1 /* $OpenBSD: pf_norm.c,v 1.219 2020/06/24 22:03:43 cheloha Exp $ */ 2 3 /* 4 * Copyright 2001 Niels Provos <provos@citi.umich.edu> 5 * Copyright 2009 Henning Brauer <henning@openbsd.org> 6 * Copyright 2011-2018 Alexander Bluhm <bluhm@openbsd.org> 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 "pflog.h" 31 32 #include <sys/param.h> 33 #include <sys/systm.h> 34 #include <sys/mbuf.h> 35 #include <sys/filio.h> 36 #include <sys/fcntl.h> 37 #include <sys/socket.h> 38 #include <sys/kernel.h> 39 #include <sys/time.h> 40 #include <sys/pool.h> 41 #include <sys/syslog.h> 42 #include <sys/mutex.h> 43 44 #include <net/if.h> 45 #include <net/if_var.h> 46 #include <net/if_pflog.h> 47 48 #include <netinet/in.h> 49 #include <netinet/ip.h> 50 #include <netinet/ip_var.h> 51 #include <netinet/ip_icmp.h> 52 #include <netinet/tcp.h> 53 #include <netinet/tcp_seq.h> 54 #include <netinet/tcp_fsm.h> 55 #include <netinet/udp.h> 56 57 #ifdef INET6 58 #include <netinet6/in6_var.h> 59 #include <netinet/ip6.h> 60 #include <netinet6/ip6_var.h> 61 #include <netinet/icmp6.h> 62 #include <netinet6/nd6.h> 63 #endif /* INET6 */ 64 65 #include <net/pfvar.h> 66 #include <net/pfvar_priv.h> 67 68 struct pf_frent { 69 TAILQ_ENTRY(pf_frent) fr_next; 70 struct mbuf *fe_m; 71 u_int16_t fe_hdrlen; /* ipv4 header length with ip options 72 ipv6, extension, fragment header */ 73 u_int16_t fe_extoff; /* last extension header offset or 0 */ 74 u_int16_t fe_len; /* fragment length */ 75 u_int16_t fe_off; /* fragment offset */ 76 u_int16_t fe_mff; /* more fragment flag */ 77 }; 78 79 RB_HEAD(pf_frag_tree, pf_fragment); 80 struct pf_frnode { 81 struct pf_addr fn_src; /* ip source address */ 82 struct pf_addr fn_dst; /* ip destination address */ 83 sa_family_t fn_af; /* address family */ 84 u_int8_t fn_proto; /* protocol for fragments in fn_tree */ 85 u_int8_t fn_direction; /* pf packet direction */ 86 u_int32_t fn_fragments; /* number of entries in fn_tree */ 87 u_int32_t fn_gen; /* fr_gen of newest entry in fn_tree */ 88 89 RB_ENTRY(pf_frnode) fn_entry; 90 struct pf_frag_tree fn_tree; /* matching fragments, lookup by id */ 91 }; 92 93 struct pf_fragment { 94 struct pf_frent *fr_firstoff[PF_FRAG_ENTRY_POINTS]; 95 /* pointers to queue element */ 96 u_int8_t fr_entries[PF_FRAG_ENTRY_POINTS]; 97 /* count entries between pointers */ 98 RB_ENTRY(pf_fragment) fr_entry; 99 TAILQ_ENTRY(pf_fragment) frag_next; 100 TAILQ_HEAD(pf_fragq, pf_frent) fr_queue; 101 u_int32_t fr_id; /* fragment id for reassemble */ 102 int32_t fr_timeout; 103 u_int32_t fr_gen; /* generation number (per pf_frnode) */ 104 u_int16_t fr_maxlen; /* maximum length of single fragment */ 105 u_int16_t fr_holes; /* number of holes in the queue */ 106 struct pf_frnode *fr_node; /* ip src/dst/proto/af for fragments */ 107 }; 108 109 struct pf_fragment_tag { 110 u_int16_t ft_hdrlen; /* header length of reassembled pkt */ 111 u_int16_t ft_extoff; /* last extension header offset or 0 */ 112 u_int16_t ft_maxlen; /* maximum fragment payload length */ 113 }; 114 115 TAILQ_HEAD(pf_fragqueue, pf_fragment) pf_fragqueue; 116 117 static __inline int pf_frnode_compare(struct pf_frnode *, 118 struct pf_frnode *); 119 RB_HEAD(pf_frnode_tree, pf_frnode) pf_frnode_tree; 120 RB_PROTOTYPE(pf_frnode_tree, pf_frnode, fn_entry, pf_frnode_compare); 121 RB_GENERATE(pf_frnode_tree, pf_frnode, fn_entry, pf_frnode_compare); 122 123 static __inline int pf_frag_compare(struct pf_fragment *, 124 struct pf_fragment *); 125 RB_PROTOTYPE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare); 126 RB_GENERATE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare); 127 128 /* Private prototypes */ 129 void pf_flush_fragments(void); 130 void pf_free_fragment(struct pf_fragment *); 131 struct pf_fragment *pf_find_fragment(struct pf_frnode *, u_int32_t); 132 struct pf_frent *pf_create_fragment(u_short *); 133 int pf_frent_holes(struct pf_frent *); 134 static inline int pf_frent_index(struct pf_frent *); 135 int pf_frent_insert(struct pf_fragment *, 136 struct pf_frent *, struct pf_frent *); 137 void pf_frent_remove(struct pf_fragment *, 138 struct pf_frent *); 139 struct pf_frent *pf_frent_previous(struct pf_fragment *, 140 struct pf_frent *); 141 struct pf_fragment *pf_fillup_fragment(struct pf_frnode *, u_int32_t, 142 struct pf_frent *, u_short *); 143 struct mbuf *pf_join_fragment(struct pf_fragment *); 144 int pf_reassemble(struct mbuf **, int, u_short *); 145 #ifdef INET6 146 int pf_reassemble6(struct mbuf **, struct ip6_frag *, 147 u_int16_t, u_int16_t, int, u_short *); 148 #endif /* INET6 */ 149 150 /* Globals */ 151 struct pool pf_frent_pl, pf_frag_pl, pf_frnode_pl; 152 struct pool pf_state_scrub_pl; 153 int pf_nfrents; 154 155 #ifdef WITH_PF_LOCK 156 struct mutex pf_frag_mtx; 157 158 #define PF_FRAG_LOCK_INIT() mtx_init(&pf_frag_mtx, IPL_SOFTNET) 159 #define PF_FRAG_LOCK() mtx_enter(&pf_frag_mtx) 160 #define PF_FRAG_UNLOCK() mtx_leave(&pf_frag_mtx) 161 #else /* !WITH_PF_LOCK */ 162 #define PF_FRAG_LOCK_INIT() (void)(0) 163 #define PF_FRAG_LOCK() (void)(0) 164 #define PF_FRAG_UNLOCK() (void)(0) 165 #endif /* WITH_PF_LOCK */ 166 167 void 168 pf_normalize_init(void) 169 { 170 pool_init(&pf_frent_pl, sizeof(struct pf_frent), 0, 171 IPL_SOFTNET, 0, "pffrent", NULL); 172 pool_init(&pf_frnode_pl, sizeof(struct pf_frnode), 0, 173 IPL_SOFTNET, 0, "pffrnode", NULL); 174 pool_init(&pf_frag_pl, sizeof(struct pf_fragment), 0, 175 IPL_SOFTNET, 0, "pffrag", NULL); 176 pool_init(&pf_state_scrub_pl, sizeof(struct pf_state_scrub), 0, 177 IPL_SOFTNET, 0, "pfstscr", NULL); 178 179 pool_sethiwat(&pf_frag_pl, PFFRAG_FRAG_HIWAT); 180 pool_sethardlimit(&pf_frent_pl, PFFRAG_FRENT_HIWAT, NULL, 0); 181 182 TAILQ_INIT(&pf_fragqueue); 183 184 PF_FRAG_LOCK_INIT(); 185 } 186 187 static __inline int 188 pf_frnode_compare(struct pf_frnode *a, struct pf_frnode *b) 189 { 190 int diff; 191 192 if ((diff = a->fn_proto - b->fn_proto) != 0) 193 return (diff); 194 if ((diff = a->fn_af - b->fn_af) != 0) 195 return (diff); 196 if ((diff = pf_addr_compare(&a->fn_src, &b->fn_src, a->fn_af)) != 0) 197 return (diff); 198 if ((diff = pf_addr_compare(&a->fn_dst, &b->fn_dst, a->fn_af)) != 0) 199 return (diff); 200 201 return (0); 202 } 203 204 static __inline int 205 pf_frag_compare(struct pf_fragment *a, struct pf_fragment *b) 206 { 207 int diff; 208 209 if ((diff = a->fr_id - b->fr_id) != 0) 210 return (diff); 211 212 return (0); 213 } 214 215 void 216 pf_purge_expired_fragments(void) 217 { 218 struct pf_fragment *frag; 219 int32_t expire; 220 221 PF_ASSERT_UNLOCKED(); 222 223 expire = getuptime() - pf_default_rule.timeout[PFTM_FRAG]; 224 225 PF_FRAG_LOCK(); 226 while ((frag = TAILQ_LAST(&pf_fragqueue, pf_fragqueue)) != NULL) { 227 if (frag->fr_timeout > expire) 228 break; 229 DPFPRINTF(LOG_NOTICE, "expiring %d(%p)", frag->fr_id, frag); 230 pf_free_fragment(frag); 231 } 232 PF_FRAG_UNLOCK(); 233 } 234 235 /* 236 * Try to flush old fragments to make space for new ones 237 */ 238 void 239 pf_flush_fragments(void) 240 { 241 struct pf_fragment *frag; 242 int goal; 243 244 goal = pf_nfrents * 9 / 10; 245 DPFPRINTF(LOG_NOTICE, "trying to free > %d frents", pf_nfrents - goal); 246 while (goal < pf_nfrents) { 247 if ((frag = TAILQ_LAST(&pf_fragqueue, pf_fragqueue)) == NULL) 248 break; 249 pf_free_fragment(frag); 250 } 251 } 252 253 /* 254 * Remove a fragment from the fragment queue, free its fragment entries, 255 * and free the fragment itself. 