1 /*- 2 * Copyright 2001 Niels Provos <provos@citi.umich.edu> 3 * Copyright 2011 Alexander Bluhm <bluhm@openbsd.org> 4 * All rights reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 16 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 17 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 18 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 19 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 20 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 21 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 22 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 23 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 24 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 25 * 26 * $OpenBSD: pf_norm.c,v 1.114 2009/01/29 14:11:45 henning Exp $ 27 */ 28 29 #include <sys/cdefs.h> 30 __FBSDID("$FreeBSD$"); 31 32 #include "opt_inet.h" 33 #include "opt_inet6.h" 34 #include "opt_pf.h" 35 36 #include <sys/param.h> 37 #include <sys/kernel.h> 38 #include <sys/lock.h> 39 #include <sys/mbuf.h> 40 #include <sys/mutex.h> 41 #include <sys/refcount.h> 42 #include <sys/rwlock.h> 43 #include <sys/socket.h> 44 45 #include <net/if.h> 46 #include <net/vnet.h> 47 #include <net/pfvar.h> 48 #include <net/if_pflog.h> 49 50 #include <netinet/in.h> 51 #include <netinet/ip.h> 52 #include <netinet/ip_var.h> 53 #include <netinet6/ip6_var.h> 54 #include <netinet/tcp.h> 55 #include <netinet/tcp_fsm.h> 56 #include <netinet/tcp_seq.h> 57 58 #ifdef INET6 59 #include <netinet/ip6.h> 60 #endif /* INET6 */ 61 62 struct pf_frent { 63 TAILQ_ENTRY(pf_frent) fr_next; 64 struct mbuf *fe_m; 65 uint16_t fe_hdrlen; /* ipv4 header lenght with ip options 66 ipv6, extension, fragment header */ 67 uint16_t fe_extoff; /* last extension header offset or 0 */ 68 uint16_t fe_len; /* fragment length */ 69 uint16_t fe_off; /* fragment offset */ 70 uint16_t fe_mff; /* more fragment flag */ 71 }; 72 73 struct pf_fragment_cmp { 74 struct pf_addr frc_src; 75 struct pf_addr frc_dst; 76 uint32_t frc_id; 77 sa_family_t frc_af; 78 uint8_t frc_proto; 79 uint8_t frc_direction; 80 }; 81 82 struct pf_fragment { 83 struct pf_fragment_cmp fr_key; 84 #define fr_src fr_key.frc_src 85 #define fr_dst fr_key.frc_dst 86 #define fr_id fr_key.frc_id 87 #define fr_af fr_key.frc_af 88 #define fr_proto fr_key.frc_proto 89 #define fr_direction fr_key.frc_direction 90 91 RB_ENTRY(pf_fragment) fr_entry; 92 TAILQ_ENTRY(pf_fragment) frag_next; 93 uint8_t fr_flags; /* status flags */ 94 #define PFFRAG_SEENLAST 0x0001 /* Seen the last fragment for this */ 95 #define PFFRAG_NOBUFFER 0x0002 /* Non-buffering fragment cache */ 96 #define PFFRAG_DROP 0x0004 /* Drop all fragments */ 97 #define BUFFER_FRAGMENTS(fr) (!((fr)->fr_flags & PFFRAG_NOBUFFER)) 98 uint16_t fr_max; /* fragment data max */ 99 uint32_t fr_timeout; 100 uint16_t fr_maxlen; /* maximum length of single fragment */ 101 TAILQ_HEAD(pf_fragq, pf_frent) fr_queue; 102 }; 103 104 struct pf_fragment_tag { 105 uint16_t ft_hdrlen; /* header length of reassembled pkt */ 106 uint16_t ft_extoff; /* last extension header offset or 0 */ 107 uint16_t ft_maxlen; /* maximum fragment payload length */ 108 uint32_t ft_id; /* fragment id */ 109 }; 110 111 static struct mtx pf_frag_mtx; 112 MTX_SYSINIT(pf_frag_mtx, &pf_frag_mtx, "pf fragments", MTX_DEF); 113 #define PF_FRAG_LOCK() mtx_lock(&pf_frag_mtx) 114 #define PF_FRAG_UNLOCK() mtx_unlock(&pf_frag_mtx) 115 #define PF_FRAG_ASSERT() mtx_assert(&pf_frag_mtx, MA_OWNED) 116 117 VNET_DEFINE(uma_zone_t, pf_state_scrub_z); /* XXX: shared with pfsync */ 118 119 static VNET_DEFINE(uma_zone_t, pf_frent_z); 120 #define V_pf_frent_z VNET(pf_frent_z) 121 static VNET_DEFINE(uma_zone_t, pf_frag_z); 122 #define V_pf_frag_z VNET(pf_frag_z) 123 124 TAILQ_HEAD(pf_fragqueue, pf_fragment); 125 TAILQ_HEAD(pf_cachequeue, pf_fragment); 126 static VNET_DEFINE(struct pf_fragqueue, pf_fragqueue); 127 #define V_pf_fragqueue VNET(pf_fragqueue) 128 static VNET_DEFINE(struct pf_cachequeue, pf_cachequeue); 129 #define V_pf_cachequeue VNET(pf_cachequeue) 130 RB_HEAD(pf_frag_tree, pf_fragment); 131 static VNET_DEFINE(struct pf_frag_tree, pf_frag_tree); 132 #define V_pf_frag_tree VNET(pf_frag_tree) 133 static VNET_DEFINE(struct pf_frag_tree, pf_cache_tree); 134 #define V_pf_cache_tree VNET(pf_cache_tree) 135 static int pf_frag_compare(struct pf_fragment *, 136 struct pf_fragment *); 137 static RB_PROTOTYPE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare); 138 static RB_GENERATE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare); 139 140 static void pf_flush_fragments(void); 141 static void pf_free_fragment(struct pf_fragment *); 142 static void pf_remove_fragment(struct pf_fragment *); 143 static int pf_normalize_tcpopt(struct pf_rule *, struct mbuf *, 144 struct tcphdr *, int, sa_family_t); 145 static struct pf_frent *pf_create_fragment(u_short *); 146 static struct pf_fragment *pf_find_fragment(struct pf_fragment_cmp *key, 147 struct pf_frag_tree *tree); 148 static struct pf_fragment *pf_fillup_fragment(struct pf_fragment_cmp *, 149 struct pf_frent *, u_short *); 150 static int pf_isfull_fragment(struct pf_fragment *); 151 static struct mbuf *pf_join_fragment(struct pf_fragment *); 152 #ifdef INET 153 static void pf_scrub_ip(struct mbuf **, uint32_t, uint8_t, uint8_t); 154 static int pf_reassemble(struct mbuf **, struct ip *, int, u_short *); 155 static struct mbuf *pf_fragcache(struct mbuf **, struct ip*, 156 struct pf_fragment **, int, int, int *); 157 #endif /* INET */ 158 #ifdef INET6 159 static int pf_reassemble6(struct mbuf **, struct ip6_hdr *, 160 struct ip6_frag *, uint16_t, uint16_t, int, u_short *); 161 static void pf_scrub_ip6(struct mbuf **, uint8_t); 162 #endif /* INET6 */ 163 164 #define DPFPRINTF(x) do { \ 165 if (V_pf_status.debug >= PF_DEBUG_MISC) { \ 166 printf("%s: ", __func__); \ 167 printf x ; \ 168 } \ 169 } while(0) 170 171 #ifdef INET 172 static void 173 pf_ip2key(struct ip *ip, int dir, struct pf_fragment_cmp *key) 174 { 175 176 key->frc_src.v4 = ip->ip_src; 177 key->frc_dst.v4 = ip->ip_dst; 178 key->frc_af = AF_INET; 179 key->frc_proto = ip->ip_p; 180 key->frc_id = ip->ip_id; 181 key->frc_direction = dir; 182 } 183 #endif /* INET */ 184 185 void 186 pf_normalize_init(void) 187 { 188 189 V_pf_frag_z = uma_zcreate("pf frags", sizeof(struct pf_fragment), 190 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 191 V_pf_frent_z = uma_zcreate("pf frag entries", sizeof(struct pf_frent), 192 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 193 V_pf_state_scrub_z = uma_zcreate("pf state scrubs", 194 sizeof(struct pf_state_scrub), NULL, NULL, NULL, NULL, 195 UMA_ALIGN_PTR, 0); 196 197 V_pf_limits[PF_LIMIT_FRAGS].zone = V_pf_frent_z; 198 V_pf_limits[PF_LIMIT_FRAGS].limit = PFFRAG_FRENT_HIWAT; 199 uma_zone_set_max(V_pf_frent_z, PFFRAG_FRENT_HIWAT); 200 uma_zone_set_warning(V_pf_frent_z, "PF frag entries limit reached"); 201 202 TAILQ_INIT(&V_pf_fragqueue); 203 TAILQ_INIT(&V_pf_cachequeue); 204 } 205 206 void 207 pf_normalize_cleanup(void) 208 { 209 210 uma_zdestroy(V_pf_state_scrub_z); 211 uma_zdestroy(V_pf_frent_z); 212 uma_zdestroy(V_pf_frag_z); 213 } 214 215 static int 216 pf_frag_compare(struct pf_fragment *a, struct pf_fragment *b) 217 { 218 int diff; 219 220 if ((diff = a->fr_id - b->fr_id) != 0) 221 return (diff); 222 if ((diff = a->fr_proto - b->fr_proto) != 0) 223 return (diff); 224 if ((diff = a->fr_af - b->fr_af) != 0) 225 return (diff); 226 if ((diff = pf_addr_cmp(&a->fr_src, &b->fr_src, a->fr_af)) != 0) 227 return (diff); 228 if ((diff = pf_addr_cmp(&a->fr_dst, &b->fr_dst, a->fr_af)) != 0) 229 return (diff); 230 return (0); 231 } 232 233 void 234 pf_purge_expired_fragments(void) 235 { 236 struct pf_fragment *frag; 237 u_int32_t expire = time_uptime - 238 V_pf_default_rule.timeout[PFTM_FRAG]; 239 240 PF_FRAG_LOCK(); 241 while ((frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue)) != NULL) { 242 KASSERT((BUFFER_FRAGMENTS(frag)), 243 ("BUFFER_FRAGMENTS(frag) == 0: %s", __FUNCTION__)); 244 if (frag->fr_timeout > expire) 245 break; 246 247 DPFPRINTF(("expiring %d(%p)\n", frag->fr_id, frag)); 248 pf_free_fragment(frag); 249 } 250 251 while ((frag = TAILQ_LAST(&V_pf_cachequeue, pf_cachequeue)) != NULL) { 252 KASSERT((!BUFFER_FRAGMENTS(frag)), 253 ("BUFFER_FRAGMENTS(frag) != 0: %s", __FUNCTION__)); 254 if (frag->fr_timeout > expire) 255 break; 256 257 DPFPRINTF(("expiring %d(%p)\n", frag->fr_id, frag)); 258 pf_free_fragment(frag); 259 KASSERT((TAILQ_EMPTY(&V_pf_cachequeue) || 260 TAILQ_LAST(&V_pf_cachequeue, pf_cachequeue) != frag), 261 ("!