 * 0 : bypass 529 * EACCES : discard packet. 530 * ENOENT : ipsec_acquire() in progress, maybe. 531 * others : error occurred. 532 */ 533 static struct secpolicy * 534 ipsec_getpolicybyaddr(struct mbuf *m, u_int dir, int flag, int *error) 535 { 536 struct secpolicyindex spidx; 537 struct secpolicy *sp; 538 539 KASSERT(m != NULL); 540 KASSERT(error != NULL); 541 KASSERTMSG(IPSEC_DIR_IS_INOROUT(dir), "invalid direction %u", dir); 542 543 sp = NULL; 544 545 /* Make an index to look for a policy. */ 546 *error = ipsec_setspidx(m, &spidx, dir, (flag & IP_FORWARDING) ? 0 : 1); 547 if (*error != 0) { 548 IPSECLOG(LOG_DEBUG, "setpidx failed, dir %u flag %u\n", dir, flag); 549 memset(&spidx, 0, sizeof(spidx)); 550 return NULL; 551 } 552 553 spidx.dir = dir; 554 555 if (key_havesp(dir)) { 556 sp = KEY_LOOKUP_SP_BYSPIDX(&spidx, dir); 557 } 558 if (sp == NULL) { 559 /* no SP found, use system default */ 560 sp = KEY_GET_DEFAULT_SP(spidx.dst.sa.sa_family); 561 } 562 563 KASSERT(sp != NULL); 564 return sp; 565 } 566 567 static struct secpolicy * 568 ipsec_checkpolicy(struct mbuf *m, u_int dir, u_int flag, int *error, 569 void *inp) 570 { 571 struct secpolicy *sp; 572 573 *error = 0; 574 575 if (inp == NULL) { 576 sp = ipsec_getpolicybyaddr(m, dir, flag, error); 577 } else { 578 struct inpcb_hdr *inph = (struct inpcb_hdr *)inp; 579 KASSERT(inph->inph_socket != NULL); 580 sp = ipsec_getpolicybysock(m, dir, inph, error); 581 } 582 if (sp == NULL) { 583 KASSERTMSG(*error != 0, "getpolicy failed w/o error"); 584 IPSEC_STATINC(IPSEC_STAT_OUT_INVAL); 585 return NULL; 586 } 587 KASSERTMSG(*error == 0, "sp w/ error set to %u", *error); 588 589 switch (sp->policy) { 590 case IPSEC_POLICY_ENTRUST: 591 default: 592 printf("%s: invalid policy %u\n", __func__, sp->policy); 593 /* fall thru... */ 594 case IPSEC_POLICY_DISCARD: 595 IPSEC_STATINC(IPSEC_STAT_OUT_POLVIO); 596 *error = -EINVAL; /* packet is discarded by caller */ 597 break; 598 case IPSEC_POLICY_BYPASS: 599 case IPSEC_POLICY_NONE: 600 KEY_SP_UNREF(&sp); 601 sp = NULL; /* NB: force NULL result */ 602 break; 603 case IPSEC_POLICY_IPSEC: 604 KASSERT(sp->req != NULL); 605 break; 606 } 607 608 if (*error != 0) { 609 KEY_SP_UNREF(&sp); 610 sp = NULL; 611 IPSECLOG(LOG_DEBUG, "done, error %d\n", *error); 612 } 613 614 return sp; 615 } 616 617 int 618 ipsec4_output(struct mbuf *m, struct inpcb *inp, int flags, 619 u_long *mtu, bool *natt_frag, bool *done) 620 { 621 struct secpolicy *sp = NULL; 622 u_long _mtu = 0; 623 int error, s; 624 625 /* 626 * Check the security policy (SP) for the packet and, if required, 627 * do IPsec-related processing. There are two cases here; the first 628 * time a packet is sent through it will be untagged and handled by 629 * ipsec_checkpolicy(). If the packet is resubmitted to ip_output 630 * (e.g. after AH, ESP, etc. processing), there will be a tag to 631 * bypass the lookup and related policy checking. 632 */ 633 if (ipsec_outdone(m)) { 634 return 0; 635 } 636 s = splsoftnet(); 637 if (inp && ipsec_pcb_skip_ipsec(inp->inp_sp, IPSEC_DIR_OUTBOUND)) { 638 splx(s); 639 return 0; 640 } 641 sp = ipsec_checkpolicy(m, IPSEC_DIR_OUTBOUND, flags, &error, inp); 642 643 /* 644 * There are four return cases: 645 * sp != NULL apply IPsec policy 646 * sp == NULL, error == 0 no IPsec handling needed 647 * sp == NULL, error == -EINVAL discard packet w/o error 648 * sp == NULL, error != 0 discard packet, report error 649 */ 650 if (sp == NULL) { 651 splx(s); 652 if (error) { 653 /* 654 * Hack: -EINVAL is used to signal that a packet 655 * should be silently discarded. This is typically 656 * because we asked key management for an SA and 657 * it was delayed (e.g. kicked up to IKE). 658 */ 659 if (error == -EINVAL) 660 error = 0; 661 m_freem(m); 662 *done = true; 663 return error; 664 } 665 /* No IPsec processing for this packet. */ 666 return 0; 667 } 668 669 /* 670 * Do delayed checksums now because we send before 671 * this is done in the normal processing path. 672 */ 673 if (m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) { 674 in_undefer_cksum_tcpudp(m); 675 m->m_pkthdr.csum_flags &= ~(M_CSUM_TCPv4|M_CSUM_UDPv4); 676 } 677 678 error = ipsec4_process_packet(m, sp->req, &_mtu); 679 if (error == 0 && _mtu != 0) { 680 /* 681 * NAT-T ESP fragmentation: do not do IPSec processing 682 * now, we will do it on each fragmented packet. 683 */ 684 *mtu = _mtu; 685 *natt_frag = true; 686 KEY_SP_UNREF(&sp); 687 splx(s); 688 return 0; 689 } 690 691 /* 692 * Preserve KAME behaviour: ENOENT can be returned 693 * when an SA acquire is in progress. Don't propagate 694 * this to user-level; it confuses applications. 