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 __FBSDID("$FreeBSD$"); 33 34 #include "opt_inet.h" 35 #include "opt_inet6.h" 36 #include "opt_pf.h" 37 38 #include <sys/param.h> 39 #include <sys/kernel.h> 40 #include <sys/lock.h> 41 #include <sys/mbuf.h> 42 #include <sys/mutex.h> 43 #include <sys/refcount.h> 44 #include <sys/socket.h> 45 46 #include <net/if.h> 47 #include <net/vnet.h> 48 #include <net/pfvar.h> 49 #include <net/if_pflog.h> 50 51 #include <netinet/in.h> 52 #include <netinet/ip.h> 53 #include <netinet/ip_var.h> 54 #include <netinet6/ip6_var.h> 55 #include <netinet6/scope6_var.h> 56 #include <netinet/tcp.h> 57 #include <netinet/tcp_fsm.h> 58 #include <netinet/tcp_seq.h> 59 #include <netinet/sctp_constants.h> 60 #include <netinet/sctp_header.h> 61 62 #ifdef INET6 63 #include <netinet/ip6.h> 64 #endif /* INET6 */ 65 66 struct pf_frent { 67 TAILQ_ENTRY(pf_frent) fr_next; 68 struct mbuf *fe_m; 69 uint16_t fe_hdrlen; /* ipv4 header length with ip options 70 ipv6, extension, fragment header */ 71 uint16_t fe_extoff; /* last extension header offset or 0 */ 72 uint16_t fe_len; /* fragment length */ 73 uint16_t fe_off; /* fragment offset */ 74 uint16_t fe_mff; /* more fragment flag */ 75 }; 76 77 struct pf_fragment_cmp { 78 struct pf_addr frc_src; 79 struct pf_addr frc_dst; 80 uint32_t frc_id; 81 sa_family_t frc_af; 82 uint8_t frc_proto; 83 }; 84 85 struct pf_fragment { 86 struct pf_fragment_cmp fr_key; 87 #define fr_src fr_key.frc_src 88 #define fr_dst fr_key.frc_dst 89 #define fr_id fr_key.frc_id 90 #define fr_af fr_key.frc_af 91 #define fr_proto fr_key.frc_proto 92 93 /* pointers to queue element */ 94 struct pf_frent *fr_firstoff[PF_FRAG_ENTRY_POINTS]; 95 /* count entries between pointers */ 96 uint8_t fr_entries[PF_FRAG_ENTRY_POINTS]; 97 RB_ENTRY(pf_fragment) fr_entry; 98 TAILQ_ENTRY(pf_fragment) frag_next; 99 uint32_t fr_timeout; 100 uint16_t fr_maxlen; /* maximum length of single fragment */ 101 u_int16_t fr_holes; /* number of holes in the queue */ 102 TAILQ_HEAD(pf_fragq, pf_frent) fr_queue; 103 }; 104 105 struct pf_fragment_tag { 106 uint16_t ft_hdrlen; /* header length of reassembled pkt */ 107 uint16_t ft_extoff; /* last extension header offset or 0 */ 108 uint16_t ft_maxlen; /* maximum fragment payload length */ 109 uint32_t ft_id; /* fragment id */ 110 }; 111 112 VNET_DEFINE_STATIC(struct mtx, pf_frag_mtx); 113 #define V_pf_frag_mtx VNET(pf_frag_mtx) 114 #define PF_FRAG_LOCK() mtx_lock(&V_pf_frag_mtx) 115 #define PF_FRAG_UNLOCK() mtx_unlock(&V_pf_frag_mtx) 116 #define PF_FRAG_ASSERT() mtx_assert(&V_pf_frag_mtx, MA_OWNED) 117 118 VNET_DEFINE(uma_zone_t, pf_state_scrub_z); /* XXX: shared with pfsync */ 119 120 VNET_DEFINE_STATIC(uma_zone_t, pf_frent_z); 121 #define V_pf_frent_z VNET(pf_frent_z) 122 VNET_DEFINE_STATIC(uma_zone_t, pf_frag_z); 123 #define V_pf_frag_z VNET(pf_frag_z) 124 125 TAILQ_HEAD(pf_fragqueue, pf_fragment); 126 TAILQ_HEAD(pf_cachequeue, pf_fragment); 127 VNET_DEFINE_STATIC(struct pf_fragqueue, pf_fragqueue); 128 #define V_pf_fragqueue VNET(pf_fragqueue) 129 RB_HEAD(pf_frag_tree, pf_fragment); 130 VNET_DEFINE_STATIC(struct pf_frag_tree, pf_frag_tree); 131 #define V_pf_frag_tree VNET(pf_frag_tree) 132 static int pf_frag_compare(struct pf_fragment *, 133 struct pf_fragment *); 134 static RB_PROTOTYPE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare); 135 static RB_GENERATE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare); 136 137 static void pf_flush_fragments(void); 138 static void pf_free_fragment(struct pf_fragment *); 139 static void pf_remove_fragment(struct pf_fragment *); 140 141 static struct pf_frent *pf_create_fragment(u_short *); 142 static int pf_frent_holes(struct pf_frent *frent); 143 static struct pf_fragment *pf_find_fragment(struct pf_fragment_cmp *key, 144 struct pf_frag_tree *tree); 145 static inline int pf_frent_index(struct pf_frent *); 146 static int pf_frent_insert(struct pf_fragment *, 147 struct pf_frent *, struct pf_frent *); 148 void pf_frent_remove(struct pf_fragment *, 149 struct pf_frent *); 150 struct pf_frent *pf_frent_previous(struct pf_fragment *, 151 struct pf_frent *); 152 static struct pf_fragment *pf_fillup_fragment(struct pf_fragment_cmp *, 153 struct pf_frent *, u_short *); 154 static struct mbuf *pf_join_fragment(struct pf_fragment *); 155 #ifdef INET 156 static int pf_reassemble(struct mbuf **, struct ip *, int, u_short *); 157 #endif /* INET */ 158 #ifdef INET6 159 static int pf_reassemble6(struct mbuf **, struct ip6_hdr *, 160 struct ip6_frag *, uint16_t, uint16_t, u_short *); 161 #endif /* INET6 */ 162 163 #define DPFPRINTF(x) do { \ 164 if (V_pf_status.debug >= PF_DEBUG_MISC) { \ 165 printf("%s: ", __func__); \ 166 printf x ; \ 167 } \ 168 } while(0) 169 170 #ifdef INET 171 static void 172 pf_ip2key(struct ip *ip, int dir, struct pf_fragment_cmp *key) 173 { 174 175 key->frc_src.v4 = ip->ip_src; 176 key->frc_dst.v4 = ip->ip_dst; 177 key->frc_af = AF_INET; 178 key->frc_proto = ip->ip_p; 179 key->frc_id = ip->ip_id; 180 } 181 #endif /* INET */ 182 183 void 184 pf_normalize_init(void) 185 { 186 187 V_pf_frag_z = uma_zcreate("pf frags", sizeof(struct pf_fragment), 188 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 189 V_pf_frent_z = uma_zcreate("pf frag entries", sizeof(struct pf_frent), 190 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 191 V_pf_state_scrub_z = uma_zcreate("pf state scrubs", 192 sizeof(struct pf_state_scrub), NULL, NULL, NULL, NULL, 193 UMA_ALIGN_PTR, 0); 194 195 mtx_init(&V_pf_frag_mtx, "pf fragments", NULL, MTX_DEF); 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 } 204 205 void 206 pf_normalize_cleanup(void) 207 { 208 209 uma_zdestroy(V_pf_state_scrub_z); 210 uma_zdestroy(V_pf_frent_z); 211 uma_zdestroy(V_pf_frag_z); 212 213 mtx_destroy(&V_pf_frag_mtx); 214 } 215 216 static int 217 pf_frag_compare(struct pf_fragment *a, struct pf_fragment *b) 218 { 219 int diff; 220 221 if ((diff = a->fr_id - b->fr_id) != 0) 222 return (diff); 223 if ((diff = a->fr_proto - b->fr_proto) != 0) 224 return (diff); 225 if ((diff = a->fr_af - b->fr_af) != 0) 226 return (diff); 227 if ((diff = pf_addr_cmp(&a->fr_src, &b->fr_src, a->fr_af)) != 0) 228 return (diff); 229 if ((diff = pf_addr_cmp(&a->fr_dst, &b->fr_dst, a->fr_af)) != 0) 230 return (diff); 231 return (0); 232 } 233 234 void 235 pf_purge_expired_fragments(void) 236 { 237 u_int32_t expire = time_uptime - 238 V_pf_default_rule.timeout[PFTM_FRAG]; 239 240 pf_purge_fragments(expire); 241 } 242 243 void 244 pf_purge_fragments(uint32_t expire) 245 { 246 struct pf_fragment *frag; 247 248 PF_FRAG_LOCK(); 249 while ((frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue)) != NULL) { 250 if (frag->fr_timeout > expire) 251 break; 252 253 DPFPRINTF(("expiring %d(%p)\n", frag->fr_id, frag)); 254 pf_free_fragment(frag); 255 } 256 257 PF_FRAG_UNLOCK(); 258 } 259 260 /* 261 * Try to flush old fragments to make space for new ones 262 */ 263 static void 264 pf_flush_fragments(void) 265 { 266 struct pf_fragment *frag; 267 int goal; 268 269 PF_FRAG_ASSERT(); 270 271 goal = uma_zone_get_cur(V_pf_frent_z) * 9 / 10; 272 DPFPRINTF(("trying to free %d frag entriess\n", goal)); 273 while (goal < uma_zone_get_cur(V_pf_frent_z)) { 274 frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue); 275 if (frag) 276 pf_free_fragment(frag); 277 else 278 break; 279 } 280 } 281 282 /* Frees the fragments and all associated entries */ 283 static void 284 pf_free_fragment(struct pf_fragment *frag) 285 { 286 struct pf_frent *frent; 287 288 PF_FRAG_ASSERT(); 289 290 /* Free all fragments */ 291 for (frent = TAILQ_FIRST(&frag->fr_queue); frent; 292 frent = TAILQ_FIRST(&frag->fr_queue)) { 293 TAILQ_REMOVE(&frag->fr_queue, frent, fr_next); 294 295 m_freem(frent->fe_m); 296 uma_zfree(V_pf_frent_z, frent); 297 } 298 299 pf_remove_fragment(frag); 300 } 301 302 static struct pf_fragment * 303 pf_find_fragment(struct pf_fragment_cmp *key, struct pf_frag_tree *tree) 304 { 305 struct pf_fragment *frag; 306 307 PF_FRAG_ASSERT(); 308 309 frag = RB_FIND(pf_frag_tree, tree, (struct pf_fragment *)key); 310 if (frag != NULL) { 311 /* XXX Are we sure we want to update the timeout? */ 312 frag->fr_timeout = time_uptime; 313 TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next); 314 TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next); 315 } 316 317 return (frag); 318 } 319 320 /* Removes a fragment from the fragment queue and frees the fragment */ 321 static void 322 pf_remove_fragment(struct pf_fragment *frag) 323 { 324 325 PF_FRAG_ASSERT(); 326 KASSERT(frag, ("frag != NULL")); 327 328 RB_REMOVE(pf_frag_tree, &V_pf_frag_tree, frag); 329 TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next); 330 uma_zfree(V_pf_frag_z, frag); 331 } 332 333 static struct pf_frent * 334 pf_create_fragment(u_short *reason) 335 { 336 struct pf_frent *frent; 337 338 PF_FRAG_ASSERT(); 339 340 frent = uma_zalloc(V_pf_frent_z, M_NOWAIT); 341 if (frent == NULL) { 342 pf_flush_fragments(); 343 frent = uma_zalloc(V_pf_frent_z, M_NOWAIT); 344 if (frent == NULL) { 345 REASON_SET(reason, PFRES_MEMORY); 346 return (NULL); 347 } 348 } 349 350 return (frent); 351 } 352 353 /* 354 * Calculate the additional holes that were created in the fragment 355 * queue by inserting this fragment. A fragment in the middle 356 * creates one more hole by splitting. For each connected side, 357 * it loses one hole. 358 * Fragment entry must be in the queue when calling this function. 