1 /* $OpenBSD: pf_norm.c,v 1.113 2008/05/07 07:07:29 markus Exp $ */ 2 3 /* 4 * Copyright (c) 2010 The DragonFly Project. All rights reserved. 5 * 6 * Copyright 2001 Niels Provos <provos@citi.umich.edu> 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 * 18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 19 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 20 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 21 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 22 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 23 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 24 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 25 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 26 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 27 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 28 */ 29 30 #include "opt_inet.h" 31 #include "opt_inet6.h" 32 33 #include <sys/param.h> 34 #include <sys/systm.h> 35 #include <sys/mbuf.h> 36 #include <sys/filio.h> 37 #include <sys/fcntl.h> 38 #include <sys/socket.h> 39 #include <sys/kernel.h> 40 #include <sys/time.h> 41 #include <vm/vm_zone.h> 42 43 #include <net/if.h> 44 #include <net/if_types.h> 45 #include <net/bpf.h> 46 #include <net/route.h> 47 #include <net/pf/if_pflog.h> 48 49 #include <netinet/in.h> 50 #include <netinet/in_var.h> 51 #include <netinet/in_systm.h> 52 #include <netinet/ip.h> 53 #include <netinet/ip_var.h> 54 #include <netinet/tcp.h> 55 #include <netinet/tcp_seq.h> 56 #include <netinet/udp.h> 57 #include <netinet/ip_icmp.h> 58 59 #ifdef INET6 60 #include <netinet/ip6.h> 61 #endif /* INET6 */ 62 63 #include <net/pf/pfvar.h> 64 65 #define PFFRAG_SEENLAST 0x0001 /* Seen the last fragment for this */ 66 #define PFFRAG_NOBUFFER 0x0002 /* Non-buffering fragment cache */ 67 #define PFFRAG_DROP 0x0004 /* Drop all fragments */ 68 #define BUFFER_FRAGMENTS(fr) (!((fr)->fr_flags & PFFRAG_NOBUFFER)) 69 70 71 TAILQ_HEAD(pf_fragqueue, pf_fragment) pf_fragqueue; 72 TAILQ_HEAD(pf_cachequeue, pf_fragment) pf_cachequeue; 73 74 static __inline int pf_frag_compare(struct pf_fragment *, 75 struct pf_fragment *); 76 RB_HEAD(pf_frag_tree, pf_fragment) pf_frag_tree, pf_cache_tree; 77 RB_PROTOTYPE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare); 78 RB_GENERATE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare); 79 80 /* Private prototypes */ 81 void pf_ip2key(struct pf_fragment *, struct ip *); 82 void pf_remove_fragment(struct pf_fragment *); 83 void pf_flush_fragments(void); 84 void pf_free_fragment(struct pf_fragment *); 85 struct pf_fragment *pf_find_fragment(struct ip *, struct pf_frag_tree *); 86 struct mbuf *pf_reassemble(struct mbuf **, struct pf_fragment **, 87 struct pf_frent *, int); 88 struct mbuf *pf_fragcache(struct mbuf **, struct ip*, 89 struct pf_fragment **, int, int, int *); 90 int pf_normalize_tcpopt(struct pf_rule *, struct mbuf *, 91 struct tcphdr *, int, sa_family_t); 92 93 #define DPFPRINTF(x) do { \ 94 if (pf_status.debug >= PF_DEBUG_MISC) { \ 95 kprintf("%s: ", __func__); \ 96 kprintf x ; \ 97 } \ 98 } while(0) 99 100 /* Globals */ 101 vm_zone_t pf_frent_pl, pf_frag_pl, pf_cache_pl, pf_cent_pl; 102 vm_zone_t pf_state_scrub_pl; 103 int pf_nfrents, pf_ncache; 104 105 void 106 pf_normalize_init(void) 107 { 108 /* XXX 109 pool_sethiwat(&pf_frag_pl, PFFRAG_FRAG_HIWAT); 110 pool_sethardlimit(&pf_frent_pl, PFFRAG_FRENT_HIWAT, NULL, 0); 111 pool_sethardlimit(&pf_cache_pl, PFFRAG_FRCACHE_HIWAT, NULL, 0); 112 pool_sethardlimit(&pf_cent_pl, PFFRAG_FRCENT_HIWAT, NULL, 0); 113 */ 114 115 TAILQ_INIT(&pf_fragqueue); 116 TAILQ_INIT(&pf_cachequeue); 117 } 118 119 static __inline int 120 pf_frag_compare(struct pf_fragment *a, struct pf_fragment *b) 121 { 122 int diff; 123 124 if ((diff = a->fr_id - b->fr_id)) 125 return (diff); 126 else if ((diff = a->fr_p - b->fr_p)) 127 return (diff); 128 else if (a->fr_src.s_addr < b->fr_src.s_addr) 129 return (-1); 130 else if (a->fr_src.s_addr > b->fr_src.s_addr) 131 return (1); 132 else if (a->fr_dst.s_addr < b->fr_dst.s_addr) 133 return (-1); 134 else if (a->fr_dst.s_addr > b->fr_dst.s_addr) 135 return (1); 136 return (0); 137 } 138 139 void 140 pf_purge_expired_fragments(void) 141 { 142 struct pf_fragment *frag; 143 u_int32_t expire = time_second - 144 pf_default_rule.timeout[PFTM_FRAG]; 145 146 while ((frag = TAILQ_LAST(&pf_fragqueue, pf_fragqueue)) != NULL) { 147 KASSERT((BUFFER_FRAGMENTS(frag)), 148 ("BUFFER_FRAGMENTS(frag) == 0: %s", __func__)); 149 if (frag->fr_timeout > expire) 150 break; 151 152 DPFPRINTF(("expiring %d(%p)\n", frag->fr_id, frag)); 153 pf_free_fragment(frag); 154 } 155 156 while ((frag = TAILQ_LAST(&pf_cachequeue, pf_cachequeue)) != NULL) { 157 KASSERT((!BUFFER_FRAGMENTS(frag)), 158 ("BUFFER_FRAGMENTS(frag) != 0: %s", __func__)); 159 if (frag->fr_timeout > expire) 160 break; 161 162 DPFPRINTF(("expiring %d(%p)\n", frag->fr_id, frag)); 163 pf_free_fragment(frag); 164 KASSERT((TAILQ_EMPTY(&pf_cachequeue) || 165 TAILQ_LAST(&pf_cachequeue, pf_cachequeue) != frag), 166 ("!(TAILQ_EMPTY() || TAILQ_LAST() == farg): %s", 167 __func__)); 168 } 169 } 170 171 /* 172 * Try to flush old fragments to make space for new ones 173 */ 174 175 void 176 pf_flush_fragments(void) 177 { 178 struct pf_fragment *frag; 179 int goal; 180 181 goal = pf_nfrents * 9 / 10; 182 DPFPRINTF(("trying to free > %d frents\n", 183 pf_nfrents - goal)); 184 while (goal < pf_nfrents) { 185 frag = TAILQ_LAST(&pf_fragqueue, pf_fragqueue); 186 if (frag == NULL) 187 break; 188 pf_free_fragment(frag); 189 } 190 191 192 goal = pf_ncache * 9 / 10; 193 DPFPRINTF(("trying to free > %d cache entries\n", 194 pf_ncache - goal)); 195 while (goal < pf_ncache) { 196 frag = TAILQ_LAST(&pf_cachequeue, pf_cachequeue); 197 if (frag == NULL) 198 break; 199 pf_free_fragment(frag); 200 } 201 } 202 203 /* Frees the fragments and all associated entries */ 204 205 void 206 pf_free_fragment(struct pf_fragment *frag) 207 { 208 struct pf_frent *frent; 209 struct pf_frcache *frcache; 210 211 /* Free all fragments */ 212 if (BUFFER_FRAGMENTS(frag)) { 213 for (frent = LIST_FIRST(&frag->fr_queue); frent; 214 frent = LIST_FIRST(&frag->fr_queue)) { 215 LIST_REMOVE(frent, fr_next); 216 217 m_freem(frent->fr_m); 218 pool_put(&pf_frent_pl, frent); 219 pf_nfrents--; 220 } 221 } else { 222 for (frcache = LIST_FIRST(&frag->fr_cache); frcache; 223 frcache = LIST_FIRST(&frag->fr_cache)) { 224 LIST_REMOVE(frcache, fr_next); 225 226 KASSERT((LIST_EMPTY(&frag->fr_cache) || 227 LIST_FIRST(&frag->fr_cache)->fr_off > 228 frcache->fr_end), 229 ("! (LIST_EMPTY() || LIST_FIRST()->fr_off >" 230 " frcache->fr_end): %s", __func__)); 231 232 pool_put(&pf_cent_pl, frcache); 233 pf_ncache--; 234 } 235 } 236 237 pf_remove_fragment(frag); 238 } 239 240 void 241 pf_ip2key(struct pf_fragment *key, struct ip *ip) 242 { 243 key->fr_p = ip->ip_p; 244 key->fr_id = ip->ip_id; 245 key->fr_src.s_addr = ip->ip_src.s_addr; 246 key->fr_dst.s_addr = ip->ip_dst.s_addr; 247 } 248 249 struct pf_fragment * 250 pf_find_fragment(struct ip *ip, struct pf_frag_tree *tree) 251 { 252 struct pf_fragment key; 253 struct pf_fragment *frag; 254 255 pf_ip2key(&key, ip); 256 257 frag = RB_FIND(pf_frag_tree, tree, &key); 258 if (frag != NULL) { 259 /* XXX Are we sure we want to update the timeout? */ 260 frag->fr_timeout = time_second; 261 if (BUFFER_FRAGMENTS(frag)) { 262 TAILQ_REMOVE(&pf_fragqueue, frag, frag_next); 263 TAILQ_INSERT_HEAD(&pf_fragqueue, frag, frag_next); 264 } else { 265 TAILQ_REMOVE(&pf_cachequeue, frag, frag_next); 266 TAILQ_INSERT_HEAD(&pf_cachequeue, frag, frag_next); 267 } 268 } 269 270 return (frag); 271 } 272 273 /* Removes a fragment from the fragment queue and frees the fragment */ 274 275 void 276 pf_remove_fragment(struct pf_fragment *frag) 277 { 278 if (BUFFER_FRAGMENTS(frag)) { 279 RB_REMOVE(pf_frag_tree, &pf_frag_tree, frag); 280 TAILQ_REMOVE(&pf_fragqueue, frag, frag_next); 281 pool_put(&pf_frag_pl, frag); 282 } else { 283 RB_REMOVE(pf_frag_tree, &pf_cache_tree, frag); 284 TAILQ_REMOVE(&pf_cachequeue, frag, frag_next); 285 pool_put(&pf_cache_pl, frag); 286 } 287 } 288 289 #define FR_IP_OFF(fr) (((fr)->fr_ip->ip_off & IP_OFFMASK) << 3) 290 struct mbuf * 291 pf_reassemble(struct mbuf **m0, struct pf_fragment **frag, 292 struct pf_frent *frent, int mff) 293 { 294 struct mbuf *m = *m0, *m2; 295 struct pf_frent *frea, *next; 296 struct pf_frent *frep = NULL; 297 struct ip *ip = frent->fr_ip; 298 int hlen = ip->ip_hl << 2; 299 u_int16_t off = (ip->ip_off & IP_OFFMASK) << 3; 300 u_int16_t ip_len = ip->ip_len - ip->ip_hl * 4; 301 u_int16_t max = ip_len + off; 302 303 KASSERT((*frag == NULL || BUFFER_FRAGMENTS(*frag)), 304 ("! (*frag == NULL || BUFFER_FRAGMENTS(*frag)): %s", __func__)); 305 306 /* Strip off ip header */ 307 m->m_data += hlen; 308 m->m_len -= hlen; 309 310 /* Create a new reassembly queue for this packet */ 311 if (*frag == NULL) { 312 *frag = pool_get(&pf_frag_pl, PR_NOWAIT); 313 if (*frag == NULL) { 314 pf_flush_fragments(); 315 *frag = pool_get(&pf_frag_pl, PR_NOWAIT); 316 if (*frag == NULL) 317 goto drop_fragment; 318 } 319 320 (*frag)->fr_flags = 0; 321 (*frag)->fr_max = 0; 322 (*frag)->fr_src = frent->fr_ip->ip_src; 323 (*frag)->fr_dst = frent->fr_ip->ip_dst; 324 (*frag)->fr_p = frent->fr_ip->ip_p; 325 (*frag)->fr_id = frent->fr_ip->ip_id; 326 (*frag)->fr_timeout = time_second; 327 LIST_INIT(&(*frag)->fr_queue); 328 329 RB_INSERT(pf_frag_tree, &pf_frag_tree, *frag); 330 TAILQ_INSERT_HEAD(&pf_fragqueue, *frag, frag_next); 331 332 /* We do not have a previous fragment */ 333 frep = NULL; 334 goto insert; 335 } 336 337 /* 338 * Find a fragment after the current one: 339 * - off contains the real shifted offset. 340 */ 341 LIST_FOREACH(frea, &(*frag)->fr_queue, fr_next) { 342 if (FR_IP_OFF(frea) > off) 343 break; 344 frep = frea; 345 } 346 347 KASSERT((frep != NULL || frea != NULL), 348 ("!(frep != NULL || frea != NULL): %s", __func__)); 349 350 if (frep != NULL && 351 FR_IP_OFF(frep) + frep->fr_ip->ip_len - frep->fr_ip->ip_hl * 352 4 > off) 353 { 354 u_int16_t precut; 355 356 precut = FR_IP_OFF(frep) + frep->fr_ip->ip_len - 357 frep->fr_ip->ip_hl * 4 - off; 358 if (precut >= ip_len) 359 goto drop_fragment; 360 m_adj(frent->fr_m, precut); 361 DPFPRINTF(("overlap -%d\n", precut)); 362 /* Enforce 8 byte boundaries */ 363 ip->ip_off = ip->ip_off + (precut >> 3); 364 off = (ip->ip_off & IP_OFFMASK) << 3; 365 ip_len -= precut; 366 ip->ip_len = ip_len; 367 } 368 369 for (; frea != NULL && ip_len + off > FR_IP_OFF(frea); 370 frea = next) 371 { 372 u_int16_t aftercut; 373 374 aftercut = ip_len + off - FR_IP_OFF(frea); 375 DPFPRINTF(("adjust overlap %d\n", aftercut)); 376 if (aftercut < frea->fr_ip->ip_len - frea->fr_ip->ip_hl 377 * 4) 378 { 379 frea->fr_ip->ip_len = 380 frea->fr_ip->ip_len - aftercut; 381 frea->fr_ip->ip_off = frea->fr_ip->ip_off + 382 (aftercut >> 3); 383 m_adj(frea->fr_m, aftercut); 384 break; 385 } 386 387 /* This fragment is completely overlapped, lose it */ 388 next = LIST_NEXT(frea, fr_next); 389 m_freem(frea->fr_m); 390 LIST_REMOVE(frea, fr_next); 391 pool_put(&pf_frent_pl, frea); 392 pf_nfrents--; 393 } 394 395 insert: 396 /* Update maximum data size */ 397 if ((*frag)->fr_max < max) 398 (*frag)->fr_max = max; 399 /* This is the last segment */ 400 if (!mff) 401 (*frag)->fr_flags |= PFFRAG_SEENLAST; 402 403 if (frep == NULL) 404 LIST_INSERT_HEAD(&(*frag)->fr_queue, frent, fr_next); 405 else 406 LIST_INSERT_AFTER(frep, frent, fr_next); 407 408 /* Check if we are completely reassembled */ 409 if (!((*frag)->fr_flags & PFFRAG_SEENLAST)) 410 return (NULL); 411 412 /* Check if we have all the data */ 413 off = 0; 414 for (frep = LIST_FIRST(&(*frag)->fr_queue); frep; frep = next) { 415 next = LIST_NEXT(frep, fr_next); 416 417 off += frep->fr_ip->ip_len - frep->fr_ip->ip_hl * 4; 418 if (off < (*frag)->fr_max && 419 (next == NULL || FR_IP_OFF(next) != off)) 420 { 421 DPFPRINTF(("missing fragment at %d, next %d, max %d\n", 422 off, next == NULL ? -1 : FR_IP_OFF(next), 423 (*frag)->fr_max)); 424 return (NULL); 425 } 426 } 427 DPFPRINTF(("%d < %d?\n", off, (*frag)->fr_max)); 428 if (off < (*frag)->fr_max) 429 return (NULL); 430 431 /* We have all the data */ 432 frent = LIST_FIRST(&(*frag)->fr_queue); 433 KASSERT((frent != NULL), ("frent == NULL: %s", __func__)); 434 if ((frent->fr_ip->ip_hl << 2) + off > IP_MAXPACKET) { 435 DPFPRINTF(("drop: too big: %d\n", off)); 436 pf_free_fragment(*frag); 437 *frag = NULL; 438 return (NULL); 439 } 440 next = LIST_NEXT(frent, fr_next); 441 442 /* Magic from ip_input */ 443 ip = frent->fr_ip; 444 m = frent->fr_m; 445 m2 = m->m_next; 446 m->m_next = NULL; 447 m_cat(m, m2); 448 pool_put(&pf_frent_pl, frent); 449 pf_nfrents--; 450 for (frent = next; frent != NULL; frent = next) { 451 next = LIST_NEXT(frent, fr_next); 452 453 m2 = frent->fr_m; 454 pool_put(&pf_frent_pl, frent); 455 pf_nfrents--; 456 m_cat(m, m2); 457 } 458 459 ip->ip_src = (*frag)->fr_src; 460 ip->ip_dst = (*frag)->fr_dst; 461 462 /* Remove from fragment queue */ 463 pf_remove_fragment(*frag); 464 *frag = NULL; 465 466 hlen = ip->ip_hl << 2; 467 ip->ip_len = off + hlen; 468 m->m_len += hlen; 469 m->m_data -= hlen; 470 471 /* some debugging cruft by sklower, below, will go away soon */ 472 /* XXX this should be done elsewhere */ 473 if (m->m_flags & M_PKTHDR) { 474 int plen = 0; 475 for (m2 = m; m2; m2 = m2->m_next) 476 plen += m2->m_len; 477 m->m_pkthdr.len = plen; 478 } 479 480 DPFPRINTF(("complete: %p(%d)\n", m, ip->ip_len)); 481 return (m); 482 483 drop_fragment: 484 /* Oops - fail safe - drop packet */ 485 pool_put(&pf_frent_pl, frent); 486 pf_nfrents--; 487 m_freem(m); 488 return (NULL); 489 } 490 491 struct mbuf * 492 pf_fragcache(struct mbuf **m0, struct ip *h, struct pf_fragment **frag, int mff, 493 int drop, int *nomem) 494 { 495 struct mbuf *m = *m0; 496 struct pf_frcache *frp, *fra, *cur = NULL; 497 int ip_len = h->ip_len - (h->ip_hl << 2); 498 u_int16_t off = h->ip_off << 3; 499 u_int16_t max = ip_len + off; 500 int hosed = 0; 501 502 KASSERT((*frag == NULL || !BUFFER_FRAGMENTS(*frag)), 503 ("!