695 * 696 * XXX this will go away when the SADB is redone. 697 */ 698 if (error == ENOENT) 699 error = 0; 700 KEY_SP_UNREF(&sp); 701 splx(s); 702 *done = true; 703 return error; 704 } 705 706 int 707 ipsec_ip_input(struct mbuf *m, bool forward) 708 { 709 struct secpolicy *sp; 710 int error, s; 711 712 s = splsoftnet(); 713 error = ipsec_in_reject(m, NULL); 714 splx(s); 715 if (error) { 716 return EINVAL; 717 } 718 719 if (!forward || !(m->m_flags & M_CANFASTFWD)) { 720 return 0; 721 } 722 723 /* 724 * Peek at the outbound SP for this packet to determine if 725 * it is a Fast Forward candidate. 726 */ 727 s = splsoftnet(); 728 sp = ipsec_checkpolicy(m, IPSEC_DIR_OUTBOUND, IP_FORWARDING, 729 &error, NULL); 730 if (sp != NULL) { 731 m->m_flags &= ~M_CANFASTFWD; 732 KEY_SP_UNREF(&sp); 733 } 734 splx(s); 735 736 return 0; 737 } 738 739 /* 740 * If the packet is routed over IPsec tunnel, tell the originator the 741 * tunnel MTU. 742 * tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz 743 * 744 * XXX: Quick hack!!! 745 * 746 * XXX: And what if the MTU goes negative? 747 */ 748 void 749 ipsec_mtu(struct mbuf *m, int *destmtu) 750 { 751 struct secpolicy *sp; 752 size_t ipsechdr; 753 int error; 754 755 sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND, IP_FORWARDING, 756 &error); 757 if (sp == NULL) { 758 return; 759 } 760 761 /* Count IPsec header size. */ 762 ipsechdr = ipsec_sp_hdrsiz(sp, m); 763 764 /* 765 * Find the correct route for outer IP header, compute tunnel MTU. 766 */ 767 if (sp->req) { 768 struct secasvar *sav; 769 770 sav = ipsec_lookup_sa(sp->req, m); 771 if (sav != NULL) { 772 struct route *ro; 773 struct rtentry *rt; 774 775 ro = &sav->sah->sa_route; 776 rt = rtcache_validate(ro); 777 if (rt && rt->rt_ifp) { 778 *destmtu = rt->rt_rmx.rmx_mtu ? 779 rt->rt_rmx.rmx_mtu : rt->rt_ifp->if_mtu; 780 *destmtu -= ipsechdr; 781 } 782 rtcache_unref(rt, ro); 783 KEY_SA_UNREF(&sav); 784 } 785 } 786 KEY_SP_UNREF(&sp); 787 } 788 789 static int 790 ipsec_setspidx_inpcb(struct mbuf *m, void *pcb) 791 { 792 struct inpcb_hdr *inph = (struct inpcb_hdr *)pcb; 793 int error; 794 795 KASSERT(inph != NULL); 796 KASSERT(inph->inph_sp != NULL); 797 KASSERT(inph->inph_sp->sp_out != NULL); 798 KASSERT(inph->inph_sp->sp_in != NULL); 799 800 error = ipsec_setspidx(m, &inph->inph_sp->sp_in->spidx, 801 IPSEC_DIR_INBOUND, 1); 802 if (error == 0) { 803 inph->inph_sp->sp_out->spidx = inph->inph_sp->sp_in->spidx; 804 inph->inph_sp->sp_out->spidx.dir = IPSEC_DIR_OUTBOUND; 805 } else { 806 memset(&inph->inph_sp->sp_in->spidx, 0, 807 sizeof(inph->inph_sp->sp_in->spidx)); 808 memset(&inph->inph_sp->sp_out->spidx, 0, 809 sizeof(inph->inph_sp->sp_out->spidx)); 810 } 811 return error; 812 } 813 814 /* 815 * configure security policy index (src/dst/proto/sport/dport) 816 * by looking at the content of mbuf. 817 * the caller is responsible for error recovery (like clearing up spidx). 818 */ 819 static int 820 ipsec_setspidx(struct mbuf *m, struct secpolicyindex *spidx, int dir, 821 int needport) 822 { 823 struct ip *ip = NULL; 824 struct ip ipbuf; 825 u_int v; 826 int error; 827 828 KASSERT(m != NULL); 829 M_VERIFY_PACKET(m); 830 831 if (m->m_pkthdr.len < sizeof(struct ip)) { 832 KEYDEBUG_PRINTF(KEYDEBUG_IPSEC_DUMP, 833 "pkthdr.len(%d) < sizeof(struct ip), ignored.\n", 834 m->m_pkthdr.len); 835 return EINVAL; 836 } 837 838 memset(spidx, 0, sizeof(*spidx)); 839 spidx->dir = dir; 840 841 if (m->m_len >= sizeof(*ip)) { 842 ip = mtod(m, struct ip *); 843 } else { 844 m_copydata(m, 0, sizeof(ipbuf), &ipbuf); 845 ip = &ipbuf; 846 } 847 v = ip->ip_v; 848 switch (v) { 849 case 4: 850 error = ipsec4_setspidx_ipaddr(m, spidx); 851 if (error) 852 return error; 853 ipsec4_get_ulp(m, spidx, needport); 854 return 0; 855 #ifdef INET6 856 case 6: 857 if (m->m_pkthdr.len < sizeof(struct ip6_hdr)) { 858 KEYDEBUG_PRINTF(KEYDEBUG_IPSEC_DUMP, 859 "pkthdr.len(%d) < sizeof(struct ip6_hdr), " 860 "ignored.\n", m->m_pkthdr.len); 861 return EINVAL; 862 } 863 error = ipsec6_setspidx_ipaddr(m, spidx); 864 if (error) 865 return error; 866 ipsec6_get_ulp(m, spidx, needport); 867 return 0; 868 #endif 869 default: 870 KEYDEBUG_PRINTF(KEYDEBUG_IPSEC_DUMP, 871 "unknown IP version %u, ignored.