1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause 3 * 4 * Copyright 2001 Niels Provos <provos@citi.umich.edu> 5 * Copyright 2011-2018 Alexander Bluhm <bluhm@openbsd.org> 6 * All rights reserved. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 18 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 19 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 20 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 21 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 22 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 26 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 27 * 28 * $OpenBSD: pf_norm.c,v 1.114 2009/01/29 14:11:45 henning Exp $ 29 */ 30 31 #include <sys/cdefs.h> 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/socket.h> 43 44 #include <net/if.h> 45 #include <net/if_var.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/in6_var.h> 54 #include <netinet6/nd6.h> 55 #include <netinet6/ip6_var.h> 56 #include <netinet6/scope6_var.h> 57 #include <netinet/tcp.h> 58 #include <netinet/tcp_fsm.h> 59 #include <netinet/tcp_seq.h> 60 #include <netinet/sctp_constants.h> 61 #include <netinet/sctp_header.h> 62 63 #ifdef INET6 64 #include <netinet/ip6.h> 65 #endif /* INET6 */ 66 67 struct pf_frent { 68 TAILQ_ENTRY(pf_frent) fr_next; 69 struct mbuf *fe_m; 70 uint16_t fe_hdrlen; /* ipv4 header length with ip options 71 ipv6, extension, fragment header */ 72 uint16_t fe_extoff; /* last extension header offset or 0 */ 73 uint16_t fe_len; /* fragment length */ 74 uint16_t fe_off; /* fragment offset */ 75 uint16_t fe_mff; /* more fragment flag */ 76 }; 77 78 struct pf_fragment_cmp { 79 struct pf_addr frc_src; 80 struct pf_addr frc_dst; 81 uint32_t frc_id; 82 sa_family_t frc_af; 83 uint8_t frc_proto; 84 }; 85 86 struct pf_fragment { 87 struct pf_fragment_cmp fr_key; 88 #define fr_src fr_key.frc_src 89 #define fr_dst fr_key.frc_dst 90 #define fr_id fr_key.frc_id 91 #define fr_af fr_key.frc_af 92 #define fr_proto fr_key.frc_proto 93 94 /* pointers to queue element */ 95 struct pf_frent *fr_firstoff[PF_FRAG_ENTRY_POINTS]; 96 /* count entries between pointers */ 97 uint8_t fr_entries[PF_FRAG_ENTRY_POINTS]; 98 RB_ENTRY(pf_fragment) fr_entry; 99 TAILQ_ENTRY(pf_fragment) frag_next; 100 uint32_t fr_timeout; 101 uint16_t fr_maxlen; /* maximum length of single fragment */ 102 u_int16_t fr_holes; /* number of holes in the queue */ 103 TAILQ_HEAD(pf_fragq, pf_frent) fr_queue; 104 }; 105 106 VNET_DEFINE_STATIC(struct mtx, pf_frag_mtx); 107 #define V_pf_frag_mtx VNET(pf_frag_mtx) 108 #define PF_FRAG_LOCK() mtx_lock(&V_pf_frag_mtx) 109 #define PF_FRAG_UNLOCK() mtx_unlock(&V_pf_frag_mtx) 110 #define PF_FRAG_ASSERT() mtx_assert(&V_pf_frag_mtx, MA_OWNED) 111 112 VNET_DEFINE(uma_zone_t, pf_state_scrub_z); /* XXX: shared with pfsync */ 113 114 VNET_DEFINE_STATIC(uma_zone_t, pf_frent_z); 115 #define V_pf_frent_z VNET(pf_frent_z) 116 VNET_DEFINE_STATIC(uma_zone_t, pf_frag_z); 117 #define V_pf_frag_z VNET(pf_frag_z) 118 119 TAILQ_HEAD(pf_fragqueue, pf_fragment); 120 TAILQ_HEAD(pf_cachequeue, pf_fragment); 121 VNET_DEFINE_STATIC(struct pf_fragqueue, pf_fragqueue); 122 #define V_pf_fragqueue VNET(pf_fragqueue) 123 RB_HEAD(pf_frag_tree, pf_fragment); 124 VNET_DEFINE_STATIC(struct pf_frag_tree, pf_frag_tree); 125 #define V_pf_frag_tree VNET(pf_frag_tree) 126 static int pf_frag_compare(struct pf_fragment *, 127 struct pf_fragment *); 128 static RB_PROTOTYPE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare); 129 static RB_GENERATE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare); 130 131 static void pf_flush_fragments(void); 132 static void pf_free_fragment(struct pf_fragment *); 133 static void pf_remove_fragment(struct pf_fragment *); 134 135 static struct pf_frent *pf_create_fragment(u_short *); 136 static int pf_frent_holes(struct pf_frent *frent); 137 static struct pf_fragment *pf_find_fragment(struct pf_fragment_cmp *key, 138 struct pf_frag_tree *tree); 139 static inline int pf_frent_index(struct pf_frent *); 140 static int pf_frent_insert(struct pf_fragment *, 141 struct pf_frent *, struct pf_frent *); 142 void pf_frent_remove(struct pf_fragment *, 143 struct pf_frent *); 144 struct pf_frent *pf_frent_previous(struct pf_fragment *, 145 struct pf_frent *); 146 static struct pf_fragment *pf_fillup_fragment(struct pf_fragment_cmp *, 147 struct pf_frent *, u_short *); 148 static struct mbuf *pf_join_fragment(struct pf_fragment *); 149 #ifdef INET 150 static int pf_reassemble(struct mbuf **, int, u_short *); 151 #endif /* INET */ 152 #ifdef INET6 153 static int pf_reassemble6(struct mbuf **, 154 struct ip6_frag *, uint16_t, uint16_t, u_short *); 155 #endif /* INET6 */ 156 157 #define DPFPRINTF(x) do { \ 158 if (V_pf_status.debug >= PF_DEBUG_MISC) { \ 159 printf("%s: ", __func__); \ 160 printf x ; \ 161 } \ 162 } while(0) 163 164 #ifdef INET 165 static void 166 pf_ip2key(struct ip *ip, int dir, struct pf_fragment_cmp *key) 167 { 168 169 key->frc_src.v4 = ip->ip_src; 170 key->frc_dst.v4 = ip->ip_dst; 171 key->frc_af = AF_INET; 172 key->frc_proto = ip->ip_p; 173 key->frc_id = ip->ip_id; 174 } 175 #endif /* INET */ 176 177 void 178 pf_normalize_init(void) 179 { 180 181 V_pf_frag_z = uma_zcreate("pf frags", sizeof(struct pf_fragment), 182 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 183 V_pf_frent_z = uma_zcreate("pf frag entries", sizeof(struct pf_frent), 184 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 185 V_pf_state_scrub_z = uma_zcreate("pf state scrubs", 186 sizeof(struct pf_state_scrub), NULL, NULL, NULL, NULL, 187 UMA_ALIGN_PTR, 0); 188 189 mtx_init(&V_pf_frag_mtx, "pf fragments", NULL, MTX_DEF); 190 191 V_pf_limits[PF_LIMIT_FRAGS].zone = V_pf_frent_z; 192 V_pf_limits[PF_LIMIT_FRAGS].limit = PFFRAG_FRENT_HIWAT; 193 uma_zone_set_max(V_pf_frent_z, PFFRAG_FRENT_HIWAT); 194 uma_zone_set_warning(V_pf_frent_z, "PF frag entries limit reached"); 195 196 TAILQ_INIT(&V_pf_fragqueue); 197 } 198 199 void 200 pf_normalize_cleanup(void) 201 { 202 203 uma_zdestroy(V_pf_state_scrub_z); 204 uma_zdestroy(V_pf_frent_z); 205 uma_zdestroy(V_pf_frag_z); 206 207 mtx_destroy(&V_pf_frag_mtx); 208 } 209 210 static int 211 pf_frag_compare(struct pf_fragment *a, struct pf_fragment *b) 212 { 213 int diff; 214 215 if ((diff = a->fr_id - b->fr_id) != 0) 216 return (diff); 217 if ((diff = a->fr_proto - b->fr_proto) != 0) 218 return (diff); 219 if ((diff = a->fr_af - b->fr_af) != 0) 220 return (diff); 221 if ((diff = pf_addr_cmp(&a->fr_src, &b->fr_src, a->fr_af)) != 0) 222 return (diff); 223 if ((diff = pf_addr_cmp(&a->fr_dst, &b->fr_dst, a->fr_af)) != 0) 224 return (diff); 225 return (0); 226 } 227 228 void 229 pf_purge_expired_fragments(void) 230 { 231 u_int32_t expire = time_uptime - 232 V_pf_default_rule.timeout[PFTM_FRAG]; 233 234 pf_purge_fragments(expire); 235 } 236 237 void 238 pf_purge_fragments(uint32_t expire) 239 { 240 struct pf_fragment *frag; 241 242 PF_FRAG_LOCK(); 243 while ((frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue)) != NULL) { 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 PF_FRAG_UNLOCK(); 252 } 253 254 /* 255 * Try to flush old fragments to make space for new ones 256 */ 257 static void 258 pf_flush_fragments(void) 259 { 260 struct pf_fragment *frag; 261 int goal; 262 263 PF_FRAG_ASSERT(); 264 265 goal = uma_zone_get_cur(V_pf_frent_z) * 9 / 10; 266 DPFPRINTF(("trying to free %d frag entriess\n", goal)); 267 while (goal < uma_zone_get_cur(V_pf_frent_z)) { 268 frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue); 269 if (frag) 270 pf_free_fragment(frag); 271 else 272 break; 273 } 274 } 275 276 /* Frees the fragments and all associated entries */ 277 static void 278 pf_free_fragment(struct pf_fragment *frag) 279 { 280 struct pf_frent *frent; 281 282 PF_FRAG_ASSERT(); 283 284 /* Free all fragments */ 285 for (frent = TAILQ_FIRST(&frag->fr_queue); frent; 286 frent = TAILQ_FIRST(&frag->fr_queue)) { 287 TAILQ_REMOVE(&frag->fr_queue, frent, fr_next); 288 289 m_freem(frent->fe_m); 290 uma_zfree(V_pf_frent_z, frent); 291 } 292 293 pf_remove_fragment(frag); 294 } 295 296 static struct pf_fragment * 297 pf_find_fragment(struct pf_fragment_cmp *key, struct pf_frag_tree *tree) 298 { 299 struct pf_fragment *frag; 300 301 PF_FRAG_ASSERT(); 302 303 frag = RB_FIND(pf_frag_tree, tree, (struct pf_fragment *)key); 304 if (frag != NULL) { 305 /* XXX Are we sure we want to update the timeout? */ 306 frag->fr_timeout = time_uptime; 307 TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next); 308 TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next); 309 } 310 311 return (frag); 312 } 313 314 /* Removes a fragment from the fragment queue and frees the fragment */ 315 static void 316 pf_remove_fragment(struct pf_fragment *frag) 317 { 318 319 PF_FRAG_ASSERT(); 320 KASSERT(frag, ("frag != NULL")); 321 322 RB_REMOVE(pf_frag_tree, &V_pf_frag_tree, frag); 323 TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next); 324 uma_zfree(V_pf_frag_z, frag); 325 } 326 327 static struct pf_frent * 328 pf_create_fragment(u_short *reason) 329 { 330 struct pf_frent *frent; 331 332 PF_FRAG_ASSERT(); 333 334 frent = uma_zalloc(V_pf_frent_z, M_NOWAIT); 335 if (frent == NULL) { 336 pf_flush_fragments(); 337 frent = uma_zalloc(V_pf_frent_z, M_NOWAIT); 338 if (frent == NULL) { 339 REASON_SET(reason, PFRES_MEMORY); 340 return (NULL); 341 } 342 } 343 344 return (frent); 345 } 346 347 /* 348 * Calculate the additional holes that were created in the fragment 349 * queue by inserting this fragment. A fragment in the middle 350 * creates one more hole by splitting. For each connected side, 351 * it loses one hole. 352 * Fragment entry must be in the queue when calling this function. 