(*frag == NULL || !BUFFER_FRAGMENTS(*frag)): %s", __func__)); 504 505 /* Create a new range queue for this packet */ 506 if (*frag == NULL) { 507 *frag = pool_get(&pf_cache_pl, PR_NOWAIT); 508 if (*frag == NULL) { 509 pf_flush_fragments(); 510 *frag = pool_get(&pf_cache_pl, PR_NOWAIT); 511 if (*frag == NULL) 512 goto no_mem; 513 } 514 515 /* Get an entry for the queue */ 516 cur = pool_get(&pf_cent_pl, PR_NOWAIT); 517 if (cur == NULL) { 518 pool_put(&pf_cache_pl, *frag); 519 *frag = NULL; 520 goto no_mem; 521 } 522 pf_ncache++; 523 524 (*frag)->fr_flags = PFFRAG_NOBUFFER; 525 (*frag)->fr_max = 0; 526 (*frag)->fr_src = h->ip_src; 527 (*frag)->fr_dst = h->ip_dst; 528 (*frag)->fr_p = h->ip_p; 529 (*frag)->fr_id = h->ip_id; 530 (*frag)->fr_timeout = time_second; 531 532 cur->fr_off = off; 533 cur->fr_end = max; 534 LIST_INIT(&(*frag)->fr_cache); 535 LIST_INSERT_HEAD(&(*frag)->fr_cache, cur, fr_next); 536 537 RB_INSERT(pf_frag_tree, &pf_cache_tree, *frag); 538 TAILQ_INSERT_HEAD(&pf_cachequeue, *frag, frag_next); 539 540 DPFPRINTF(("fragcache[%d]: new %d-%d\n", h->ip_id, off, max)); 541 542 goto pass; 543 } 544 545 /* 546 * Find a fragment after the current one: 547 * - off contains the real shifted offset. 548 */ 549 frp = NULL; 550 LIST_FOREACH(fra, &(*frag)->fr_cache, fr_next) { 551 if (fra->fr_off > off) 552 break; 553 frp = fra; 554 } 555 556 KASSERT((frp != NULL || fra != NULL), 557 ("!(frp != NULL || fra != NULL): %s", __func__)); 558 559 if (frp != NULL) { 560 int precut; 561 562 precut = frp->fr_end - off; 563 if (precut >= ip_len) { 564 /* Fragment is entirely a duplicate */ 565 DPFPRINTF(("fragcache[%d]: dead (%d-%d) %d-%d\n", 566 h->ip_id, frp->fr_off, frp->fr_end, off, max)); 567 goto drop_fragment; 568 } 569 if (precut == 0) { 570 /* They are adjacent. Fixup cache entry */ 571 DPFPRINTF(("fragcache[%d]: adjacent (%d-%d) %d-%d\n", 572 h->ip_id, frp->fr_off, frp->fr_end, off, max)); 573 frp->fr_end = max; 574 } else if (precut > 0) { 575 /* The first part of this payload overlaps with a 576 * fragment that has already been passed. 577 * Need to trim off the first part of the payload. 578 * But to do so easily, we need to create another 579 * mbuf to throw the original header into. 580 */ 581 582 DPFPRINTF(("fragcache[%d]: chop %d (%d-%d) %d-%d\n", 583 h->ip_id, precut, frp->fr_off, frp->fr_end, off, 584 max)); 585 586 off += precut; 587 max -= precut; 588 /* Update the previous frag to encompass this one */ 589 frp->fr_end = max; 590 591 if (!drop) { 592 /* XXX Optimization opportunity 593 * This is a very heavy way to trim the payload. 594 * we could do it much faster by diddling mbuf 595 * internals but that would be even less legible 596 * than this mbuf magic. For my next trick, 597 * I'll pull a rabbit out of my laptop. 598 */ 599 *m0 = m_dup(m, MB_DONTWAIT); 600 /* From KAME Project : We have missed this! */ 601 m_adj(*m0, (h->ip_hl << 2) - 602 (*m0)->m_pkthdr.len); 603 if (*m0 == NULL) 604 goto no_mem; 605 KASSERT(((*m0)->m_next == NULL), 606 ("(*m0)->m_next != NULL: %s", 607 __func__)); 608 m_adj(m, precut + (h->ip_hl << 2)); 609 m_cat(*m0, m); 610 m = *m0; 611 if (m->m_flags & M_PKTHDR) { 612 int plen = 0; 613 struct mbuf *t; 614 for (t = m; t; t = t->m_next) 615 plen += t->m_len; 616 m->m_pkthdr.len = plen; 617 } 618 619 620 h = mtod(m, struct ip *); 621 622 KASSERT(((int)m->m_len == 623 h->ip_len - precut), 624 ("m->m_len != h->ip_len - precut: %s", 625 __func__)); 626 h->ip_off = h->ip_off + 627 (precut >> 3); 628 h->ip_len = h->ip_len - precut; 629 } else { 630 hosed++; 631 } 632 } else { 633 /* There is a gap between fragments */ 634 635 DPFPRINTF(("fragcache[%d]: gap %d (%d-%d) %d-%d\n", 636 h->ip_id, -precut, frp->fr_off, frp->fr_end, off, 637 max)); 638 639 cur = pool_get(&pf_cent_pl, PR_NOWAIT); 640 if (cur == NULL) 641 goto no_mem; 642 pf_ncache++; 643 644 cur->fr_off = off; 645 cur->fr_end = max; 646 LIST_INSERT_AFTER(frp, cur, fr_next); 647 } 648 } 649 650 if (fra != NULL) { 651 int aftercut; 652 int merge = 0; 653 654 aftercut = max - fra->fr_off; 655 if (aftercut == 0) { 656 /* Adjacent fragments */ 657 DPFPRINTF(("fragcache[%d]: adjacent %d-%d (%d-%d)\n", 658 h->ip_id, off, max, fra->fr_off, fra->fr_end)); 659 fra->fr_off = off; 660 merge = 1; 661 } else if (aftercut > 0) { 662 /* Need to chop off the tail of this fragment */ 663 DPFPRINTF(("fragcache[%d]: chop %d %d-%d (%d-%d)\n", 664 h->ip_id, aftercut, off, max, fra->fr_off, 665 fra->fr_end)); 666 fra->fr_off = off; 667 max -= aftercut; 668 669 merge = 1; 670 671 if (!drop) { 672 m_adj(m, -aftercut); 673 if (m->m_flags & M_PKTHDR) { 674 int plen = 0; 675 struct mbuf *t; 676 for (t = m; t; t = t->m_next) 677 plen += t->m_len; 678 m->m_pkthdr.len = plen; 679 } 680 h = mtod(m, struct ip *); 681 KASSERT(((int)m->m_len == h->ip_len - aftercut), 682 ("m->m_len != h->ip_len - aftercut: %s", 683 __func__)); 684 h->ip_len = h->ip_len - aftercut; 685 } else { 686 hosed++; 687 } 688 } else if (frp == NULL) { 689 /* There is a gap between fragments */ 690 DPFPRINTF(("fragcache[%d]: gap %d %d-%d (%d-%d)\n", 691 h->ip_id, -aftercut, off, max, fra->fr_off, 692 fra->fr_end)); 693 694 cur = pool_get(&pf_cent_pl, PR_NOWAIT); 695 if (cur == NULL) 696 goto no_mem; 697 pf_ncache++; 698 699 cur->fr_off = off; 700 cur->fr_end = max; 701 LIST_INSERT_BEFORE(fra, cur, fr_next); 702 } 703 704 705 /* Need to glue together two separate fragment descriptors */ 706 if (merge) { 707 if (cur && fra->fr_off <= cur->fr_end) { 708 /* Need to merge in a previous 'cur' */ 709 DPFPRINTF(("fragcache[%d]: adjacent(merge " 710 "%d-%d) %d-%d (%d-%d)\n", 711 h->ip_id, cur->fr_off, cur->fr_end, off, 712 max, fra->fr_off, fra->fr_end)); 713 fra->fr_off = cur->fr_off; 714 LIST_REMOVE(cur, fr_next); 715 pool_put(&pf_cent_pl, cur); 716 pf_ncache--; 717 cur = NULL; 718 719 } else if (frp && fra->fr_off <= frp->fr_end) { 720 /* Need to merge in a modified 'frp' */ 721 KASSERT((cur == NULL), ("cur != NULL: %s", 722 __func__)); 723 DPFPRINTF(("fragcache[%d]: adjacent(merge " 724 "%d-%d) %d-%d (%d-%d)\n", 725 h->ip_id, frp->fr_off, frp->fr_end, off, 726 max, fra->fr_off, fra->fr_end)); 727 fra->fr_off = frp->fr_off; 728 LIST_REMOVE(frp, fr_next); 729 pool_put(&pf_cent_pl, frp); 730 pf_ncache--; 731 frp = NULL; 732 733 } 734 } 735 } 736 737 if (hosed) { 738 /* 739 * We must keep tracking the overall fragment even when 740 * we're going to drop it anyway so that we know when to 741 * free the overall descriptor. Thus we drop the frag late. 742 */ 743 goto drop_fragment; 744 } 745 746 747 pass: 748 /* Update maximum data size */ 749 if ((*frag)->fr_max < max) 750 (*frag)->fr_max = max; 751 752 /* This is the last segment */ 753 if (!mff) 754 (*frag)->fr_flags |= PFFRAG_SEENLAST; 755 756 /* Check if we are completely