\n", v); 872 return EINVAL; 873 } 874 } 875 876 static void 877 ipsec4_get_ulp(struct mbuf *m, struct secpolicyindex *spidx, int needport) 878 { 879 u_int8_t nxt; 880 int off; 881 882 KASSERT(m != NULL); 883 KASSERTMSG(m->m_pkthdr.len >= sizeof(struct ip), "packet too short"); 884 885 /* NB: ip_input() flips it into host endian XXX need more checking */ 886 if (m->m_len >= sizeof(struct ip)) { 887 struct ip *ip = mtod(m, struct ip *); 888 if (ip->ip_off & htons(IP_MF | IP_OFFMASK)) 889 goto done; 890 off = ip->ip_hl << 2; 891 nxt = ip->ip_p; 892 } else { 893 struct ip ih; 894 895 m_copydata(m, 0, sizeof(struct ip), &ih); 896 if (ih.ip_off & htons(IP_MF | IP_OFFMASK)) 897 goto done; 898 off = ih.ip_hl << 2; 899 nxt = ih.ip_p; 900 } 901 902 while (off < m->m_pkthdr.len) { 903 struct ip6_ext ip6e; 904 struct tcphdr th; 905 struct udphdr uh; 906 struct icmp icmph; 907 908 switch (nxt) { 909 case IPPROTO_TCP: 910 spidx->ul_proto = nxt; 911 if (!needport) 912 goto done_proto; 913 if (off + sizeof(struct tcphdr) > m->m_pkthdr.len) 914 goto done; 915 m_copydata(m, off, sizeof(th), &th); 916 spidx->src.sin.sin_port = th.th_sport; 917 spidx->dst.sin.sin_port = th.th_dport; 918 return; 919 case IPPROTO_UDP: 920 spidx->ul_proto = nxt; 921 if (!needport) 922 goto done_proto; 923 if (off + sizeof(struct udphdr) > m->m_pkthdr.len) 924 goto done; 925 m_copydata(m, off, sizeof(uh), &uh); 926 spidx->src.sin.sin_port = uh.uh_sport; 927 spidx->dst.sin.sin_port = uh.uh_dport; 928 return; 929 case IPPROTO_AH: 930 if (off + sizeof(ip6e) > m->m_pkthdr.len) 931 goto done; 932 /* XXX sigh, this works but is totally bogus */ 933 m_copydata(m, off, sizeof(ip6e), &ip6e); 934 off += (ip6e.ip6e_len + 2) << 2; 935 nxt = ip6e.ip6e_nxt; 936 break; 937 case IPPROTO_ICMP: 938 spidx->ul_proto = nxt; 939 if (off + sizeof(struct icmp) > m->m_pkthdr.len) 940 goto done; 941 m_copydata(m, off, sizeof(icmph), &icmph); 942 ((struct sockaddr_in *)&spidx->src)->sin_port = 943 htons((uint16_t)icmph.icmp_type); 944 ((struct sockaddr_in *)&spidx->dst)->sin_port = 945 htons((uint16_t)icmph.icmp_code); 946 return; 947 default: 948 /* XXX intermediate headers??? */ 949 spidx->ul_proto = nxt; 950 goto done_proto; 951 } 952 } 953 done: 954 spidx->ul_proto = IPSEC_ULPROTO_ANY; 955 done_proto: 956 spidx->src.sin.sin_port = IPSEC_PORT_ANY; 957 spidx->dst.sin.sin_port = IPSEC_PORT_ANY; 958 } 959 960 static int 961 ipsec4_setspidx_ipaddr(struct mbuf *m, struct secpolicyindex *spidx) 962 { 963 static const struct sockaddr_in template = { 964 sizeof(struct sockaddr_in), 965 AF_INET, 966 0, { 0 }, { 0, 0, 0, 0, 0, 0, 0, 0 } 967 }; 968 969 spidx->src.sin = template; 970 spidx->dst.sin = template; 971 972 if (m->m_len < sizeof(struct ip)) { 973 m_copydata(m, offsetof(struct ip, ip_src), 974 sizeof(struct in_addr), &spidx->src.sin.sin_addr); 975 m_copydata(m, offsetof(struct ip, ip_dst), 976 sizeof(struct in_addr), &spidx->dst.sin.sin_addr); 977 } else { 978 struct ip *ip = mtod(m, struct ip *); 979 spidx->src.sin.sin_addr = ip->ip_src; 980 spidx->dst.sin.sin_addr = ip->ip_dst; 981 } 982 983 spidx->prefs = sizeof(struct in_addr) << 3; 984 spidx->prefd = sizeof(struct in_addr) << 3; 985 986 return 0; 987 } 988 989 #ifdef INET6 990 static void 991 ipsec6_get_ulp(struct mbuf *m, struct secpolicyindex *spidx, int needport) 992 { 993 int off, nxt; 994 struct tcphdr th; 995 struct udphdr uh; 996 struct icmp6_hdr icmph; 997 998 KASSERT(m != NULL); 999 1000 if (KEYDEBUG_ON(KEYDEBUG_IPSEC_DUMP)) { 1001 kdebug_mbuf(__func__, m); 1002 } 1003 1004 /* set default */ 1005 spidx->ul_proto = IPSEC_ULPROTO_ANY; 1006 ((struct sockaddr_in6 *)&spidx->src)->sin6_port = IPSEC_PORT_ANY; 1007 ((struct sockaddr_in6 *)&spidx->dst)->sin6_port = IPSEC_PORT_ANY; 1008 1009 nxt = -1; 1010 off = ip6_lasthdr(m, 0, IPPROTO_IPV6, &nxt); 1011 if (off < 0 || m->m_pkthdr.len < off) 1012 return; 1013 1014 switch (nxt) { 1015 case IPPROTO_TCP: 1016 spidx->ul_proto = nxt; 1017 if (!needport) 1018 break; 1019 if (off + sizeof(struct tcphdr) > m->m_pkthdr.len) 1020 break; 1021 m_copydata(m, off, sizeof(th), &th); 1022 ((struct sockaddr_in6 *)&spidx->src)->sin6_port = th.th_sport; 1023 ((struct sockaddr_in6 *)&spidx->dst)->sin6_port = th.th_dport; 1024 break; 1025 case IPPROTO_UDP: 1026 spidx->ul_proto = nxt; 1027 if (!needport) 1028 break; 1029 if (off + sizeof(struct udphdr) > m->m_pkthdr.len) 1030 break; 1031 m_copydata(m, off, sizeof(uh), &uh); 1032 ((struct sockaddr_in6 *)&spidx->src)->sin6_port = uh.uh_sport; 1033 ((struct sockaddr_in6 *)&spidx->dst)->sin6_port = uh.uh_dport; 1034 break; 1035 case IPPROTO_ICMPV6: 1036 spidx->ul_proto = nxt; 1037 if (off + sizeof(struct icmp6_hdr) > m->m_pkthdr.len) 1038 break; 1039 m_copydata(m, off, sizeof(icmph), &icmph); 1040 ((struct sockaddr_in6 *)&spidx->src)->sin6_port = 1041 htons((uint16_t)icmph.icmp6_type); 1042 ((struct sockaddr_in6 *)&spidx->dst)->sin6_port = 1043 htons((uint16_t)icmph.icmp6_code); 1044 break; 1045 default: 1046 /* XXX intermediate headers??? */ 1047 spidx->ul_proto = nxt; 