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 42 #include <net/if.h> 43 #include <net/if_types.h> 44 #include <net/bpf.h> 45 #include <net/route.h> 46 #include <net/pf/if_pflog.h> 47 48 #include <netinet/in.h> 49 #include <netinet/in_var.h> 50 #include <netinet/in_systm.h> 51 #include <netinet/ip.h> 52 #include <netinet/ip_var.h> 53 #include <netinet/tcp.h> 54 #include <netinet/tcp_seq.h> 55 #include <netinet/udp.h> 56 #include <netinet/ip_icmp.h> 57 58 #ifdef INET6 59 #include <netinet/ip6.h> 60 #endif /* INET6 */ 61 62 #include <net/pf/pfvar.h> 63 64 #define PFFRAG_SEENLAST 0x0001 /* Seen the last fragment for this */ 65 #define PFFRAG_NOBUFFER 0x0002 /* Non-buffering fragment cache */ 66 #define PFFRAG_DROP 0x0004 /* Drop all fragments */ 67 #define BUFFER_FRAGMENTS(fr) (!((fr)->fr_flags & PFFRAG_NOBUFFER)) 68 69 70 TAILQ_HEAD(pf_fragqueue, pf_fragment) pf_fragqueue; 71 TAILQ_HEAD(pf_cachequeue, pf_fragment) pf_cachequeue; 72 73 static __inline int pf_frag_compare(struct pf_fragment *, 74 struct pf_fragment *); 75 RB_HEAD(pf_frag_tree, pf_fragment) pf_frag_tree, pf_cache_tree; 76 RB_PROTOTYPE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare); 77 RB_GENERATE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare); 78 79 /* Private prototypes */ 80 void pf_ip2key(struct pf_fragment *, struct ip *); 81 void pf_remove_fragment(struct pf_fragment *); 82 void pf_flush_fragments(void); 83 void pf_free_fragment(struct pf_fragment *); 84 struct pf_fragment *pf_find_fragment(struct ip *, struct pf_frag_tree *); 85 struct mbuf *pf_reassemble(struct mbuf **, struct pf_fragment **, 86 struct pf_frent *, int); 87 struct mbuf *pf_fragcache(struct mbuf **, struct ip*, 88 struct pf_fragment **, int, int, int *); 89 int pf_normalize_tcpopt(struct pf_rule *, struct mbuf *, 90 struct tcphdr *, int, sa_family_t); 91 92 #define DPFPRINTF(x) do { \ 93 if (pf_status.debug >= PF_DEBUG_MISC) { \ 94 kprintf("%s: ", __func__); \ 95 kprintf x ; \ 96 } \ 97 } while(0) 98 99 static MALLOC_DEFINE(M_PFFRAGPL, "pffrag", "pf fragment pool list"); 100 static MALLOC_DEFINE(M_PFCACHEPL, "pffrcache", "pf fragment cache pool list"); 101 static MALLOC_DEFINE(M_PFFRENTPL, "pffrent", "pf frent pool list"); 102 static MALLOC_DEFINE(M_PFCENTPL, "pffrcent", "pf fragment cent pool list"); 103 static MALLOC_DEFINE(M_PFSTATESCRUBPL, "pfstatescrub", "pf state scrub pool list"); 104 105 /* Globals */ 106 struct malloc_type *pf_frent_pl, *pf_frag_pl, *pf_cache_pl, *pf_cent_pl; 107 struct malloc_type *pf_state_scrub_pl; 108 int pf_nfrents, pf_ncache; 109 110 void 111 pf_normalize_init(void) 112 { 113 /* XXX 114 pool_sethiwat(&pf_frag_pl, PFFRAG_FRAG_HIWAT); 115 pool_sethardlimit(&pf_frent_pl, PFFRAG_FRENT_HIWAT, NULL, 0); 116 pool_sethardlimit(&pf_cache_pl, PFFRAG_FRCACHE_HIWAT, NULL, 0); 117 pool_sethardlimit(&pf_cent_pl, PFFRAG_FRCENT_HIWAT, NULL, 0); 118 */ 119 120 TAILQ_INIT(&pf_fragqueue); 121 TAILQ_INIT(&pf_cachequeue); 122 } 123 124 static __inline int 125 pf_frag_compare(struct pf_fragment *a, struct pf_fragment *b) 126 { 127 int diff; 128 129 if ((diff = a->fr_id - b->fr_id)) 130 return (diff); 131 else if ((diff = a->fr_p - b->fr_p)) 132 return (diff); 133 else if (a->fr_src.s_addr < b->fr_src.s_addr) 134 return (-1); 135 else if (a->fr_src.s_addr > b->fr_src.s_addr) 136 return (1); 137 else if (a->fr_dst.s_addr < b->fr_dst.s_addr) 138 return (-1); 139 else if (a->fr_dst.s_addr > b->fr_dst.s_addr) 140 return (1); 141 return (0); 142 } 143 144 void 145 pf_purge_expired_fragments(void) 146 { 147 struct pf_fragment *frag; 148 u_int32_t expire = time_second - 149 pf_default_rule.timeout[PFTM_FRAG]; 150 151 while ((frag = TAILQ_LAST(&pf_fragqueue, pf_fragqueue)) != NULL) { 152 KASSERT((BUFFER_FRAGMENTS(frag)), 153 ("BUFFER_FRAGMENTS(frag) == 0: %s", __func__)); 154 if (frag->fr_timeout > expire) 155 break; 156 157 DPFPRINTF(("expiring %d(%p)\n", frag->fr_id, frag)); 158 pf_free_fragment(frag); 159 } 160 161 while ((frag = TAILQ_LAST(&pf_cachequeue, pf_cachequeue)) != NULL) { 162 KASSERT((!BUFFER_FRAGMENTS(frag)), 163 ("BUFFER_FRAGMENTS(frag) != 0: %s", __func__)); 164 if (frag->fr_timeout > expire) 165 break; 166 167 DPFPRINTF(("expiring %d(%p)\n", frag->fr_id, frag)); 168 pf_free_fragment(frag); 169 KASSERT((TAILQ_EMPTY(&pf_cachequeue) || 170 TAILQ_LAST(&pf_cachequeue, pf_cachequeue) != frag), 171 ("!(TAILQ_EMPTY() || TAILQ_LAST() == farg): %s", 172 __func__)); 173 } 174 } 175 176 /* 177 * Try to flush old fragments to make space for new ones 178 */ 179 180 void 181 pf_flush_fragments(void) 182 { 183 struct pf_fragment *frag; 184 int goal; 185 186 goal = pf_nfrents * 9 / 10; 187 DPFPRINTF(("trying to free > %d frents\n", 188 pf_nfrents - goal)); 189 while (goal < pf_nfrents) { 190 frag = TAILQ_LAST(&pf_fragqueue, pf_fragqueue); 191 if (frag == NULL) 192 break; 193 pf_free_fragment(frag); 194 } 195 196 197 goal = pf_ncache * 9 / 10; 198 DPFPRINTF(("trying to free > %d cache entries\n", 199 pf_ncache - goal)); 200 while (goal < pf_ncache) { 201 frag = TAILQ_LAST(&pf_cachequeue, pf_cachequeue); 202 if (frag == NULL) 203 break; 204 pf_free_fragment(frag); 205 } 206 } 207 208 /* Frees the fragments and all associated entries */ 209 210 void 211 pf_free_fragment(struct pf_fragment *frag) 212 { 213 struct pf_frent *frent; 214 struct pf_frcache *frcache; 215 216 /* Free all fragments */ 217 if (BUFFER_FRAGMENTS(frag)) { 218 for (frent = LIST_FIRST(&frag->fr_queue); frent; 219 frent = LIST_FIRST(&frag->fr_queue)) { 220 LIST_REMOVE(frent, fr_next); 221 222 m_freem(frent->fr_m); 223 kfree(frent, M_PFFRENTPL); 224 pf_nfrents--; 225 } 226 } else { 227 for (frcache = LIST_FIRST(&frag->fr_cache); frcache; 228 frcache = LIST_FIRST(&frag->fr_cache)) { 229 LIST_REMOVE(frcache, fr_next); 230 231 KASSERT((LIST_EMPTY(&frag->fr_cache) || 232 LIST_FIRST(&frag->fr_cache)->fr_off > 233 frcache->fr_end), 234 ("! (LIST_EMPTY() || LIST_FIRST()->fr_off >" 235 " frcache->fr_end): %s", __func__)); 236 237 kfree(frcache, M_PFCENTPL); 238 pf_ncache--; 239 } 240 } 241 242 pf_remove_fragment(frag); 243 } 244 245 void 246 pf_ip2key(struct pf_fragment *key, struct ip *ip) 247 { 248 key->fr_p = ip->ip_p; 249 key->fr_id = ip->ip_id; 250 key->fr_src.s_addr = ip->ip_src.s_addr; 251 key->fr_dst.s_addr = ip->ip_dst.s_addr; 252 } 253 254 struct pf_fragment * 255 pf_find_fragment(struct ip *ip, struct pf_frag_tree *tree) 256 { 257 struct pf_fragment key; 258 struct pf_fragment *frag; 259 260 pf_ip2key(&key, ip); 261 262 frag = RB_FIND(pf_frag_tree, tree, &key); 263 if (frag != NULL) { 264 /* XXX Are we sure we want to update the timeout? */ 265 frag->fr_timeout = time_second; 266 if (BUFFER_FRAGMENTS(frag)) { 267 TAILQ_REMOVE(&pf_fragqueue, frag, frag_next); 268 TAILQ_INSERT_HEAD(&pf_fragqueue, frag, frag_next); 269 } else { 270 TAILQ_REMOVE(&pf_cachequeue, frag, frag_next); 271 TAILQ_INSERT_HEAD(&pf_cachequeue, frag, frag_next); 