(TAILQ_EMPTY() || TAILQ_LAST() == farg): %s", 262 __FUNCTION__)); 263 } 264 PF_FRAG_UNLOCK(); 265 } 266 267 /* 268 * Try to flush old fragments to make space for new ones 269 */ 270 static void 271 pf_flush_fragments(void) 272 { 273 struct pf_fragment *frag, *cache; 274 int goal; 275 276 PF_FRAG_ASSERT(); 277 278 goal = uma_zone_get_cur(V_pf_frent_z) * 9 / 10; 279 DPFPRINTF(("trying to free %d frag entriess\n", goal)); 280 while (goal < uma_zone_get_cur(V_pf_frent_z)) { 281 frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue); 282 if (frag) 283 pf_free_fragment(frag); 284 cache = TAILQ_LAST(&V_pf_cachequeue, pf_cachequeue); 285 if (cache) 286 pf_free_fragment(cache); 287 if (frag == NULL && cache == NULL) 288 break; 289 } 290 } 291 292 /* Frees the fragments and all associated entries */ 293 static void 294 pf_free_fragment(struct pf_fragment *frag) 295 { 296 struct pf_frent *frent; 297 298 PF_FRAG_ASSERT(); 299 300 /* Free all fragments */ 301 if (BUFFER_FRAGMENTS(frag)) { 302 for (frent = TAILQ_FIRST(&frag->fr_queue); frent; 303 frent = TAILQ_FIRST(&frag->fr_queue)) { 304 TAILQ_REMOVE(&frag->fr_queue, frent, fr_next); 305 306 m_freem(frent->fe_m); 307 uma_zfree(V_pf_frent_z, frent); 308 } 309 } else { 310 for (frent = TAILQ_FIRST(&frag->fr_queue); frent; 311 frent = TAILQ_FIRST(&frag->fr_queue)) { 312 TAILQ_REMOVE(&frag->fr_queue, frent, fr_next); 313 314 KASSERT((TAILQ_EMPTY(&frag->fr_queue) || 315 TAILQ_FIRST(&frag->fr_queue)->fe_off > 316 frent->fe_len), 317 ("! (TAILQ_EMPTY() || TAILQ_FIRST()->fe_off >" 318 " frent->fe_len): %s", __func__)); 319 320 uma_zfree(V_pf_frent_z, frent); 321 } 322 } 323 324 pf_remove_fragment(frag); 325 } 326 327 static struct pf_fragment * 328 pf_find_fragment(struct pf_fragment_cmp *key, struct pf_frag_tree *tree) 329 { 330 struct pf_fragment *frag; 331 332 PF_FRAG_ASSERT(); 333 334 frag = RB_FIND(pf_frag_tree, tree, (struct pf_fragment *)key); 335 if (frag != NULL) { 336 /* XXX Are we sure we want to update the timeout? */ 337 frag->fr_timeout = time_uptime; 338 if (BUFFER_FRAGMENTS(frag)) { 339 TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next); 340 TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next); 341 } else { 342 TAILQ_REMOVE(&V_pf_cachequeue, frag, frag_next); 343 TAILQ_INSERT_HEAD(&V_pf_cachequeue, frag, frag_next); 344 } 345 } 346 347 return (frag); 348 } 349 350 /* Removes a fragment from the fragment queue and frees the fragment */ 351 static void 352 pf_remove_fragment(struct pf_fragment *frag) 353 { 354 355 PF_FRAG_ASSERT(); 356 357 if (BUFFER_FRAGMENTS(frag)) { 358 RB_REMOVE(pf_frag_tree, &V_pf_frag_tree, frag); 359 TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next); 360 uma_zfree(V_pf_frag_z, frag); 361 } else { 362 RB_REMOVE(pf_frag_tree, &V_pf_cache_tree, frag); 363 TAILQ_REMOVE(&V_pf_cachequeue, frag, frag_next); 364 uma_zfree(V_pf_frag_z, frag); 365 } 366 } 367 368 static struct pf_frent * 369 pf_create_fragment(u_short *reason) 370 { 371 struct pf_frent *frent; 372 373 PF_FRAG_ASSERT(); 374 375 frent = uma_zalloc(V_pf_frent_z, M_NOWAIT); 376 if (frent == NULL) { 377 pf_flush_fragments(); 378 frent = uma_zalloc(V_pf_frent_z, M_NOWAIT); 379 if (frent == NULL) { 380 REASON_SET(reason, PFRES_MEMORY); 381 return (NULL); 382 } 383 } 384 385 return (frent); 386 } 387 388 static struct pf_fragment * 389 pf_fillup_fragment(struct pf_fragment_cmp *key, struct pf_frent *frent, 390 u_short *reason) 391 { 392 struct pf_frent *after, *next, *prev; 393 struct pf_fragment *frag; 394 uint16_t total; 395 396 PF_FRAG_ASSERT(); 397 398 /* No empty fragments. */ 399 if (frent->fe_len == 0) { 400 DPFPRINTF(("bad fragment: len 0")); 401 goto bad_fragment; 402 } 403 404 /* All fragments are 8 byte aligned. */ 405 if (frent->fe_mff && (frent->fe_len & 0x7)) { 406 DPFPRINTF(("bad fragment: mff and len %d", frent->fe_len)); 407 goto bad_fragment; 408 } 409 410 /* Respect maximum length, IP_MAXPACKET == IPV6_MAXPACKET. */ 411 if (frent->fe_off + frent->fe_len > IP_MAXPACKET) { 412 DPFPRINTF(("bad fragment: max packet %d", 413 frent->fe_off + frent->fe_len)); 414 goto bad_fragment; 415 } 416 417 DPFPRINTF((key->frc_af == AF_INET ? 418 "reass frag %d @ %d-%d" : "reass frag %#08x @ %d-%d", 419 key->frc_id, frent->fe_off, frent->fe_off + frent->fe_len)); 420 421 /* Fully buffer all of the fragments in this fragment queue. */ 422 frag = pf_find_fragment(key, &V_pf_frag_tree); 423 424 /* Create a new reassembly queue for this packet. */ 425 if (frag == NULL) { 426 frag = uma_zalloc(V_pf_frag_z, M_NOWAIT); 427 if (frag == NULL) { 428 pf_flush_fragments(); 429 frag = uma_zalloc(V_pf_frag_z, M_NOWAIT); 430 if (frag == NULL) { 431 REASON_SET(reason, PFRES_MEMORY); 432 goto drop_fragment; 433 } 434 } 435 436 *(struct pf_fragment_cmp *)frag = *key; 437 frag->fr_timeout = time_second; 438 frag->fr_maxlen = frent->fe_len; 439 TAILQ_INIT(&frag->fr_queue); 440 441 RB_INSERT(pf_frag_tree, &V_pf_frag_tree, frag); 442 TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next); 443 444 /* We do not have a previous fragment. */ 445 TAILQ_INSERT_HEAD(&frag->fr_queue, frent, fr_next); 446 447 return (frag); 448 } 449 450 KASSERT(!TAILQ_EMPTY(&frag->fr_queue), ("!TAILQ_EMPTY()->fr_queue")); 451 452 /* Remember maximum fragment len for refragmentation. */ 453 if (frent->fe_len > frag->fr_maxlen) 454 frag->fr_maxlen = frent->fe_len; 455 456 /* Maximum data we have seen already. */ 457 total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off + 458 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len; 459 460 /* Non terminal fragments must have more fragments flag. */ 461 if (frent->fe_off + frent->fe_len < total && !frent->fe_mff) 462 goto bad_fragment; 463 464 /* Check if we saw the last fragment already. */ 465 if (!TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_mff) { 466 if (frent->fe_off + frent->fe_len > total || 467 (frent->fe_off + frent->fe_len == total && frent->fe_mff)) 468 goto bad_fragment; 469 } else { 470 if (frent->fe_off + frent->fe_len == total && !frent->fe_mff) 471 goto bad_fragment; 472 } 473 474 /* Find a fragment after the current one. */ 475 prev = NULL; 476 TAILQ_FOREACH(after, &frag->fr_queue, fr_next) { 477 if (after->fe_off > frent->fe_off) 478 break; 479 prev = after; 480 } 481 482 KASSERT(prev != NULL || after != NULL, 483 ("prev != NULL || after != NULL")); 484 485 if (prev != NULL && prev->fe_off + prev->fe_len > frent->fe_off) { 486 uint16_t precut; 487 488 precut = prev->fe_off + prev->fe_len - frent->fe_off; 489 if (precut >= frent->fe_len) 490 goto bad_fragment; 491 DPFPRINTF(("overlap -%d", precut)); 492 m_adj(frent->fe_m, precut); 493 frent->fe_off += precut; 494 frent->fe_len -= precut; 495 } 496 497 for (; after != NULL && frent->fe_off + frent->fe_len > after->fe_off; 498 after = next) { 499 uint16_t aftercut; 500 501 aftercut = frent->fe_off + frent->fe_len - after->fe_off; 502 DPFPRINTF(("adjust overlap %d", aftercut)); 503 if (aftercut < after->fe_len) { 504 m_adj(after->fe_m, aftercut); 505 after->fe_off += aftercut; 506 after->fe_len -= aftercut; 507 break; 508 } 509 510 /* This fragment is completely overlapped, lose it. */ 511 next = TAILQ_NEXT(after, fr_next); 512 m_freem(after->fe_m); 513 TAILQ_REMOVE(&frag->fr_queue, after, fr_next); 514 uma_zfree(V_pf_frent_z, after); 515 } 516 517 if (prev == NULL) 518 TAILQ_INSERT_HEAD(&frag->fr_queue, frent, fr_next); 519 else 520 TAILQ_INSERT_AFTER(&frag->fr_queue, prev, frent, fr_next); 521 522 return (frag); 523 524 bad_fragment: 525 REASON_SET(reason, PFRES_FRAG); 526 drop_fragment: 527 uma_zfree(V_pf_frent_z, frent); 528 return (NULL); 529 } 530 531 static int 532 pf_isfull_fragment(struct pf_fragment *frag) 533 { 534 struct pf_frent *frent, *next; 535 uint16_t off, total; 536 537 /* Check if we are completely reassembled */ 538 if (TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_mff) 539 return (0); 540 541 /* Maximum data we have seen already */ 542 total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off + 543 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len; 544 545 /* Check if we have all the data */ 546 off = 0; 547 for (frent = TAILQ_FIRST(&frag->fr_queue); frent; frent = next) { 548 next = TAILQ_NEXT(frent, fr_next); 549 550 off += frent->fe_len; 551 if (off < total && (next == NULL || next->fe_off != off)) { 552 DPFPRINTF(("missing fragment at %d, next %d, total %d", 553 off, next == NULL ? -1 : next->fe_off, total)); 554 return (0); 555 } 556 } 557 DPFPRINTF(("%d < %d?", off, total)); 558 if (off < total) 559 return (0); 560 KASSERT(off == total, ("off == total")); 561 562 return (1); 