1048 break; 1049 } 1050 } 1051 1052 static int 1053 ipsec6_setspidx_ipaddr(struct mbuf *m, struct secpolicyindex *spidx) 1054 { 1055 struct ip6_hdr *ip6 = NULL; 1056 struct ip6_hdr ip6buf; 1057 struct sockaddr_in6 *sin6; 1058 1059 if (m->m_len >= sizeof(*ip6)) { 1060 ip6 = mtod(m, struct ip6_hdr *); 1061 } else { 1062 m_copydata(m, 0, sizeof(ip6buf), &ip6buf); 1063 ip6 = &ip6buf; 1064 } 1065 1066 sin6 = (struct sockaddr_in6 *)&spidx->src; 1067 memset(sin6, 0, sizeof(*sin6)); 1068 sin6->sin6_family = AF_INET6; 1069 sin6->sin6_len = sizeof(struct sockaddr_in6); 1070 memcpy(&sin6->sin6_addr, &ip6->ip6_src, sizeof(ip6->ip6_src)); 1071 if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) { 1072 sin6->sin6_addr.s6_addr16[1] = 0; 1073 sin6->sin6_scope_id = ntohs(ip6->ip6_src.s6_addr16[1]); 1074 } 1075 spidx->prefs = sizeof(struct in6_addr) << 3; 1076 1077 sin6 = (struct sockaddr_in6 *)&spidx->dst; 1078 memset(sin6, 0, sizeof(*sin6)); 1079 sin6->sin6_family = AF_INET6; 1080 sin6->sin6_len = sizeof(struct sockaddr_in6); 1081 memcpy(&sin6->sin6_addr, &ip6->ip6_dst, sizeof(ip6->ip6_dst)); 1082 if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_dst)) { 1083 sin6->sin6_addr.s6_addr16[1] = 0; 1084 sin6->sin6_scope_id = ntohs(ip6->ip6_dst.s6_addr16[1]); 1085 } 1086 spidx->prefd = sizeof(struct in6_addr) << 3; 1087 1088 return 0; 1089 } 1090 #endif 1091 1092 static void 1093 ipsec_delpcbpolicy(struct inpcbpolicy *p) 1094 { 1095 1096 kmem_intr_free(p, sizeof(*p)); 1097 } 1098 1099 int 1100 ipsec_init_pcbpolicy(struct socket *so, struct inpcbpolicy **policy) 1101 { 1102 struct inpcbpolicy *new; 1103 1104 KASSERT(so != NULL); 1105 KASSERT(policy != NULL); 1106 1107 new = kmem_intr_zalloc(sizeof(*new), KM_NOSLEEP); 1108 if (new == NULL) { 1109 IPSECLOG(LOG_DEBUG, "No more memory.\n"); 1110 return ENOBUFS; 1111 } 1112 1113 if (IPSEC_PRIVILEGED_SO(so)) 1114 new->priv = 1; 1115 else 1116 new->priv = 0; 1117 1118 /* 1119 * Set dummy SPs. Actual SPs will be allocated later if needed. 1120 */ 1121 new->sp_in = &ipsec_dummy_sp; 1122 new->sp_out = &ipsec_dummy_sp; 1123 1124 *policy = new; 1125 1126 return 0; 1127 } 1128 1129 static void 1130 ipsec_destroy_policy(struct secpolicy *sp) 1131 { 1132 1133 if (sp == &ipsec_dummy_sp) { 1134 ; /* It's dummy. No need to free it. */ 1135 } else { 1136 /* 1137 * We cannot destroy here because it can be called in 1138 * softint. So mark the SP as DEAD and let the timer 1139 * destroy it. See key_timehandler_spd. 1140 */ 1141 sp->state = IPSEC_SPSTATE_DEAD; 1142 } 1143 } 1144 1145 int 1146 ipsec_set_policy(void *inp, const void *request, size_t len, 1147 kauth_cred_t cred) 1148 { 1149 struct inpcb_hdr *inph = (struct inpcb_hdr *)inp; 1150 const struct sadb_x_policy *xpl; 1151 struct secpolicy *newsp, *oldsp; 1152 struct secpolicy **policy; 1153 int error; 1154 1155 KASSERT(!cpu_softintr_p()); 1156 KASSERT(inph != NULL); 1157 KASSERT(inph_locked(inph)); 1158 KASSERT(request != NULL); 1159 1160 if (len < sizeof(*xpl)) 1161 return EINVAL; 1162 xpl = (const struct sadb_x_policy *)request; 1163 1164 KASSERT(inph->inph_sp != NULL); 1165 1166 /* select direction */ 1167 switch (xpl->sadb_x_policy_dir) { 1168 case IPSEC_DIR_INBOUND: 1169 policy = &inph->inph_sp->sp_in; 1170 break; 1171 case IPSEC_DIR_OUTBOUND: 1172 policy = &inph->inph_sp->sp_out; 1173 break; 1174 default: 1175 IPSECLOG(LOG_ERR, "invalid direction=%u\n", 1176 xpl->sadb_x_policy_dir); 1177 return EINVAL; 1178 } 1179 1180 /* sanity check. */ 1181 if (policy == NULL || *policy == NULL) 1182 return EINVAL; 1183 1184 if (KEYDEBUG_ON(KEYDEBUG_IPSEC_DUMP)) { 1185 kdebug_sadb_xpolicy("set passed policy", request); 1186 } 1187 1188 /* check policy type */ 1189 /* ipsec_set_policy() accepts IPSEC, ENTRUST and BYPASS. */ 1190 if (xpl->sadb_x_policy_type == IPSEC_POLICY_DISCARD || 1191 xpl->sadb_x_policy_type == IPSEC_POLICY_NONE) 1192 return EINVAL; 1193 1194 /* check privileged socket */ 1195 if (xpl->sadb_x_policy_type == IPSEC_POLICY_BYPASS) { 1196 error = kauth_authorize_network(cred, KAUTH_NETWORK_IPSEC, 1197 KAUTH_REQ_NETWORK_IPSEC_BYPASS, NULL, NULL, NULL); 1198 if (error) 1199 return error; 1200 } 1201 1202 /* allocation new SP entry */ 1203 if ((newsp = key_msg2sp(xpl, len, &error)) == NULL) 1204 return error; 1205 1206 key_init_sp(newsp); 1207 newsp->created = time_uptime; 1208 /* Insert the global list for SPs for sockets */ 1209 key_socksplist_add(newsp); 1210 1211 /* clear old SP and set new SP */ 1212 oldsp = *policy; 1213 *policy = newsp; 1214 ipsec_destroy_policy(oldsp); 1215 1216 if (KEYDEBUG_ON(KEYDEBUG_IPSEC_DUMP)) { 1217 printf("%s: new policy\n", __func__); 