353 */ 354 static int 355 pf_frent_holes(struct pf_frent *frent) 356 { 357 struct pf_frent *prev = TAILQ_PREV(frent, pf_fragq, fr_next); 358 struct pf_frent *next = TAILQ_NEXT(frent, fr_next); 359 int holes = 1; 360 361 if (prev == NULL) { 362 if (frent->fe_off == 0) 363 holes--; 364 } else { 365 KASSERT(frent->fe_off != 0, ("frent->fe_off != 0")); 366 if (frent->fe_off == prev->fe_off + prev->fe_len) 367 holes--; 368 } 369 if (next == NULL) { 370 if (!frent->fe_mff) 371 holes--; 372 } else { 373 KASSERT(frent->fe_mff, ("frent->fe_mff")); 374 if (next->fe_off == frent->fe_off + frent->fe_len) 375 holes--; 376 } 377 return holes; 378 } 379 380 static inline int 381 pf_frent_index(struct pf_frent *frent) 382 { 383 /* 384 * We have an array of 16 entry points to the queue. A full size 385 * 65535 octet IP packet can have 8192 fragments. So the queue 386 * traversal length is at most 512 and at most 16 entry points are 387 * checked. We need 128 additional bytes on a 64 bit architecture. 388 */ 389 CTASSERT(((u_int16_t)0xffff &~ 7) / (0x10000 / PF_FRAG_ENTRY_POINTS) == 390 16 - 1); 391 CTASSERT(((u_int16_t)0xffff >> 3) / PF_FRAG_ENTRY_POINTS == 512 - 1); 392 393 return frent->fe_off / (0x10000 / PF_FRAG_ENTRY_POINTS); 394 } 395 396 static int 397 pf_frent_insert(struct pf_fragment *frag, struct pf_frent *frent, 398 struct pf_frent *prev) 399 { 400 int index; 401 402 CTASSERT(PF_FRAG_ENTRY_LIMIT <= 0xff); 403 404 /* 405 * A packet has at most 65536 octets. With 16 entry points, each one 406 * spawns 4096 octets. We limit these to 64 fragments each, which 407 * means on average every fragment must have at least 64 octets. 408 */ 409 index = pf_frent_index(frent); 410 if (frag->fr_entries[index] >= PF_FRAG_ENTRY_LIMIT) 411 return ENOBUFS; 412 frag->fr_entries[index]++; 413 414 if (prev == NULL) { 415 TAILQ_INSERT_HEAD(&frag->fr_queue, frent, fr_next); 416 } else { 417 KASSERT(prev->fe_off + prev->fe_len <= frent->fe_off, 418 ("overlapping fragment")); 419 TAILQ_INSERT_AFTER(&frag->fr_queue, prev, frent, fr_next); 420 } 421 422 if (frag->fr_firstoff[index] == NULL) { 423 KASSERT(prev == NULL || pf_frent_index(prev) < index, 424 ("prev == NULL || pf_frent_index(pref) < index")); 425 frag->fr_firstoff[index] = frent; 426 } else { 427 if (frent->fe_off < frag->fr_firstoff[index]->fe_off) { 428 KASSERT(prev == NULL || pf_frent_index(prev) < index, 429 ("prev == NULL || pf_frent_index(pref) < index")); 430 frag->fr_firstoff[index] = frent; 431 } else { 432 KASSERT(prev != NULL, ("prev != NULL")); 433 KASSERT(pf_frent_index(prev) == index, 434 ("pf_frent_index(prev) == index")); 435 } 436 } 437 438 frag->fr_holes += pf_frent_holes(frent); 439 440 return 0; 441 } 442 443 void 444 pf_frent_remove(struct pf_fragment *frag, struct pf_frent *frent) 445 { 446 #ifdef INVARIANTS 447 struct pf_frent *prev = TAILQ_PREV(frent, pf_fragq, fr_next); 448 #endif 449 struct pf_frent *next = TAILQ_NEXT(frent, fr_next); 450 int index; 451 452 frag->fr_holes -= pf_frent_holes(frent); 453 454 index = pf_frent_index(frent); 455 KASSERT(frag->fr_firstoff[index] != NULL, ("frent not found")); 456 if (frag->fr_firstoff[index]->fe_off == frent->fe_off) { 457 if (next == NULL) { 458 frag->fr_firstoff[index] = NULL; 459 } else { 460 KASSERT(frent->fe_off + frent->fe_len <= next->fe_off, 461 ("overlapping fragment")); 462 if (pf_frent_index(next) == index) { 463 frag->fr_firstoff[index] = next; 464 } else { 465 frag->fr_firstoff[index] = NULL; 466 } 467 } 468 } else { 469 KASSERT(frag->fr_firstoff[index]->fe_off < frent->fe_off, 470 ("frag->fr_firstoff[index]->fe_off < frent->fe_off")); 471 KASSERT(prev != NULL, ("prev != NULL")); 472 KASSERT(prev->fe_off + prev->fe_len <= frent->fe_off, 473 ("overlapping fragment")); 474 KASSERT(pf_frent_index(prev) == index, 475 ("pf_frent_index(prev) == index")); 476 } 477 478 TAILQ_REMOVE(&frag->fr_queue, frent, fr_next); 479 480 KASSERT(frag->fr_entries[index] > 0, ("No fragments remaining")); 481 frag->fr_entries[index]--; 482 } 483 484 struct pf_frent * 485 pf_frent_previous(struct pf_fragment *frag, struct pf_frent *frent) 486 { 487 struct pf_frent *prev, *next; 488 int index; 489 490 /* 491 * If there are no fragments after frag, take the final one. Assume 492 * that the global queue is not empty. 493 */ 494 prev = TAILQ_LAST(&frag->fr_queue, pf_fragq); 495 KASSERT(prev != NULL, ("prev != NULL")); 496 if (prev->fe_off <= frent->fe_off) 497 return prev; 498 /* 499 * We want to find a fragment entry that is before frag, but still 500 * close to it. Find the first fragment entry that is in the same 501 * entry point or in the first entry point after that. As we have 502 * already checked that there are entries behind frag, this will 503 * succeed. 504 */ 505 for (index = pf_frent_index(frent); index < PF_FRAG_ENTRY_POINTS; 506 index++) { 507 prev = frag->fr_firstoff[index]; 508 if (prev != NULL) 509 break; 510 } 511 KASSERT(prev != NULL, ("prev != NULL")); 512 /* 513 * In prev we may have a fragment from the same entry point that is 514 * before frent, or one that is just one position behind frent. 515 * In the latter case, we go back one step and have the predecessor. 516 * There may be none if the new fragment will be the first one. 517 */ 518 if (prev->fe_off > frent->fe_off) { 519 prev = TAILQ_PREV(prev, pf_fragq, fr_next); 520 if (prev == NULL) 521 return NULL; 522 KASSERT(prev->fe_off <= frent->fe_off, 523 ("prev->fe_off <= frent->fe_off")); 524 return prev; 525 } 526 /* 527 * In prev is the first fragment of the entry point. The offset 528 * of frag is behind it. Find the closest previous fragment. 529 */ 530 for (next = TAILQ_NEXT(prev, fr_next); next != NULL; 531 next = TAILQ_NEXT(next, fr_next)) { 532 if (next->fe_off > frent->fe_off) 533 break; 534 prev = next; 535 } 536 return prev; 537 } 538 539 static struct pf_fragment * 540 pf_fillup_fragment(struct pf_fragment_cmp *key, struct pf_frent *frent, 541 u_short *reason) 542 { 543 struct pf_frent *after, *next, *prev; 544 struct pf_fragment *frag; 545 uint16_t total; 546 int old_index, new_index; 547 548 PF_FRAG_ASSERT(); 549 550 /* No empty fragments. */ 551 if (frent->fe_len == 0) { 552 DPFPRINTF(("bad fragment: len 0\n")); 553 goto bad_fragment; 554 } 555 556 /* All fragments are 8 byte aligned. */ 557 if (frent->fe_mff && (frent->fe_len & 0x7)) { 558 DPFPRINTF(("bad fragment: mff and len %d\n", frent->fe_len)); 559 goto bad_fragment; 560 } 561 562 /* Respect maximum length, IP_MAXPACKET == IPV6_MAXPACKET. */ 563 if (frent->fe_off + frent->fe_len > IP_MAXPACKET) { 564 DPFPRINTF(("bad fragment: max packet %d\n", 565 frent->fe_off + frent->fe_len)); 566 goto bad_fragment; 567 } 568 569 DPFPRINTF((key->frc_af == AF_INET ? 