359 */ 360 static int 361 pf_frent_holes(struct pf_frent *frent) 362 { 363 struct pf_frent *prev = TAILQ_PREV(frent, pf_fragq, fr_next); 364 struct pf_frent *next = TAILQ_NEXT(frent, fr_next); 365 int holes = 1; 366 367 if (prev == NULL) { 368 if (frent->fe_off == 0) 369 holes--; 370 } else { 371 KASSERT(frent->fe_off != 0, ("frent->fe_off != 0")); 372 if (frent->fe_off == prev->fe_off + prev->fe_len) 373 holes--; 374 } 375 if (next == NULL) { 376 if (!frent->fe_mff) 377 holes--; 378 } else { 379 KASSERT(frent->fe_mff, ("frent->fe_mff")); 380 if (next->fe_off == frent->fe_off + frent->fe_len) 381 holes--; 382 } 383 return holes; 384 } 385 386 static inline int 387 pf_frent_index(struct pf_frent *frent) 388 { 389 /* 390 * We have an array of 16 entry points to the queue. A full size 391 * 65535 octet IP packet can have 8192 fragments. So the queue 392 * traversal length is at most 512 and at most 16 entry points are 393 * checked. We need 128 additional bytes on a 64 bit architecture. 394 */ 395 CTASSERT(((u_int16_t)0xffff &~ 7) / (0x10000 / PF_FRAG_ENTRY_POINTS) == 396 16 - 1); 397 CTASSERT(((u_int16_t)0xffff >> 3) / PF_FRAG_ENTRY_POINTS == 512 - 1); 398 399 return frent->fe_off / (0x10000 / PF_FRAG_ENTRY_POINTS); 400 } 401 402 static int 403 pf_frent_insert(struct pf_fragment *frag, struct pf_frent *frent, 404 struct pf_frent *prev) 405 { 406 int index; 407 408 CTASSERT(PF_FRAG_ENTRY_LIMIT <= 0xff); 409 410 /* 411 * A packet has at most 65536 octets. With 16 entry points, each one 412 * spawns 4096 octets. We limit these to 64 fragments each, which 413 * means on average every fragment must have at least 64 octets. 414 */ 415 index = pf_frent_index(frent); 416 if (frag->fr_entries[index] >= PF_FRAG_ENTRY_LIMIT) 417 return ENOBUFS; 418 frag->fr_entries[index]++; 419 420 if (prev == NULL) { 421 TAILQ_INSERT_HEAD(&frag->fr_queue, frent, fr_next); 422 } else { 423 KASSERT(prev->fe_off + prev->fe_len <= frent->fe_off, 424 ("overlapping fragment")); 425 TAILQ_INSERT_AFTER(&frag->fr_queue, prev, frent, fr_next); 426 } 427 428 if (frag->fr_firstoff[index] == NULL) { 429 KASSERT(prev == NULL || pf_frent_index(prev) < index, 430 ("prev == NULL || pf_frent_index(pref) < index")); 431 frag->fr_firstoff[index] = frent; 432 } else { 433 if (frent->fe_off < frag->fr_firstoff[index]->fe_off) { 434 KASSERT(prev == NULL || pf_frent_index(prev) < index, 435 ("prev == NULL || pf_frent_index(pref) < index")); 436 frag->fr_firstoff[index] = frent; 437 } else { 438 KASSERT(prev != NULL, ("prev != NULL")); 439 KASSERT(pf_frent_index(prev) == index, 440 ("pf_frent_index(prev) == index")); 441 } 442 } 443 444 frag->fr_holes += pf_frent_holes(frent); 445 446 return 0; 447 } 448 449 void 450 pf_frent_remove(struct pf_fragment *frag, struct pf_frent *frent) 451 { 452 #ifdef INVARIANTS 453 struct pf_frent *prev = TAILQ_PREV(frent, pf_fragq, fr_next); 454 #endif 455 struct pf_frent *next = TAILQ_NEXT(frent, fr_next); 456 int index; 457 458 frag->fr_holes -= pf_frent_holes(frent); 459 460 index = pf_frent_index(frent); 461 KASSERT(frag->fr_firstoff[index] != NULL, ("frent not found")); 462 if (frag->fr_firstoff[index]->fe_off == frent->fe_off) { 463 if (next == NULL) { 464 frag->fr_firstoff[index] = NULL; 465 } else { 466 KASSERT(frent->fe_off + frent->fe_len <= next->fe_off, 467 ("overlapping fragment")); 468 if (pf_frent_index(next) == index) { 469 frag->fr_firstoff[index] = next; 470 } else { 471 frag->fr_firstoff[index] = NULL; 472 } 473 } 474 } else { 475 KASSERT(frag->fr_firstoff[index]->fe_off < frent->fe_off, 476 ("frag->fr_firstoff[index]->fe_off < frent->fe_off")); 477 KASSERT(prev != NULL, ("prev != NULL")); 478 KASSERT(prev->fe_off + prev->fe_len <= frent->fe_off, 479 ("overlapping fragment")); 480 KASSERT(pf_frent_index(prev) == index, 481 ("pf_frent_index(prev) == index")); 482 } 483 484 TAILQ_REMOVE(&frag->fr_queue, frent, fr_next); 485 486 KASSERT(frag->fr_entries[index] > 0, ("No fragments remaining")); 487 frag->fr_entries[index]--; 488 } 489 490 struct pf_frent * 491 pf_frent_previous(struct pf_fragment *frag, struct pf_frent *frent) 492 { 493 struct pf_frent *prev, *next; 494 int index; 495 496 /* 497 * If there are no fragments after frag, take the final one. Assume 498 * that the global queue is not empty. 499 */ 500 prev = TAILQ_LAST(&frag->fr_queue, pf_fragq); 501 KASSERT(prev != NULL, ("prev != NULL")); 502 if (prev->fe_off <= frent->fe_off) 503 return prev; 504 /* 505 * We want to find a fragment entry that is before frag, but still 506 * close to it. Find the first fragment entry that is in the same 507 * entry point or in the first entry point after that. As we have 508 * already checked that there are entries behind frag, this will 509 * succeed. 510 */ 511 for (index = pf_frent_index(frent); index < PF_FRAG_ENTRY_POINTS; 512 index++) { 513 prev = frag->fr_firstoff[index]; 514 if (prev != NULL) 515 break; 516 } 517 KASSERT(prev != NULL, ("prev != NULL")); 518 /* 519 * In prev we may have a fragment from the same entry point that is 520 * before frent, or one that is just one position behind frent. 521 * In the latter case, we go back one step and have the predecessor. 522 * There may be none if the new fragment will be the first one. 523 */ 524 if (prev->fe_off > frent->fe_off) { 525 prev = TAILQ_PREV(prev, pf_fragq, fr_next); 526 if (prev == NULL) 527 return NULL; 528 KASSERT(prev->fe_off <= frent->fe_off, 529 ("prev->fe_off <= frent->fe_off")); 530 return prev; 531 } 532 /* 533 * In prev is the first fragment of the entry point. The offset 534 * of frag is behind it. Find the closest previous fragment. 535 */ 536 for (next = TAILQ_NEXT(prev, fr_next); next != NULL; 537 next = TAILQ_NEXT(next, fr_next)) { 538 if (next->fe_off > frent->fe_off) 539 break; 540 prev = next; 541 } 542 return prev; 543 } 544 545 static struct pf_fragment * 546 pf_fillup_fragment(struct pf_fragment_cmp *key, struct pf_frent *frent, 547 u_short *reason) 548 { 549 struct pf_frent *after, *next, *prev; 550 struct pf_fragment *frag; 551 uint16_t total; 552 int old_index, new_index; 553 554 PF_FRAG_ASSERT(); 555 556 /* No empty fragments. */ 557 if (frent->fe_len == 0) { 558 DPFPRINTF(("bad fragment: len 0\n")); 559 goto bad_fragment; 560 } 561 562 /* All fragments are 8 byte aligned. */ 563 if (frent->fe_mff && (frent->fe_len & 0x7)) { 564 DPFPRINTF(("bad fragment: mff and len %d\n", frent->fe_len)); 565 goto bad_fragment; 566 } 567 568 /* Respect maximum length, IP_MAXPACKET == IPV6_MAXPACKET. */ 569 if (frent->fe_off + frent->fe_len > IP_MAXPACKET) { 570 DPFPRINTF(("bad fragment: max packet %d\n", 571 frent->fe_off + frent->fe_len)); 572 goto bad_fragment; 573 } 574 575 DPFPRINTF((key->frc_af == AF_INET ? 