1 /*- 2 * Copyright 2001 Niels Provos <provos@citi.umich.edu> 3 * Copyright 2011 Alexander Bluhm <bluhm@openbsd.org> 4 * All rights reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 16 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 17 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 18 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 19 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 20 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 21 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 22 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 23 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 24 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 25 * 26 * $OpenBSD: pf_norm.c,v 1.114 2009/01/29 14:11:45 henning Exp $ 27 */ 28 29 #include <sys/cdefs.h> 30 __FBSDID("$FreeBSD$"); 31 32 #include "opt_inet.h" 33 #include "opt_inet6.h" 34 #include "opt_pf.h" 35 36 #include <sys/param.h> 37 #include <sys/kernel.h> 38 #include <sys/lock.h> 39 #include <sys/mbuf.h> 40 #include <sys/mutex.h> 41 #include <sys/refcount.h> 42 #include <sys/rwlock.h> 43 #include <sys/socket.h> 44 45 #include <net/if.h> 46 #include <net/vnet.h> 47 #include <net/pfvar.h> 48 #include <net/if_pflog.h> 49 50 #include <netinet/in.h> 51 #include <netinet/ip.h> 52 #include <netinet/ip_var.h> 53 #include <netinet6/ip6_var.h> 54 #include <netinet/tcp.h> 55 #include <netinet/tcp_fsm.h> 56 #include <netinet/tcp_seq.h> 57 58 #ifdef INET6 59 #include <netinet/ip6.h> 60 #endif /* INET6 */ 61 62 struct pf_frent { 63 TAILQ_ENTRY(pf_frent) fr_next; 64 struct mbuf *fe_m; 65 uint16_t fe_hdrlen; /* ipv4 header length with ip options 66 ipv6, extension, fragment header */ 67 uint16_t fe_extoff; /* last extension header offset or 0 */ 68 uint16_t fe_len; /* fragment length */ 69 uint16_t fe_off; /* fragment offset */ 70 uint16_t fe_mff; /* more fragment flag */ 71 }; 72 73 struct pf_fragment_cmp { 74 struct pf_addr frc_src; 75 struct pf_addr frc_dst; 76 uint32_t frc_id; 77 sa_family_t frc_af; 78 uint8_t frc_proto; 79 }; 80 81 struct pf_fragment { 82 struct pf_fragment_cmp fr_key; 83 #define fr_src fr_key.frc_src 84 #define fr_dst fr_key.frc_dst 85 #define fr_id fr_key.frc_id 86 #define fr_af fr_key.frc_af 87 #define fr_proto fr_key.frc_proto 88 89 RB_ENTRY(pf_fragment) fr_entry; 90 TAILQ_ENTRY(pf_fragment) frag_next; 91 uint32_t fr_timeout; 92 uint16_t fr_maxlen; /* maximum length of single fragment */ 93 TAILQ_HEAD(pf_fragq, pf_frent) fr_queue; 94 }; 95 96 struct pf_fragment_tag { 97 uint16_t ft_hdrlen; /* header length of reassembled pkt */ 98 uint16_t ft_extoff; /* last extension header offset or 0 */ 99 uint16_t ft_maxlen; /* maximum fragment payload length */ 100 uint32_t ft_id; /* fragment id */ 101 }; 102 103 static struct mtx pf_frag_mtx; 104 MTX_SYSINIT(pf_frag_mtx, &pf_frag_mtx, "pf fragments", MTX_DEF); 105 #define PF_FRAG_LOCK() mtx_lock(&pf_frag_mtx) 106 #define PF_FRAG_UNLOCK() mtx_unlock(&pf_frag_mtx) 107 #define PF_FRAG_ASSERT() mtx_assert(&pf_frag_mtx, MA_OWNED) 108 109 VNET_DEFINE(uma_zone_t, pf_state_scrub_z); /* XXX: shared with pfsync */ 110 111 static VNET_DEFINE(uma_zone_t, pf_frent_z); 112 #define V_pf_frent_z VNET(pf_frent_z) 113 static VNET_DEFINE(uma_zone_t, pf_frag_z); 114 #define V_pf_frag_z VNET(pf_frag_z) 115 116 TAILQ_HEAD(pf_fragqueue, pf_fragment); 117 TAILQ_HEAD(pf_cachequeue, pf_fragment); 118 static VNET_DEFINE(struct pf_fragqueue, pf_fragqueue); 119 #define V_pf_fragqueue VNET(pf_fragqueue) 120 RB_HEAD(pf_frag_tree, pf_fragment); 121 static VNET_DEFINE(struct pf_frag_tree, pf_frag_tree); 122 #define V_pf_frag_tree VNET(pf_frag_tree) 123 static int pf_frag_compare(struct pf_fragment *, 124 struct pf_fragment *); 125 static RB_PROTOTYPE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare); 126 static RB_GENERATE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare); 127 128 static void pf_flush_fragments(void); 129 static void pf_free_fragment(struct pf_fragment *); 130 static void pf_remove_fragment(struct pf_fragment *); 131 static int pf_normalize_tcpopt(struct pf_rule *, struct mbuf *, 132 struct tcphdr *, int, sa_family_t); 133 static struct pf_frent *pf_create_fragment(u_short *); 134 static struct pf_fragment *pf_find_fragment(struct pf_fragment_cmp *key, 135 struct pf_frag_tree *tree); 136 static struct pf_fragment *pf_fillup_fragment(struct pf_fragment_cmp *, 137 struct pf_frent *, u_short *); 138 static int pf_isfull_fragment(struct pf_fragment *); 139 static struct mbuf *pf_join_fragment(struct pf_fragment *); 140 #ifdef INET 141 static void pf_scrub_ip(struct mbuf **, uint32_t, uint8_t, uint8_t); 142 static int pf_reassemble(struct mbuf **, struct ip *, int, u_short *); 143 #endif /* INET */ 144 #ifdef INET6 145 static int pf_reassemble6(struct mbuf **, struct ip6_hdr *, 146 struct ip6_frag *, uint16_t, uint16_t, u_short *); 147 static void pf_scrub_ip6(struct mbuf **, uint8_t); 148 #endif /* INET6 */ 149 150 #define DPFPRINTF(x) do { \ 151 if (V_pf_status.debug >= PF_DEBUG_MISC) { \ 152 printf("%s: ", __func__); \ 153 printf x ; \ 154 } \ 155 } while(0) 156 157 #ifdef INET 158 static void 159 pf_ip2key(struct ip *ip, int dir, struct pf_fragment_cmp *key) 160 { 161 162 key->frc_src.v4 = ip->ip_src; 163 key->frc_dst.v4 = ip->ip_dst; 164 key->frc_af = AF_INET; 165 key->frc_proto = ip->ip_p; 166 key->frc_id = ip->ip_id; 167 } 168 #endif /* INET */ 169 170 void 171 pf_normalize_init(void) 172 { 173 174 V_pf_frag_z = uma_zcreate("pf frags", sizeof(struct pf_fragment), 175 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 176 V_pf_frent_z = uma_zcreate("pf frag entries", sizeof(struct pf_frent), 177 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 178 V_pf_state_scrub_z = uma_zcreate("pf state scrubs", 179 sizeof(struct pf_state_scrub), NULL, NULL, NULL, NULL, 180 UMA_ALIGN_PTR, 0); 181 182 V_pf_limits[PF_LIMIT_FRAGS].zone = V_pf_frent_z; 183 V_pf_limits[PF_LIMIT_FRAGS].limit = PFFRAG_FRENT_HIWAT; 184 uma_zone_set_max(V_pf_frent_z, PFFRAG_FRENT_HIWAT); 185 uma_zone_set_warning(V_pf_frent_z, "PF frag entries limit reached"); 186 187 TAILQ_INIT(&V_pf_fragqueue); 188 } 189 190 void 191 pf_normalize_cleanup(void) 192 { 193 194 uma_zdestroy(V_pf_state_scrub_z); 195 uma_zdestroy(V_pf_frent_z); 196 uma_zdestroy(V_pf_frag_z); 197 } 198 199 static int 200 pf_frag_compare(struct pf_fragment *a, struct pf_fragment *b) 201 { 202 int diff; 203 204 if ((diff = a->fr_id - b->fr_id) != 0) 205 return (diff); 206 if ((diff = a->fr_proto - b->fr_proto) != 0) 207 return (diff); 208 if ((diff = a->fr_af - b->fr_af) != 0) 209 return (diff); 210 if ((diff = pf_addr_cmp(&a->fr_src, &b->fr_src, a->fr_af)) != 0) 211 return (diff); 212 if ((diff = pf_addr_cmp(&a->fr_dst, &b->fr_dst, a->fr_af)) != 0) 213 return (diff); 214 return (0); 215 } 216 217 void 218 pf_purge_expired_fragments(void) 219 { 220 struct pf_fragment *frag; 221 u_int32_t expire = time_uptime - 222 