reassembled */ 757 if (((*frag)->fr_flags & PFFRAG_SEENLAST) && 758 LIST_FIRST(&(*frag)->fr_cache)->fr_off == 0 && 759 LIST_FIRST(&(*frag)->fr_cache)->fr_end == (*frag)->fr_max) { 760 /* Remove from fragment queue */ 761 DPFPRINTF(("fragcache[%d]: done 0-%d\n", h->ip_id, 762 (*frag)->fr_max)); 763 pf_free_fragment(*frag); 764 *frag = NULL; 765 } 766 767 return (m); 768 769 no_mem: 770 *nomem = 1; 771 772 /* Still need to pay attention to !IP_MF */ 773 if (!mff && *frag != NULL) 774 (*frag)->fr_flags |= PFFRAG_SEENLAST; 775 776 m_freem(m); 777 return (NULL); 778 779 drop_fragment: 780 781 /* Still need to pay attention to !IP_MF */ 782 if (!mff && *frag != NULL) 783 (*frag)->fr_flags |= PFFRAG_SEENLAST; 784 785 if (drop) { 786 /* This fragment has been deemed bad. Don't reass */ 787 if (((*frag)->fr_flags & PFFRAG_DROP) == 0) 788 DPFPRINTF(("fragcache[%d]: dropping overall fragment\n", 789 h->ip_id)); 790 (*frag)->fr_flags |= PFFRAG_DROP; 791 } 792 793 m_freem(m); 794 return (NULL); 795 } 796 797 int 798 pf_normalize_ip(struct mbuf **m0, int dir, struct pfi_kif *kif, u_short *reason, 799 struct pf_pdesc *pd) 800 { 801 struct mbuf *m = *m0; 802 struct pf_rule *r; 803 struct pf_frent *frent; 804 struct pf_fragment *frag = NULL; 805 struct ip *h = mtod(m, struct ip *); 806 int mff = (h->ip_off & IP_MF); 807 int hlen = h->ip_hl << 2; 808 u_int16_t fragoff = (h->ip_off & IP_OFFMASK) << 3; 809 u_int16_t max; 810 int ip_len; 811 int ip_off; 812 int tag = -1; 813 814 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr); 815 while (r != NULL) { 816 r->evaluations++; 817 if (pfi_kif_match(r->kif, kif) == r->ifnot) 818 r = r->skip[PF_SKIP_IFP].ptr; 819 else if (r->direction && r->direction != dir) 820 r = r->skip[PF_SKIP_DIR].ptr; 821 else if (r->af && r->af != AF_INET) 822 r = r->skip[PF_SKIP_AF].ptr; 823 else if (r->proto && r->proto != h->ip_p) 824 r = r->skip[PF_SKIP_PROTO].ptr; 825 else if (PF_MISMATCHAW(&r->src.addr, 826 (struct pf_addr *)&h->ip_src.s_addr, AF_INET, 827 r->src.neg, kif)) 828 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 829 else if (PF_MISMATCHAW(&r->dst.addr, 830 (struct pf_addr *)&h->ip_dst.s_addr, AF_INET, 831 r->dst.neg, NULL)) 832 r = r->skip[PF_SKIP_DST_ADDR].ptr; 833 else if (r->match_tag && !pf_match_tag(m, r, &tag)) 834 r = TAILQ_NEXT(r, entries); 835 else 836 break; 837 } 838 839 if (r == NULL || r->action == PF_NOSCRUB) 840 return (PF_PASS); 841 else { 842 r->packets[dir == PF_OUT]++; 843 r->bytes[dir == PF_OUT] += pd->tot_len; 844 } 845 846 /* Check for illegal packets */ 847 if (hlen < (int)sizeof(struct ip)) 848 goto drop; 849 850 if (hlen > h->ip_len) 851 goto drop; 852 853 /* Clear IP_DF if the rule uses the no-df option */ 854 if (r->rule_flag & PFRULE_NODF && h->ip_off & IP_DF) { 855 u_int16_t ip_off = h->ip_off; 856 857 h->ip_off &= ~IP_DF; 858 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0); 859 } 860 861 /* We will need other tests here */ 862 if (!fragoff && !mff) 863 goto no_fragment; 864 865 /* We're dealing with a fragment now. Don't allow fragments 866 * with IP_DF to enter the cache. If the flag was cleared by 867 * no-df above, fine. Otherwise drop it. 868 */ 869 if (h->ip_off & IP_DF) { 870 DPFPRINTF(("IP_DF\n")); 871 goto bad; 872 } 873 874 ip_len = h->ip_len - hlen; 875 ip_off = (h->ip_off & IP_OFFMASK) << 3; 876 877 /* All fragments are 8 byte aligned */ 878 if (mff && (ip_len & 0x7)) { 879 DPFPRINTF(("mff and %d\n", ip_len)); 880 goto bad; 881 } 882 883 /* Respect maximum length */ 884 if (fragoff + ip_len > IP_MAXPACKET) { 885 DPFPRINTF(("max packet %d\n", fragoff + ip_len)); 886 goto bad; 887 } 888 max = fragoff + ip_len; 889 890 if ((r->rule_flag & (PFRULE_FRAGCROP|PFRULE_FRAGDROP)) == 0) { 891 /* Fully buffer all of the fragments */ 892 893 frag = pf_find_fragment(h, &pf_frag_tree); 894 895 /* Check if we saw the last fragment already */ 896 if (frag != NULL && (frag->fr_flags & PFFRAG_SEENLAST) && 897 max > frag->fr_max) 898 goto bad; 899 900 /* Get an entry for the fragment queue */ 901 frent = pool_get(&pf_frent_pl, PR_NOWAIT); 902 if (frent == NULL) { 903 REASON_SET(reason, PFRES_MEMORY); 904 return (PF_DROP); 905 } 906 pf_nfrents++; 907 frent->fr_ip = h; 908 frent->fr_m = m; 909 910 /* Might return a completely reassembled mbuf, or NULL */ 911 DPFPRINTF(("reass frag %d @ %d-%d\n", h->ip_id, fragoff, max)); 912 *m0 = m = pf_reassemble(m0, &frag, frent, mff); 913 914 if (m == NULL) 915 return (PF_DROP); 916 917 if (frag != NULL && (frag->fr_flags & PFFRAG_DROP)) 918 goto drop; 919 920 h = mtod(m, struct ip *); 921 } else { 922 /* non-buffering fragment cache (drops or masks overlaps) */ 923 int nomem = 0; 924 925 if (dir == PF_OUT && m->m_pkthdr.pf.flags & PF_TAG_FRAGCACHE) { 926 /* 927 * Already passed the fragment cache in the 928 * input direction. If we continued, it would 929 * appear to be a dup and would be dropped. 930 */ 931 goto fragment_pass; 932 } 933 934 frag = pf_find_fragment(h, &pf_cache_tree); 935 936 /* Check if we saw the last fragment already */ 937 if (frag != NULL && (frag->fr_flags & PFFRAG_SEENLAST) && 938 max > frag->fr_max) { 939 if (r->rule_flag & PFRULE_FRAGDROP) 940 frag->fr_flags |= PFFRAG_DROP; 941 goto bad; 942 } 943 944 *m0 = m = pf_fragcache(m0, h, &frag, mff, 945 (r->rule_flag & PFRULE_FRAGDROP) ? 1 : 0, &nomem); 946 if (m == NULL) { 947 if (nomem) 948 goto no_mem; 949 goto drop; 950 } 951 952 if (dir == PF_IN) 953 m->m_pkthdr.pf.flags |= PF_TAG_FRAGCACHE; 954 955 if (frag != NULL && (frag->fr_flags & PFFRAG_DROP)) 956 goto drop; 957 goto fragment_pass; 958 } 959 960 no_fragment: 961 /* At this point, only IP_DF is allowed in ip_off */ 962 if (h->ip_off & ~IP_DF) { 963 u_int16_t ip_off = h->ip_off; 964 965 h->ip_off &= IP_DF; 966 h->ip_sum = pf_cksum_fixup(h->ip_sum, htons(ip_off), htons(h->ip_off), 0); 967 } 968 969 /* Enforce a minimum ttl, may cause endless packet loops */ 970 if (r->min_ttl && h->ip_ttl < r->min_ttl) { 971 u_int16_t ip_ttl = h->ip_ttl; 972 973 h->ip_ttl = r->min_ttl; 974 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_ttl, h->ip_ttl, 0); 975 } 976 977 /* Enforce tos */ 978 if (r->rule_flag & PFRULE_SET_TOS) { 979 u_int16_t ov, nv; 980 981 ov = *(u_int16_t *)h; 982 h->ip_tos = r->set_tos; 983 nv = *(u_int16_t *)h; 984 985 h->ip_sum = pf_cksum_fixup(h->ip_sum, ov, nv, 0); 986 } 987 988 if (r->rule_flag & PFRULE_RANDOMID) { 989 u_int16_t ip_id = h->ip_id; 990 991 h->ip_id = ip_randomid(); 992 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_id, h->ip_id, 0); 993 } 994 if ((r->rule_flag & (PFRULE_FRAGCROP|PFRULE_FRAGDROP)) == 0) 995 pd->flags |= PFDESC_IP_REAS; 996 997 return (PF_PASS); 998 999 fragment_pass: 1000 /* Enforce a minimum ttl, may cause endless packet loops */ 1001 if (r->min_ttl && h->ip_ttl < r->min_ttl) { 1002 u_int16_t ip_ttl = h->ip_ttl; 1003 1004 h->ip_ttl = r->min_ttl; 1005 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_ttl, h->ip_ttl, 0); 1006 } 1007 /* Enforce tos */ 1008 if (r->rule_flag & PFRULE_SET_TOS) { 1009 u_int16_t ov, nv; 1010 1011 ov = *(u_int16_t *)h; 1012 h->ip_tos = r->set_tos; 1013 nv = *(u_int16_t *)h; 1014 1015 h->ip_sum = pf_cksum_fixup(h->ip_sum, ov, nv, 0); 1016 } 1017 if ((r->rule_flag & (PFRULE_FRAGCROP|PFRULE_FRAGDROP)) == 0) 1018 pd->flags |= PFDESC_IP_REAS; 1019 return (PF_PASS); 1020 1021 no_mem: 1022 REASON_SET(reason, PFRES_MEMORY); 1023 if (r != NULL && r->log) 1024 PFLOG_PACKET(kif, h, m, AF_INET, dir, *reason, r, NULL, NULL, pd); 1025 return (PF_DROP); 1026 1027 drop: 1028 REASON_SET(reason, PFRES_NORM); 1029 if (r != NULL && r->log) 1030 PFLOG_PACKET(kif, h, m, AF_INET, dir, *reason, r, NULL, NULL, pd); 1031 return (PF_DROP); 1032 1033 bad: 1034 DPFPRINTF(("dropping bad fragment\n")); 1035 1036 /* Free associated fragments */ 1037 if (frag != NULL) 1038 pf_free_fragment(frag); 1039 1040 REASON_SET(reason, PFRES_FRAG); 1041 if (r != NULL && r->log) 1042 PFLOG_PACKET(kif, h, m, AF_INET, dir, *reason, r, NULL, NULL, pd); 1043 1044 return (PF_DROP); 1045 } 1046 1047 #ifdef INET6 1048 int 1049 pf_normalize_ip6(struct mbuf **m0, int dir, struct pfi_kif *kif, 1050 u_short *reason, struct pf_pdesc *pd) 1051 { 1052 struct mbuf *m = *m0; 1053 struct pf_rule *r; 1054 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 1055 int off; 1056 struct ip6_ext ext; 1057 struct ip6_opt opt; 1058 struct ip6_opt_jumbo jumbo; 1059 struct ip6_frag frag; 1060 u_int32_t jumbolen = 0, plen; 1061 u_int16_t fragoff = 0; 1062 int optend; 1063 int ooff; 1064 u_int8_t proto; 1065 int terminal; 1066 1067 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr); 1068 while (r != NULL) { 1069 r->evaluations++; 1070 if (pfi_kif_match(r->kif, kif) == r->ifnot) 1071 r = r->skip[PF_SKIP_IFP].ptr; 1072 else if (r->direction && r->direction != dir) 1073 r = r->skip[PF_SKIP_DIR].ptr; 1074 else if (r->af && r->af != AF_INET6) 1075 r = r->skip[PF_SKIP_AF].ptr; 1076 #if 0 /* header chain! */ 1077 else if (r->proto && r->proto != h->ip6_nxt) 1078 r = r->skip[PF_SKIP_PROTO].ptr; 1079 #endif 1080 else if (PF_MISMATCHAW(&r->src.addr, 1081 (struct pf_addr *)&h->ip6_src, AF_INET6, 1082 r->src.neg, kif)) 1083 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 1084 else if (PF_MISMATCHAW(&r->dst.addr, 1085 (struct pf_addr *)&h->ip6_dst, AF_INET6, 1086 r->dst.neg, NULL)) 1087 r = r->skip[PF_SKIP_DST_ADDR].ptr; 1088 else 1089 break; 1090 } 1091 1092 if (r == NULL || r->action == PF_NOSCRUB) 1093 return (PF_PASS); 1094 else { 1095 r->packets[dir == PF_OUT]++; 1096 r->bytes[dir == PF_OUT] += pd->tot_len; 1097 } 1098 1099 /* Check for illegal packets */ 1100 if (sizeof(struct ip6_hdr) + IPV6_MAXPACKET < m->m_pkthdr.len) 1101 goto drop; 1102 1103 off = sizeof(struct ip6_hdr); 1104 proto = h->ip6_nxt; 1105 terminal = 0; 1106 do { 1107 switch (proto) { 1108 case IPPROTO_FRAGMENT: 1109 goto fragment; 1110 break; 1111 case IPPROTO_AH: 1112 case IPPROTO_ROUTING: 1113 case IPPROTO_DSTOPTS: 1114 if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL, 1115 NULL, AF_INET6)) 1116 goto shortpkt; 1117 if (proto == IPPROTO_AH) 1118 off += (ext.ip6e_len + 2) * 4; 1119 else 1120 off += (ext.ip6e_len + 1) * 8; 1121 proto = ext.ip6e_nxt; 1122 break; 1123 case IPPROTO_HOPOPTS: 1124 if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL, 1125 NULL, AF_INET6)) 1126 goto shortpkt; 1127 optend = off + (ext.ip6e_len + 1) * 8; 1128 ooff = off + sizeof(ext); 1129 do { 1130 if (!pf_pull_hdr(m, ooff, &opt.ip6o_type, 1131 sizeof(opt.ip6o_type), NULL, NULL, 1132 AF_INET6)) 1133 goto shortpkt; 1134 if (opt.ip6o_type == IP6OPT_PAD1) { 1135 ooff++; 1136 continue; 1137 } 1138 if (!pf_pull_hdr(m, ooff, &opt, sizeof(opt), 1139 NULL, NULL, AF_INET6)) 1140 goto shortpkt; 1141 if (ooff + sizeof(opt) + opt.ip6o_len > optend) 1142 goto drop; 1143 switch (opt.ip6o_type) { 1144 case IP6OPT_JUMBO: 1145 if (h->ip6_plen != 0) 1146 goto drop; 1147 if (!pf_pull_hdr(m, ooff, &jumbo, 1148 sizeof(jumbo), NULL, NULL, 1149 AF_INET6)) 1150 goto shortpkt; 1151 memcpy(&jumbolen, jumbo.ip6oj_jumbo_len, 1152 sizeof(jumbolen)); 1153 jumbolen = ntohl(jumbolen); 1154 if (jumbolen <= IPV6_MAXPACKET) 1155 goto drop; 1156 if (sizeof(struct ip6_hdr) + jumbolen != 1157 m->m_pkthdr.len) 1158 goto drop; 1159 break; 1160 default: 1161 break; 1162 } 1163 ooff += sizeof(opt) + opt.ip6o_len; 1164 } while (ooff < optend); 1165 1166 off = optend; 1167 proto = ext.ip6e_nxt; 1168 break; 1169 default: 1170 terminal = 1; 1171 break; 1172 } 1173 } while (!terminal); 1174 1175 /* jumbo payload option must be present, or plen > 0 */ 1176 if (ntohs(h->ip6_plen) == 0) 1177 plen = jumbolen; 1178 else 1179 plen = ntohs(h->ip6_plen); 1180 if (plen == 0) 1181 goto drop; 1182 if (sizeof(struct ip6_hdr) + plen > m->m_pkthdr.len) 1183 goto shortpkt; 1184 1185 /* Enforce a minimum ttl, may cause endless packet loops */ 1186 if (r->min_ttl && h->ip6_hlim < r->min_ttl) 1187 h->ip6_hlim = r->min_ttl; 1188 1189 return (PF_PASS); 1190 1191 fragment: 1192 if (ntohs(h->ip6_plen) == 0 || jumbolen) 1193 goto drop; 1194 plen = ntohs(h->ip6_plen); 1195 1196 if (!pf_pull_hdr(m, off, &frag, sizeof(frag), NULL, NULL, AF_INET6)) 1197 goto shortpkt; 1198 fragoff = ntohs(frag.ip6f_offlg & IP6F_OFF_MASK); 1199 if (fragoff + (plen - off - sizeof(frag)) > IPV6_MAXPACKET) 1200 goto badfrag; 1201 1202 /* do something about it */ 1203 /* remember to set pd->flags |= PFDESC_IP_REAS */ 1204 return (PF_PASS); 1205 1206 shortpkt: 1207 REASON_SET(reason, PFRES_SHORT); 1208 if (r != NULL && r->log) 1209 PFLOG_PACKET(kif, h, m, AF_INET6, dir, *reason, r, NULL, NULL, pd); 1210 return (PF_DROP); 1211 1212 drop: 1213 REASON_SET(reason, PFRES_NORM); 1214 if (r != NULL && r->log) 1215 PFLOG_PACKET(kif, h, m, AF_INET6, dir, *reason, r, NULL, NULL, pd); 1216 return (PF_DROP); 1217 1218 badfrag: 1219 REASON_SET(reason, PFRES_FRAG); 1220 if (r != NULL && r->log) 1221 PFLOG_PACKET(kif, h, m, AF_INET6, dir, *reason, r, NULL, NULL, pd); 1222 return (PF_DROP); 1223 } 1224 #endif /* INET6 */ 1225 1226 int 1227 pf_normalize_tcp(int dir, struct pfi_kif *kif, struct mbuf *m, int ipoff, 1228 int off, void *h, struct pf_pdesc *pd) 1229 { 1230 struct pf_rule *r, *rm = NULL; 1231 struct tcphdr *th = pd->hdr.tcp; 1232 int rewrite = 0; 1233 u_short reason; 1234 u_int8_t flags; 1235 sa_family_t af = pd->af; 1236 1237 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr); 1238 while (r != NULL) { 1239 r->evaluations++; 1240 if (pfi_kif_match(r->kif, kif) == r->ifnot) 1241 r = r->skip[PF_SKIP_IFP].ptr; 1242 else if (r->direction && r->direction != dir) 1243 r = r->skip[PF_SKIP_DIR].ptr; 1244 else if (r->af && r->af != af) 1245 r = r->skip[PF_SKIP_AF].ptr; 1246 else if (r->proto && r->proto != pd->proto) 1247 r = r->skip[PF_SKIP_PROTO].ptr; 1248 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af, 1249 r->src.neg, kif)) 1250 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 1251 else if (r->src.port_op && !pf_match_port(r->src.port_op, 1252 r->src.port[0], r->src.port[1], th->th_sport)) 1253 r = r->skip[PF_SKIP_SRC_PORT].ptr; 1254 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af, 1255 r->dst.neg, NULL)) 1256 r = r->skip[PF_SKIP_DST_ADDR].ptr; 1257 else if (r->dst.port_op && !pf_match_port(r->dst.port_op, 1258 r->dst.port[0], r->dst.port[1], th->th_dport)) 1259 r = r->skip[PF_SKIP_DST_PORT].ptr; 1260 else if (r->os_fingerprint != PF_OSFP_ANY && !pf_osfp_match( 1261 pf_osfp_fingerprint(pd, m, off, th), 1262 r->os_fingerprint)) 1263 r = TAILQ_NEXT(r, entries); 1264 else { 1265 rm = r; 1266 break; 1267 } 1268 } 1269 1270 if (rm == NULL || rm->action == PF_NOSCRUB) 1271 return (PF_PASS); 1272 else { 1273 r->packets[dir == PF_OUT]++; 1274 r->bytes[dir == PF_OUT] += pd->tot_len; 1275 } 1276 1277 if (rm->rule_flag & PFRULE_REASSEMBLE_TCP) 1278 pd->flags |= PFDESC_TCP_NORM; 1279 1280 flags = th->th_flags; 1281 if (flags & TH_SYN) { 1282 /* Illegal packet */ 1283 if (flags & TH_RST) 1284 goto tcp_drop; 1285 1286 if (flags & TH_FIN) 1287 flags &= ~TH_FIN; 1288 } else { 1289 /* Illegal packet */ 1290 if (!(flags & (TH_ACK|TH_RST))) 1291 goto tcp_drop; 1292 } 1293 1294 if (!(flags & TH_ACK)) { 1295 /* These flags are only valid if ACK is set */ 1296 if ((flags & TH_FIN) || (flags & TH_PUSH) || (flags & TH_URG)) 1297 goto tcp_drop; 1298 } 1299 1300 /* Check for illegal header length */ 1301 if (th->th_off < (sizeof(struct tcphdr) >> 2)) 1302 goto tcp_drop; 1303 1304 /* If flags changed, or reserved data set, then adjust */ 1305 if (flags != th->th_flags || th->th_x2 != 0) { 1306 u_int16_t ov, nv; 1307 1308 ov = *(u_int16_t *)(&th->th_ack + 1); 1309 th->th_flags = flags; 1310 th->th_x2 = 0; 1311 nv = *(u_int16_t *)(&th->th_ack + 1); 1312 1313 th->th_sum = pf_cksum_fixup(th->th_sum, ov, nv, 0); 1314 rewrite = 1; 1315 } 1316 1317 /* Remove urgent pointer, if TH_URG is not set */ 1318 if (!(flags & TH_URG) && th->th_urp) { 1319 th->th_sum = pf_cksum_fixup(th->th_sum, th->th_urp, 0, 0); 1320 th->th_urp = 0; 1321 rewrite = 1; 1322 } 1323 1324 /* Process options */ 1325 if (r->max_mss && pf_normalize_tcpopt(r, m, th, off, pd->af)) 1326 rewrite = 1; 1327 1328 /* copy back packet headers if we sanitized */ 1329 if (rewrite) 1330 m_copyback(m, off, sizeof(*th), (caddr_t)th); 1331 1332 return (PF_PASS); 1333 1334 tcp_drop: 1335 REASON_SET(&reason, PFRES_NORM); 1336 if (rm != NULL && r->log) 1337 PFLOG_PACKET(kif, h, m, AF_INET, dir, reason, r, NULL, NULL, pd); 1338 return (PF_DROP); 1339 } 1340 1341 int 1342 pf_normalize_tcp_init(struct mbuf *m, int off, struct pf_pdesc *pd, 1343 struct tcphdr *th, struct pf_state_peer *src, struct pf_state_peer *dst) 1344 { 1345 u_int32_t tsval, tsecr; 1346 u_int8_t hdr[60]; 1347 u_int8_t *opt; 1348 1349 KASSERT((src->scrub == NULL), 1350 ("pf_normalize_tcp_init: src->scrub != NULL")); 1351 1352 src->scrub = pool_get(&pf_state_scrub_pl, PR_NOWAIT); 1353 if (src->scrub == NULL) 1354 return (1); 1355 bzero(src->scrub, sizeof(*src->scrub)); 1356 1357 switch (pd->af) { 1358 #ifdef INET 1359 case AF_INET: { 1360 struct ip *h = mtod(m, struct ip *); 1361 src->scrub->pfss_ttl = h->ip_ttl; 1362 break; 1363 } 1364 #endif /* INET */ 1365 #ifdef INET6 1366 case AF_INET6: { 1367 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 1368 src->scrub->pfss_ttl = h->ip6_hlim; 1369 break; 1370 } 1371 #endif /* INET6 */ 1372 } 1373 1374 1375 /* 1376 * All normalizations below are only begun if we see the start of 1377 * the connections. They must all set an enabled bit in pfss_flags 1378 */ 1379 if ((th->th_flags & TH_SYN) == 0) 1380 return (0); 1381 1382 1383 if (th->th_off > (sizeof(struct tcphdr) >> 2) && src->scrub && 1384 pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) { 1385 /* Diddle with TCP options */ 1386 int hlen; 1387 opt = hdr + sizeof(struct tcphdr); 1388 hlen = (th->th_off << 2) - sizeof(struct tcphdr); 1389 while (hlen >= TCPOLEN_TIMESTAMP) { 1390 switch (*opt) { 1391 case TCPOPT_EOL: /* FALLTHROUGH */ 1392 case TCPOPT_NOP: 1393 opt++; 1394 hlen--; 1395 break; 1396 case TCPOPT_TIMESTAMP: 1397 if (opt[1] >= TCPOLEN_TIMESTAMP) { 1398 src->scrub->pfss_flags |= 1399 PFSS_TIMESTAMP; 1400 src->scrub->pfss_ts_mod = karc4random(); 1401 1402 /* note PFSS_PAWS not set yet */ 1403 memcpy(&tsval, &opt[2], 1404 sizeof(u_int32_t)); 1405 memcpy(&tsecr, &opt[6], 1406 sizeof(u_int32_t)); 1407 src->scrub->pfss_tsval0 = ntohl(tsval); 1408 src->scrub->pfss_tsval = ntohl(tsval); 1409 src->scrub->pfss_tsecr = ntohl(tsecr); 1410 getmicrouptime(&src->scrub->pfss_last); 1411 } 1412 /* FALLTHROUGH */ 1413 default: 1414 hlen -= MAX(opt[1], 2); 1415 opt += MAX(opt[1], 2); 1416 break; 1417 } 1418 } 1419 } 1420 1421 return (0); 1422 } 1423 1424 void 1425 pf_normalize_tcp_cleanup(struct pf_state *state) 1426 { 1427 if (state->src.scrub) 1428 pool_put(&pf_state_scrub_pl, state->src.scrub); 1429 if (state->dst.scrub) 1430 pool_put(&pf_state_scrub_pl, state->dst.scrub); 1431 1432 /* Someday... flush the TCP segment reassembly descriptors. */ 1433 } 1434 1435 int 1436 pf_normalize_tcp_stateful(struct mbuf *m, int off, struct pf_pdesc *pd, 1437 u_short *reason, struct tcphdr *th, struct pf_state *state, 1438 struct pf_state_peer *src, struct pf_state_peer *dst, int *writeback) 1439 { 1440 struct timeval uptime; 1441 u_int32_t tsval, tsecr; 1442 u_int tsval_from_last; 1443 u_int8_t hdr[60]; 1444 u_int8_t *opt; 1445 int copyback = 0; 1446 int got_ts = 0; 1447 1448 KASSERT((src->scrub || dst->scrub), 1449 ("pf_normalize_tcp_statefull: src->scrub && dst->scrub!")); 1450 1451 /* 1452 * Enforce the minimum TTL seen for this connection. Negate a common 1453 * technique to evade an intrusion detection system and confuse 1454 * firewall state code. 1455 */ 1456 switch (pd->af) { 1457 #ifdef INET 1458 case AF_INET: { 1459 if (src->scrub) { 1460 struct ip *h = mtod(m, struct ip *); 1461 if (h->ip_ttl > src->scrub->pfss_ttl) 1462 src->scrub->pfss_ttl = h->ip_ttl; 1463 h->ip_ttl = src->scrub->pfss_ttl; 1464 } 1465 break; 1466 } 1467 #endif /* INET */ 1468 #ifdef INET6 1469 case AF_INET6: { 1470 if (src->scrub) { 1471 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 1472 if (h->ip6_hlim > src->scrub->pfss_ttl) 1473 src->scrub->pfss_ttl = h->ip6_hlim; 1474 h->ip6_hlim = src->scrub->pfss_ttl; 1475 } 1476 break; 1477 } 1478 #endif /* INET6 */ 1479 } 1480 1481 if (th->th_off > (sizeof(struct tcphdr) >> 2) && 1482 ((src->scrub && (src->scrub->pfss_flags & PFSS_TIMESTAMP)) || 1483 (dst->scrub && (dst->scrub->pfss_flags & PFSS_TIMESTAMP))) && 1484 pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) { 1485 /* Diddle with TCP options */ 1486 int hlen; 1487 opt = hdr + sizeof(struct tcphdr); 1488 hlen = (th->th_off << 2) - sizeof(struct tcphdr); 1489 while (hlen >= TCPOLEN_TIMESTAMP) { 1490 switch (*opt) { 1491 case TCPOPT_EOL: /* FALLTHROUGH */ 1492 case TCPOPT_NOP: 1493 opt++; 1494 hlen--; 1495 break; 1496 case TCPOPT_TIMESTAMP: 1497 /* Modulate the timestamps. Can be used for 1498 * NAT detection, OS uptime determination or 1499 * reboot detection. 