563 } 564 565 static struct mbuf * 566 pf_join_fragment(struct pf_fragment *frag) 567 { 568 struct mbuf *m, *m2; 569 struct pf_frent *frent, *next; 570 571 frent = TAILQ_FIRST(&frag->fr_queue); 572 next = TAILQ_NEXT(frent, fr_next); 573 574 m = frent->fe_m; 575 m_adj(m, (frent->fe_hdrlen + frent->fe_len) - m->m_pkthdr.len); 576 uma_zfree(V_pf_frent_z, frent); 577 for (frent = next; frent != NULL; frent = next) { 578 next = TAILQ_NEXT(frent, fr_next); 579 580 m2 = frent->fe_m; 581 /* Strip off ip header. */ 582 m_adj(m2, frent->fe_hdrlen); 583 /* Strip off any trailing bytes. */ 584 m_adj(m2, frent->fe_len - m2->m_pkthdr.len); 585 586 uma_zfree(V_pf_frent_z, frent); 587 m_cat(m, m2); 588 } 589 590 /* Remove from fragment queue. */ 591 pf_remove_fragment(frag); 592 593 return (m); 594 } 595 596 #ifdef INET 597 static int 598 pf_reassemble(struct mbuf **m0, struct ip *ip, int dir, u_short *reason) 599 { 600 struct mbuf *m = *m0; 601 struct pf_frent *frent; 602 struct pf_fragment *frag; 603 struct pf_fragment_cmp key; 604 uint16_t total, hdrlen; 605 606 /* Get an entry for the fragment queue */ 607 if ((frent = pf_create_fragment(reason)) == NULL) 608 return (PF_DROP); 609 610 frent->fe_m = m; 611 frent->fe_hdrlen = ip->ip_hl << 2; 612 frent->fe_extoff = 0; 613 frent->fe_len = ntohs(ip->ip_len) - (ip->ip_hl << 2); 614 frent->fe_off = (ntohs(ip->ip_off) & IP_OFFMASK) << 3; 615 frent->fe_mff = ntohs(ip->ip_off) & IP_MF; 616 617 pf_ip2key(ip, dir, &key); 618 619 if ((frag = pf_fillup_fragment(&key, frent, reason)) == NULL) 620 return (PF_DROP); 621 622 /* The mbuf is part of the fragment entry, no direct free or access */ 623 m = *m0 = NULL; 624 625 if (!pf_isfull_fragment(frag)) 626 return (PF_PASS); /* drop because *m0 is NULL, no error */ 627 628 /* We have all the data */ 629 frent = TAILQ_FIRST(&frag->fr_queue); 630 KASSERT(frent != NULL, ("frent != NULL")); 631 total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off + 632 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len; 633 hdrlen = frent->fe_hdrlen; 634 635 m = *m0 = pf_join_fragment(frag); 636 frag = NULL; 637 638 if (m->m_flags & M_PKTHDR) { 639 int plen = 0; 640 for (m = *m0; m; m = m->m_next) 641 plen += m->m_len; 642 m = *m0; 643 m->m_pkthdr.len = plen; 644 } 645 646 ip = mtod(m, struct ip *); 647 ip->ip_len = htons(hdrlen + total); 648 ip->ip_off &= ~(IP_MF|IP_OFFMASK); 649 650 if (hdrlen + total > IP_MAXPACKET) { 651 DPFPRINTF(("drop: too big: %d", total)); 652 ip->ip_len = 0; 653 REASON_SET(reason, PFRES_SHORT); 654 /* PF_DROP requires a valid mbuf *m0 in pf_test() */ 655 return (PF_DROP); 656 } 657 658 DPFPRINTF(("complete: %p(%d)\n", m, ntohs(ip->ip_len))); 659 return (PF_PASS); 660 } 661 #endif /* INET */ 662 663 #ifdef INET6 664 static int 665 pf_reassemble6(struct mbuf **m0, struct ip6_hdr *ip6, struct ip6_frag *fraghdr, 666 uint16_t hdrlen, uint16_t extoff, int dir, u_short *reason) 667 { 668 struct mbuf *m = *m0; 669 struct pf_frent *frent; 670 struct pf_fragment *frag; 671 struct pf_fragment_cmp key; 672 struct m_tag *mtag; 673 struct pf_fragment_tag *ftag; 674 int off; 675 uint32_t frag_id; 676 uint16_t total, maxlen; 677 uint8_t proto; 678 679 PF_FRAG_LOCK(); 680 681 /* Get an entry for the fragment queue. */ 682 if ((frent = pf_create_fragment(reason)) == NULL) { 683 PF_FRAG_UNLOCK(); 684 return (PF_DROP); 685 } 686 687 frent->fe_m = m; 688 frent->fe_hdrlen = hdrlen; 689 frent->fe_extoff = extoff; 690 frent->fe_len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - hdrlen; 691 frent->fe_off = ntohs(fraghdr->ip6f_offlg & IP6F_OFF_MASK); 692 frent->fe_mff = fraghdr->ip6f_offlg & IP6F_MORE_FRAG; 693 694 key.frc_src.v6 = ip6->ip6_src; 695 key.frc_dst.v6 = ip6->ip6_dst; 696 key.frc_af = AF_INET6; 697 /* Only the first fragment's protocol is relevant. */ 698 key.frc_proto = 0; 699 key.frc_id = fraghdr->ip6f_ident; 700 key.frc_direction = dir; 701 702 if ((frag = pf_fillup_fragment(&key, frent, reason)) == NULL) { 703 PF_FRAG_UNLOCK(); 704 return (PF_DROP); 705 } 706 707 /* The mbuf is part of the fragment entry, no direct free or access. */ 708 m = *m0 = NULL; 709 710 if (!pf_isfull_fragment(frag)) { 711 PF_FRAG_UNLOCK(); 712 return (PF_PASS); /* Drop because *m0 is NULL, no error. */ 713 } 714 715 /* We have all the data. */ 716 extoff = frent->fe_extoff; 717 maxlen = frag->fr_maxlen; 718 frag_id = frag->fr_id; 719 frent = TAILQ_FIRST(&frag->fr_queue); 720 KASSERT(frent != NULL, ("frent != NULL")); 721 total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off + 722 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len; 723 hdrlen = frent->fe_hdrlen - sizeof(struct ip6_frag); 724 725 m = *m0 = pf_join_fragment(frag); 726 frag = NULL; 727 728 PF_FRAG_UNLOCK(); 729 730 /* Take protocol from first fragment header. */ 731 m = m_getptr(m, hdrlen + offsetof(struct ip6_frag, ip6f_nxt), &off); 732 KASSERT(m, ("%s: short mbuf chain", __func__)); 733 proto = *(mtod(m, caddr_t) + off); 734 m = *m0; 735 736 /* Delete frag6 header */ 737 if (ip6_deletefraghdr(m, hdrlen, M_NOWAIT) != 0) 738 goto fail; 739 740 if (m->m_flags & M_PKTHDR) { 741 int plen = 0; 742 for (m = *m0; m; m = m->m_next) 743 plen += m->m_len; 744 m = *m0; 745 m->m_pkthdr.len = plen; 746 } 747 748 if ((mtag = m_tag_get(PF_REASSEMBLED, sizeof(struct pf_fragment_tag), 749 M_NOWAIT)) == NULL) 750 goto fail; 751 ftag = (struct pf_fragment_tag *)(mtag + 1); 752 ftag->ft_hdrlen = hdrlen; 753 ftag->ft_extoff = extoff; 754 ftag->ft_maxlen = maxlen; 755 ftag->ft_id = frag_id; 756 m_tag_prepend(m, mtag); 757 758 ip6 = mtod(m, struct ip6_hdr *); 759 ip6->ip6_plen = htons(hdrlen - sizeof(struct ip6_hdr) + total); 760 if (extoff) { 761 /* Write protocol into next field of last extension header. */ 762 m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt), 763 &off); 764 KASSERT(m, ("%s: short mbuf chain", __func__)); 765 *(mtod(m, char *) + off) = proto; 766 m = *m0; 767 } else 768 ip6->ip6_nxt = proto; 769 770 if (hdrlen - sizeof(struct ip6_hdr) + total > IPV6_MAXPACKET) { 771 DPFPRINTF(("drop: too big: %d", total)); 772 ip6->ip6_plen = 0; 773 REASON_SET(reason, PFRES_SHORT); 774 /* PF_DROP requires a valid mbuf *m0 in pf_test6(). */ 775 return (PF_DROP); 776 } 777 778 DPFPRINTF(("complete: %p(%d)", m, ntohs(ip6->ip6_plen))); 779 return (PF_PASS); 780 781 fail: 782 REASON_SET(reason, PFRES_MEMORY); 783 /* PF_DROP requires a valid mbuf *m0 in pf_test6(), will free later. */ 784 return (PF_DROP); 785 } 786 #endif /* INET6 */ 787 788 #ifdef INET 789 static struct mbuf * 790 pf_fragcache(struct mbuf **m0, struct ip *h, struct pf_fragment **frag, int mff, 791 int drop, int *nomem) 792 { 793 struct mbuf *m = *m0; 794 struct pf_frent *frp, *fra, *cur = NULL; 795 int ip_len = ntohs(h->ip_len) - (h->ip_hl << 2); 796 u_int16_t off = ntohs(h->ip_off) << 3; 797 u_int16_t max = ip_len + off; 798 int hosed = 0; 799 800 PF_FRAG_ASSERT(); 801 KASSERT((*frag == NULL || !BUFFER_FRAGMENTS(*frag)), 802 ("!(*frag == NULL || !BUFFER_FRAGMENTS(*frag)): %s", __FUNCTION__)); 803 804 /* Create a new range queue for this packet */ 805 if (*frag == NULL) { 806 *frag = uma_zalloc(V_pf_frag_z, M_NOWAIT); 807 if (*frag == NULL) { 808 pf_flush_fragments(); 809 *frag = uma_zalloc(V_pf_frag_z, M_NOWAIT); 810 if (*frag == NULL) 811 goto no_mem; 812 } 813 814 /* Get an entry for the queue */ 815 cur = uma_zalloc(V_pf_frent_z, M_NOWAIT); 816 if (cur == NULL) { 817 uma_zfree(V_pf_frag_z, *frag); 818 *frag = NULL; 819 goto no_mem; 820 } 821 822 (*frag)->fr_flags = PFFRAG_NOBUFFER; 823 (*frag)->fr_max = 0; 824 (*frag)->fr_src.v4 = h->ip_src; 825 (*frag)->fr_dst.v4 = h->ip_dst; 826 (*frag)->fr_id = h->ip_id; 827 (*frag)->fr_timeout = time_uptime; 828 829 cur->fe_off = off; 830 cur->fe_len = max; /* TODO: fe_len = max - off ? */ 831 TAILQ_INIT(&(*frag)->fr_queue); 832 TAILQ_INSERT_HEAD(&(*frag)->fr_queue, cur, fr_next); 833 834 RB_INSERT(pf_frag_tree, &V_pf_cache_tree, *frag); 835 TAILQ_INSERT_HEAD(&V_pf_cachequeue, *frag, frag_next); 836 837 DPFPRINTF(("fragcache[%d]: new %d-%d\n", h->ip_id, off, max)); 838 839 goto pass; 840 } 841 842 /* 843 * Find a fragment after the current one: 844 * - off contains the real shifted offset. 845 */ 846 frp = NULL; 847 TAILQ_FOREACH(fra, &(*frag)->fr_queue, fr_next) { 848 if (fra->fe_off > off) 849 break; 850 frp = fra; 851 } 852 853 KASSERT((frp != NULL || fra != NULL), 854 ("!