272 } 273 } 274 275 return (frag); 276 } 277 278 /* Removes a fragment from the fragment queue and frees the fragment */ 279 280 void 281 pf_remove_fragment(struct pf_fragment *frag) 282 { 283 if (BUFFER_FRAGMENTS(frag)) { 284 RB_REMOVE(pf_frag_tree, &pf_frag_tree, frag); 285 TAILQ_REMOVE(&pf_fragqueue, frag, frag_next); 286 kfree(frag, M_PFFRAGPL); 287 } else { 288 RB_REMOVE(pf_frag_tree, &pf_cache_tree, frag); 289 TAILQ_REMOVE(&pf_cachequeue, frag, frag_next); 290 kfree(frag, M_PFCACHEPL); 291 } 292 } 293 294 #define FR_IP_OFF(fr) (((fr)->fr_ip->ip_off & IP_OFFMASK) << 3) 295 struct mbuf * 296 pf_reassemble(struct mbuf **m0, struct pf_fragment **frag, 297 struct pf_frent *frent, int mff) 298 { 299 struct mbuf *m = *m0, *m2; 300 struct pf_frent *frea, *next; 301 struct pf_frent *frep = NULL; 302 struct ip *ip = frent->fr_ip; 303 int hlen = ip->ip_hl << 2; 304 u_int16_t off = (ip->ip_off & IP_OFFMASK) << 3; 305 u_int16_t ip_len = ip->ip_len - ip->ip_hl * 4; 306 u_int16_t max = ip_len + off; 307 308 KASSERT((*frag == NULL || BUFFER_FRAGMENTS(*frag)), 309 ("! (*frag == NULL || BUFFER_FRAGMENTS(*frag)): %s", __func__)); 310 311 /* Strip off ip header */ 312 m->m_data += hlen; 313 m->m_len -= hlen; 314 315 /* Create a new reassembly queue for this packet */ 316 if (*frag == NULL) { 317 *frag = kmalloc(sizeof(struct pf_fragment), M_PFFRAGPL, M_NOWAIT); 318 if (*frag == NULL) { 319 pf_flush_fragments(); 320 *frag = kmalloc(sizeof(struct pf_fragment), M_PFFRAGPL, M_NOWAIT); 321 if (*frag == NULL) 322 goto drop_fragment; 323 } 324 325 (*frag)->fr_flags = 0; 326 (*frag)->fr_max = 0; 327 (*frag)->fr_src = frent->fr_ip->ip_src; 328 (*frag)->fr_dst = frent->fr_ip->ip_dst; 329 (*frag)->fr_p = frent->fr_ip->ip_p; 330 (*frag)->fr_id = frent->fr_ip->ip_id; 331 (*frag)->fr_timeout = time_second; 332 LIST_INIT(&(*frag)->fr_queue); 333 334 RB_INSERT(pf_frag_tree, &pf_frag_tree, *frag); 335 TAILQ_INSERT_HEAD(&pf_fragqueue, *frag, frag_next); 336 337 /* We do not have a previous fragment */ 338 frep = NULL; 339 goto insert; 340 } 341 342 /* 343 * Find a fragment after the current one: 344 * - off contains the real shifted offset. 345 */ 346 LIST_FOREACH(frea, &(*frag)->fr_queue, fr_next) { 347 if (FR_IP_OFF(frea) > off) 348 break; 349 frep = frea; 350 } 351 352 KASSERT((frep != NULL || frea != NULL), 353 ("!(frep != NULL || frea != NULL): %s", __func__)); 354 355 if (frep != NULL && 356 FR_IP_OFF(frep) + frep->fr_ip->ip_len - frep->fr_ip->ip_hl * 357 4 > off) 358 { 359 u_int16_t precut; 360 361 precut = FR_IP_OFF(frep) + frep->fr_ip->ip_len - 362 frep->fr_ip->ip_hl * 4 - off; 363 if (precut >= ip_len) 364 goto drop_fragment; 365 m_adj(frent->fr_m, precut); 366 DPFPRINTF(("overlap -%d\n", precut)); 367 /* Enforce 8 byte boundaries */ 368 ip->ip_off = ip->ip_off + (precut >> 3); 369 off = (ip->ip_off & IP_OFFMASK) << 3; 370 ip_len -= precut; 371 ip->ip_len = ip_len; 372 } 373 374 for (; frea != NULL && ip_len + off > FR_IP_OFF(frea); 375 frea = next) 376 { 377 u_int16_t aftercut; 378 379 aftercut = ip_len + off - FR_IP_OFF(frea); 380 DPFPRINTF(("adjust overlap %d\n", aftercut)); 381 if (aftercut < frea->fr_ip->ip_len - frea->fr_ip->ip_hl 382 * 4) 383 { 384 frea->fr_ip->ip_len = 385 frea->fr_ip->ip_len - aftercut; 386 frea->fr_ip->ip_off = frea->fr_ip->ip_off + 387 (aftercut >> 3); 388 m_adj(frea->fr_m, aftercut); 389 break; 390 } 391 392 /* This fragment is completely overlapped, lose it */ 393 next = LIST_NEXT(frea, fr_next); 394 m_freem(frea->fr_m); 395 LIST_REMOVE(frea, fr_next); 396 kfree(frea, M_PFFRENTPL); 397 pf_nfrents--; 398 } 399 400 insert: 401 /* Update maximum data size */ 402 if ((*frag)->fr_max < max) 403 (*frag)->fr_max = max; 404 /* This is the last segment */ 405 if (!mff) 406 (*frag)->fr_flags |= PFFRAG_SEENLAST; 407 408 if (frep == NULL) 409 LIST_INSERT_HEAD(&(*frag)->fr_queue, frent, fr_next); 410 else 411 LIST_INSERT_AFTER(frep, frent, fr_next); 412 413 /* Check if we are completely reassembled */ 414 if (!((*frag)->fr_flags & PFFRAG_SEENLAST)) 415 return (NULL); 416 417 /* Check if we have all the data */ 418 off = 0; 419 for (frep = LIST_FIRST(&(*frag)->fr_queue); frep; frep = next) { 420 next = LIST_NEXT(frep, fr_next); 421 422 off += frep->fr_ip->ip_len - frep->fr_ip->ip_hl * 4; 423 if (off < (*frag)->fr_max && 424 (next == NULL || FR_IP_OFF(next) != off)) 425 { 426 DPFPRINTF(("missing fragment at %d, next %d, max %d\n", 427 off, next == NULL ? -1 : FR_IP_OFF(next), 428 (*frag)->fr_max)); 429 return (NULL); 430 } 431 } 432 DPFPRINTF(("%d < %d?\n", off, (*frag)->fr_max)); 433 if (off < (*frag)->fr_max) 434 return (NULL); 435 436 /* We have all the data */ 437 frent = LIST_FIRST(&(*frag)->fr_queue); 438 KASSERT((frent != NULL), ("frent == NULL: %s", __func__)); 439 if ((frent->fr_ip->ip_hl << 2) + off > IP_MAXPACKET) { 440 DPFPRINTF(("drop: too big: %d\n", off)); 441 pf_free_fragment(*frag); 442 *frag = NULL; 443 return (NULL); 444 } 445 next = LIST_NEXT(frent, fr_next); 446 447 /* Magic from ip_input */ 448 ip = frent->fr_ip; 449 m = frent->fr_m; 450 m2 = m->m_next; 451 m->m_next = NULL; 452 m_cat(m, m2); 453 kfree(frent, M_PFFRENTPL); 454 pf_nfrents--; 455 for (frent = next; frent != NULL; frent = next) { 456 next = LIST_NEXT(frent, fr_next); 457 458 m2 = frent->fr_m; 459 kfree(frent, M_PFFRENTPL); 460 pf_nfrents--; 461 m_cat(m, m2); 462 } 463 464 ip->ip_src = (*frag)->fr_src; 465 ip->ip_dst = (*frag)->fr_dst; 466 467 /* Remove from fragment queue */ 468 pf_remove_fragment(*frag); 469 *frag = NULL; 470 471 hlen = ip->ip_hl << 2; 472 ip->ip_len = off + hlen; 473 m->m_len += hlen; 474 m->m_data -= hlen; 475 476 /* some debugging cruft by sklower, below, will go away soon */ 477 /* XXX this should be done elsewhere */ 478 if (m->m_flags & M_PKTHDR) { 479 int plen = 0; 480 for (m2 = m; m2; m2 = m2->m_next) 481 plen += m2->m_len; 482 m->m_pkthdr.len = plen; 483 } 484 485 DPFPRINTF(("complete: %p(%d)\n", m, ip->ip_len)); 486 return (m); 487 488 drop_fragment: 489 /* Oops - fail safe - drop packet */ 490 kfree(frent, M_PFFRENTPL); 491 pf_nfrents--; 492 m_freem(m); 493 return (NULL); 494 } 495 496 struct mbuf * 497 pf_fragcache(struct mbuf **m0, struct ip *h, struct pf_fragment **frag, int mff, 498 int drop, int *nomem) 499 { 500 struct mbuf *m = *m0; 501 struct pf_frcache *frp, *fra, *cur = NULL; 502 int ip_len = h->ip_len - (h->ip_hl << 2); 503 u_int16_t off = h->ip_off << 3; 504 u_int16_t max = ip_len + off; 505 int hosed = 0; 506 507 KASSERT((*frag == NULL || !BUFFER_FRAGMENTS(*frag)), 508 ("!