570 "reass frag %d @ %d-%d\n" : "reass frag %#08x @ %d-%d\n", 571 key->frc_id, frent->fe_off, frent->fe_off + frent->fe_len)); 572 573 /* Fully buffer all of the fragments in this fragment queue. */ 574 frag = pf_find_fragment(key, &V_pf_frag_tree); 575 576 /* Create a new reassembly queue for this packet. */ 577 if (frag == NULL) { 578 frag = uma_zalloc(V_pf_frag_z, M_NOWAIT); 579 if (frag == NULL) { 580 pf_flush_fragments(); 581 frag = uma_zalloc(V_pf_frag_z, M_NOWAIT); 582 if (frag == NULL) { 583 REASON_SET(reason, PFRES_MEMORY); 584 goto drop_fragment; 585 } 586 } 587 588 *(struct pf_fragment_cmp *)frag = *key; 589 memset(frag->fr_firstoff, 0, sizeof(frag->fr_firstoff)); 590 memset(frag->fr_entries, 0, sizeof(frag->fr_entries)); 591 frag->fr_timeout = time_uptime; 592 frag->fr_maxlen = frent->fe_len; 593 frag->fr_holes = 1; 594 TAILQ_INIT(&frag->fr_queue); 595 596 RB_INSERT(pf_frag_tree, &V_pf_frag_tree, frag); 597 TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next); 598 599 /* We do not have a previous fragment, cannot fail. */ 600 pf_frent_insert(frag, frent, NULL); 601 602 return (frag); 603 } 604 605 KASSERT(!TAILQ_EMPTY(&frag->fr_queue), ("!TAILQ_EMPTY()->fr_queue")); 606 607 /* Remember maximum fragment len for refragmentation. */ 608 if (frent->fe_len > frag->fr_maxlen) 609 frag->fr_maxlen = frent->fe_len; 610 611 /* Maximum data we have seen already. */ 612 total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off + 613 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len; 614 615 /* Non terminal fragments must have more fragments flag. */ 616 if (frent->fe_off + frent->fe_len < total && !frent->fe_mff) 617 goto bad_fragment; 618 619 /* Check if we saw the last fragment already. */ 620 if (!TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_mff) { 621 if (frent->fe_off + frent->fe_len > total || 622 (frent->fe_off + frent->fe_len == total && frent->fe_mff)) 623 goto bad_fragment; 624 } else { 625 if (frent->fe_off + frent->fe_len == total && !frent->fe_mff) 626 goto bad_fragment; 627 } 628 629 /* Find neighbors for newly inserted fragment */ 630 prev = pf_frent_previous(frag, frent); 631 if (prev == NULL) { 632 after = TAILQ_FIRST(&frag->fr_queue); 633 KASSERT(after != NULL, ("after != NULL")); 634 } else { 635 after = TAILQ_NEXT(prev, fr_next); 636 } 637 638 if (prev != NULL && prev->fe_off + prev->fe_len > frent->fe_off) { 639 uint16_t precut; 640 641 precut = prev->fe_off + prev->fe_len - frent->fe_off; 642 if (precut >= frent->fe_len) 643 goto bad_fragment; 644 DPFPRINTF(("overlap -%d\n", precut)); 645 m_adj(frent->fe_m, precut); 646 frent->fe_off += precut; 647 frent->fe_len -= precut; 648 } 649 650 for (; after != NULL && frent->fe_off + frent->fe_len > after->fe_off; 651 after = next) { 652 uint16_t aftercut; 653 654 aftercut = frent->fe_off + frent->fe_len - after->fe_off; 655 DPFPRINTF(("adjust overlap %d\n", aftercut)); 656 if (aftercut < after->fe_len) { 657 m_adj(after->fe_m, aftercut); 658 old_index = pf_frent_index(after); 659 after->fe_off += aftercut; 660 after->fe_len -= aftercut; 661 new_index = pf_frent_index(after); 662 if (old_index != new_index) { 663 DPFPRINTF(("frag index %d, new %d", 664 old_index, new_index)); 665 /* Fragment switched queue as fe_off changed */ 666 after->fe_off -= aftercut; 667 after->fe_len += aftercut; 668 /* Remove restored fragment from old queue */ 669 pf_frent_remove(frag, after); 670 after->fe_off += aftercut; 671 after->fe_len -= aftercut; 672 /* Insert into correct queue */ 673 if (pf_frent_insert(frag, after, prev)) { 674 DPFPRINTF( 675 ("fragment requeue limit exceeded")); 676 m_freem(after->fe_m); 677 uma_zfree(V_pf_frent_z, after); 678 /* There is not way to recover */ 679 goto bad_fragment; 680 } 681 } 682 break; 683 } 684 685 /* This fragment is completely overlapped, lose it. */ 686 next = TAILQ_NEXT(after, fr_next); 687 pf_frent_remove(frag, after); 688 m_freem(after->fe_m); 689 uma_zfree(V_pf_frent_z, after); 690 } 691 692 /* If part of the queue gets too long, there is not way to recover. */ 693 if (pf_frent_insert(frag, frent, prev)) { 694 DPFPRINTF(("fragment queue limit exceeded\n")); 695 goto bad_fragment; 696 } 697 698 return (frag); 699 700 bad_fragment: 701 REASON_SET(reason, PFRES_FRAG); 702 drop_fragment: 703 uma_zfree(V_pf_frent_z, frent); 704 return (NULL); 705 } 706 707 static struct mbuf * 708 pf_join_fragment(struct pf_fragment *frag) 709 { 710 struct mbuf *m, *m2; 711 struct pf_frent *frent, *next; 712 713 frent = TAILQ_FIRST(&frag->fr_queue); 714 next = TAILQ_NEXT(frent, fr_next); 715 716 m = frent->fe_m; 717 m_adj(m, (frent->fe_hdrlen + frent->fe_len) - m->m_pkthdr.len); 718 uma_zfree(V_pf_frent_z, frent); 719 for (frent = next; frent != NULL; frent = next) { 720 next = TAILQ_NEXT(frent, fr_next); 721 722 m2 = frent->fe_m; 723 /* Strip off ip header. */ 724 m_adj(m2, frent->fe_hdrlen); 725 /* Strip off any trailing bytes. */ 726 m_adj(m2, frent->fe_len - m2->m_pkthdr.len); 727 728 uma_zfree(V_pf_frent_z, frent); 729 m_cat(m, m2); 730 } 731 732 /* Remove from fragment queue. */ 733 pf_remove_fragment(frag); 734 735 return (m); 736 } 737 738 #ifdef INET 739 static int 740 pf_reassemble(struct mbuf **m0, int dir, u_short *reason) 741 { 742 struct mbuf *m = *m0; 743 struct ip *ip = mtod(m, struct ip *); 744 struct pf_frent *frent; 745 struct pf_fragment *frag; 746 struct m_tag *mtag; 747 struct pf_fragment_tag *ftag; 748 struct pf_fragment_cmp key; 749 uint16_t total, hdrlen; 750 uint32_t frag_id; 751 uint16_t maxlen; 752 753 /* Get an entry for the fragment queue */ 754 if ((frent = pf_create_fragment(reason)) == NULL) 755 return (PF_DROP); 756 757 frent->fe_m = m; 758 frent->fe_hdrlen = ip->ip_hl << 2; 759 frent->fe_extoff = 0; 760 frent->fe_len = ntohs(ip->ip_len) - (ip->ip_hl << 2); 761 frent->fe_off = (ntohs(ip->ip_off) & IP_OFFMASK) << 3; 762 frent->fe_mff = ntohs(ip->ip_off) & IP_MF; 763 764 pf_ip2key(ip, dir, &key); 765 766 if ((frag = pf_fillup_fragment(&key, frent, reason)) == NULL) 767 return (PF_DROP); 768 769 /* The mbuf is part of the fragment entry, no direct free or access */ 770 m = *m0 = NULL; 771 772 if (frag->fr_holes) { 773 DPFPRINTF(("frag %d, holes %d\n", frag->fr_id, frag->fr_holes)); 774 return (PF_PASS); /* drop because *m0 is NULL, no error */ 775 } 776 777 /* We have all the data */ 778 frent = TAILQ_FIRST(&frag->fr_queue); 779 KASSERT(frent != NULL, ("frent != NULL")); 780 total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off + 781 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len; 782 hdrlen = frent->fe_hdrlen; 783 784 maxlen = frag->fr_maxlen; 785 frag_id = frag->fr_id; 786 m = *m0 = pf_join_fragment(frag); 787 frag = NULL; 788 789 if (m->m_flags & M_PKTHDR) { 790 int plen = 0; 791 for (m = *m0; m; m = m->m_next) 792 plen += m->m_len; 793 m = *m0; 794 m->m_pkthdr.len = plen; 795 } 796 797 if ((mtag = m_tag_get(PACKET_TAG_PF_REASSEMBLED, 798 sizeof(struct pf_fragment_tag), M_NOWAIT)) == NULL) { 799 REASON_SET(reason, PFRES_SHORT); 800 /* PF_DROP requires a valid mbuf *m0 in pf_test() */ 801 return (PF_DROP); 802 } 803 ftag = (struct pf_fragment_tag *)(mtag + 1); 804 ftag->ft_hdrlen = hdrlen; 805 ftag->ft_extoff = 0; 806 ftag->ft_maxlen = maxlen; 807 ftag->ft_id = frag_id; 808 m_tag_prepend(m, mtag); 809 810 ip = mtod(m, struct ip *); 811 ip->ip_sum = pf_cksum_fixup(ip->ip_sum, ip->ip_len, 812 htons(hdrlen + total), 0); 813 ip->ip_len = htons(hdrlen + total); 814 ip->ip_sum = pf_cksum_fixup(ip->ip_sum, ip->ip_off, 815 ip->ip_off & ~(IP_MF|IP_OFFMASK), 0); 816 ip->ip_off &= ~(IP_MF|IP_OFFMASK); 817 818 if (hdrlen + total > IP_MAXPACKET) { 819 DPFPRINTF(("drop: too big: %d\n", total)); 820 ip->ip_len = 0; 821 REASON_SET(reason, PFRES_SHORT); 822 /* PF_DROP requires a valid mbuf *m0 in pf_test() */ 823 return (PF_DROP); 824 } 825 826 DPFPRINTF(("complete: %p(%d)\n", m, ntohs(ip->ip_len))); 827 return (PF_PASS); 828 } 829 #endif /* INET */ 830 831 #ifdef INET6 832 static int 833 pf_reassemble6(struct mbuf **m0, struct ip6_frag *fraghdr, 834 uint16_t hdrlen, uint16_t extoff, u_short *reason) 835 { 836 struct mbuf *m = *m0; 837 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 838 struct pf_frent *frent; 839 struct pf_fragment *frag; 840 struct pf_fragment_cmp key; 841 struct m_tag *mtag; 842 struct pf_fragment_tag *ftag; 843 int off; 844 uint32_t frag_id; 845 uint16_t total, maxlen; 846 uint8_t proto; 847 848 PF_FRAG_LOCK(); 849 850 /* Get an entry for the fragment queue. */ 851 if ((frent = pf_create_fragment(reason)) == NULL) { 852 PF_FRAG_UNLOCK(); 853 return (PF_DROP); 854 } 855 856 frent->fe_m = m; 857 frent->fe_hdrlen = hdrlen; 858 frent->fe_extoff = extoff; 859 frent->fe_len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - hdrlen; 860 frent->fe_off = ntohs(fraghdr->ip6f_offlg & IP6F_OFF_MASK); 861 frent->fe_mff = fraghdr->ip6f_offlg & IP6F_MORE_FRAG; 862 863 key.frc_src.v6 = ip6->ip6_src; 864 key.frc_dst.v6 = ip6->ip6_dst; 865 key.frc_af = AF_INET6; 866 /* Only the first fragment's protocol is relevant. */ 867 key.frc_proto = 0; 868 key.frc_id = fraghdr->ip6f_ident; 869 870 if ((frag = pf_fillup_fragment(&key, frent, reason)) == NULL) { 871 PF_FRAG_UNLOCK(); 872 return (PF_DROP); 873 } 874 875 /* The mbuf is part of the fragment entry, no direct free or access. */ 876 m = *m0 = NULL; 877 878 if (frag->fr_holes) { 879 DPFPRINTF(("frag %d, holes %d\n", frag->fr_id, 880 frag->fr_holes)); 881 PF_FRAG_UNLOCK(); 882 return (PF_PASS); /* Drop because *m0 is NULL, no error. */ 883 } 884 885 /* We have all the data. */ 886 frent = TAILQ_FIRST(&frag->fr_queue); 887 KASSERT(frent != NULL, ("frent != NULL")); 888 extoff = frent->fe_extoff; 889 maxlen = frag->fr_maxlen; 890 frag_id = frag->fr_id; 891 total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off + 892 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len; 893 hdrlen = frent->fe_hdrlen - sizeof(struct ip6_frag); 894 895 m = *m0 = pf_join_fragment(frag); 896 frag = NULL; 897 898 PF_FRAG_UNLOCK(); 899 900 /* Take protocol from first fragment header. */ 901 m = m_getptr(m, hdrlen + offsetof(struct ip6_frag, ip6f_nxt), &off); 902 KASSERT(m, ("%s: short mbuf chain", __func__)); 903 proto = *(mtod(m, uint8_t *) + off); 904 m = *m0; 905 906 /* Delete frag6 header */ 907 if (ip6_deletefraghdr(m, hdrlen, M_NOWAIT) != 0) 908 goto fail; 909 910 if (m->m_flags & M_PKTHDR) { 911 int plen = 0; 912 for (m = *m0; m; m = m->m_next) 913 plen += m->m_len; 914 m = *m0; 915 m->m_pkthdr.len = plen; 916 } 917 918 if ((mtag = m_tag_get(PACKET_TAG_PF_REASSEMBLED, 919 sizeof(struct pf_fragment_tag), M_NOWAIT)) == NULL) 920 goto fail; 921 ftag = (struct pf_fragment_tag *)(mtag + 1); 922 ftag->ft_hdrlen = hdrlen; 923 ftag->ft_extoff = extoff; 924 ftag->ft_maxlen = maxlen; 925 ftag->ft_id = frag_id; 926 m_tag_prepend(m, mtag); 927 928 ip6 = mtod(m, struct ip6_hdr *); 929 ip6->ip6_plen = htons(hdrlen - sizeof(struct ip6_hdr) + total); 930 if (extoff) { 931 /* Write protocol into next field of last extension header. */ 932 m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt), 933 &off); 934 KASSERT(m, ("%s: short mbuf chain", __func__)); 935 *(mtod(m, char *) + off) = proto; 936 m = *m0; 937 } else 938 ip6->ip6_nxt = proto; 939 940 if (hdrlen - sizeof(struct ip6_hdr) + total > IPV6_MAXPACKET) { 941 DPFPRINTF(("drop: too big: %d\n", total)); 942 ip6->ip6_plen = 0; 943 REASON_SET(reason, PFRES_SHORT); 944 /* PF_DROP requires a valid mbuf *m0 in pf_test6(). */ 945 return (PF_DROP); 946 } 947 948 DPFPRINTF(("complete: %p(%d)\n", m, ntohs(ip6->ip6_plen))); 949 return (PF_PASS); 950 951 fail: 952 REASON_SET(reason, PFRES_MEMORY); 953 /* PF_DROP requires a valid mbuf *m0 in pf_test6(), will free later. */ 954 return (PF_DROP); 955 } 956 #endif /* INET6 */ 957 958 #ifdef INET6 959 int 960 pf_max_frag_size(struct mbuf *m) 961 { 962 struct m_tag *tag; 963 struct pf_fragment_tag *ftag; 964 965 tag = m_tag_find(m, PACKET_TAG_PF_REASSEMBLED, NULL); 966 if (tag == NULL) 967 return (m->m_pkthdr.len); 968 969 ftag = (struct pf_fragment_tag *)(tag + 1); 970 971 return (ftag->ft_maxlen); 972 } 973 974 int 975 pf_refragment6(struct ifnet *ifp, struct mbuf **m0, struct m_tag *mtag, 976 struct ifnet *rt, bool forward) 977 { 978 struct mbuf *m = *m0, *t; 979 struct ip6_hdr *hdr; 980 struct pf_fragment_tag *ftag = (struct pf_fragment_tag *)(mtag + 1); 981 struct pf_pdesc pd; 982 uint32_t frag_id; 983 uint16_t hdrlen, extoff, maxlen; 984 uint8_t proto; 985 int error, action; 986 987 hdrlen = ftag->ft_hdrlen; 988 extoff = ftag->ft_extoff; 989 maxlen = ftag->ft_maxlen; 990 frag_id = ftag->ft_id; 991 m_tag_delete(m, mtag); 992 mtag = NULL; 993 ftag = NULL; 994 995 if (extoff) { 996 int off; 997 998 /* Use protocol from next field of last extension header */ 999 m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt), 1000 &off); 1001 KASSERT((m != NULL), ("pf_refragment6: short mbuf chain")); 1002 proto = *(mtod(m, uint8_t *) + off); 1003 *(mtod(m, char *) + off) = IPPROTO_FRAGMENT; 1004 m = *m0; 1005 } else { 1006 hdr = mtod(m, struct ip6_hdr *); 1007 proto = hdr->ip6_nxt; 1008 hdr->ip6_nxt = IPPROTO_FRAGMENT; 1009 } 1010 1011 /* In case of link-local traffic we'll need a scope set. */ 1012 hdr = mtod(m, struct ip6_hdr *); 1013 1014 in6_setscope(&hdr->ip6_src, ifp, NULL); 1015 in6_setscope(&hdr->ip6_dst, ifp, NULL); 1016 1017 /* The MTU must be a multiple of 8 bytes, or we risk doing the 1018 * fragmentation wrong. */ 1019 maxlen = maxlen & ~7; 1020 1021 /* 1022 * Maxlen may be less than 8 if there was only a single 1023 * fragment. As it was fragmented before, add a fragment 1024 * header also for a single fragment. If total or maxlen 1025 * is less than 8, ip6_fragment() will return EMSGSIZE and 1026 * we drop the packet. 1027 */ 1028 error = ip6_fragment(ifp, m, hdrlen, proto, maxlen, frag_id); 1029 m = (*m0)->m_nextpkt; 1030 (*m0)->m_nextpkt = NULL; 1031 if (error == 0) { 1032 /* The first mbuf contains the unfragmented packet. */ 1033 m_freem(*m0); 1034 *m0 = NULL; 1035 action = PF_PASS; 1036 } else { 1037 /* Drop expects an mbuf to free. */ 1038 DPFPRINTF(("refragment error %d\n", error)); 1039 action = PF_DROP; 1040 } 1041 for (; m; m = t) { 1042 t = m->m_nextpkt; 1043 m->m_nextpkt = NULL; 1044 m->m_flags |= M_SKIP_FIREWALL; 1045 memset(&pd, 0, sizeof(pd)); 1046 pd.pf_mtag = pf_find_mtag(m); 1047 if (error != 0) { 1048 m_freem(m); 1049 continue; 1050 } 1051 if (rt != NULL) { 1052 struct sockaddr_in6 dst; 1053 hdr = mtod(m, struct ip6_hdr *); 1054 1055 bzero(&dst, sizeof(dst)); 1056 dst.sin6_family = AF_INET6; 1057 dst.sin6_len = sizeof(dst); 1058 dst.sin6_addr = hdr->ip6_dst; 1059 1060 nd6_output_ifp(rt, rt, m, &dst, NULL); 1061 } else if (forward) { 1062 MPASS(m->m_pkthdr.rcvif != NULL); 1063 ip6_forward(m, 0); 1064 } else { 1065 (void)ip6_output(m, NULL, NULL, 0, NULL, NULL, 1066 NULL); 1067 } 1068 } 1069 1070 return (action); 1071 } 1072 #endif /* INET6 */ 1073 1074 #ifdef INET 1075 int 1076 pf_normalize_ip(struct mbuf **m0, u_short *reason, 1077 struct pf_pdesc *pd) 1078 { 1079 struct pf_krule *r; 1080 struct ip *h = mtod(*m0, struct ip *); 1081 int mff = (ntohs(h->ip_off) & IP_MF); 1082 int hlen = h->ip_hl << 2; 1083 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3; 1084 u_int16_t max; 1085 int ip_len; 1086 int tag = -1; 1087 int verdict; 1088 bool scrub_compat; 1089 1090 PF_RULES_RASSERT(); 1091 1092 MPASS(pd->m == *m0); 1093 1094 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr); 1095 /* 1096 * Check if there are any scrub rules, matching or not. 1097 * Lack of scrub rules means: 1098 * - enforced packet normalization operation just like in OpenBSD 1099 * - fragment reassembly depends on V_pf_status.reass 1100 * With scrub rules: 1101 * - packet normalization is performed if there is a matching scrub rule 1102 * - fragment reassembly is performed if the matching rule has no 1103 * PFRULE_FRAGMENT_NOREASS flag 1104 */ 1105 scrub_compat = (r != NULL); 1106 while (r != NULL) { 1107 pf_counter_u64_add(&r->evaluations, 1); 1108 if (pfi_kkif_match(r->kif, pd->kif) == r->ifnot) 1109 r = r->skip[PF_SKIP_IFP]; 1110 else if (r->direction && r->direction != pd->dir) 1111 r = r->skip[PF_SKIP_DIR]; 1112 else if (r->af && r->af != AF_INET) 1113 r = r->skip[PF_SKIP_AF]; 1114 else if (r->proto && r->proto != h->ip_p) 1115 r = r->skip[PF_SKIP_PROTO]; 1116 else if (PF_MISMATCHAW(&r->src.addr, 1117 (struct pf_addr *)&h->ip_src.s_addr, AF_INET, 1118 r->src.neg, pd->kif, M_GETFIB(pd->m))) 1119 r = r->skip[PF_SKIP_SRC_ADDR]; 1120 else if (PF_MISMATCHAW(&r->dst.addr, 1121 (struct pf_addr *)&h->ip_dst.s_addr, AF_INET, 1122 r->dst.neg, NULL, M_GETFIB(pd->m))) 1123 r = r->skip[PF_SKIP_DST_ADDR]; 1124 else if (r->match_tag && !pf_match_tag(pd->m, r, &tag, 1125 pd->pf_mtag ? pd->pf_mtag->tag : 0)) 1126 r = TAILQ_NEXT(r, entries); 1127 else 1128 break; 1129 } 1130 1131 if (scrub_compat) { 1132 /* With scrub rules present IPv4 normalization happens only 1133 * if one of rules has matched and it's not a "no scrub" rule */ 1134 if (r == NULL || r->action == PF_NOSCRUB) 1135 return (PF_PASS); 1136 1137 pf_counter_u64_critical_enter(); 1138 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1); 1139 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len); 1140 pf_counter_u64_critical_exit(); 1141 pf_rule_to_actions(r, &pd->act); 1142 } 1143 1144 /* Check for illegal packets */ 1145 if (hlen < (int)sizeof(struct ip)) { 1146 REASON_SET(reason, PFRES_NORM); 1147 goto drop; 1148 } 1149 1150 if (hlen > ntohs(h->ip_len)) { 1151 REASON_SET(reason, PFRES_NORM); 1152 goto drop; 1153 } 1154 1155 /* Clear IP_DF if the rule uses the no-df option or we're in no-df mode */ 1156 if (((!scrub_compat && V_pf_status.reass & PF_REASS_NODF) || 1157 (r != NULL && r->rule_flag & PFRULE_NODF)) && 1158 (h->ip_off & htons(IP_DF)) 1159 ) { 1160 u_int16_t ip_off = h->ip_off; 1161 1162 h->ip_off &= htons(~IP_DF); 1163 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0); 1164 } 1165 1166 /* We will need other tests here */ 1167 if (!fragoff && !mff) 1168 goto no_fragment; 1169 1170 /* We're dealing with a fragment now. Don't allow fragments 1171 * with IP_DF to enter the cache. If the flag was cleared by 1172 * no-df above, fine. Otherwise drop it. 1173 */ 1174 if (h->ip_off & htons(IP_DF)) { 1175 DPFPRINTF(("IP_DF\n")); 1176 goto bad; 1177 } 1178 1179 ip_len = ntohs(h->ip_len) - hlen; 1180 1181 /* All fragments are 8 byte aligned */ 1182 if (mff && (ip_len & 0x7)) { 1183 DPFPRINTF(("mff and %d\n", ip_len)); 1184 goto bad; 1185 } 1186 1187 /* Respect maximum length */ 1188 if (fragoff + ip_len > IP_MAXPACKET) { 1189 DPFPRINTF(("max packet %d\n", fragoff + ip_len)); 1190 goto bad; 1191 } 1192 1193 if ((!scrub_compat && V_pf_status.reass) || 1194 (r != NULL && !