(frp != NULL || fra != NULL): %s", __FUNCTION__)); 855 856 if (frp != NULL) { 857 int precut; 858 859 precut = frp->fe_len - off; 860 if (precut >= ip_len) { 861 /* Fragment is entirely a duplicate */ 862 DPFPRINTF(("fragcache[%d]: dead (%d-%d) %d-%d\n", 863 h->ip_id, frp->fe_off, frp->fe_len, off, max)); 864 goto drop_fragment; 865 } 866 if (precut == 0) { 867 /* They are adjacent. Fixup cache entry */ 868 DPFPRINTF(("fragcache[%d]: adjacent (%d-%d) %d-%d\n", 869 h->ip_id, frp->fe_off, frp->fe_len, off, max)); 870 frp->fe_len = max; 871 } else if (precut > 0) { 872 /* The first part of this payload overlaps with a 873 * fragment that has already been passed. 874 * Need to trim off the first part of the payload. 875 * But to do so easily, we need to create another 876 * mbuf to throw the original header into. 877 */ 878 879 DPFPRINTF(("fragcache[%d]: chop %d (%d-%d) %d-%d\n", 880 h->ip_id, precut, frp->fe_off, frp->fe_len, off, 881 max)); 882 883 off += precut; 884 max -= precut; 885 /* Update the previous frag to encompass this one */ 886 frp->fe_len = max; 887 888 if (!drop) { 889 /* XXX Optimization opportunity 890 * This is a very heavy way to trim the payload. 891 * we could do it much faster by diddling mbuf 892 * internals but that would be even less legible 893 * than this mbuf magic. For my next trick, 894 * I'll pull a rabbit out of my laptop. 895 */ 896 *m0 = m_dup(m, M_NOWAIT); 897 if (*m0 == NULL) 898 goto no_mem; 899 /* From KAME Project : We have missed this! */ 900 m_adj(*m0, (h->ip_hl << 2) - 901 (*m0)->m_pkthdr.len); 902 903 KASSERT(((*m0)->m_next == NULL), 904 ("(*m0)->m_next != NULL: %s", 905 __FUNCTION__)); 906 m_adj(m, precut + (h->ip_hl << 2)); 907 m_cat(*m0, m); 908 m = *m0; 909 if (m->m_flags & M_PKTHDR) { 910 int plen = 0; 911 struct mbuf *t; 912 for (t = m; t; t = t->m_next) 913 plen += t->m_len; 914 m->m_pkthdr.len = plen; 915 } 916 917 918 h = mtod(m, struct ip *); 919 920 KASSERT(((int)m->m_len == 921 ntohs(h->ip_len) - precut), 922 ("m->m_len != ntohs(h->ip_len) - precut: %s", 923 __FUNCTION__)); 924 h->ip_off = htons(ntohs(h->ip_off) + 925 (precut >> 3)); 926 h->ip_len = htons(ntohs(h->ip_len) - precut); 927 } else { 928 hosed++; 929 } 930 } else { 931 /* There is a gap between fragments */ 932 933 DPFPRINTF(("fragcache[%d]: gap %d (%d-%d) %d-%d\n", 934 h->ip_id, -precut, frp->fe_off, frp->fe_len, off, 935 max)); 936 937 cur = uma_zalloc(V_pf_frent_z, M_NOWAIT); 938 if (cur == NULL) 939 goto no_mem; 940 941 cur->fe_off = off; 942 cur->fe_len = max; 943 TAILQ_INSERT_AFTER(&(*frag)->fr_queue, frp, cur, fr_next); 944 } 945 } 946 947 if (fra != NULL) { 948 int aftercut; 949 int merge = 0; 950 951 aftercut = max - fra->fe_off; 952 if (aftercut == 0) { 953 /* Adjacent fragments */ 954 DPFPRINTF(("fragcache[%d]: adjacent %d-%d (%d-%d)\n", 955 h->ip_id, off, max, fra->fe_off, fra->fe_len)); 956 fra->fe_off = off; 957 merge = 1; 958 } else if (aftercut > 0) { 959 /* Need to chop off the tail of this fragment */ 960 DPFPRINTF(("fragcache[%d]: chop %d %d-%d (%d-%d)\n", 961 h->ip_id, aftercut, off, max, fra->fe_off, 962 fra->fe_len)); 963 fra->fe_off = off; 964 max -= aftercut; 965 966 merge = 1; 967 968 if (!drop) { 969 m_adj(m, -aftercut); 970 if (m->m_flags & M_PKTHDR) { 971 int plen = 0; 972 struct mbuf *t; 973 for (t = m; t; t = t->m_next) 974 plen += t->m_len; 975 m->m_pkthdr.len = plen; 976 } 977 h = mtod(m, struct ip *); 978 KASSERT(((int)m->m_len == ntohs(h->ip_len) - aftercut), 979 ("m->m_len != ntohs(h->ip_len) - aftercut: %s", 980 __FUNCTION__)); 981 h->ip_len = htons(ntohs(h->ip_len) - aftercut); 982 } else { 983 hosed++; 984 } 985 } else if (frp == NULL) { 986 /* There is a gap between fragments */ 987 DPFPRINTF(("fragcache[%d]: gap %d %d-%d (%d-%d)\n", 988 h->ip_id, -aftercut, off, max, fra->fe_off, 989 fra->fe_len)); 990 991 cur = uma_zalloc(V_pf_frent_z, M_NOWAIT); 992 if (cur == NULL) 993 goto no_mem; 994 995 cur->fe_off = off; 996 cur->fe_len = max; 997 TAILQ_INSERT_HEAD(&(*frag)->fr_queue, cur, fr_next); 998 } 999 1000 1001 /* Need to glue together two separate fragment descriptors */ 1002 if (merge) { 1003 if (cur && fra->fe_off <= cur->fe_len) { 1004 /* Need to merge in a previous 'cur' */ 1005 DPFPRINTF(("fragcache[%d]: adjacent(merge " 1006 "%d-%d) %d-%d (%d-%d)\n", 1007 h->ip_id, cur->fe_off, cur->fe_len, off, 1008 max, fra->fe_off, fra->fe_len)); 1009 fra->fe_off = cur->fe_off; 1010 TAILQ_REMOVE(&(*frag)->fr_queue, cur, fr_next); 1011 uma_zfree(V_pf_frent_z, cur); 1012 cur = NULL; 1013 1014 } else if (frp && fra->fe_off <= frp->fe_len) { 1015 /* Need to merge in a modified 'frp' */ 1016 KASSERT((cur == NULL), ("cur != NULL: %s", 1017 __FUNCTION__)); 1018 DPFPRINTF(("fragcache[%d]: adjacent(merge " 1019 "%d-%d) %d-%d (%d-%d)\n", 1020 h->ip_id, frp->fe_off, frp->fe_len, off, 1021 max, fra->fe_off, fra->fe_len)); 1022 fra->fe_off = frp->fe_off; 1023 TAILQ_REMOVE(&(*frag)->fr_queue, frp, fr_next); 1024 uma_zfree(V_pf_frent_z, frp); 1025 frp = NULL; 1026 1027 } 1028 } 1029 } 1030 1031 if (hosed) { 1032 /* 1033 * We must keep tracking the overall fragment even when 1034 * we're going to drop it anyway so that we know when to 1035 * free the overall descriptor. Thus we drop the frag late. 1036 */ 1037 goto drop_fragment; 1038 } 1039 1040 1041 pass: 1042 /* Update maximum data size */ 1043 if ((*frag)->fr_max < max) 1044 (*frag)->fr_max = max; 1045 1046 /* This is the last segment */ 1047 if (!mff) 1048 (*frag)->fr_flags |= PFFRAG_SEENLAST; 1049 1050 /* Check if we are completely reassembled */ 1051 if (((*frag)->fr_flags & PFFRAG_SEENLAST) && 1052 TAILQ_FIRST(&(*frag)->fr_queue)->fe_off == 0 && 1053 TAILQ_FIRST(&(*frag)->fr_queue)->fe_len == (*frag)->fr_max) { 1054 /* Remove from fragment queue */ 1055 DPFPRINTF(("fragcache[%d]: done 0-%d\n", h->ip_id, 1056 (*frag)->fr_max)); 1057 pf_free_fragment(*frag); 1058 *frag = NULL; 1059 } 1060 1061 return (m); 1062 1063 no_mem: 1064 *nomem = 1; 1065 1066 /* Still need to pay attention to !IP_MF */ 1067 if (!mff && *frag != NULL) 1068 (*frag)->fr_flags |= PFFRAG_SEENLAST; 1069 1070 m_freem(m); 1071 return (NULL); 1072 1073 drop_fragment: 1074 1075 /* Still need to pay attention to !IP_MF */ 1076 if (!mff && *frag != NULL) 1077 (*frag)->fr_flags |= PFFRAG_SEENLAST; 1078 1079 if (drop) { 1080 /* This fragment has been deemed bad. Don't reass */ 1081 if (((*frag)->fr_flags & PFFRAG_DROP) == 0) 1082 DPFPRINTF(("fragcache[%d]: dropping overall fragment\n", 1083 h->ip_id)); 1084 (*frag)->fr_flags |= PFFRAG_DROP; 1085 } 1086 1087 m_freem(m); 1088 return (NULL); 1089 } 1090 #endif /* INET */ 1091 1092 #ifdef INET6 1093 int 1094 pf_refragment6(struct ifnet *ifp, struct mbuf **m0, struct m_tag *mtag) 1095 { 1096 struct mbuf *m = *m0, *t; 1097 struct pf_fragment_tag *ftag = (struct pf_fragment_tag *)(mtag + 1); 1098 struct pf_pdesc pd; 1099 uint32_t frag_id; 1100 uint16_t hdrlen, extoff, maxlen; 1101 uint8_t proto; 1102 int error, action; 1103 1104 hdrlen = ftag->ft_hdrlen; 1105 extoff = ftag->ft_extoff; 1106 maxlen = ftag->ft_maxlen; 1107 frag_id = ftag->ft_id; 1108 m_tag_delete(m, mtag); 1109 mtag = NULL; 1110 ftag = NULL; 1111 1112 if (extoff) { 1113 int off; 1114 1115 /* Use protocol from next field of last extension header */ 1116 m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt), 1117 &off); 1118 KASSERT((m != NULL), ("pf_refragment6: short mbuf chain")); 1119 proto = *(mtod(m, caddr_t) + off); 1120 *(mtod(m, char *) + off) = IPPROTO_FRAGMENT; 1121 m = *m0; 1122 } else { 1123 struct ip6_hdr *hdr; 1124 1125 hdr = mtod(m, struct ip6_hdr *); 1126 proto = hdr->ip6_nxt; 1127 hdr->ip6_nxt = IPPROTO_FRAGMENT; 1128 } 1129 1130 /* 1131 * Maxlen may be less than 8 if there was only a single 1132 * fragment. As it was fragmented before, add a fragment 1133 * header also for a single fragment. If total or maxlen 1134 * is less than 8, ip6_fragment() will return EMSGSIZE and 1135 * we drop the packet. 