(*frag == NULL || !BUFFER_FRAGMENTS(*frag)): %s", __func__)); 509 510 /* Create a new range queue for this packet */ 511 if (*frag == NULL) { 512 *frag = kmalloc(sizeof(struct pf_fragment), M_PFCACHEPL, M_NOWAIT); 513 if (*frag == NULL) { 514 pf_flush_fragments(); 515 *frag = kmalloc(sizeof(struct pf_fragment), M_PFCACHEPL, M_NOWAIT); 516 if (*frag == NULL) 517 goto no_mem; 518 } 519 520 /* Get an entry for the queue */ 521 cur = kmalloc(sizeof(struct pf_frcache), M_PFCENTPL, M_NOWAIT); 522 if (cur == NULL) { 523 kfree(*frag, M_PFCACHEPL); 524 *frag = NULL; 525 goto no_mem; 526 } 527 pf_ncache++; 528 529 (*frag)->fr_flags = PFFRAG_NOBUFFER; 530 (*frag)->fr_max = 0; 531 (*frag)->fr_src = h->ip_src; 532 (*frag)->fr_dst = h->ip_dst; 533 (*frag)->fr_p = h->ip_p; 534 (*frag)->fr_id = h->ip_id; 535 (*frag)->fr_timeout = time_second; 536 537 cur->fr_off = off; 538 cur->fr_end = max; 539 LIST_INIT(&(*frag)->fr_cache); 540 LIST_INSERT_HEAD(&(*frag)->fr_cache, cur, fr_next); 541 542 RB_INSERT(pf_frag_tree, &pf_cache_tree, *frag); 543 TAILQ_INSERT_HEAD(&pf_cachequeue, *frag, frag_next); 544 545 DPFPRINTF(("fragcache[%d]: new %d-%d\n", h->ip_id, off, max)); 546 547 goto pass; 548 } 549 550 /* 551 * Find a fragment after the current one: 552 * - off contains the real shifted offset. 553 */ 554 frp = NULL; 555 LIST_FOREACH(fra, &(*frag)->fr_cache, fr_next) { 556 if (fra->fr_off > off) 557 break; 558 frp = fra; 559 } 560 561 KASSERT((frp != NULL || fra != NULL), 562 ("!(frp != NULL || fra != NULL): %s", __func__)); 563 564 if (frp != NULL) { 565 int precut; 566 567 precut = frp->fr_end - off; 568 if (precut >= ip_len) { 569 /* Fragment is entirely a duplicate */ 570 DPFPRINTF(("fragcache[%d]: dead (%d-%d) %d-%d\n", 571 h->ip_id, frp->fr_off, frp->fr_end, off, max)); 572 goto drop_fragment; 573 } 574 if (precut == 0) { 575 /* They are adjacent. Fixup cache entry */ 576 DPFPRINTF(("fragcache[%d]: adjacent (%d-%d) %d-%d\n", 577 h->ip_id, frp->fr_off, frp->fr_end, off, max)); 578 frp->fr_end = max; 579 } else if (precut > 0) { 580 /* The first part of this payload overlaps with a 581 * fragment that has already been passed. 582 * Need to trim off the first part of the payload. 583 * But to do so easily, we need to create another 584 * mbuf to throw the original header into. 585 */ 586 587 DPFPRINTF(("fragcache[%d]: chop %d (%d-%d) %d-%d\n", 588 h->ip_id, precut, frp->fr_off, frp->fr_end, off, 589 max)); 590 591 off += precut; 592 max -= precut; 593 /* Update the previous frag to encompass this one */ 594 frp->fr_end = max; 595 596 if (!drop) { 597 /* XXX Optimization opportunity 598 * This is a very heavy way to trim the payload. 599 * we could do it much faster by diddling mbuf 600 * internals but that would be even less legible 601 * than this mbuf magic. For my next trick, 602 * I'll pull a rabbit out of my laptop. 603 */ 604 *m0 = m_dup(m, MB_DONTWAIT); 605 /* From KAME Project : We have missed this! */ 606 m_adj(*m0, (h->ip_hl << 2) - 607 (*m0)->m_pkthdr.len); 608 if (*m0 == NULL) 609 goto no_mem; 610 KASSERT(((*m0)->m_next == NULL), 611 ("(*m0)->m_next != NULL: %s", 612 __func__)); 613 m_adj(m, precut + (h->ip_hl << 2)); 614 m_cat(*m0, m); 615 m = *m0; 616 if (m->m_flags & M_PKTHDR) { 617 int plen = 0; 618 struct mbuf *t; 619 for (t = m; t; t = t->m_next) 620 plen += t->m_len; 621 m->m_pkthdr.len = plen; 622 } 623 624 625 h = mtod(m, struct ip *); 626 627 KASSERT(((int)m->m_len == 628 h->ip_len - precut), 629 ("m->m_len != h->ip_len - precut: %s", 630 __func__)); 631 h->ip_off = h->ip_off + 632 (precut >> 3); 633 h->ip_len = h->ip_len - precut; 634 } else { 635 hosed++; 636 } 637 } else { 638 /* There is a gap between fragments */ 639 640 DPFPRINTF(("fragcache[%d]: gap %d (%d-%d) %d-%d\n", 641 h->ip_id, -precut, frp->fr_off, frp->fr_end, off, 642 max)); 643 644 cur = kmalloc(sizeof(struct pf_frcache), M_PFCENTPL, M_NOWAIT); 645 if (cur == NULL) 646 goto no_mem; 647 pf_ncache++; 648 649 cur->fr_off = off; 650 cur->fr_end = max; 651 LIST_INSERT_AFTER(frp, cur, fr_next); 652 } 653 } 654 655 if (fra != NULL) { 656 int aftercut; 657 int merge = 0; 658 659 aftercut = max - fra->fr_off; 660 if (aftercut == 0) { 661 /* Adjacent fragments */ 662 DPFPRINTF(("fragcache[%d]: adjacent %d-%d (%d-%d)\n", 663 h->ip_id, off, max, fra->fr_off, fra->fr_end)); 664 fra->fr_off = off; 665 merge = 1; 666 } else if (aftercut > 0) { 667 /* Need to chop off the tail of this fragment */ 668 DPFPRINTF(("fragcache[%d]: chop %d %d-%d (%d-%d)\n", 669 h->ip_id, aftercut, off, max, fra->fr_off, 670 fra->fr_end)); 671 fra->fr_off = off; 672 max -= aftercut; 673 674 merge = 1; 675 676 if (!drop) { 677 m_adj(m, -aftercut); 678 if (m->m_flags & M_PKTHDR) { 679 int plen = 0; 680 struct mbuf *t; 681 for (t = m; t; t = t->m_next) 682 plen += t->m_len; 683 m->m_pkthdr.len = plen; 684 } 685 h = mtod(m, struct ip *); 686 KASSERT(((int)m->m_len == h->ip_len - aftercut), 687 ("m->m_len != h->ip_len - aftercut: %s", 688 __func__)); 689 h->ip_len = h->ip_len - aftercut; 690 } else { 691 hosed++; 692 } 693 } else if (frp == NULL) { 694 /* There is a gap between fragments */ 695 DPFPRINTF(("fragcache[%d]: gap %d %d-%d (%d-%d)\n", 696 h->ip_id, -aftercut, off, max, fra->fr_off, 697 fra->fr_end)); 698 699 cur = kmalloc(sizeof(struct pf_frcache), M_PFCENTPL, M_NOWAIT); 700 if (cur == NULL) 701 goto no_mem; 702 pf_ncache++; 703 704 cur->fr_off = off; 705 cur->fr_end = max; 706 LIST_INSERT_BEFORE(fra, cur, fr_next); 707 } 708 709 710 /* Need to glue together two separate fragment descriptors */ 711 if (merge) { 712 if (cur && fra->fr_off <= cur->fr_end) { 713 /* Need to merge in a previous 'cur' */ 714 DPFPRINTF(("fragcache[%d]: adjacent(merge " 715 "%d-%d) %d-%d (%d-%d)\n", 716 h->ip_id, cur->fr_off, cur->fr_end, off, 717 max, fra->fr_off, fra->fr_end)); 718 fra->fr_off = cur->fr_off; 719 LIST_REMOVE(cur, fr_next); 720 kfree(cur, M_PFCENTPL); 721 pf_ncache--; 722 cur = NULL; 723 724 } else if (frp && fra->fr_off <= frp->fr_end) { 725 /* Need to merge in a modified 'frp' */ 726 KASSERT((cur == NULL), ("cur != NULL: %s", 727 __func__)); 728 DPFPRINTF(("fragcache[%d]: adjacent(merge " 729 "%d-%d) %d-%d (%d-%d)\n", 730 h->ip_id, frp->fr_off, frp->fr_end, off, 731 max, fra->fr_off, fra->fr_end)); 732 fra->fr_off = frp->fr_off; 733 LIST_REMOVE(frp, fr_next); 734 kfree(frp, M_PFCENTPL); 735 pf_ncache--; 736 frp = NULL; 737 738 } 739 } 740 } 741 742 if (hosed) { 743 /* 744 * We must keep tracking the overall fragment even when 745 * we're going to drop it anyway so that we know when to 746 * free the overall descriptor. Thus we drop the frag late. 747 */ 748 goto drop_fragment; 749 } 750 751 752 pass: 753 /* Update maximum data size */ 754 if ((*frag)->fr_max < max) 755 (*frag)->fr_max = max; 756 757 /* This is the last segment */ 758 if (!mff) 759 (*frag)->fr_flags |= PFFRAG_SEENLAST; 760 761 /* Check if we are completely reassembled */ 762 if (((*frag)->fr_flags & PFFRAG_SEENLAST) && 763 LIST_FIRST(&(*frag)->fr_cache)->fr_off == 0 && 764 LIST_FIRST(&(*frag)->fr_cache)->fr_end == (*frag)->fr_max) { 765 /* Remove from fragment queue */ 766 DPFPRINTF(("fragcache[%d]: done 0-%d\n", h->ip_id, 767 (*frag)->fr_max)); 768 pf_free_fragment(*frag); 769 *frag = NULL; 770 } 771 772 return (m); 773 774 no_mem: 775 *nomem = 1; 776 777 /* Still need to pay attention to !IP_MF */ 778 if (!mff && *frag != NULL) 779 (*frag)->fr_flags |= PFFRAG_SEENLAST; 780 781 m_freem(m); 782 return (NULL); 783 784 drop_fragment: 785 786 /* Still need to pay attention to !IP_MF */ 787 if (!mff && *frag != NULL) 788 (*frag)->fr_flags |= PFFRAG_SEENLAST; 789 790 if (drop) { 791 /* This fragment has been deemed bad. Don't reass */ 792 if (((*frag)->fr_flags & PFFRAG_DROP) == 0) 793 DPFPRINTF(("fragcache[%d]: dropping overall fragment\n", 794 h->ip_id)); 795 (*frag)->fr_flags |= PFFRAG_DROP; 796 } 797 798 m_freem(m); 799 return (NULL); 800 } 801 802 int 803 pf_normalize_ip(struct mbuf **m0, int dir, struct pfi_kif *kif, u_short *reason, 804 struct pf_pdesc *pd) 805 { 806 struct mbuf *m = *m0; 807 struct pf_rule *r; 808 struct pf_frent *frent; 809 struct pf_fragment *frag = NULL; 810 struct ip *h = mtod(m, struct ip *); 811 int mff = (h->ip_off & IP_MF); 812 int hlen = h->ip_hl << 2; 813 u_int16_t fragoff = (h->ip_off & IP_OFFMASK) << 3; 814 u_int16_t max; 815 int ip_len; 816 int ip_off; 817 int tag = -1; 818 819 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr); 820 while (r != NULL) { 821 r->evaluations++; 822 if (pfi_kif_match(r->kif, kif) == r->ifnot) 823 r = r->skip[PF_SKIP_IFP].ptr; 824 else if (r->direction && r->direction != dir) 825 r = r->skip[PF_SKIP_DIR].ptr; 826 else if (r->af && r->af != AF_INET) 827 r = r->skip[PF_SKIP_AF].ptr; 828 else if (r->proto && r->proto != h->ip_p) 829 r = r->skip[PF_SKIP_PROTO].ptr; 830 else if (PF_MISMATCHAW(&r->src.addr, 831 (struct pf_addr *)&h->ip_src.s_addr, AF_INET, 832 r->src.neg, kif)) 833 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 834 else if (PF_MISMATCHAW(&r->dst.addr, 835 (struct pf_addr *)&h->ip_dst.s_addr, AF_INET, 836 r->dst.neg, NULL)) 837 r = r->skip[PF_SKIP_DST_ADDR].ptr; 838 else if (r->match_tag && !pf_match_tag(m, r, &tag)) 839 r = TAILQ_NEXT(r, entries); 840 else 841 break; 842 } 843 844 if (r == NULL || r->action == PF_NOSCRUB) 845 return (PF_PASS); 846 else { 847 r->packets[dir == PF_OUT]++; 848 r->bytes[dir == PF_OUT] += pd->tot_len; 849 } 850 851 /* Check for illegal packets */ 852 if (hlen < (int)sizeof(struct ip)) 853 goto drop; 854 855 if (hlen > h->ip_len) 856 goto drop; 857 858 /* Clear IP_DF if the rule uses the no-df option */ 859 if (r->rule_flag & PFRULE_NODF && h->ip_off & IP_DF) { 860 u_int16_t ip_off = h->ip_off; 861 862 h->ip_off &= ~IP_DF; 863 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0); 864 } 865 866 /* We will need other tests here */ 867 if (!fragoff && !mff) 868 goto no_fragment; 869 870 /* We're dealing with a fragment now. Don't allow fragments 871 * with IP_DF to enter the cache. If the flag was cleared by 872 * no-df above, fine. Otherwise drop it. 873 */ 874 if (h->ip_off & IP_DF) { 875 DPFPRINTF(("IP_DF\n")); 876 goto bad; 877 } 878 879 ip_len = h->ip_len - hlen; 880 ip_off = (h->ip_off & IP_OFFMASK) << 3; 881 882 /* All fragments are 8 byte aligned */ 883 if (mff && (ip_len & 0x7)) { 884 DPFPRINTF(("mff and %d\n", ip_len)); 885 goto bad; 886 } 887 888 /* Respect maximum length */ 889 if (fragoff + ip_len > IP_MAXPACKET) { 890 DPFPRINTF(("max packet %d\n", fragoff + ip_len)); 891 goto bad; 892 } 893 max = fragoff + ip_len; 894 895 if ((r->rule_flag & (PFRULE_FRAGCROP|PFRULE_FRAGDROP)) == 0) { 896 /* Fully buffer all of the fragments */ 897 898 frag = pf_find_fragment(h, &pf_frag_tree); 899 900 /* Check if we saw the last fragment already */ 901 if (frag != NULL && (frag->fr_flags & PFFRAG_SEENLAST) && 902 max > frag->fr_max) 903 goto bad; 904 905 /* Get an entry for the fragment queue */ 906 frent = kmalloc(sizeof(struct pf_frent), M_PFFRENTPL, M_NOWAIT); 907 if (frent == NULL) { 908 REASON_SET(reason, PFRES_MEMORY); 909 return (PF_DROP); 910 } 911 pf_nfrents++; 912 frent->fr_ip = h; 913 frent->fr_m = m; 914 915 /* Might return a completely reassembled mbuf, or NULL */ 916 DPFPRINTF(("reass frag %d @ %d-%d\n", h->ip_id, fragoff, max)); 917 *m0 = m = pf_reassemble(m0, &frag, frent, mff); 918 919 if (m == NULL) 920 return (PF_DROP); 921 922 if (frag != NULL && (frag->fr_flags & PFFRAG_DROP)) 923 goto drop; 924 925 h = mtod(m, struct ip *); 926 } else { 927 /* non-buffering fragment cache (drops or masks overlaps) */ 928 int nomem = 0; 929 930 if (dir == PF_OUT && m->m_pkthdr.pf.flags & PF_TAG_FRAGCACHE) { 931 /* 932 * Already passed the fragment cache in the 933 * input direction. If we continued, it would 934 * appear to be a dup and would be dropped. 935 */ 936 goto fragment_pass; 937 } 938 939 frag = pf_find_fragment(h, &pf_cache_tree); 940 941 /* Check if we saw the last fragment already */ 942 if (frag != NULL && (frag->fr_flags & PFFRAG_SEENLAST) && 943 max > frag->fr_max) { 944 if (r->rule_flag & PFRULE_FRAGDROP) 945 frag->fr_flags |= PFFRAG_DROP; 946 goto bad; 947 } 948 949 *m0 = m = pf_fragcache(m0, h, &frag, mff, 950 (r->rule_flag & PFRULE_FRAGDROP) ? 1 : 0, &nomem); 951 if (m == NULL) { 952 if (nomem) 953 goto no_mem; 954 goto drop; 955 } 956 957 if (dir == PF_IN) 958 m->m_pkthdr.pf.flags |= PF_TAG_FRAGCACHE; 959 960 if (frag != NULL && (frag->fr_flags & PFFRAG_DROP)) 961 goto drop; 962 goto fragment_pass; 963 } 964 965 no_fragment: 966 /* At this point, only IP_DF is allowed in ip_off */ 967 if (h->ip_off & ~IP_DF) { 968 u_int16_t ip_off = h->ip_off; 969 970 h->ip_off &= IP_DF; 971 h->ip_sum = pf_cksum_fixup(h->ip_sum, htons(ip_off), htons(h->ip_off), 0); 972 } 973 974 /* Enforce a minimum ttl, may cause endless packet loops */ 975 if (r->min_ttl && h->ip_ttl < r->min_ttl) { 976 u_int16_t ip_ttl = h->ip_ttl; 977 978 h->ip_ttl = r->min_ttl; 979 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_ttl, h->ip_ttl, 0); 980 } 981 982 /* Enforce tos */ 983 if (r->rule_flag & PFRULE_SET_TOS) { 984 u_int16_t ov, nv; 985 986 ov = *(u_int16_t *)h; 987 h->ip_tos = r->set_tos; 988 nv = *(u_int16_t *)h; 989 990 h->ip_sum = pf_cksum_fixup(h->ip_sum, ov, nv, 0); 991 } 992 993 if (r->rule_flag & PFRULE_RANDOMID) { 994 u_int16_t ip_id = h->ip_id; 995 996 h->ip_id = ip_randomid(); 997 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_id, h->ip_id, 0); 998 } 999 if ((r->rule_flag & (PFRULE_FRAGCROP|PFRULE_FRAGDROP)) == 0) 1000 pd->flags |= PFDESC_IP_REAS; 1001 1002 return (PF_PASS); 1003 1004 fragment_pass: 1005 /* Enforce a minimum ttl, may cause endless packet loops */ 1006 if (r->min_ttl && h->ip_ttl < r->min_ttl) { 1007 u_int16_t ip_ttl = h->ip_ttl; 1008 1009 h->ip_ttl = r->min_ttl; 1010 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_ttl, h->ip_ttl, 0); 1011 } 1012 /* Enforce tos */ 1013 if (r->rule_flag & PFRULE_SET_TOS) { 1014 u_int16_t ov, nv; 1015 1016 ov = *(u_int16_t *)h; 1017 h->ip_tos = r->set_tos; 1018 nv = *(u_int16_t *)h; 1019 1020 h->ip_sum = pf_cksum_fixup(h->ip_sum, ov, nv, 0); 1021 } 1022 