1/*- 2 * SPDX-License-Identifier: BSD-2-Clause 3 * 4 * Copyright (c) 2002, 2003 Sam Leffler, Errno Consulting 5 * Copyright (c) 2016 Andrey V. Elsukov <ae@FreeBSD.org> 6 * All rights reserved. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 */ 29 30/* 31 * IPsec output processing. 32 */ 33#include "opt_inet.h" 34#include "opt_inet6.h" 35#include "opt_ipsec.h" 36#include "opt_sctp.h" 37 38#include <sys/param.h> 39#include <sys/systm.h> 40#include <sys/mbuf.h> 41#include <sys/domain.h> 42#include <sys/protosw.h> 43#include <sys/socket.h> 44#include <sys/errno.h> 45#include <sys/hhook.h> 46#include <sys/syslog.h> 47 48#include <net/if.h> 49#include <net/if_enc.h> 50#include <net/if_var.h> 51#include <net/vnet.h> 52 53#include <netinet/in.h> 54#include <netinet/in_pcb.h> 55#include <netinet/in_systm.h> 56#include <netinet/ip.h> 57#include <netinet/ip_var.h> 58#include <netinet/in_var.h> 59#include <netinet/ip_ecn.h> 60#ifdef INET6 61#include <netinet6/ip6_ecn.h> 62#endif 63#include <netinet/ip_icmp.h> 64#include <netinet/tcp_var.h> 65 66#include <netinet/ip6.h> 67#ifdef INET6 68#include <netinet6/ip6_var.h> 69#include <netinet6/scope6_var.h> 70#endif 71#include <netinet/in_pcb.h> 72#ifdef INET6 73#include <netinet/icmp6.h> 74#endif 75#if defined(SCTP) || defined(SCTP_SUPPORT) 76#include <netinet/sctp_crc32.h> 77#endif 78 79#include <netinet/udp.h> 80#include <netipsec/ah.h> 81#include <netipsec/esp.h> 82#include <netipsec/ipsec.h> 83#ifdef INET6 84#include <netipsec/ipsec6.h> 85#endif 86#include <netipsec/ipsec_support.h> 87#include <netipsec/ah_var.h> 88#include <netipsec/esp_var.h> 89#include <netipsec/ipcomp_var.h> 90 91#include <netipsec/xform.h> 92 93#include <netipsec/key.h> 94#include <netipsec/keydb.h> 95#include <netipsec/key_debug.h> 96 97#include <machine/in_cksum.h> 98 99#define IPSEC_OSTAT_INC(proto, name) do { \ 100 if ((proto) == IPPROTO_ESP) \ 101 ESPSTAT_INC(esps_##name); \ 102 else if ((proto) == IPPROTO_AH)\ 103 AHSTAT_INC(ahs_##name); \ 104 else \ 105 IPCOMPSTAT_INC(ipcomps_##name); \ 106} while (0) 107 108static int ipsec_encap(struct mbuf **mp, struct secasindex *saidx); 109static size_t ipsec_get_pmtu(struct secasvar *sav); 110 111#ifdef INET 112static struct secasvar * 113ipsec4_allocsa(struct mbuf *m, struct secpolicy *sp, u_int *pidx, int *error) 114{ 115 struct secasindex *saidx, tmpsaidx; 116 struct ipsecrequest *isr; 117 struct sockaddr_in *sin; 118 struct secasvar *sav; 119 struct ip *ip; 120 121 /* 122 * Check system global policy controls. 123 */ 124next: 125 isr = sp->req[*pidx]; 126 if ((isr->saidx.proto == IPPROTO_ESP && !V_esp_enable) || 127 (isr->saidx.proto == IPPROTO_AH && !V_ah_enable) || 128 (isr->saidx.proto == IPPROTO_IPCOMP && !V_ipcomp_enable)) { 129 DPRINTF(("%s: IPsec outbound packet dropped due" 130 " to policy (check your sysctls)\n", __func__)); 131 IPSEC_OSTAT_INC(isr->saidx.proto, pdrops); 132 *error = EHOSTUNREACH; 133 return (NULL); 134 } 135 /* 136 * Craft SA index to search for proper SA. Note that 137 * we only initialize unspecified SA peers for transport 138 * mode; for tunnel mode they must already be filled in. 139 */ 140 if (isr->saidx.mode == IPSEC_MODE_TRANSPORT) { 141 saidx = &tmpsaidx; 142 *saidx = isr->saidx; 143 ip = mtod(m, struct ip *); 144 if (saidx->src.sa.sa_len == 0) { 145 sin = &saidx->src.sin; 146 sin->sin_len = sizeof(*sin); 147 sin->sin_family = AF_INET; 148 sin->sin_port = IPSEC_PORT_ANY; 149 sin->sin_addr = ip->ip_src; 150 } 151 if (saidx->dst.sa.sa_len == 0) { 152 sin = &saidx->dst.sin; 153 sin->sin_len = sizeof(*sin); 154 sin->sin_family = AF_INET; 155 sin->sin_port = IPSEC_PORT_ANY; 156 sin->sin_addr = ip->ip_dst; 157 } 158 } else 159 saidx = &sp->req[*pidx]->saidx; 160 /* 161 * Lookup SA and validate it. 162 */ 163 sav = key_allocsa_policy(sp, saidx, error); 164 if (sav == NULL) { 165 IPSECSTAT_INC(ips_out_nosa); 166 if (*error != 0) 167 return (NULL); 168 if (ipsec_get_reqlevel(sp, *pidx) != IPSEC_LEVEL_REQUIRE) { 169 /* 170 * We have no SA and policy that doesn't require 171 * this IPsec transform, thus we can continue w/o 172 * IPsec processing, i.e. return EJUSTRETURN. 