1136 */ 1137 error = ip6_fragment(ifp, m, hdrlen, proto, maxlen, frag_id); 1138 m = (*m0)->m_nextpkt; 1139 (*m0)->m_nextpkt = NULL; 1140 if (error == 0) { 1141 /* The first mbuf contains the unfragmented packet. */ 1142 m_freem(*m0); 1143 *m0 = NULL; 1144 action = PF_PASS; 1145 } else { 1146 /* Drop expects an mbuf to free. */ 1147 DPFPRINTF(("refragment error %d", error)); 1148 action = PF_DROP; 1149 } 1150 for (t = m; m; m = t) { 1151 t = m->m_nextpkt; 1152 m->m_nextpkt = NULL; 1153 m->m_flags |= M_SKIP_FIREWALL; 1154 memset(&pd, 0, sizeof(pd)); 1155 pd.pf_mtag = pf_find_mtag(m); 1156 if (error == 0) 1157 ip6_forward(m, 0); 1158 else 1159 m_freem(m); 1160 } 1161 1162 return (action); 1163 } 1164 #endif /* INET6 */ 1165 1166 #ifdef INET 1167 int 1168 pf_normalize_ip(struct mbuf **m0, int dir, struct pfi_kif *kif, u_short *reason, 1169 struct pf_pdesc *pd) 1170 { 1171 struct mbuf *m = *m0; 1172 struct pf_rule *r; 1173 struct pf_fragment *frag = NULL; 1174 struct pf_fragment_cmp key; 1175 struct ip *h = mtod(m, struct ip *); 1176 int mff = (ntohs(h->ip_off) & IP_MF); 1177 int hlen = h->ip_hl << 2; 1178 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3; 1179 u_int16_t max; 1180 int ip_len; 1181 int ip_off; 1182 int tag = -1; 1183 int verdict; 1184 1185 PF_RULES_RASSERT(); 1186 1187 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr); 1188 while (r != NULL) { 1189 r->evaluations++; 1190 if (pfi_kif_match(r->kif, kif) == r->ifnot) 1191 r = r->skip[PF_SKIP_IFP].ptr; 1192 else if (r->direction && r->direction != dir) 1193 r = r->skip[PF_SKIP_DIR].ptr; 1194 else if (r->af && r->af != AF_INET) 1195 r = r->skip[PF_SKIP_AF].ptr; 1196 else if (r->proto && r->proto != h->ip_p) 1197 r = r->skip[PF_SKIP_PROTO].ptr; 1198 else if (PF_MISMATCHAW(&r->src.addr, 1199 (struct pf_addr *)&h->ip_src.s_addr, AF_INET, 1200 r->src.neg, kif, M_GETFIB(m))) 1201 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 1202 else if (PF_MISMATCHAW(&r->dst.addr, 1203 (struct pf_addr *)&h->ip_dst.s_addr, AF_INET, 1204 r->dst.neg, NULL, M_GETFIB(m))) 1205 r = r->skip[PF_SKIP_DST_ADDR].ptr; 1206 else if (r->match_tag && !pf_match_tag(m, r, &tag, 1207 pd->pf_mtag ? pd->pf_mtag->tag : 0)) 1208 r = TAILQ_NEXT(r, entries); 1209 else 1210 break; 1211 } 1212 1213 if (r == NULL || r->action == PF_NOSCRUB) 1214 return (PF_PASS); 1215 else { 1216 r->packets[dir == PF_OUT]++; 1217 r->bytes[dir == PF_OUT] += pd->tot_len; 1218 } 1219 1220 /* Check for illegal packets */ 1221 if (hlen < (int)sizeof(struct ip)) 1222 goto drop; 1223 1224 if (hlen > ntohs(h->ip_len)) 1225 goto drop; 1226 1227 /* Clear IP_DF if the rule uses the no-df option */ 1228 if (r->rule_flag & PFRULE_NODF && h->ip_off & htons(IP_DF)) { 1229 u_int16_t ip_off = h->ip_off; 1230 1231 h->ip_off &= htons(~IP_DF); 1232 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0); 1233 } 1234 1235 /* We will need other tests here */ 1236 if (!fragoff && !mff) 1237 goto no_fragment; 1238 1239 /* We're dealing with a fragment now. Don't allow fragments 1240 * with IP_DF to enter the cache. If the flag was cleared by 1241 * no-df above, fine. Otherwise drop it. 1242 */ 1243 if (h->ip_off & htons(IP_DF)) { 1244 DPFPRINTF(("IP_DF\n")); 1245 goto bad; 1246 } 1247 1248 ip_len = ntohs(h->ip_len) - hlen; 1249 ip_off = (ntohs(h->ip_off) & IP_OFFMASK) << 3; 1250 1251 /* All fragments are 8 byte aligned */ 1252 if (mff && (ip_len & 0x7)) { 1253 DPFPRINTF(("mff and %d\n", ip_len)); 1254 goto bad; 1255 } 1256 1257 /* Respect maximum length */ 1258 if (fragoff + ip_len > IP_MAXPACKET) { 1259 DPFPRINTF(("max packet %d\n", fragoff + ip_len)); 1260 goto bad; 1261 } 1262 max = fragoff + ip_len; 1263 1264 if ((r->rule_flag & (PFRULE_FRAGCROP|PFRULE_FRAGDROP)) == 0) { 1265 1266 /* Fully buffer all of the fragments */ 1267 PF_FRAG_LOCK(); 1268 1269 pf_ip2key(h, dir, &key); 1270 frag = pf_find_fragment(&key, &V_pf_frag_tree); 1271 1272 /* Check if we saw the last fragment already */ 1273 if (frag != NULL && (frag->fr_flags & PFFRAG_SEENLAST) && 1274 max > frag->fr_max) 1275 goto bad; 1276 1277 /* Might return a completely reassembled mbuf, or NULL */ 1278 DPFPRINTF(("reass frag %d @ %d-%d\n", h->ip_id, fragoff, max)); 1279 verdict = pf_reassemble(m0, h, dir, reason); 1280 PF_FRAG_UNLOCK(); 1281 1282 if (verdict != PF_PASS) 1283 return (PF_DROP); 1284 1285 m = *m0; 1286 if (m == NULL) 1287 return (PF_DROP); 1288 1289 if (frag != NULL && (frag->fr_flags & PFFRAG_DROP)) 1290 goto drop; 1291 1292 h = mtod(m, struct ip *); 1293 } else { 1294 /* non-buffering fragment cache (drops or masks overlaps) */ 1295 int nomem = 0; 1296 1297 if (dir == PF_OUT && pd->pf_mtag && 1298 pd->pf_mtag->flags & PF_TAG_FRAGCACHE) { 1299 /* 1300 * Already passed the fragment cache in the 1301 * input direction. If we continued, it would 1302 * appear to be a dup and would be dropped. 1303 */ 1304 goto fragment_pass; 1305 } 1306 1307 PF_FRAG_LOCK(); 1308 pf_ip2key(h, dir, &key); 1309 frag = pf_find_fragment(&key, &V_pf_cache_tree); 1310 1311 /* Check if we saw the last fragment already */ 1312 if (frag != NULL && (frag->fr_flags & PFFRAG_SEENLAST) && 1313 max > frag->fr_max) { 1314 if (r->rule_flag & PFRULE_FRAGDROP) 1315 frag->fr_flags |= PFFRAG_DROP; 1316 goto bad; 1317 } 1318 1319 *m0 = m = pf_fragcache(m0, h, &frag, mff, 1320 (r->rule_flag & PFRULE_FRAGDROP) ? 1 : 0, &nomem); 1321 PF_FRAG_UNLOCK(); 1322 if (m == NULL) { 1323 if (nomem) 1324 goto no_mem; 1325 goto drop; 1326 } 1327 1328 if (dir == PF_IN) { 1329 /* Use mtag from copied and trimmed mbuf chain. */ 1330 pd->pf_mtag = pf_get_mtag(m); 1331 if (pd->pf_mtag == NULL) { 1332 m_freem(m); 1333 *m0 = NULL; 1334 goto no_mem; 1335 } 1336 pd->pf_mtag->flags |= PF_TAG_FRAGCACHE; 1337 } 1338 1339 if (frag != NULL && (frag->fr_flags & PFFRAG_DROP)) 1340 goto drop; 1341 goto fragment_pass; 1342 } 1343 1344 no_fragment: 1345 /* At this point, only IP_DF is allowed in ip_off */ 1346 if (h->ip_off & ~htons(IP_DF)) { 1347 u_int16_t ip_off = h->ip_off; 1348 1349 h->ip_off &= htons(IP_DF); 1350 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0); 1351 } 1352 1353 /* not missing a return here */ 1354 1355 fragment_pass: 1356 pf_scrub_ip(&m, r->rule_flag, r->min_ttl, r->set_tos); 1357 1358 if ((r->rule_flag & (PFRULE_FRAGCROP|PFRULE_FRAGDROP)) == 0) 1359 pd->flags |= PFDESC_IP_REAS; 1360 return (PF_PASS); 1361 1362 no_mem: 1363 REASON_SET(reason, PFRES_MEMORY); 1364 if (r != NULL && r->log) 1365 PFLOG_PACKET(kif, m, AF_INET, dir, *reason, r, NULL, NULL, pd, 1366 1); 1367 return (PF_DROP); 1368 1369 drop: 1370 REASON_SET(reason, PFRES_NORM); 1371 if (r != NULL && r->log) 1372 PFLOG_PACKET(kif, m, AF_INET, dir, *reason, r, NULL, NULL, pd, 1373 1); 1374 return (PF_DROP); 1375 1376 bad: 1377 DPFPRINTF(("dropping bad fragment\n")); 1378 1379 /* Free associated fragments */ 1380 if (frag != NULL) { 1381 pf_free_fragment(frag); 1382 PF_FRAG_UNLOCK(); 1383 } 1384 1385 REASON_SET(reason, PFRES_FRAG); 1386 if (r != NULL && r->log) 1387 PFLOG_PACKET(kif, m, AF_INET, dir, *reason, r, NULL, NULL, pd, 1388 1); 1389 1390 return (PF_DROP); 1391 } 1392 #endif 1393 1394 #ifdef INET6 1395 int 1396 pf_normalize_ip6(struct mbuf **m0, int dir, struct pfi_kif *kif, 1397 u_short *reason, struct pf_pdesc *pd) 1398 { 1399 struct mbuf *m = *m0; 1400 struct pf_rule *r; 1401 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 1402 int extoff; 1403 int off; 1404 struct ip6_ext ext; 1405 struct ip6_opt opt; 1406 struct ip6_opt_jumbo jumbo; 1407 struct ip6_frag frag; 1408 u_int32_t jumbolen = 0, plen; 1409 int optend; 1410 int ooff; 1411 u_int8_t proto; 1412 int terminal; 1413 1414 PF_RULES_RASSERT(); 1415 1416 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr); 1417 while (r != NULL) { 1418 r->evaluations++; 1419 if (pfi_kif_match(r->kif, kif) == r->ifnot) 1420 r = r->skip[PF_SKIP_IFP].ptr; 1421 else if (r->direction && r->direction != dir) 1422 r = r->skip[PF_SKIP_DIR].ptr; 1423 else if (r->af && r->af != AF_INET6) 1424 r = r->skip[PF_SKIP_AF].ptr; 1425 #if 0 /* header chain! */ 1426 else if (r->proto && r->proto != h->ip6_nxt) 1427 r = r->skip[PF_SKIP_PROTO].ptr; 1428 #endif 1429 else if (PF_MISMATCHAW(&r->src.addr, 1430 (struct pf_addr *)&h->ip6_src, AF_INET6, 1431 r->src.neg, kif, M_GETFIB(m))) 1432 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 1433 else if (PF_MISMATCHAW(&r->dst.addr, 1434 (struct pf_addr *)&h->ip6_dst, AF_INET6, 1435 r->dst.neg, NULL, M_GETFIB(m))) 1436 r = r->skip[PF_SKIP_DST_ADDR].ptr; 1437 else 1438 break; 1439 } 1440 1441 if (r == NULL || r->action == PF_NOSCRUB) 1442 return (PF_PASS); 1443 else { 1444 r->packets[dir == PF_OUT]++; 1445 r->bytes[dir == PF_OUT] += pd->tot_len; 1446 } 1447 1448 /* Check for illegal packets */ 1449 if (sizeof(struct ip6_hdr) + IPV6_MAXPACKET < m->m_pkthdr.len) 1450 goto drop; 1451 1452 extoff = 0; 1453 off = sizeof(struct ip6_hdr); 1454 proto = h->ip6_nxt; 1455 terminal = 0; 1456 do { 1457 switch (proto) { 1458 case IPPROTO_FRAGMENT: 1459 goto fragment; 1460 break; 1461 case IPPROTO_AH: 1462 case IPPROTO_ROUTING: 1463 case IPPROTO_DSTOPTS: 1464 if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL, 1465 NULL, AF_INET6)) 1466 goto shortpkt; 1467 extoff = off; 1468 if (proto == IPPROTO_AH) 1469 off += (ext.ip6e_len + 2) * 4; 1470 else 1471 off += (ext.ip6e_len + 1) * 8; 1472 proto = ext.ip6e_nxt; 1473 break; 1474 case IPPROTO_HOPOPTS: 1475 if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL, 1476 NULL, AF_INET6)) 1477 