576 "reass frag %d @ %d-%d\n" : "reass frag %#08x @ %d-%d\n", 577 key->frc_id, frent->fe_off, frent->fe_off + frent->fe_len)); 578 579 /* Fully buffer all of the fragments in this fragment queue. */ 580 frag = pf_find_fragment(key, &V_pf_frag_tree); 581 582 /* Create a new reassembly queue for this packet. */ 583 if (frag == NULL) { 584 frag = uma_zalloc(V_pf_frag_z, M_NOWAIT); 585 if (frag == NULL) { 586 pf_flush_fragments(); 587 frag = uma_zalloc(V_pf_frag_z, M_NOWAIT); 588 if (frag == NULL) { 589 REASON_SET(reason, PFRES_MEMORY); 590 goto drop_fragment; 591 } 592 } 593 594 *(struct pf_fragment_cmp *)frag = *key; 595 memset(frag->fr_firstoff, 0, sizeof(frag->fr_firstoff)); 596 memset(frag->fr_entries, 0, sizeof(frag->fr_entries)); 597 frag->fr_timeout = time_uptime; 598 frag->fr_maxlen = frent->fe_len; 599 frag->fr_holes = 1; 600 TAILQ_INIT(&frag->fr_queue); 601 602 RB_INSERT(pf_frag_tree, &V_pf_frag_tree, frag); 603 TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next); 604 605 /* We do not have a previous fragment, cannot fail. */ 606 pf_frent_insert(frag, frent, NULL); 607 608 return (frag); 609 } 610 611 KASSERT(!TAILQ_EMPTY(&frag->fr_queue), ("!TAILQ_EMPTY()->fr_queue")); 612 613 /* Remember maximum fragment len for refragmentation. */ 614 if (frent->fe_len > frag->fr_maxlen) 615 frag->fr_maxlen = frent->fe_len; 616 617 /* Maximum data we have seen already. */ 618 total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off + 619 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len; 620 621 /* Non terminal fragments must have more fragments flag. */ 622 if (frent->fe_off + frent->fe_len < total && !frent->fe_mff) 623 goto bad_fragment; 624 625 /* Check if we saw the last fragment already. */ 626 if (!TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_mff) { 627 if (frent->fe_off + frent->fe_len > total || 628 (frent->fe_off + frent->fe_len == total && frent->fe_mff)) 629 goto bad_fragment; 630 } else { 631 if (frent->fe_off + frent->fe_len == total && !frent->fe_mff) 632 goto bad_fragment; 633 } 634 635 /* Find neighbors for newly inserted fragment */ 636 prev = pf_frent_previous(frag, frent); 637 if (prev == NULL) { 638 after = TAILQ_FIRST(&frag->fr_queue); 639 KASSERT(after != NULL, ("after != NULL")); 640 } else { 641 after = TAILQ_NEXT(prev, fr_next); 642 } 643 644 if (prev != NULL && prev->fe_off + prev->fe_len > frent->fe_off) { 645 uint16_t precut; 646 647 precut = prev->fe_off + prev->fe_len - frent->fe_off; 648 if (precut >= frent->fe_len) 649 goto bad_fragment; 650 DPFPRINTF(("overlap -%d\n", precut)); 651 m_adj(frent->fe_m, precut); 652 frent->fe_off += precut; 653 frent->fe_len -= precut; 654 } 655 656 for (; after != NULL && frent->fe_off + frent->fe_len > after->fe_off; 657 after = next) { 658 uint16_t aftercut; 659 660 aftercut = frent->fe_off + frent->fe_len - after->fe_off; 661 DPFPRINTF(("adjust overlap %d\n", aftercut)); 662 if (aftercut < after->fe_len) { 663 m_adj(after->fe_m, aftercut); 664 old_index = pf_frent_index(after); 665 after->fe_off += aftercut; 666 after->fe_len -= aftercut; 667 new_index = pf_frent_index(after); 668 if (old_index != new_index) { 669 DPFPRINTF(("frag index %d, new %d", 670 old_index, new_index)); 671 /* Fragment switched queue as fe_off changed */ 672 after->fe_off -= aftercut; 673 after->fe_len += aftercut; 674 /* Remove restored fragment from old queue */ 675 pf_frent_remove(frag, after); 676 after->fe_off += aftercut; 677 after->fe_len -= aftercut; 678 /* Insert into correct queue */ 679 if (pf_frent_insert(frag, after, prev)) { 680 DPFPRINTF( 681 ("fragment requeue limit exceeded")); 682 m_freem(after->fe_m); 683 uma_zfree(V_pf_frent_z, after); 684 /* There is not way to recover */ 685 goto bad_fragment; 686 } 687 } 688 break; 689 } 690 691 /* This fragment is completely overlapped, lose it. */ 692 next = TAILQ_NEXT(after, fr_next); 693 pf_frent_remove(frag, after); 694 m_freem(after->fe_m); 695 uma_zfree(V_pf_frent_z, after); 696 } 697 698 /* If part of the queue gets too long, there is not way to recover. */ 699 if (pf_frent_insert(frag, frent, prev)) { 700 DPFPRINTF(("fragment queue limit exceeded\n")); 701 goto bad_fragment; 702 } 703 704 return (frag); 705 706 bad_fragment: 707 REASON_SET(reason, PFRES_FRAG); 708 drop_fragment: 709 uma_zfree(V_pf_frent_z, frent); 710 return (NULL); 711 } 712 713 static struct mbuf * 714 pf_join_fragment(struct pf_fragment *frag) 715 { 716 struct mbuf *m, *m2; 717 struct pf_frent *frent, *next; 718 719 frent = TAILQ_FIRST(&frag->fr_queue); 720 next = TAILQ_NEXT(frent, fr_next); 721 722 m = frent->fe_m; 723 m_adj(m, (frent->fe_hdrlen + frent->fe_len) - m->m_pkthdr.len); 724 uma_zfree(V_pf_frent_z, frent); 725 for (frent = next; frent != NULL; frent = next) { 726 next = TAILQ_NEXT(frent, fr_next); 727 728 m2 = frent->fe_m; 729 /* Strip off ip header. */ 730 m_adj(m2, frent->fe_hdrlen); 731 /* Strip off any trailing bytes. */ 732 m_adj(m2, frent->fe_len - m2->m_pkthdr.len); 733 734 uma_zfree(V_pf_frent_z, frent); 735 m_cat(m, m2); 736 } 737 738 /* Remove from fragment queue. */ 739 pf_remove_fragment(frag); 740 741 return (m); 742 } 743 744 #ifdef INET 745 static int 746 pf_reassemble(struct mbuf **m0, struct ip *ip, int dir, u_short *reason) 747 { 748 struct mbuf *m = *m0; 749 struct pf_frent *frent; 750 struct pf_fragment *frag; 751 struct pf_fragment_cmp key; 752 uint16_t total, hdrlen; 753 754 /* Get an entry for the fragment queue */ 755 if ((frent = pf_create_fragment(reason)) == NULL) 756 return (PF_DROP); 757 758 frent->fe_m = m; 759 frent->fe_hdrlen = ip->ip_hl << 2; 760 frent->fe_extoff = 0; 761 frent->fe_len = ntohs(ip->ip_len) - (ip->ip_hl << 2); 762 frent->fe_off = (ntohs(ip->ip_off) & IP_OFFMASK) << 3; 763 frent->fe_mff = ntohs(ip->ip_off) & IP_MF; 764 765 pf_ip2key(ip, dir, &key); 766 767 if ((frag = pf_fillup_fragment(&key, frent, reason)) == NULL) 768 return (PF_DROP); 769 770 /* The mbuf is part of the fragment entry, no direct free or access */ 771 m = *m0 = NULL; 772 773 if (frag->fr_holes) { 774 DPFPRINTF(("frag %d, holes %d\n", frag->fr_id, frag->fr_holes)); 775 return (PF_PASS); /* drop because *m0 is NULL, no error */ 776 } 777 778 /* We have all the data */ 779 frent = TAILQ_FIRST(&frag->fr_queue); 780 KASSERT(frent != NULL, ("frent != NULL")); 781 total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off + 782 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len; 783 hdrlen = frent->fe_hdrlen; 784 785 m = *m0 = pf_join_fragment(frag); 786 frag = NULL; 787 788 if (m->m_flags & M_PKTHDR) { 789 int plen = 0; 790 for (m = *m0; m; m = m->m_next) 791 plen += m->m_len; 792 m = *m0; 793 m->m_pkthdr.len = plen; 794 } 795 796 ip = mtod(m, struct ip *); 797 ip->ip_sum = pf_cksum_fixup(ip->ip_sum, ip->ip_len, 798 htons(hdrlen + total), 0); 799 ip->ip_len = htons(hdrlen + total); 800 ip->ip_sum = pf_cksum_fixup(ip->ip_sum, ip->ip_off, 801 ip->ip_off & ~(IP_MF|IP_OFFMASK), 0); 802 ip->ip_off &= ~(IP_MF|IP_OFFMASK); 803 804 if (hdrlen + total > IP_MAXPACKET) { 805 DPFPRINTF(("drop: too big: %d\n", total)); 806 ip->ip_len = 0; 807 REASON_SET(reason, PFRES_SHORT); 808 /* PF_DROP requires a valid mbuf *m0 in pf_test() */ 809 return (PF_DROP); 810 } 811 812 DPFPRINTF(("complete: %p(%d)\n", m, ntohs(ip->ip_len))); 813 return (PF_PASS); 814 } 815 #endif /* INET */ 816 817 #ifdef INET6 818 static int 819 pf_reassemble6(struct mbuf **m0, struct ip6_hdr *ip6, struct ip6_frag *fraghdr, 820 uint16_t hdrlen, uint16_t extoff, u_short *reason) 821 { 822 struct mbuf *m = *m0; 823 struct pf_frent *frent; 824 struct pf_fragment *frag; 825 struct pf_fragment_cmp key; 826 struct m_tag *mtag; 827 struct pf_fragment_tag *ftag; 828 int off; 829 uint32_t frag_id; 830 uint16_t total, maxlen; 831 uint8_t proto; 832 833 PF_FRAG_LOCK(); 834 835 /* Get an entry for the fragment queue. */ 836 if ((frent = pf_create_fragment(reason)) == NULL) { 837 PF_FRAG_UNLOCK(); 838 return (PF_DROP); 839 } 840 841 frent->fe_m = m; 842 frent->fe_hdrlen = hdrlen; 843 frent->fe_extoff = extoff; 844 frent->fe_len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - hdrlen; 845 frent->fe_off = ntohs(fraghdr->ip6f_offlg & IP6F_OFF_MASK); 846 frent->fe_mff = fraghdr->ip6f_offlg & IP6F_MORE_FRAG; 847 848 key.frc_src.v6 = ip6->ip6_src; 849 key.frc_dst.v6 = ip6->ip6_dst; 850 key.frc_af = AF_INET6; 851 /* Only the first fragment's protocol is relevant. */ 852 key.frc_proto = 0; 853 key.frc_id = fraghdr->ip6f_ident; 854 855 if ((frag = pf_fillup_fragment(&key, frent, reason)) == NULL) { 856 PF_FRAG_UNLOCK(); 857 return (PF_DROP); 858 } 859 860 /* The mbuf is part of the fragment entry, no direct free or access. */ 861 m = *m0 = NULL; 862 863 if (frag->fr_holes) { 864 DPFPRINTF(("frag %d, holes %d\n", frag->fr_id, 865 frag->fr_holes)); 866 PF_FRAG_UNLOCK(); 867 return (PF_PASS); /* Drop because *m0 is NULL, no error. */ 868 } 869 870 /* We have all the data. */ 871 frent = TAILQ_FIRST(&frag->fr_queue); 872 KASSERT(frent != NULL, ("frent != NULL")); 873 extoff = frent->fe_extoff; 874 maxlen = frag->fr_maxlen; 875 frag_id = frag->fr_id; 876 total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off + 877 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len; 878 hdrlen = frent->fe_hdrlen - sizeof(struct ip6_frag); 879 880 m = *m0 = pf_join_fragment(frag); 881 frag = NULL; 882 883 PF_FRAG_UNLOCK(); 884 885 /* Take protocol from first fragment header. */ 886 m = m_getptr(m, hdrlen + offsetof(struct ip6_frag, ip6f_nxt), &off); 887 KASSERT(m, ("%s: short mbuf chain", __func__)); 888 proto = *(mtod(m, uint8_t *) + off); 889 