256 */ 257 void 258 pf_free_fragment(struct pf_fragment *frag) 259 { 260 struct pf_frent *frent; 261 struct pf_frnode *frnode; 262 263 frnode = frag->fr_node; 264 RB_REMOVE(pf_frag_tree, &frnode->fn_tree, frag); 265 KASSERT(frnode->fn_fragments >= 1); 266 frnode->fn_fragments--; 267 if (frnode->fn_fragments == 0) { 268 KASSERT(RB_EMPTY(&frnode->fn_tree)); 269 RB_REMOVE(pf_frnode_tree, &pf_frnode_tree, frnode); 270 pool_put(&pf_frnode_pl, frnode); 271 } 272 TAILQ_REMOVE(&pf_fragqueue, frag, frag_next); 273 274 /* Free all fragment entries */ 275 while ((frent = TAILQ_FIRST(&frag->fr_queue)) != NULL) { 276 TAILQ_REMOVE(&frag->fr_queue, frent, fr_next); 277 m_freem(frent->fe_m); 278 pool_put(&pf_frent_pl, frent); 279 pf_nfrents--; 280 } 281 pool_put(&pf_frag_pl, frag); 282 } 283 284 struct pf_fragment * 285 pf_find_fragment(struct pf_frnode *key, u_int32_t id) 286 { 287 struct pf_fragment *frag, idkey; 288 struct pf_frnode *frnode; 289 u_int32_t stale; 290 291 frnode = RB_FIND(pf_frnode_tree, &pf_frnode_tree, key); 292 if (frnode == NULL) 293 return (NULL); 294 KASSERT(frnode->fn_fragments >= 1); 295 idkey.fr_id = id; 296 frag = RB_FIND(pf_frag_tree, &frnode->fn_tree, &idkey); 297 if (frag == NULL) 298 return (NULL); 299 /* 300 * Limit the number of fragments we accept for each (proto,src,dst,af) 301 * combination (aka pf_frnode), so we can deal better with a high rate 302 * of fragments. Problem analysis is in RFC 4963. 303 * Store the current generation for each pf_frnode in fn_gen and on 304 * lookup discard 'stale' fragments (pf_fragment, based on the fr_gen 305 * member). Instead of adding another button interpret the pf fragment 306 * timeout in multiples of 200 fragments. This way the default of 60s 307 * means: pf_fragment objects older than 60*200 = 12,000 generations 308 * are considered stale. 309 */ 310 stale = pf_default_rule.timeout[PFTM_FRAG] * PF_FRAG_STALE; 311 if ((frnode->fn_gen - frag->fr_gen) >= stale) { 312 DPFPRINTF(LOG_NOTICE, "stale fragment %d(%p), gen %u, num %u", 313 frag->fr_id, frag, frag->fr_gen, frnode->fn_fragments); 314 pf_free_fragment(frag); 315 return (NULL); 316 } 317 TAILQ_REMOVE(&pf_fragqueue, frag, frag_next); 318 TAILQ_INSERT_HEAD(&pf_fragqueue, frag, frag_next); 319 320 return (frag); 321 } 322 323 struct pf_frent * 324 pf_create_fragment(u_short *reason) 325 { 326 struct pf_frent *frent; 327 328 frent = pool_get(&pf_frent_pl, PR_NOWAIT); 329 if (frent == NULL) { 330 pf_flush_fragments(); 331 frent = pool_get(&pf_frent_pl, PR_NOWAIT); 332 if (frent == NULL) { 333 REASON_SET(reason, PFRES_MEMORY); 334 return (NULL); 335 } 336 } 337 pf_nfrents++; 338 339 return (frent); 340 } 341 342 /* 343 * Calculate the additional holes that were created in the fragment 344 * queue by inserting this fragment. A fragment in the middle 345 * creates one more hole by splitting. For each connected side, 346 * it loses one hole. 347 * Fragment entry must be in the queue when calling this function. 348 */ 349 int 350 pf_frent_holes(struct pf_frent *frent) 351 { 352 struct pf_frent *prev = TAILQ_PREV(frent, pf_fragq, fr_next); 353 struct pf_frent *next = TAILQ_NEXT(frent, fr_next); 354 int holes = 1; 355 356 if (prev == NULL) { 357 if (frent->fe_off == 0) 358 holes--; 359 } else { 360 KASSERT(frent->fe_off != 0); 361 if (frent->fe_off == prev->fe_off + prev->fe_len) 362 holes--; 363 } 364 if (next == NULL) { 365 if (!frent->fe_mff) 366 holes--; 367 } else { 368 KASSERT(frent->fe_mff); 369 if (next->fe_off == frent->fe_off + frent->fe_len) 370 holes--; 371 } 372 return holes; 373 } 374 375 static inline int 376 pf_frent_index(struct pf_frent *frent) 377 { 378 /* 379 * We have an array of 16 entry points to the queue. A full size 380 * 65535 octet IP packet can have 8192 fragments. So the queue 381 * traversal length is at most 512 and at most 16 entry points are 382 * checked. We need 128 additional bytes on a 64 bit architecture. 383 */ 384 CTASSERT(((u_int16_t)0xffff &~ 7) / (0x10000 / PF_FRAG_ENTRY_POINTS) == 385 16 - 1); 386 CTASSERT(((u_int16_t)0xffff >> 3) / PF_FRAG_ENTRY_POINTS == 512 - 1); 387 388 return frent->fe_off / (0x10000 / PF_FRAG_ENTRY_POINTS); 389 } 390 391 int 392 pf_frent_insert(struct pf_fragment *frag, struct pf_frent *frent, 393 struct pf_frent *prev) 394 { 395 CTASSERT(PF_FRAG_ENTRY_LIMIT <= 0xff); 396 int index; 397 398 /* 399 * A packet has at most 65536 octets. With 16 entry points, each one 400 * spawns 4096 octets. We limit these to 64 fragments each, which 401 * means on average every fragment must have at least 64 octets. 402 */ 403 index = pf_frent_index(frent); 404 if (frag->fr_entries[index] >= PF_FRAG_ENTRY_LIMIT) 405 return ENOBUFS; 406 frag->fr_entries[index]++; 407 408 if (prev == NULL) { 409 TAILQ_INSERT_HEAD(&frag->fr_queue, frent, fr_next); 410 } else { 411 KASSERT(prev->fe_off + prev->fe_len <= frent->fe_off); 412 TAILQ_INSERT_AFTER(&frag->fr_queue, prev, frent, fr_next); 413 } 414 415 if (frag->fr_firstoff[index] == NULL) { 416 KASSERT(prev == NULL || pf_frent_index(prev) < index); 417 frag->fr_firstoff[index] = frent; 418 } else { 419 if (frent->fe_off < frag->fr_firstoff[index]->fe_off) { 420 KASSERT(prev == NULL || pf_frent_index(prev) < index); 421 frag->fr_firstoff[index] = frent; 422 } else { 423 KASSERT(prev != NULL); 424 KASSERT(pf_frent_index(prev) == index); 425 } 426 } 427 428 frag->fr_holes += pf_frent_holes(frent); 429 430 return 0; 431 } 432 433 void 434 pf_frent_remove(struct pf_fragment *frag, struct pf_frent *frent) 435 { 436 #ifdef DIAGNOSTIC 437 struct pf_frent *prev = TAILQ_PREV(frent, pf_fragq, fr_next); 438 #endif 439 struct pf_frent *next = TAILQ_NEXT(frent, fr_next); 440 int index; 441 442 frag->fr_holes -= pf_frent_holes(frent); 443 444 index = pf_frent_index(frent); 445 