goto shortpkt; 1478 extoff = off; 1479 optend = off + (ext.ip6e_len + 1) * 8; 1480 ooff = off + sizeof(ext); 1481 do { 1482 if (!pf_pull_hdr(m, ooff, &opt.ip6o_type, 1483 sizeof(opt.ip6o_type), NULL, NULL, 1484 AF_INET6)) 1485 goto shortpkt; 1486 if (opt.ip6o_type == IP6OPT_PAD1) { 1487 ooff++; 1488 continue; 1489 } 1490 if (!pf_pull_hdr(m, ooff, &opt, sizeof(opt), 1491 NULL, NULL, AF_INET6)) 1492 goto shortpkt; 1493 if (ooff + sizeof(opt) + opt.ip6o_len > optend) 1494 goto drop; 1495 switch (opt.ip6o_type) { 1496 case IP6OPT_JUMBO: 1497 if (h->ip6_plen != 0) 1498 goto drop; 1499 if (!pf_pull_hdr(m, ooff, &jumbo, 1500 sizeof(jumbo), NULL, NULL, 1501 AF_INET6)) 1502 goto shortpkt; 1503 memcpy(&jumbolen, jumbo.ip6oj_jumbo_len, 1504 sizeof(jumbolen)); 1505 jumbolen = ntohl(jumbolen); 1506 if (jumbolen <= IPV6_MAXPACKET) 1507 goto drop; 1508 if (sizeof(struct ip6_hdr) + jumbolen != 1509 m->m_pkthdr.len) 1510 goto drop; 1511 break; 1512 default: 1513 break; 1514 } 1515 ooff += sizeof(opt) + opt.ip6o_len; 1516 } while (ooff < optend); 1517 1518 off = optend; 1519 proto = ext.ip6e_nxt; 1520 break; 1521 default: 1522 terminal = 1; 1523 break; 1524 } 1525 } while (!terminal); 1526 1527 /* jumbo payload option must be present, or plen > 0 */ 1528 if (ntohs(h->ip6_plen) == 0) 1529 plen = jumbolen; 1530 else 1531 plen = ntohs(h->ip6_plen); 1532 if (plen == 0) 1533 goto drop; 1534 if (sizeof(struct ip6_hdr) + plen > m->m_pkthdr.len) 1535 goto shortpkt; 1536 1537 pf_scrub_ip6(&m, r->min_ttl); 1538 1539 return (PF_PASS); 1540 1541 fragment: 1542 /* Jumbo payload packets cannot be fragmented. */ 1543 plen = ntohs(h->ip6_plen); 1544 if (plen == 0 || jumbolen) 1545 goto drop; 1546 if (sizeof(struct ip6_hdr) + plen > m->m_pkthdr.len) 1547 goto shortpkt; 1548 1549 if (!pf_pull_hdr(m, off, &frag, sizeof(frag), NULL, NULL, AF_INET6)) 1550 goto shortpkt; 1551 1552 /* Offset now points to data portion. */ 1553 off += sizeof(frag); 1554 1555 /* Returns PF_DROP or *m0 is NULL or completely reassembled mbuf. */ 1556 if (pf_reassemble6(m0, h, &frag, off, extoff, dir, reason) != PF_PASS) 1557 return (PF_DROP); 1558 m = *m0; 1559 if (m == NULL) 1560 return (PF_DROP); 1561 1562 pd->flags |= PFDESC_IP_REAS; 1563 return (PF_PASS); 1564 1565 shortpkt: 1566 REASON_SET(reason, PFRES_SHORT); 1567 if (r != NULL && r->log) 1568 PFLOG_PACKET(kif, m, AF_INET6, dir, *reason, r, NULL, NULL, pd, 1569 1); 1570 return (PF_DROP); 1571 1572 drop: 1573 REASON_SET(reason, PFRES_NORM); 1574 if (r != NULL && r->log) 1575 PFLOG_PACKET(kif, m, AF_INET6, dir, *reason, r, NULL, NULL, pd, 1576 1); 1577 return (PF_DROP); 1578 } 1579 #endif /* INET6 */ 1580 1581 int 1582 pf_normalize_tcp(int dir, struct pfi_kif *kif, struct mbuf *m, int ipoff, 1583 int off, void *h, struct pf_pdesc *pd) 1584 { 1585 struct pf_rule *r, *rm = NULL; 1586 struct tcphdr *th = pd->hdr.tcp; 1587 int rewrite = 0; 1588 u_short reason; 1589 u_int8_t flags; 1590 sa_family_t af = pd->af; 1591 1592 PF_RULES_RASSERT(); 1593 1594 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr); 1595 while (r != NULL) { 1596 r->evaluations++; 1597 if (pfi_kif_match(r->kif, kif) == r->ifnot) 1598 r = r->skip[PF_SKIP_IFP].ptr; 1599 else if (r->direction && r->direction != dir) 1600 r = r->skip[PF_SKIP_DIR].ptr; 1601 else if (r->af && r->af != af) 1602 r = r->skip[PF_SKIP_AF].ptr; 1603 else if (r->proto && r->proto != pd->proto) 1604 r = r->skip[PF_SKIP_PROTO].ptr; 1605 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af, 1606 r->src.neg, kif, M_GETFIB(m))) 1607 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 1608 else if (r->src.port_op && !pf_match_port(r->src.port_op, 1609 r->src.port[0], r->src.port[1], th->th_sport)) 1610 r = r->skip[PF_SKIP_SRC_PORT].ptr; 1611 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af, 1612 r->dst.neg, NULL, M_GETFIB(m))) 1613 r = r->skip[PF_SKIP_DST_ADDR].ptr; 1614 else if (r->dst.port_op && !pf_match_port(r->dst.port_op, 1615 r->dst.port[0], r->dst.port[1], th->th_dport)) 1616 r = r->skip[PF_SKIP_DST_PORT].ptr; 1617 else if (r->os_fingerprint != PF_OSFP_ANY && !pf_osfp_match( 1618 pf_osfp_fingerprint(pd, m, off, th), 1619 r->os_fingerprint)) 1620 r = TAILQ_NEXT(r, entries); 1621 else { 1622 rm = r; 1623 break; 1624 } 1625 } 1626 1627 if (rm == NULL || rm->action == PF_NOSCRUB) 1628 return (PF_PASS); 1629 else { 1630 r->packets[dir == PF_OUT]++; 1631 r->bytes[dir == PF_OUT] += pd->tot_len; 1632 } 1633 1634 if (rm->rule_flag & PFRULE_REASSEMBLE_TCP) 1635 pd->flags |= PFDESC_TCP_NORM; 1636 1637 flags = th->th_flags; 1638 if (flags & TH_SYN) { 1639 /* Illegal packet */ 1640 if (flags & TH_RST) 1641 goto tcp_drop; 1642 1643 if (flags & TH_FIN) 1644 goto tcp_drop; 1645 } else { 1646 /* Illegal packet */ 1647 if (!(flags & (TH_ACK|TH_RST))) 1648 goto tcp_drop; 1649 } 1650 1651 if (!(flags & TH_ACK)) { 1652 /* These flags are only valid if ACK is set */ 1653 if ((flags & TH_FIN) || (flags & TH_PUSH) || (flags & TH_URG)) 1654 goto tcp_drop; 1655 } 1656 1657 /* Check for illegal header length */ 1658 if (th->th_off < (sizeof(struct tcphdr) >> 2)) 1659 goto tcp_drop; 1660 1661 /* If flags changed, or reserved data set, then adjust */ 1662 if (flags != th->th_flags || th->th_x2 != 0) { 1663 u_int16_t ov, nv; 1664 1665 ov = *(u_int16_t *)(&th->th_ack + 1); 1666 th->th_flags = flags; 1667 th->th_x2 = 0; 1668 nv = *(u_int16_t *)(&th->th_ack + 1); 1669 1670 th->th_sum = pf_cksum_fixup(th->th_sum, ov, nv, 0); 1671 rewrite = 1; 1672 } 1673 1674 /* Remove urgent pointer, if TH_URG is not set */ 1675 if (!(flags & TH_URG) && th->th_urp) { 1676 th->th_sum = pf_cksum_fixup(th->th_sum, th->th_urp, 0, 0); 1677 th->th_urp = 0; 1678 rewrite = 1; 1679 } 1680 1681 /* Process options */ 1682 if (r->max_mss && pf_normalize_tcpopt(r, m, th, off, pd->af)) 1683 rewrite = 1; 1684 1685 /* copy back packet headers if we sanitized */ 1686 if (rewrite) 1687 m_copyback(m, off, sizeof(*th), (caddr_t)th); 1688 1689 return (PF_PASS); 1690 1691 tcp_drop: 1692 REASON_SET(&reason, PFRES_NORM); 1693 if (rm != NULL && r->log) 1694 PFLOG_PACKET(kif, m, AF_INET, dir, reason, r, NULL, NULL, pd, 1695 1); 1696 return (PF_DROP); 1697 } 1698 1699 int 1700 pf_normalize_tcp_init(struct mbuf *m, int off, struct pf_pdesc *pd, 1701 struct tcphdr *th, struct pf_state_peer *src, struct pf_state_peer *dst) 1702 { 1703 u_int32_t tsval, tsecr; 1704 u_int8_t hdr[60]; 1705 u_int8_t *opt; 1706 1707 KASSERT((src->scrub == NULL), 1708 ("pf_normalize_tcp_init: src->scrub != NULL")); 1709 1710 src->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT); 1711 if (src->scrub == NULL) 1712 return (1); 1713 1714 switch (pd->af) { 1715 #ifdef INET 1716 case AF_INET: { 1717 struct ip *h = mtod(m, struct ip *); 1718 src->scrub->pfss_ttl = h->ip_ttl; 1719 break; 1720 } 1721 #endif /* INET */ 1722 #ifdef INET6 1723 case AF_INET6: { 1724 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 1725 src->scrub->pfss_ttl = h->ip6_hlim; 1726 break; 1727 } 1728 #endif /* INET6 */ 1729 } 1730 1731 1732 /* 1733 * All normalizations below are only begun if we see the start of 1734 * the connections. They must all set an enabled bit in pfss_flags 1735 */ 1736 if ((th->th_flags & TH_SYN) == 0) 1737 return (0); 1738 1739 1740 if (th->th_off > (sizeof(struct tcphdr) >> 2) && src->scrub && 1741 pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) { 1742 /* Diddle with TCP options */ 1743 int hlen; 1744 opt = hdr + sizeof(struct tcphdr); 1745 hlen = (th->th_off << 2) - sizeof(struct tcphdr); 1746 while (hlen >= TCPOLEN_TIMESTAMP) { 1747 switch (*opt) { 1748 case TCPOPT_EOL: /* FALLTHROUGH */ 1749 case TCPOPT_NOP: 1750 opt++; 1751 hlen--; 1752 break; 1753 case TCPOPT_TIMESTAMP: 1754 if (opt[1] >= TCPOLEN_TIMESTAMP) { 1755 src->scrub->pfss_flags |= 1756 PFSS_TIMESTAMP; 1757 src->scrub->pfss_ts_mod = 1758 htonl(arc4random()); 1759 1760 /* note PFSS_PAWS not set yet */ 1761 memcpy(&tsval, &opt[2], 1762 sizeof(u_int32_t)); 1763 memcpy(&tsecr, &opt[6], 1764 sizeof(u_int32_t)); 1765 src->scrub->pfss_tsval0 = ntohl(tsval); 1766 src->scrub->pfss_tsval = ntohl(tsval); 1767 src->scrub->pfss_tsecr = ntohl(tsecr); 1768 getmicrouptime(&src->scrub->pfss_last); 1769 } 1770 /* FALLTHROUGH */ 1771 default: 1772 hlen -= MAX(opt[1], 2); 1773 opt += MAX(opt[1], 2); 1774 break; 1775 } 1776 } 1777 } 1778 1779 return (0); 1780 } 1781 1782 void 1783 pf_normalize_tcp_cleanup(struct pf_state *state) 1784 { 1785 if (state->src.scrub) 1786 uma_zfree(V_pf_state_scrub_z, state->src.scrub); 1787 if (state->dst.scrub) 1788 uma_zfree(V_pf_state_scrub_z, state->dst.scrub); 1789 1790 /* Someday... flush