(r->rule_flag & PFRULE_FRAGMENT_NOREASS)) 1195 ) { 1196 max = fragoff + ip_len; 1197 1198 /* Fully buffer all of the fragments 1199 * Might return a completely reassembled mbuf, or NULL */ 1200 PF_FRAG_LOCK(); 1201 DPFPRINTF(("reass frag %d @ %d-%d\n", h->ip_id, fragoff, max)); 1202 verdict = pf_reassemble(m0, pd->dir, reason); 1203 PF_FRAG_UNLOCK(); 1204 1205 if (verdict != PF_PASS) 1206 return (PF_DROP); 1207 1208 pd->m = *m0; 1209 if (pd->m == NULL) 1210 return (PF_DROP); 1211 1212 h = mtod(pd->m, struct ip *); 1213 pd->tot_len = htons(h->ip_len); 1214 1215 no_fragment: 1216 /* At this point, only IP_DF is allowed in ip_off */ 1217 if (h->ip_off & ~htons(IP_DF)) { 1218 u_int16_t ip_off = h->ip_off; 1219 1220 h->ip_off &= htons(IP_DF); 1221 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0); 1222 } 1223 } 1224 1225 return (PF_PASS); 1226 1227 bad: 1228 DPFPRINTF(("dropping bad fragment\n")); 1229 REASON_SET(reason, PFRES_FRAG); 1230 drop: 1231 if (r != NULL && r->log) 1232 PFLOG_PACKET(PF_DROP, *reason, r, NULL, NULL, pd, 1); 1233 1234 return (PF_DROP); 1235 } 1236 #endif 1237 1238 #ifdef INET6 1239 int 1240 pf_normalize_ip6(struct mbuf **m0, int off, u_short *reason, 1241 struct pf_pdesc *pd) 1242 { 1243 struct pf_krule *r; 1244 struct ip6_hdr *h; 1245 struct ip6_frag frag; 1246 bool scrub_compat; 1247 1248 PF_RULES_RASSERT(); 1249 1250 pd->m = *m0; 1251 1252 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr); 1253 /* 1254 * Check if there are any scrub rules, matching or not. 1255 * Lack of scrub rules means: 1256 * - enforced packet normalization operation just like in OpenBSD 1257 * With scrub rules: 1258 * - packet normalization is performed if there is a matching scrub rule 1259 * XXX: Fragment reassembly always performed for IPv6! 1260 */ 1261 scrub_compat = (r != NULL); 1262 while (r != NULL) { 1263 pf_counter_u64_add(&r->evaluations, 1); 1264 if (pfi_kkif_match(r->kif, pd->kif) == r->ifnot) 1265 r = r->skip[PF_SKIP_IFP]; 1266 else if (r->direction && r->direction != pd->dir) 1267 r = r->skip[PF_SKIP_DIR]; 1268 else if (r->af && r->af != AF_INET6) 1269 r = r->skip[PF_SKIP_AF]; 1270 else if (r->proto && r->proto != pd->proto) 1271 r = r->skip[PF_SKIP_PROTO]; 1272 else if (PF_MISMATCHAW(&r->src.addr, 1273 (struct pf_addr *)&pd->src, AF_INET6, 1274 r->src.neg, pd->kif, M_GETFIB(pd->m))) 1275 r = r->skip[PF_SKIP_SRC_ADDR]; 1276 else if (PF_MISMATCHAW(&r->dst.addr, 1277 (struct pf_addr *)&pd->dst, AF_INET6, 1278 r->dst.neg, NULL, M_GETFIB(pd->m))) 1279 r = r->skip[PF_SKIP_DST_ADDR]; 1280 else 1281 break; 1282 } 1283 1284 if (scrub_compat) { 1285 /* With scrub rules present IPv6 normalization happens only 1286 * if one of rules has matched and it's not a "no scrub" rule */ 1287 if (r == NULL || r->action == PF_NOSCRUB) 1288 return (PF_PASS); 1289 1290 pf_counter_u64_critical_enter(); 1291 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1); 1292 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len); 1293 pf_counter_u64_critical_exit(); 1294 pf_rule_to_actions(r, &pd->act); 1295 } 1296 1297 if (!pf_pull_hdr(pd->m, off, &frag, sizeof(frag), NULL, reason, AF_INET6)) 1298 return (PF_DROP); 1299 1300 /* Offset now points to data portion. */ 1301 off += sizeof(frag); 1302 1303 if (pd->virtual_proto == PF_VPROTO_FRAGMENT) { 1304 /* Returns PF_DROP or *m0 is NULL or completely reassembled 1305 * mbuf. */ 1306 if (pf_reassemble6(m0, &frag, off, pd->extoff, reason) != PF_PASS) 1307 return (PF_DROP); 1308 pd->m = *m0; 1309 if (pd->m == NULL) 1310 return (PF_DROP); 1311 h = mtod(pd->m, struct ip6_hdr *); 1312 pd->tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr); 1313 } 1314 1315 return (PF_PASS); 1316 } 1317 #endif /* INET6 */ 1318 1319 int 1320 pf_normalize_tcp(struct pf_pdesc *pd) 1321 { 1322 struct pf_krule *r, *rm = NULL; 1323 struct tcphdr *th = &pd->hdr.tcp; 1324 int rewrite = 0; 1325 u_short reason; 1326 u_int16_t flags; 1327 sa_family_t af = pd->af; 1328 int srs; 1329 1330 PF_RULES_RASSERT(); 1331 1332 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr); 1333 /* Check if there any scrub rules. Lack of scrub rules means enforced 1334 * packet normalization operation just like in OpenBSD. */ 1335 srs = (r != NULL); 1336 while (r != NULL) { 1337 pf_counter_u64_add(&r->evaluations, 1); 1338 if (pfi_kkif_match(r->kif, pd->kif) == r->ifnot) 1339 r = r->skip[PF_SKIP_IFP]; 1340 else if (r->direction && r->direction != pd->dir) 1341 r = r->skip[PF_SKIP_DIR]; 1342 else if (r->af && r->af != af) 1343 r = r->skip[PF_SKIP_AF]; 1344 else if (r->proto && r->proto != pd->proto) 1345 r = r->skip[PF_SKIP_PROTO]; 1346 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af, 1347 r->src.neg, pd->kif, M_GETFIB(pd->m))) 1348 r = r->skip[PF_SKIP_SRC_ADDR]; 1349 else if (r->src.port_op && !pf_match_port(r->src.port_op, 1350 r->src.port[0], r->src.port[1], th->th_sport)) 1351 r = r->skip[PF_SKIP_SRC_PORT]; 1352 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af, 1353 r->dst.neg, NULL, M_GETFIB(pd->m))) 1354 r = r->skip[PF_SKIP_DST_ADDR]; 1355 else if (r->dst.port_op && !pf_match_port(r->dst.port_op, 1356 r->dst.port[0], r->dst.port[1], th->th_dport)) 1357 r = r->skip[PF_SKIP_DST_PORT]; 1358 else if (r->os_fingerprint != PF_OSFP_ANY && !pf_osfp_match( 1359 pf_osfp_fingerprint(pd, th), 1360 r->os_fingerprint)) 1361 r = TAILQ_NEXT(r, entries); 1362 else { 1363 rm = r; 1364 break; 1365 } 1366 } 1367 1368 if (srs) { 1369 /* With scrub rules present TCP normalization happens only 1370 * if one of rules has matched and it's not a "no scrub" rule */ 1371 if (rm == NULL || rm->action == PF_NOSCRUB) 1372 return (PF_PASS); 1373 1374 pf_counter_u64_critical_enter(); 1375 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1); 1376 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len); 1377 pf_counter_u64_critical_exit(); 1378 pf_rule_to_actions(rm, &pd->act); 1379 } 1380 1381 if (rm && rm->rule_flag & PFRULE_REASSEMBLE_TCP) 1382 pd->flags |= PFDESC_TCP_NORM; 1383 1384 flags = tcp_get_flags(th); 1385 if (flags & TH_SYN) { 1386 /* Illegal packet */ 1387 if (flags & TH_RST) 1388 goto tcp_drop; 1389 1390 if (flags & TH_FIN) 1391 goto tcp_drop; 1392 } else { 1393 /* Illegal packet */ 1394 if (!(flags & (TH_ACK|TH_RST))) 1395 goto tcp_drop; 1396 } 1397 1398 if (!(flags & TH_ACK)) { 1399 /* These flags are only valid if ACK is set */ 1400 if ((flags & TH_FIN) || (flags & TH_PUSH) || (flags & TH_URG)) 1401 goto tcp_drop; 1402 } 1403 1404 /* Check for illegal header length */ 1405 if (th->th_off < (sizeof(struct tcphdr) >> 2)) 1406 goto tcp_drop; 1407 1408 /* If flags changed, or reserved data set, then adjust */ 1409 if (flags != tcp_get_flags(th) || 1410 (tcp_get_flags(th) & (TH_RES1|TH_RES2|TH_RES2)) != 0) { 1411 u_int16_t ov, nv; 1412 1413 ov = *(u_int16_t *)(&th->th_ack + 1); 1414 flags &= ~(TH_RES1 | TH_RES2 | TH_RES3); 1415 tcp_set_flags(th, flags); 1416 nv = *(u_int16_t *)(&th->th_ack + 1); 1417 1418 th->th_sum = pf_proto_cksum_fixup(pd->m, th->th_sum, ov, nv, 0); 1419 rewrite = 1; 1420 } 1421 1422 /* Remove urgent pointer, if TH_URG is not set */ 1423 if (!