V_pf_default_rule.timeout[PFTM_FRAG]; 223 224 PF_FRAG_LOCK(); 225 while ((frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue)) != NULL) { 226 if (frag->fr_timeout > expire) 227 break; 228 229 DPFPRINTF(("expiring %d(%p)\n", frag->fr_id, frag)); 230 pf_free_fragment(frag); 231 } 232 233 PF_FRAG_UNLOCK(); 234 } 235 236 /* 237 * Try to flush old fragments to make space for new ones 238 */ 239 static void 240 pf_flush_fragments(void) 241 { 242 struct pf_fragment *frag; 243 int goal; 244 245 PF_FRAG_ASSERT(); 246 247 goal = uma_zone_get_cur(V_pf_frent_z) * 9 / 10; 248 DPFPRINTF(("trying to free %d frag entriess\n", goal)); 249 while (goal < uma_zone_get_cur(V_pf_frent_z)) { 250 frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue); 251 if (frag) 252 pf_free_fragment(frag); 253 else 254 break; 255 } 256 } 257 258 /* Frees the fragments and all associated entries */ 259 static void 260 pf_free_fragment(struct pf_fragment *frag) 261 { 262 struct pf_frent *frent; 263 264 PF_FRAG_ASSERT(); 265 266 /* Free all fragments */ 267 for (frent = TAILQ_FIRST(&frag->fr_queue); frent; 268 frent = TAILQ_FIRST(&frag->fr_queue)) { 269 TAILQ_REMOVE(&frag->fr_queue, frent, fr_next); 270 271 m_freem(frent->fe_m); 272 uma_zfree(V_pf_frent_z, frent); 273 } 274 275 pf_remove_fragment(frag); 276 } 277 278 static struct pf_fragment * 279 pf_find_fragment(struct pf_fragment_cmp *key, struct pf_frag_tree *tree) 280 { 281 struct pf_fragment *frag; 282 283 PF_FRAG_ASSERT(); 284 285 frag = RB_FIND(pf_frag_tree, tree, (struct pf_fragment *)key); 286 if (frag != NULL) { 287 /* XXX Are we sure we want to update the timeout? */ 288 frag->fr_timeout = time_uptime; 289 TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next); 290 TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next); 291 } 292 293 return (frag); 294 } 295 296 /* Removes a fragment from the fragment queue and frees the fragment */ 297 static void 298 pf_remove_fragment(struct pf_fragment *frag) 299 { 300 301 PF_FRAG_ASSERT(); 302 303 RB_REMOVE(pf_frag_tree, &V_pf_frag_tree, frag); 304 TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next); 305 uma_zfree(V_pf_frag_z, frag); 306 } 307 308 static struct pf_frent * 309 pf_create_fragment(u_short *reason) 310 { 311 struct pf_frent *frent; 312 313 PF_FRAG_ASSERT(); 314 315 frent = uma_zalloc(V_pf_frent_z, M_NOWAIT); 316 if (frent == NULL) { 317 pf_flush_fragments(); 318 frent = uma_zalloc(V_pf_frent_z, M_NOWAIT); 319 if (frent == NULL) { 320 REASON_SET(reason, PFRES_MEMORY); 321 return (NULL); 322 } 323 } 324 325 return (frent); 326 } 327 328 static struct pf_fragment * 329 pf_fillup_fragment(struct pf_fragment_cmp *key, struct pf_frent *frent, 330 u_short *reason) 331 { 332 struct pf_frent *after, *next, *prev; 333 struct pf_fragment *frag; 334 uint16_t total; 335 336 PF_FRAG_ASSERT(); 337 338 /* No empty fragments. */ 339 if (frent->fe_len == 0) { 340 DPFPRINTF(("bad fragment: len 0")); 341 goto bad_fragment; 342 } 343 344 /* All fragments are 8 byte aligned. */ 345 if (frent->fe_mff && (frent->fe_len & 0x7)) { 346 DPFPRINTF(("bad fragment: mff and len %d", frent->fe_len)); 347 goto bad_fragment; 348 } 349 350 /* Respect maximum length, IP_MAXPACKET == IPV6_MAXPACKET. */ 351 if (frent->fe_off + frent->fe_len > IP_MAXPACKET) { 352 DPFPRINTF(("bad fragment: max packet %d", 353 frent->fe_off + frent->fe_len)); 354 goto bad_fragment; 355 } 356 357 DPFPRINTF((key->frc_af == AF_INET ? 358 "reass frag %d @ %d-%d" : "reass frag %#08x @ %d-%d", 359 key->frc_id, frent->fe_off, frent->fe_off + frent->fe_len)); 360 361 /* Fully buffer all of the fragments in this fragment queue. */ 362 frag = pf_find_fragment(key, &V_pf_frag_tree); 363 364 /* Create a new reassembly queue for this packet. */ 365 if (frag == NULL) { 366 frag = uma_zalloc(V_pf_frag_z, M_NOWAIT); 367 if (frag == NULL) { 368 pf_flush_fragments(); 369 frag = uma_zalloc(V_pf_frag_z, M_NOWAIT); 370 if (frag == NULL) { 371 REASON_SET(reason, PFRES_MEMORY); 372 goto drop_fragment; 373 } 374 } 375 376 *(struct pf_fragment_cmp *)frag = *key; 377 frag->fr_timeout = time_uptime; 378 frag->fr_maxlen = frent->fe_len; 379 TAILQ_INIT(&frag->fr_queue); 380 381 RB_INSERT(pf_frag_tree, &V_pf_frag_tree, frag); 382 TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next); 383 384 /* We do not have a previous fragment. */ 385 TAILQ_INSERT_HEAD(&frag->fr_queue, frent, fr_next); 386 387 return (frag); 388 } 389 390 KASSERT(!TAILQ_EMPTY(&frag->fr_queue), ("!TAILQ_EMPTY()->fr_queue")); 391 392 /* Remember maximum fragment len for refragmentation. */ 393 if (frent->fe_len > frag->fr_maxlen) 394 frag->fr_maxlen = frent->fe_len; 395 396 /* Maximum data we have seen already. */ 397 total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off + 398 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len; 399 400 /* Non terminal fragments must have more fragments flag. */ 401 if (frent->fe_off + frent->fe_len < total && !frent->fe_mff) 402 goto bad_fragment; 403 404 /* Check if we saw the last fragment already. */ 405 if (!TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_mff) { 406 if (frent->fe_off + frent->fe_len > total || 407 (frent->fe_off + frent->fe_len == total && frent->fe_mff)) 408 goto bad_fragment; 409 } else { 410 if (frent->fe_off + frent->fe_len == total && !frent->fe_mff) 411 goto bad_fragment; 412 } 413 414 /* Find a fragment after the current one. */ 415 prev = NULL; 416 TAILQ_FOREACH(after, &frag->fr_queue, fr_next) { 417 if (after->fe_off > frent->fe_off) 418 break; 419 prev = after; 420 } 421 422 KASSERT(prev != NULL || after != NULL, 423 ("prev != NULL || after != NULL")); 424 425 if (prev != NULL && prev->fe_off + prev->fe_len > frent->fe_off) { 426 uint16_t precut; 427 428 precut = prev->fe_off + prev->fe_len - frent->fe_off; 429 if (precut >= frent->fe_len) 430 goto bad_fragment; 431 DPFPRINTF(("overlap -%d", precut)); 432 m_adj(frent->fe_m, precut); 433 frent->fe_off += precut; 434 frent->fe_len -= precut; 435 } 436 437 for (; after != NULL && frent->fe_off + frent->fe_len > after->fe_off; 438 after = next) { 439 uint16_t aftercut; 440 441 aftercut = frent->fe_off + frent->fe_len - after->fe_off; 442 DPFPRINTF(("adjust overlap %d", aftercut)); 443 if (aftercut < after->fe_len) { 444 m_adj(after->fe_m, aftercut); 445 after->fe_off += aftercut; 446 after->fe_len -= aftercut; 447 break; 448 } 449 450 /* This fragment is completely overlapped, lose it. */ 451 next = TAILQ_NEXT(after, fr_next); 452 m_freem(after->fe_m); 453 TAILQ_REMOVE(&frag->fr_queue, after, fr_next); 454 uma_zfree(V_pf_frent_z, after); 455 } 456 457 if (prev == NULL) 458 TAILQ_INSERT_HEAD(&frag->fr_queue, frent, fr_next); 459 else 460 TAILQ_INSERT_AFTER(&frag->fr_queue, prev, frent, fr_next); 461 462 return (frag); 463 464 bad_fragment: 465 REASON_SET(reason, PFRES_FRAG); 466 drop_fragment: 467 uma_zfree(V_pf_frent_z, frent); 468 return (NULL); 469 } 470 471 static int 472 