173 * But first check if there is some bundled transform. 174 */ 175 if (sp->tcount > ++(*pidx)) 176 goto next; 177 *error = EJUSTRETURN; 178 } 179 return (NULL); 180 } 181 IPSEC_ASSERT(sav->tdb_xform != NULL, ("SA with NULL tdb_xform")); 182 return (sav); 183} 184 185/* 186 * IPsec output logic for IPv4. 187 */ 188static int 189ipsec4_perform_request(struct mbuf *m, struct secpolicy *sp, 190 struct inpcb *inp, u_int idx) 191{ 192 struct ipsec_ctx_data ctx; 193 union sockaddr_union *dst; 194 struct secasvar *sav; 195 struct ip *ip; 196 int error, i, off; 197 198 IPSEC_ASSERT(idx < sp->tcount, ("Wrong IPsec request index %d", idx)); 199 200 /* 201 * We hold the reference to SP. Content of SP couldn't be changed. 202 * Craft secasindex and do lookup for suitable SA. 203 * Then do encapsulation if needed and call xform's output. 204 * We need to store SP in the xform callback parameters. 205 * In xform callback we will extract SP and it can be used to 206 * determine next transform. At the end of transform we can 207 * release reference to SP. 208 */ 209 sav = ipsec4_allocsa(m, sp, &idx, &error); 210 if (sav == NULL) { 211 if (error == EJUSTRETURN) { /* No IPsec required */ 212 key_freesp(&sp); 213 return (error); 214 } 215 goto bad; 216 } 217 /* 218 * XXXAE: most likely ip_sum at this point is wrong. 219 */ 220 IPSEC_INIT_CTX(&ctx, &m, inp, sav, AF_INET, IPSEC_ENC_BEFORE); 221 if ((error = ipsec_run_hhooks(&ctx, HHOOK_TYPE_IPSEC_OUT)) != 0) 222 goto bad; 223 224 ip = mtod(m, struct ip *); 225 dst = &sav->sah->saidx.dst; 226 /* Do the appropriate encapsulation, if necessary */ 227 if (sp->req[idx]->saidx.mode == IPSEC_MODE_TUNNEL || /* Tunnel requ'd */ 228 dst->sa.sa_family != AF_INET || /* PF mismatch */ 229 (dst->sa.sa_family == AF_INET && /* Proxy */ 230 dst->sin.sin_addr.s_addr != INADDR_ANY && 231 dst->sin.sin_addr.s_addr != ip->ip_dst.s_addr)) { 232 /* Fix IPv4 header checksum and length */ 233 ip->ip_len = htons(m->m_pkthdr.len); 234 ip->ip_sum = 0; 235 ip->ip_sum = in_cksum(m, ip->ip_hl << 2); 236 error = ipsec_encap(&m, &sav->sah->saidx); 237 if (error != 0) { 238 DPRINTF(("%s: encapsulation for SPI 0x%08x failed " 239 "with error %d\n", __func__, ntohl(sav->spi), 240 error)); 241 /* XXXAE: IPSEC_OSTAT_INC(tunnel); */ 242 goto bad; 243 } 244 inp = NULL; 245 } 246 247 IPSEC_INIT_CTX(&ctx, &m, inp, sav, dst->sa.sa_family, IPSEC_ENC_AFTER); 248 if ((error = ipsec_run_hhooks(&ctx, HHOOK_TYPE_IPSEC_OUT)) != 0) 249 goto bad; 250 251 /* 252 * Dispatch to the appropriate IPsec transform logic. The 253 * packet will be returned for transmission after crypto 254 * processing, etc. are completed. 255 * 256 * NB: m & sav are ``passed to caller'' who's responsible for 257 * reclaiming their resources. 258 */ 259 switch(dst->sa.sa_family) { 260 case AF_INET: 261 ip = mtod(m, struct ip *); 262 i = ip->ip_hl << 2; 263 off = offsetof(struct ip, ip_p); 264 break; 265#ifdef INET6 266 case AF_INET6: 267 i = sizeof(struct ip6_hdr); 268 off = offsetof(struct ip6_hdr, ip6_nxt); 269 break; 270#endif /* INET6 */ 271 default: 272 DPRINTF(("%s: unsupported protocol family %u\n", 273 __func__, dst->sa.sa_family)); 274 error = EPFNOSUPPORT; 275 IPSEC_OSTAT_INC(sav->sah->saidx.proto, nopf); 276 goto bad; 277 } 278 error = (*sav->tdb_xform->xf_output)(m, sp, sav, idx, i, off); 279 return (error); 280bad: 281 IPSECSTAT_INC(ips_out_inval); 282 if (m != NULL) 283 m_freem(m); 284 if (sav != NULL) 285 key_freesav(&sav); 286 key_freesp(&sp); 287 return (error); 288} 289 290int 291ipsec4_process_packet(struct mbuf *m, struct secpolicy *sp, 292 struct inpcb *inp) 293{ 294 295 return (ipsec4_perform_request(m, sp, inp, 0)); 296} 297 298int 299ipsec4_check_pmtu(struct mbuf *m, struct secpolicy *sp, int forwarding) 300{ 301 struct secasvar *sav; 302 struct ip *ip; 303 size_t hlen, pmtu; 304 uint32_t idx; 305 int error; 306 307 /* Don't check PMTU if the frame won't have DF bit set. */ 308 if (!V_ip4_ipsec_dfbit) 309 return (0); 310 if (V_ip4_ipsec_dfbit == 1) 311 goto setdf; 312 313 /* V_ip4_ipsec_dfbit > 1 - we will copy it from inner header. */ 314 ip = mtod(m, struct ip *); 315 if (!