if ((r->rule_flag & (PFRULE_FRAGCROP|PFRULE_FRAGDROP)) == 0) 1023 pd->flags |= PFDESC_IP_REAS; 1024 return (PF_PASS); 1025 1026 no_mem: 1027 REASON_SET(reason, PFRES_MEMORY); 1028 if (r != NULL && r->log) 1029 PFLOG_PACKET(kif, h, m, AF_INET, dir, *reason, r, NULL, NULL, pd); 1030 return (PF_DROP); 1031 1032 drop: 1033 REASON_SET(reason, PFRES_NORM); 1034 if (r != NULL && r->log) 1035 PFLOG_PACKET(kif, h, m, AF_INET, dir, *reason, r, NULL, NULL, pd); 1036 return (PF_DROP); 1037 1038 bad: 1039 DPFPRINTF(("dropping bad fragment\n")); 1040 1041 /* Free associated fragments */ 1042 if (frag != NULL) 1043 pf_free_fragment(frag); 1044 1045 REASON_SET(reason, PFRES_FRAG); 1046 if (r != NULL && r->log) 1047 PFLOG_PACKET(kif, h, m, AF_INET, dir, *reason, r, NULL, NULL, pd); 1048 1049 return (PF_DROP); 1050 } 1051 1052 #ifdef INET6 1053 int 1054 pf_normalize_ip6(struct mbuf **m0, int dir, struct pfi_kif *kif, 1055 u_short *reason, struct pf_pdesc *pd) 1056 { 1057 struct mbuf *m = *m0; 1058 struct pf_rule *r; 1059 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 1060 int off; 1061 struct ip6_ext ext; 1062 struct ip6_opt opt; 1063 struct ip6_opt_jumbo jumbo; 1064 struct ip6_frag frag; 1065 u_int32_t jumbolen = 0, plen; 1066 u_int16_t fragoff = 0; 1067 int optend; 1068 int ooff; 1069 u_int8_t proto; 1070 int terminal; 1071 1072 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr); 1073 while (r != NULL) { 1074 r->evaluations++; 1075 if (pfi_kif_match(r->kif, kif) == r->ifnot) 1076 r = r->skip[PF_SKIP_IFP].ptr; 1077 else if (r->direction && r->direction != dir) 1078 r = r->skip[PF_SKIP_DIR].ptr; 1079 else if (r->af && r->af != AF_INET6) 1080 r = r->skip[PF_SKIP_AF].ptr; 1081 #if 0 /* header chain! */ 1082 else if (r->proto && r->proto != h->ip6_nxt) 1083 r = r->skip[PF_SKIP_PROTO].ptr; 1084 #endif 1085 else if (PF_MISMATCHAW(&r->src.addr, 1086 (struct pf_addr *)&h->ip6_src, AF_INET6, 1087 r->src.neg, kif)) 1088 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 1089 else if (PF_MISMATCHAW(&r->dst.addr, 1090 (struct pf_addr *)&h->ip6_dst, AF_INET6, 1091 r->dst.neg, NULL)) 1092 r = r->skip[PF_SKIP_DST_ADDR].ptr; 1093 else 1094 break; 1095 } 1096 1097 if (r == NULL || r->action == PF_NOSCRUB) 1098 return (PF_PASS); 1099 else { 1100 r->packets[dir == PF_OUT]++; 1101 r->bytes[dir == PF_OUT] += pd->tot_len; 1102 } 1103 1104 /* Check for illegal packets */ 1105 if (sizeof(struct ip6_hdr) + IPV6_MAXPACKET < m->m_pkthdr.len) 1106 goto drop; 1107 1108 off = sizeof(struct ip6_hdr); 1109 proto = h->ip6_nxt; 1110 terminal = 0; 1111 do { 1112 switch (proto) { 1113 case IPPROTO_FRAGMENT: 1114 goto fragment; 1115 break; 1116 case IPPROTO_AH: 1117 case IPPROTO_ROUTING: 1118 case IPPROTO_DSTOPTS: 1119 if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL, 1120 NULL, AF_INET6)) 1121 goto shortpkt; 1122 if (proto == IPPROTO_AH) 1123 off += (ext.ip6e_len + 2) * 4; 1124 else 1125 off += (ext.ip6e_len + 1) * 8; 1126 proto = ext.ip6e_nxt; 1127 break; 1128 case IPPROTO_HOPOPTS: 1129 if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL, 1130 NULL, AF_INET6)) 1131 goto shortpkt; 1132 optend = off + (ext.ip6e_len + 1) * 8; 1133 ooff = off + sizeof(ext); 1134 do { 1135 if (!pf_pull_hdr(m, ooff, &opt.ip6o_type, 1136 sizeof(opt.ip6o_type), NULL, NULL, 1137 AF_INET6)) 1138 goto shortpkt; 1139 if (opt.ip6o_type == IP6OPT_PAD1) { 1140 ooff++; 1141 continue; 1142 } 1143 if (!pf_pull_hdr(m, ooff, &opt, sizeof(opt), 1144 NULL, NULL, AF_INET6)) 1145 goto shortpkt; 1146 if (ooff + sizeof(opt) + opt.ip6o_len > optend) 1147 goto drop; 1148 switch (opt.ip6o_type) { 1149 case IP6OPT_JUMBO: 1150 if (h->ip6_plen != 0) 1151 goto drop; 1152 if (!pf_pull_hdr(m, ooff, &jumbo, 1153 sizeof(jumbo), NULL, NULL, 1154 AF_INET6)) 1155 goto shortpkt; 1156 memcpy(&jumbolen, jumbo.ip6oj_jumbo_len, 1157 sizeof(jumbolen)); 1158 jumbolen = ntohl(jumbolen); 1159 if (jumbolen <= IPV6_MAXPACKET) 1160 goto drop; 1161 if (sizeof(struct ip6_hdr) + jumbolen != 1162 m->m_pkthdr.len) 1163 goto drop; 1164 break; 1165 default: 1166 break; 1167 } 1168 ooff += sizeof(opt) + opt.ip6o_len; 1169 } while (ooff < optend); 1170 1171 off = optend; 1172 proto = ext.ip6e_nxt; 1173 break; 1174 default: 1175 terminal = 1; 1176 break; 1177 } 1178 } while (!terminal); 1179 1180 /* jumbo payload option must be present, or plen > 0 */ 1181 if (ntohs(h->ip6_plen) == 0) 1182 plen = jumbolen; 1183 else 1184 plen = ntohs(h->ip6_plen); 1185 if (plen == 0) 1186 goto drop; 1187 if (sizeof(struct ip6_hdr) + plen > m->m_pkthdr.len) 1188 goto shortpkt; 1189 1190 /* Enforce a minimum ttl, may cause endless packet loops */ 1191 if (r->min_ttl && h->ip6_hlim < r->min_ttl) 1192 h->ip6_hlim = r->min_ttl; 1193 1194 return (PF_PASS); 1195 1196 fragment: 1197 if (ntohs(h->ip6_plen) == 0 || jumbolen) 1198 goto drop; 1199 plen = ntohs(h->ip6_plen); 1200 1201 if (!pf_pull_hdr(m, off, &frag, sizeof(frag), NULL, NULL, AF_INET6)) 1202 goto shortpkt; 1203 fragoff = ntohs(frag.ip6f_offlg & IP6F_OFF_MASK); 1204 if (fragoff + (plen - off - sizeof(frag)) > IPV6_MAXPACKET) 1205 goto badfrag; 1206 1207 /* do something about it */ 1208 /* remember to set pd->flags |= PFDESC_IP_REAS */ 1209 return (PF_PASS); 1210 1211 shortpkt: 1212 REASON_SET(reason, PFRES_SHORT); 1213 if (r != NULL && r->log) 1214 PFLOG_PACKET(kif, h, m, AF_INET6, dir, *reason, r, NULL, NULL, pd); 1215 return (PF_DROP); 1216 1217 drop: 1218 REASON_SET(reason, PFRES_NORM); 1219 if (r != NULL && r->log) 1220 PFLOG_PACKET(kif, h, m, AF_INET6, dir, *reason, r, NULL, NULL, pd); 1221 return (PF_DROP); 1222 1223 badfrag: 1224 REASON_SET(reason, PFRES_FRAG); 1225 if (r != NULL && r->log) 1226 PFLOG_PACKET(kif, h, m, AF_INET6, dir, *reason, r, NULL, NULL, pd); 1227 return (PF_DROP); 1228 } 1229 #endif /* INET6 */ 1230 1231 int 1232 pf_normalize_tcp(int dir, struct pfi_kif *kif, struct mbuf *m, int ipoff, 1233 int off, void *h, struct pf_pdesc *pd) 1234 { 1235 struct pf_rule *r, *rm = NULL; 1236 struct tcphdr *th = pd->hdr.tcp; 1237 int rewrite = 0; 1238 u_short reason; 1239 u_int8_t flags; 1240 sa_family_t af = pd->af; 1241 1242 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr); 1243 while (r != NULL) { 1244 r->evaluations++; 1245 if (pfi_kif_match(r->kif, kif) == r->ifnot) 1246 r = r->skip[PF_SKIP_IFP].ptr; 1247 else if (r->direction && r->direction != dir) 1248 r = r->skip[PF_SKIP_DIR].ptr; 1249 else if (r->af && r->af != af) 1250 r = r->skip[PF_SKIP_AF].ptr; 1251 else if (r->proto && r->proto != pd->proto) 1252 r = r->skip[PF_SKIP_PROTO].ptr; 1253 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af, 1254 r->src.neg, kif)) 1255 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 1256 else if (r->src.port_op && !pf_match_port(r->src.port_op, 1257 r->src.port[0], r->src.port[1], th->th_sport)) 1258 r = r->skip[PF_SKIP_SRC_PORT].ptr; 1259 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af, 1260 r->dst.neg, NULL)) 1261 r = r->skip[PF_SKIP_DST_ADDR].ptr; 1262 else if (r->dst.port_op && !pf_match_port(r->dst.port_op, 1263 r->dst.port[0], r->dst.port[1], th->th_dport)) 1264 r = r->skip[PF_SKIP_DST_PORT].ptr; 1265 else if (r->os_fingerprint != PF_OSFP_ANY && !pf_osfp_match( 1266 pf_osfp_fingerprint(pd, m, off, th), 1267 r->os_fingerprint)) 1268 r = TAILQ_NEXT(r, entries); 1269 else { 1270 rm = r; 1271 break; 1272 } 1273 } 1274 1275 if (rm == NULL || rm->action == PF_NOSCRUB) 1276 return (PF_PASS); 1277 else { 1278 r->packets[dir == PF_OUT]++; 1279 r->bytes[dir == PF_OUT] += pd->tot_len; 1280 } 1281 1282 if (rm->rule_flag & PFRULE_REASSEMBLE_TCP) 1283 pd->flags |= PFDESC_TCP_NORM; 1284 1285 flags = th->th_flags; 1286 if (flags & TH_SYN) { 1287 /* Illegal packet */ 1288 if (flags & TH_RST) 1289 goto tcp_drop; 1290 1291 if (flags & TH_FIN) 1292 flags &= ~TH_FIN; 1293 } else { 1294 /* Illegal packet */ 1295 if (!