pf_isfull_fragment(struct pf_fragment *frag) 473 { 474 struct pf_frent *frent, *next; 475 uint16_t off, total; 476 477 /* Check if we are completely reassembled */ 478 if (TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_mff) 479 return (0); 480 481 /* Maximum data we have seen already */ 482 total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off + 483 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len; 484 485 /* Check if we have all the data */ 486 off = 0; 487 for (frent = TAILQ_FIRST(&frag->fr_queue); frent; frent = next) { 488 next = TAILQ_NEXT(frent, fr_next); 489 490 off += frent->fe_len; 491 if (off < total && (next == NULL || next->fe_off != off)) { 492 DPFPRINTF(("missing fragment at %d, next %d, total %d", 493 off, next == NULL ? -1 : next->fe_off, total)); 494 return (0); 495 } 496 } 497 DPFPRINTF(("%d < %d?", off, total)); 498 if (off < total) 499 return (0); 500 KASSERT(off == total, ("off == total")); 501 502 return (1); 503 } 504 505 static struct mbuf * 506 pf_join_fragment(struct pf_fragment *frag) 507 { 508 struct mbuf *m, *m2; 509 struct pf_frent *frent, *next; 510 511 frent = TAILQ_FIRST(&frag->fr_queue); 512 next = TAILQ_NEXT(frent, fr_next); 513 514 m = frent->fe_m; 515 m_adj(m, (frent->fe_hdrlen + frent->fe_len) - m->m_pkthdr.len); 516 uma_zfree(V_pf_frent_z, frent); 517 for (frent = next; frent != NULL; frent = next) { 518 next = TAILQ_NEXT(frent, fr_next); 519 520 m2 = frent->fe_m; 521 /* Strip off ip header. */ 522 m_adj(m2, frent->fe_hdrlen); 523 /* Strip off any trailing bytes. */ 524 m_adj(m2, frent->fe_len - m2->m_pkthdr.len); 525 526 uma_zfree(V_pf_frent_z, frent); 527 m_cat(m, m2); 528 } 529 530 /* Remove from fragment queue. */ 531 pf_remove_fragment(frag); 532 533 return (m); 534 } 535 536 #ifdef INET 537 static int 538 pf_reassemble(struct mbuf **m0, struct ip *ip, int dir, u_short *reason) 539 { 540 struct mbuf *m = *m0; 541 struct pf_frent *frent; 542 struct pf_fragment *frag; 543 struct pf_fragment_cmp key; 544 uint16_t total, hdrlen; 545 546 /* Get an entry for the fragment queue */ 547 if ((frent = pf_create_fragment(reason)) == NULL) 548 return (PF_DROP); 549 550 frent->fe_m = m; 551 frent->fe_hdrlen = ip->ip_hl << 2; 552 frent->fe_extoff = 0; 553 frent->fe_len = ntohs(ip->ip_len) - (ip->ip_hl << 2); 554 frent->fe_off = (ntohs(ip->ip_off) & IP_OFFMASK) << 3; 555 frent->fe_mff = ntohs(ip->ip_off) & IP_MF; 556 557 pf_ip2key(ip, dir, &key); 558 559 if ((frag = pf_fillup_fragment(&key, frent, reason)) == NULL) 560 return (PF_DROP); 561 562 /* The mbuf is part of the fragment entry, no direct free or access */ 563 m = *m0 = NULL; 564 565 if (!pf_isfull_fragment(frag)) 566 return (PF_PASS); /* drop because *m0 is NULL, no error */ 567 568 /* We have all the data */ 569 frent = TAILQ_FIRST(&frag->fr_queue); 570 KASSERT(frent != NULL, ("frent != NULL")); 571 total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off + 572 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len; 573 hdrlen = frent->fe_hdrlen; 574 575 m = *m0 = pf_join_fragment(frag); 576 frag = NULL; 577 578 if (m->m_flags & M_PKTHDR) { 579 int plen = 0; 580 for (m = *m0; m; m = m->m_next) 581 plen += m->m_len; 582 m = *m0; 583 m->m_pkthdr.len = plen; 584 } 585 586 ip = mtod(m, struct ip *); 587 ip->ip_len = htons(hdrlen + total); 588 ip->ip_off &= ~(IP_MF|IP_OFFMASK); 589 590 if (hdrlen + total > IP_MAXPACKET) { 591 DPFPRINTF(("drop: too big: %d", total)); 592 ip->ip_len = 0; 593 REASON_SET(reason, PFRES_SHORT); 594 /* PF_DROP requires a valid mbuf *m0 in pf_test() */ 595 return (PF_DROP); 596 } 597 598 DPFPRINTF(("complete: %p(%d)\n", m, ntohs(ip->ip_len))); 599 return (PF_PASS); 600 } 601 #endif /* INET */ 602 603 #ifdef INET6 604 static int 605 pf_reassemble6(struct mbuf **m0, struct ip6_hdr *ip6, struct ip6_frag *fraghdr, 606 uint16_t hdrlen, uint16_t extoff, u_short *reason) 607 { 608 struct mbuf *m = *m0; 609 struct pf_frent *frent; 610 struct pf_fragment *frag; 611 struct pf_fragment_cmp key; 612 struct m_tag *mtag; 613 struct pf_fragment_tag *ftag; 614 int off; 615 uint32_t frag_id; 616 uint16_t total, maxlen; 617 uint8_t proto; 618 619 PF_FRAG_LOCK(); 620 621 /* Get an entry for the fragment queue. */ 622 if ((frent = pf_create_fragment(reason)) == NULL) { 623 PF_FRAG_UNLOCK(); 624 return (PF_DROP); 625 } 626 627 frent->fe_m = m; 628 frent->fe_hdrlen = hdrlen; 629 frent->fe_extoff = extoff; 630 frent->fe_len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - hdrlen; 631 frent->fe_off = ntohs(fraghdr->ip6f_offlg & IP6F_OFF_MASK); 632 frent->fe_mff = fraghdr->ip6f_offlg & IP6F_MORE_FRAG; 633 634 key.frc_src.v6 = ip6->ip6_src; 635 key.frc_dst.v6 = ip6->ip6_dst; 636 key.frc_af = AF_INET6; 637 /* Only the first fragment's protocol is relevant. */ 638 key.frc_proto = 0; 639 key.frc_id = fraghdr->ip6f_ident; 640 641 if ((frag = pf_fillup_fragment(&key, frent, reason)) == NULL) { 642 PF_FRAG_UNLOCK(); 643 return (PF_DROP); 644 } 645 646 /* The mbuf is part of the fragment entry, no direct free or access. */ 647 m = *m0 = NULL; 648 649 if (!pf_isfull_fragment(frag)) { 650 PF_FRAG_UNLOCK(); 651 return (PF_PASS); /* Drop because *m0 is NULL, no error. */ 652 } 653 654 /* We have all the data. */ 655 extoff = frent->fe_extoff; 656 maxlen = frag->fr_maxlen; 657 frag_id = frag->fr_id; 658 frent = TAILQ_FIRST(&frag->fr_queue); 659 KASSERT(frent != NULL, ("frent != NULL")); 660 total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off + 661 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len; 662 hdrlen = frent->fe_hdrlen - sizeof(struct ip6_frag); 663 664 m = *m0 = pf_join_fragment(frag); 665 frag = NULL; 666 667 PF_FRAG_UNLOCK(); 668 669 /* Take protocol from first fragment header. */ 670 m = m_getptr(m, hdrlen + offsetof(struct ip6_frag, ip6f_nxt), &off); 671 KASSERT(m, ("%s: short mbuf chain", __func__)); 672 proto = *(mtod(m, caddr_t) + off); 673 m = *m0; 674 675 /* Delete frag6 header */ 676 if (ip6_deletefraghdr(m, hdrlen, M_NOWAIT) != 0) 677 goto fail; 678 679 if (m->m_flags & M_PKTHDR) { 680 int plen = 0; 681 for (m = *m0; m; m = m->m_next) 682 plen += m->m_len; 683 m = *m0; 684 m->m_pkthdr.len = plen; 685 } 686 687 if ((mtag = m_tag_get(PF_REASSEMBLED, sizeof(struct pf_fragment_tag), 688 M_NOWAIT)) == NULL) 689 goto fail; 690 ftag = (struct pf_fragment_tag *)(mtag + 1); 691 ftag->ft_hdrlen = hdrlen; 692 ftag->ft_extoff = extoff; 693 ftag->ft_maxlen = maxlen; 694 ftag->ft_id = frag_id; 695 m_tag_prepend(m, mtag); 696 697 ip6 = mtod(m, struct ip6_hdr *); 698 ip6->ip6_plen = htons(hdrlen - sizeof(struct ip6_hdr) + total); 699 if (extoff) { 700 /* Write protocol into next field of last extension header. */ 701 m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt), 702 &off); 703 KASSERT(m, ("%s: short mbuf chain", __func__)); 704 *(mtod(m, char *) + off) = proto; 705 m = *m0; 706 } else 707 ip6->ip6_nxt = proto; 708 709 if (hdrlen - sizeof(struct ip6_hdr) + total > IPV6_MAXPACKET) { 710 DPFPRINTF(("drop: too big: %d", total)); 711 ip6->ip6_plen = 0; 712 REASON_SET(reason, PFRES_SHORT); 713 /* PF_DROP requires a valid mbuf *m0 in pf_test6(). */ 714 return (PF_DROP); 715 } 716 717 DPFPRINTF(("complete: %p(%d)", m, ntohs(ip6->ip6_plen))); 718 return (PF_PASS); 719 720 fail: 721 REASON_SET(reason, PFRES_MEMORY); 722 /* PF_DROP requires a valid mbuf *m0 in pf_test6(), will free later. */ 723 return (PF_DROP); 724 } 725 #endif /* INET6 */ 726 727 #ifdef INET6 728 int 729 pf_refragment6(struct ifnet *ifp, struct mbuf **m0, struct m_tag *mtag) 730 { 731 struct mbuf *m = *m0, *t; 732 struct pf_fragment_tag *ftag = (struct pf_fragment_tag *)(mtag + 1); 733 struct pf_pdesc pd; 734 uint32_t frag_id; 735 uint16_t hdrlen, extoff, maxlen; 736 uint8_t proto; 737 int error, action; 738 739 hdrlen = ftag->ft_hdrlen; 740 extoff = ftag->ft_extoff; 741 maxlen = ftag->ft_maxlen; 742 frag_id = ftag->ft_id; 743 m_tag_delete(m, mtag); 744 mtag = NULL; 745 ftag = NULL; 746 747 if (extoff) { 748 int off; 749 750 /* Use protocol from next field of last extension header */ 751 m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt), 752 &off); 753 KASSERT((m != NULL), ("pf_refragment6: short mbuf chain")); 754 proto = *(mtod(m, caddr_t) + off); 755 *(mtod(m, char *) + off) = IPPROTO_FRAGMENT; 756 m = *m0; 757 } else { 758 struct ip6_hdr *hdr; 759 760 hdr = mtod(m, struct ip6_hdr *); 761 proto = hdr->ip6_nxt; 762 hdr->ip6_nxt = IPPROTO_FRAGMENT; 763 } 764 765 /* The MTU must be a multiple of 8 bytes, or we risk doing the 766 * fragmentation wrong. */ 767 maxlen = maxlen & ~7; 768 769 /* 770 * Maxlen may be less than 8 if there was only a single 771 * fragment. As it was fragmented before, add a fragment 772 * header also for a single fragment. If total or maxlen 773 * is less than 8, ip6_fragment() will return EMSGSIZE and 774 * we drop the packet. 775 */ 776 error = ip6_fragment(ifp, m, hdrlen, proto, maxlen, frag_id); 777 m = (*m0)->m_nextpkt; 778 (*m0)->m_nextpkt = NULL; 779 if (error == 0) { 780 /* The first mbuf contains the unfragmented packet. */ 781 m_freem(*m0); 782 *m0 = NULL; 783 action = PF_PASS; 784 } else { 785 /* Drop expects an mbuf to free. */ 786 DPFPRINTF(("refragment error %d", error)); 787 action = PF_DROP; 788 } 789 for (t = m; m; m = t) { 790 t = m->m_nextpkt; 791 m->m_nextpkt = NULL; 792 m->m_flags |= M_SKIP_FIREWALL; 793 memset(&pd, 0, sizeof(pd)); 794 pd.pf_mtag = pf_find_mtag(m); 795 if (error == 0) 796 ip6_forward(m, 0); 797 else 798 m_freem(m); 799 } 800 801 return (action); 802 } 803 #endif /* INET6 */ 804 805 #ifdef INET 806 int 807 pf_normalize_ip(struct mbuf **m0, int dir, struct pfi_kif *kif, u_short *reason, 808 struct pf_pdesc *pd) 809 { 810 struct mbuf *m = *m0; 811 struct pf_rule *r; 812 struct ip *h = mtod(m, struct ip *); 813 int mff = (ntohs(h->ip_off) & IP_MF); 814 int hlen = h->ip_hl << 2; 815 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3; 816 u_int16_t max; 817 int ip_len; 818 int ip_off; 819 int tag = -1; 820 int verdict; 821 822 PF_RULES_RASSERT(); 823 824 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr); 825 while (r != NULL) { 826 r->evaluations++; 827 if (pfi_kif_match(r->kif, kif) == r->ifnot) 828 r = r->skip[PF_SKIP_IFP].ptr; 829 else if (r->direction && r->direction != dir) 830 r = r->skip[PF_SKIP_DIR].ptr; 831 else if (r->af && r->af != AF_INET) 832 r = r->skip[PF_SKIP_AF].ptr; 833 else if (r->proto && r->proto != h->ip_p) 834 r = r->skip[PF_SKIP_PROTO].ptr; 835 else if (PF_MISMATCHAW(&r->src.addr, 836 (struct pf_addr *)&h->ip_src.s_addr, AF_INET, 837 r->src.neg, kif, M_GETFIB(m))) 838 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 839 else if (PF_MISMATCHAW(&r->dst.addr, 840 (struct pf_addr *)&h->ip_dst.s_addr, AF_INET, 841 r->dst.neg, NULL, M_GETFIB(m))) 842 r = r->skip[PF_SKIP_DST_ADDR].ptr; 843 else if (r->match_tag && !pf_match_tag(m, r, &tag, 844 pd->pf_mtag ? pd->pf_mtag->tag : 0)) 845 r = TAILQ_NEXT(r, entries); 846 else 847 break; 848 } 849 850 if (r == NULL || r->action == PF_NOSCRUB) 851 return (PF_PASS); 852 else { 853 r->packets[dir == PF_OUT]++; 854 r->bytes[dir == PF_OUT] += pd->tot_len; 855 } 856 857 /* Check for illegal packets */ 858 if (hlen < (int)sizeof(struct ip)) { 859 REASON_SET(reason, PFRES_NORM); 860 goto drop; 861 } 862 863 if (hlen > ntohs(h->ip_len)) { 864 REASON_SET(reason, PFRES_NORM); 865 goto drop; 866 } 867 868 /* Clear IP_DF if the rule uses the no-df option */ 869 if (r->rule_flag & PFRULE_NODF && h->ip_off & htons(IP_DF)) { 870 u_int16_t ip_off = h->ip_off; 871 872 h->ip_off &= htons(~IP_DF); 873 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0); 874 } 875 876 /* We will need other tests here */ 877 if (!fragoff && !mff) 878 goto no_fragment; 879 880 /* We're dealing with a fragment now. Don't allow fragments 881 * with IP_DF to enter the cache. If the flag was cleared by 882 * no-df above, fine. Otherwise drop it. 883 */ 884 if (h->ip_off & htons(IP_DF)) { 885 DPFPRINTF(("IP_DF\n")); 886 goto bad; 887 } 888 889 ip_len = ntohs(h->ip_len) - hlen; 890 ip_off = (ntohs(h->ip_off) & IP_OFFMASK) << 3; 891 892 /* All fragments are 8 byte aligned */ 893 if (mff && (ip_len & 0x7)) { 894 DPFPRINTF(("mff and %d\n", ip_len)); 895 goto bad; 896 } 897 898 /* Respect maximum length */ 899 if (fragoff + ip_len > IP_MAXPACKET) { 900 DPFPRINTF(("max packet %d\n", fragoff + ip_len)); 901 goto bad; 902 } 903 max = fragoff + ip_len; 904 905 /* Fully buffer all of the fragments 906 * Might return a completely reassembled mbuf, or NULL */ 907 PF_FRAG_LOCK(); 908 DPFPRINTF(("reass frag %d @ %d-%d\n", h->ip_id, fragoff, max)); 909 verdict = pf_reassemble(m0, h, dir, reason); 910 PF_FRAG_UNLOCK(); 911 912 if (verdict != PF_PASS) 913 return (PF_DROP); 914 915 m = *m0; 916 if (m == NULL) 917 return (PF_DROP); 918 919 h = mtod(m, struct ip *); 920 921 no_fragment: 922 /* At this point, only IP_DF is allowed in ip_off */ 923 if (h->ip_off & ~htons(IP_DF)) { 924 u_int16_t ip_off = h->ip_off; 925 926 h->ip_off &= htons(IP_DF); 927 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0); 928 } 929 930 pf_scrub_ip(&m, r->rule_flag, r->min_ttl, r->set_tos); 931 932 return (PF_PASS); 933 934 bad: 935 DPFPRINTF(("dropping bad fragment\n")); 936 REASON_SET(reason, PFRES_FRAG); 937 drop: 938 if (r != NULL && r->log) 939 PFLOG_PACKET(kif, m, AF_INET, dir, *reason, r, NULL, NULL, pd, 940 1); 941 942 return (PF_DROP); 943 } 944 #endif 945 946 #ifdef INET6 947 int 948 pf_normalize_ip6(struct mbuf **m0, int dir, struct pfi_kif *kif, 949 u_short *reason, struct pf_pdesc *pd) 950 { 951 struct mbuf *m = *m0; 952 struct pf_rule *r; 953 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 954 int extoff; 955 int off; 956 struct ip6_ext ext; 957 struct ip6_opt opt; 958 struct ip6_opt_jumbo jumbo; 959 struct ip6_frag frag; 960 u_int32_t jumbolen = 0, plen; 961 int optend; 962 int ooff; 963 u_int8_t proto; 