1500 */ 1501 1502 if (got_ts) { 1503 /* Huh? Multiple timestamps!? */ 1504 if (pf_status.debug >= PF_DEBUG_MISC) { 1505 DPFPRINTF(("multiple TS??")); 1506 pf_print_state(state); 1507 kprintf("\n"); 1508 } 1509 REASON_SET(reason, PFRES_TS); 1510 return (PF_DROP); 1511 } 1512 if (opt[1] >= TCPOLEN_TIMESTAMP) { 1513 memcpy(&tsval, &opt[2], 1514 sizeof(u_int32_t)); 1515 if (tsval && src->scrub && 1516 (src->scrub->pfss_flags & 1517 PFSS_TIMESTAMP)) { 1518 tsval = ntohl(tsval); 1519 pf_change_a(&opt[2], 1520 &th->th_sum, 1521 htonl(tsval + 1522 src->scrub->pfss_ts_mod), 1523 0); 1524 copyback = 1; 1525 } 1526 1527 /* Modulate TS reply iff valid (!0) */ 1528 memcpy(&tsecr, &opt[6], 1529 sizeof(u_int32_t)); 1530 if (tsecr && dst->scrub && 1531 (dst->scrub->pfss_flags & 1532 PFSS_TIMESTAMP)) { 1533 tsecr = ntohl(tsecr) 1534 - dst->scrub->pfss_ts_mod; 1535 pf_change_a(&opt[6], 1536 &th->th_sum, htonl(tsecr), 1537 0); 1538 copyback = 1; 1539 } 1540 got_ts = 1; 1541 } 1542 /* FALLTHROUGH */ 1543 default: 1544 hlen -= MAX(opt[1], 2); 1545 opt += MAX(opt[1], 2); 1546 break; 1547 } 1548 } 1549 if (copyback) { 1550 /* Copyback the options, caller copys back header */ 1551 *writeback = 1; 1552 m_copyback(m, off + sizeof(struct tcphdr), 1553 (th->th_off << 2) - sizeof(struct tcphdr), hdr + 1554 sizeof(struct tcphdr)); 1555 } 1556 } 1557 1558 1559 /* 1560 * Must invalidate PAWS checks on connections idle for too long. 1561 * The fastest allowed timestamp clock is 1ms. That turns out to 1562 * be about 24 days before it wraps. XXX Right now our lowerbound 1563 * TS echo check only works for the first 12 days of a connection 1564 * when the TS has exhausted half its 32bit space 1565 */ 1566 #define TS_MAX_IDLE (24*24*60*60) 1567 #define TS_MAX_CONN (12*24*60*60) /* XXX remove when better tsecr check */ 1568 1569 getmicrouptime(&uptime); 1570 if (src->scrub && (src->scrub->pfss_flags & PFSS_PAWS) && 1571 (uptime.tv_sec - src->scrub->pfss_last.tv_sec > TS_MAX_IDLE || 1572 time_second - state->creation > TS_MAX_CONN)) { 1573 if (pf_status.debug >= PF_DEBUG_MISC) { 1574 DPFPRINTF(("src idled out of PAWS\n")); 1575 pf_print_state(state); 1576 kprintf("\n"); 1577 } 1578 src->scrub->pfss_flags = (src->scrub->pfss_flags & ~PFSS_PAWS) 1579 | PFSS_PAWS_IDLED; 1580 } 1581 if (dst->scrub && (dst->scrub->pfss_flags & PFSS_PAWS) && 1582 uptime.tv_sec - dst->scrub->pfss_last.tv_sec > TS_MAX_IDLE) { 1583 if (pf_status.debug >= PF_DEBUG_MISC) { 1584 DPFPRINTF(("dst idled out of PAWS\n")); 1585 pf_print_state(state); 1586 kprintf("\n"); 1587 } 1588 dst->scrub->pfss_flags = (dst->scrub->pfss_flags & ~PFSS_PAWS) 1589 | PFSS_PAWS_IDLED; 1590 } 1591 1592 if (got_ts && src->scrub && dst->scrub && 1593 (src->scrub->pfss_flags & PFSS_PAWS) && 1594 (dst->scrub->pfss_flags & PFSS_PAWS)) { 1595 /* Validate that the timestamps are "in-window". 1596 * RFC1323 describes TCP Timestamp options that allow 1597 * measurement of RTT (round trip time) and PAWS 1598 * (protection against wrapped sequence numbers). PAWS 1599 * gives us a set of rules for rejecting packets on 1600 * long fat pipes (packets that were somehow delayed 1601 * in transit longer than the time it took to send the 1602 * full TCP sequence space of 4Gb). We can use these 1603 * rules and infer a few others that will let us treat 1604 * the 32bit timestamp and the 32bit echoed timestamp 1605 * as sequence numbers to prevent a blind attacker from 1606 * inserting packets into a connection. 1607 * 1608 * RFC1323 tells us: 1609 * - The timestamp on this packet must be greater than 1610 * or equal to the last value echoed by the other 1611 * endpoint. The RFC says those will be discarded 1612 * since it is a dup that has already been acked. 1613 * This gives us a lowerbound on the timestamp. 1614 * timestamp >= other last echoed timestamp 1615 * - The timestamp will be less than or equal to 1616 * the last timestamp plus the time between the 1617 * last packet and now. The RFC defines the max 1618 * clock rate as 1ms. We will allow clocks to be 1619 * up to 10% fast and will allow a total difference 1620 * or 30 seconds due to a route change. And this 1621 * gives us an upperbound on the timestamp. 1622 * timestamp <= last timestamp + max ticks 1623 * We have to be careful here. Windows will send an 1624 * initial timestamp of zero and then initialize it 1625 * to a random value after the 3whs; presumably to 1626 * avoid a DoS by having to call an expensive RNG 1627 * during a SYN flood. Proof MS has at least one 1628 * good security geek. 1629 * 1630 * - The TCP timestamp option must also echo the other 1631 * endpoints timestamp. The timestamp echoed is the 1632 * one carried on the earliest unacknowledged segment 1633 * on the left edge of the sequence window. The RFC 1634 * states that the host will reject any echoed 1635 * timestamps that were larger than any ever sent. 1636 * This gives us an upperbound on the TS echo. 1637 * tescr <= largest_tsval 1638 * - The lowerbound on the TS echo is a little more 1639 * tricky to determine. The other endpoint's echoed 1640 * values will not decrease. But there may be 1641 * network conditions that re-order packets and 1642 * cause our view of them to decrease. For now the 1643 * only lowerbound we can safely determine is that 1644 * the TS echo will never be less than the original 1645 * TS. XXX There is probably a better lowerbound. 1646 * Remove TS_MAX_CONN with better lowerbound check. 1647 * tescr >= other original TS 1648 * 1649 * It is also important to note that the fastest 1650 * timestamp clock of 1ms will wrap its 32bit space in 1651 * 24 days. So we just disable TS checking after 24 1652 * days of idle time. We actually must use a 12d 1653 * connection limit until we can come up with a better 1654 * lowerbound to the TS echo check. 1655 */ 1656 struct timeval delta_ts; 1657 int ts_fudge; 1658 1659 1660 /* 1661 * PFTM_TS_DIFF is how many seconds of leeway to allow 1662 * a host's timestamp. This can happen if the previous 1663 * packet got delayed in transit for much longer than 1664 * this packet. 1665 */ 1666 if ((ts_fudge = state->rule.ptr->timeout[PFTM_TS_DIFF]) == 0) 1667 ts_fudge = pf_default_rule.timeout[PFTM_TS_DIFF]; 1668 1669 1670 /* Calculate max ticks since the last timestamp */ 1671 #define TS_MAXFREQ 1100 /* RFC max TS freq of 1Khz + 10% skew */ 1672 #define TS_MICROSECS 1000000 /* microseconds per second */ 1673 #ifndef timersub 1674 #define timersub(tvp, uvp, vvp) \ 1675 do { \ 1676 (vvp)->tv_sec = (tvp)->tv_sec - (uvp)->tv_sec; \ 1677 (vvp)->tv_usec = (tvp)->tv_usec - (uvp)->tv_usec; \ 1678 if ((vvp)->tv_usec < 0) { \ 1679 (vvp)->tv_sec--; \ 1680 (vvp)->tv_usec += 1000000; \ 1681 } \ 1682 } while (0) 1683 #endif 1684 1685 timersub(&uptime, &src->scrub->pfss_last, &delta_ts); 1686 tsval_from_last = (delta_ts.tv_sec + ts_fudge) * TS_MAXFREQ; 1687 tsval_from_last += delta_ts.tv_usec / (TS_MICROSECS/TS_MAXFREQ); 1688 1689 1690 if ((src->state >= TCPS_ESTABLISHED && 1691 dst->state >= TCPS_ESTABLISHED) && 1692 (SEQ_LT(tsval, dst->scrub->pfss_tsecr) || 1693 SEQ_GT(tsval, src->scrub->pfss_tsval + tsval_from_last) || 1694 (tsecr && (SEQ_GT(tsecr, dst->scrub->pfss_tsval) || 1695 SEQ_LT(tsecr, dst->scrub->pfss_tsval0))))) { 1696 /* Bad RFC1323 implementation or an insertion attack. 