(ip->ip_off & htons(IP_DF))) 316 return (0); 317 318setdf: 319 idx = sp->tcount - 1; 320 sav = ipsec4_allocsa(m, sp, &idx, &error); 321 if (sav == NULL) { 322 key_freesp(&sp); 323 /* 324 * No matching SA was found and SADB_ACQUIRE message was generated. 325 * Since we have matched a SP to this packet drop it silently. 326 */ 327 if (error == 0) 328 error = EINPROGRESS; 329 if (error != EJUSTRETURN) 330 m_freem(m); 331 332 return (error); 333 } 334 335 pmtu = ipsec_get_pmtu(sav); 336 if (pmtu == 0) { 337 key_freesav(&sav); 338 return (0); 339 } 340 341 hlen = ipsec_hdrsiz_internal(sp); 342 key_freesav(&sav); 343 344 if (m_length(m, NULL) + hlen > pmtu) { 345 /* 346 * If we're forwarding generate ICMP message here, 347 * so that it contains pmtu subtracted by header size. 348 * Set error to EINPROGRESS, in order for the frame 349 * to be dropped silently. 350 */ 351 if (forwarding) { 352 if (pmtu > hlen) 353 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 354 0, pmtu - hlen); 355 else 356 m_freem(m); 357 358 key_freesp(&sp); 359 return (EINPROGRESS); /* Pretend that we consumed it. */ 360 } else { 361 m_freem(m); 362 key_freesp(&sp); 363 return (EMSGSIZE); 364 } 365 } 366 367 return (0); 368} 369 370static int 371ipsec4_common_output(struct mbuf *m, struct inpcb *inp, int forwarding) 372{ 373 struct secpolicy *sp; 374 int error; 375 376 /* Lookup for the corresponding outbound security policy */ 377 sp = ipsec4_checkpolicy(m, inp, &error, !forwarding); 378 if (sp == NULL) { 379 if (error == -EINVAL) { 380 /* Discarded by policy. */ 381 m_freem(m); 382 return (EACCES); 383 } 384 return (0); /* No IPsec required. */ 385 } 386 387 /* 388 * Usually we have to have tunnel mode IPsec security policy 389 * when we are forwarding a packet. Otherwise we could not handle 390 * encrypted replies, because they are not destined for us. But 391 * some users are doing source address translation for forwarded 392 * packets, and thus, even if they are forwarded, the replies will 393 * return back to us. 394 */ 395 if (!forwarding) { 396 /* 397 * Do delayed checksums now because we send before 398 * this is done in the normal processing path. 399 */ 400 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 401 in_delayed_cksum(m); 402 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 403 } 404#if defined(SCTP) || defined(SCTP_SUPPORT) 405 if (m->m_pkthdr.csum_flags & CSUM_SCTP) { 406 struct ip *ip; 407 408 ip = mtod(m, struct ip *); 409 sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2)); 410 m->m_pkthdr.csum_flags &= ~CSUM_SCTP; 411 } 412#endif 413 } 414 /* NB: callee frees mbuf and releases reference to SP */ 415 error = ipsec4_check_pmtu(m, sp, forwarding); 416 if (error != 0) { 417 if (error == EJUSTRETURN) 418 return (0); 419 420 return (error); 421 } 422 423 error = ipsec4_process_packet(m, sp, inp); 424 if (error == EJUSTRETURN) { 425 /* 426 * We had a SP with a level of 'use' and no SA. We 427 * will just continue to process the packet without 428 * IPsec processing and return without error. 429 */ 430 return (0); 431 } 432 if (error == 0) 433 return (EINPROGRESS); /* consumed by IPsec */ 434 return (error); 435} 436 437/* 438 * IPSEC_OUTPUT() method implementation for IPv4. 439 * 0 - no IPsec handling needed 440 * other values - mbuf consumed by IPsec. 441 */ 442int 443ipsec4_output(struct mbuf *m, struct inpcb *inp) 444{ 445 446 /* 447 * If the packet is resubmitted to ip_output (e.g. after 448 * AH, ESP, etc. processing), there will be a tag to bypass 449 * the lookup and related policy checking. 450 */ 451 if (m_tag_find(m, PACKET_TAG_IPSEC_OUT_DONE, NULL) != NULL) 452 return (0); 453 454 return (ipsec4_common_output(m, inp, 0)); 455} 456 457/* 458 * IPSEC_FORWARD() method implementation for IPv4. 