964 int terminal; 965 966 PF_RULES_RASSERT(); 967 968 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr); 969 while (r != NULL) { 970 r->evaluations++; 971 if (pfi_kif_match(r->kif, kif) == r->ifnot) 972 r = r->skip[PF_SKIP_IFP].ptr; 973 else if (r->direction && r->direction != dir) 974 r = r->skip[PF_SKIP_DIR].ptr; 975 else if (r->af && r->af != AF_INET6) 976 r = r->skip[PF_SKIP_AF].ptr; 977 #if 0 /* header chain! */ 978 else if (r->proto && r->proto != h->ip6_nxt) 979 r = r->skip[PF_SKIP_PROTO].ptr; 980 #endif 981 else if (PF_MISMATCHAW(&r->src.addr, 982 (struct pf_addr *)&h->ip6_src, AF_INET6, 983 r->src.neg, kif, M_GETFIB(m))) 984 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 985 else if (PF_MISMATCHAW(&r->dst.addr, 986 (struct pf_addr *)&h->ip6_dst, AF_INET6, 987 r->dst.neg, NULL, M_GETFIB(m))) 988 r = r->skip[PF_SKIP_DST_ADDR].ptr; 989 else 990 break; 991 } 992 993 if (r == NULL || r->action == PF_NOSCRUB) 994 return (PF_PASS); 995 else { 996 r->packets[dir == PF_OUT]++; 997 r->bytes[dir == PF_OUT] += pd->tot_len; 998 } 999 1000 /* Check for illegal packets */ 1001 if (sizeof(struct ip6_hdr) + IPV6_MAXPACKET < m->m_pkthdr.len) 1002 goto drop; 1003 1004 extoff = 0; 1005 off = sizeof(struct ip6_hdr); 1006 proto = h->ip6_nxt; 1007 terminal = 0; 1008 do { 1009 switch (proto) { 1010 case IPPROTO_FRAGMENT: 1011 goto fragment; 1012 break; 1013 case IPPROTO_AH: 1014 case IPPROTO_ROUTING: 1015 case IPPROTO_DSTOPTS: 1016 if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL, 1017 NULL, AF_INET6)) 1018 goto shortpkt; 1019 extoff = off; 1020 if (proto == IPPROTO_AH) 1021 off += (ext.ip6e_len + 2) * 4; 1022 else 1023 off += (ext.ip6e_len + 1) * 8; 1024 proto = ext.ip6e_nxt; 1025 break; 1026 case IPPROTO_HOPOPTS: 1027 if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL, 1028 NULL, AF_INET6)) 1029 goto shortpkt; 1030 extoff = off; 1031 optend = off + (ext.ip6e_len + 1) * 8; 1032 ooff = off + sizeof(ext); 1033 do { 1034 if (!pf_pull_hdr(m, ooff, &opt.ip6o_type, 1035 sizeof(opt.ip6o_type), NULL, NULL, 1036 AF_INET6)) 1037 goto shortpkt; 1038 if (opt.ip6o_type == IP6OPT_PAD1) { 1039 ooff++; 1040 continue; 1041 } 1042 if (!pf_pull_hdr(m, ooff, &opt, sizeof(opt), 1043 NULL, NULL, AF_INET6)) 1044 goto shortpkt; 1045 if (ooff + sizeof(opt) + opt.ip6o_len > optend) 1046 goto drop; 1047 switch (opt.ip6o_type) { 1048 case IP6OPT_JUMBO: 1049 if (h->ip6_plen != 0) 1050 goto drop; 1051 if (!pf_pull_hdr(m, ooff, &jumbo, 1052 sizeof(jumbo), NULL, NULL, 1053 AF_INET6)) 1054 goto shortpkt; 1055 memcpy(&jumbolen, jumbo.ip6oj_jumbo_len, 1056 sizeof(jumbolen)); 1057 jumbolen = ntohl(jumbolen); 1058 if (jumbolen <= IPV6_MAXPACKET) 1059 goto drop; 1060 if (sizeof(struct ip6_hdr) + jumbolen != 1061 m->m_pkthdr.len) 1062 goto drop; 1063 break; 1064 default: 1065 break; 1066 } 1067 ooff += sizeof(opt) + opt.ip6o_len; 1068 } while (ooff < optend); 1069 1070 off = optend; 1071 proto = ext.ip6e_nxt; 1072 break; 1073 default: 1074 terminal = 1; 1075 break; 1076 } 1077 } while (!terminal); 1078 1079 /* jumbo payload option must be present, or plen > 0 */ 1080 if (ntohs(h->ip6_plen) == 0) 1081 plen = jumbolen; 1082 else 1083 plen = ntohs(h->ip6_plen); 1084 if (plen == 0) 1085 goto drop; 1086 if (sizeof(struct ip6_hdr) + plen > m->m_pkthdr.len) 1087 goto shortpkt; 1088 1089 pf_scrub_ip6(&m, r->min_ttl); 1090 1091 return (PF_PASS); 1092 1093 fragment: 1094 /* Jumbo payload packets cannot be fragmented. */ 1095 plen = ntohs(h->ip6_plen); 1096 if (plen == 0 || jumbolen) 1097 goto drop; 1098 if (sizeof(struct ip6_hdr) + plen > m->m_pkthdr.len) 1099 goto shortpkt; 1100 1101 if (!pf_pull_hdr(m, off, &frag, sizeof(frag), NULL, NULL, AF_INET6)) 1102 goto shortpkt; 1103 1104 /* Offset now points to data portion. */ 1105 off += sizeof(frag); 1106 1107 /* Returns PF_DROP or *m0 is NULL or completely reassembled mbuf. */ 1108 if (pf_reassemble6(m0, h, &frag, off, extoff, reason) != PF_PASS) 1109 return (PF_DROP); 1110 m = *m0; 1111 if (m == NULL) 1112 return (PF_DROP); 1113 1114 pd->flags |= PFDESC_IP_REAS; 1115 return (PF_PASS); 1116 1117 shortpkt: 1118 REASON_SET(reason, PFRES_SHORT); 1119 if (r != NULL && r->log) 1120 PFLOG_PACKET(kif, m, AF_INET6, dir, *reason, r, NULL, NULL, pd, 1121 1); 1122 return (PF_DROP); 1123 1124 drop: 1125 REASON_SET(reason, PFRES_NORM); 1126 if (r != NULL && r->log) 1127 PFLOG_PACKET(kif, m, AF_INET6, dir, *reason, r, NULL, NULL, pd, 1128 1); 1129 return (PF_DROP); 1130 } 1131 #endif /* INET6 */ 1132 1133 int 1134 pf_normalize_tcp(int dir, struct pfi_kif *kif, struct mbuf *m, int ipoff, 1135 int off, void *h, struct pf_pdesc *pd) 1136 { 1137 struct pf_rule *r, *rm = NULL; 1138 struct tcphdr *th = pd->hdr.tcp; 1139 int rewrite = 0; 1140 u_short reason; 1141 u_int8_t flags; 1142 sa_family_t af = pd->af; 1143 1144 PF_RULES_RASSERT(); 1145 1146 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr); 1147 while (r != NULL) { 1148 r->evaluations++; 1149 if (pfi_kif_match(r->kif, kif) == r->ifnot) 1150 r = r->skip[PF_SKIP_IFP].ptr; 1151 else if (r->direction && r->direction != dir) 1152 r = r->skip[PF_SKIP_DIR].ptr; 1153 else if (r->af && r->af != af) 1154 r = r->skip[PF_SKIP_AF].ptr; 1155 else if (r->proto && r->proto != pd->proto) 1156 r = r->skip[PF_SKIP_PROTO].ptr; 1157 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af, 1158 r->src.neg, kif, M_GETFIB(m))) 1159 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 1160 else if (r->src.port_op && !pf_match_port(r->src.port_op, 1161 r->src.port[0], r->src.port[1], th->th_sport)) 1162 r = r->skip[PF_SKIP_SRC_PORT].ptr; 1163 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af, 1164 r->dst.neg, NULL, M_GETFIB(m))) 1165 r = r->skip[PF_SKIP_DST_ADDR].ptr; 1166 else if (r->dst.port_op && !pf_match_port(r->dst.port_op, 1167 r->dst.port[0], r->dst.port[1], th->th_dport)) 1168 r = r->skip[PF_SKIP_DST_PORT].ptr; 1169 else if (r->os_fingerprint != PF_OSFP_ANY && !pf_osfp_match( 1170 pf_osfp_fingerprint(pd, m, off, th), 1171 r->os_fingerprint)) 1172 r = TAILQ_NEXT(r, entries); 1173 else { 1174 rm = r; 1175 break; 1176 } 1177 } 1178 1179 if (rm == NULL || rm->action == PF_NOSCRUB) 1180 return (PF_PASS); 1181 else { 1182 r->packets[dir == PF_OUT]++; 1183 r->bytes[dir == PF_OUT] += pd->tot_len; 1184 } 1185 1186 if (rm->rule_flag & PFRULE_REASSEMBLE_TCP) 1187 pd->flags |= PFDESC_TCP_NORM; 1188 1189 flags = th->th_flags; 1190 if (flags & TH_SYN) { 1191 /* Illegal packet */ 1192 if (flags & TH_RST) 1193 goto tcp_drop; 1194 1195 if (flags & TH_FIN) 1196 goto tcp_drop; 1197 } else { 1198 /* Illegal packet */ 1199 if (!(flags & (TH_ACK|TH_RST))) 1200 goto tcp_drop; 1201 } 1202 1203 if (!