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/vnet.h> 46#include <net/pfvar.h> 47#include <net/if_pflog.h> 48 49#include <netinet/in.h> 50#include <netinet/ip.h> 51#include <netinet/ip_var.h> 52#include <netinet6/ip6_var.h> 53#include <netinet6/scope6_var.h> 54#include <netinet/tcp.h> 55#include <netinet/tcp_fsm.h> 56#include <netinet/tcp_seq.h> 57#include <netinet/sctp_constants.h> 58#include <netinet/sctp_header.h> 59 60#ifdef INET6 61#include <netinet/ip6.h> 62#endif /* INET6 */ 63 64struct pf_frent { 65 TAILQ_ENTRY(pf_frent) fr_next; 66 struct mbuf *fe_m; 67 uint16_t fe_hdrlen; /* ipv4 header length with ip options 68 ipv6, extension, fragment header */ 69 uint16_t fe_extoff; /* last extension header offset or 0 */ 70 uint16_t fe_len; /* fragment length */ 71 uint16_t fe_off; /* fragment offset */ 72 uint16_t fe_mff; /* more fragment flag */ 73}; 74 75struct pf_fragment_cmp { 76 struct pf_addr frc_src; 77 struct pf_addr frc_dst; 78 uint32_t frc_id; 79 sa_family_t frc_af; 80 uint8_t frc_proto; 81}; 82 83struct pf_fragment { 84 struct pf_fragment_cmp fr_key; 85#define fr_src fr_key.frc_src 86#define fr_dst fr_key.frc_dst 87#define fr_id fr_key.frc_id 88#define fr_af fr_key.frc_af 89#define fr_proto fr_key.frc_proto 90 91 /* pointers to queue element */ 92 struct pf_frent *fr_firstoff[PF_FRAG_ENTRY_POINTS]; 93 /* count entries between pointers */ 94 uint8_t fr_entries[PF_FRAG_ENTRY_POINTS]; 95 RB_ENTRY(pf_fragment) fr_entry; 96 TAILQ_ENTRY(pf_fragment) frag_next; 97 uint32_t fr_timeout; 98 uint16_t fr_maxlen; /* maximum length of single fragment */ 99 u_int16_t fr_holes; /* number of holes in the queue */ 100 TAILQ_HEAD(pf_fragq, pf_frent) fr_queue; 101}; 102 103struct pf_fragment_tag { 104 uint16_t ft_hdrlen; /* header length of reassembled pkt */ 105 uint16_t ft_extoff; /* last extension header offset or 0 */ 106 uint16_t ft_maxlen; /* maximum fragment payload length */ 107 uint32_t ft_id; /* fragment id */ 108}; 109 110VNET_DEFINE_STATIC(struct mtx, pf_frag_mtx); 111#define V_pf_frag_mtx VNET(pf_frag_mtx) 112#define PF_FRAG_LOCK() mtx_lock(&V_pf_frag_mtx) 113#define PF_FRAG_UNLOCK() mtx_unlock(&V_pf_frag_mtx) 114#define PF_FRAG_ASSERT() mtx_assert(&V_pf_frag_mtx, MA_OWNED) 115 116VNET_DEFINE(uma_zone_t, pf_state_scrub_z); /* XXX: shared with pfsync */ 117 118VNET_DEFINE_STATIC(uma_zone_t, pf_frent_z); 119#define V_pf_frent_z VNET(pf_frent_z) 120VNET_DEFINE_STATIC(uma_zone_t, pf_frag_z); 121#define V_pf_frag_z VNET(pf_frag_z) 122 123TAILQ_HEAD(pf_fragqueue, pf_fragment); 124TAILQ_HEAD(pf_cachequeue, pf_fragment); 125VNET_DEFINE_STATIC(struct pf_fragqueue, pf_fragqueue); 126#define V_pf_fragqueue VNET(pf_fragqueue) 127RB_HEAD(pf_frag_tree, pf_fragment); 128VNET_DEFINE_STATIC(struct pf_frag_tree, pf_frag_tree); 129#define V_pf_frag_tree VNET(pf_frag_tree) 130static int pf_frag_compare(struct pf_fragment *, 131 struct pf_fragment *); 132static RB_PROTOTYPE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare); 133static RB_GENERATE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare); 134 135static void pf_flush_fragments(void); 136static void pf_free_fragment(struct pf_fragment *); 137static void pf_remove_fragment(struct pf_fragment *); 138 139static struct pf_frent *pf_create_fragment(u_short *); 140static int pf_frent_holes(struct pf_frent *frent); 141static struct pf_fragment *pf_find_fragment(struct pf_fragment_cmp *key, 142 struct pf_frag_tree *tree); 143static inline int pf_frent_index(struct pf_frent *); 144static int pf_frent_insert(struct pf_fragment *, 145 struct pf_frent *, struct pf_frent *); 146void pf_frent_remove(struct pf_fragment *, 147 struct pf_frent *); 148struct pf_frent *pf_frent_previous(struct pf_fragment *, 149 struct pf_frent *); 150static struct pf_fragment *pf_fillup_fragment(struct pf_fragment_cmp *, 151 struct pf_frent *, u_short *); 152static