(flags & TH_URG) && th->th_urp) { 1424 th->th_sum = pf_proto_cksum_fixup(pd->m, th->th_sum, th->th_urp, 1425 0, 0); 1426 th->th_urp = 0; 1427 rewrite = 1; 1428 } 1429 1430 /* copy back packet headers if we sanitized */ 1431 if (rewrite) 1432 m_copyback(pd->m, pd->off, sizeof(*th), (caddr_t)th); 1433 1434 return (PF_PASS); 1435 1436 tcp_drop: 1437 REASON_SET(&reason, PFRES_NORM); 1438 if (rm != NULL && r->log) 1439 PFLOG_PACKET(PF_DROP, reason, r, NULL, NULL, pd, 1); 1440 return (PF_DROP); 1441 } 1442 1443 int 1444 pf_normalize_tcp_init(struct pf_pdesc *pd, struct tcphdr *th, 1445 struct pf_state_peer *src, struct pf_state_peer *dst) 1446 { 1447 u_int32_t tsval, tsecr; 1448 u_int8_t hdr[60]; 1449 u_int8_t *opt; 1450 1451 KASSERT((src->scrub == NULL), 1452 ("pf_normalize_tcp_init: src->scrub != NULL")); 1453 1454 src->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT); 1455 if (src->scrub == NULL) 1456 return (1); 1457 1458 switch (pd->af) { 1459 #ifdef INET 1460 case AF_INET: { 1461 struct ip *h = mtod(pd->m, struct ip *); 1462 src->scrub->pfss_ttl = h->ip_ttl; 1463 break; 1464 } 1465 #endif /* INET */ 1466 #ifdef INET6 1467 case AF_INET6: { 1468 struct ip6_hdr *h = mtod(pd->m, struct ip6_hdr *); 1469 src->scrub->pfss_ttl = h->ip6_hlim; 1470 break; 1471 } 1472 #endif /* INET6 */ 1473 } 1474 1475 /* 1476 * All normalizations below are only begun if we see the start of 1477 * the connections. They must all set an enabled bit in pfss_flags 1478 */ 1479 if ((tcp_get_flags(th) & TH_SYN) == 0) 1480 return (0); 1481 1482 if (th->th_off > (sizeof(struct tcphdr) >> 2) && src->scrub && 1483 pf_pull_hdr(pd->m, pd->off, hdr, th->th_off << 2, NULL, NULL, pd->af)) { 1484 /* Diddle with TCP options */ 1485 int hlen; 1486 opt = hdr + sizeof(struct tcphdr); 1487 hlen = (th->th_off << 2) - sizeof(struct tcphdr); 1488 while (hlen >= TCPOLEN_TIMESTAMP) { 1489 switch (*opt) { 1490 case TCPOPT_EOL: /* FALLTHROUGH */ 1491 case TCPOPT_NOP: 1492 opt++; 1493 hlen--; 1494 break; 1495 case TCPOPT_TIMESTAMP: 1496 if (opt[1] >= TCPOLEN_TIMESTAMP) { 1497 src->scrub->pfss_flags |= 1498 PFSS_TIMESTAMP; 1499 src->scrub->pfss_ts_mod = 1500 htonl(arc4random()); 1501 1502 /* note PFSS_PAWS not set yet */ 1503 memcpy(&tsval, &opt[2], 1504 sizeof(u_int32_t)); 1505 memcpy(&tsecr, &opt[6], 1506 sizeof(u_int32_t)); 1507 src->scrub->pfss_tsval0 = ntohl(tsval); 1508 src->scrub->pfss_tsval = ntohl(tsval); 1509 src->scrub->pfss_tsecr = ntohl(tsecr); 1510 getmicrouptime(&src->scrub->pfss_last); 1511 } 1512 /* FALLTHROUGH */ 1513 default: 1514 hlen -= MAX(opt[1], 2); 1515 opt += MAX(opt[1], 2); 1516 break; 1517 } 1518 } 1519 } 1520 1521 return (0); 1522 } 1523 1524 void 1525 pf_normalize_tcp_cleanup(struct pf_kstate *state) 1526 { 1527 /* XXX Note: this also cleans up SCTP. */ 1528 uma_zfree(V_pf_state_scrub_z, state->src.scrub); 1529 uma_zfree(V_pf_state_scrub_z, state->dst.scrub); 1530 1531 /* Someday... flush the TCP segment reassembly descriptors. */ 1532 } 1533 int 1534 pf_normalize_sctp_init(struct pf_pdesc *pd, struct pf_state_peer *src, 1535 struct pf_state_peer *dst) 1536 { 1537 src->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT); 1538 if (src->scrub == NULL) 1539 return (1); 1540 1541 dst->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT); 1542 if (dst->scrub == NULL) { 1543 uma_zfree(V_pf_state_scrub_z, src); 1544 return (1); 1545 } 1546 1547 dst->scrub->pfss_v_tag = pd->sctp_initiate_tag; 1548 1549 return (0); 1550 } 1551 1552 int 1553 pf_normalize_tcp_stateful(struct pf_pdesc *pd, 1554 u_short *reason, struct tcphdr *th, struct pf_kstate *state, 1555 struct pf_state_peer *src, struct pf_state_peer *dst, int *writeback) 1556 { 1557 struct timeval uptime; 1558 u_int32_t tsval, tsecr; 1559 u_int tsval_from_last; 1560 u_int8_t hdr[60]; 1561 u_int8_t *opt; 1562 int copyback = 0; 1563 int got_ts = 0; 1564 size_t startoff; 1565 1566 KASSERT((src->scrub || dst->scrub), 1567 ("%s: src->scrub && dst->scrub!", __func__)); 1568 1569 /* 1570 * Enforce the minimum TTL seen for this connection. Negate a common 1571 * technique to evade an intrusion detection system and confuse 1572 * firewall state code. 1573 */ 1574 switch (pd->af) { 1575 #ifdef INET 1576 case AF_INET: { 1577 if (src->scrub) { 1578 struct ip *h = mtod(pd->m, struct ip *); 1579 if (h->ip_ttl > src->scrub->pfss_ttl) 1580 src->scrub->pfss_ttl = h->ip_ttl; 1581 h->ip_ttl = src->scrub->pfss_ttl; 1582 } 1583 break; 1584 } 1585 #endif /* INET */ 1586 #ifdef INET6 1587 case AF_INET6: { 1588 if (src->scrub) { 1589 struct ip6_hdr *h = mtod(pd->m, struct ip6_hdr *); 1590 if (h->ip6_hlim > src->scrub->pfss_ttl) 1591 src->scrub->pfss_ttl = h->ip6_hlim; 1592 h->ip6_hlim = src->scrub->pfss_ttl; 1593 } 1594 break; 1595 } 1596 #endif /* INET6 */ 1597 } 1598 1599 if (th->th_off > (sizeof(struct tcphdr) >> 2) && 1600 ((src->scrub && (src->scrub->pfss_flags & PFSS_TIMESTAMP)) || 1601 (dst->scrub && (dst->scrub->pfss_flags & PFSS_TIMESTAMP))) && 1602 pf_pull_hdr(pd->m, pd->off, hdr, th->th_off << 2, NULL, NULL, pd->af)) { 1603 /* Diddle with TCP options */ 1604 int hlen; 1605 opt = hdr + sizeof(struct tcphdr); 1606 hlen = (th->th_off << 2) - sizeof(struct tcphdr); 1607 while (hlen >= TCPOLEN_TIMESTAMP) { 1608 startoff = opt - (hdr + sizeof(struct tcphdr)); 1609 switch (*opt) { 1610 case TCPOPT_EOL: /* FALLTHROUGH */ 1611 case TCPOPT_NOP: 1612 opt++; 1613 hlen--; 1614 break; 1615 case TCPOPT_TIMESTAMP: 1616 /* Modulate the timestamps. Can be used for 1617 * NAT detection, OS uptime determination or 1618 * reboot detection. 1619 */ 1620 1621 if (got_ts) { 1622 /* Huh? Multiple timestamps!? */ 1623 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1624 DPFPRINTF(("multiple TS??\n")); 1625 pf_print_state(state); 1626 printf("\n"); 1627 } 1628 REASON_SET(reason, PFRES_TS); 1629 return (PF_DROP); 1630 } 1631 if (opt[1] >= TCPOLEN_TIMESTAMP) { 1632 memcpy(&tsval, &opt[2], 1633 sizeof(u_int32_t)); 1634 if (tsval && src->scrub && 1635 (src->scrub->pfss_flags & 1636 PFSS_TIMESTAMP)) { 1637 tsval = ntohl(tsval); 1638 pf_patch_32_unaligned(pd->m, 1639 &th->th_sum, 1640 &opt[2], 1641 htonl(tsval + 1642 src->scrub->pfss_ts_mod), 1643 PF_ALGNMNT(startoff), 1644 0); 1645 copyback = 1; 1646 } 1647 1648 /* Modulate TS reply iff valid (!0) */ 1649 memcpy(&tsecr, &opt[6], 1650 sizeof(u_int32_t)); 1651 if (tsecr && dst->scrub && 1652 (dst->scrub->pfss_flags & 1653 PFSS_TIMESTAMP)) { 1654 tsecr = ntohl(tsecr) 1655 - dst->scrub->pfss_ts_mod; 1656 pf_patch_32_unaligned(pd->m, 1657 &th->th_sum, 1658 &opt[6], 1659 htonl(tsecr), 1660 PF_ALGNMNT(startoff), 1661 0); 1662 copyback = 1; 1663 } 1664 got_ts = 1; 1665 } 1666 /* FALLTHROUGH */ 1667 default: 1668 hlen -= MAX(opt[1], 2); 1669 opt += MAX(opt[1], 2); 1670 break; 1671 } 1672 } 1673 if (copyback) { 1674 /* Copyback the options, caller copys back header */ 1675 *writeback = 1; 1676 m_copyback(pd->m, pd->off + sizeof(struct tcphdr), 1677 (th->th_off << 2) - sizeof(struct tcphdr), hdr + 1678 sizeof(struct tcphdr)); 1679 } 1680 } 1681 1682 /* 1683 * Must invalidate PAWS checks on connections idle for too long. 1684 * The fastest allowed timestamp clock is 1ms. That turns out to 1685 * be about 24 days before it wraps. XXX Right now our lowerbound 1686 * TS echo check only works for the first 12 days of a connection 1687 * when the TS has exhausted half its 32bit space 1688 */ 1689 #define TS_MAX_IDLE (24*24*60*60) 1690 #define TS_MAX_CONN (12*24*60*60) /* XXX remove when better tsecr check */ 1691 1692 getmicrouptime(&uptime); 1693 if (src->scrub && (src->scrub->pfss_flags & PFSS_PAWS) && 1694 (uptime.tv_sec - src->scrub->pfss_last.tv_sec > TS_MAX_IDLE || 1695 time_uptime - (state->creation / 1000) > TS_MAX_CONN)) { 1696 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1697 DPFPRINTF(("src idled out of PAWS\n")); 1698 pf_print_state(state); 1699 printf("\n"); 1700 } 1701 src->scrub->pfss_flags = (src->scrub->pfss_flags & ~PFSS_PAWS) 1702 | PFSS_PAWS_IDLED; 1703 } 1704 if (dst->scrub && (dst->scrub->pfss_flags & PFSS_PAWS) && 1705 uptime.tv_sec - dst->scrub->pfss_last.tv_sec > TS_MAX_IDLE) { 1706 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1707 DPFPRINTF(("dst idled out of PAWS\n")); 1708 pf_print_state(state); 1709 printf("\n"); 1710 } 1711 dst->scrub->pfss_flags = (dst->scrub->pfss_flags & ~PFSS_PAWS) 1712 | PFSS_PAWS_IDLED; 1713 } 1714 1715 if (got_ts && src->scrub && dst->scrub && 1716 (src->scrub->pfss_flags & PFSS_PAWS) && 1717 (dst->scrub->pfss_flags & PFSS_PAWS)) { 1718 /* Validate that the timestamps are "in-window". 1719 * RFC1323 describes TCP Timestamp options that allow 1720 * measurement of RTT (round trip time) and PAWS 1721 * (protection against wrapped sequence numbers). PAWS 1722 * gives us a set of rules for rejecting packets on 1723 * long fat pipes (packets that were somehow delayed 1724 * in transit longer than the time it took to send the 1725 * full TCP sequence space of 4Gb). We can use these 1726 * rules and infer a few others that will let us treat 1727 * the 32bit timestamp and the 32bit echoed timestamp 1728 * as sequence numbers to prevent a blind attacker from 1729 * inserting packets into a connection. 1730 * 1731 * RFC1323 tells us: 1732 * - The timestamp on this packet must be greater than 1733 * or equal to the last value echoed by the other 1734 * endpoint. The RFC says those will be discarded 1735 * since it is a dup that has already been acked. 1736 * This gives us a lowerbound on the timestamp. 1737 * timestamp >= other last echoed timestamp 1738 * - The timestamp will be less than or equal to 1739 * the last timestamp plus the time between the 1740 * last packet and now. The RFC defines the max 1741 * clock rate as 1ms. We will allow clocks to be 1742 * up to 10% fast and will allow a total difference 1743 * or 30 seconds due to a route change. And this 1744 * gives us an upperbound on the timestamp. 