459 * 0 - no IPsec handling needed 460 * other values - mbuf consumed by IPsec. 461 */ 462int 463ipsec4_forward(struct mbuf *m) 464{ 465 466 /* 467 * Check if this packet has an active inbound SP and needs to be 468 * dropped instead of forwarded. 469 */ 470 if (ipsec4_in_reject(m, NULL) != 0) { 471 m_freem(m); 472 return (EACCES); 473 } 474 return (ipsec4_common_output(m, NULL, 1)); 475} 476#endif 477 478#ifdef INET6 479static int 480in6_sa_equal_addrwithscope(const struct sockaddr_in6 *sa, 481 const struct in6_addr *ia) 482{ 483 struct in6_addr ia2; 484 485 if (IN6_IS_SCOPE_LINKLOCAL(&sa->sin6_addr)) { 486 memcpy(&ia2, &sa->sin6_addr, sizeof(ia2)); 487 ia2.s6_addr16[1] = htons(sa->sin6_scope_id); 488 return (IN6_ARE_ADDR_EQUAL(ia, &ia2)); 489 } 490 return (IN6_ARE_ADDR_EQUAL(&sa->sin6_addr, ia)); 491} 492 493static struct secasvar * 494ipsec6_allocsa(struct mbuf *m, struct secpolicy *sp, u_int *pidx, int *error) 495{ 496 struct secasindex *saidx, tmpsaidx; 497 struct ipsecrequest *isr; 498 struct sockaddr_in6 *sin6; 499 struct secasvar *sav; 500 struct ip6_hdr *ip6; 501 502 /* 503 * Check system global policy controls. 504 */ 505next: 506 isr = sp->req[*pidx]; 507 if ((isr->saidx.proto == IPPROTO_ESP && !V_esp_enable) || 508 (isr->saidx.proto == IPPROTO_AH && !V_ah_enable) || 509 (isr->saidx.proto == IPPROTO_IPCOMP && !V_ipcomp_enable)) { 510 DPRINTF(("%s: IPsec outbound packet dropped due" 511 " to policy (check your sysctls)\n", __func__)); 512 IPSEC_OSTAT_INC(isr->saidx.proto, pdrops); 513 *error = EHOSTUNREACH; 514 return (NULL); 515 } 516 /* 517 * Craft SA index to search for proper SA. Note that 518 * we only fillin unspecified SA peers for transport 519 * mode; for tunnel mode they must already be filled in. 520 */ 521 if (isr->saidx.mode == IPSEC_MODE_TRANSPORT) { 522 saidx = &tmpsaidx; 523 *saidx = isr->saidx; 524 ip6 = mtod(m, struct ip6_hdr *); 525 if (saidx->src.sin6.sin6_len == 0) { 526 sin6 = (struct sockaddr_in6 *)&saidx->src; 527 sin6->sin6_len = sizeof(*sin6); 528 sin6->sin6_family = AF_INET6; 529 sin6->sin6_port = IPSEC_PORT_ANY; 530 sin6->sin6_addr = ip6->ip6_src; 531 if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) { 532 /* fix scope id for comparing SPD */ 533 sin6->sin6_addr.s6_addr16[1] = 0; 534 sin6->sin6_scope_id = 535 ntohs(ip6->ip6_src.s6_addr16[1]); 536 } 537 } 538 if (saidx->dst.sin6.sin6_len == 0) { 539 sin6 = (struct sockaddr_in6 *)&saidx->dst; 540 sin6->sin6_len = sizeof(*sin6); 541 sin6->sin6_family = AF_INET6; 542 sin6->sin6_port = IPSEC_PORT_ANY; 543 sin6->sin6_addr = ip6->ip6_dst; 544 if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_dst)) { 545 /* fix scope id for comparing SPD */ 546 sin6->sin6_addr.s6_addr16[1] = 0; 547 sin6->sin6_scope_id = 548 ntohs(ip6->ip6_dst.s6_addr16[1]); 549 } 550 } 551 } else 552 saidx = &sp->req[*pidx]->saidx; 553 /* 554 * Lookup SA and validate it. 555 */ 556 sav = key_allocsa_policy(sp, saidx, error); 557 if (sav == NULL) { 558 IPSEC6STAT_INC(ips_out_nosa); 559 if (*error != 0) 560 return (NULL); 561 if (ipsec_get_reqlevel(sp, *pidx) != IPSEC_LEVEL_REQUIRE) { 562 /* 563 * We have no SA and policy that doesn't require 564 * this IPsec transform, thus we can continue w/o 565 * IPsec processing, i.e. return EJUSTRETURN. 566 * But first check if there is some bundled transform. 567 */ 568 if (sp->tcount > ++(*pidx)) 569 goto next; 570 *error = EJUSTRETURN; 571 } 572 return (NULL); 573 } 574 IPSEC_ASSERT(sav->tdb_xform != NULL, ("SA with NULL tdb_xform")); 575 return (sav); 576} 577 578/* 579 * IPsec output logic for IPv6. 