m = *m0; 890 891 /* Delete frag6 header */ 892 if (ip6_deletefraghdr(m, hdrlen, M_NOWAIT) != 0) 893 goto fail; 894 895 if (m->m_flags & M_PKTHDR) { 896 int plen = 0; 897 for (m = *m0; m; m = m->m_next) 898 plen += m->m_len; 899 m = *m0; 900 m->m_pkthdr.len = plen; 901 } 902 903 if ((mtag = m_tag_get(PACKET_TAG_PF_REASSEMBLED, 904 sizeof(struct pf_fragment_tag), M_NOWAIT)) == NULL) 905 goto fail; 906 ftag = (struct pf_fragment_tag *)(mtag + 1); 907 ftag->ft_hdrlen = hdrlen; 908 ftag->ft_extoff = extoff; 909 ftag->ft_maxlen = maxlen; 910 ftag->ft_id = frag_id; 911 m_tag_prepend(m, mtag); 912 913 ip6 = mtod(m, struct ip6_hdr *); 914 ip6->ip6_plen = htons(hdrlen - sizeof(struct ip6_hdr) + total); 915 if (extoff) { 916 /* Write protocol into next field of last extension header. */ 917 m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt), 918 &off); 919 KASSERT(m, ("%s: short mbuf chain", __func__)); 920 *(mtod(m, char *) + off) = proto; 921 m = *m0; 922 } else 923 ip6->ip6_nxt = proto; 924 925 if (hdrlen - sizeof(struct ip6_hdr) + total > IPV6_MAXPACKET) { 926 DPFPRINTF(("drop: too big: %d\n", total)); 927 ip6->ip6_plen = 0; 928 REASON_SET(reason, PFRES_SHORT); 929 /* PF_DROP requires a valid mbuf *m0 in pf_test6(). */ 930 return (PF_DROP); 931 } 932 933 DPFPRINTF(("complete: %p(%d)\n", m, ntohs(ip6->ip6_plen))); 934 return (PF_PASS); 935 936 fail: 937 REASON_SET(reason, PFRES_MEMORY); 938 /* PF_DROP requires a valid mbuf *m0 in pf_test6(), will free later. */ 939 return (PF_DROP); 940 } 941 #endif /* INET6 */ 942 943 #ifdef INET6 944 int 945 pf_refragment6(struct ifnet *ifp, struct mbuf **m0, struct m_tag *mtag, 946 bool forward) 947 { 948 struct mbuf *m = *m0, *t; 949 struct ip6_hdr *hdr; 950 struct pf_fragment_tag *ftag = (struct pf_fragment_tag *)(mtag + 1); 951 struct pf_pdesc pd; 952 uint32_t frag_id; 953 uint16_t hdrlen, extoff, maxlen; 954 uint8_t proto; 955 int error, action; 956 957 hdrlen = ftag->ft_hdrlen; 958 extoff = ftag->ft_extoff; 959 maxlen = ftag->ft_maxlen; 960 frag_id = ftag->ft_id; 961 m_tag_delete(m, mtag); 962 mtag = NULL; 963 ftag = NULL; 964 965 if (extoff) { 966 int off; 967 968 /* Use protocol from next field of last extension header */ 969 m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt), 970 &off); 971 KASSERT((m != NULL), ("pf_refragment6: short mbuf chain")); 972 proto = *(mtod(m, uint8_t *) + off); 973 *(mtod(m, char *) + off) = IPPROTO_FRAGMENT; 974 m = *m0; 975 } else { 976 hdr = mtod(m, struct ip6_hdr *); 977 proto = hdr->ip6_nxt; 978 hdr->ip6_nxt = IPPROTO_FRAGMENT; 979 } 980 981 /* In case of link-local traffic we'll need a scope set. */ 982 hdr = mtod(m, struct ip6_hdr *); 983 984 in6_setscope(&hdr->ip6_src, ifp, NULL); 985 in6_setscope(&hdr->ip6_dst, ifp, NULL); 986 987 /* The MTU must be a multiple of 8 bytes, or we risk doing the 988 * fragmentation wrong. */ 989 maxlen = maxlen & ~7; 990 991 /* 992 * Maxlen may be less than 8 if there was only a single 993 * fragment. As it was fragmented before, add a fragment 994 * header also for a single fragment. If total or maxlen 995 * is less than 8, ip6_fragment() will return EMSGSIZE and 996 * we drop the packet. 997 */ 998 error = ip6_fragment(ifp, m, hdrlen, proto, maxlen, frag_id); 999 m = (*m0)->m_nextpkt; 1000 (*m0)->m_nextpkt = NULL; 1001 if (error == 0) { 1002 /* The first mbuf contains the unfragmented packet. */ 1003 m_freem(*m0); 1004 *m0 = NULL; 1005 action = PF_PASS; 1006 } else { 1007 /* Drop expects an mbuf to free. */ 1008 DPFPRINTF(("refragment error %d\n", error)); 1009 action = PF_DROP; 1010 } 1011 for (; m; m = t) { 1012 t = m->m_nextpkt; 1013 m->m_nextpkt = NULL; 1014 m->m_flags |= M_SKIP_FIREWALL; 1015 memset(&pd, 0, sizeof(pd)); 1016 pd.pf_mtag = pf_find_mtag(m); 1017 if (error == 0) 1018 if (forward) { 1019 MPASS(m->m_pkthdr.rcvif != NULL); 1020 ip6_forward(m, 0); 1021 } else { 1022 (void)ip6_output(m, NULL, NULL, 0, NULL, NULL, 1023 NULL); 1024 } 1025 else 1026 m_freem(m); 1027 } 1028 1029 return (action); 1030 } 1031 #endif /* INET6 */ 1032 1033 #ifdef INET 1034 int 1035 pf_normalize_ip(struct mbuf **m0, struct pfi_kkif *kif, u_short *reason, 1036 struct pf_pdesc *pd) 1037 { 1038 struct mbuf *m = *m0; 1039 struct pf_krule *r; 1040 struct ip *h = mtod(m, struct ip *); 1041 int mff = (ntohs(h->ip_off) & IP_MF); 1042 int hlen = h->ip_hl << 2; 1043 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3; 1044 u_int16_t max; 1045 int ip_len; 1046 int tag = -1; 1047 int verdict; 1048 int srs; 1049 1050 PF_RULES_RASSERT(); 1051 1052 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr); 1053 /* Check if there any scrub rules. Lack of scrub rules means enforced 1054 * packet normalization operation just like in OpenBSD. */ 1055 srs = (r != NULL); 1056 while (r != NULL) { 1057 pf_counter_u64_add(&r->evaluations, 1); 1058 if (pfi_kkif_match(r->kif, kif) == r->ifnot) 1059 r = r->skip[PF_SKIP_IFP].ptr; 1060 else if (r->direction && r->direction != pd->dir) 1061 r = r->skip[PF_SKIP_DIR].ptr; 1062 else if (r->af && r->af != AF_INET) 1063 r = r->skip[PF_SKIP_AF].ptr; 1064 else if (r->proto && r->proto != h->ip_p) 1065 r = r->skip[PF_SKIP_PROTO].ptr; 1066 else if (PF_MISMATCHAW(&r->src.addr, 1067 (struct pf_addr *)&h->ip_src.s_addr, AF_INET, 1068 r->src.neg, kif, M_GETFIB(m))) 1069 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 1070 else if (PF_MISMATCHAW(&r->dst.addr, 1071 (struct pf_addr *)&h->ip_dst.s_addr, AF_INET, 1072 r->dst.neg, NULL, M_GETFIB(m))) 1073 r = r->skip[PF_SKIP_DST_ADDR].ptr; 1074 else if (r->match_tag && !pf_match_tag(m, r, &tag, 1075 pd->pf_mtag ? pd->pf_mtag->tag : 0)) 1076 r = TAILQ_NEXT(r, entries); 1077 else 1078 break; 1079 } 1080 1081 if (srs) { 1082 /* With scrub rules present IPv4 normalization happens only 1083 * if one of rules has matched and it's not a "no scrub" rule */ 1084 if (r == NULL || r->action == PF_NOSCRUB) 1085 return (PF_PASS); 1086 1087 pf_counter_u64_critical_enter(); 1088 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1); 1089 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len); 1090 pf_counter_u64_critical_exit(); 1091 pf_rule_to_actions(r, &pd->act); 1092 } else if ((!V_pf_status.reass && (h->ip_off & htons(IP_MF | IP_OFFMASK)))) { 1093 /* With no scrub rules IPv4 fragment reassembly depends on the 1094 * global switch. Fragments can be dropped early if reassembly 1095 * is disabled. */ 1096 REASON_SET(reason, PFRES_NORM); 1097 goto drop; 1098 } 1099 1100 /* Check for illegal packets */ 1101 if (hlen < (int)sizeof(struct ip)) { 1102 REASON_SET(reason, PFRES_NORM); 1103 goto drop; 1104 } 1105 1106 if (hlen > ntohs(h->ip_len)) { 1107 REASON_SET(reason, PFRES_NORM); 1108 goto drop; 1109 } 1110 1111 /* Clear IP_DF if the rule uses the no-df option or we're in no-df mode */ 1112 if ((((r && r->rule_flag & PFRULE_NODF) || 1113 (V_pf_status.reass & PF_REASS_NODF)) && h->ip_off & htons(IP_DF) 1114 )) { 1115 u_int16_t ip_off = h->ip_off; 1116 1117 h->ip_off &= htons(~IP_DF); 1118 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0); 1119 } 1120 1121 /* We will need other tests here */ 1122 if (!fragoff && !mff) 1123 goto no_fragment; 1124 1125 /* We're dealing with a fragment now. Don't allow fragments 1126 * with IP_DF to enter the cache. If the flag was cleared by 1127 * no-df above, fine. Otherwise drop it. 1128 */ 1129 if (h->ip_off & htons(IP_DF)) { 1130 DPFPRINTF(("IP_DF\n")); 1131 goto bad; 1132 } 1133 1134 ip_len = ntohs(h->ip_len) - hlen; 1135 1136 /* All fragments are 8 byte aligned */ 1137 if (mff && (ip_len & 0x7)) { 1138 DPFPRINTF(("mff and %d\n", ip_len)); 1139 goto bad; 1140 } 1141 1142 /* Respect maximum length */ 1143 if (fragoff + ip_len > IP_MAXPACKET) { 1144 DPFPRINTF(("max packet %d\n", fragoff + ip_len)); 1145 goto bad; 1146 } 1147 1148 if (r==NULL || !