KASSERT(frag->fr_firstoff[index] != NULL); 446 if (frag->fr_firstoff[index]->fe_off == frent->fe_off) { 447 if (next == NULL) { 448 frag->fr_firstoff[index] = NULL; 449 } else { 450 KASSERT(frent->fe_off + frent->fe_len <= next->fe_off); 451 if (pf_frent_index(next) == index) { 452 frag->fr_firstoff[index] = next; 453 } else { 454 frag->fr_firstoff[index] = NULL; 455 } 456 } 457 } else { 458 KASSERT(frag->fr_firstoff[index]->fe_off < frent->fe_off); 459 KASSERT(prev != NULL); 460 KASSERT(prev->fe_off + prev->fe_len <= frent->fe_off); 461 KASSERT(pf_frent_index(prev) == index); 462 } 463 464 TAILQ_REMOVE(&frag->fr_queue, frent, fr_next); 465 466 KASSERT(frag->fr_entries[index] > 0); 467 frag->fr_entries[index]--; 468 } 469 470 struct pf_frent * 471 pf_frent_previous(struct pf_fragment *frag, struct pf_frent *frent) 472 { 473 struct pf_frent *prev, *next; 474 int index; 475 476 /* 477 * If there are no fragments after frag, take the final one. Assume 478 * that the global queue is not empty. 479 */ 480 prev = TAILQ_LAST(&frag->fr_queue, pf_fragq); 481 KASSERT(prev != NULL); 482 if (prev->fe_off <= frent->fe_off) 483 return prev; 484 /* 485 * We want to find a fragment entry that is before frag, but still 486 * close to it. Find the first fragment entry that is in the same 487 * entry point or in the first entry point after that. As we have 488 * already checked that there are entries behind frag, this will 489 * succeed. 490 */ 491 for (index = pf_frent_index(frent); index < PF_FRAG_ENTRY_POINTS; 492 index++) { 493 prev = frag->fr_firstoff[index]; 494 if (prev != NULL) 495 break; 496 } 497 KASSERT(prev != NULL); 498 /* 499 * In prev we may have a fragment from the same entry point that is 500 * before frent, or one that is just one position behind frent. 501 * In the latter case, we go back one step and have the predecessor. 502 * There may be none if the new fragment will be the first one. 503 */ 504 if (prev->fe_off > frent->fe_off) { 505 prev = TAILQ_PREV(prev, pf_fragq, fr_next); 506 if (prev == NULL) 507 return NULL; 508 KASSERT(prev->fe_off <= frent->fe_off); 509 return prev; 510 } 511 /* 512 * In prev is the first fragment of the entry point. The offset 513 * of frag is behind it. Find the closest previous fragment. 514 */ 515 for (next = TAILQ_NEXT(prev, fr_next); next != NULL; 516 next = TAILQ_NEXT(next, fr_next)) { 517 if (next->fe_off > frent->fe_off) 518 break; 519 prev = next; 520 } 521 return prev; 522 } 523 524 struct pf_fragment * 525 pf_fillup_fragment(struct pf_frnode *key, u_int32_t id, 526 struct pf_frent *frent, u_short *reason) 527 { 528 struct pf_frent *after, *next, *prev; 529 struct pf_fragment *frag; 530 struct pf_frnode *frnode; 531 u_int16_t total; 532 533 /* No empty fragments */ 534 if (frent->fe_len == 0) { 535 DPFPRINTF(LOG_NOTICE, "bad fragment: len 0"); 536 goto bad_fragment; 537 } 538 539 /* All fragments are 8 byte aligned */ 540 if (frent->fe_mff && (frent->fe_len & 0x7)) { 541 DPFPRINTF(LOG_NOTICE, "bad fragment: mff and len %d", 542 frent->fe_len); 543 goto bad_fragment; 544 } 545 546 /* Respect maximum length, IP_MAXPACKET == IPV6_MAXPACKET */ 547 if (frent->fe_off + frent->fe_len > IP_MAXPACKET) { 548 DPFPRINTF(LOG_NOTICE, "bad fragment: max packet %d", 549 frent->fe_off + frent->fe_len); 550 goto bad_fragment; 551 } 552 553 DPFPRINTF(LOG_INFO, key->fn_af == AF_INET ? 554 "reass frag %d @ %d-%d" : "reass frag %#08x @ %d-%d", 555 id, frent->fe_off, frent->fe_off + frent->fe_len); 556 557 /* Fully buffer all of the fragments in this fragment queue */ 558 frag = pf_find_fragment(key, id); 559 560 /* Create a new reassembly queue for this packet */ 561 if (frag == NULL) { 562 frag = pool_get(&pf_frag_pl, PR_NOWAIT); 563 if (frag == NULL) { 564 pf_flush_fragments(); 565 frag = pool_get(&pf_frag_pl, PR_NOWAIT); 566 if (frag == NULL) { 567 REASON_SET(reason, PFRES_MEMORY); 568 goto drop_fragment; 569 } 570 } 571 frnode = RB_FIND(pf_frnode_tree, &pf_frnode_tree, key); 572 if (frnode == NULL) { 573 frnode = pool_get(&pf_frnode_pl, PR_NOWAIT); 574 if (frnode == NULL) { 575 pf_flush_fragments(); 576 frnode = pool_get(&pf_frnode_pl, PR_NOWAIT); 577 if (frnode == NULL) { 578 REASON_SET(reason, PFRES_MEMORY); 579 pool_put(&pf_frag_pl, frag); 580 goto drop_fragment; 581 } 582 } 583 *frnode = *key; 584 RB_INIT(&frnode->fn_tree); 585 frnode->fn_fragments = 0; 586 frnode->fn_gen = 0; 587 } 588 memset(frag->fr_firstoff, 0, sizeof(frag->fr_firstoff)); 589 memset(frag->fr_entries, 0, sizeof(frag->fr_entries)); 590 TAILQ_INIT(&frag->fr_queue); 591 frag->fr_id = id; 592 frag->fr_timeout = getuptime(); 593 frag->fr_gen = frnode->fn_gen++; 594 frag->fr_maxlen = frent->fe_len; 595 frag->fr_holes = 1; 596 frag->fr_node = frnode; 597 /* RB_INSERT cannot fail as pf_find_fragment() found nothing */ 598 RB_INSERT(pf_frag_tree, &frnode->fn_tree, frag); 599 frnode->fn_fragments++; 600 if (frnode->fn_fragments == 1) 601 RB_INSERT(pf_frnode_tree, &pf_frnode_tree, frnode); 602 TAILQ_INSERT_HEAD(&pf_fragqueue, frag, frag_next); 603 604 /* We do not have a previous fragment, cannot fail. */ 605 pf_frent_insert(frag, frent, NULL); 606 607 return (frag); 608 } 609 610 KASSERT(!TAILQ_EMPTY(&frag->fr_queue)); 611 KASSERT(frag->fr_node); 612 613 /* Remember maximum fragment len for refragmentation */ 614 if (frent->fe_len > frag->fr_maxlen) 615 frag->fr_maxlen = frent->fe_len; 616 617 /* Maximum data we have seen already */ 618 total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off + 619 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len; 620 621 /* Non terminal fragments must have more fragments flag */ 622 if (frent->fe_off + frent->fe_len < total && !frent->fe_mff) 623 goto free_ipv6_fragment; 624 625 /* Check if we saw the last fragment already */ 626 if (!TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_mff) { 627 if (frent->fe_off + frent->fe_len > total || 628 (frent->fe_off + frent->fe_len == total && frent->fe_mff)) 629 goto free_ipv6_fragment; 630 } else { 631 if (frent->fe_off + frent->fe_len == total && !frent->fe_mff) 632 goto free_ipv6_fragment; 633 } 634 635 /* Find neighbors for newly inserted fragment */ 636 prev = pf_frent_previous(frag, frent); 637 if (prev == NULL) { 638 after = TAILQ_FIRST(&frag->fr_queue); 639 KASSERT(after != NULL); 640 } else { 641 after = TAILQ_NEXT(prev, fr_next); 642 } 643 644 if (prev != NULL && prev->fe_off + prev->fe_len > frent->fe_off) { 645 u_int16_t precut; 646 647 #ifdef INET6 648 if (frag->fr_node->fn_af == AF_INET6) 649 goto free_ipv6_fragment; 650 #endif /* INET6 */ 651 652 precut = prev->fe_off + prev->fe_len - frent->fe_off; 653 if (precut >= frent->fe_len) { 654 DPFPRINTF(LOG_NOTICE, "new frag overlapped"); 655 goto drop_fragment; 656 } 657 DPFPRINTF(LOG_NOTICE, "frag head overlap %d", precut); 658 m_adj(frent->fe_m, precut); 659 frent->fe_off += precut; 660 frent->fe_len -= precut; 661 } 662 663 for (; after != NULL && frent->fe_off + frent->fe_len > after->fe_off; 664 after = next) { 665 u_int16_t aftercut; 666 667 #ifdef INET6 668 if (frag->fr_node->fn_af == AF_INET6) 669 goto free_ipv6_fragment; 670 #endif /* INET6 */ 671 672 aftercut = frent->fe_off + frent->fe_len - after->fe_off; 673 if (aftercut < after->fe_len) { 674 DPFPRINTF(LOG_NOTICE, "frag tail overlap %d", aftercut); 675 m_adj(after->fe_m, aftercut); 676 after->fe_off += aftercut; 677 after->fe_len -= aftercut; 678 break; 679 } 680 681 /* This fragment is completely overlapped, lose it */ 682 DPFPRINTF(LOG_NOTICE, "old frag overlapped"); 683 next = TAILQ_NEXT(after, fr_next); 684 pf_frent_remove(frag, after); 685 m_freem(after->fe_m); 686 pool_put(&pf_frent_pl, after); 687 pf_nfrents--; 688 } 689 690 /* If part of the queue gets too long, there is not way to recover. */ 691 if (pf_frent_insert(frag, frent, prev)) { 692 DPFPRINTF(LOG_WARNING, "fragment queue limit exceeded"); 693 goto free_fragment; 694 } 695 696 return (frag); 697 698 free_ipv6_fragment: 699 if (frag->fr_node->fn_af == AF_INET) 700 goto bad_fragment; 701 /* 702 * RFC 5722, Errata 3089: When reassembling an IPv6 datagram, if one 703 * or more its constituent fragments is determined to be an overlapping 704 * fragment, the entire datagram (and any constituent fragments) MUST 705 * be silently discarded. 706 */ 707 DPFPRINTF(LOG_NOTICE, "flush overlapping fragments"); 708 free_fragment: 709 pf_free_fragment(frag); 710 bad_fragment: 711 REASON_SET(reason, PFRES_FRAG); 712 drop_fragment: 713 pool_put(&pf_frent_pl, frent); 714 pf_nfrents--; 715 return (NULL); 716 } 717 718 struct mbuf * 719 pf_join_fragment(struct pf_fragment *frag) 720 { 721 struct mbuf *m, *m2; 722 struct pf_frent *frent; 723 724 frent = TAILQ_FIRST(&frag->fr_queue); 725 TAILQ_REMOVE(&frag->fr_queue, frent, fr_next); 726 727 m = frent->fe_m; 728 /* Strip off any trailing bytes */ 729 if ((frent->fe_hdrlen + frent->fe_len) < m->m_pkthdr.len) 730 m_adj(m, (frent->fe_hdrlen + frent->fe_len) - m->m_pkthdr.len); 731 /* Magic from ip_input */ 732 m2 = m->m_next; 733 m->m_next = NULL; 734 m_cat(m, m2); 735 pool_put(&pf_frent_pl, frent); 736 pf_nfrents--; 737 738 while ((frent = TAILQ_FIRST(&frag->fr_queue)) != NULL) { 739 TAILQ_REMOVE(&frag->fr_queue, frent, fr_next); 740 m2 = frent->fe_m; 741 /* Strip off ip header */ 742 m_adj(m2, frent->fe_hdrlen); 743 /* Strip off any trailing bytes */ 744 if (frent->fe_len < m2->m_pkthdr.len) 745 m_adj(m2, frent->fe_len - m2->m_pkthdr.len); 746 pool_put(&pf_frent_pl, frent); 747 pf_nfrents--; 748 m_removehdr(m2); 749 m_cat(m, m2); 750 } 751 752 /* Remove from fragment queue */ 753 pf_free_fragment(frag); 754 755 return (m); 756 } 757 758 int 759 pf_reassemble(struct mbuf **m0, int dir, u_short *reason) 760 { 761 struct mbuf *m = *m0; 762 struct ip *ip = mtod(m, struct ip *); 763 struct pf_frent *frent; 764 struct pf_fragment *frag; 765 struct pf_frnode key; 766 u_int16_t total, hdrlen; 767 768 /* Get an entry for the fragment queue */ 769 if ((frent = pf_create_fragment(reason)) == NULL) 770 return (PF_DROP); 771 772 frent->fe_m = m; 773 frent->fe_hdrlen = ip->ip_hl << 2; 774 frent->fe_extoff = 0; 775 frent->fe_len = ntohs(ip->ip_len) - (ip->ip_hl << 2); 776 frent->fe_off = (ntohs(ip->ip_off) & IP_OFFMASK) << 3; 777 frent->fe_mff = ntohs(ip->ip_off) & IP_MF; 778 779 key.fn_src.v4 = ip->ip_src; 780 key.fn_dst.v4 = ip->ip_dst; 781 key.fn_af = AF_INET; 782 key.fn_proto = ip->ip_p; 783 key.fn_direction = dir; 784 785 PF_FRAG_LOCK(); 786 if ((frag = pf_fillup_fragment(&key, ip->ip_id, frent, reason)) 787 == NULL) { 788 PF_FRAG_UNLOCK(); 789 return (PF_DROP); 790 } 791 792 /* The mbuf is part of the fragment entry, no direct free or access */ 793 m = *m0 = NULL; 794 795 if (frag->fr_holes) { 796 DPFPRINTF(LOG_DEBUG, "frag %d, holes %d", 797 frag->fr_id, frag->fr_holes); 798 PF_FRAG_UNLOCK(); 799 return (PF_PASS); /* drop because *m0 is NULL, no error */ 800 } 801 802 /* We have all the data */ 803 frent = TAILQ_FIRST(&frag->fr_queue); 804 KASSERT(frent != NULL); 805 total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off + 806 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len; 807 hdrlen = frent->fe_hdrlen; 808 m = *m0 = pf_join_fragment(frag); 809 frag = NULL; 810 m_calchdrlen(m); 811 812 ip = mtod(m, struct ip *); 813 ip->ip_len = htons(hdrlen + total); 814 ip->ip_off &= ~(IP_MF|IP_OFFMASK); 815 816 if (hdrlen + total > IP_MAXPACKET) { 817 PF_FRAG_UNLOCK(); 818 DPFPRINTF(LOG_NOTICE, "drop: too big: %d", total); 819 ip->ip_len = 0; 820 REASON_SET(reason, PFRES_SHORT); 821 /* PF_DROP requires a valid mbuf *m0 in pf_test() */ 822 return (PF_DROP); 823 } 824 825 PF_FRAG_UNLOCK(); 826 DPFPRINTF(LOG_INFO, "complete: %p(%d)", m, ntohs(ip->ip_len)); 827 return (PF_PASS); 828 } 829 830 #ifdef INET6 831 int 832 pf_reassemble6(struct mbuf **m0, struct ip6_frag *fraghdr, 833 u_int16_t hdrlen, u_int16_t extoff, int dir, u_short *reason) 834 { 835 struct mbuf *m = *m0; 836 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 837 struct m_tag *mtag; 838 struct pf_fragment_tag *ftag; 839 struct pf_frent *frent; 840 struct pf_fragment *frag; 841 struct pf_frnode key; 842 int off; 843 u_int16_t total, maxlen; 844 u_int8_t proto; 845 846 /* Get an entry for the fragment queue */ 847 if ((frent = pf_create_fragment(reason)) == NULL) 848 return (PF_DROP); 849 850 frent->fe_m = m; 851 frent->fe_hdrlen = hdrlen; 852 frent->fe_extoff = extoff; 853 frent->fe_len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - hdrlen; 854 frent->fe_off = ntohs(fraghdr->ip6f_offlg & IP6F_OFF_MASK); 855 frent->fe_mff = fraghdr->ip6f_offlg & IP6F_MORE_FRAG; 856 857 key.fn_src.v6 = ip6->ip6_src; 858 key.fn_dst.v6 = ip6->ip6_dst; 859 key.fn_af = AF_INET6; 860 /* Only the first fragment's protocol is relevant */ 861 key.fn_proto = 0; 862 key.fn_direction = dir; 863 864 PF_FRAG_LOCK(); 865 if ((frag = pf_fillup_fragment(&key, fraghdr->ip6f_ident, frent, 866 reason)) == NULL) { 867 PF_FRAG_UNLOCK(); 868 return (PF_DROP); 869 } 870 871 /* The mbuf is part of the fragment entry, no direct free or access */ 872 m = *m0 = NULL; 873 874 if (frag->fr_holes) { 875 DPFPRINTF(LOG_DEBUG, "frag %#08x, holes %d", 876 frag->fr_id, frag->fr_holes); 877 PF_FRAG_UNLOCK(); 878 return (PF_PASS); /* drop