580 */ 581static int 582ipsec6_perform_request(struct mbuf *m, struct secpolicy *sp, 583 struct inpcb *inp, u_int idx) 584{ 585 struct ipsec_ctx_data ctx; 586 union sockaddr_union *dst; 587 struct secasvar *sav; 588 struct ip6_hdr *ip6; 589 int error, i, off; 590 591 IPSEC_ASSERT(idx < sp->tcount, ("Wrong IPsec request index %d", idx)); 592 593 sav = ipsec6_allocsa(m, sp, &idx, &error); 594 if (sav == NULL) { 595 if (error == EJUSTRETURN) { /* No IPsec required */ 596 key_freesp(&sp); 597 return (error); 598 } 599 goto bad; 600 } 601 602 /* Fix IP length in case if it is not set yet. */ 603 ip6 = mtod(m, struct ip6_hdr *); 604 ip6->ip6_plen = htons(m->m_pkthdr.len - sizeof(*ip6)); 605 606 IPSEC_INIT_CTX(&ctx, &m, inp, sav, AF_INET6, IPSEC_ENC_BEFORE); 607 if ((error = ipsec_run_hhooks(&ctx, HHOOK_TYPE_IPSEC_OUT)) != 0) 608 goto bad; 609 610 ip6 = mtod(m, struct ip6_hdr *); /* pfil can change mbuf */ 611 dst = &sav->sah->saidx.dst; 612 613 /* Do the appropriate encapsulation, if necessary */ 614 if (sp->req[idx]->saidx.mode == IPSEC_MODE_TUNNEL || /* Tunnel requ'd */ 615 dst->sa.sa_family != AF_INET6 || /* PF mismatch */ 616 ((dst->sa.sa_family == AF_INET6) && 617 (!IN6_IS_ADDR_UNSPECIFIED(&dst->sin6.sin6_addr)) && 618 (!in6_sa_equal_addrwithscope(&dst->sin6, &ip6->ip6_dst)))) { 619 if (m->m_pkthdr.len - sizeof(*ip6) > IPV6_MAXPACKET) { 620 /* No jumbogram support. */ 621 error = ENXIO; /*XXX*/ 622 goto bad; 623 } 624 error = ipsec_encap(&m, &sav->sah->saidx); 625 if (error != 0) { 626 DPRINTF(("%s: encapsulation for SPI 0x%08x failed " 627 "with error %d\n", __func__, ntohl(sav->spi), 628 error)); 629 /* XXXAE: IPSEC_OSTAT_INC(tunnel); */ 630 goto bad; 631 } 632 inp = NULL; 633 } 634 635 IPSEC_INIT_CTX(&ctx, &m, inp, sav, dst->sa.sa_family, IPSEC_ENC_AFTER); 636 if ((error = ipsec_run_hhooks(&ctx, HHOOK_TYPE_IPSEC_OUT)) != 0) 637 goto bad; 638 639 switch(dst->sa.sa_family) { 640#ifdef INET 641 case AF_INET: 642 { 643 struct ip *ip; 644 ip = mtod(m, struct ip *); 645 i = ip->ip_hl << 2; 646 off = offsetof(struct ip, ip_p); 647 } 648 break; 649#endif /* AF_INET */ 650 case AF_INET6: 651 i = sizeof(struct ip6_hdr); 652 off = offsetof(struct ip6_hdr, ip6_nxt); 653 break; 654 default: 655 DPRINTF(("%s: unsupported protocol family %u\n", 656 __func__, dst->sa.sa_family)); 657 error = EPFNOSUPPORT; 658 IPSEC_OSTAT_INC(sav->sah->saidx.proto, nopf); 659 goto bad; 660 } 661 error = (*sav->tdb_xform->xf_output)(m, sp, sav, idx, i, off); 662 return (error); 663bad: 664 IPSEC6STAT_INC(ips_out_inval); 665 if (m != NULL) 666 m_freem(m); 667 if (sav != NULL) 668 key_freesav(&sav); 669 key_freesp(&sp); 670 return (error); 671} 672 673int 674ipsec6_process_packet(struct mbuf *m, struct secpolicy *sp, 675 struct inpcb *inp) 676{ 677 678 return (ipsec6_perform_request(m, sp, inp, 0)); 679} 680 681/* 682 * IPv6 implementation is based on IPv4 implementation. 683 */ 684int 685ipsec6_check_pmtu(struct mbuf *m, struct secpolicy *sp, int forwarding) 686{ 687 struct secasvar *sav; 688 size_t hlen, pmtu; 689 uint32_t idx; 690 int error; 691 692 /* 693 * According to RFC8200 L3 fragmentation is supposed to be done only on 694 * locally generated packets. During L3 forwarding packets that are too 695 * big are always supposed to be dropped, with an ICMPv6 packet being 696 * sent back. 697 */ 698 if (!forwarding) 699 return (0); 700 701 idx = sp->tcount - 1; 702 sav = ipsec6_allocsa(m, sp, &idx, &error); 703 if (sav == NULL) { 704 key_freesp(&sp); 705 /* 706 * No matching SA was found and SADB_ACQUIRE message was generated. 707 * Since we have matched a SP to this packet drop it silently. 708 */ 709 if (error == 0) 710 error = EINPROGRESS; 711 if (error != EJUSTRETURN) 712 m_freem(m); 713 714 return (error); 715 } 716 717 pmtu = ipsec_get_pmtu(sav); 718 if (pmtu == 0) { 719 key_freesav(&sav); 720 return (0); 721 } 722 723 hlen = ipsec_hdrsiz_internal(sp); 724 key_freesav(&sav); 725 726 if (m_length(m, NULL) + hlen > pmtu) { 727 /* 728 * If we're forwarding generate ICMPv6 message here, 729 * so that it contains pmtu subtracted by header size. 730 * Set error to EINPROGRESS, in order for the frame 731 * to be dropped silently. 