(flags & TH_ACK)) { 1204 /* These flags are only valid if ACK is set */ 1205 if ((flags & TH_FIN) || (flags & TH_PUSH) || (flags & TH_URG)) 1206 goto tcp_drop; 1207 } 1208 1209 /* Check for illegal header length */ 1210 if (th->th_off < (sizeof(struct tcphdr) >> 2)) 1211 goto tcp_drop; 1212 1213 /* If flags changed, or reserved data set, then adjust */ 1214 if (flags != th->th_flags || th->th_x2 != 0) { 1215 u_int16_t ov, nv; 1216 1217 ov = *(u_int16_t *)(&th->th_ack + 1); 1218 th->th_flags = flags; 1219 th->th_x2 = 0; 1220 nv = *(u_int16_t *)(&th->th_ack + 1); 1221 1222 th->th_sum = pf_proto_cksum_fixup(m, th->th_sum, ov, nv, 0); 1223 rewrite = 1; 1224 } 1225 1226 /* Remove urgent pointer, if TH_URG is not set */ 1227 if (!(flags & TH_URG) && th->th_urp) { 1228 th->th_sum = pf_proto_cksum_fixup(m, th->th_sum, th->th_urp, 1229 0, 0); 1230 th->th_urp = 0; 1231 rewrite = 1; 1232 } 1233 1234 /* Process options */ 1235 if (r->max_mss && pf_normalize_tcpopt(r, m, th, off, pd->af)) 1236 rewrite = 1; 1237 1238 /* copy back packet headers if we sanitized */ 1239 if (rewrite) 1240 m_copyback(m, off, sizeof(*th), (caddr_t)th); 1241 1242 return (PF_PASS); 1243 1244 tcp_drop: 1245 REASON_SET(&reason, PFRES_NORM); 1246 if (rm != NULL && r->log) 1247 PFLOG_PACKET(kif, m, AF_INET, dir, reason, r, NULL, NULL, pd, 1248 1); 1249 return (PF_DROP); 1250 } 1251 1252 int 1253 pf_normalize_tcp_init(struct mbuf *m, int off, struct pf_pdesc *pd, 1254 struct tcphdr *th, struct pf_state_peer *src, struct pf_state_peer *dst) 1255 { 1256 u_int32_t tsval, tsecr; 1257 u_int8_t hdr[60]; 1258 u_int8_t *opt; 1259 1260 KASSERT((src->scrub == NULL), 1261 ("pf_normalize_tcp_init: src->scrub != NULL")); 1262 1263 src->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT); 1264 if (src->scrub == NULL) 1265 return (1); 1266 1267 switch (pd->af) { 1268 #ifdef INET 1269 case AF_INET: { 1270 struct ip *h = mtod(m, struct ip *); 1271 src->scrub->pfss_ttl = h->ip_ttl; 1272 break; 1273 } 1274 #endif /* INET */ 1275 #ifdef INET6 1276 case AF_INET6: { 1277 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 1278 src->scrub->pfss_ttl = h->ip6_hlim; 1279 break; 1280 } 1281 #endif /* INET6 */ 1282 } 1283 1284 1285 /* 1286 * All normalizations below are only begun if we see the start of 1287 * the connections. They must all set an enabled bit in pfss_flags 1288 */ 1289 if ((th->th_flags & TH_SYN) == 0) 1290 return (0); 1291 1292 1293 if (th->th_off > (sizeof(struct tcphdr) >> 2) && src->scrub && 1294 pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) { 1295 /* Diddle with TCP options */ 1296 int hlen; 1297 opt = hdr + sizeof(struct tcphdr); 1298 hlen = (th->th_off << 2) - sizeof(struct tcphdr); 1299 while (hlen >= TCPOLEN_TIMESTAMP) { 1300 switch (*opt) { 1301 case TCPOPT_EOL: /* FALLTHROUGH */ 1302 case TCPOPT_NOP: 1303 opt++; 1304 hlen--; 1305 break; 1306 case TCPOPT_TIMESTAMP: 1307 if (opt[1] >= TCPOLEN_TIMESTAMP) { 1308 src->scrub->pfss_flags |= 1309 PFSS_TIMESTAMP; 1310 src->scrub->pfss_ts_mod = 1311 htonl(arc4random()); 1312 1313 /* note PFSS_PAWS not set yet */ 1314 memcpy(&tsval, &opt[2], 1315 sizeof(u_int32_t)); 1316 memcpy(&tsecr, &opt[6], 1317 sizeof(u_int32_t)); 1318 src->scrub->pfss_tsval0 = ntohl(tsval); 1319 src->scrub->pfss_tsval = ntohl(tsval); 1320 src->scrub->pfss_tsecr = ntohl(tsecr); 1321 getmicrouptime(&src->scrub->pfss_last); 1322 } 1323 /* FALLTHROUGH */ 1324 default: 1325 hlen -= MAX(opt[1], 2); 1326 opt += MAX(opt[1], 2); 1327 break; 1328 } 1329 } 1330 } 1331 1332 return (0); 1333 } 1334 1335 void 1336 pf_normalize_tcp_cleanup(struct pf_state *state) 1337 { 1338 if (state->src.scrub) 1339 uma_zfree(V_pf_state_scrub_z, state->src.scrub); 1340 if (state->dst.scrub) 1341 uma_zfree(V_pf_state_scrub_z, state->dst.scrub); 1342 1343 /* Someday... flush the TCP segment reassembly descriptors. */ 1344 } 1345 1346 int 1347 pf_normalize_tcp_stateful(struct mbuf *m, int off, struct pf_pdesc *pd, 1348 u_short *reason, struct tcphdr *th, struct pf_state *state, 1349 struct pf_state_peer *src, struct pf_state_peer *dst, int *writeback) 1350 { 1351 struct timeval uptime; 1352 u_int32_t tsval, tsecr; 1353 u_int tsval_from_last; 1354 u_int8_t hdr[60]; 1355 u_int8_t *opt; 1356 int copyback = 0; 1357 int got_ts = 0; 1358 1359 KASSERT((src->scrub || dst->scrub), 1360 ("%s: src->scrub && dst->scrub!", __func__)); 1361 1362 /* 1363 * Enforce the minimum TTL seen for this connection. Negate a common 1364 * technique to evade an intrusion detection system and confuse 1365 * firewall state code. 1366 */ 1367 switch (pd->af) { 1368 #ifdef INET 1369 case AF_INET: { 1370 if (src->scrub) { 1371 struct ip *h = mtod(m, struct ip *); 1372 if (h->ip_ttl > src->scrub->pfss_ttl) 1373 src->scrub->pfss_ttl = h->ip_ttl; 1374 h->ip_ttl = src->scrub->pfss_ttl; 1375 } 1376 break; 1377 } 1378 #endif /* INET */ 1379 #ifdef INET6 1380 case AF_INET6: { 1381 if (src->scrub) { 1382 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 1383 if (h->ip6_hlim > src->scrub->pfss_ttl) 1384 src->scrub->pfss_ttl = h->ip6_hlim; 1385 h->ip6_hlim = src->scrub->pfss_ttl; 1386 } 1387 break; 1388 } 1389 #endif /* INET6 */ 1390 } 1391 1392 if (th->th_off > (sizeof(struct tcphdr) >> 2) && 1393 ((src->scrub && (src->scrub->pfss_flags & PFSS_TIMESTAMP)) || 1394 (dst->scrub && (dst->scrub->pfss_flags & PFSS_TIMESTAMP))) && 1395 pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) { 1396 /* Diddle with TCP options */ 1397 int hlen; 1398 opt = hdr + sizeof(struct tcphdr); 1399 hlen = (th->th_off << 2) - sizeof(struct tcphdr); 1400 while (hlen >= TCPOLEN_TIMESTAMP) { 1401 switch (*opt) { 1402 case TCPOPT_EOL: /* FALLTHROUGH */ 1403 case TCPOPT_NOP: 1404 opt++; 1405 hlen--; 1406 break; 1407 case TCPOPT_TIMESTAMP: 1408 /* Modulate the timestamps. Can be used for 1409 * NAT detection, OS uptime determination or 1410 * reboot detection. 1411 */ 1412 1413 if (got_ts) { 1414 /* Huh? Multiple timestamps!? */ 1415 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1416 DPFPRINTF(("multiple TS??")); 1417 pf_print_state(state); 1418 printf("\n"); 1419 } 1420 REASON_SET(reason, PFRES_TS); 1421 return (PF_DROP); 1422 } 1423 if (opt[1] >= TCPOLEN_TIMESTAMP) { 1424 memcpy(&tsval, &opt[2], 1425 sizeof(u_int32_t)); 1426 if (tsval && src->scrub && 1427 (src->scrub->pfss_flags & 1428 PFSS_TIMESTAMP)) { 1429 tsval = ntohl(tsval); 1430 pf_change_proto_a(m, &opt[2], 1431 &th->th_sum, 1432 htonl(tsval + 1433 src->scrub->pfss_ts_mod), 1434 0); 1435 copyback = 1; 1436 } 1437 1438 /* Modulate TS reply iff valid (!0) */ 1439 memcpy(&tsecr, &opt[6], 1440 sizeof(u_int32_t)); 1441 if (tsecr && dst->scrub && 1442 (dst->scrub->pfss_flags & 1443 PFSS_TIMESTAMP)) { 1444 tsecr = ntohl(tsecr) 1445 - dst->scrub->pfss_ts_mod; 1446 pf_change_proto_a(m, &opt[6], 1447 &th->th_sum, htonl(tsecr), 1448 0); 1449 copyback = 1; 1450 } 1451 got_ts = 1; 1452 } 1453 /* FALLTHROUGH */ 1454 default: 1455 hlen -= MAX(opt[1], 2); 1456 opt += MAX(opt[1], 2); 1457 break; 1458 } 1459 } 1460 if (copyback) { 1461 /* Copyback the options, caller copys back header */ 1462 *writeback = 1; 1463 m_copyback(m, off + sizeof(struct tcphdr), 1464 (th->th_off << 2) - sizeof(struct tcphdr), hdr + 1465 