struct mbuf *pf_join_fragment(struct pf_fragment *); 153#ifdef INET 154static int pf_reassemble(struct mbuf **, struct ip *, int, u_short *); 155#endif /* INET */ 156#ifdef INET6 157static int pf_reassemble6(struct mbuf **, struct ip6_hdr *, 158 struct ip6_frag *, uint16_t, uint16_t, u_short *); 159#endif /* INET6 */ 160 161#define DPFPRINTF(x) do { \ 162 if (V_pf_status.debug >= PF_DEBUG_MISC) { \ 163 printf("%s: ", __func__); \ 164 printf x ; \ 165 } \ 166} while(0) 167 168#ifdef INET 169static void 170pf_ip2key(struct ip *ip, int dir, struct pf_fragment_cmp *key) 171{ 172 173 key->frc_src.v4 = ip->ip_src; 174 key->frc_dst.v4 = ip->ip_dst; 175 key->frc_af = AF_INET; 176 key->frc_proto = ip->ip_p; 177 key->frc_id = ip->ip_id; 178} 179#endif /* INET */ 180 181void 182pf_normalize_init(void) 183{ 184 185 V_pf_frag_z = uma_zcreate("pf frags", sizeof(struct pf_fragment), 186 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 187 V_pf_frent_z = uma_zcreate("pf frag entries", sizeof(struct pf_frent), 188 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 189 V_pf_state_scrub_z = uma_zcreate("pf state scrubs", 190 sizeof(struct pf_state_scrub), NULL, NULL, NULL, NULL, 191 UMA_ALIGN_PTR, 0); 192 193 mtx_init(&V_pf_frag_mtx, "pf fragments", NULL, MTX_DEF); 194 195 V_pf_limits[PF_LIMIT_FRAGS].zone = V_pf_frent_z; 196 V_pf_limits[PF_LIMIT_FRAGS].limit = PFFRAG_FRENT_HIWAT; 197 uma_zone_set_max(V_pf_frent_z, PFFRAG_FRENT_HIWAT); 198 uma_zone_set_warning(V_pf_frent_z, "PF frag entries limit reached"); 199 200 TAILQ_INIT(&V_pf_fragqueue); 201} 202 203void 204pf_normalize_cleanup(void) 205{ 206 207 uma_zdestroy(V_pf_state_scrub_z); 208 uma_zdestroy(V_pf_frent_z); 209 uma_zdestroy(V_pf_frag_z); 210 211 mtx_destroy(&V_pf_frag_mtx); 212} 213 214static int 215pf_frag_compare(struct pf_fragment *a, struct pf_fragment *b) 216{ 217 int diff; 218 219 if ((diff = a->fr_id - b->fr_id) != 0) 220 return (diff); 221 if ((diff = a->fr_proto - b->fr_proto) != 0) 222 return (diff); 223 if ((diff = a->fr_af - b->fr_af) != 0) 224 return (diff); 225 if ((diff = pf_addr_cmp(&a->fr_src, &b->fr_src, a->fr_af)) != 0) 226 return (diff); 227 if ((diff = pf_addr_cmp(&a->fr_dst, &b->fr_dst, a->fr_af)) != 0) 228 return (diff); 229 return (0); 230} 231 232void 233pf_purge_expired_fragments(void) 234{ 235 u_int32_t expire = time_uptime - 236 V_pf_default_rule.timeout[PFTM_FRAG]; 237 238 pf_purge_fragments(expire); 239} 240 241void 242pf_purge_fragments(uint32_t expire) 243{ 244 struct pf_fragment *frag; 245 246 PF_FRAG_LOCK(); 247 while ((frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue)) != NULL) { 248 if (frag->fr_timeout > expire) 249 break; 250 251 DPFPRINTF(("expiring %d(%p)\n", frag->fr_id, frag)); 252 pf_free_fragment(frag); 253 } 254 255 PF_FRAG_UNLOCK(); 256} 257 258/* 259 * Try to flush old fragments to make space for new ones 260 */ 261static void 262pf_flush_fragments(void) 263{ 264 struct pf_fragment *frag; 265 int goal; 266 267 PF_FRAG_ASSERT(); 268 269 goal = uma_zone_get_cur(V_pf_frent_z) * 9 / 10; 270 DPFPRINTF(("trying to free %d frag entriess\n", goal)); 271 while (goal < uma_zone_get_cur(V_pf_frent_z)) { 272 frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue); 273 if (frag) 274 pf_free_fragment(frag); 275 else 276 break; 277 } 278} 279 280/* Frees the fragments and all associated entries */ 281static void 282pf_free_fragment(struct pf_fragment *frag) 283{ 284 struct pf_frent *frent; 285 286 PF_FRAG_ASSERT(); 287 288 /* Free all fragments */ 289 for (frent = TAILQ_FIRST(&frag->fr_queue); frent; 290 frent = TAILQ_FIRST(&frag->fr_queue)) { 291 TAILQ_REMOVE(&frag->fr_queue, frent, fr_next); 292 293 m_freem(frent->fe_m); 294 uma_zfree(V_pf_frent_z, frent); 295 } 296 297 pf_remove_fragment(frag); 298} 299 300static struct pf_fragment * 301pf_find_fragment(struct pf_fragment_cmp *key, struct