(r->rule_flag & PFRULE_FRAGMENT_NOREASS)) { 1149 max = fragoff + ip_len; 1150 1151 /* Fully buffer all of the fragments 1152 * Might return a completely reassembled mbuf, or NULL */ 1153 PF_FRAG_LOCK(); 1154 DPFPRINTF(("reass frag %d @ %d-%d\n", h->ip_id, fragoff, max)); 1155 verdict = pf_reassemble(m0, h, pd->dir, reason); 1156 PF_FRAG_UNLOCK(); 1157 1158 if (verdict != PF_PASS) 1159 return (PF_DROP); 1160 1161 m = *m0; 1162 if (m == NULL) 1163 return (PF_DROP); 1164 1165 h = mtod(m, struct ip *); 1166 1167 no_fragment: 1168 /* At this point, only IP_DF is allowed in ip_off */ 1169 if (h->ip_off & ~htons(IP_DF)) { 1170 u_int16_t ip_off = h->ip_off; 1171 1172 h->ip_off &= htons(IP_DF); 1173 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0); 1174 } 1175 } 1176 1177 return (PF_PASS); 1178 1179 bad: 1180 DPFPRINTF(("dropping bad fragment\n")); 1181 REASON_SET(reason, PFRES_FRAG); 1182 drop: 1183 if (r != NULL && r->log) 1184 PFLOG_PACKET(kif, m, AF_INET, *reason, r, NULL, NULL, pd, 1); 1185 1186 return (PF_DROP); 1187 } 1188 #endif 1189 1190 #ifdef INET6 1191 int 1192 pf_normalize_ip6(struct mbuf **m0, struct pfi_kkif *kif, 1193 u_short *reason, struct pf_pdesc *pd) 1194 { 1195 struct mbuf *m = *m0; 1196 struct pf_krule *r; 1197 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 1198 int extoff; 1199 int off; 1200 struct ip6_ext ext; 1201 struct ip6_opt opt; 1202 struct ip6_frag frag; 1203 u_int32_t plen; 1204 int optend; 1205 int ooff; 1206 u_int8_t proto; 1207 int terminal; 1208 int srs; 1209 1210 PF_RULES_RASSERT(); 1211 1212 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr); 1213 /* Check if there any scrub rules. Lack of scrub rules means enforced 1214 * packet normalization operation just like in OpenBSD. */ 1215 srs = (r != NULL); 1216 while (r != NULL) { 1217 pf_counter_u64_add(&r->evaluations, 1); 1218 if (pfi_kkif_match(r->kif, kif) == r->ifnot) 1219 r = r->skip[PF_SKIP_IFP].ptr; 1220 else if (r->direction && r->direction != pd->dir) 1221 r = r->skip[PF_SKIP_DIR].ptr; 1222 else if (r->af && r->af != AF_INET6) 1223 r = r->skip[PF_SKIP_AF].ptr; 1224 #if 0 /* header chain! */ 1225 else if (r->proto && r->proto != h->ip6_nxt) 1226 r = r->skip[PF_SKIP_PROTO].ptr; 1227 #endif 1228 else if (PF_MISMATCHAW(&r->src.addr, 1229 (struct pf_addr *)&h->ip6_src, AF_INET6, 1230 r->src.neg, kif, M_GETFIB(m))) 1231 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 1232 else if (PF_MISMATCHAW(&r->dst.addr, 1233 (struct pf_addr *)&h->ip6_dst, AF_INET6, 1234 r->dst.neg, NULL, M_GETFIB(m))) 1235 r = r->skip[PF_SKIP_DST_ADDR].ptr; 1236 else 1237 break; 1238 } 1239 1240 if (srs) { 1241 /* With scrub rules present IPv6 normalization happens only 1242 * if one of rules has matched and it's not a "no scrub" rule */ 1243 if (r == NULL || r->action == PF_NOSCRUB) 1244 return (PF_PASS); 1245 1246 pf_counter_u64_critical_enter(); 1247 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1); 1248 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len); 1249 pf_counter_u64_critical_exit(); 1250 pf_rule_to_actions(r, &pd->act); 1251 } 1252 1253 /* Check for illegal packets */ 1254 if (sizeof(struct ip6_hdr) + IPV6_MAXPACKET < m->m_pkthdr.len) 1255 goto drop; 1256 1257 again: 1258 h = mtod(m, struct ip6_hdr *); 1259 plen = ntohs(h->ip6_plen); 1260 /* jumbo payload option not supported */ 1261 if (plen == 0) 1262 goto drop; 1263 1264 extoff = 0; 1265 off = sizeof(struct ip6_hdr); 1266 proto = h->ip6_nxt; 1267 terminal = 0; 1268 do { 1269 switch (proto) { 1270 case IPPROTO_FRAGMENT: 1271 goto fragment; 1272 break; 1273 case IPPROTO_AH: 1274 case IPPROTO_ROUTING: 1275 case IPPROTO_DSTOPTS: 1276 if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL, 1277 NULL, AF_INET6)) 1278 goto shortpkt; 1279 extoff = off; 1280 if (proto == IPPROTO_AH) 1281 off += (ext.ip6e_len + 2) * 4; 1282 else 1283 off += (ext.ip6e_len + 1) * 8; 1284 proto = ext.ip6e_nxt; 1285 break; 1286 case IPPROTO_HOPOPTS: 1287 if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL, 1288 NULL, AF_INET6)) 1289 goto shortpkt; 1290 extoff = off; 1291 optend = off + (ext.ip6e_len + 1) * 8; 1292 ooff = off + sizeof(ext); 1293 do { 1294 if (!pf_pull_hdr(m, ooff, &opt.ip6o_type, 1295 sizeof(opt.ip6o_type), NULL, NULL, 1296 AF_INET6)) 1297 goto shortpkt; 1298 if (opt.ip6o_type == IP6OPT_PAD1) { 1299 ooff++; 1300 continue; 1301 } 1302 if (!pf_pull_hdr(m, ooff, &opt, sizeof(opt), 1303 NULL, NULL, AF_INET6)) 1304 goto shortpkt; 1305 if (ooff + sizeof(opt) + opt.ip6o_len > optend) 1306 goto drop; 1307 if (opt.ip6o_type == IP6OPT_JUMBO) 1308 goto drop; 1309 ooff += sizeof(opt) + opt.ip6o_len; 1310 } while (ooff < optend); 1311 1312 off = optend; 1313 proto = ext.ip6e_nxt; 1314 break; 1315 default: 1316 terminal = 1; 1317 break; 1318 } 1319 } while (!terminal); 1320 1321 if (sizeof(struct ip6_hdr) + plen > m->m_pkthdr.len) 1322 goto shortpkt; 1323 1324 return (PF_PASS); 1325 1326 fragment: 1327 if (pd->flags & PFDESC_IP_REAS) 1328 return (PF_DROP); 1329 if (sizeof(struct ip6_hdr) + plen > m->m_pkthdr.len) 1330 goto shortpkt; 1331 1332 if (!pf_pull_hdr(m, off, &frag, sizeof(frag), NULL, NULL, AF_INET6)) 1333 goto shortpkt; 1334 1335 /* Offset now points to data portion. */ 1336 off += sizeof(frag); 1337 1338 /* Returns PF_DROP or *m0 is NULL or completely reassembled mbuf. */ 1339 if (pf_reassemble6(m0, h, &frag, off, extoff, reason) != PF_PASS) 1340 return (PF_DROP); 1341 m = *m0; 1342 if (m == NULL) 1343 return (PF_DROP); 1344 1345 pd->flags |= PFDESC_IP_REAS; 1346 goto again; 1347 1348 shortpkt: 1349 REASON_SET(reason, PFRES_SHORT); 1350 if (r != NULL && r->log) 1351 PFLOG_PACKET(kif, m, AF_INET6, *reason, r, NULL, NULL, pd, 1); 1352 return (PF_DROP); 1353 1354 drop: 1355 REASON_SET(reason, PFRES_NORM); 1356 if (r != NULL && r->log) 1357 PFLOG_PACKET(kif, m, AF_INET6, *reason, r, NULL, NULL, pd, 1); 1358 return (PF_DROP); 1359 } 1360 #endif /* INET6 */ 1361 1362 int 1363 pf_normalize_tcp(struct pfi_kkif *kif, struct mbuf *m, int ipoff, 1364 int off, void *h, struct pf_pdesc *pd) 1365 { 1366 struct pf_krule *r, *rm = NULL; 1367 struct tcphdr *th = &pd->hdr.tcp; 1368 int rewrite = 0; 1369 u_short reason; 1370 u_int8_t flags; 1371 sa_family_t af = pd->af; 1372 int srs; 1373 1374 PF_RULES_RASSERT(); 1375 1376 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr); 1377 /* Check if there any scrub rules. Lack of scrub rules means enforced 1378 * packet normalization operation just like in OpenBSD. */ 1379 srs = (r != NULL); 1380 while (r != NULL) { 1381 pf_counter_u64_add(&r->evaluations, 1); 1382 if (pfi_kkif_match(r->kif, kif) == r->ifnot) 1383 r = r->skip[PF_SKIP_IFP].ptr; 1384 else if (r->direction && r->direction != pd->dir) 1385 r = r->skip[PF_SKIP_DIR].ptr; 1386 else if (r->af && r->af != af) 1387 r = r->skip[PF_SKIP_AF].ptr; 1388 else if (r->proto && r->proto != pd->proto) 1389 r = r->skip[PF_SKIP_PROTO].ptr; 1390 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af, 1391 r->src.neg, kif, M_GETFIB(m))) 1392 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 1393 else if (r->src.port_op && !pf_match_port(r->src.port_op, 1394 r->src.port[0], r->src.port[1], th->th_sport)) 1395 r = r->skip[PF_SKIP_SRC_PORT].ptr; 1396 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af, 1397 r->dst.neg, NULL, M_GETFIB(m))) 1398 r = r->skip[PF_SKIP_DST_ADDR].ptr; 1399 else if (r->dst.port_op && !pf_match_port(r->dst.port_op, 1400 r->dst.port[0], r->dst.port[1], th->th_dport)) 1401 r = r->skip[PF_SKIP_DST_PORT].ptr; 1402 else if (r->os_fingerprint != PF_OSFP_ANY && !pf_osfp_match( 1403 pf_osfp_fingerprint(pd, m, off, th), 1404 r->os_fingerprint)) 1405 r = TAILQ_NEXT(r, entries); 1406 else { 1407 rm = r; 1408 break; 1409 } 1410 } 1411 1412 if (srs) { 1413 /* With scrub rules present TCP normalization happens only 1414 * if one of rules has matched and it's not a "no scrub" rule */ 1415 if (rm == NULL || rm->action == PF_NOSCRUB) 1416 return (PF_PASS); 1417 1418 pf_counter_u64_critical_enter(); 1419 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1); 1420 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len); 1421 pf_counter_u64_critical_exit(); 1422 pf_rule_to_actions(rm, &pd->act); 1423 } 1424 1425 if (rm && rm->rule_flag & PFRULE_REASSEMBLE_TCP) 1426 pd->flags |= PFDESC_TCP_NORM; 1427 1428 flags = th->th_flags; 1429 if (flags & TH_SYN) { 1430 /* Illegal packet */ 1431 if (flags & TH_RST) 1432 goto tcp_drop; 1433 1434 if (flags & TH_FIN) 1435 goto tcp_drop; 1436 } else { 1437 /* Illegal packet */ 1438 if (!(flags & (TH_ACK|TH_RST))) 1439 goto tcp_drop; 1440 } 1441 1442 if (!(flags & TH_ACK)) { 1443 /* These flags are only valid if ACK is set */ 1444 if ((flags & TH_FIN) || (flags & TH_PUSH) || (flags & TH_URG)) 1445 goto tcp_drop; 1446 } 1447 1448 /* Check for illegal header length */ 1449 if (th->th_off < (sizeof(struct tcphdr) >> 2)) 1450 goto tcp_drop; 1451 1452 /* If flags changed, or reserved data set, then adjust */ 1453 if (flags != th->th_flags || th->th_x2 != 0) { 1454 u_int16_t ov, nv; 1455 1456 ov = *(u_int16_t *)(&th->th_ack + 1); 1457 th->th_flags = flags; 1458 th->th_x2 = 0; 1459 nv = *(u_int16_t *)(&th->th_ack + 1); 1460 1461 th->th_sum = pf_proto_cksum_fixup(m, th->th_sum, ov, nv, 0); 1462 rewrite = 1; 1463 } 1464 1465 /* Remove urgent pointer, if TH_URG is not set */ 1466 if (!