because *m0 is NULL, no error */ 879 } 880 881 /* We have all the data */ 882 frent = TAILQ_FIRST(&frag->fr_queue); 883 KASSERT(frent != NULL); 884 extoff = frent->fe_extoff; 885 maxlen = frag->fr_maxlen; 886 total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off + 887 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len; 888 hdrlen = frent->fe_hdrlen - sizeof(struct ip6_frag); 889 m = *m0 = pf_join_fragment(frag); 890 frag = NULL; 891 892 /* Take protocol from first fragment header */ 893 if ((m = m_getptr(m, hdrlen + offsetof(struct ip6_frag, ip6f_nxt), 894 &off)) == NULL) 895 panic("%s: short frag mbuf chain", __func__); 896 proto = *(mtod(m, caddr_t) + off); 897 m = *m0; 898 899 /* Delete frag6 header */ 900 if (frag6_deletefraghdr(m, hdrlen) != 0) 901 goto fail; 902 903 m_calchdrlen(m); 904 905 if ((mtag = m_tag_get(PACKET_TAG_PF_REASSEMBLED, sizeof(struct 906 pf_fragment_tag), M_NOWAIT)) == NULL) 907 goto fail; 908 ftag = (struct pf_fragment_tag *)(mtag + 1); 909 ftag->ft_hdrlen = hdrlen; 910 ftag->ft_extoff = extoff; 911 ftag->ft_maxlen = maxlen; 912 m_tag_prepend(m, mtag); 913 914 ip6 = mtod(m, struct ip6_hdr *); 915 ip6->ip6_plen = htons(hdrlen - sizeof(struct ip6_hdr) + total); 916 if (extoff) { 917 /* Write protocol into next field of last extension header */ 918 if ((m = m_getptr(m, extoff + offsetof(struct ip6_ext, 919 ip6e_nxt), &off)) == NULL) 920 panic("%s: short ext mbuf chain", __func__); 921 *(mtod(m, caddr_t) + off) = proto; 922 m = *m0; 923 } else 924 ip6->ip6_nxt = proto; 925 926 if (hdrlen - sizeof(struct ip6_hdr) + total > IPV6_MAXPACKET) { 927 PF_FRAG_UNLOCK(); 928 DPFPRINTF(LOG_NOTICE, "drop: too big: %d", total); 929 ip6->ip6_plen = 0; 930 REASON_SET(reason, PFRES_SHORT); 931 /* PF_DROP requires a valid mbuf *m0 in pf_test6() */ 932 return (PF_DROP); 933 } 934 PF_FRAG_UNLOCK(); 935 936 DPFPRINTF(LOG_INFO, "complete: %p(%d)", m, ntohs(ip6->ip6_plen)); 937 return (PF_PASS); 938 939 fail: 940 PF_FRAG_UNLOCK(); 941 REASON_SET(reason, PFRES_MEMORY); 942 /* PF_DROP requires a valid mbuf *m0 in pf_test6(), will free later */ 943 return (PF_DROP); 944 } 945 946 int 947 pf_refragment6(struct mbuf **m0, struct m_tag *mtag, struct sockaddr_in6 *dst, 948 struct ifnet *ifp, struct rtentry *rt) 949 { 950 struct mbuf *m = *m0, *t; 951 struct pf_fragment_tag *ftag = (struct pf_fragment_tag *)(mtag + 1); 952 u_int32_t mtu; 953 u_int16_t hdrlen, extoff, maxlen; 954 u_int8_t proto; 955 int error, action; 956 957 hdrlen = ftag->ft_hdrlen; 958 extoff = ftag->ft_extoff; 959 maxlen = ftag->ft_maxlen; 960 m_tag_delete(m, mtag); 961 mtag = NULL; 962 ftag = NULL; 963 964 /* Checksum must be calculated for the whole packet */ 965 in6_proto_cksum_out(m, NULL); 966 967 if (extoff) { 968 int off; 969 970 /* Use protocol from next field of last extension header */ 971 if ((m = m_getptr(m, extoff + offsetof(struct ip6_ext, 972 ip6e_nxt), &off)) == NULL) 973 panic("%s: short ext mbuf chain", __func__); 974 proto = *(mtod(m, caddr_t) + off); 975 *(mtod(m, caddr_t) + off) = IPPROTO_FRAGMENT; 976 m = *m0; 977 } else { 978 struct ip6_hdr *hdr; 979 980 hdr = mtod(m, struct ip6_hdr *); 981 proto = hdr->ip6_nxt; 982 hdr->ip6_nxt = IPPROTO_FRAGMENT; 983 } 984 985 /* 986 * Maxlen may be less than 8 iff there was only a single 987 * fragment. As it was fragmented before, add a fragment 988 * header also for a single fragment. If total or maxlen 989 * is less than 8, ip6_fragment() will return EMSGSIZE and 990 * we drop the packet. 991 */ 992 mtu = hdrlen + sizeof(struct ip6_frag) + maxlen; 993 error = ip6_fragment(m, hdrlen, proto, mtu); 994 995 m = (*m0)->m_nextpkt; 996 (*m0)->m_nextpkt = NULL; 997 if (error == 0) { 998 /* The first mbuf contains the unfragmented packet */ 999 m_freemp(m0); 1000 action = PF_PASS; 1001 } else { 1002 /* Drop expects an mbuf to free */ 1003 DPFPRINTF(LOG_NOTICE, "refragment error %d", error); 1004 action = PF_DROP; 1005 } 1006 1007 for (t = m; m; m = t) { 1008 t = m->m_nextpkt; 1009 m->m_nextpkt = NULL; 1010 m->m_pkthdr.pf.flags |= PF_TAG_REFRAGMENTED; 1011 if (error == 0) { 1012 if (ifp == NULL) { 1013 ip6_forward(m, NULL, 0); 1014 } else if ((u_long)m->m_pkthdr.len <= ifp->if_mtu) { 1015 ifp->if_output(ifp, m, sin6tosa(dst), rt); 1016 } else { 1017 icmp6_error(m, ICMP6_PACKET_TOO_BIG, 0, 1018 ifp->if_mtu); 1019 } 1020 } else { 1021 m_freem(m); 1022 } 1023 } 1024 1025 return (action); 1026 } 1027 #endif /* INET6 */ 1028 1029 int 1030 pf_normalize_ip(struct pf_pdesc *pd, u_short *reason) 1031 { 1032 struct ip *h = mtod(pd->m, struct ip *); 1033 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3; 1034 u_int16_t mff = (ntohs(h->ip_off) & IP_MF); 1035 1036 if (!fragoff && !mff) 1037 goto no_fragment; 1038 1039 /* Clear IP_DF if we're in no-df mode */ 1040 if (pf_status.reass & PF_REASS_NODF && h->ip_off & htons(IP_DF)) 1041 h->ip_off &= htons(~IP_DF); 1042 1043 /* We're dealing with a fragment now. Don't allow fragments 1044 * with IP_DF to enter the cache. If the flag was cleared by 1045 * no-df above, fine. Otherwise drop it. 1046 */ 1047 if (h->ip_off & htons(IP_DF)) { 1048 DPFPRINTF(LOG_NOTICE, "bad fragment: IP_DF"); 1049 REASON_SET(reason, PFRES_FRAG); 1050 return (PF_DROP); 1051 } 1052 1053 if (!pf_status.reass) 1054 return (PF_PASS); /* no reassembly */ 1055 1056 /* Returns PF_DROP or m is NULL or completely reassembled mbuf */ 1057 if (pf_reassemble(&pd->m, pd->dir, reason) != PF_PASS) 1058 return (PF_DROP); 1059 if (pd->m == NULL) 1060 return (PF_PASS); /* packet has been reassembled, no error */ 1061 1062 h = mtod(pd->m, struct ip *); 1063 1064 no_fragment: 1065 /* At this point, only IP_DF is allowed in ip_off */ 1066 if (h->ip_off & ~htons(IP_DF)) 1067 h->ip_off &= htons(IP_DF); 1068 1069 return (PF_PASS); 1070 } 1071 1072 #ifdef INET6 1073 int 1074 pf_normalize_ip6(struct pf_pdesc *pd, u_short *reason) 1075 { 1076 struct ip6_frag frag; 1077 1078 if (pd->fragoff == 0) 1079 goto no_fragment; 1080 1081 if (!pf_pull_hdr(pd->m, pd->fragoff, &frag, sizeof(frag), NULL, reason, 1082 AF_INET6)) 1083 return (PF_DROP); 1084 1085 if (!pf_status.reass) 1086 return (PF_PASS); /* no reassembly */ 1087 1088 /* Returns PF_DROP or m is NULL or completely reassembled mbuf */ 1089 if (pf_reassemble6(&pd->m, &frag, pd->fragoff + sizeof(frag), 1090 pd->extoff, pd->dir, reason) != PF_PASS) 1091 return (PF_DROP); 1092 if (pd->m == NULL) 1093 return (PF_PASS); /* packet has been reassembled, no error */ 1094 1095 no_fragment: 1096 return (PF_PASS); 1097 } 1098 #endif /* INET6 */ 1099 1100 int 1101 pf_normalize_tcp(struct pf_pdesc *pd) 1102 { 1103 struct tcphdr *th = &pd->hdr.tcp; 1104 u_short reason; 1105 u_int8_t flags; 1106 u_int rewrite = 0; 1107 1108 flags = th->th_flags; 1109 if (flags & TH_SYN) { 1110 /* Illegal packet */ 1111 if (flags & TH_RST) 1112 goto tcp_drop; 1113 1114 if (flags & TH_FIN) /* XXX why clear instead of drop? */ 1115 flags &= ~TH_FIN; 1116 } else { 1117 /* Illegal packet */ 1118 if (!