pf_frag_tree *tree) 302{ 303 struct pf_fragment *frag; 304 305 PF_FRAG_ASSERT(); 306 307 frag = RB_FIND(pf_frag_tree, tree, (struct pf_fragment *)key); 308 if (frag != NULL) { 309 /* XXX Are we sure we want to update the timeout? */ 310 frag->fr_timeout = time_uptime; 311 TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next); 312 TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next); 313 } 314 315 return (frag); 316} 317 318/* Removes a fragment from the fragment queue and frees the fragment */ 319static void 320pf_remove_fragment(struct pf_fragment *frag) 321{ 322 323 PF_FRAG_ASSERT(); 324 KASSERT(frag, ("frag != NULL")); 325 326 RB_REMOVE(pf_frag_tree, &V_pf_frag_tree, frag); 327 TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next); 328 uma_zfree(V_pf_frag_z, frag); 329} 330 331static struct pf_frent * 332pf_create_fragment(u_short *reason) 333{ 334 struct pf_frent *frent; 335 336 PF_FRAG_ASSERT(); 337 338 frent = uma_zalloc(V_pf_frent_z, M_NOWAIT); 339 if (frent == NULL) { 340 pf_flush_fragments(); 341 frent = uma_zalloc(V_pf_frent_z, M_NOWAIT); 342 if (frent == NULL) { 343 REASON_SET(reason, PFRES_MEMORY); 344 return (NULL); 345 } 346 } 347 348 return (frent); 349} 350 351/* 352 * Calculate the additional holes that were created in the fragment 353 * queue by inserting this fragment. A fragment in the middle 354 * creates one more hole by splitting. For each connected side, 355 * it loses one hole. 356 * Fragment entry must be in the queue when calling this function. 357 */ 358static int 359pf_frent_holes(struct pf_frent *frent) 360{ 361 struct pf_frent *prev = TAILQ_PREV(frent, pf_fragq, fr_next); 362 struct pf_frent *next = TAILQ_NEXT(frent, fr_next); 363 int holes = 1; 364 365 if (prev == NULL) { 366 if (frent->fe_off == 0) 367 holes--; 368 } else { 369 KASSERT(frent->fe_off != 0, ("frent->fe_off != 0")); 370 if (frent->fe_off == prev->fe_off + prev->fe_len) 371 holes--; 372 } 373 if (next == NULL) { 374 if (!frent->fe_mff) 375 holes--; 376 } else { 377 KASSERT(frent->fe_mff, ("frent->fe_mff")); 378 if (next->fe_off == frent->fe_off + frent->fe_len) 379 holes--; 380 } 381 return holes; 382} 383 384static inline int 385pf_frent_index(struct pf_frent *frent) 386{ 387 /* 388 * We have an array of 16 entry points to the queue. A full size 389 * 65535 octet IP packet can have 8192 fragments. So the queue 390 * traversal length is at most 512 and at most 16 entry points are 391 * checked. We need 128 additional bytes on a 64 bit architecture. 392 */ 393 CTASSERT(((u_int16_t)0xffff &~ 7) / (0x10000 / PF_FRAG_ENTRY_POINTS) == 394 16 - 1); 395 CTASSERT(((u_int16_t)0xffff >> 3) / PF_FRAG_ENTRY_POINTS == 512 - 1); 396 397 return frent->fe_off / (0x10000 / PF_FRAG_ENTRY_POINTS); 398} 399 400static int 401pf_frent_insert(struct pf_fragment *frag, struct pf_frent *frent, 402 struct pf_frent *prev) 403{ 404 int index; 405 406 CTASSERT(PF_FRAG_ENTRY_LIMIT <= 0xff); 407 408 /* 409 * A packet has at most 65536 octets. With 16 entry points, each one 410 * spawns 4096 octets. We limit these to 64 fragments each, which 411 * means on average every fragment must have at least 64 octets. 412 */ 413 index = pf_frent_index(frent); 414 if (frag->fr_entries[index] >= PF_FRAG_ENTRY_LIMIT) 415 return ENOBUFS; 416 frag->fr_entries[index]++; 417 418 if (prev == NULL) { 419 TAILQ_INSERT_HEAD(&frag->fr_queue, frent, fr_next); 420 } else { 421 KASSERT(prev->fe_off + prev->fe_len <= frent->fe_off, 422 ("overlapping fragment")); 423 TAILQ_INSERT_AFTER(&frag->fr_queue, prev, frent, fr_next); 424 } 425 426 if (frag->fr_firstoff[index] == NULL) { 427 KASSERT(prev == NULL || pf_frent_index(prev) < index, 428 ("prev == NULL || pf_frent_index(pref) < index")); 429 frag->fr_firstoff[index] = frent; 430 } else { 431 if (frent->fe_off < frag->fr_firstoff[index]->fe_off) { 432 KASSERT(prev == NULL || pf_frent_index(prev) < index, 433 ("prev == NULL || pf_frent_index(pref) < index")); 434 frag->fr_firstoff[index] = frent; 435 } else { 436 KASSERT(prev != NULL, ("prev != NULL")); 437 KASSERT(pf_frent_index(prev) == index, 438 ("pf_frent_index(prev) == index")); 439 } 440 } 441 442 frag->fr_holes += pf_frent_holes(frent); 443 444 return 0; 445} 446 447void 448pf_frent_remove(struct pf_fragment *frag, struct pf_frent *frent) 449{ 450#ifdef INVARIANTS 451 struct pf_frent *prev = TAILQ_PREV(frent, pf_fragq, fr_next); 452#endif 453 struct pf_frent *next = TAILQ_NEXT(frent, fr_next); 454 int index; 455 456 frag->fr_holes -= pf_frent_holes(frent); 457 458 index = pf_frent_index(frent); 459 KASSERT(frag->fr_firstoff[index] != NULL, ("frent not found")); 460 if (frag->fr_firstoff[index]->fe_off == frent->fe_off) { 461 if (next == NULL) { 462 frag->fr_firstoff[index] = NULL; 463 } else { 464 KASSERT(frent->fe_off + frent->fe_len <= next->fe_off, 465 ("overlapping fragment")); 466 if (pf_frent_index(next) == index) { 467 frag->fr_firstoff[index] = next; 468 } else { 469 frag->fr_firstoff[index] = NULL; 470 } 471 } 472 } else { 473 KASSERT(frag->fr_firstoff[index]->fe_off < frent->fe_off, 474 ("frag->fr_firstoff[index]->fe_off < frent->fe_off")); 475 KASSERT(prev != NULL, ("prev != NULL")); 476 KASSERT(prev->fe_off + prev->fe_len <= frent->fe_off, 477 ("overlapping fragment")); 478 KASSERT(pf_frent_index(prev) == index, 479 ("pf_frent_index(prev) == index")); 480 } 481 482 TAILQ_REMOVE(&frag->fr_queue, frent, fr_next); 483 484 KASSERT(frag->fr_entries[index] > 0, ("No fragments remaining")); 485 frag->fr_entries[index]--; 486} 487 488struct pf_frent * 489pf_frent_previous(struct pf_fragment *frag, struct pf_frent *frent) 490{ 491 struct pf_frent *prev, *next; 492 int index; 493 494 /* 495 * If there are no fragments after frag, take the final one. Assume 496 * that the global queue is not empty. 497 */ 498 prev = TAILQ_LAST(&frag->fr_queue, pf_fragq); 499 KASSERT(prev != NULL, ("prev != NULL")); 500 if (prev->fe_off <= frent->fe_off) 501 return prev; 502 /* 503 * We want to find a fragment entry that is before frag, but still 504 * close to it. Find the first fragment entry that is in the same 505 * entry point or in the first entry point after that. As we have 506 * already checked that there are entries behind frag, this will 507 * succeed. 508 */ 509 for (index = pf_frent_index(frent); index < PF_FRAG_ENTRY_POINTS; 510 index++) { 511 prev = frag->fr_firstoff[index]; 512 if (prev != NULL) 513 break; 514 } 515 KASSERT(prev != NULL, ("prev != NULL")); 516 /* 517 * In prev we may have a fragment from the same entry point that is 518 * before frent, or one that is just one position behind frent. 519 * In the latter case, we go back one step and have the predecessor. 520 * There may be none if the new fragment will be the first one. 521 */ 522 if (prev->fe_off > frent->fe_off) { 523 prev = TAILQ_PREV(prev, pf_fragq, fr_next); 524 if (prev == NULL) 525 return NULL; 526 KASSERT(prev->fe_off <= frent->fe_off, 527 ("prev->fe_off <= frent->fe_off")); 528 return prev; 529 } 530 /* 531 * In prev is the first fragment of the entry point. The offset 532 * of frag is behind it. Find the closest previous fragment. 533 */ 534 for (next = TAILQ_NEXT(prev, fr_next); next != NULL; 535 next = TAILQ_NEXT(next, fr_next)) { 536 if (next->fe_off > frent->fe_off) 537 break; 538 prev = next; 539 } 540 return prev; 541} 542 543static struct pf_fragment * 544pf_fillup_fragment(struct pf_fragment_cmp *key, struct pf_frent *frent, 545 u_short *reason) 546{ 547 struct pf_frent *after, *next, *prev; 548 struct pf_fragment *frag; 549 uint16_t total; 550 int old_index, new_index; 551 552 PF_FRAG_ASSERT(); 553 554 /* No empty fragments. */ 555 if (frent->fe_len == 0) { 556 DPFPRINTF(("bad fragment: len 0\n")); 557 goto bad_fragment; 558 } 559 560 /* All fragments are 8 byte aligned. */ 561 if (frent->fe_mff && (frent->fe_len & 0x7)) { 562 DPFPRINTF(("bad fragment: mff and len %d\n", frent->fe_len)); 563 goto bad_fragment; 564 } 565 566 /* Respect maximum length, IP_MAXPACKET == IPV6_MAXPACKET. */ 567 if (frent->fe_off + frent->fe_len > IP_MAXPACKET) { 568 DPFPRINTF(("bad fragment: max packet %d\n", 569 frent->fe_off + frent->fe_len)); 570 goto bad_fragment; 571 } 572 573 DPFPRINTF((key->frc_af == AF_INET ? 