sizeof(struct tcphdr)); 1466 } 1467 } 1468 1469 1470 /* 1471 * Must invalidate PAWS checks on connections idle for too long. 1472 * The fastest allowed timestamp clock is 1ms. That turns out to 1473 * be about 24 days before it wraps. XXX Right now our lowerbound 1474 * TS echo check only works for the first 12 days of a connection 1475 * when the TS has exhausted half its 32bit space 1476 */ 1477 #define TS_MAX_IDLE (24*24*60*60) 1478 #define TS_MAX_CONN (12*24*60*60) /* XXX remove when better tsecr check */ 1479 1480 getmicrouptime(&uptime); 1481 if (src->scrub && (src->scrub->pfss_flags & PFSS_PAWS) && 1482 (uptime.tv_sec - src->scrub->pfss_last.tv_sec > TS_MAX_IDLE || 1483 time_uptime - state->creation > TS_MAX_CONN)) { 1484 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1485 DPFPRINTF(("src idled out of PAWS\n")); 1486 pf_print_state(state); 1487 printf("\n"); 1488 } 1489 src->scrub->pfss_flags = (src->scrub->pfss_flags & ~PFSS_PAWS) 1490 | PFSS_PAWS_IDLED; 1491 } 1492 if (dst->scrub && (dst->scrub->pfss_flags & PFSS_PAWS) && 1493 uptime.tv_sec - dst->scrub->pfss_last.tv_sec > TS_MAX_IDLE) { 1494 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1495 DPFPRINTF(("dst idled out of PAWS\n")); 1496 pf_print_state(state); 1497 printf("\n"); 1498 } 1499 dst->scrub->pfss_flags = (dst->scrub->pfss_flags & ~PFSS_PAWS) 1500 | PFSS_PAWS_IDLED; 1501 } 1502 1503 if (got_ts && src->scrub && dst->scrub && 1504 (src->scrub->pfss_flags & PFSS_PAWS) && 1505 (dst->scrub->pfss_flags & PFSS_PAWS)) { 1506 /* Validate that the timestamps are "in-window". 1507 * RFC1323 describes TCP Timestamp options that allow 1508 * measurement of RTT (round trip time) and PAWS 1509 * (protection against wrapped sequence numbers). PAWS 1510 * gives us a set of rules for rejecting packets on 1511 * long fat pipes (packets that were somehow delayed 1512 * in transit longer than the time it took to send the 1513 * full TCP sequence space of 4Gb). We can use these 1514 * rules and infer a few others that will let us treat 1515 * the 32bit timestamp and the 32bit echoed timestamp 1516 * as sequence numbers to prevent a blind attacker from 1517 * inserting packets into a connection. 1518 * 1519 * RFC1323 tells us: 1520 * - The timestamp on this packet must be greater than 1521 * or equal to the last value echoed by the other 1522 * endpoint. The RFC says those will be discarded 1523 * since it is a dup that has already been acked. 1524 * This gives us a lowerbound on the timestamp. 1525 * timestamp >= other last echoed timestamp 1526 * - The timestamp will be less than or equal to 1527 * the last timestamp plus the time between the 1528 * last packet and now. The RFC defines the max 1529 * clock rate as 1ms. We will allow clocks to be 1530 * up to 10% fast and will allow a total difference 1531 * or 30 seconds due to a route change. And this 1532 * gives us an upperbound on the timestamp. 1533 * timestamp <= last timestamp + max ticks 1534 * We have to be careful here. Windows will send an 1535 * initial timestamp of zero and then initialize it 1536 * to a random value after the 3whs; presumably to 1537 * avoid a DoS by having to call an expensive RNG 1538 * during a SYN flood. Proof MS has at least one 1539 * good security geek. 1540 * 1541 * - The TCP timestamp option must also echo the other 1542 * endpoints timestamp. The timestamp echoed is the 1543 * one carried on the earliest unacknowledged segment 1544 * on the left edge of the sequence window. The RFC 1545 * states that the host will reject any echoed 1546 * timestamps that were larger than any ever sent. 1547 * This gives us an upperbound on the TS echo. 1548 * tescr <= largest_tsval 1549 * - The lowerbound on the TS echo is a little more 1550 * tricky to determine. The other endpoint's echoed 1551 * values will not decrease. But there may be 1552 * network conditions that re-order packets and 1553 * cause our view of them to decrease. For now the 1554 * only lowerbound we can safely determine is that 1555 * the TS echo will never be less than the original 1556 * TS. XXX There is probably a better lowerbound. 1557 * Remove TS_MAX_CONN with better lowerbound check. 1558 * tescr >= other original TS 1559 * 1560 * It is also important to note that the fastest 1561 * timestamp clock of 1ms will wrap its 32bit space in 1562 * 24 days. So we just disable TS checking after 24 1563 * days of idle time. We actually must use a 12d 1564 * connection limit until we can come up with a better 1565 * lowerbound to the TS echo check. 1566 */ 1567 struct timeval delta_ts; 1568 int ts_fudge; 1569 1570 1571 /* 1572 * PFTM_TS_DIFF is how many seconds of leeway to allow 1573 * a host's timestamp. This can happen if the previous 1574 * packet got delayed in transit for much longer than 1575 * this packet. 1576 */ 1577 if ((ts_fudge = state->rule.ptr->timeout[PFTM_TS_DIFF]) == 0) 1578 ts_fudge = V_pf_default_rule.timeout[PFTM_TS_DIFF]; 1579 1580 /* Calculate max ticks since the last timestamp */ 1581 #define TS_MAXFREQ 1100 /* RFC max TS freq of 1Khz + 10% skew */ 1582 #define TS_MICROSECS 1000000 /* microseconds per second */ 1583 delta_ts = uptime; 1584 timevalsub(&delta_ts, &src->scrub->pfss_last); 1585 tsval_from_last = (delta_ts.tv_sec + ts_fudge) * TS_MAXFREQ; 1586 tsval_from_last += delta_ts.tv_usec / (TS_MICROSECS/TS_MAXFREQ); 1587 1588 if ((src->state >= TCPS_ESTABLISHED && 1589 dst->state >= TCPS_ESTABLISHED) && 1590 (SEQ_LT(tsval, dst->scrub->pfss_tsecr) || 1591 SEQ_GT(tsval, src->scrub->pfss_tsval + tsval_from_last) || 1592 (tsecr && (SEQ_GT(tsecr, dst->scrub->pfss_tsval) || 1593 SEQ_LT(tsecr, dst->scrub->pfss_tsval0))))) { 1594 /* Bad RFC1323 implementation or an insertion attack. 1595 * 1596 * - Solaris 2.6 and 2.7 are known to send another ACK 1597 * after the FIN,FIN|ACK,ACK closing that carries 1598 * an old timestamp. 1599 */ 1600 1601 DPFPRINTF(("Timestamp failed %c%c%c%c\n", 1602 SEQ_LT(tsval, dst->scrub->pfss_tsecr) ? '0' : ' ', 1603 SEQ_GT(tsval, src->scrub->pfss_tsval + 1604 tsval_from_last) ? '1' : ' ', 1605 SEQ_GT(tsecr, dst->scrub->pfss_tsval) ? '2' : ' ', 1606 SEQ_LT(tsecr, dst->scrub->pfss_tsval0)? '3' : ' ')); 1607 DPFPRINTF((" tsval: %u tsecr: %u +ticks: %u " 1608 "idle: %jus %lums\n", 1609 tsval, tsecr, tsval_from_last, 1610 (uintmax_t)delta_ts.tv_sec, 1611 delta_ts.tv_usec / 1000)); 1612 DPFPRINTF((" src->tsval: %u tsecr: %u\n", 1613 src->scrub->pfss_tsval, src->scrub->pfss_tsecr)); 1614 DPFPRINTF((" dst->tsval: %u tsecr: %u tsval0: %u" 1615 "\n", dst->scrub->pfss_tsval, 1616 dst->scrub->pfss_tsecr, dst->scrub->pfss_tsval0)); 1617 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1618 pf_print_state(state); 1619 pf_print_flags(th->th_flags); 1620 printf("\n"); 1621 } 1622 REASON_SET(reason, PFRES_TS); 1623 return (PF_DROP); 1624 } 1625 1626 /* XXX I'd really like to require tsecr but it's optional */ 1627 1628 } else if (!got_ts && (th->th_flags & TH_RST) == 0 && 1629 ((src->state == TCPS_ESTABLISHED && dst->state == TCPS_ESTABLISHED) 1630 || pd->p_len > 0 || (th->th_flags & TH_SYN)) && 1631 src->scrub && dst->scrub && 1632 (src->scrub->pfss_flags & PFSS_PAWS) && 1633 (dst->scrub->pfss_flags & PFSS_PAWS)) { 1634 /* Didn't send a timestamp. Timestamps aren't really useful 1635 * when: 1636 * - connection opening or closing (often not even sent). 