1218 kdebug_secpolicy(newsp); 1219 } 1220 1221 return 0; 1222 } 1223 1224 int 1225 ipsec_get_policy(void *inp, const void *request, size_t len, 1226 struct mbuf **mp) 1227 { 1228 struct inpcb_hdr *inph = (struct inpcb_hdr *)inp; 1229 const struct sadb_x_policy *xpl; 1230 struct secpolicy *policy; 1231 1232 /* sanity check. */ 1233 if (inph == NULL || request == NULL || mp == NULL) 1234 return EINVAL; 1235 KASSERT(inph->inph_sp != NULL); 1236 if (len < sizeof(*xpl)) 1237 return EINVAL; 1238 xpl = (const struct sadb_x_policy *)request; 1239 1240 /* select direction */ 1241 switch (xpl->sadb_x_policy_dir) { 1242 case IPSEC_DIR_INBOUND: 1243 policy = inph->inph_sp->sp_in; 1244 break; 1245 case IPSEC_DIR_OUTBOUND: 1246 policy = inph->inph_sp->sp_out; 1247 break; 1248 default: 1249 IPSECLOG(LOG_ERR, "invalid direction=%u\n", 1250 xpl->sadb_x_policy_dir); 1251 return EINVAL; 1252 } 1253 1254 if (policy == NULL) 1255 return EINVAL; 1256 1257 *mp = key_sp2msg(policy, M_NOWAIT); 1258 if (!*mp) { 1259 IPSECLOG(LOG_DEBUG, "No more memory.\n"); 1260 return ENOBUFS; 1261 } 1262 1263 if (KEYDEBUG_ON(KEYDEBUG_IPSEC_DUMP)) { 1264 kdebug_mbuf(__func__, *mp); 1265 } 1266 1267 return 0; 1268 } 1269 1270 int 1271 ipsec_delete_pcbpolicy(void *inp) 1272 { 1273 struct inpcb_hdr *inph = (struct inpcb_hdr *)inp; 1274 1275 KASSERT(inph != NULL); 1276 1277 if (inph->inph_sp == NULL) 1278 return 0; 1279 1280 if (inph->inph_sp->sp_in != NULL) 1281 ipsec_destroy_policy(inph->inph_sp->sp_in); 1282 1283 if (inph->inph_sp->sp_out != NULL) 1284 ipsec_destroy_policy(inph->inph_sp->sp_out); 1285 1286 ipsec_invalpcbcache(inph->inph_sp, IPSEC_DIR_ANY); 1287 1288 ipsec_delpcbpolicy(inph->inph_sp); 1289 inph->inph_sp = NULL; 1290 1291 return 0; 1292 } 1293 1294 /* 1295 * Return the current level (either IPSEC_LEVEL_USE or IPSEC_LEVEL_REQUIRE). 1296 */ 1297 u_int 1298 ipsec_get_reqlevel(const struct ipsecrequest *isr) 1299 { 1300 u_int level = 0; 1301 u_int esp_trans_deflev, esp_net_deflev; 1302 u_int ah_trans_deflev, ah_net_deflev; 1303 1304 KASSERT(isr != NULL); 1305 KASSERT(isr->sp != NULL); 1306 KASSERTMSG( 1307 isr->sp->spidx.src.sa.sa_family == isr->sp->spidx.dst.sa.sa_family, 1308 "af family mismatch, src %u, dst %u", 1309 isr->sp->spidx.src.sa.sa_family, isr->sp->spidx.dst.sa.sa_family); 1310 1311 /* XXX note that we have ipseclog() expanded here - code sync issue */ 1312 #define IPSEC_CHECK_DEFAULT(lev) \ 1313 (((lev) != IPSEC_LEVEL_USE && (lev) != IPSEC_LEVEL_REQUIRE \ 1314 && (lev) != IPSEC_LEVEL_UNIQUE) ? \ 1315 (ipsec_debug ? log(LOG_INFO, "fixed system default level " #lev \ 1316 ":%d->%d\n", (lev), IPSEC_LEVEL_REQUIRE) : (void)0), \ 1317 (lev) = IPSEC_LEVEL_REQUIRE, (lev) \ 1318 : (lev)) 1319 1320 /* set default level */ 1321 switch (((struct sockaddr *)&isr->sp->spidx.src)->sa_family) { 1322 #ifdef INET 1323 case AF_INET: 1324 esp_trans_deflev = IPSEC_CHECK_DEFAULT(ip4_esp_trans_deflev); 1325 esp_net_deflev = IPSEC_CHECK_DEFAULT(ip4_esp_net_deflev); 1326 ah_trans_deflev = IPSEC_CHECK_DEFAULT(ip4_ah_trans_deflev); 1327 ah_net_deflev = IPSEC_CHECK_DEFAULT(ip4_ah_net_deflev); 1328 break; 1329 #endif 1330 #ifdef INET6 1331 case AF_INET6: 1332 esp_trans_deflev = IPSEC_CHECK_DEFAULT(ip6_esp_trans_deflev); 1333 esp_net_deflev = IPSEC_CHECK_DEFAULT(ip6_esp_net_deflev); 1334 ah_trans_deflev = IPSEC_CHECK_DEFAULT(ip6_ah_trans_deflev); 1335 ah_net_deflev = IPSEC_CHECK_DEFAULT(ip6_ah_net_deflev); 1336 break; 1337 #endif 1338 default: 1339 panic("%s: unknown af %u", __func__, 1340 isr->sp->spidx.src.sa.sa_family); 1341 } 1342 1343 #undef IPSEC_CHECK_DEFAULT 1344 1345 /* set level */ 1346 switch (isr->level) { 1347 case IPSEC_LEVEL_DEFAULT: 1348 switch (isr->saidx.proto) { 1349 case IPPROTO_ESP: 1350 if (isr->saidx.mode == IPSEC_MODE_TUNNEL) 1351 level = esp_net_deflev; 1352 else 1353 level = esp_trans_deflev; 1354 break; 1355 case IPPROTO_AH: 1356 if (isr->saidx.mode == IPSEC_MODE_TUNNEL) 1357 level = ah_net_deflev; 1358 else 1359 level = ah_trans_deflev; 1360 break; 1361 case IPPROTO_IPCOMP: 1362 /* 1363 * we don't really care, as IPcomp document says that 1364 * we shouldn't compress small packets 1365 */ 1366 level = IPSEC_LEVEL_USE; 1367 break; 1368 default: 1369 panic("%s: Illegal protocol defined %u", __func__, 1370 isr->saidx.proto); 1371 } 1372 break; 1373 1374 case IPSEC_LEVEL_USE: 1375 case IPSEC_LEVEL_REQUIRE: 1376 level = isr->level; 1377 break; 1378 case IPSEC_LEVEL_UNIQUE: 1379 level = IPSEC_LEVEL_REQUIRE; 1380 break; 1381 1382 default: 1383 panic("%s: Illegal IPsec level %u", __func__, isr->level); 1384 } 1385 1386 return level; 1387 } 1388 1389 /* 1390 * Check security policy requirements against the actual packet contents. 