574 "reass frag %d @ %d-%d\n" : "reass frag %#08x @ %d-%d\n", 575 key->frc_id, frent->fe_off, frent->fe_off + frent->fe_len)); 576 577 /* Fully buffer all of the fragments in this fragment queue. */ 578 frag = pf_find_fragment(key, &V_pf_frag_tree); 579 580 /* Create a new reassembly queue for this packet. */ 581 if (frag == NULL) { 582 frag = uma_zalloc(V_pf_frag_z, M_NOWAIT); 583 if (frag == NULL) { 584 pf_flush_fragments(); 585 frag = uma_zalloc(V_pf_frag_z, M_NOWAIT); 586 if (frag == NULL) { 587 REASON_SET(reason, PFRES_MEMORY); 588 goto drop_fragment; 589 } 590 } 591 592 *(struct pf_fragment_cmp *)frag = *key; 593 memset(frag->fr_firstoff, 0, sizeof(frag->fr_firstoff)); 594 memset(frag->fr_entries, 0, sizeof(frag->fr_entries)); 595 frag->fr_timeout = time_uptime; 596 frag->fr_maxlen = frent->fe_len; 597 frag->fr_holes = 1; 598 TAILQ_INIT(&frag->fr_queue); 599 600 RB_INSERT(pf_frag_tree, &V_pf_frag_tree, frag); 601 TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next); 602 603 /* We do not have a previous fragment, cannot fail. */ 604 pf_frent_insert(frag, frent, NULL); 605 606 return (frag); 607 } 608 609 KASSERT(!TAILQ_EMPTY(&frag->fr_queue), ("!TAILQ_EMPTY()->fr_queue")); 610 611 /* Remember maximum fragment len for refragmentation. */ 612 if (frent->fe_len > frag->fr_maxlen) 613 frag->fr_maxlen = frent->fe_len; 614 615 /* Maximum data we have seen already. */ 616 total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off + 617 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len; 618 619 /* Non terminal fragments must have more fragments flag. */ 620 if (frent->fe_off + frent->fe_len < total && !frent->fe_mff) 621 goto bad_fragment; 622 623 /* Check if we saw the last fragment already. */ 624 if (!TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_mff) { 625 if (frent->fe_off + frent->fe_len > total || 626 (frent->fe_off + frent->fe_len == total && frent->fe_mff)) 627 goto bad_fragment; 628 } else { 629 if (frent->fe_off + frent->fe_len == total && !frent->fe_mff) 630 goto bad_fragment; 631 } 632 633 /* Find neighbors for newly inserted fragment */ 634 prev = pf_frent_previous(frag, frent); 635 if (prev == NULL) { 636 after = TAILQ_FIRST(&frag->fr_queue); 637 KASSERT(after != NULL, ("after != NULL")); 638 } else { 639 after = TAILQ_NEXT(prev, fr_next); 640 } 641 642 if (prev != NULL && prev->fe_off + prev->fe_len > frent->fe_off) { 643 uint16_t precut; 644 645 precut = prev->fe_off + prev->fe_len - frent->fe_off; 646 if (precut >= frent->fe_len) 647 goto bad_fragment; 648 DPFPRINTF(("overlap -%d\n", precut)); 649 m_adj(frent->fe_m, precut); 650 frent->fe_off += precut; 651 frent->fe_len -= precut; 652 } 653 654 for (; after != NULL && frent->fe_off + frent->fe_len > after->fe_off; 655 after = next) { 656 uint16_t aftercut; 657 658 aftercut = frent->fe_off + frent->fe_len - after->fe_off; 659 DPFPRINTF(("adjust overlap %d\n", aftercut)); 660 if (aftercut < after->fe_len) { 661 m_adj(after->fe_m, aftercut); 662 old_index = pf_frent_index(after); 663 after->fe_off += aftercut; 664 after->fe_len -= aftercut; 665 new_index = pf_frent_index(after); 666 if (old_index != new_index) { 667 DPFPRINTF(("frag index %d, new %d", 668 old_index, new_index)); 669 /* Fragment switched queue as fe_off changed */ 670 after->fe_off -= aftercut; 671 after->fe_len += aftercut; 672 /* Remove restored fragment