(flags & (TH_ACK|TH_RST))) 1296 goto tcp_drop; 1297 } 1298 1299 if (!(flags & TH_ACK)) { 1300 /* These flags are only valid if ACK is set */ 1301 if ((flags & TH_FIN) || (flags & TH_PUSH) || (flags & TH_URG)) 1302 goto tcp_drop; 1303 } 1304 1305 /* Check for illegal header length */ 1306 if (th->th_off < (sizeof(struct tcphdr) >> 2)) 1307 goto tcp_drop; 1308 1309 /* If flags changed, or reserved data set, then adjust */ 1310 if (flags != th->th_flags || th->th_x2 != 0) { 1311 u_int16_t ov, nv; 1312 1313 ov = *(u_int16_t *)(&th->th_ack + 1); 1314 th->th_flags = flags; 1315 th->th_x2 = 0; 1316 nv = *(u_int16_t *)(&th->th_ack + 1); 1317 1318 th->th_sum = pf_cksum_fixup(th->th_sum, ov, nv, 0); 1319 rewrite = 1; 1320 } 1321 1322 /* Remove urgent pointer, if TH_URG is not set */ 1323 if (!(flags & TH_URG) && th->th_urp) { 1324 th->th_sum = pf_cksum_fixup(th->th_sum, th->th_urp, 0, 0); 1325 th->th_urp = 0; 1326 rewrite = 1; 1327 } 1328 1329 /* Process options */ 1330 if (r->max_mss && pf_normalize_tcpopt(r, m, th, off, pd->af)) 1331 rewrite = 1; 1332 1333 /* copy back packet headers if we sanitized */ 1334 if (rewrite) 1335 m_copyback(m, off, sizeof(*th), (caddr_t)th); 1336 1337 return (PF_PASS); 1338 1339 tcp_drop: 1340 REASON_SET(&reason, PFRES_NORM); 1341 if (rm != NULL && r->log) 1342 PFLOG_PACKET(kif, h, m, AF_INET, dir, reason, r, NULL, NULL, pd); 1343 return (PF_DROP); 1344 } 1345 1346 int 1347 pf_normalize_tcp_init(struct mbuf *m, int off, struct pf_pdesc *pd, 1348 struct tcphdr *th, struct pf_state_peer *src, struct pf_state_peer *dst) 1349 { 1350 u_int32_t tsval, tsecr; 1351 u_int8_t hdr[60]; 1352 u_int8_t *opt; 1353 1354 KASSERT((src->scrub == NULL), 1355 ("pf_normalize_tcp_init: src->scrub != NULL")); 1356 1357 src->scrub = kmalloc(sizeof(struct pf_state_scrub), M_PFSTATESCRUBPL, 1358 M_NOWAIT | M_ZERO); 1359 if (src->scrub == NULL) 1360 return (1); 1361 1362 switch (pd->af) { 1363 #ifdef INET 1364 case AF_INET: { 1365 struct ip *h = mtod(m, struct ip *); 1366 src->scrub->pfss_ttl = h->ip_ttl; 1367 break; 1368 } 1369 #endif /* INET */ 1370 #ifdef INET6 1371 case AF_INET6: { 1372 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 1373 src->scrub->pfss_ttl = h->ip6_hlim; 1374 break; 1375 } 1376 #endif /* INET6 */ 1377 } 1378 1379 1380 /* 1381 * All normalizations below are only begun if we see the start of 1382 * the connections. They must all set an enabled bit in pfss_flags 1383 */ 1384 if ((th->th_flags & TH_SYN) == 0) 1385 return (0); 1386 1387 1388 if (th->th_off > (sizeof(struct tcphdr) >> 2) && src->scrub && 1389 pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) { 1390 /* Diddle with TCP options */ 1391 int hlen; 1392 opt = hdr + sizeof(struct tcphdr); 1393 hlen = (th->th_off << 2) - sizeof(struct tcphdr); 1394 while (hlen >= TCPOLEN_TIMESTAMP) { 1395 switch (*opt) { 1396 case TCPOPT_EOL: /* FALLTHROUGH */ 1397 case TCPOPT_NOP: 1398 opt++; 1399 hlen--; 1400 break; 1401 case TCPOPT_TIMESTAMP: 1402 if (opt[1] >= TCPOLEN_TIMESTAMP) { 1403 src->scrub->pfss_flags |= 1404 PFSS_TIMESTAMP; 1405 src->scrub->pfss_ts_mod = karc4random(); 1406 1407 /* note PFSS_PAWS not set yet */ 1408 memcpy(&tsval, &opt[2], 1409 sizeof(u_int32_t)); 1410 memcpy(&tsecr, &opt[6], 1411 sizeof(u_int32_t)); 1412 src->scrub->pfss_tsval0 = ntohl(tsval); 1413 src->scrub->pfss_tsval = ntohl(tsval); 1414 src->scrub->pfss_tsecr = ntohl(tsecr); 1415 getmicrouptime(&src->scrub->pfss_last); 1416 } 1417 /* FALLTHROUGH */ 1418 default: 1419 hlen -= MAX(opt[1], 2); 1420 opt += MAX(opt[1], 2); 1421 break; 1422 } 1423 } 1424 } 1425 1426 return (0); 1427 } 1428 1429 void 1430 pf_normalize_tcp_cleanup(struct pf_state *state) 1431 { 1432 if (state->src.scrub) 1433 kfree(state->src.scrub, M_PFSTATESCRUBPL); 1434 if (state->dst.scrub) 1435 kfree(state->dst.scrub, M_PFSTATESCRUBPL); 1436 1437 /* Someday... flush the TCP segment reassembly descriptors. */ 1438 } 1439 1440 int 1441 pf_normalize_tcp_stateful(struct mbuf *m, int off, struct pf_pdesc *pd, 1442 u_short *reason, struct tcphdr *th, struct pf_state *state, 1443 struct pf_state_peer *src, struct pf_state_peer *dst, int *writeback) 1444 { 1445 struct timeval uptime; 1446 u_int32_t tsval, tsecr; 1447 u_int tsval_from_last; 1448 u_int8_t hdr[60]; 1449 u_int8_t *opt; 1450 int copyback = 0; 1451 int got_ts = 0; 1452 1453 KASSERT((src->scrub || dst->scrub), 1454 ("pf_normalize_tcp_statefull: src->scrub && dst->scrub!")); 1455 1456 /* 1457 * Enforce the minimum TTL seen for this connection. Negate a common 1458 * technique to evade an intrusion detection system and confuse 1459 * firewall state code. 1460 */ 1461 switch (pd->af) { 1462 #ifdef INET 1463 case AF_INET: { 1464 if (src->scrub) { 1465 struct ip *h = mtod(m, struct ip *); 1466 if (h->ip_ttl > src->scrub->pfss_ttl) 1467 src->scrub->pfss_ttl = h->ip_ttl; 1468 h->ip_ttl = src->scrub->pfss_ttl; 1469 } 1470 break; 1471 } 1472 #endif /* INET */ 1473 #ifdef INET6 1474 case AF_INET6: { 1475 if (src->scrub) { 1476 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 1477 if (h->ip6_hlim > src->scrub->pfss_ttl) 1478 src->scrub->pfss_ttl = h->ip6_hlim; 1479 h->ip6_hlim = src->scrub->pfss_ttl; 1480 } 1481 break; 1482 } 1483 #endif /* INET6 */ 1484 } 1485 1486 if (th->th_off > (sizeof(struct tcphdr) >> 2) && 1487 ((src->scrub && (src->scrub->pfss_flags & PFSS_TIMESTAMP)) || 1488 (dst->scrub && (dst->scrub->pfss_flags & PFSS_TIMESTAMP))) && 1489 pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) { 1490 /* Diddle with TCP options */ 1491 int hlen; 1492 opt = hdr + sizeof(struct tcphdr); 1493 hlen = (th->th_off << 2) - sizeof(struct tcphdr); 1494 while (hlen >= TCPOLEN_TIMESTAMP) { 1495 switch (*opt) { 1496 case TCPOPT_EOL: /* FALLTHROUGH */ 1497 case TCPOPT_NOP: 1498 opt++; 1499 hlen--; 1500 break; 1501 case TCPOPT_TIMESTAMP: 1502 /* Modulate the timestamps. Can be used for 1503 * NAT detection, OS uptime determination or 1504 * reboot detection. 