1391 * 1392 * If the SP requires an IPsec packet, and the packet was neither AH nor ESP, 1393 * then kick it. 1394 */ 1395 static int 1396 ipsec_sp_reject(const struct secpolicy *sp, const struct mbuf *m) 1397 { 1398 struct ipsecrequest *isr; 1399 1400 if (KEYDEBUG_ON(KEYDEBUG_IPSEC_DATA)) { 1401 printf("%s: using SP\n", __func__); 1402 kdebug_secpolicy(sp); 1403 } 1404 1405 /* check policy */ 1406 switch (sp->policy) { 1407 case IPSEC_POLICY_DISCARD: 1408 return 1; 1409 case IPSEC_POLICY_BYPASS: 1410 case IPSEC_POLICY_NONE: 1411 return 0; 1412 } 1413 1414 KASSERTMSG(sp->policy == IPSEC_POLICY_IPSEC, 1415 "invalid policy %u", sp->policy); 1416 1417 /* XXX should compare policy against ipsec header history */ 1418 1419 for (isr = sp->req; isr != NULL; isr = isr->next) { 1420 if (ipsec_get_reqlevel(isr) != IPSEC_LEVEL_REQUIRE) 1421 continue; 1422 switch (isr->saidx.proto) { 1423 case IPPROTO_ESP: 1424 if ((m->m_flags & M_DECRYPTED) == 0) { 1425 KEYDEBUG_PRINTF(KEYDEBUG_IPSEC_DUMP, 1426 "ESP m_flags:%x\n", m->m_flags); 1427 return 1; 1428 } 1429 break; 1430 case IPPROTO_AH: 1431 if ((m->m_flags & M_AUTHIPHDR) == 0) { 1432 KEYDEBUG_PRINTF(KEYDEBUG_IPSEC_DUMP, 1433 "AH m_flags:%x\n", m->m_flags); 1434 return 1; 1435 } 1436 break; 1437 case IPPROTO_IPCOMP: 1438 /* 1439 * We don't really care, as IPcomp document 1440 * says that we shouldn't compress small 1441 * packets, IPComp policy should always be 1442 * treated as being in "use" level. 1443 */ 1444 break; 1445 } 1446 } 1447 1448 return 0; 1449 } 1450 1451 /* 1452 * Check security policy requirements. 1453 */ 1454 int 1455 ipsec_in_reject(struct mbuf *m, void *inp) 1456 { 1457 struct inpcb_hdr *inph = (struct inpcb_hdr *)inp; 1458 struct secpolicy *sp; 1459 int error; 1460 int result; 1461 1462 KASSERT(m != NULL); 1463 1464 if (inph == NULL) 1465 sp = ipsec_getpolicybyaddr(m, IPSEC_DIR_INBOUND, 1466 IP_FORWARDING, &error); 1467 else 1468 sp = ipsec_getpolicybysock(m, IPSEC_DIR_INBOUND, 1469 inph, &error); 1470 1471 if (sp != NULL) { 1472 result = ipsec_sp_reject(sp, m); 1473 if (result) 1474 IPSEC_STATINC(IPSEC_STAT_IN_POLVIO); 1475 KEY_SP_UNREF(&sp); 1476 } else { 1477 result = 0; 1478 } 1479 return result; 1480 } 1481 1482 /* 1483 * Compute the byte size to be occupied by the IPsec header. If it is 1484 * tunneled, it includes the size of outer IP header. 1485 */ 1486 static size_t 1487 ipsec_sp_hdrsiz(const struct secpolicy *sp, const struct mbuf *m) 1488 { 1489 struct ipsecrequest *isr; 1490 size_t siz; 1491 1492 if (KEYDEBUG_ON(KEYDEBUG_IPSEC_DATA)) { 1493 printf("%s: using SP\n", __func__); 1494 kdebug_secpolicy(sp); 1495 } 1496 1497 switch (sp->policy) { 1498 case IPSEC_POLICY_DISCARD: 1499 case IPSEC_POLICY_BYPASS: 1500 case IPSEC_POLICY_NONE: 1501 return 0; 1502 } 1503 1504 KASSERTMSG(sp->policy == IPSEC_POLICY_IPSEC, 1505 "invalid policy %u", sp->policy); 1506 1507 siz = 0; 1508 for (isr = sp->req; isr != NULL; isr = isr->next) { 1509 size_t clen = 0; 1510 struct secasvar *sav; 1511 1512 switch (isr->saidx.proto) { 1513 case IPPROTO_ESP: 1514 sav = ipsec_lookup_sa(isr, m); 1515 if (sav != NULL) { 1516 clen = esp_hdrsiz(sav); 1517 KEY_SA_UNREF(&sav); 1518 } else 1519 clen = esp_hdrsiz(NULL); 1520 break; 1521 case IPPROTO_AH: 1522 sav = ipsec_lookup_sa(isr, m); 1523 if (sav != NULL) { 1524 clen = ah_hdrsiz(sav); 1525 KEY_SA_UNREF(&sav); 1526 } else 1527 clen = ah_hdrsiz(NULL); 1528 break; 1529 case IPPROTO_IPCOMP: 1530 clen = sizeof(struct ipcomp); 1531 break; 1532 } 1533 1534 if (isr->saidx.mode == IPSEC_MODE_TUNNEL) { 1535 switch (isr->saidx.dst.sa.sa_family) { 1536 case AF_INET: 1537 clen += sizeof(struct ip); 1538 break; 1539 #ifdef INET6 1540 case AF_INET6: 1541 clen += sizeof(struct ip6_hdr); 1542 break; 1543 #endif 1544 default: 1545 IPSECLOG(LOG_ERR, "unknown AF %d in " 1546 "IPsec tunnel SA\n", 1547 ((const struct sockaddr *)&isr->saidx.dst) 1548 ->sa_family); 1549 break; 1550 } 1551 } 1552 siz += clen; 1553 } 1554 1555 return siz; 1556 } 1557 1558 size_t 1559 ipsec_hdrsiz(struct mbuf *m, u_int dir, void *inp) 1560 { 1561 struct inpcb_hdr *inph = (struct inpcb_hdr *)inp; 1562 struct secpolicy *sp; 1563 int error; 1564 size_t size; 1565 1566 KASSERT(m != NULL); 1567 KASSERTMSG(inph == NULL || inph->inph_socket != NULL, 1568 "socket w/o inpcb"); 1569 1570 if (inph == NULL) 1571 sp = ipsec_getpolicybyaddr(m, dir, IP_FORWARDING, &error); 1572 else 1573 sp = ipsec_getpolicybysock(m, dir, inph, &error); 1574 1575 if (sp != NULL) { 1576 size = ipsec_sp_hdrsiz(sp, m); 1577 KEYDEBUG_PRINTF(KEYDEBUG_IPSEC_DATA, "size:%zu.