the TCP segment reassembly descriptors. */ 1791 } 1792 1793 int 1794 pf_normalize_tcp_stateful(struct mbuf *m, int off, struct pf_pdesc *pd, 1795 u_short *reason, struct tcphdr *th, struct pf_state *state, 1796 struct pf_state_peer *src, struct pf_state_peer *dst, int *writeback) 1797 { 1798 struct timeval uptime; 1799 u_int32_t tsval, tsecr; 1800 u_int tsval_from_last; 1801 u_int8_t hdr[60]; 1802 u_int8_t *opt; 1803 int copyback = 0; 1804 int got_ts = 0; 1805 1806 KASSERT((src->scrub || dst->scrub), 1807 ("%s: src->scrub && dst->scrub!", __func__)); 1808 1809 /* 1810 * Enforce the minimum TTL seen for this connection. Negate a common 1811 * technique to evade an intrusion detection system and confuse 1812 * firewall state code. 1813 */ 1814 switch (pd->af) { 1815 #ifdef INET 1816 case AF_INET: { 1817 if (src->scrub) { 1818 struct ip *h = mtod(m, struct ip *); 1819 if (h->ip_ttl > src->scrub->pfss_ttl) 1820 src->scrub->pfss_ttl = h->ip_ttl; 1821 h->ip_ttl = src->scrub->pfss_ttl; 1822 } 1823 break; 1824 } 1825 #endif /* INET */ 1826 #ifdef INET6 1827 case AF_INET6: { 1828 if (src->scrub) { 1829 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 1830 if (h->ip6_hlim > src->scrub->pfss_ttl) 1831 src->scrub->pfss_ttl = h->ip6_hlim; 1832 h->ip6_hlim = src->scrub->pfss_ttl; 1833 } 1834 break; 1835 } 1836 #endif /* INET6 */ 1837 } 1838 1839 if (th->th_off > (sizeof(struct tcphdr) >> 2) && 1840 ((src->scrub && (src->scrub->pfss_flags & PFSS_TIMESTAMP)) || 1841 (dst->scrub && (dst->scrub->pfss_flags & PFSS_TIMESTAMP))) && 1842 pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) { 1843 /* Diddle with TCP options */ 1844 int hlen; 1845 opt = hdr + sizeof(struct tcphdr); 1846 hlen = (th->th_off << 2) - sizeof(struct tcphdr); 1847 while (hlen >= TCPOLEN_TIMESTAMP) { 1848 switch (*opt) { 1849 case TCPOPT_EOL: /* FALLTHROUGH */ 1850 case TCPOPT_NOP: 1851 opt++; 1852 hlen--; 1853 break; 1854 case TCPOPT_TIMESTAMP: 1855 /* Modulate the timestamps. Can be used for 1856 * NAT detection, OS uptime determination or 1857 * reboot detection. 1858 */ 1859 1860 if (got_ts) { 1861 /* Huh? Multiple timestamps!? */ 1862 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1863 DPFPRINTF(("multiple TS??")); 1864 pf_print_state(state); 1865 printf("\n"); 1866 } 1867 REASON_SET(reason, PFRES_TS); 1868 return (PF_DROP); 1869 } 1870 if (opt[1] >= TCPOLEN_TIMESTAMP) { 1871 memcpy(&tsval, &opt[2], 1872 sizeof(u_int32_t)); 1873 if (tsval && src->scrub && 1874 (src->scrub->pfss_flags & 1875 PFSS_TIMESTAMP)) { 1876 tsval = ntohl(tsval); 1877 pf_change_a(&opt[2], 1878 &th->th_sum, 1879 htonl(tsval + 1880 src->scrub->pfss_ts_mod), 1881 0); 1882 copyback = 1; 1883 } 1884 1885 /* Modulate TS reply iff valid (!0) */ 1886 memcpy(&tsecr, &opt[6], 1887 sizeof(u_int32_t)); 1888 if (tsecr && dst->scrub && 1889 (dst->scrub->pfss_flags & 1890 PFSS_TIMESTAMP)) { 1891 tsecr = ntohl(tsecr) 1892 - dst->scrub->pfss_ts_mod; 1893 pf_change_a(&opt[6], 1894 &th->th_sum, htonl(tsecr), 1895 0); 1896 copyback = 1; 1897 } 1898 got_ts = 1; 1899 } 1900 /* FALLTHROUGH */ 1901 default: 1902 hlen -= MAX(opt[1], 2); 1903 opt += MAX(opt[1], 2); 1904 break; 1905 } 1906 } 1907 if (copyback) { 1908 /* Copyback the options, caller copys back header */ 1909 *writeback = 1; 1910 m_copyback(m, off + sizeof(struct tcphdr), 1911 (th->th_off << 2) - sizeof(struct tcphdr), hdr + 1912 sizeof(struct tcphdr)); 1913 } 1914 } 1915 1916 1917 /* 1918 * Must invalidate PAWS checks on connections idle for too long. 1919 * The fastest allowed timestamp clock is 1ms. That turns out to 1920 * be about 24 days before it wraps. XXX Right now our lowerbound 1921 * TS echo check only works for the first 12 days of a connection 1922 * when the TS has exhausted half its 32bit space 1923 */ 1924 #define TS_MAX_IDLE (24*24*60*60) 1925 #define TS_MAX_CONN (12*24*60*60) /* XXX remove when better tsecr check */ 1926 1927 getmicrouptime(&uptime); 1928 if (src->scrub && (src->scrub->pfss_flags & PFSS_PAWS) && 1929 (uptime.tv_sec - src->scrub->pfss_last.tv_sec > TS_MAX_IDLE || 1930 time_uptime - state->creation > TS_MAX_CONN)) { 1931 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1932 DPFPRINTF(("src idled out of PAWS\n")); 1933 pf_print_state(state); 1934 printf("\n"); 1935 } 1936 src->scrub->pfss_flags = (src->scrub->pfss_flags & ~PFSS_PAWS) 1937 | PFSS_PAWS_IDLED; 1938 } 1939 if (dst->scrub && (dst->scrub->pfss_flags & PFSS_PAWS) && 1940 uptime.tv_sec - dst->scrub->pfss_last.tv_sec > TS_MAX_IDLE) { 1941 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1942 DPFPRINTF(("dst idled out of PAWS\n")); 1943 pf_print_state(state); 1944 printf("\n"); 1945 } 1946 dst->scrub->pfss_flags = (dst->scrub->pfss_flags & ~PFSS_PAWS) 1947 | PFSS_PAWS_IDLED; 1948 } 1949 1950 if (got_ts && src->scrub && dst->scrub && 1951 (src->scrub->pfss_flags & PFSS_PAWS) && 1952 (dst->scrub->pfss_flags & PFSS_PAWS)) { 1953 /* Validate that the timestamps are "in-window". 1954 * RFC1323 describes TCP Timestamp options that allow 1955 * measurement of RTT (round trip time) and PAWS 1956 * (protection against wrapped sequence numbers). PAWS 1957 * gives us a set of rules for rejecting packets on 1958 * long fat pipes (packets that were somehow delayed 1959 * in transit longer than the time it took to send the 1960 * full TCP sequence space of 4Gb). We can use these 1961 * rules and infer a few others that will let us treat 1962 * the 32bit timestamp and the 32bit echoed timestamp 1963 * as sequence numbers to prevent a blind attacker from 1964 * inserting packets into a connection. 1965 * 1966 * RFC1323 tells us: 1967 * - The timestamp on this packet must be greater than 1968 * or equal to the last value echoed by the other 1969 * endpoint. The RFC says those will be discarded 1970 * since it is a dup that has already been acked. 1971 * This gives us a lowerbound on the timestamp. 1972 * timestamp >= other last echoed timestamp 1973 * - The timestamp will be less than or equal to 1974 * the last timestamp plus the time between the 1975 * last packet and now. The RFC defines the max 1976 * clock rate as 1ms. We will allow clocks to be 1977 * up to 10% fast and will allow a total difference 1978 * or 30 seconds due to a route change. And this 1979 * gives us an upperbound on the timestamp. 1980 * timestamp <= last timestamp + max ticks 1981 * We have to be careful here. Windows will send an 1982 * initial timestamp of zero and then initialize it 1983 * to a random value after the 3whs; presumably to 1984 * avoid a DoS by having to call an expensive RNG 1985 * during a SYN flood. Proof MS has at least one 1986 * good security geek. 1987 * 1988 * - The TCP timestamp option must also echo the other 1989 * endpoints timestamp. The timestamp echoed is the 1990 * one carried on the earliest unacknowledged segment 1991 * on the left edge of the sequence window. The RFC 1992 * states that the host will reject any echoed 1993 * timestamps that were larger than any ever sent. 1994 * This gives us an upperbound on the TS echo. 1995 * tescr <= largest_tsval 1996 * - The lowerbound on the TS echo is a little more 1997 * tricky to determine. The other endpoint's echoed 1998 * values will not decrease. But there may be 1999 * network conditions that re-order packets and 2000 * cause our view of them to decrease. For now the 2001 * only lowerbound we can safely determine is that 2002 * the TS echo will never be less than the original 2003 * TS. XXX There is probably a better lowerbound. 2004 * Remove TS_MAX_CONN with better lowerbound check. 2005 * tescr >= other original TS 2006 * 2007 * It is also important to note that the fastest 2008 * timestamp clock of 1ms will wrap its 32bit space in 2009 * 24 days. So we just disable TS checking after 24 2010 * days of idle time. We actually must use a 12d 2011 * connection limit until we can come up with a better 2012 * lowerbound to the TS echo check. 2013 */ 2014 struct timeval delta_ts; 2015 int ts_fudge; 2016 2017 2018 /* 2019 * PFTM_TS_DIFF is how many seconds of leeway to allow 2020 * a host's timestamp. This can happen if the previous 2021 * packet got delayed in transit for much longer than 2022 * this packet. 