732 */ 733 if (forwarding) { 734 if (pmtu > hlen) 735 icmp6_error(m, ICMP6_PACKET_TOO_BIG, 0, pmtu - hlen); 736 else 737 m_freem(m); 738 739 key_freesp(&sp); 740 return (EINPROGRESS); /* Pretend that we consumed it. */ 741 } 742 } 743 744 return (0); 745} 746 747static int 748ipsec6_common_output(struct mbuf *m, struct inpcb *inp, int forwarding) 749{ 750 struct secpolicy *sp; 751 int error; 752 753 /* Lookup for the corresponding outbound security policy */ 754 sp = ipsec6_checkpolicy(m, inp, &error, !forwarding); 755 if (sp == NULL) { 756 if (error == -EINVAL) { 757 /* Discarded by policy. */ 758 m_freem(m); 759 return (EACCES); 760 } 761 return (0); /* No IPsec required. */ 762 } 763 764 if (!forwarding) { 765 /* 766 * Do delayed checksums now because we send before 767 * this is done in the normal processing path. 768 */ 769 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6) { 770 in6_delayed_cksum(m, m->m_pkthdr.len - 771 sizeof(struct ip6_hdr), sizeof(struct ip6_hdr)); 772 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6; 773 } 774#if defined(SCTP) || defined(SCTP_SUPPORT) 775 if (m->m_pkthdr.csum_flags & CSUM_SCTP_IPV6) { 776 sctp_delayed_cksum(m, sizeof(struct ip6_hdr)); 777 m->m_pkthdr.csum_flags &= ~CSUM_SCTP_IPV6; 778 } 779#endif 780 } 781 782 error = ipsec6_check_pmtu(m, sp, forwarding); 783 if (error != 0) { 784 if (error == EJUSTRETURN) 785 return (0); 786 787 return (error); 788 } 789 790 /* NB: callee frees mbuf and releases reference to SP */ 791 error = ipsec6_process_packet(m, sp, inp); 792 if (error == EJUSTRETURN) { 793 /* 794 * We had a SP with a level of 'use' and no SA. We 795 * will just continue to process the packet without 796 * IPsec processing and return without error. 797 */ 798 return (0); 799 } 800 if (error == 0) 801 return (EINPROGRESS); /* consumed by IPsec */ 802 return (error); 803} 804 805/* 806 * IPSEC_OUTPUT() method implementation for IPv6. 807 * 0 - no IPsec handling needed 808 * other values - mbuf consumed by IPsec. 809 */ 810int 811ipsec6_output(struct mbuf *m, struct inpcb *inp) 812{ 813 814 /* 815 * If the packet is resubmitted to ip_output (e.g. after 816 * AH, ESP, etc. processing), there will be a tag to bypass 817 * the lookup and related policy checking. 818 */ 819 if (m_tag_find(m, PACKET_TAG_IPSEC_OUT_DONE, NULL) != NULL) 820 return (0); 821 822 return (ipsec6_common_output(m, inp, 0)); 823} 824 825/* 826 * IPSEC_FORWARD() method implementation for IPv6. 827 * 0 - no IPsec handling needed 828 * other values - mbuf consumed by IPsec. 829 */ 830int 831ipsec6_forward(struct mbuf *m) 832{ 833 834 /* 835 * Check if this packet has an active inbound SP and needs to be 836 * dropped instead of forwarded. 837 */ 838 if (ipsec6_in_reject(m, NULL) != 0) { 839 m_freem(m); 840 return (EACCES); 841 } 842 return (ipsec6_common_output(m, NULL, 1)); 843} 844#endif /* INET6 */ 845 846int 847ipsec_process_done(struct mbuf *m, struct secpolicy *sp, struct secasvar *sav, 848 u_int idx) 849{ 850 struct epoch_tracker et; 851 struct xform_history *xh; 852 struct secasindex *saidx; 853 struct m_tag *mtag; 854 int error; 855 856 saidx = &sav->sah->saidx; 857 switch (saidx->dst.sa.sa_family) { 858#ifdef INET 859 case AF_INET: 860 /* Fix the header length, for AH processing. */ 861 mtod(m, struct ip *)->ip_len = htons(m->m_pkthdr.len); 862 break; 863#endif /* INET */ 864#ifdef INET6 865 case AF_INET6: 866 /* Fix the header length, for AH processing. */ 867 if (m->m_pkthdr.len < sizeof (struct ip6_hdr)) { 868 error = ENXIO; 869 goto bad; 870 } 871 if (m->m_pkthdr.len - sizeof (struct ip6_hdr) > IPV6_MAXPACKET) { 872 /* No jumbogram support. */ 873 error = ENXIO; /*?*/ 874 goto bad; 875 } 876 mtod(m, struct ip6_hdr *)->ip6_plen = 877 htons(m->m_pkthdr.len - sizeof(struct ip6_hdr)); 878 break; 879#endif /* INET6 */ 880 default: 881 DPRINTF(("%s: unknown protocol family %u\n", __func__, 882 saidx->dst.sa.sa_family)); 883 error = ENXIO; 884 goto bad; 885 } 886 887 /* 888 * Add a record of what we've done to the packet. 