(flags & TH_URG) && th->th_urp) { 1467 th->th_sum = pf_proto_cksum_fixup(m, th->th_sum, th->th_urp, 1468 0, 0); 1469 th->th_urp = 0; 1470 rewrite = 1; 1471 } 1472 1473 /* copy back packet headers if we sanitized */ 1474 if (rewrite) 1475 m_copyback(m, off, sizeof(*th), (caddr_t)th); 1476 1477 return (PF_PASS); 1478 1479 tcp_drop: 1480 REASON_SET(&reason, PFRES_NORM); 1481 if (rm != NULL && r->log) 1482 PFLOG_PACKET(kif, m, AF_INET, reason, r, NULL, NULL, pd, 1); 1483 return (PF_DROP); 1484 } 1485 1486 int 1487 pf_normalize_tcp_init(struct mbuf *m, int off, struct pf_pdesc *pd, 1488 struct tcphdr *th, struct pf_state_peer *src, struct pf_state_peer *dst) 1489 { 1490 u_int32_t tsval, tsecr; 1491 u_int8_t hdr[60]; 1492 u_int8_t *opt; 1493 1494 KASSERT((src->scrub == NULL), 1495 ("pf_normalize_tcp_init: src->scrub != NULL")); 1496 1497 src->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT); 1498 if (src->scrub == NULL) 1499 return (1); 1500 1501 switch (pd->af) { 1502 #ifdef INET 1503 case AF_INET: { 1504 struct ip *h = mtod(m, struct ip *); 1505 src->scrub->pfss_ttl = h->ip_ttl; 1506 break; 1507 } 1508 #endif /* INET */ 1509 #ifdef INET6 1510 case AF_INET6: { 1511 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 1512 src->scrub->pfss_ttl = h->ip6_hlim; 1513 break; 1514 } 1515 #endif /* INET6 */ 1516 } 1517 1518 /* 1519 * All normalizations below are only begun if we see the start of 1520 * the connections. They must all set an enabled bit in pfss_flags 1521 */ 1522 if ((th->th_flags & TH_SYN) == 0) 1523 return (0); 1524 1525 if (th->th_off > (sizeof(struct tcphdr) >> 2) && src->scrub && 1526 pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) { 1527 /* Diddle with TCP options */ 1528 int hlen; 1529 opt = hdr + sizeof(struct tcphdr); 1530 hlen = (th->th_off << 2) - sizeof(struct tcphdr); 1531 while (hlen >= TCPOLEN_TIMESTAMP) { 1532 switch (*opt) { 1533 case TCPOPT_EOL: /* FALLTHROUGH */ 1534 case TCPOPT_NOP: 1535 opt++; 1536 hlen--; 1537 break; 1538 case TCPOPT_TIMESTAMP: 1539 if (opt[1] >= TCPOLEN_TIMESTAMP) { 1540 src->scrub->pfss_flags |= 1541 PFSS_TIMESTAMP; 1542 src->scrub->pfss_ts_mod = 1543 htonl(arc4random()); 1544 1545 /* note PFSS_PAWS not set yet */ 1546 memcpy(&tsval, &opt[2], 1547 sizeof(u_int32_t)); 1548 memcpy(&tsecr, &opt[6], 1549 sizeof(u_int32_t)); 1550 src->scrub->pfss_tsval0 = ntohl(tsval); 1551 src->scrub->pfss_tsval = ntohl(tsval); 1552 src->scrub->pfss_tsecr = ntohl(tsecr); 1553 getmicrouptime(&src->scrub->pfss_last); 1554 } 1555 /* FALLTHROUGH */ 1556 default: 1557 hlen -= MAX(opt[1], 2); 1558 opt += MAX(opt[1], 2); 1559 break; 1560 } 1561 } 1562 } 1563 1564 return (0); 1565 } 1566 1567 void 1568 pf_normalize_tcp_cleanup(struct pf_kstate *state) 1569 { 1570 uma_zfree(V_pf_state_scrub_z, state->src.scrub); 1571 uma_zfree(V_pf_state_scrub_z, state->dst.scrub); 1572 1573 /* Someday... flush the TCP segment reassembly descriptors. */ 1574 } 1575 1576 int 1577 pf_normalize_tcp_stateful(struct mbuf *m, int off, struct pf_pdesc *pd, 1578 u_short *reason, struct tcphdr *th, struct pf_kstate *state, 1579 struct pf_state_peer *src, struct pf_state_peer *dst, int *writeback) 1580 { 1581 struct timeval uptime; 1582 u_int32_t tsval, tsecr; 1583 u_int tsval_from_last; 1584 u_int8_t hdr[60]; 1585 u_int8_t *opt; 1586 int copyback = 0; 1587 int got_ts = 0; 1588 size_t startoff; 1589 1590 KASSERT((src->scrub || dst->scrub), 1591 ("%s: src->scrub && dst->scrub!", __func__)); 1592 1593 /* 1594 * Enforce the minimum TTL seen for this connection. Negate a common 1595 * technique to evade an intrusion detection system and confuse 1596 * firewall state code. 1597 */ 1598 switch (pd->af) { 1599 #ifdef INET 1600 case AF_INET: { 1601 if (src->scrub) { 1602 struct ip *h = mtod(m, struct ip *); 1603 if (h->ip_ttl > src->scrub->pfss_ttl) 1604 src->scrub->pfss_ttl = h->ip_ttl; 1605 h->ip_ttl = src->scrub->pfss_ttl; 1606 } 1607 break; 1608 } 1609 #endif /* INET */ 1610 #ifdef INET6 1611 case AF_INET6: { 1612 if (src->scrub) { 1613 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 1614 if (h->ip6_hlim > src->scrub->pfss_ttl) 1615 src->scrub->pfss_ttl = h->ip6_hlim; 1616 h->ip6_hlim = src->scrub->pfss_ttl; 1617 } 1618 break; 1619 } 1620 #endif /* INET6 */ 1621 } 1622 1623 if (th->th_off > (sizeof(struct tcphdr) >> 2) && 1624 ((src->scrub && (src->scrub->pfss_flags & PFSS_TIMESTAMP)) || 1625 (dst->scrub && (dst->scrub->pfss_flags & PFSS_TIMESTAMP))) && 1626 pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) { 1627 /* Diddle with TCP options */ 1628 int hlen; 1629 opt = hdr + sizeof(struct tcphdr); 1630 hlen = (th->th_off << 2) - sizeof(struct tcphdr); 1631 while (hlen >= TCPOLEN_TIMESTAMP) { 1632 startoff = opt - (hdr + sizeof(struct tcphdr)); 1633 switch (*opt) { 1634 case TCPOPT_EOL: /* FALLTHROUGH */ 1635 case TCPOPT_NOP: 1636 opt++; 1637 hlen--; 1638 break; 1639 case TCPOPT_TIMESTAMP: 1640 /* Modulate the timestamps. Can be used for 1641 * NAT detection, OS uptime determination or 1642 * reboot detection. 1643 */ 1644 1645 if (got_ts) { 1646 /* Huh? Multiple timestamps!? */ 1647 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1648 DPFPRINTF(("multiple TS??\n")); 1649 pf_print_state(state); 1650 printf("\n"); 1651 } 1652 REASON_SET(reason, PFRES_TS); 1653 return (PF_DROP); 1654 } 1655 if (opt[1] >= TCPOLEN_TIMESTAMP) { 1656 memcpy(&tsval, &opt[2], 1657 sizeof(u_int32_t)); 1658 if (tsval && src->scrub && 1659 (src->scrub->pfss_flags & 1660 PFSS_TIMESTAMP)) { 1661 tsval = ntohl(tsval); 1662 pf_patch_32_unaligned(m, 1663 &th->th_sum, 1664 &opt[2], 1665 htonl(tsval + 1666 src->scrub->pfss_ts_mod), 1667 PF_ALGNMNT(startoff), 1668 0); 1669 copyback = 1; 1670 } 1671 1672 /* Modulate TS reply iff valid (!0) */ 1673 memcpy(&tsecr, &opt[6], 1674 sizeof(u_int32_t)); 1675 if (tsecr && dst->scrub && 1676 (dst->scrub->pfss_flags & 1677 PFSS_TIMESTAMP)) { 1678 tsecr = ntohl(tsecr) 1679 - dst->scrub->pfss_ts_mod; 1680 pf_patch_32_unaligned(m, 1681 &th->th_sum, 1682 &opt[6], 1683 htonl(tsecr), 1684 PF_ALGNMNT(startoff), 1685 0); 1686 copyback = 1; 1687 } 1688 got_ts = 1; 1689 } 1690 /* FALLTHROUGH */ 1691 default: 1692 hlen -= MAX(opt[1], 2); 1693 opt += MAX(opt[1], 2); 1694 break; 1695 } 1696 } 1697 if (copyback) { 1698 /* Copyback the options, caller copys back header */ 1699 *writeback = 1; 1700 m_copyback(m, off + sizeof(struct tcphdr), 1701 (th->th_off << 2) - sizeof(struct tcphdr), hdr + 1702 sizeof(struct tcphdr)); 1703 } 1704 } 1705 1706 /* 1707 * Must invalidate PAWS checks on connections idle for too long. 1708 * The fastest allowed timestamp clock is 1ms. That turns out to 1709 * be about 24 days before it wraps. XXX Right now our lowerbound 1710 * TS echo check only works for the first 12 days of a connection 1711 * when the TS has exhausted half its 32bit space 1712 */ 1713 #define TS_MAX_IDLE (24*24*60*60) 1714 #define TS_MAX_CONN (12*24*60*60) /* XXX remove when better tsecr check */ 1715 1716 getmicrouptime(&uptime); 1717 if (src->scrub && (src->scrub->pfss_flags & PFSS_PAWS) && 1718 (uptime.tv_sec - src->scrub->pfss_last.tv_sec > TS_MAX_IDLE || 1719 time_uptime - state->creation > TS_MAX_CONN)) { 1720 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1721 DPFPRINTF(("src idled out of PAWS\n")); 1722 pf_print_state(state); 1723 printf("\n"); 1724 } 1725 src->scrub->pfss_flags = (src->scrub->pfss_flags & ~PFSS_PAWS) 1726 | PFSS_PAWS_IDLED; 1727 } 1728 if (dst->scrub && (dst->scrub->pfss_flags & PFSS_PAWS) && 1729 uptime.tv_sec - dst->scrub->pfss_last.tv_sec > TS_MAX_IDLE) { 1730 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1731 DPFPRINTF(("dst idled out of PAWS\n")); 1732 pf_print_state(state); 1733 printf("\n"); 1734 } 1735 dst->scrub->pfss_flags = (dst->scrub->pfss_flags & ~PFSS_PAWS) 1736 | PFSS_PAWS_IDLED; 1737 } 1738 1739 if (got_ts && src->scrub && dst->scrub && 1740 (src->scrub->pfss_flags & PFSS_PAWS) && 1741 (dst->scrub->pfss_flags & PFSS_PAWS)) { 1742 /* Validate that the timestamps are "in-window". 1743 * RFC1323 describes TCP Timestamp options that allow 1744 * measurement of RTT (round trip time) and PAWS 1745 * (protection against wrapped sequence numbers). PAWS 1746 * gives us a set of rules for rejecting packets on 1747 * long fat pipes (packets that were somehow delayed 1748 * in transit longer than the time it took to send the 1749 * full TCP sequence space of 4Gb). We can use these 1750 * rules and infer a few others that will let us treat 1751 * the 32bit timestamp and the 32bit echoed timestamp 1752 * as sequence numbers to prevent a blind attacker from 1753 * inserting packets into a connection. 