(flags & (TH_ACK|TH_RST))) 1119 goto tcp_drop; 1120 } 1121 1122 if (!(flags & TH_ACK)) { 1123 /* These flags are only valid if ACK is set */ 1124 if (flags & (TH_FIN|TH_PUSH|TH_URG)) 1125 goto tcp_drop; 1126 } 1127 1128 /* If flags changed, or reserved data set, then adjust */ 1129 if (flags != th->th_flags || th->th_x2 != 0) { 1130 /* hack: set 4-bit th_x2 = 0 */ 1131 u_int8_t *th_off = (u_int8_t*)(&th->th_ack+1); 1132 pf_patch_8(pd, th_off, th->th_off << 4, PF_HI); 1133 1134 pf_patch_8(pd, &th->th_flags, flags, PF_LO); 1135 rewrite = 1; 1136 } 1137 1138 /* Remove urgent pointer, if TH_URG is not set */ 1139 if (!(flags & TH_URG) && th->th_urp) { 1140 pf_patch_16(pd, &th->th_urp, 0); 1141 rewrite = 1; 1142 } 1143 1144 /* copy back packet headers if we sanitized */ 1145 if (rewrite) { 1146 m_copyback(pd->m, pd->off, sizeof(*th), th, M_NOWAIT); 1147 } 1148 1149 return (PF_PASS); 1150 1151 tcp_drop: 1152 REASON_SET(&reason, PFRES_NORM); 1153 return (PF_DROP); 1154 } 1155 1156 int 1157 pf_normalize_tcp_init(struct pf_pdesc *pd, struct pf_state_peer *src) 1158 { 1159 struct tcphdr *th = &pd->hdr.tcp; 1160 u_int32_t tsval, tsecr; 1161 int olen; 1162 u_int8_t opts[MAX_TCPOPTLEN], *opt; 1163 1164 1165 KASSERT(src->scrub == NULL); 1166 1167 src->scrub = pool_get(&pf_state_scrub_pl, PR_NOWAIT); 1168 if (src->scrub == NULL) 1169 return (1); 1170 memset(src->scrub, 0, sizeof(*src->scrub)); 1171 1172 switch (pd->af) { 1173 case AF_INET: { 1174 struct ip *h = mtod(pd->m, struct ip *); 1175 src->scrub->pfss_ttl = h->ip_ttl; 1176 break; 1177 } 1178 #ifdef INET6 1179 case AF_INET6: { 1180 struct ip6_hdr *h = mtod(pd->m, struct ip6_hdr *); 1181 src->scrub->pfss_ttl = h->ip6_hlim; 1182 break; 1183 } 1184 #endif /* INET6 */ 1185 default: 1186 unhandled_af(pd->af); 1187 } 1188 1189 /* 1190 * All normalizations below are only begun if we see the start of 1191 * the connections. They must all set an enabled bit in pfss_flags 1192 */ 1193 if ((th->th_flags & TH_SYN) == 0) 1194 return (0); 1195 1196 olen = (th->th_off << 2) - sizeof(*th); 1197 if (olen < TCPOLEN_TIMESTAMP || !pf_pull_hdr(pd->m, 1198 pd->off + sizeof(*th), opts, olen, NULL, NULL, pd->af)) 1199 return (0); 1200 1201 opt = opts; 1202 while ((opt = pf_find_tcpopt(opt, opts, olen, 1203 TCPOPT_TIMESTAMP, TCPOLEN_TIMESTAMP)) != NULL) { 1204 1205 src->scrub->pfss_flags |= PFSS_TIMESTAMP; 1206 src->scrub->pfss_ts_mod = arc4random(); 1207 /* note PFSS_PAWS not set yet */ 1208 memcpy(&tsval, &opt[2], sizeof(u_int32_t)); 1209 memcpy(&tsecr, &opt[6], sizeof(u_int32_t)); 1210 src->scrub->pfss_tsval0 = ntohl(tsval); 1211 src->scrub->pfss_tsval = ntohl(tsval); 1212 src->scrub->pfss_tsecr = ntohl(tsecr); 1213 getmicrouptime(&src->scrub->pfss_last); 1214 1215 opt += opt[1]; 1216 } 1217 1218 return (0); 1219 } 1220 1221 void 1222 pf_normalize_tcp_cleanup(struct pf_state *state) 1223 { 1224 if (state->src.scrub) 1225 pool_put(&pf_state_scrub_pl, state->src.scrub); 1226 if (state->dst.scrub) 1227 pool_put(&pf_state_scrub_pl, state->dst.scrub); 1228 1229 /* Someday... flush the TCP segment reassembly descriptors. */ 1230 } 1231 1232 int 1233 pf_normalize_tcp_stateful(struct pf_pdesc *pd, u_short *reason, 1234 struct pf_state *state, struct pf_state_peer *src, 1235 struct pf_state_peer *dst, int *writeback) 1236 { 1237 struct tcphdr *th = &pd->hdr.tcp; 1238 struct timeval uptime; 1239 u_int tsval_from_last; 1240 u_int32_t tsval, tsecr; 1241 int copyback = 0; 1242 int got_ts = 0; 1243 int olen; 1244 u_int8_t opts[MAX_TCPOPTLEN], *opt; 1245 1246 KASSERT(src->scrub || dst->scrub); 1247 1248 /* 1249 * Enforce the minimum TTL seen for this connection. Negate a common 1250 * technique to evade an intrusion detection system and confuse 1251 * firewall state code. 1252 */ 1253 switch (pd->af) { 1254 case AF_INET: 1255 if (src->scrub) { 1256 struct ip *h = mtod(pd->m, struct ip *); 1257 if (h->ip_ttl > src->scrub->pfss_ttl) 1258 src->scrub->pfss_ttl = h->ip_ttl; 1259 h->ip_ttl = src->scrub->pfss_ttl; 1260 } 1261 break; 1262 #ifdef INET6 1263 case AF_INET6: 1264 if (src->scrub) { 1265 struct ip6_hdr *h = mtod(pd->m, struct ip6_hdr *); 1266 if (h->ip6_hlim > src->scrub->pfss_ttl) 1267 src->scrub->pfss_ttl = h->ip6_hlim; 1268 h->ip6_hlim = src->scrub->pfss_ttl; 1269 } 1270 break; 1271 #endif /* INET6 */ 1272 default: 1273 unhandled_af(pd->af); 1274 } 1275 1276 olen = (th->th_off << 2) - sizeof(*th); 1277 1278 if (olen >= TCPOLEN_TIMESTAMP && 1279 ((src->scrub && (src->scrub->pfss_flags & PFSS_TIMESTAMP)) || 1280 (dst->scrub && (dst->scrub->pfss_flags & PFSS_TIMESTAMP))) && 1281 pf_pull_hdr(pd->m, pd->off + sizeof(*th), opts, olen, NULL, NULL, 1282 pd->af)) { 1283 1284 /* Modulate the timestamps. Can be used for NAT detection, OS 1285 * uptime determination or reboot detection. 1286 */ 1287 opt = opts; 1288 while ((opt = pf_find_tcpopt(opt, opts, olen, 1289 TCPOPT_TIMESTAMP, TCPOLEN_TIMESTAMP)) != NULL) { 1290 1291 u_int8_t *ts = opt + 2; 1292 u_int8_t *tsr = opt + 6; 1293 1294 if (got_ts) { 1295 /* Huh? Multiple timestamps!? */ 1296 if (pf_status.debug >= LOG_NOTICE) { 1297 log(LOG_NOTICE, 1298 "pf: %s: multiple TS??", __func__); 1299 pf_print_state(state); 1300 addlog("\n"); 1301 } 1302 REASON_SET(reason, PFRES_TS); 1303 return (PF_DROP); 1304 } 1305 1306 memcpy(&tsval, ts, sizeof(u_int32_t)); 1307 memcpy(&tsecr, tsr, sizeof(u_int32_t)); 1308 1309 /* modulate TS */ 1310 if (tsval && src->scrub && 1311 (src->scrub->pfss_flags & PFSS_TIMESTAMP)) { 1312 /* tsval used further on */ 1313 tsval = ntohl(tsval); 1314 pf_patch_32_unaligned(pd, 1315 ts, htonl(tsval + src->scrub->pfss_ts_mod), 1316 PF_ALGNMNT(ts - opts)); 1317 copyback = 1; 1318 } 1319 1320 /* modulate TS reply if any (!0) */ 1321 if (tsecr && dst->scrub && 1322 (dst->scrub->pfss_flags & PFSS_TIMESTAMP)) { 1323 /* tsecr used further on */ 1324 tsecr = ntohl(tsecr) - dst->scrub->pfss_ts_mod; 1325 pf_patch_32_unaligned(pd, 1326 tsr, htonl(tsecr), PF_ALGNMNT(tsr - opts)); 1327 copyback = 1; 1328 } 1329 1330 got_ts = 1; 1331 opt += opt[1]; 1332 } 1333 1334 if (copyback) { 1335 /* Copyback the options, caller copys back header */ 1336 *writeback = 1; 1337 m_copyback(pd->m, pd->off + sizeof(*th), olen, opts, M_NOWAIT); 1338 } 1339 } 1340 1341 