1745 * timestamp <= last timestamp + max ticks 1746 * We have to be careful here. Windows will send an 1747 * initial timestamp of zero and then initialize it 1748 * to a random value after the 3whs; presumably to 1749 * avoid a DoS by having to call an expensive RNG 1750 * during a SYN flood. Proof MS has at least one 1751 * good security geek. 1752 * 1753 * - The TCP timestamp option must also echo the other 1754 * endpoints timestamp. The timestamp echoed is the 1755 * one carried on the earliest unacknowledged segment 1756 * on the left edge of the sequence window. The RFC 1757 * states that the host will reject any echoed 1758 * timestamps that were larger than any ever sent. 1759 * This gives us an upperbound on the TS echo. 1760 * tescr <= largest_tsval 1761 * - The lowerbound on the TS echo is a little more 1762 * tricky to determine. The other endpoint's echoed 1763 * values will not decrease. But there may be 1764 * network conditions that re-order packets and 1765 * cause our view of them to decrease. For now the 1766 * only lowerbound we can safely determine is that 1767 * the TS echo will never be less than the original 1768 * TS. XXX There is probably a better lowerbound. 1769 * Remove TS_MAX_CONN with better lowerbound check. 1770 * tescr >= other original TS 1771 * 1772 * It is also important to note that the fastest 1773 * timestamp clock of 1ms will wrap its 32bit space in 1774 * 24 days. So we just disable TS checking after 24 1775 * days of idle time. We actually must use a 12d 1776 * connection limit until we can come up with a better 1777 * lowerbound to the TS echo check. 1778 */ 1779 struct timeval delta_ts; 1780 int ts_fudge; 1781 1782 /* 1783 * PFTM_TS_DIFF is how many seconds of leeway to allow 1784 * a host's timestamp. This can happen if the previous 1785 * packet got delayed in transit for much longer than 1786 * this packet. 1787 */ 1788 if ((ts_fudge = state->rule->timeout[PFTM_TS_DIFF]) == 0) 1789 ts_fudge = V_pf_default_rule.timeout[PFTM_TS_DIFF]; 1790 1791 /* Calculate max ticks since the last timestamp */ 1792 #define TS_MAXFREQ 1100 /* RFC max TS freq of 1Khz + 10% skew */ 1793 #define TS_MICROSECS 1000000 /* microseconds per second */ 1794 delta_ts = uptime; 1795 timevalsub(&delta_ts, &src->scrub->pfss_last); 1796 tsval_from_last = (delta_ts.tv_sec + ts_fudge) * TS_MAXFREQ; 1797 tsval_from_last += delta_ts.tv_usec / (TS_MICROSECS/TS_MAXFREQ); 1798 1799 if ((src->state >= TCPS_ESTABLISHED && 1800 dst->state >= TCPS_ESTABLISHED) && 1801 (SEQ_LT(tsval, dst->scrub->pfss_tsecr) || 1802 SEQ_GT(tsval, src->scrub->pfss_tsval + tsval_from_last) || 1803 (tsecr && (SEQ_GT(tsecr, dst->scrub->pfss_tsval) || 1804 SEQ_LT(tsecr, dst->scrub->pfss_tsval0))))) { 1805 /* Bad RFC1323 implementation or an insertion attack. 1806 * 1807 * - Solaris 2.6 and 2.7 are known to send another ACK 1808 * after the FIN,FIN|ACK,ACK closing that carries 1809 * an old timestamp. 1810 */ 1811 1812 DPFPRINTF(("Timestamp failed %c%c%c%c\n", 1813 SEQ_LT(tsval, dst->scrub->pfss_tsecr) ? '0' : ' ', 1814 SEQ_GT(tsval, src->scrub->pfss_tsval + 1815 tsval_from_last) ? '1' : ' ', 1816 SEQ_GT(tsecr, dst->scrub->pfss_tsval) ? '2' : ' ', 1817 SEQ_LT(tsecr, dst->scrub->pfss_tsval0)? '3' : ' ')); 1818 DPFPRINTF((" tsval: %u tsecr: %u +ticks: %u " 1819 "idle: %jus %lums\n", 1820 tsval, tsecr, tsval_from_last, 1821 (uintmax_t)delta_ts.tv_sec, 1822 delta_ts.tv_usec / 1000)); 1823 DPFPRINTF((" src->tsval: %u tsecr: %u\n", 1824 src->scrub->pfss_tsval, src->scrub->pfss_tsecr)); 1825 DPFPRINTF((" dst->tsval: %u tsecr: %u tsval0: %u" 1826 "\n", dst->scrub->pfss_tsval, 1827 dst->scrub->pfss_tsecr, dst->scrub->pfss_tsval0)); 1828 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1829 pf_print_state(state); 1830 pf_print_flags(tcp_get_flags(th)); 1831 printf("\n"); 1832 } 1833 REASON_SET(reason, PFRES_TS); 1834 return (PF_DROP); 1835 } 1836 1837 /* XXX I'd really like to require tsecr but it's optional */ 1838 1839 } else if (!got_ts && (tcp_get_flags(th) & TH_RST) == 0 && 1840 ((src->state == TCPS_ESTABLISHED && dst->state == TCPS_ESTABLISHED) 1841 || pd->p_len > 0 || (tcp_get_flags(th) & TH_SYN)) && 1842 src->scrub && dst->scrub && 1843 (src->scrub->pfss_flags & PFSS_PAWS) && 1844 (dst->scrub->pfss_flags & PFSS_PAWS)) { 1845 /* Didn't send a timestamp. Timestamps aren't really useful 1846 * when: 1847 * - connection opening or closing (often not even sent). 1848 * but we must not let an attacker to put a FIN on a 1849 * data packet to sneak it through our ESTABLISHED check. 1850 * - on a TCP reset. RFC suggests not even looking at TS. 1851 * - on an empty ACK. The TS will not be echoed so it will 1852 * probably not help keep the RTT calculation in sync and 1853 * there isn't as much danger when the sequence numbers 1854 * got wrapped. So some stacks don't include TS on empty 1855 * ACKs :-( 1856 * 1857 * To minimize the disruption to mostly RFC1323 conformant 1858 * stacks, we will only require timestamps on data packets. 1859 * 1860 * And what do ya know, we cannot require timestamps on data 1861 * packets. There appear to be devices that do legitimate 1862 * TCP connection hijacking. There are HTTP devices that allow 1863 * a 3whs (with timestamps) and then buffer the HTTP request. 1864 * If the intermediate device has the HTTP response cache, it 1865 * will spoof the response but not bother timestamping its 1866 * packets. So we can look for the presence of a timestamp in 1867 * the first data packet and if there, require it in all future 1868 * packets. 1869 */ 1870 1871 if (pd->p_len > 0 && (src->scrub->pfss_flags & PFSS_DATA_TS)) { 1872 /* 1873 * Hey! Someone tried to sneak a packet in. Or the 1874 * stack changed its RFC1323 behavior?!?! 1875 */ 1876 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1877 DPFPRINTF(("Did not receive expected RFC1323 " 1878 "timestamp\n")); 1879 pf_print_state(state); 1880 pf_print_flags(tcp_get_flags(th)); 1881 printf("\n"); 1882 } 1883 REASON_SET(reason, PFRES_TS); 1884 return (PF_DROP); 1885 } 1886 } 1887 1888 /* 1889 * We will note if a host sends his data packets with or without 1890 * timestamps. And require all data packets to contain a timestamp 1891 * if the first does. PAWS implicitly requires that all data packets be 1892 * timestamped. But I think there are middle-man devices that hijack 1893 * TCP streams immediately after the 3whs and don't timestamp their 1894 * packets (seen in a WWW accelerator or cache). 1895 */ 1896 if (pd->p_len > 0 && src->scrub && (src->scrub->pfss_flags & 1897 (PFSS_TIMESTAMP|PFSS_DATA_TS|PFSS_DATA_NOTS)) == PFSS_TIMESTAMP) { 1898 if (got_ts) 1899 src->scrub->pfss_flags |= PFSS_DATA_TS; 1900 else { 1901 src->scrub->pfss_flags |= PFSS_DATA_NOTS; 1902 if (V_pf_status.debug >= PF_DEBUG_MISC && dst->scrub && 1903 (dst->scrub->pfss_flags & PFSS_TIMESTAMP)) { 1904 /* Don't warn if other host rejected RFC1323 */ 1905 DPFPRINTF(("Broken RFC1323 stack did not " 1906 "timestamp data packet. Disabled PAWS " 1907 "security.\n")); 1908 pf_print_state(state); 1909 pf_print_flags(tcp_get_flags(th)); 1910 printf("\n"); 1911 } 1912 } 1913 } 1914 1915 /* 1916 * Update PAWS values 1917 */ 1918 if (got_ts && src->scrub && PFSS_TIMESTAMP == (src->scrub->pfss_flags & 1919 (PFSS_PAWS_IDLED|PFSS_TIMESTAMP))) { 1920 getmicrouptime(&src->scrub->pfss_last); 1921 if (SEQ_GEQ(tsval, src->scrub->pfss_tsval) || 1922 (src->scrub->pfss_flags & PFSS_PAWS) == 0) 1923 src->scrub->pfss_tsval = tsval; 1924 1925 if (tsecr) { 1926 if (SEQ_GEQ(tsecr, src->scrub->pfss_tsecr) || 1927 (src->scrub->pfss_flags & PFSS_PAWS) == 0) 1928 src->scrub->pfss_tsecr = tsecr; 1929 1930 if ((src->scrub->pfss_flags & PFSS_PAWS) == 0 && 1931 (SEQ_LT(tsval, src->scrub->pfss_tsval0) || 1932 src->scrub->pfss_tsval0 == 0)) { 1933 /* tsval0 MUST be the lowest timestamp */ 1934 src->scrub->pfss_tsval0 = tsval; 1935 } 1936 1937 /* Only fully initialized after a TS gets echoed */ 1938 if ((src->scrub->pfss_flags & PFSS_PAWS) == 0) 1939 src->scrub->pfss_flags |= PFSS_PAWS; 1940 } 1941 } 1942 1943 /* I have a dream.... TCP segment reassembly.... */ 1944 return (0); 1945 } 1946 1947 int 1948 pf_normalize_mss(struct pf_pdesc *pd) 1949 { 1950 struct tcphdr *th = &pd->hdr.tcp; 1951 u_int16_t *mss; 1952 int thoff; 1953 int opt, cnt, optlen = 0; 1954 u_char opts[TCP_MAXOLEN]; 1955 u_char *optp = opts; 1956 size_t startoff; 1957 1958 thoff = th->th_off << 2; 1959 cnt = thoff - sizeof(struct tcphdr); 1960 1961 if (cnt > 0 && !pf_pull_hdr(pd->m, pd->off + sizeof(*th), opts, cnt, 1962 NULL, NULL, pd->af)) 1963 