1754 * 1755 * RFC1323 tells us: 1756 * - The timestamp on this packet must be greater than 1757 * or equal to the last value echoed by the other 1758 * endpoint. The RFC says those will be discarded 1759 * since it is a dup that has already been acked. 1760 * This gives us a lowerbound on the timestamp. 1761 * timestamp >= other last echoed timestamp 1762 * - The timestamp will be less than or equal to 1763 * the last timestamp plus the time between the 1764 * last packet and now. The RFC defines the max 1765 * clock rate as 1ms. We will allow clocks to be 1766 * up to 10% fast and will allow a total difference 1767 * or 30 seconds due to a route change. And this 1768 * gives us an upperbound on the timestamp. 1769 * timestamp <= last timestamp + max ticks 1770 * We have to be careful here. Windows will send an 1771 * initial timestamp of zero and then initialize it 1772 * to a random value after the 3whs; presumably to 1773 * avoid a DoS by having to call an expensive RNG 1774 * during a SYN flood. Proof MS has at least one 1775 * good security geek. 1776 * 1777 * - The TCP timestamp option must also echo the other 1778 * endpoints timestamp. The timestamp echoed is the 1779 * one carried on the earliest unacknowledged segment 1780 * on the left edge of the sequence window. The RFC 1781 * states that the host will reject any echoed 1782 * timestamps that were larger than any ever sent. 1783 * This gives us an upperbound on the TS echo. 1784 * tescr <= largest_tsval 1785 * - The lowerbound on the TS echo is a little more 1786 * tricky to determine. The other endpoint's echoed 1787 * values will not decrease. But there may be 1788 * network conditions that re-order packets and 1789 * cause our view of them to decrease. For now the 1790 * only lowerbound we can safely determine is that 1791 * the TS echo will never be less than the original 1792 * TS. XXX There is probably a better lowerbound. 1793 * Remove TS_MAX_CONN with better lowerbound check. 1794 * tescr >= other original TS 1795 * 1796 * It is also important to note that the fastest 1797 * timestamp clock of 1ms will wrap its 32bit space in 1798 * 24 days. So we just disable TS checking after 24 1799 * days of idle time. We actually must use a 12d 1800 * connection limit until we can come up with a better 1801 * lowerbound to the TS echo check. 1802 */ 1803 struct timeval delta_ts; 1804 int ts_fudge; 1805 1806 /* 1807 * PFTM_TS_DIFF is how many seconds of leeway to allow 1808 * a host's timestamp. This can happen if the previous 1809 * packet got delayed in transit for much longer than 1810 * this packet. 1811 */ 1812 if ((ts_fudge = state->rule.ptr->timeout[PFTM_TS_DIFF]) == 0) 1813 ts_fudge = V_pf_default_rule.timeout[PFTM_TS_DIFF]; 1814 1815 /* Calculate max ticks since the last timestamp */ 1816 #define TS_MAXFREQ 1100 /* RFC max TS freq of 1Khz + 10% skew */ 1817 #define TS_MICROSECS 1000000 /* microseconds per second */ 1818 delta_ts = uptime; 1819 timevalsub(&delta_ts, &src->scrub->pfss_last); 1820 tsval_from_last = (delta_ts.tv_sec + ts_fudge) * TS_MAXFREQ; 1821 tsval_from_last += delta_ts.tv_usec / (TS_MICROSECS/TS_MAXFREQ); 1822 1823 if ((src->state >= TCPS_ESTABLISHED && 1824 dst->state >= TCPS_ESTABLISHED) && 1825 (SEQ_LT(tsval, dst->scrub->pfss_tsecr) || 1826 SEQ_GT(tsval, src->scrub->pfss_tsval + tsval_from_last) || 1827 (tsecr && (SEQ_GT(tsecr, dst->scrub->pfss_tsval) || 1828 SEQ_LT(tsecr, dst->scrub->pfss_tsval0))))) { 1829 /* Bad RFC1323 implementation or an insertion attack. 1830 * 1831 * - Solaris 2.6 and 2.7 are known to send another ACK 1832 * after the FIN,FIN|ACK,ACK closing that carries 1833 * an old timestamp. 1834 */ 1835 1836 DPFPRINTF(("Timestamp failed %c%c%c%c\n", 1837 SEQ_LT(tsval, dst->scrub->pfss_tsecr) ? '0' : ' ', 1838 SEQ_GT(tsval, src->scrub->pfss_tsval + 1839 tsval_from_last) ? '1' : ' ', 1840 SEQ_GT(tsecr, dst->scrub->pfss_tsval) ? '2' : ' ', 1841 SEQ_LT(tsecr, dst->scrub->pfss_tsval0)? '3' : ' ')); 1842 DPFPRINTF((" tsval: %u tsecr: %u +ticks: %u " 1843 "idle: %jus %lums\n", 1844 tsval, tsecr, tsval_from_last, 1845 (uintmax_t)delta_ts.tv_sec, 1846 delta_ts.tv_usec / 1000)); 1847 DPFPRINTF((" src->tsval: %u tsecr: %u\n", 1848 src->scrub->pfss_tsval, src->scrub->pfss_tsecr)); 1849 DPFPRINTF((" dst->tsval: %u tsecr: %u tsval0: %u" 1850 "\n", dst->scrub->pfss_tsval, 1851 dst->scrub->pfss_tsecr, dst->scrub->pfss_tsval0)); 1852 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1853 pf_print_state(state); 1854 pf_print_flags(th->th_flags); 1855 printf("\n"); 1856 } 1857 REASON_SET(reason, PFRES_TS); 1858 return (PF_DROP); 1859 } 1860 1861 /* XXX I'd really like to require tsecr but it's optional */ 1862 1863 } else if (!got_ts && (th->th_flags & TH_RST) == 0 && 1864 ((src->state == TCPS_ESTABLISHED && dst->state == TCPS_ESTABLISHED) 1865 || pd->p_len > 0 || (th->th_flags & TH_SYN)) && 1866 src->scrub && dst->scrub && 1867 (src->scrub->pfss_flags & PFSS_PAWS) && 1868 (dst->scrub->pfss_flags & PFSS_PAWS)) { 1869 /* Didn't send a timestamp. Timestamps aren't really useful 1870 * when: 1871 * - connection opening or closing (often not even sent). 1872 * but we must not let an attacker to put a FIN on a 1873 * data packet to sneak it through our ESTABLISHED check. 1874 * - on a TCP reset. RFC suggests not even looking at TS. 1875 * - on an empty ACK. The TS will not be echoed so it will 1876 * probably not help keep the RTT calculation in sync and 1877 * there isn't as much danger when the sequence numbers 1878 * got wrapped. So some stacks don't include TS on empty 1879 * ACKs :-( 1880 * 1881 * To minimize the disruption to mostly RFC1323 conformant 1882 * stacks, we will only require timestamps on data packets. 1883 * 1884 * And what do ya know, we cannot require timestamps on data 1885 * packets. There appear to be devices that do legitimate 1886 * TCP connection hijacking. There are HTTP devices that allow 1887 * a 3whs (with timestamps) and then buffer the HTTP request. 1888 * If the intermediate device has the HTTP response cache, it 1889 * will spoof the response but not bother timestamping its 1890 * packets. So we can look for the presence of a timestamp in 1891 * the first data packet and if there, require it in all future 1892 * packets. 1893 */ 1894 1895 if (pd->p_len > 0 && (src->scrub->pfss_flags & PFSS_DATA_TS)) { 1896 /* 1897 * Hey! Someone tried to sneak a packet in. Or the 1898 * stack changed its RFC1323 behavior?!?! 1899 */ 1900 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1901 DPFPRINTF(("Did not receive expected RFC1323 " 1902 "timestamp\n")); 1903 pf_print_state(state); 1904 pf_print_flags(th->th_flags); 1905 printf("\n"); 1906 } 1907 REASON_SET(reason, PFRES_TS); 1908 return (PF_DROP); 1909 } 1910 } 1911 1912 /* 1913 * We will note if a host sends his data packets with or without 1914 * timestamps. And require all data packets to contain a timestamp 1915 * if the first does. PAWS implicitly requires that all data packets be 1916 * timestamped. But I think there are middle-man devices that hijack 1917 * TCP streams immediately after the 3whs and don't timestamp their 1918 * packets (seen in a WWW accelerator or cache). 1919 */ 1920 if (pd->p_len > 0 && src->scrub && (src->scrub->pfss_flags & 1921 (PFSS_TIMESTAMP|PFSS_DATA_TS|PFSS_DATA_NOTS)) == PFSS_TIMESTAMP) { 1922 if (got_ts) 1923 src->scrub->pfss_flags |= PFSS_DATA_TS; 1924 else { 1925 src->scrub->pfss_flags |= PFSS_DATA_NOTS; 1926 if (V_pf_status.debug >= PF_DEBUG_MISC && dst->scrub && 1927 (dst->scrub->pfss_flags & PFSS_TIMESTAMP)) { 1928 /* Don't warn if other host rejected RFC1323 */ 1929 DPFPRINTF(("Broken RFC1323 stack did not " 1930 "timestamp data packet. Disabled PAWS " 1931 "security.