1697 * 1698 * - Solaris 2.6 and 2.7 are known to send another ACK 1699 * after the FIN,FIN|ACK,ACK closing that carries 1700 * an old timestamp. 1701 */ 1702 1703 DPFPRINTF(("Timestamp failed %c%c%c%c\n", 1704 SEQ_LT(tsval, dst->scrub->pfss_tsecr) ? '0' : ' ', 1705 SEQ_GT(tsval, src->scrub->pfss_tsval + 1706 tsval_from_last) ? '1' : ' ', 1707 SEQ_GT(tsecr, dst->scrub->pfss_tsval) ? '2' : ' ', 1708 SEQ_LT(tsecr, dst->scrub->pfss_tsval0)? '3' : ' ')); 1709 DPFPRINTF((" tsval: %u tsecr: %u +ticks: %u " 1710 "idle: %lus %lums\n", 1711 tsval, tsecr, tsval_from_last, delta_ts.tv_sec, 1712 delta_ts.tv_usec / 1000)); 1713 DPFPRINTF((" src->tsval: %u tsecr: %u\n", 1714 src->scrub->pfss_tsval, src->scrub->pfss_tsecr)); 1715 DPFPRINTF((" dst->tsval: %u tsecr: %u tsval0: %u" 1716 "\n", dst->scrub->pfss_tsval, 1717 dst->scrub->pfss_tsecr, dst->scrub->pfss_tsval0)); 1718 if (pf_status.debug >= PF_DEBUG_MISC) { 1719 pf_print_state(state); 1720 pf_print_flags(th->th_flags); 1721 kprintf("\n"); 1722 } 1723 REASON_SET(reason, PFRES_TS); 1724 return (PF_DROP); 1725 } 1726 1727 /* XXX I'd really like to require tsecr but it's optional */ 1728 1729 } else if (!got_ts && (th->th_flags & TH_RST) == 0 && 1730 ((src->state == TCPS_ESTABLISHED && dst->state == TCPS_ESTABLISHED) 1731 || pd->p_len > 0 || (th->th_flags & TH_SYN)) && 1732 src->scrub && dst->scrub && 1733 (src->scrub->pfss_flags & PFSS_PAWS) && 1734 (dst->scrub->pfss_flags & PFSS_PAWS)) { 1735 /* Didn't send a timestamp. Timestamps aren't really useful 1736 * when: 1737 * - connection opening or closing (often not even sent). 1738 * but we must not let an attacker to put a FIN on a 1739 * data packet to sneak it through our ESTABLISHED check. 1740 * - on a TCP reset. RFC suggests not even looking at TS. 1741 * - on an empty ACK. The TS will not be echoed so it will 1742 * probably not help keep the RTT calculation in sync and 1743 * there isn't as much danger when the sequence numbers 1744 * got wrapped. So some stacks don't include TS on empty 1745 * ACKs :-( 1746 * 1747 * To minimize the disruption to mostly RFC1323 conformant 1748 * stacks, we will only require timestamps on data packets. 1749 * 1750 * And what do ya know, we cannot require timestamps on data 1751 * packets. There appear to be devices that do legitimate 1752 * TCP connection hijacking. There are HTTP devices that allow 1753 * a 3whs (with timestamps) and then buffer the HTTP request. 1754 * If the intermediate device has the HTTP response cache, it 1755 * will spoof the response but not bother timestamping its 1756 * packets. So we can look for the presence of a timestamp in 1757 * the first data packet and if there, require it in all future 1758 * packets. 1759 */ 1760 1761 if (pd->p_len > 0 && (src->scrub->pfss_flags & PFSS_DATA_TS)) { 1762 /* 1763 * Hey! Someone tried to sneak a packet in. Or the 1764 * stack changed its RFC1323 behavior?!?! 1765 */ 1766 if (pf_status.debug >= PF_DEBUG_MISC) { 1767 DPFPRINTF(("Did not receive expected RFC1323 " 1768 "timestamp\n")); 1769 pf_print_state(state); 1770 pf_print_flags(th->th_flags); 1771 kprintf("\n"); 1772 } 1773 REASON_SET(reason, PFRES_TS); 1774 return (PF_DROP); 1775 } 1776 } 1777 1778 1779 /* 1780 * We will note if a host sends his data packets with or without 1781 * timestamps. And require all data packets to contain a timestamp 1782 * if the first does. PAWS implicitly requires that all data packets be 1783 * timestamped. But I think there are middle-man devices that hijack 1784 * TCP streams immediately after the 3whs and don't timestamp their 1785 * packets (seen in a WWW accelerator or cache). 1786 */ 1787 if (pd->p_len > 0 && src->scrub && (src->scrub->pfss_flags & 1788 (PFSS_TIMESTAMP|PFSS_DATA_TS|PFSS_DATA_NOTS)) == PFSS_TIMESTAMP) { 1789 if (got_ts) 1790 src->scrub->pfss_flags |= PFSS_DATA_TS; 1791 else { 1792 src->scrub->pfss_flags |= PFSS_DATA_NOTS; 1793 if (pf_status.debug >= PF_DEBUG_MISC && dst->scrub && 1794 (dst->scrub->pfss_flags & PFSS_TIMESTAMP)) { 1795 /* Don't warn if other host rejected RFC1323 */ 1796 DPFPRINTF(("Broken RFC1323 stack did not " 1797 "timestamp data packet. Disabled PAWS " 1798 "security.\n")); 1799 pf_print_state(state); 1800 pf_print_flags(th->th_flags); 1801 kprintf("\n"); 1802 } 1803 } 1804 } 1805 1806 1807 /* 1808 * Update PAWS values 1809 */ 1810 if (got_ts && src->scrub && PFSS_TIMESTAMP == (src->scrub->pfss_flags & 1811 (PFSS_PAWS_IDLED|PFSS_TIMESTAMP))) { 1812 getmicrouptime(&src->scrub->pfss_last); 1813 if (SEQ_GEQ(tsval, src->scrub->pfss_tsval) || 1814 (src->scrub->pfss_flags & PFSS_PAWS) == 0) 1815 src->scrub->pfss_tsval = tsval; 1816 1817 if (tsecr) { 1818 if (SEQ_GEQ(tsecr, src->scrub->pfss_tsecr) || 1819 (src->scrub->pfss_flags & PFSS_PAWS) == 0) 1820 src->scrub->pfss_tsecr = tsecr; 1821 1822 if ((src->scrub->pfss_flags & PFSS_PAWS) == 0 && 1823 (SEQ_LT(tsval, src->scrub->pfss_tsval0) || 1824 src->scrub->pfss_tsval0 == 0)) { 1825 /* tsval0 MUST be the lowest timestamp */ 1826 src->scrub->pfss_tsval0 = tsval; 1827 } 1828 1829 /* Only fully initialized after a TS gets echoed */ 1830 if ((src->scrub->pfss_flags & PFSS_PAWS) == 0) 1831 src->scrub->pfss_flags |= PFSS_PAWS; 1832 } 1833 } 1834 1835 /* I have a dream.... TCP segment reassembly.... */ 1836 return (0); 1837 } 1838 1839 int 1840 pf_normalize_tcpopt(struct pf_rule *r, struct mbuf *m, struct tcphdr *th, 1841 int off, sa_family_t af) 1842 { 1843 u_int16_t *mss; 1844 int thoff; 1845 int opt, cnt, optlen = 0; 1846 int rewrite = 0; 1847 u_char opts[TCP_MAXOLEN]; 1848 u_char *optp = opts; 1849 1850 thoff = th->th_off << 2; 1851 cnt = thoff - sizeof(struct tcphdr); 1852 1853 if (cnt > 0 && !pf_pull_hdr(m, off + sizeof(*th), opts, cnt, 1854 NULL, NULL, af)) 1855 return (rewrite); 1856 1857 for (; cnt > 0; cnt -= optlen, optp += optlen) { 1858 opt = optp[0]; 1859 if (opt == TCPOPT_EOL) 1860 break; 1861 if (opt == TCPOPT_NOP) 1862 optlen = 1; 1863 else { 1864 if (cnt < 2) 1865 break; 1866 optlen = optp[1]; 1867 if (optlen < 2 || optlen > cnt) 1868 break; 1869 } 1870 switch (opt) { 1871 case TCPOPT_MAXSEG: 1872 mss = (u_int16_t *)(optp + 2); 1873 if ((ntohs(*mss)) > r->max_mss) { 1874 th->th_sum = pf_cksum_fixup(th->th_sum, 1875 *mss, htons(r->max_mss), 0); 1876 *mss = htons(r->max_mss); 1877 rewrite = 1; 1878 } 1879 break; 1880 default: 1881 break; 1882 } 1883 } 1884 1885 if (rewrite) 1886 m_copyback(m, off + sizeof(*th), thoff - sizeof(*th), opts); 1887 1888 return (rewrite); 1889 } 1890