\n", size); 1578 KEY_SP_UNREF(&sp); 1579 } else { 1580 size = 0; 1581 } 1582 1583 return size; 1584 } 1585 1586 /* 1587 * Check the variable replay window. 1588 * ipsec_chkreplay() performs replay check before ICV verification. 1589 * ipsec_updatereplay() updates replay bitmap. This must be called after 1590 * ICV verification (it also performs replay check, which is usually done 1591 * beforehand). 1592 * 0 (zero) is returned if packet disallowed, 1 if packet permitted. 1593 * 1594 * based on RFC 2401. 1595 */ 1596 int 1597 ipsec_chkreplay(u_int32_t seq, const struct secasvar *sav) 1598 { 1599 const struct secreplay *replay; 1600 u_int32_t diff; 1601 int fr; 1602 u_int32_t wsizeb; /* constant: bits of window size */ 1603 int frlast; /* constant: last frame */ 1604 1605 KASSERT(sav != NULL); 1606 KASSERT(sav->replay != NULL); 1607 1608 replay = sav->replay; 1609 1610 if (replay->wsize == 0) 1611 return 1; /* no need to check replay. */ 1612 1613 /* constant */ 1614 frlast = replay->wsize - 1; 1615 wsizeb = replay->wsize << 3; 1616 1617 /* sequence number of 0 is invalid */ 1618 if (seq == 0) 1619 return 0; 1620 1621 /* first time is always okay */ 1622 if (replay->count == 0) 1623 return 1; 1624 1625 if (seq > replay->lastseq) { 1626 /* larger sequences are okay */ 1627 return 1; 1628 } else { 1629 /* seq is equal or less than lastseq. */ 1630 diff = replay->lastseq - seq; 1631 1632 /* over range to check, i.e. too old or wrapped */ 1633 if (diff >= wsizeb) 1634 return 0; 1635 1636 fr = frlast - diff / 8; 1637 1638 /* this packet already seen ? */ 1639 if ((replay->bitmap)[fr] & (1 << (diff % 8))) 1640 return 0; 1641 1642 /* out of order but good */ 1643 return 1; 1644 } 1645 } 1646 1647 /* 1648 * check replay counter whether to update or not. 1649 * OUT: 0: OK 1650 * 1: NG 1651 */ 1652 int 1653 ipsec_updatereplay(u_int32_t seq, const struct secasvar *sav) 1654 { 1655 struct secreplay *replay; 1656 u_int32_t diff; 1657 int fr; 1658 u_int32_t wsizeb; /* constant: bits of window size */ 1659 int frlast; /* constant: last frame */ 1660 1661 KASSERT(sav != NULL); 1662 KASSERT(sav->replay != NULL); 1663 1664 replay = sav->replay; 1665 1666 if (replay->wsize == 0) 1667 goto ok; /* no need to check replay. */ 1668 1669 /* constant */ 1670 frlast = replay->wsize - 1; 1671 wsizeb = replay->wsize << 3; 1672 1673 /* sequence number of 0 is invalid */ 1674 if (seq == 0) 1675 return 1; 1676 1677 /* first time */ 1678 if (replay->count == 0) { 1679 replay->lastseq = seq; 1680 memset(replay->bitmap, 0, replay->wsize); 1681 (replay->bitmap)[frlast] = 1; 1682 goto ok; 1683 } 1684 1685 if (seq > replay->lastseq) { 1686 /* seq is larger than lastseq. */ 1687 diff = seq - replay->lastseq; 1688 1689 /* new larger sequence number */ 1690 if (diff < wsizeb) { 1691 /* In window */ 1692 /* set bit for this packet */ 1693 vshiftl(replay->bitmap, diff, replay->wsize); 1694 (replay->bitmap)[frlast] |= 1; 1695 } else { 1696 /* this packet has a "way larger" */ 1697 memset(replay->bitmap, 0, replay->wsize); 1698 (replay->bitmap)[frlast] = 1; 1699 } 1700 replay->lastseq = seq; 1701 1702 /* larger is good */ 1703 } else { 1704 /* seq is equal or less than lastseq. */ 1705 diff = replay->lastseq - seq; 1706 1707 /* over range to check, i.e. too old or wrapped */ 1708 if (diff >= wsizeb) 1709 return 1; 1710 1711 fr = frlast - diff / 8; 1712 1713 /* this packet already seen ? */ 1714 if ((replay->bitmap)[fr] & (1 << (diff % 8))) 1715 return 1; 1716 1717 /* mark as seen */ 1718 (replay->bitmap)[fr] |= (1 << (diff % 8)); 1719 1720 /* out of order but good */ 1721 } 1722 1723 ok: 1724 if (replay->count == ~0) { 1725 char buf[IPSEC_LOGSASTRLEN]; 1726 1727 /* set overflow flag */ 1728 replay->overflow++; 1729 1730 /* don't increment, no more packets accepted */ 1731 if ((sav->flags & SADB_X_EXT_CYCSEQ) == 0) 1732 return 1; 1733 1734 IPSECLOG(LOG_WARNING, "replay counter made %d cycle. %s\n", 1735 replay->overflow, ipsec_logsastr(sav, buf, sizeof(buf))); 1736 } 1737 1738 replay->count++; 1739 1740 return 0; 1741 } 1742 1743 /* 1744 * shift variable length buffer to left. 1745 * IN: bitmap: pointer to the buffer 1746 * nbit: the number of to shift. 1747 * wsize: buffer size (bytes). 