1342 /* 1343 * Must invalidate PAWS checks on connections idle for too long. 1344 * The fastest allowed timestamp clock is 1ms. That turns out to 1345 * be about 24 days before it wraps. XXX Right now our lowerbound 1346 * TS echo check only works for the first 12 days of a connection 1347 * when the TS has exhausted half its 32bit space 1348 */ 1349 #define TS_MAX_IDLE (24*24*60*60) 1350 #define TS_MAX_CONN (12*24*60*60) /* XXX remove when better tsecr check */ 1351 1352 getmicrouptime(&uptime); 1353 if (src->scrub && (src->scrub->pfss_flags & PFSS_PAWS) && 1354 (uptime.tv_sec - src->scrub->pfss_last.tv_sec > TS_MAX_IDLE || 1355 getuptime() - state->creation > TS_MAX_CONN)) { 1356 if (pf_status.debug >= LOG_NOTICE) { 1357 log(LOG_NOTICE, "pf: src idled out of PAWS "); 1358 pf_print_state(state); 1359 addlog("\n"); 1360 } 1361 src->scrub->pfss_flags = 1362 (src->scrub->pfss_flags & ~PFSS_PAWS) | PFSS_PAWS_IDLED; 1363 } 1364 if (dst->scrub && (dst->scrub->pfss_flags & PFSS_PAWS) && 1365 uptime.tv_sec - dst->scrub->pfss_last.tv_sec > TS_MAX_IDLE) { 1366 if (pf_status.debug >= LOG_NOTICE) { 1367 log(LOG_NOTICE, "pf: dst idled out of PAWS "); 1368 pf_print_state(state); 1369 addlog("\n"); 1370 } 1371 dst->scrub->pfss_flags = 1372 (dst->scrub->pfss_flags & ~PFSS_PAWS) | PFSS_PAWS_IDLED; 1373 } 1374 1375 if (got_ts && src->scrub && dst->scrub && 1376 (src->scrub->pfss_flags & PFSS_PAWS) && 1377 (dst->scrub->pfss_flags & PFSS_PAWS)) { 1378 /* Validate that the timestamps are "in-window". 1379 * RFC1323 describes TCP Timestamp options that allow 1380 * measurement of RTT (round trip time) and PAWS 1381 * (protection against wrapped sequence numbers). PAWS 1382 * gives us a set of rules for rejecting packets on 1383 * long fat pipes (packets that were somehow delayed 1384 * in transit longer than the time it took to send the 1385 * full TCP sequence space of 4Gb). We can use these 1386 * rules and infer a few others that will let us treat 1387 * the 32bit timestamp and the 32bit echoed timestamp 1388 * as sequence numbers to prevent a blind attacker from 1389 * inserting packets into a connection. 1390 * 1391 * RFC1323 tells us: 1392 * - The timestamp on this packet must be greater than 1393 * or equal to the last value echoed by the other 1394 * endpoint. The RFC says those will be discarded 1395 * since it is a dup that has already been acked. 1396 * This gives us a lowerbound on the timestamp. 1397 * timestamp >= other last echoed timestamp 1398 * - The timestamp will be less than or equal to 1399 * the last timestamp plus the time between the 1400 * last packet and now. The RFC defines the max 1401 * clock rate as 1ms. We will allow clocks to be 1402 * up to 10% fast and will allow a total difference 1403 * or 30 seconds due to a route change. And this 1404 * gives us an upperbound on the timestamp. 1405 * timestamp <= last timestamp + max ticks 1406 * We have to be careful here. Windows will send an 1407 * initial timestamp of zero and then initialize it 1408 * to a random value after the 3whs; presumably to 1409 * avoid a DoS by having to call an expensive RNG 1410 * during a SYN flood. Proof MS has at least one 1411 * good security geek. 1412 * 1413 * - The TCP timestamp option must also echo the other 1414 * endpoints timestamp. The timestamp echoed is the 1415 * one carried on the earliest unacknowledged segment 1416 * on the left edge of the sequence window. The RFC 1417 * states that the host will reject any echoed 1418 * timestamps that were larger than any ever sent. 1419 * This gives us an upperbound on the TS echo. 1420 * tescr <= largest_tsval 1421 * - The lowerbound on the TS echo is a little more 1422 * tricky to determine. The other endpoint's echoed 1423 * values will not decrease. But there may be 1424 * network conditions that re-order packets and 1425 * cause our view of them to decrease. For now the 1426 * only lowerbound we can safely determine is that 1427 * the TS echo will never be less than the original 1428 * TS. XXX There is probably a better lowerbound. 1429 * Remove TS_MAX_CONN with better lowerbound check. 1430 * tescr >= other original TS 1431 * 1432 * It is also important to note that the fastest 1433 * timestamp clock of 1ms will wrap its 32bit space in 1434 * 24 days. So we just disable TS checking after 24 1435 * days of idle time. We actually must use a 12d 1436 * connection limit until we can come up with a better 1437 * lowerbound to the TS echo check. 1438 */ 1439 struct timeval delta_ts; 1440 int ts_fudge; 1441 1442 /* 1443 * PFTM_TS_DIFF is how many seconds of leeway to allow 1444 * a host's timestamp. This can happen if the previous 1445 * packet got delayed in transit for much longer than 1446 * this packet. 1447 */ 1448 if ((ts_fudge = state->rule.ptr->timeout[PFTM_TS_DIFF]) == 0) 1449 ts_fudge = pf_default_rule.timeout[PFTM_TS_DIFF]; 1450 1451 /* Calculate max ticks since the last timestamp */ 1452 #define TS_MAXFREQ 1100 /* RFC max TS freq of 1Khz + 10% skew */ 1453 #define TS_MICROSECS 1000000 /* microseconds per second */ 1454 timersub(&uptime, &src->scrub->pfss_last, &delta_ts); 1455 tsval_from_last = (delta_ts.tv_sec + ts_fudge) * TS_MAXFREQ; 1456 tsval_from_last += delta_ts.tv_usec / (TS_MICROSECS/TS_MAXFREQ); 1457 1458 if ((src->state >= TCPS_ESTABLISHED && 1459 dst->state >= TCPS_ESTABLISHED) && 1460 (SEQ_LT(tsval, dst->scrub->pfss_tsecr) || 1461 SEQ_GT(tsval, src->scrub->pfss_tsval + tsval_from_last) || 1462 (tsecr && (SEQ_GT(tsecr, dst->scrub->pfss_tsval) || 1463 SEQ_LT(tsecr, dst->scrub->pfss_tsval0))))) { 1464 /* Bad RFC1323 implementation or an insertion attack. 1465 * 1466 * - Solaris 2.6 and 2.7 are known to send another ACK 1467 * after the FIN,FIN|ACK,ACK closing that carries 1468 * an old timestamp. 1469 */ 1470 1471 DPFPRINTF(LOG_NOTICE, "Timestamp failed %c%c%c%c", 1472 SEQ_LT(tsval, dst->scrub->pfss_tsecr) ? '0' : ' ', 1473 SEQ_GT(tsval, src->scrub->pfss_tsval + 1474 tsval_from_last) ? '1' : ' ', 1475 SEQ_GT(tsecr, dst->scrub->pfss_tsval) ? '2' : ' ', 1476 SEQ_LT(tsecr, dst->scrub->pfss_tsval0)? '3' : ' '); 1477 DPFPRINTF(LOG_NOTICE, " tsval: %u tsecr: %u " 1478 "+ticks: %u idle: %llu.