return (0); 1964 1965 for (; cnt > 0; cnt -= optlen, optp += optlen) { 1966 startoff = optp - opts; 1967 opt = optp[0]; 1968 if (opt == TCPOPT_EOL) 1969 break; 1970 if (opt == TCPOPT_NOP) 1971 optlen = 1; 1972 else { 1973 if (cnt < 2) 1974 break; 1975 optlen = optp[1]; 1976 if (optlen < 2 || optlen > cnt) 1977 break; 1978 } 1979 switch (opt) { 1980 case TCPOPT_MAXSEG: 1981 mss = (u_int16_t *)(optp + 2); 1982 if ((ntohs(*mss)) > pd->act.max_mss) { 1983 pf_patch_16_unaligned(pd->m, 1984 &th->th_sum, 1985 mss, htons(pd->act.max_mss), 1986 PF_ALGNMNT(startoff), 1987 0); 1988 m_copyback(pd->m, pd->off + sizeof(*th), 1989 thoff - sizeof(*th), opts); 1990 m_copyback(pd->m, pd->off, sizeof(*th), (caddr_t)th); 1991 } 1992 break; 1993 default: 1994 break; 1995 } 1996 } 1997 1998 return (0); 1999 } 2000 2001 int 2002 pf_scan_sctp(struct pf_pdesc *pd) 2003 { 2004 struct sctp_chunkhdr ch = { }; 2005 int chunk_off = sizeof(struct sctphdr); 2006 int chunk_start; 2007 int ret; 2008 2009 while (pd->off + chunk_off < pd->tot_len) { 2010 if (!pf_pull_hdr(pd->m, pd->off + chunk_off, &ch, sizeof(ch), NULL, 2011 NULL, pd->af)) 2012 return (PF_DROP); 2013 2014 /* Length includes the header, this must be at least 4. */ 2015 if (ntohs(ch.chunk_length) < 4) 2016 return (PF_DROP); 2017 2018 chunk_start = chunk_off; 2019 chunk_off += roundup(ntohs(ch.chunk_length), 4); 2020 2021 switch (ch.chunk_type) { 2022 case SCTP_INITIATION: 2023 case SCTP_INITIATION_ACK: { 2024 struct sctp_init_chunk init; 2025 2026 if (!pf_pull_hdr(pd->m, pd->off + chunk_start, &init, 2027 sizeof(init), NULL, NULL, pd->af)) 2028 return (PF_DROP); 2029 2030 /* 2031 * RFC 9620, Section 3.3.2, "The Initiate Tag is allowed to have 2032 * any value except 0." 2033 */ 2034 if (init.init.initiate_tag == 0) 2035 return (PF_DROP); 2036 if (init.init.num_inbound_streams == 0) 2037 return (PF_DROP); 2038 if (init.init.num_outbound_streams == 0) 2039 return (PF_DROP); 2040 if (ntohl(init.init.a_rwnd) < SCTP_MIN_RWND) 2041 return (PF_DROP); 2042 2043 /* 2044 * RFC 9260, Section 3.1, INIT chunks MUST have zero 2045 * verification tag. 2046 */ 2047 if (ch.chunk_type == SCTP_INITIATION && 2048 pd->hdr.sctp.v_tag != 0) 2049 return (PF_DROP); 2050 2051 pd->sctp_initiate_tag = init.init.initiate_tag; 2052 2053 if (ch.chunk_type == SCTP_INITIATION) 2054 pd->sctp_flags |= PFDESC_SCTP_INIT; 2055 else 2056 pd->sctp_flags |= PFDESC_SCTP_INIT_ACK; 2057 2058 ret = pf_multihome_scan_init(pd->off + chunk_start, 2059 ntohs(init.ch.chunk_length), pd); 2060 if (ret != PF_PASS) 2061 return (ret); 2062 2063 break; 2064 } 2065 case SCTP_ABORT_ASSOCIATION: 2066 pd->sctp_flags |= PFDESC_SCTP_ABORT; 2067 break; 2068 case SCTP_SHUTDOWN: 2069 case SCTP_SHUTDOWN_ACK: 2070 pd->sctp_flags |= PFDESC_SCTP_SHUTDOWN; 2071 break; 2072 case SCTP_SHUTDOWN_COMPLETE: 2073 pd->sctp_flags |= PFDESC_SCTP_SHUTDOWN_COMPLETE; 2074 break; 2075 case SCTP_COOKIE_ECHO: 2076 pd->sctp_flags |= PFDESC_SCTP_COOKIE; 2077 break; 2078 case SCTP_COOKIE_ACK: 2079 pd->sctp_flags |= PFDESC_SCTP_COOKIE_ACK; 2080 break; 2081 case SCTP_DATA: 2082 pd->sctp_flags |= PFDESC_SCTP_DATA; 2083 break; 2084 case SCTP_HEARTBEAT_REQUEST: 2085 pd->sctp_flags |= PFDESC_SCTP_HEARTBEAT; 2086 break; 2087 case SCTP_HEARTBEAT_ACK: 2088 pd->sctp_flags |= PFDESC_SCTP_HEARTBEAT_ACK; 2089 break; 2090 case SCTP_ASCONF: 2091 pd->sctp_flags |= PFDESC_SCTP_ASCONF; 2092 2093 ret = pf_multihome_scan_asconf(pd->off + chunk_start, 2094 ntohs(ch.chunk_length), pd); 2095 if (ret != PF_PASS) 2096 return (ret); 2097 break; 2098 default: 2099 pd->sctp_flags |= PFDESC_SCTP_OTHER; 2100 break; 2101 } 2102 } 2103 2104 /* Validate chunk lengths vs. packet length. */ 2105 if (pd->off + chunk_off != pd->tot_len) 2106 return (PF_DROP); 2107 2108 /* 2109 * INIT, INIT_ACK or SHUTDOWN_COMPLETE chunks must always be the only 2110 * one in a packet. 2111 */ 2112 if ((pd->sctp_flags & PFDESC_SCTP_INIT) && 2113 (pd->sctp_flags & ~PFDESC_SCTP_INIT)) 2114 return (PF_DROP); 2115 if ((pd->sctp_flags & PFDESC_SCTP_INIT_ACK) && 2116 (pd->sctp_flags & ~PFDESC_SCTP_INIT_ACK)) 2117 return (PF_DROP); 2118 if ((pd->sctp_flags & PFDESC_SCTP_SHUTDOWN_COMPLETE) && 2119 (pd->sctp_flags & ~PFDESC_SCTP_SHUTDOWN_COMPLETE)) 2120 return (PF_DROP); 2121 2122 return (PF_PASS); 2123 } 2124 2125 int 2126 pf_normalize_sctp(struct pf_pdesc *pd) 2127 { 2128 struct pf_krule *r, *rm = NULL; 2129 struct sctphdr *sh = &pd->hdr.sctp; 2130 u_short reason; 2131 sa_family_t af = pd->af; 2132 int srs; 2133 2134 PF_RULES_RASSERT(); 2135 2136 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr); 2137 /* Check if there any scrub rules. Lack of scrub rules means enforced 2138 * packet normalization operation just like in OpenBSD. */ 2139 srs = (r != NULL); 2140 while (r != NULL) { 2141 pf_counter_u64_add(&r->evaluations, 1); 2142 if (pfi_kkif_match(r->kif, pd->kif) == r->ifnot) 2143 r = r->skip[PF_SKIP_IFP]; 2144 else if (r->direction && r->direction != pd->dir) 2145 r = r->skip[PF_SKIP_DIR]; 2146 else if (r->af && r->af != af) 2147 r = r->skip[PF_SKIP_AF]; 2148 else if (r->proto && r->proto != pd->proto) 2149 r = r->skip[PF_SKIP_PROTO]; 2150 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af, 2151 r->src.neg, pd->kif, M_GETFIB(pd->m))) 2152 r = r->skip[PF_SKIP_SRC_ADDR]; 2153 else if (r->src.port_op && !pf_match_port(r->src.port_op, 2154 r->src.port[0], r->src.port[1], sh->src_port)) 2155 r = r->skip[PF_SKIP_SRC_PORT]; 2156 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af, 2157 r->dst.neg, NULL, M_GETFIB(pd->m))) 2158 r = r->skip[PF_SKIP_DST_ADDR]; 2159 else if (r->dst.port_op && !pf_match_port(r->dst.port_op, 2160 r->dst.port[0], r->dst.port[1], sh->dest_port)) 2161 r = r->skip[PF_SKIP_DST_PORT]; 2162 else { 2163 rm = r; 2164 break; 2165 } 2166 } 2167 2168 if (srs) { 2169 /* With scrub rules present SCTP normalization happens only 2170 * if one of rules has matched and it's not a "no scrub" rule */ 2171 if (rm == NULL || rm->action == PF_NOSCRUB) 2172 return (PF_PASS); 2173 2174 pf_counter_u64_critical_enter(); 2175 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1); 2176 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len); 2177 pf_counter_u64_critical_exit(); 2178 } 2179 2180 /* Verify we're a multiple of 4 bytes long */ 2181 if ((pd->tot_len - pd->off - sizeof(struct sctphdr)) % 4) 2182 goto sctp_drop; 2183 2184 /* INIT chunk needs to be the only chunk */ 2185 if (pd->sctp_flags & PFDESC_SCTP_INIT) 2186 if (pd->sctp_flags & ~PFDESC_SCTP_INIT) 2187 goto sctp_drop; 2188 2189 return (PF_PASS); 2190 2191 sctp_drop: 2192 REASON_SET(&reason, PFRES_NORM); 2193 if (rm != NULL && r->log) 2194 PFLOG_PACKET(PF_DROP, reason, r, NULL, NULL, pd, 2195 1); 2196 2197 return (PF_DROP); 2198 } 2199 2200 #if defined(INET) || defined(INET6) 2201 void 2202 pf_scrub(struct pf_pdesc *pd) 2203 { 2204 2205 struct ip *h = mtod(pd->m, struct ip *); 2206 #ifdef INET6 2207 struct ip6_hdr *h6 = mtod(pd->m, struct ip6_hdr *); 2208 #endif 2209 2210 /* Clear IP_DF if no-df was requested */ 2211 if (pd->af == AF_INET && pd->act.flags & PFSTATE_NODF && 2212 h->ip_off & htons(IP_DF)) 2213 { 2214 u_int16_t ip_off = h->ip_off; 2215 2216 h->ip_off &= htons(~IP_DF); 2217 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0); 2218 } 2219 2220 /* Enforce a minimum ttl, may cause endless packet loops */ 2221 if (pd->af == AF_INET && pd->act.min_ttl && 2222 h->ip_ttl < pd->act.min_ttl) { 2223 u_int16_t ip_ttl = h->ip_ttl; 2224 2225 h->ip_ttl = pd->act.min_ttl; 2226 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_ttl, h->ip_ttl, 0); 2227 } 2228 #ifdef INET6 2229 /* Enforce a minimum ttl, may cause endless packet loops */ 2230 if (pd->af == AF_INET6 && pd->act.min_ttl && 2231 h6->ip6_hlim < pd->act.min_ttl) 2232 h6->ip6_hlim = pd->act.min_ttl; 2233 #endif 2234 /* Enforce tos */ 2235 if (pd->act.flags & PFSTATE_SETTOS) { 2236 switch (pd->af) { 2237 case AF_INET: { 2238 u_int16_t ov, nv; 2239 2240 ov = *(u_int16_t *)h; 2241 h->ip_tos = pd->act.set_tos | (h->ip_tos & IPTOS_ECN_MASK); 2242 nv = *(u_int16_t *)h; 2243 2244 h->ip_sum = pf_cksum_fixup(h->ip_sum, ov, nv, 0); 2245 break; 2246 } 2247 #ifdef INET6 2248 case AF_INET6: 2249 h6->ip6_flow &= IPV6_FLOWLABEL_MASK | IPV6_VERSION_MASK; 2250 h6->ip6_flow |= htonl((pd->act.set_tos | IPV6_ECN(h6)) << 20); 2251 break; 2252 #endif 2253 } 2254 } 2255 2256 /* random-id, but not for fragments */ 2257 #ifdef INET 2258 if (pd->af == AF_INET && 2259 pd->act.flags & PFSTATE_RANDOMID && !(h->ip_off & ~htons(IP_DF))) { 2260 uint16_t ip_id = h->ip_id; 2261 2262 ip_fillid(h); 2263 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_id, h->ip_id, 0); 2264 } 2265 #endif 2266 } 2267 #endif 2268