1505 */ 1506 1507 if (got_ts) { 1508 /* Huh? Multiple timestamps!? */ 1509 if (pf_status.debug >= PF_DEBUG_MISC) { 1510 DPFPRINTF(("multiple TS??")); 1511 pf_print_state(state); 1512 kprintf("\n"); 1513 } 1514 REASON_SET(reason, PFRES_TS); 1515 return (PF_DROP); 1516 } 1517 if (opt[1] >= TCPOLEN_TIMESTAMP) { 1518 memcpy(&tsval, &opt[2], 1519 sizeof(u_int32_t)); 1520 if (tsval && src->scrub && 1521 (src->scrub->pfss_flags & 1522 PFSS_TIMESTAMP)) { 1523 tsval = ntohl(tsval); 1524 pf_change_a(&opt[2], 1525 &th->th_sum, 1526 htonl(tsval + 1527 src->scrub->pfss_ts_mod), 1528 0); 1529 copyback = 1; 1530 } 1531 1532 /* Modulate TS reply iff valid (!0) */ 1533 memcpy(&tsecr, &opt[6], 1534 sizeof(u_int32_t)); 1535 if (tsecr && dst->scrub && 1536 (dst->scrub->pfss_flags & 1537 PFSS_TIMESTAMP)) { 1538 tsecr = ntohl(tsecr) 1539 - dst->scrub->pfss_ts_mod; 1540 pf_change_a(&opt[6], 1541 &th->th_sum, htonl(tsecr), 1542 0); 1543 copyback = 1; 1544 } 1545 got_ts = 1; 1546 } 1547 /* FALLTHROUGH */ 1548 default: 1549 hlen -= MAX(opt[1], 2); 1550 opt += MAX(opt[1], 2); 1551 break; 1552 } 1553 } 1554 if (copyback) { 1555 /* Copyback the options, caller copys back header */ 1556 *writeback = 1; 1557 m_copyback(m, off + sizeof(struct tcphdr), 1558 (th->th_off << 2) - sizeof(struct tcphdr), hdr + 1559 sizeof(struct tcphdr)); 1560 } 1561 } 1562 1563 1564 /* 1565 * Must invalidate PAWS checks on connections idle for too long. 1566 * The fastest allowed timestamp clock is 1ms. That turns out to 1567 * be about 24 days before it wraps. XXX Right now our lowerbound 1568 * TS echo check only works for the first 12 days of a connection 1569 * when the TS has exhausted half its 32bit space 1570 */ 1571 #define TS_MAX_IDLE (24*24*60*60) 1572 #define TS_MAX_CONN (12*24*60*60) /* XXX remove when better tsecr check */ 1573 1574 getmicrouptime(&uptime); 1575 if (src->scrub && (src->scrub->pfss_flags & PFSS_PAWS) && 1576 (uptime.tv_sec - src->scrub->pfss_last.tv_sec > TS_MAX_IDLE || 1577 time_second - state->creation > TS_MAX_CONN)) { 1578 if (pf_status.debug >= PF_DEBUG_MISC) { 1579 DPFPRINTF(("src idled out of PAWS\n")); 1580 pf_print_state(state); 1581 kprintf("\n"); 1582 } 1583 src->scrub->pfss_flags = (src->scrub->pfss_flags & ~PFSS_PAWS) 1584 | PFSS_PAWS_IDLED; 1585 } 1586 if (dst->scrub && (dst->scrub->pfss_flags & PFSS_PAWS) && 1587 uptime.tv_sec - dst->scrub->pfss_last.tv_sec > TS_MAX_IDLE) { 1588 if (pf_status.debug >= PF_DEBUG_MISC) { 1589 DPFPRINTF(("dst idled out of PAWS\n")); 1590 pf_print_state(state); 1591 kprintf("\n"); 1592 } 1593 dst->scrub->pfss_flags = (dst->scrub->pfss_flags & ~PFSS_PAWS) 1594 | PFSS_PAWS_IDLED; 1595 } 1596 1597 if (got_ts && src->scrub && dst->scrub && 1598 (src->scrub->pfss_flags & PFSS_PAWS) && 1599 (dst->scrub->pfss_flags & PFSS_PAWS)) { 1600 /* Validate that the timestamps are "in-window". 1601 * RFC1323 describes TCP Timestamp options that allow 1602 * measurement of RTT (round trip time) and PAWS 1603 * (protection against wrapped sequence numbers). PAWS 1604 * gives us a set of rules for rejecting packets on 1605 * long fat pipes (packets that were somehow delayed 1606 * in transit longer than the time it took to send the 1607 * full TCP sequence space of 4Gb). We can use these 1608 * rules and infer a few others that will let us treat 1609 * the 32bit timestamp and the 32bit echoed timestamp 1610 * as sequence numbers to prevent a blind attacker from 1611 * inserting packets into a connection. 1612 * 1613 * RFC1323 tells us: 1614 * - The timestamp on this packet must be greater than 1615 * or equal to the last value echoed by the other 1616 * endpoint. The RFC says those will be discarded 1617 * since it is a dup that has already been acked. 1618 * This gives us a lowerbound on the timestamp. 1619 * timestamp >= other last echoed timestamp 1620 * - The timestamp will be less than or equal to 1621 * the last timestamp plus the time between the 1622 * last packet and now. The RFC defines the max 1623 * clock rate as 1ms. We will allow clocks to be 1624 * up to 10% fast and will allow a total difference 1625 * or 30 seconds due to a route change. And this 1626 * gives us an upperbound on the timestamp. 1627 * timestamp <= last timestamp + max ticks 1628 * We have to be careful here. Windows will send an 1629 * initial timestamp of zero and then initialize it 1630 * to a random value after the 3whs; presumably to 1631 * avoid a DoS by having to call an expensive RNG 1632 * during a SYN flood. Proof MS has at least one 1633 * good security geek. 1634 * 1635 * - The TCP timestamp option must also echo the other 1636 * endpoints timestamp. The timestamp echoed is the 1637 * one carried on the earliest unacknowledged segment 1638 * on the left edge of the sequence window. The RFC 1639 * states that the host will reject any echoed 1640 * timestamps that were larger than any ever sent. 1641 * This gives us an upperbound on the TS echo. 1642 * tescr <= largest_tsval 1643 * - The lowerbound on the TS echo is a little more 1644 * tricky to determine. The other endpoint's echoed 1645 * values will not decrease. But there may be 1646 * network conditions that re-order packets and 1647 * cause our view of them to decrease. For now the 1648 * only lowerbound we can safely determine is that 1649 * the TS echo will never be less than the original 1650 * TS. XXX There is probably a better lowerbound. 1651 * Remove TS_MAX_CONN with better lowerbound check. 1652 * tescr >= other original TS 1653 * 1654 * It is also important to note that the fastest 1655 * timestamp clock of 1ms will wrap its 32bit space in 1656 * 24 days. So we just disable TS checking after 24 1657 * days of idle time. We actually must use a 12d 1658 * connection limit until we can come up with a better 1659 * lowerbound to the TS echo check. 1660 */ 1661 struct timeval delta_ts; 1662 int ts_fudge; 1663 1664 1665 /* 1666 * PFTM_TS_DIFF is how many seconds of leeway to allow 1667 * a host's timestamp. This can happen if the previous 1668 * packet got delayed in transit for much longer than 1669 * this packet. 1670 */ 1671 if ((ts_fudge = state->rule.ptr->timeout[PFTM_TS_DIFF]) == 0) 1672 ts_fudge = pf_default_rule.timeout[PFTM_TS_DIFF]; 1673 1674 1675 /* Calculate max ticks since the last timestamp */ 1676 #define TS_MAXFREQ 1100 /* RFC max TS freq of 1Khz + 10% skew */ 1677 #define TS_MICROSECS 1000000 /* microseconds per second */ 1678 #ifndef timersub 1679 #define timersub(tvp, uvp, vvp) \ 1680 do { \ 1681 (vvp)->tv_sec = (tvp)->tv_sec - (uvp)->tv_sec; \ 1682 (vvp)->tv_usec = (tvp)->tv_usec - (uvp)->tv_usec; \ 1683 if ((vvp)->tv_usec < 0) { \ 1684 (vvp)->tv_sec--; \ 1685 (vvp)->tv_usec += 1000000; \ 1686 } \ 1687 } while (0) 1688 #endif 1689 1690 timersub(&uptime, &src->scrub->pfss_last, &delta_ts); 1691 tsval_from_last = (delta_ts.tv_sec + ts_fudge) * TS_MAXFREQ; 1692 tsval_from_last += delta_ts.tv_usec / (TS_MICROSECS/TS_MAXFREQ); 1693 1694 1695 if ((src->state >= TCPS_ESTABLISHED && 1696 dst->state >= TCPS_ESTABLISHED) && 1697 (SEQ_LT(tsval, dst->scrub->pfss_tsecr) || 1698 SEQ_GT(tsval, src->scrub->pfss_tsval + tsval_from_last) || 1699 (tsecr && (SEQ_GT(tsecr, dst->scrub->pfss_tsval) || 1700 SEQ_LT(tsecr, dst->scrub->pfss_tsval0))))) { 1701 /* Bad RFC1323 implementation or an insertion attack. 