%06lus", tsval, tsecr, 1479 tsval_from_last, (long long)delta_ts.tv_sec, 1480 delta_ts.tv_usec); 1481 DPFPRINTF(LOG_NOTICE, " src->tsval: %u tsecr: %u", 1482 src->scrub->pfss_tsval, src->scrub->pfss_tsecr); 1483 DPFPRINTF(LOG_NOTICE, " dst->tsval: %u tsecr: %u " 1484 "tsval0: %u", dst->scrub->pfss_tsval, 1485 dst->scrub->pfss_tsecr, dst->scrub->pfss_tsval0); 1486 if (pf_status.debug >= LOG_NOTICE) { 1487 log(LOG_NOTICE, "pf: "); 1488 pf_print_state(state); 1489 pf_print_flags(th->th_flags); 1490 addlog("\n"); 1491 } 1492 REASON_SET(reason, PFRES_TS); 1493 return (PF_DROP); 1494 } 1495 /* XXX I'd really like to require tsecr but it's optional */ 1496 } else if (!got_ts && (th->th_flags & TH_RST) == 0 && 1497 ((src->state == TCPS_ESTABLISHED && dst->state == TCPS_ESTABLISHED) 1498 || pd->p_len > 0 || (th->th_flags & TH_SYN)) && 1499 src->scrub && dst->scrub && 1500 (src->scrub->pfss_flags & PFSS_PAWS) && 1501 (dst->scrub->pfss_flags & PFSS_PAWS)) { 1502 /* Didn't send a timestamp. Timestamps aren't really useful 1503 * when: 1504 * - connection opening or closing (often not even sent). 1505 * but we must not let an attacker to put a FIN on a 1506 * data packet to sneak it through our ESTABLISHED check. 1507 * - on a TCP reset. RFC suggests not even looking at TS. 1508 * - on an empty ACK. The TS will not be echoed so it will 1509 * probably not help keep the RTT calculation in sync and 1510 * there isn't as much danger when the sequence numbers 1511 * got wrapped. So some stacks don't include TS on empty 1512 * ACKs :-( 1513 * 1514 * To minimize the disruption to mostly RFC1323 conformant 1515 * stacks, we will only require timestamps on data packets. 1516 * 1517 * And what do ya know, we cannot require timestamps on data 1518 * packets. There appear to be devices that do legitimate 1519 * TCP connection hijacking. There are HTTP devices that allow 1520 * a 3whs (with timestamps) and then buffer the HTTP request. 1521 * If the intermediate device has the HTTP response cache, it 1522 * will spoof the response but not bother timestamping its 1523 * packets. So we can look for the presence of a timestamp in 1524 * the first data packet and if there, require it in all future 1525 * packets. 1526 */ 1527 1528 if (pd->p_len > 0 && (src->scrub->pfss_flags & PFSS_DATA_TS)) { 1529 /* 1530 * Hey! Someone tried to sneak a packet in. Or the 1531 * stack changed its RFC1323 behavior?!?! 1532 */ 1533 if (pf_status.debug >= LOG_NOTICE) { 1534 log(LOG_NOTICE, 1535 "pf: did not receive expected RFC1323 " 1536 "timestamp"); 1537 pf_print_state(state); 1538 pf_print_flags(th->th_flags); 1539 addlog("\n"); 1540 } 1541 REASON_SET(reason, PFRES_TS); 1542 return (PF_DROP); 1543 } 1544 } 1545 1546 /* 1547 * We will note if a host sends his data packets with or without 1548 * timestamps. And require all data packets to contain a timestamp 1549 * if the first does. PAWS implicitly requires that all data packets be 1550 * timestamped. But I think there are middle-man devices that hijack 1551 * TCP streams immediately after the 3whs and don't timestamp their 1552 * packets (seen in a WWW accelerator or cache). 1553 */ 1554 if (pd->p_len > 0 && src->scrub && (src->scrub->pfss_flags & 1555 (PFSS_TIMESTAMP|PFSS_DATA_TS|PFSS_DATA_NOTS)) == PFSS_TIMESTAMP) { 1556 if (got_ts) 1557 src->scrub->pfss_flags |= PFSS_DATA_TS; 1558 else { 1559 src->scrub->pfss_flags |= PFSS_DATA_NOTS; 1560 if (pf_status.debug >= LOG_NOTICE && dst->scrub && 1561 (dst->scrub->pfss_flags & PFSS_TIMESTAMP)) { 1562 /* Don't warn if other host rejected RFC1323 */ 1563 log(LOG_NOTICE, 1564 "pf: broken RFC1323 stack did not " 1565 "timestamp data packet. Disabled PAWS " 1566 "security."); 1567 pf_print_state(state); 1568 pf_print_flags(th->th_flags); 1569 addlog("\n"); 1570 } 1571 } 1572 } 1573 1574 /* 1575 * Update PAWS values 1576 */ 1577 if (got_ts && src->scrub && PFSS_TIMESTAMP == (src->scrub->pfss_flags & 1578 (PFSS_PAWS_IDLED|PFSS_TIMESTAMP))) { 1579 getmicrouptime(&src->scrub->pfss_last); 1580 if (SEQ_GEQ(tsval, src->scrub->pfss_tsval) || 1581 (src->scrub->pfss_flags & PFSS_PAWS) == 0) 1582 src->scrub->pfss_tsval = tsval; 1583 1584 if (tsecr) { 1585 if (SEQ_GEQ(tsecr, src->scrub->pfss_tsecr) || 1586 (src->scrub->pfss_flags & PFSS_PAWS) == 0) 1587 src->scrub->pfss_tsecr = tsecr; 1588 1589 if ((src->scrub->pfss_flags & PFSS_PAWS) == 0 && 1590 (SEQ_LT(tsval, src->scrub->pfss_tsval0) || 1591 src->scrub->pfss_tsval0 == 0)) { 1592 /* tsval0 MUST be the lowest timestamp */ 1593 src->scrub->pfss_tsval0 = tsval; 1594 } 1595 1596 /* Only fully initialized after a TS gets echoed */ 1597 if ((src->scrub->pfss_flags & PFSS_PAWS) == 0) 1598 src->scrub->pfss_flags |= PFSS_PAWS; 1599 } 1600 } 1601 1602 /* I have a dream.... TCP segment reassembly.... */ 1603 return (0); 1604 } 1605 1606 int 1607 pf_normalize_mss(struct pf_pdesc *pd, u_int16_t maxmss) 1608 { 1609 int olen, optsoff; 1610 u_int8_t opts[MAX_TCPOPTLEN], *opt; 1611 1612 olen = (pd->hdr.tcp.th_off << 2) - sizeof(struct tcphdr); 1613 optsoff = pd->off + sizeof(struct tcphdr); 1614 if (olen < TCPOLEN_MAXSEG || 1615 !pf_pull_hdr(pd->m, optsoff, opts, olen, NULL, NULL, pd->af)) 1616 return (0); 1617 1618 opt = opts; 1619 while ((opt = pf_find_tcpopt(opt, opts, olen, 1620 TCPOPT_MAXSEG, TCPOLEN_MAXSEG)) != NULL) { 1621 u_int16_t mss; 1622 u_int8_t *mssp = opt + 2; 1623 memcpy(&mss, mssp, sizeof(mss)); 1624 if (ntohs(mss) > maxmss) { 1625 size_t mssoffopts = mssp - opts; 1626 pf_patch_16_unaligned(pd, &mss, 1627 htons(maxmss), PF_ALGNMNT(mssoffopts)); 1628 m_copyback(pd->m, optsoff + mssoffopts, 1629 sizeof(mss), &mss, M_NOWAIT); 1630 m_copyback(pd->m, pd->off, 1631 sizeof(struct tcphdr), &pd->hdr.tcp, M_NOWAIT); 1632 } 1633 1634 opt += opt[1]; 1635 } 1636 1637 return (0); 1638 } 1639 1640 void 1641 pf_scrub(struct mbuf *m, u_int16_t flags, sa_family_t af, u_int8_t min_ttl, 1642 u_int8_t tos) 1643 { 1644 struct ip *h = mtod(m, struct ip *); 1645 #ifdef INET6 1646 struct ip6_hdr *h6 = mtod(m, struct ip6_hdr *); 1647 #endif /* INET6 */ 1648 1649 /* Clear IP_DF if no-df was requested */ 1650 if (flags & PFSTATE_NODF && af == AF_INET && h->ip_off & htons(IP_DF)) 1651 h->ip_off &= htons(~IP_DF); 1652 1653 /* Enforce a minimum ttl, may cause endless packet loops */ 1654 if (min_ttl && af == AF_INET && h->ip_ttl < min_ttl) 1655 h->ip_ttl = min_ttl; 1656 #ifdef INET6 1657 if (min_ttl && af == AF_INET6 && h6->ip6_hlim < min_ttl) 1658 h6->ip6_hlim = min_ttl; 1659 #endif /* INET6 */ 1660 1661 /* Enforce tos */ 1662 if (flags & PFSTATE_SETTOS) { 1663 if (af == AF_INET) 1664 h->ip_tos = tos | (h->ip_tos & IPTOS_ECN_MASK); 1665 #ifdef INET6 1666 if (af == AF_INET6) { 1667 /* drugs are unable to explain such idiocy */ 1668 h6->ip6_flow &= ~htonl(0x0fc00000); 1669 h6->ip6_flow |= htonl(((u_int32_t)tos) << 20); 1670 } 1671 #endif /* INET6 */ 1672 } 1673 1674 /* random-id, but not for fragments */ 1675 if (flags & PFSTATE_RANDOMID && af == AF_INET && 1676 !(h->ip_off & ~htons(IP_DF))) 1677 h->ip_id = htons(ip_randomid()); 1678 } 1679