2023 */ 2024 if ((ts_fudge = state->rule.ptr->timeout[PFTM_TS_DIFF]) == 0) 2025 ts_fudge = V_pf_default_rule.timeout[PFTM_TS_DIFF]; 2026 2027 /* Calculate max ticks since the last timestamp */ 2028 #define TS_MAXFREQ 1100 /* RFC max TS freq of 1Khz + 10% skew */ 2029 #define TS_MICROSECS 1000000 /* microseconds per second */ 2030 delta_ts = uptime; 2031 timevalsub(&delta_ts, &src->scrub->pfss_last); 2032 tsval_from_last = (delta_ts.tv_sec + ts_fudge) * TS_MAXFREQ; 2033 tsval_from_last += delta_ts.tv_usec / (TS_MICROSECS/TS_MAXFREQ); 2034 2035 if ((src->state >= TCPS_ESTABLISHED && 2036 dst->state >= TCPS_ESTABLISHED) && 2037 (SEQ_LT(tsval, dst->scrub->pfss_tsecr) || 2038 SEQ_GT(tsval, src->scrub->pfss_tsval + tsval_from_last) || 2039 (tsecr && (SEQ_GT(tsecr, dst->scrub->pfss_tsval) || 2040 SEQ_LT(tsecr, dst->scrub->pfss_tsval0))))) { 2041 /* Bad RFC1323 implementation or an insertion attack. 2042 * 2043 * - Solaris 2.6 and 2.7 are known to send another ACK 2044 * after the FIN,FIN|ACK,ACK closing that carries 2045 * an old timestamp. 2046 */ 2047 2048 DPFPRINTF(("Timestamp failed %c%c%c%c\n", 2049 SEQ_LT(tsval, dst->scrub->pfss_tsecr) ? '0' : ' ', 2050 SEQ_GT(tsval, src->scrub->pfss_tsval + 2051 tsval_from_last) ? '1' : ' ', 2052 SEQ_GT(tsecr, dst->scrub->pfss_tsval) ? '2' : ' ', 2053 SEQ_LT(tsecr, dst->scrub->pfss_tsval0)? '3' : ' ')); 2054 DPFPRINTF((" tsval: %u tsecr: %u +ticks: %u " 2055 "idle: %jus %lums\n", 2056 tsval, tsecr, tsval_from_last, 2057 (uintmax_t)delta_ts.tv_sec, 2058 delta_ts.tv_usec / 1000)); 2059 DPFPRINTF((" src->tsval: %u tsecr: %u\n", 2060 src->scrub->pfss_tsval, src->scrub->pfss_tsecr)); 2061 DPFPRINTF((" dst->tsval: %u tsecr: %u tsval0: %u" 2062 "\n", dst->scrub->pfss_tsval, 2063 dst->scrub->pfss_tsecr, dst->scrub->pfss_tsval0)); 2064 if (V_pf_status.debug >= PF_DEBUG_MISC) { 2065 pf_print_state(state); 2066 pf_print_flags(th->th_flags); 2067 printf("\n"); 2068 } 2069 REASON_SET(reason, PFRES_TS); 2070 return (PF_DROP); 2071 } 2072 2073 /* XXX I'd really like to require tsecr but it's optional */ 2074 2075 } else if (!got_ts && (th->th_flags & TH_RST) == 0 && 2076 ((src->state == TCPS_ESTABLISHED && dst->state == TCPS_ESTABLISHED) 2077 || pd->p_len > 0 || (th->th_flags & TH_SYN)) && 2078 src->scrub && dst->scrub && 2079 (src->scrub->pfss_flags & PFSS_PAWS) && 2080 (dst->scrub->pfss_flags & PFSS_PAWS)) { 2081 /* Didn't send a timestamp. Timestamps aren't really useful 2082 * when: 2083 * - connection opening or closing (often not even sent). 2084 * but we must not let an attacker to put a FIN on a 2085 * data packet to sneak it through our ESTABLISHED check. 2086 * - on a TCP reset. RFC suggests not even looking at TS. 2087 * - on an empty ACK. The TS will not be echoed so it will 2088 * probably not help keep the RTT calculation in sync and 2089 * there isn't as much danger when the sequence numbers 2090 * got wrapped. So some stacks don't include TS on empty 2091 * ACKs :-( 2092 * 2093 * To minimize the disruption to mostly RFC1323 conformant 2094 * stacks, we will only require timestamps on data packets. 2095 * 2096 * And what do ya know, we cannot require timestamps on data 2097 * packets. There appear to be devices that do legitimate 2098 * TCP connection hijacking. There are HTTP devices that allow 2099 * a 3whs (with timestamps) and then buffer the HTTP request. 2100 * If the intermediate device has the HTTP response cache, it 2101 * will spoof the response but not bother timestamping its 2102 * packets. So we can look for the presence of a timestamp in 2103 * the first data packet and if there, require it in all future 2104 * packets. 2105 */ 2106 2107 if (pd->p_len > 0 && (src->scrub->pfss_flags & PFSS_DATA_TS)) { 2108 /* 2109 * Hey! Someone tried to sneak a packet in. Or the 2110 * stack changed its RFC1323 behavior?!?! 2111 */ 2112 if (V_pf_status.debug >= PF_DEBUG_MISC) { 2113 DPFPRINTF(("Did not receive expected RFC1323 " 2114 "timestamp\n")); 2115 pf_print_state(state); 2116 pf_print_flags(th->th_flags); 2117 printf("\n"); 2118 } 2119 REASON_SET(reason, PFRES_TS); 2120 return (PF_DROP); 2121 } 2122 } 2123 2124 2125 /* 2126 * We will note if a host sends his data packets with or without 2127 * timestamps. And require all data packets to contain a timestamp 2128 * if the first does. PAWS implicitly requires that all data packets be 2129 * timestamped. But I think there are middle-man devices that hijack 2130 * TCP streams immediately after the 3whs and don't timestamp their 2131 * packets (seen in a WWW accelerator or cache). 2132 */ 2133 if (pd->p_len > 0 && src->scrub && (src->scrub->pfss_flags & 2134 (PFSS_TIMESTAMP|PFSS_DATA_TS|PFSS_DATA_NOTS)) == PFSS_TIMESTAMP) { 2135 if (got_ts) 2136 src->scrub->pfss_flags |= PFSS_DATA_TS; 2137 else { 2138 src->scrub->pfss_flags |= PFSS_DATA_NOTS; 2139 if (V_pf_status.debug >= PF_DEBUG_MISC && dst->scrub && 2140 (dst->scrub->pfss_flags & PFSS_TIMESTAMP)) { 2141 /* Don't warn if other host rejected RFC1323 */ 2142 DPFPRINTF(("Broken RFC1323 stack did not " 2143 "timestamp data packet. Disabled PAWS " 2144 "security.\n")); 2145 pf_print_state(state); 2146 pf_print_flags(th->th_flags); 2147 printf("\n"); 2148 } 2149 } 2150 } 2151 2152 2153 /* 2154 * Update PAWS values 2155 */ 2156 if (got_ts && src->scrub && PFSS_TIMESTAMP == (src->scrub->pfss_flags & 2157 (PFSS_PAWS_IDLED|PFSS_TIMESTAMP))) { 2158 getmicrouptime(&src->scrub->pfss_last); 2159 if (SEQ_GEQ(tsval, src->scrub->pfss_tsval) || 2160 (src->scrub->pfss_flags & PFSS_PAWS) == 0) 2161 src->scrub->pfss_tsval = tsval; 2162 2163 if (tsecr) { 2164 if (SEQ_GEQ(tsecr, src->scrub->pfss_tsecr) || 2165 (src->scrub->pfss_flags & PFSS_PAWS) == 0) 2166 src->scrub->pfss_tsecr = tsecr; 2167 2168 if ((src->scrub->pfss_flags & PFSS_PAWS) == 0 && 2169 (SEQ_LT(tsval, src->scrub->pfss_tsval0) || 2170 src->scrub->pfss_tsval0 == 0)) { 2171 /* tsval0 MUST be the lowest timestamp */ 2172 src->scrub->pfss_tsval0 = tsval; 2173 } 2174 2175 /* Only fully initialized after a TS gets echoed */ 2176 if ((src->scrub->pfss_flags & PFSS_PAWS) == 0) 2177 src->scrub->pfss_flags |= PFSS_PAWS; 2178 } 2179 } 2180 2181 /* I have a dream.... TCP segment reassembly.... */ 2182 return (0); 2183 } 2184 2185 static int 2186 pf_normalize_tcpopt(struct pf_rule *r, struct mbuf *m, struct tcphdr *th, 2187 int off, sa_family_t af) 2188 { 2189 u_int16_t *mss; 2190 int thoff; 2191 int opt, cnt, optlen = 0; 2192 int rewrite = 0; 2193 u_char opts[TCP_MAXOLEN]; 2194 u_char *optp = opts; 2195 2196 thoff = th->th_off << 2; 2197 cnt = thoff - sizeof(struct tcphdr); 2198 2199 if (cnt > 0 && !pf_pull_hdr(m, off + sizeof(*th), opts, cnt, 2200 NULL, NULL, af)) 2201 return (rewrite); 2202 2203 for (; cnt > 0; cnt -= optlen, optp += optlen) { 2204 opt = optp[0]; 2205 if (opt == TCPOPT_EOL) 2206 break; 2207 if (opt == TCPOPT_NOP) 2208 optlen = 1; 2209 else { 2210 if (cnt < 2) 2211 break; 2212 optlen = optp[1]; 2213 if (optlen < 2 || optlen > cnt) 2214 break; 2215 } 2216 switch (opt) { 2217 case TCPOPT_MAXSEG: 2218 mss = (u_int16_t *)(optp + 2); 2219 if ((ntohs(*mss)) > r->max_mss) { 2220 th->th_sum = pf_cksum_fixup(th->th_sum, 2221 *mss, htons(r->max_mss), 0); 2222 *mss = htons(r->max_mss); 2223 rewrite = 1; 2224 } 2225 break; 2226 default: 2227 break; 2228 } 2229 } 2230 2231 if (rewrite) 2232 m_copyback(m, off + sizeof(*th), thoff - sizeof(*th), opts); 2233 2234 return (rewrite); 2235 } 2236 2237 #ifdef INET 2238 static void 2239 pf_scrub_ip(struct mbuf **m0, u_int32_t flags, u_int8_t min_ttl, u_int8_t tos) 2240 { 2241 struct mbuf *m = *m0; 2242 struct ip *h = mtod(m, struct ip *); 2243 2244 /* Clear IP_DF if no-df was requested */ 2245 if (flags & PFRULE_NODF && h->ip_off & htons(IP_DF)) { 2246 u_int16_t ip_off = h->ip_off; 2247 2248 h->ip_off &= htons(~IP_DF); 2249 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0); 2250 } 2251 2252 /* Enforce a minimum ttl, may cause endless packet loops */ 2253 if (min_ttl && h->ip_ttl < min_ttl) { 2254 u_int16_t ip_ttl = h->ip_ttl; 2255 2256 h->ip_ttl = min_ttl; 2257 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_ttl, h->ip_ttl, 0); 2258 } 2259 2260 /* Enforce tos */ 2261 if (flags & PFRULE_SET_TOS) { 2262 u_int16_t ov, nv; 2263 2264 ov = *(u_int16_t *)h; 2265 h->ip_tos = tos; 2266 nv = *(u_int16_t *)h; 2267 2268 h->ip_sum = pf_cksum_fixup(h->ip_sum, ov, nv, 0); 2269 } 2270 2271 /* random-id, but not for fragments */ 2272 if (flags & PFRULE_RANDOMID && !(h->ip_off & ~htons(IP_DF))) { 2273 uint16_t ip_id = h->ip_id; 2274 2275 ip_fillid(h); 2276 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_id, h->ip_id, 0); 2277 } 2278 } 2279 #endif /* INET */ 2280 2281 #ifdef INET6 2282 static void 2283 pf_scrub_ip6(struct mbuf **m0, u_int8_t min_ttl) 2284 { 2285 struct mbuf *m = *m0; 2286 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 2287 2288 /* Enforce a minimum ttl, may cause endless packet loops */ 2289 if (min_ttl && h->ip6_hlim < min_ttl) 2290 h->ip6_hlim = min_ttl; 2291 } 2292 #endif 2293