1748 */ 1749 static void 1750 vshiftl(unsigned char *bitmap, int nbit, int wsize) 1751 { 1752 int s, j, i; 1753 unsigned char over; 1754 1755 for (j = 0; j < nbit; j += 8) { 1756 s = (nbit - j < 8) ? (nbit - j): 8; 1757 bitmap[0] <<= s; 1758 for (i = 1; i < wsize; i++) { 1759 over = (bitmap[i] >> (8 - s)); 1760 bitmap[i] <<= s; 1761 bitmap[i-1] |= over; 1762 } 1763 } 1764 1765 return; 1766 } 1767 1768 /* Return a printable string for the address. */ 1769 const char * 1770 ipsec_address(const union sockaddr_union *sa, char *buf, size_t size) 1771 { 1772 switch (sa->sa.sa_family) { 1773 case AF_INET: 1774 in_print(buf, size, &sa->sin.sin_addr); 1775 return buf; 1776 #if INET6 1777 case AF_INET6: 1778 in6_print(buf, size, &sa->sin6.sin6_addr); 1779 return buf; 1780 #endif 1781 default: 1782 return "(unknown address family)"; 1783 } 1784 } 1785 1786 const char * 1787 ipsec_logsastr(const struct secasvar *sav, char *buf, size_t size) 1788 { 1789 const struct secasindex *saidx = &sav->sah->saidx; 1790 char sbuf[IPSEC_ADDRSTRLEN], dbuf[IPSEC_ADDRSTRLEN]; 1791 1792 KASSERTMSG(saidx->src.sa.sa_family == saidx->dst.sa.sa_family, 1793 "af family mismatch, src %u, dst %u", 1794 saidx->src.sa.sa_family, saidx->dst.sa.sa_family); 1795 1796 snprintf(buf, size, "SA(SPI=%u src=%s dst=%s)", 1797 (u_int32_t)ntohl(sav->spi), 1798 ipsec_address(&saidx->src, sbuf, sizeof(sbuf)), 1799 ipsec_address(&saidx->dst, dbuf, sizeof(dbuf))); 1800 1801 return buf; 1802 } 1803 1804 #ifdef INET6 1805 struct secpolicy * 1806 ipsec6_check_policy(struct mbuf *m, struct in6pcb *in6p, int flags, 1807 int *needipsecp, int *errorp) 1808 { 1809 struct secpolicy *sp = NULL; 1810 int s; 1811 int error = 0; 1812 int needipsec = 0; 1813 1814 if (ipsec_outdone(m)) { 1815 goto skippolicycheck; 1816 } 1817 s = splsoftnet(); 1818 if (in6p && ipsec_pcb_skip_ipsec(in6p->in6p_sp, IPSEC_DIR_OUTBOUND)) { 1819 splx(s); 1820 goto skippolicycheck; 1821 } 1822 sp = ipsec_checkpolicy(m, IPSEC_DIR_OUTBOUND, flags, &error, in6p); 1823 splx(s); 1824 1825 /* 1826 * There are four return cases: 1827 * sp != NULL apply IPsec policy 1828 * sp == NULL, error == 0 no IPsec handling needed 1829 * sp == NULL, error == -EINVAL discard packet w/o error 1830 * sp == NULL, error != 0 discard packet, report error 1831 */ 1832 if (sp == NULL) { 1833 needipsec = 0; 1834 } else { 1835 needipsec = 1; 1836 } 1837 1838 skippolicycheck: 1839 *errorp = error; 1840 *needipsecp = needipsec; 1841 return sp; 1842 } 1843 1844 /* 1845 * calculate UDP checksum for UDP encapsulated ESP for IPv6. 1846 * 1847 * RFC2460(Internet Protocol, Version 6 Specification) says: 1848 * 1849 * IPv6 receivers MUST discard UDP packets with a zero checksum. 1850 * 1851 * There is more relaxed speficication RFC6935(IPv6 and UDP Checksums for 1852 * Tunneled Packets). The document allows zero checksum. It's too 1853 * late to publish, there are a lot of interoperability problems... 1854 */ 1855 void 1856 ipsec6_udp_cksum(struct mbuf *m) 1857 { 1858 struct ip6_hdr *ip6; 1859 uint16_t plen, uh_sum; 1860 int off; 1861 1862 /* must called after m_pullup() */ 1863 KASSERT(m->m_len >= sizeof(struct ip6_hdr)); 1864 1865 ip6 = mtod(m, struct ip6_hdr *); 1866 KASSERT(ip6->ip6_nxt == IPPROTO_UDP); 1867 1868 /* ip6->ip6_plen can not be updated before ip6_output() */ 1869 plen = m->m_pkthdr.len - sizeof(*ip6); 1870 KASSERT(plen >= sizeof(struct udphdr)); 1871 1872 uh_sum = in6_cksum(m, IPPROTO_UDP, sizeof(*ip6), plen); 1873 if (uh_sum == 0) 1874 uh_sum = 0xffff; 1875 1876 off = sizeof(*ip6) + offsetof(struct udphdr, uh_sum); 1877 m_copyback(m, off, sizeof(uh_sum), (void *)&uh_sum); 1878 } 1879 #endif /* INET6 */ 1880 1881 /* 1882 * ----------------------------------------------------------------------------- 1883 */ 1884 1885 /* XXX this stuff doesn't belong here... */ 1886 1887 static struct xformsw *xforms = NULL; 1888 1889 /* 1890 * Register a transform; typically at system startup. 1891 */ 1892 void 1893 xform_register(struct xformsw *xsp) 1894 { 1895 xsp->xf_next = xforms; 1896 xforms = xsp; 1897 } 1898 1899 /* 1900 * Initialize transform support in an sav. 1901 */ 1902 int 1903 xform_init(struct secasvar *sav, int xftype) 1904 { 1905 struct xformsw *xsp; 1906 1907 if (sav->tdb_xform != NULL) /* previously initialized */ 1908 return 0; 1909 for (xsp = xforms; xsp; xsp = xsp->xf_next) 1910 if (xsp->xf_type == xftype) 1911 return (*xsp->xf_init)(sav, xsp); 1912 1913 IPSECLOG(LOG_DEBUG, "no match for xform type %d\n", xftype); 1914 return EINVAL; 1915 } 1916 1917 /* 1918 * XXXJRT This should be done as a protosw init call. 1919 */ 1920 void 1921 ipsec_attach(void) 1922 { 1923 1924 ipsec_output_init(); 1925 1926 ipsecstat_percpu = percpu_alloc(sizeof(uint64_t) * IPSEC_NSTATS); 1927 1928 sysctl_net_inet_ipsec_setup(NULL); 1929 #ifdef INET6 1930 sysctl_net_inet6_ipsec6_setup(NULL); 1931 #endif 1932 1933 ah_attach(); 1934 esp_attach(); 1935 ipcomp_attach(); 1936 ipe4_attach(); 1937 #ifdef TCP_SIGNATURE 1938 tcpsignature_attach(); 1939 #endif 1940 } 1941