1702 * 1703 * - Solaris 2.6 and 2.7 are known to send another ACK 1704 * after the FIN,FIN|ACK,ACK closing that carries 1705 * an old timestamp. 1706 */ 1707 1708 DPFPRINTF(("Timestamp failed %c%c%c%c\n", 1709 SEQ_LT(tsval, dst->scrub->pfss_tsecr) ? '0' : ' ', 1710 SEQ_GT(tsval, src->scrub->pfss_tsval + 1711 tsval_from_last) ? '1' : ' ', 1712 SEQ_GT(tsecr, dst->scrub->pfss_tsval) ? '2' : ' ', 1713 SEQ_LT(tsecr, dst->scrub->pfss_tsval0)? '3' : ' ')); 1714 DPFPRINTF((" tsval: %u tsecr: %u +ticks: %u " 1715 "idle: %lus %lums\n", 1716 tsval, tsecr, tsval_from_last, delta_ts.tv_sec, 1717 delta_ts.tv_usec / 1000)); 1718 DPFPRINTF((" src->tsval: %u tsecr: %u\n", 1719 src->scrub->pfss_tsval, src->scrub->pfss_tsecr)); 1720 DPFPRINTF((" dst->tsval: %u tsecr: %u tsval0: %u" 1721 "\n", dst->scrub->pfss_tsval, 1722 dst->scrub->pfss_tsecr, dst->scrub->pfss_tsval0)); 1723 if (pf_status.debug >= PF_DEBUG_MISC) { 1724 pf_print_state(state); 1725 pf_print_flags(th->th_flags); 1726 kprintf("\n"); 1727 } 1728 REASON_SET(reason, PFRES_TS); 1729 return (PF_DROP); 1730 } 1731 1732 /* XXX I'd really like to require tsecr but it's optional */ 1733 1734 } else if (!got_ts && (th->th_flags & TH_RST) == 0 && 1735 ((src->state == TCPS_ESTABLISHED && dst->state == TCPS_ESTABLISHED) 1736 || pd->p_len > 0 || (th->th_flags & TH_SYN)) && 1737 src->scrub && dst->scrub && 1738 (src->scrub->pfss_flags & PFSS_PAWS) && 1739 (dst->scrub->pfss_flags & PFSS_PAWS)) { 1740 /* Didn't send a timestamp. Timestamps aren't really useful 1741 * when: 1742 * - connection opening or closing (often not even sent). 1743 * but we must not let an attacker to put a FIN on a 1744 * data packet to sneak it through our ESTABLISHED check. 1745 * - on a TCP reset. RFC suggests not even looking at TS. 1746 * - on an empty ACK. The TS will not be echoed so it will 1747 * probably not help keep the RTT calculation in sync and 1748 * there isn't as much danger when the sequence numbers 1749 * got wrapped. So some stacks don't include TS on empty 1750 * ACKs :-( 1751 * 1752 * To minimize the disruption to mostly RFC1323 conformant 1753 * stacks, we will only require timestamps on data packets. 1754 * 1755 * And what do ya know, we cannot require timestamps on data 1756 * packets. There appear to be devices that do legitimate 1757 * TCP connection hijacking. There are HTTP devices that allow 1758 * a 3whs (with timestamps) and then buffer the HTTP request. 1759 * If the intermediate device has the HTTP response cache, it 1760 * will spoof the response but not bother timestamping its 1761 * packets. So we can look for the presence of a timestamp in 1762 * the first data packet and if there, require it in all future 1763 * packets. 1764 */ 1765 1766 if (pd->p_len > 0 && (src->scrub->pfss_flags & PFSS_DATA_TS)) { 1767 /* 1768 * Hey! Someone tried to sneak a packet in. Or the 1769 * stack changed its RFC1323 behavior?!?! 1770 */ 1771 if (pf_status.debug >= PF_DEBUG_MISC) { 1772 DPFPRINTF(("Did not receive expected RFC1323 " 1773 "timestamp\n")); 1774 pf_print_state(state); 1775 pf_print_flags(th->th_flags); 1776 kprintf("\n"); 1777 } 1778 REASON_SET(reason, PFRES_TS); 1779 return (PF_DROP); 1780 } 1781 } 1782 1783 1784 /* 1785 * We will note if a host sends his data packets with or without 1786 * timestamps. And require all data packets to contain a timestamp 1787 * if the first does. PAWS implicitly requires that all data packets be 1788 * timestamped. But I think there are middle-man devices that hijack 1789 * TCP streams immediately after the 3whs and don't timestamp their 1790 * packets (seen in a WWW accelerator or cache). 1791 */ 1792 if (pd->p_len > 0 && src->scrub && (src->scrub->pfss_flags & 1793 (PFSS_TIMESTAMP|PFSS_DATA_TS|PFSS_DATA_NOTS)) == PFSS_TIMESTAMP) { 1794 if (got_ts) 1795 src->scrub->pfss_flags |= PFSS_DATA_TS; 1796 else { 1797 src->scrub->pfss_flags |= PFSS_DATA_NOTS; 1798 if (pf_status.debug >= PF_DEBUG_MISC && dst->scrub && 1799 (dst->scrub->pfss_flags & PFSS_TIMESTAMP)) { 1800 /* Don't warn if other host rejected RFC1323 */ 1801 DPFPRINTF(("Broken RFC1323 stack did not " 1802 "timestamp data packet. Disabled PAWS " 1803 "security.\n")); 1804 pf_print_state(state); 1805 pf_print_flags(th->th_flags); 1806 kprintf("\n"); 1807 } 1808 } 1809 } 1810 1811 1812 /* 1813 * Update PAWS values 1814 */ 1815 if (got_ts && src->scrub && PFSS_TIMESTAMP == (src->scrub->pfss_flags & 1816 (PFSS_PAWS_IDLED|PFSS_TIMESTAMP))) { 1817 getmicrouptime(&src->scrub->pfss_last); 1818 if (SEQ_GEQ(tsval, src->scrub->pfss_tsval) || 1819 (src->scrub->pfss_flags & PFSS_PAWS) == 0) 1820 src->scrub->pfss_tsval = tsval; 1821 1822 if (tsecr) { 1823 if (SEQ_GEQ(tsecr, src->scrub->pfss_tsecr) || 1824 (src->scrub->pfss_flags & PFSS_PAWS) == 0) 1825 src->scrub->pfss_tsecr = tsecr; 1826 1827 if ((src->scrub->pfss_flags & PFSS_PAWS) == 0 && 1828 (SEQ_LT(tsval, src->scrub->pfss_tsval0) || 1829 src->scrub->pfss_tsval0 == 0)) { 1830 /* tsval0 MUST be the lowest timestamp */ 1831 src->scrub->pfss_tsval0 = tsval; 1832 } 1833 1834 /* Only fully initialized after a TS gets echoed */ 1835 if ((src->scrub->pfss_flags & PFSS_PAWS) == 0) 1836 src->scrub->pfss_flags |= PFSS_PAWS; 1837 } 1838 } 1839 1840 /* I have a dream.... TCP segment reassembly.... */ 1841 return (0); 1842 } 1843 1844 int 1845 pf_normalize_tcpopt(struct pf_rule *r, struct mbuf *m, struct tcphdr *th, 1846 int off, sa_family_t af) 1847 { 1848 u_int16_t *mss; 1849 int thoff; 1850 int opt, cnt, optlen = 0; 1851 int rewrite = 0; 1852 u_char opts[TCP_MAXOLEN]; 1853 u_char *optp = opts; 1854 1855 thoff = th->th_off << 2; 1856 cnt = thoff - sizeof(struct tcphdr); 1857 1858 if (cnt > 0 && !pf_pull_hdr(m, off + sizeof(*th), opts, cnt, 1859 NULL, NULL, af)) 1860 return (rewrite); 1861 1862 for (; cnt > 0; cnt -= optlen, optp += optlen) { 1863 opt = optp[0]; 1864 if (opt == TCPOPT_EOL) 1865 break; 1866 if (opt == TCPOPT_NOP) 1867 optlen = 1; 1868 else { 1869 if (cnt < 2) 1870 break; 1871 optlen = optp[1]; 1872 if (optlen < 2 || optlen > cnt) 1873 break; 1874 } 1875 switch (opt) { 1876 case TCPOPT_MAXSEG: 1877 mss = (u_int16_t *)(optp + 2); 1878 if ((ntohs(*mss)) > r->max_mss) { 1879 th->th_sum = pf_cksum_fixup(th->th_sum, 1880 *mss, htons(r->max_mss), 0); 1881 *mss = htons(r->max_mss); 1882 rewrite = 1; 1883 } 1884 break; 1885 default: 1886 break; 1887 } 1888 } 1889 1890 if (rewrite) 1891 m_copyback(m, off + sizeof(*th), thoff - sizeof(*th), opts); 1892 1893 return (rewrite); 1894 } 1895