889 */ 890 mtag = m_tag_get(PACKET_TAG_IPSEC_OUT_DONE, sizeof(*xh), M_NOWAIT); 891 if (mtag == NULL) { 892 DPRINTF(("%s: could not get packet tag\n", __func__)); 893 error = ENOMEM; 894 goto bad; 895 } 896 897 xh = (struct xform_history *)(mtag + 1); 898 xh->dst = saidx->dst; 899 xh->proto = saidx->proto; 900 xh->mode = saidx->mode; 901 xh->spi = sav->spi; 902 m_tag_prepend(m, mtag); 903 904 key_sa_recordxfer(sav, m); /* record data transfer */ 905 906 /* 907 * If there's another (bundled) SA to apply, do so. 908 * Note that this puts a burden on the kernel stack size. 909 * If this is a problem we'll need to introduce a queue 910 * to set the packet on so we can unwind the stack before 911 * doing further processing. 912 */ 913 if (++idx < sp->tcount) { 914 switch (saidx->dst.sa.sa_family) { 915#ifdef INET 916 case AF_INET: 917 key_freesav(&sav); 918 IPSECSTAT_INC(ips_out_bundlesa); 919 return (ipsec4_perform_request(m, sp, NULL, idx)); 920 /* NOTREACHED */ 921#endif 922#ifdef INET6 923 case AF_INET6: 924 key_freesav(&sav); 925 IPSEC6STAT_INC(ips_out_bundlesa); 926 return (ipsec6_perform_request(m, sp, NULL, idx)); 927 /* NOTREACHED */ 928#endif /* INET6 */ 929 default: 930 DPRINTF(("%s: unknown protocol family %u\n", __func__, 931 saidx->dst.sa.sa_family)); 932 error = EPFNOSUPPORT; 933 goto bad; 934 } 935 } 936 937 key_freesp(&sp), sp = NULL; /* Release reference to SP */ 938#if defined(INET) || defined(INET6) 939 /* 940 * Do UDP encapsulation if SA requires it. 941 */ 942 if (sav->natt != NULL) { 943 error = udp_ipsec_output(m, sav); 944 if (error != 0) 945 goto bad; 946 } 947#endif /* INET || INET6 */ 948 /* 949 * We're done with IPsec processing, transmit the packet using the 950 * appropriate network protocol (IP or IPv6). 951 */ 952 NET_EPOCH_ENTER(et); 953 switch (saidx->dst.sa.sa_family) { 954#ifdef INET 955 case AF_INET: 956 key_freesav(&sav); 957 error = ip_output(m, NULL, NULL, IP_RAWOUTPUT, NULL, NULL); 958 break; 959#endif /* INET */ 960#ifdef INET6 961 case AF_INET6: 962 key_freesav(&sav); 963 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL); 964 break; 965#endif /* INET6 */ 966 default: 967 panic("ipsec_process_done"); 968 } 969 NET_EPOCH_EXIT(et); 970 return (error); 971bad: 972 m_freem(m); 973 key_freesav(&sav); 974 if (sp != NULL) 975 key_freesp(&sp); 976 return (error); 977} 978 979/* 980 * ipsec_prepend() is optimized version of M_PREPEND(). 981 * ipsec_encap() is called by IPsec output routine for tunnel mode SA. 982 * It is expected that after IP encapsulation some IPsec transform will 983 * be performed. Each IPsec transform inserts its variable length header 984 * just after outer IP header using m_makespace(). If given mbuf has not 985 * enough free space at the beginning, we allocate new mbuf and reserve 986 * some space at the beginning and at the end. 987 * This helps avoid allocating of new mbuf and data copying in m_makespace(), 988 * we place outer header in the middle of mbuf's data with reserved leading 989 * and trailing space: 990 * [ LEADINGSPACE ][ Outer IP header ][ TRAILINGSPACE ] 991 * LEADINGSPACE will be used to add ethernet header, TRAILINGSPACE will 992 * be used to inject AH/ESP/IPCOMP header. 993 */ 994#define IPSEC_TRAILINGSPACE (sizeof(struct udphdr) +/* NAT-T */ \ 995 max(sizeof(struct newesp) + EALG_MAX_BLOCK_LEN, /* ESP + IV */ \ 996 sizeof(struct newah) + HASH_MAX_LEN /* AH + ICV */)) 997static struct mbuf * 998ipsec_prepend(struct mbuf *m, int len, int how) 999{ 1000 struct mbuf *n; 1001 1002 M_ASSERTPKTHDR(m); 1003 IPSEC_ASSERT(len < MHLEN, ("wrong length")); 1004 if (M_LEADINGSPACE(m) >= len) { 1005 /* No need to allocate new mbuf. */ 1006 m->m_data -= len; 1007 m->m_len += len; 1008 m->m_pkthdr.len += len; 1009 return (m); 1010 } 1011 n = m_gethdr(how, m->m_type); 1012 if (n == NULL) { 1013 m_freem(m); 1014 return (NULL); 1015 } 1016 m_move_pkthdr(n, m); 1017 n->m_next = m; 1018 if (len + IPSEC_TRAILINGSPACE < M_SIZE(n)) 1019 m_align(n, len + IPSEC_TRAILINGSPACE); 1020 n->m_len = len; 1021 n->m_pkthdr.len += len; 1022 return (n); 1023} 1024 1025static size_t 1026ipsec_get_pmtu(struct secasvar *sav) 1027{ 1028 union sockaddr_union *dst; 1029 struct in_conninfo inc; 1030 size_t pmtu; 1031 1032 dst = &sav->sah->saidx.dst; 1033 memset(&inc, 0, sizeof(inc)); 1034 1035 switch (dst->sa.sa_family) { 1036#ifdef INET 1037 case