\n")); 1932 pf_print_state(state); 1933 pf_print_flags(th->th_flags); 1934 printf("\n"); 1935 } 1936 } 1937 } 1938 1939 /* 1940 * Update PAWS values 1941 */ 1942 if (got_ts && src->scrub && PFSS_TIMESTAMP == (src->scrub->pfss_flags & 1943 (PFSS_PAWS_IDLED|PFSS_TIMESTAMP))) { 1944 getmicrouptime(&src->scrub->pfss_last); 1945 if (SEQ_GEQ(tsval, src->scrub->pfss_tsval) || 1946 (src->scrub->pfss_flags & PFSS_PAWS) == 0) 1947 src->scrub->pfss_tsval = tsval; 1948 1949 if (tsecr) { 1950 if (SEQ_GEQ(tsecr, src->scrub->pfss_tsecr) || 1951 (src->scrub->pfss_flags & PFSS_PAWS) == 0) 1952 src->scrub->pfss_tsecr = tsecr; 1953 1954 if ((src->scrub->pfss_flags & PFSS_PAWS) == 0 && 1955 (SEQ_LT(tsval, src->scrub->pfss_tsval0) || 1956 src->scrub->pfss_tsval0 == 0)) { 1957 /* tsval0 MUST be the lowest timestamp */ 1958 src->scrub->pfss_tsval0 = tsval; 1959 } 1960 1961 /* Only fully initialized after a TS gets echoed */ 1962 if ((src->scrub->pfss_flags & PFSS_PAWS) == 0) 1963 src->scrub->pfss_flags |= PFSS_PAWS; 1964 } 1965 } 1966 1967 /* I have a dream.... TCP segment reassembly.... */ 1968 return (0); 1969 } 1970 1971 int 1972 pf_normalize_mss(struct mbuf *m, int off, struct pf_pdesc *pd) 1973 { 1974 struct tcphdr *th = &pd->hdr.tcp; 1975 u_int16_t *mss; 1976 int thoff; 1977 int opt, cnt, optlen = 0; 1978 u_char opts[TCP_MAXOLEN]; 1979 u_char *optp = opts; 1980 size_t startoff; 1981 1982 thoff = th->th_off << 2; 1983 cnt = thoff - sizeof(struct tcphdr); 1984 1985 if (cnt > 0 && !pf_pull_hdr(m, off + sizeof(*th), opts, cnt, 1986 NULL, NULL, pd->af)) 1987 return (0); 1988 1989 for (; cnt > 0; cnt -= optlen, optp += optlen) { 1990 startoff = optp - opts; 1991 opt = optp[0]; 1992 if (opt == TCPOPT_EOL) 1993 break; 1994 if (opt == TCPOPT_NOP) 1995 optlen = 1; 1996 else { 1997 if (cnt < 2) 1998 break; 1999 optlen = optp[1]; 2000 if (optlen < 2 || optlen > cnt) 2001 break; 2002 } 2003 switch (opt) { 2004 case TCPOPT_MAXSEG: 2005 mss = (u_int16_t *)(optp + 2); 2006 if ((ntohs(*mss)) > pd->act.max_mss) { 2007 pf_patch_16_unaligned(m, 2008 &th->th_sum, 2009 mss, htons(pd->act.max_mss), 2010 PF_ALGNMNT(startoff), 2011 0); 2012 m_copyback(m, off + sizeof(*th), 2013 thoff - sizeof(*th), opts); 2014 m_copyback(m, off, sizeof(*th), (caddr_t)th); 2015 } 2016 break; 2017 default: 2018 break; 2019 } 2020 } 2021 2022 return (0); 2023 } 2024 2025 static int 2026 pf_scan_sctp(struct mbuf *m, int ipoff, int off, struct pf_pdesc *pd) 2027 { 2028 struct sctp_chunkhdr ch = { }; 2029 int chunk_off = sizeof(struct sctphdr); 2030 int chunk_start; 2031 2032 while (off + chunk_off < pd->tot_len) { 2033 if (!pf_pull_hdr(m, off + chunk_off, &ch, sizeof(ch), NULL, 2034 NULL, pd->af)) 2035 return (PF_DROP); 2036 2037 /* Length includes the header, this must be at least 4. */ 2038 if (ntohs(ch.chunk_length) < 4) 2039 return (PF_DROP); 2040 2041 chunk_start = chunk_off; 2042 chunk_off += roundup(ntohs(ch.chunk_length), 4); 2043 2044 switch (ch.chunk_type) { 2045 case SCTP_INITIATION: { 2046 struct sctp_init_chunk init; 2047 2048 if (!pf_pull_hdr(m, off + chunk_start, &init, 2049 sizeof(init), NULL, NULL, pd->af)) 2050 return (PF_DROP); 2051 2052 /* 2053 * RFC 9620, Section 3.3.2, "The Initiate Tag is allowed to have 2054 * any value except 0." 2055 */ 2056 if (init.init.initiate_tag == 0) 2057 return (PF_DROP); 2058 if (init.init.num_inbound_streams == 0) 2059 return (PF_DROP); 2060 if (init.init.num_outbound_streams == 0) 2061 return (PF_DROP); 2062 if (ntohl(init.init.a_rwnd) < SCTP_MIN_RWND) 2063 return (PF_DROP); 2064 2065 /* 2066 * RFC 9260, Section 3.1, INIT chunks MUST have zero 2067 * verification tag. 2068 */ 2069 if (pd->hdr.sctp.v_tag != 0) 2070 return (PF_DROP); 2071 2072 pd->sctp_initiate_tag = init.init.initiate_tag; 2073 2074 pd->sctp_flags |= PFDESC_SCTP_INIT; 2075 break; 2076 } 2077 case SCTP_INITIATION_ACK: 2078 pd->sctp_flags |= PFDESC_SCTP_INIT_ACK; 2079 break; 2080 case SCTP_ABORT_ASSOCIATION: 2081 pd->sctp_flags |= PFDESC_SCTP_ABORT; 2082 break; 2083 case SCTP_SHUTDOWN: 2084 case SCTP_SHUTDOWN_ACK: 2085 pd->sctp_flags |= PFDESC_SCTP_SHUTDOWN; 2086 break; 2087 case SCTP_SHUTDOWN_COMPLETE: 2088 pd->sctp_flags |= PFDESC_SCTP_SHUTDOWN_COMPLETE; 2089 break; 2090 case SCTP_COOKIE_ECHO: 2091 case SCTP_COOKIE_ACK: 2092 pd->sctp_flags |= PFDESC_SCTP_COOKIE; 2093 break; 2094 case SCTP_DATA: 2095 pd->sctp_flags |= PFDESC_SCTP_DATA; 2096 break; 2097 default: 2098 pd->sctp_flags |= PFDESC_SCTP_OTHER; 2099 break; 2100 } 2101 } 2102 2103 /* Validate chunk lengths vs. packet length. */ 2104 if (off + chunk_off != pd->tot_len) 2105 return (PF_DROP); 2106 2107 /* 2108 * INIT, INIT_ACK or SHUTDOWN_COMPLETE chunks must always be the only 2109 * one in a packet. 2110 */ 2111 if ((pd->sctp_flags & PFDESC_SCTP_INIT) && 2112 (pd->sctp_flags & ~PFDESC_SCTP_INIT)) 2113 return (PF_DROP); 2114 if ((pd->sctp_flags & PFDESC_SCTP_INIT_ACK) && 2115 (pd->sctp_flags & ~PFDESC_SCTP_INIT_ACK)) 2116 return (PF_DROP); 2117 if ((pd->sctp_flags & PFDESC_SCTP_SHUTDOWN_COMPLETE) && 2118 (pd->sctp_flags & ~PFDESC_SCTP_SHUTDOWN_COMPLETE)) 2119 return (PF_DROP); 2120 2121 return (PF_PASS); 2122 } 2123 2124 int 2125 pf_normalize_sctp(int dir, struct pfi_kkif *kif, struct mbuf *m, int ipoff, 2126 int off, void *h, 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 /* Unconditionally scan the SCTP packet, because we need to look for 2137 * things like shutdown and asconf chunks. */ 2138 if (pf_scan_sctp(m, ipoff, off, pd) != PF_PASS) 2139 goto sctp_drop; 2140 2141 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr); 2142 /* Check if there any scrub rules. Lack of scrub rules means enforced 2143 * packet normalization operation just like in OpenBSD. */ 2144 srs = (r != NULL); 2145 while (r != NULL) { 2146 pf_counter_u64_add(&r->evaluations, 1); 2147 if (pfi_kkif_match(r->kif, kif) == r->ifnot) 2148 r = r->skip[PF_SKIP_IFP].ptr; 2149 else if (r->direction && r->direction != dir) 2150 r = r->skip[PF_SKIP_DIR].ptr; 2151 else if (r->af && r->af != af) 2152 r = r->skip[PF_SKIP_AF].ptr; 2153 else if (r->proto && r->proto != pd->proto) 2154 r = r->skip[PF_SKIP_PROTO].ptr; 2155 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af, 2156 r->src.neg, kif, M_GETFIB(m))) 2157 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 2158 else if (r->src.port_op && !pf_match_port(r->src.port_op, 2159 r->src.port[0], r->src.port[1], sh->src_port)) 2160 r = r->skip[PF_SKIP_SRC_PORT].ptr; 2161 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af, 2162 r->dst.neg, NULL, M_GETFIB(m))) 2163 r = r->skip[PF_SKIP_DST_ADDR].ptr; 2164 else if (r->dst.port_op && !pf_match_port(r->dst.port_op, 2165 r->dst.port[0], r->dst.port[1], sh->dest_port)) 2166 r = r->skip[PF_SKIP_DST_PORT].ptr; 2167 else { 2168 rm = r; 2169 break; 2170 } 2171 } 2172 2173 if (srs) { 2174 /* With scrub rules present SCTP normalization happens only 2175 * if one of rules has matched and it's not a "no scrub" rule */ 2176 if (rm == NULL || rm->action == PF_NOSCRUB) 2177 return (PF_PASS); 2178 2179 pf_counter_u64_critical_enter(); 2180 pf_counter_u64_add_protected(&r->packets[dir == PF_OUT], 1); 2181 pf_counter_u64_add_protected(&r->bytes[dir == PF_OUT], pd->tot_len); 2182 pf_counter_u64_critical_exit(); 2183 } 2184 2185 /* Verify we're a multiple of 4 bytes long */ 2186 if ((pd->tot_len - off - sizeof(struct sctphdr)) % 4) 2187 goto sctp_drop; 2188 2189 /* INIT chunk needs to be the only chunk */ 2190 if (pd->sctp_flags & PFDESC_SCTP_INIT) 2191 if (pd->sctp_flags & ~PFDESC_SCTP_INIT) 2192 goto sctp_drop; 2193 2194 return (PF_PASS); 2195 2196 sctp_drop: 2197 REASON_SET(&reason, PFRES_NORM); 2198 if (rm != NULL && r->log) 2199 PFLOG_PACKET(kif, m, AF_INET, reason, r, NULL, NULL, pd, 2200 1); 2201 2202 return (PF_DROP); 2203 } 2204 2205 #ifdef INET 2206 void 2207 pf_scrub_ip(struct mbuf **m0, struct pf_pdesc *pd) 2208 { 2209 struct mbuf *m = *m0; 2210 struct ip *h = mtod(m, struct ip *); 2211 2212 /* Clear IP_DF if no-df was requested */ 2213 if (pd->act.flags & PFSTATE_NODF && h->ip_off & htons(IP_DF)) { 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->act.min_ttl && h->ip_ttl < pd->act.min_ttl) { 2222 u_int16_t ip_ttl = h->ip_ttl; 2223 2224 h->ip_ttl = pd->act.min_ttl; 2225 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_ttl, h->ip_ttl, 0); 2226 } 2227 2228 /* Enforce tos */ 2229 if (pd->act.flags & PFSTATE_SETTOS) { 2230 u_int16_t ov, nv; 2231 2232 ov = *(u_int16_t *)h; 2233 h->ip_tos = pd->act.set_tos | (h->ip_tos & IPTOS_ECN_MASK); 2234 nv = *(u_int16_t *)h; 2235 2236 h->ip_sum = pf_cksum_fixup(h->ip_sum, ov, nv, 0); 2237 } 2238 2239 /* random-id, but not for fragments */ 2240 if (pd->act.flags & PFSTATE_RANDOMID && !(h->ip_off & ~htons(IP_DF))) { 2241 uint16_t ip_id = h->ip_id; 2242 2243 ip_fillid(h); 2244 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_id, h->ip_id, 0); 2245 } 2246 } 2247 #endif /* INET */ 2248 2249 #ifdef INET6 2250 void 2251 pf_scrub_ip6(struct mbuf **m0, struct pf_pdesc *pd) 2252 { 2253 struct mbuf *m = *m0; 2254 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 2255 2256 /* Enforce a minimum ttl, may cause endless packet loops */ 2257 if (pd->act.min_ttl && h->ip6_hlim < pd->act.min_ttl) 2258 h->ip6_hlim = pd->act.min_ttl; 2259 2260 /* Enforce tos. Set traffic class bits */ 2261 if (pd->act.flags & PFSTATE_SETTOS) { 2262 h->ip6_flow &= IPV6_FLOWLABEL_MASK | IPV6_VERSION_MASK; 2263 h->ip6_flow |= htonl((pd->act.set_tos | IPV6_ECN(h)) << 20); 2264 } 2265 } 2266 #endif 2267