1637 * but we must not let an attacker to put a FIN on a 1638 * data packet to sneak it through our ESTABLISHED check. 1639 * - on a TCP reset. RFC suggests not even looking at TS. 1640 * - on an empty ACK. The TS will not be echoed so it will 1641 * probably not help keep the RTT calculation in sync and 1642 * there isn't as much danger when the sequence numbers 1643 * got wrapped. So some stacks don't include TS on empty 1644 * ACKs :-( 1645 * 1646 * To minimize the disruption to mostly RFC1323 conformant 1647 * stacks, we will only require timestamps on data packets. 1648 * 1649 * And what do ya know, we cannot require timestamps on data 1650 * packets. There appear to be devices that do legitimate 1651 * TCP connection hijacking. There are HTTP devices that allow 1652 * a 3whs (with timestamps) and then buffer the HTTP request. 1653 * If the intermediate device has the HTTP response cache, it 1654 * will spoof the response but not bother timestamping its 1655 * packets. So we can look for the presence of a timestamp in 1656 * the first data packet and if there, require it in all future 1657 * packets. 1658 */ 1659 1660 if (pd->p_len > 0 && (src->scrub->pfss_flags & PFSS_DATA_TS)) { 1661 /* 1662 * Hey! Someone tried to sneak a packet in. Or the 1663 * stack changed its RFC1323 behavior?!?! 1664 */ 1665 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1666 DPFPRINTF(("Did not receive expected RFC1323 " 1667 "timestamp\n")); 1668 pf_print_state(state); 1669 pf_print_flags(th->th_flags); 1670 printf("\n"); 1671 } 1672 REASON_SET(reason, PFRES_TS); 1673 return (PF_DROP); 1674 } 1675 } 1676 1677 1678 /* 1679 * We will note if a host sends his data packets with or without 1680 * timestamps. And require all data packets to contain a timestamp 1681 * if the first does. PAWS implicitly requires that all data packets be 1682 * timestamped. But I think there are middle-man devices that hijack 1683 * TCP streams immediately after the 3whs and don't timestamp their 1684 * packets (seen in a WWW accelerator or cache). 1685 */ 1686 if (pd->p_len > 0 && src->scrub && (src->scrub->pfss_flags & 1687 (PFSS_TIMESTAMP|PFSS_DATA_TS|PFSS_DATA_NOTS)) == PFSS_TIMESTAMP) { 1688 if (got_ts) 1689 src->scrub->pfss_flags |= PFSS_DATA_TS; 1690 else { 1691 src->scrub->pfss_flags |= PFSS_DATA_NOTS; 1692 if (V_pf_status.debug >= PF_DEBUG_MISC && dst->scrub && 1693 (dst->scrub->pfss_flags & PFSS_TIMESTAMP)) { 1694 /* Don't warn if other host rejected RFC1323 */ 1695 DPFPRINTF(("Broken RFC1323 stack did not " 1696 "timestamp data packet. Disabled PAWS " 1697 "security.\n")); 1698 pf_print_state(state); 1699 pf_print_flags(th->th_flags); 1700 printf("\n"); 1701 } 1702 } 1703 } 1704 1705 1706 /* 1707 * Update PAWS values 1708 */ 1709 if (got_ts && src->scrub && PFSS_TIMESTAMP == (src->scrub->pfss_flags & 1710 (PFSS_PAWS_IDLED|PFSS_TIMESTAMP))) { 1711 getmicrouptime(&src->scrub->pfss_last); 1712 if (SEQ_GEQ(tsval, src->scrub->pfss_tsval) || 1713 (src->scrub->pfss_flags & PFSS_PAWS) == 0) 1714 src->scrub->pfss_tsval = tsval; 1715 1716 if (tsecr) { 1717 if (SEQ_GEQ(tsecr, src->scrub->pfss_tsecr) || 1718 (src->scrub->pfss_flags & PFSS_PAWS) == 0) 1719 src->scrub->pfss_tsecr = tsecr; 1720 1721 if ((src->scrub->pfss_flags & PFSS_PAWS) == 0 && 1722 (SEQ_LT(tsval, src->scrub->pfss_tsval0) || 1723 src->scrub->pfss_tsval0 == 0)) { 1724 /* tsval0 MUST be the lowest timestamp */ 1725 src->scrub->pfss_tsval0 = tsval; 1726 } 1727 1728 /* Only fully initialized after a TS gets echoed */ 1729 if ((src->scrub->pfss_flags & PFSS_PAWS) == 0) 1730 src->scrub->pfss_flags |= PFSS_PAWS; 1731 } 1732 } 1733 1734 /* I have a dream.... TCP segment reassembly.... */ 1735 return (0); 1736 } 1737 1738 static int 1739 pf_normalize_tcpopt(struct pf_rule *r, struct mbuf *m, struct tcphdr *th, 1740 int off, sa_family_t af) 1741 { 1742 u_int16_t *mss; 1743 int thoff; 1744 int opt, cnt, optlen = 0; 1745 int rewrite = 0; 1746 u_char opts[TCP_MAXOLEN]; 1747 u_char *optp = opts; 1748 1749 thoff = th->th_off << 2; 1750 cnt = thoff - sizeof(struct tcphdr); 1751 1752 if (cnt > 0 && !pf_pull_hdr(m, off + sizeof(*th), opts, cnt, 1753 NULL, NULL, af)) 1754 return (rewrite); 1755 1756 for (; cnt > 0; cnt -= optlen, optp += optlen) { 1757 opt = optp[0]; 1758 if (opt == TCPOPT_EOL) 1759 break; 1760 if (opt == TCPOPT_NOP) 1761 optlen = 1; 1762 else { 1763 if (cnt < 2) 1764 break; 1765 optlen = optp[1]; 1766 if (optlen < 2 || optlen > cnt) 1767 break; 1768 } 1769 switch (opt) { 1770 case TCPOPT_MAXSEG: 1771 mss = (u_int16_t *)(optp + 2); 1772 if ((ntohs(*mss)) > r->max_mss) { 1773 th->th_sum = pf_proto_cksum_fixup(m, 1774 th->th_sum, *mss, htons(r->max_mss), 0); 1775 *mss = htons(r->max_mss); 1776 rewrite = 1; 1777 } 1778 break; 1779 default: 1780 break; 1781 } 1782 } 1783 1784 if (rewrite) 1785 m_copyback(m, off + sizeof(*th), thoff - sizeof(*th), opts); 1786 1787 return (rewrite); 1788 } 1789 1790 #ifdef INET 1791 static void 1792 pf_scrub_ip(struct mbuf **m0, u_int32_t flags, u_int8_t min_ttl, u_int8_t tos) 1793 { 1794 struct mbuf *m = *m0; 1795 struct ip *h = mtod(m, struct ip *); 1796 1797 /* Clear IP_DF if no-df was requested */ 1798 if (flags & PFRULE_NODF && h->ip_off & htons(IP_DF)) { 1799 u_int16_t ip_off = h->ip_off; 1800 1801 h->ip_off &= htons(~IP_DF); 1802 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0); 1803 } 1804 1805 /* Enforce a minimum ttl, may cause endless packet loops */ 1806 if (min_ttl && h->ip_ttl < min_ttl) { 1807 u_int16_t ip_ttl = h->ip_ttl; 1808 1809 h->ip_ttl = min_ttl; 1810 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_ttl, h->ip_ttl, 0); 1811 } 1812 1813 /* Enforce tos */ 1814 if (flags & PFRULE_SET_TOS) { 1815 u_int16_t ov, nv; 1816 1817 ov = *(u_int16_t *)h; 1818 h->ip_tos = tos | (h->ip_tos & IPTOS_ECN_MASK); 1819 nv = *(u_int16_t *)h; 1820 1821 h->ip_sum = pf_cksum_fixup(h->ip_sum, ov, nv, 0); 1822 } 1823 1824 /* random-id, but not for fragments */ 1825 if (flags & PFRULE_RANDOMID && !(h->ip_off & ~htons(IP_DF))) { 1826 uint16_t ip_id = h->ip_id; 1827 1828 ip_fillid(h); 1829 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_id, h->ip_id, 0); 1830 } 1831 } 1832 #endif /* INET */ 1833 1834 #ifdef INET6 1835 static void 1836 pf_scrub_ip6(struct mbuf **m0, u_int8_t min_ttl) 1837 { 1838 struct mbuf *m = *m0; 1839 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 1840 1841 /* Enforce a minimum ttl, may cause endless packet loops */ 1842 if (min_ttl && h->ip6_hlim < min_ttl) 1843 h->ip6_hlim = min_ttl; 1844 } 1845 #endif 1846