AF_INET: 1038 inc.inc_faddr = satosin(&dst->sa)->sin_addr; 1039 break; 1040#endif 1041#ifdef INET6 1042 case AF_INET6: 1043 inc.inc6_faddr = satosin6(&dst->sa)->sin6_addr; 1044 inc.inc_flags |= INC_ISIPV6; 1045 break; 1046#endif 1047 default: 1048 return (0); 1049 } 1050 1051 pmtu = tcp_hc_getmtu(&inc); 1052 if (pmtu != 0) 1053 return (pmtu); 1054 1055 /* No entry in hostcache. Assume that PMTU is equal to link's MTU */ 1056 switch (dst->sa.sa_family) { 1057#ifdef INET 1058 case AF_INET: 1059 pmtu = tcp_maxmtu(&inc, NULL); 1060 break; 1061#endif 1062#ifdef INET6 1063 case AF_INET6: 1064 pmtu = tcp_maxmtu6(&inc, NULL); 1065 break; 1066#endif 1067 default: 1068 return (0); 1069 } 1070 if (pmtu == 0) 1071 return (0); 1072 1073 tcp_hc_updatemtu(&inc, pmtu); 1074 1075 return (pmtu); 1076} 1077 1078static int 1079ipsec_encap(struct mbuf **mp, struct secasindex *saidx) 1080{ 1081#ifdef INET6 1082 struct ip6_hdr *ip6; 1083#endif 1084 struct ip *ip; 1085#ifdef INET 1086 int setdf; 1087#endif 1088 uint8_t itos, proto; 1089 1090 ip = mtod(*mp, struct ip *); 1091 switch (ip->ip_v) { 1092#ifdef INET 1093 case IPVERSION: 1094 proto = IPPROTO_IPIP; 1095 /* 1096 * Collect IP_DF state from the inner header 1097 * and honor system-wide control of how to handle it. 1098 */ 1099 switch (V_ip4_ipsec_dfbit) { 1100 case 0: /* clear in outer header */ 1101 case 1: /* set in outer header */ 1102 setdf = V_ip4_ipsec_dfbit; 1103 break; 1104 default:/* propagate to outer header */ 1105 setdf = (ip->ip_off & htons(IP_DF)) != 0; 1106 } 1107 itos = ip->ip_tos; 1108 break; 1109#endif 1110#ifdef INET6 1111 case (IPV6_VERSION >> 4): 1112 proto = IPPROTO_IPV6; 1113 ip6 = mtod(*mp, struct ip6_hdr *); 1114 itos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 1115 /* scoped address handling */ 1116 in6_clearscope(&ip6->ip6_src); 1117 in6_clearscope(&ip6->ip6_dst); 1118 break; 1119#endif 1120 default: 1121 return (EAFNOSUPPORT); 1122 } 1123 switch (saidx->dst.sa.sa_family) { 1124#ifdef INET 1125 case AF_INET: 1126 if (saidx->src.sa.sa_family != AF_INET || 1127 saidx->src.sin.sin_addr.s_addr == INADDR_ANY || 1128 saidx->dst.sin.sin_addr.s_addr == INADDR_ANY) 1129 return (EINVAL); 1130 *mp = ipsec_prepend(*mp, sizeof(struct ip), M_NOWAIT); 1131 if (*mp == NULL) 1132 return (ENOBUFS); 1133 ip = mtod(*mp, struct ip *); 1134 ip->ip_v = IPVERSION; 1135 ip->ip_hl = sizeof(struct ip) >> 2; 1136 ip->ip_p = proto; 1137 ip->ip_len = htons((*mp)->m_pkthdr.len); 1138 ip->ip_ttl = V_ip_defttl; 1139 ip->ip_sum = 0; 1140 ip->ip_off = setdf ? htons(IP_DF): 0; 1141 ip->ip_src = saidx->src.sin.sin_addr; 1142 ip->ip_dst = saidx->dst.sin.sin_addr; 1143 ip_ecn_ingress(V_ip4_ipsec_ecn, &ip->ip_tos, &itos); 1144 ip_fillid(ip); 1145 break; 1146#endif /* INET */ 1147#ifdef INET6 1148 case AF_INET6: 1149 if (saidx->src.sa.sa_family != AF_INET6 || 1150 IN6_IS_ADDR_UNSPECIFIED(&saidx->src.sin6.sin6_addr) || 1151 IN6_IS_ADDR_UNSPECIFIED(&saidx->dst.sin6.sin6_addr)) 1152 return (EINVAL); 1153 *mp = ipsec_prepend(*mp, sizeof(struct ip6_hdr), M_NOWAIT); 1154 if (*mp == NULL) 1155 return (ENOBUFS); 1156 ip6 = mtod(*mp, struct ip6_hdr *); 1157 ip6->ip6_flow = 0; 1158 ip6->ip6_vfc = IPV6_VERSION; 1159 ip6->ip6_hlim = V_ip6_defhlim; 1160 ip6->ip6_nxt = proto; 1161 ip6->ip6_dst = saidx->dst.sin6.sin6_addr; 1162 /* For link-local address embed scope zone id */ 1163 if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_dst)) 1164 ip6->ip6_dst.s6_addr16[1] = 1165 htons(saidx->dst.sin6.sin6_scope_id & 0xffff); 1166 ip6->ip6_src = saidx->src.sin6.sin6_addr; 1167 if (IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) 1168 ip6->ip6_src.s6_addr16[1] = 1169 htons(saidx->src.sin6.sin6_scope_id & 0xffff); 1170 ip6->ip6_plen = htons((*mp)->m_pkthdr.len - sizeof(*ip6)); 1171 ip_ecn_ingress(V_ip6_ipsec_ecn, &proto, &itos); 1172 ip6->ip6_flow |= htonl((uint32_t)proto << 20); 1173 break; 1174#endif /* INET6 */ 1175 default: 1176 return (EAFNOSUPPORT); 1177 } 1178 (*mp)->m_flags &= ~(M_BCAST | M_MCAST); 1179 return (0); 1180} 1181