from old queue */ 673 pf_frent_remove(frag, after); 674 after->fe_off += aftercut; 675 after->fe_len -= aftercut; 676 /* Insert into correct queue */ 677 if (pf_frent_insert(frag, after, prev)) { 678 DPFPRINTF( 679 ("fragment requeue limit exceeded")); 680 m_freem(after->fe_m); 681 uma_zfree(V_pf_frent_z, after); 682 /* There is not way to recover */ 683 goto bad_fragment; 684 } 685 } 686 break; 687 } 688 689 /* This fragment is completely overlapped, lose it. */ 690 next = TAILQ_NEXT(after, fr_next); 691 pf_frent_remove(frag, after); 692 m_freem(after->fe_m); 693 uma_zfree(V_pf_frent_z, after); 694 } 695 696 /* If part of the queue gets too long, there is not way to recover. */ 697 if (pf_frent_insert(frag, frent, prev)) { 698 DPFPRINTF(("fragment queue limit exceeded\n")); 699 goto bad_fragment; 700 } 701 702 return (frag); 703 704bad_fragment: 705 REASON_SET(reason, PFRES_FRAG); 706drop_fragment: 707 uma_zfree(V_pf_frent_z, frent); 708 return (NULL); 709} 710 711static struct mbuf * 712pf_join_fragment(struct pf_fragment *frag) 713{ 714 struct mbuf *m, *m2; 715 struct pf_frent *frent, *next; 716 717 frent = TAILQ_FIRST(&frag->fr_queue); 718 next = TAILQ_NEXT(frent, fr_next); 719 720 m = frent->fe_m; 721 m_adj(m, (frent->fe_hdrlen + frent->fe_len) - m->m_pkthdr.len); 722 uma_zfree(V_pf_frent_z, frent); 723 for (frent = next; frent != NULL; frent = next) { 724 next = TAILQ_NEXT(frent, fr_next); 725 726 m2 = frent->fe_m; 727 /* Strip off ip header. */ 728 m_adj(m2, frent->fe_hdrlen); 729 /* Strip off any trailing bytes. */ 730 m_adj(m2, frent->fe_len - m2->m_pkthdr.len); 731 732 uma_zfree(V_pf_frent_z, frent); 733 m_cat(m, m2); 734 } 735 736 /* Remove from fragment queue. */ 737 pf_remove_fragment(frag); 738 739 return (m); 740} 741 742#ifdef INET 743static int 744pf_reassemble(struct mbuf **m0, struct ip *ip, int dir, u_short *reason) 745{ 746 struct mbuf *m = *m0; 747 struct pf_frent *frent; 748 struct pf_fragment *frag; 749 struct pf_fragment_cmp key; 750 uint16_t total, hdrlen; 751 752 /* Get an entry for the fragment queue */ 753 if ((frent = pf_create_fragment(reason)) == NULL) 754 return (PF_DROP); 755 756 frent->fe_m = m; 757 frent->fe_hdrlen = ip->ip_hl << 2; 758 frent->fe_extoff = 0; 759 frent->fe_len = ntohs(ip->ip_len) - (ip->ip_hl << 2); 760 frent->fe_off = (ntohs(ip->ip_off) & IP_OFFMASK) << 3; 761 frent->fe_mff = ntohs(ip->ip_off) & IP_MF; 762 763 pf_ip2key(ip, dir, &key); 764 765 if ((frag = pf_fillup_fragment(&key, frent, reason)) == NULL) 766 return (PF_DROP); 767 768 /* The mbuf is part of the fragment entry, no direct free or access */ 769 m = *m0 = NULL; 770 771 if (frag->fr_holes) { 772 DPFPRINTF(("frag %d, holes %d\n", frag->fr_id, frag->fr_holes)); 773 return (PF_PASS); /* drop because *m0 is NULL, no error */ 774 } 775 776 /* We have all the data */ 777 frent = TAILQ_FIRST(&frag->fr_queue); 778 KASSERT(frent != NULL, ("frent != NULL")); 779 total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off + 780 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len; 781 hdrlen = frent->fe_hdrlen; 782 783 m = *m0 = pf_join_fragment(frag); 784 frag = NULL; 785 786 if (m->m_flags & M_PKTHDR) { 787 int plen = 0; 788 for (m = *m0; m; m = m->m_next) 789 plen += m->m_len; 790 m = *m0; 791 m->m_pkthdr.len = plen; 792 } 793 794 ip = mtod(m, struct ip *); 795 ip->ip_sum = pf_cksum_fixup(ip->ip_sum, ip->ip_len, 796 htons(hdrlen + total), 0); 797 ip->ip_len = htons(hdrlen + total); 798 ip->ip_sum = pf_cksum_fixup(ip->ip_sum, ip->ip_off, 799 ip->ip_off & ~(IP_MF|IP_OFFMASK), 0); 800 ip->ip_off &= ~(IP_MF|IP_OFFMASK); 801 802 if (hdrlen + total > IP_MAXPACKET) { 803 DPFPRINTF(("drop: too big: %d\n", total)); 804 ip->ip_len = 0; 805 REASON_SET(reason, PFRES_SHORT); 806 /* PF_DROP requires a valid mbuf *m0 in pf_test() */ 807 return (PF_DROP); 808 } 809 810 DPFPRINTF(("complete: %p(%d)\n", m, ntohs(ip->ip_len))); 811 return (PF_PASS); 812} 813#endif /* INET */ 814 815#ifdef INET6 816static int 817pf_reassemble6(struct mbuf **m0, struct ip6_hdr *ip6, struct ip6_frag *fraghdr, 818 uint16_t hdrlen, uint16_t extoff, u_short *reason) 819{ 820 struct mbuf *m = *m0; 821 struct pf_frent *frent; 822 struct pf_fragment *frag; 823 struct pf_fragment_cmp key; 824 struct m_tag *mtag; 825 struct pf_fragment_tag *ftag; 826 int off; 827 uint32_t frag_id; 828 uint16_t total, maxlen; 829 uint8_t proto; 830 831 PF_FRAG_LOCK(); 832 833 /* Get an entry for the fragment queue. */ 834 if ((frent = pf_create_fragment(reason)) == NULL) { 835 PF_FRAG_UNLOCK(); 836 return (PF_DROP); 837 } 838 839 frent->fe_m = m; 840 frent->fe_hdrlen = hdrlen; 841 frent->fe_extoff = extoff; 842 frent->fe_len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - hdrlen; 843 frent->fe_off = ntohs(fraghdr->ip6f_offlg & IP6F_OFF_MASK); 844 frent->fe_mff = fraghdr->ip6f_offlg & IP6F_MORE_FRAG; 845 846 key.frc_src.v6 = ip6->ip6_src; 847 key.frc_dst.v6 = ip6->ip6_dst; 848 key.frc_af = AF_INET6; 849 /* Only the first fragment's protocol is relevant. */ 850 key.frc_proto = 0; 851 key.frc_id = fraghdr->ip6f_ident; 852 853 if ((frag = pf_fillup_fragment(&key, frent, reason)) == NULL) { 854 PF_FRAG_UNLOCK(); 855 return (PF_DROP); 856 } 857 858 /* The mbuf is part of the fragment entry, no direct free or access. */ 859 m = *m0 = NULL; 860 861 if (frag->fr_holes) { 862 DPFPRINTF(("frag %d, holes %d\n", frag->fr_id, 863 frag->fr_holes)); 864 PF_FRAG_UNLOCK(); 865 return (PF_PASS); /* Drop because *m0 is NULL, no error. */ 866 } 867 868 /* We have all the data. */ 869 frent = TAILQ_FIRST(&frag->fr_queue); 870 KASSERT(frent != NULL, ("frent != NULL")); 871 extoff = frent->fe_extoff; 872 maxlen = frag->fr_maxlen; 873 frag_id = frag->fr_id; 874 total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off + 875 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len; 876 hdrlen = frent->fe_hdrlen - sizeof(struct ip6_frag); 877 878 m = *m0 = pf_join_fragment(frag); 879 frag = NULL; 880 881 PF_FRAG_UNLOCK(); 882 883 /* Take protocol from first fragment header. */ 884 m = m_getptr(m, hdrlen + offsetof(struct ip6_frag, ip6f_nxt), &off); 885 KASSERT(m, ("%s: short mbuf chain", __func__)); 886 proto = *(mtod(m, uint8_t *) + off); 887 m = *m0; 888 889 /* Delete frag6 header */ 890 if (ip6_deletefraghdr(m, hdrlen, M_NOWAIT) != 0) 891 goto fail; 892 893 if (m->m_flags & M_PKTHDR) { 894 int plen = 0; 895 for (m = *m0; m; m = m->m_next) 896 plen += m->m_len; 897 m = *m0; 898 m->m_pkthdr.len = plen; 899 } 900 901 if ((mtag = m_tag_get(PACKET_TAG_PF_REASSEMBLED, 902 sizeof(struct pf_fragment_tag), M_NOWAIT)) == NULL) 903 goto fail; 904 ftag = (struct pf_fragment_tag *)(mtag + 1); 905 ftag->ft_hdrlen = hdrlen; 906 ftag->ft_extoff = extoff; 907 ftag->ft_maxlen = maxlen; 908 ftag->ft_id = frag_id; 909 m_tag_prepend(m, mtag); 910 911 ip6 = mtod(m, struct ip6_hdr *); 912 ip6->ip6_plen = htons(hdrlen - sizeof(struct ip6_hdr) + total); 913 if (extoff) { 914 /* Write protocol into next field of last extension header. */ 915 m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt), 916 &off); 917 KASSERT(m, ("%s: short mbuf chain", __func__)); 918 *(mtod(m, char *) + off) = proto; 919 m = *m0; 920 } else 921 ip6->ip6_nxt = proto; 922 923 if (hdrlen - sizeof(struct ip6_hdr) + total > IPV6_MAXPACKET) { 924 DPFPRINTF(("drop: too big: %d\n", total)); 925 ip6->ip6_plen = 0; 926 REASON_SET(reason, PFRES_SHORT); 927 /* PF_DROP requires a valid mbuf *m0 in pf_test6(). */ 928 return (PF_DROP); 929 } 930 931 DPFPRINTF(("complete: %p(%d)\n", m, ntohs(ip6->ip6_plen))); 932 return (PF_PASS); 933 934fail: 935 REASON_SET(reason, PFRES_MEMORY); 936 /* PF_DROP requires a valid mbuf *m0 in pf_test6(), will free later. */ 937 return (PF_DROP); 938} 939#endif /* INET6 */ 940 941#ifdef INET6 942int 943pf_max_frag_size(struct mbuf *m) 944{ 945 struct m_tag *tag; 946 struct pf_fragment_tag *ftag; 947 948 tag = m_tag_find(m, PACKET_TAG_PF_REASSEMBLED, NULL); 949 if (tag == NULL) 950 return (m->m_pkthdr.len); 951 952 ftag = (struct pf_fragment_tag *)(tag + 1); 953 954 return (ftag->ft_maxlen); 955} 956 957int 958pf_refragment6(struct ifnet *ifp, struct mbuf **m0, struct m_tag *mtag, 959 bool forward) 960{ 961 struct mbuf *m = *m0, *t; 962 struct ip6_hdr *hdr; 963 struct pf_fragment_tag *ftag = (struct pf_fragment_tag *)(mtag + 1); 964 struct pf_pdesc pd; 965 uint32_t frag_id; 966 uint16_t hdrlen, extoff, maxlen; 967 uint8_t proto; 968 int error, action; 969 970 hdrlen = ftag->ft_hdrlen; 971 extoff = ftag->ft_extoff; 972 maxlen = ftag->ft_maxlen; 973 frag_id = ftag->ft_id; 974 m_tag_delete(m, mtag); 975 mtag = NULL; 976 ftag = NULL; 977 978 if (extoff) { 979 int off; 980 981 /* Use protocol from next field of last extension header */ 982 m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt), 983 &off); 984 KASSERT((m != NULL), ("pf_refragment6: short mbuf chain")); 985 proto = *(mtod(m, uint8_t *) + off); 986 *(mtod(m, char *) + off) = IPPROTO_FRAGMENT; 987 m = *m0; 988 } else { 989 hdr = mtod(m, struct ip6_hdr *); 990 proto = hdr->ip6_nxt; 991 hdr->ip6_nxt = IPPROTO_FRAGMENT; 992 } 993 994 /* In case of link-local traffic we'll need a scope set. */ 995 hdr = mtod(m, struct ip6_hdr *); 996 997 in6_setscope(&hdr->ip6_src, ifp, NULL); 998 in6_setscope(&hdr->ip6_dst, ifp, NULL); 999 1000 /* The MTU must be a multiple of 8 bytes, or we risk doing the 1001 * fragmentation wrong. */ 1002 maxlen = maxlen & ~7; 1003 1004 /* 1005 * Maxlen may be less than 8 if there was only a single 1006 * fragment. As it was fragmented before, add a fragment 1007 * header also for a single fragment. If total or maxlen 1008 * is less than 8, ip6_fragment() will return EMSGSIZE and 1009 * we drop the packet. 1010 */ 1011 error = ip6_fragment(ifp, m, hdrlen, proto, maxlen, frag_id); 1012 m = (*m0)->m_nextpkt; 1013 (*m0)->m_nextpkt = NULL; 1014 if (error == 0) { 1015 /* The first mbuf contains the unfragmented packet. */ 1016 m_freem(*m0); 1017 *m0 = NULL; 1018 action = PF_PASS; 1019 } else { 1020 /* Drop expects an mbuf to free. */ 1021 DPFPRINTF(("refragment error %d\n", error)); 1022 action = PF_DROP; 1023 } 1024 for (; m; m = t) { 1025 t = m->m_nextpkt; 1026 m->m_nextpkt = NULL; 1027 m->m_flags |= M_SKIP_FIREWALL; 1028 memset(&pd, 0, sizeof(pd)); 1029 pd.pf_mtag = pf_find_mtag(m); 1030 if (error == 0) 1031 if (forward) { 1032 MPASS(m->m_pkthdr.rcvif != NULL); 1033 ip6_forward(m, 0); 1034 } else { 1035 (void)ip6_output(m, NULL, NULL, 0, NULL, NULL, 1036 NULL); 1037 } 1038 else 1039 m_freem(m); 1040 } 1041 1042 return (action); 1043} 1044#endif /* INET6 */ 1045 1046#ifdef INET 1047int 1048pf_normalize_ip(struct mbuf **m0, struct pfi_kkif *kif, u_short *reason, 1049 struct pf_pdesc *pd) 1050{ 1051 struct mbuf *m = *m0; 1052 struct pf_krule *r; 1053 struct ip *h = mtod(m, struct ip *); 1054 int mff = (ntohs(h->ip_off) & IP_MF); 1055 int hlen = h->ip_hl << 2; 1056 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3; 1057 u_int16_t max; 1058 int ip_len; 1059 int tag = -1; 1060 int verdict; 1061 bool scrub_compat; 1062 1063 PF_RULES_RASSERT(); 1064 1065 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr); 1066 /* 1067 * Check if there are any scrub rules, matching or not. 1068 * Lack of scrub rules means: 1069 * - enforced packet normalization operation just like in OpenBSD 1070 * - fragment reassembly depends on V_pf_status.reass 1071 * With scrub rules: 1072 * - packet normalization is performed if there is a matching scrub rule 1073 * - fragment reassembly is performed if the matching rule has no 1074 * PFRULE_FRAGMENT_NOREASS flag 1075 */ 1076 scrub_compat = (r != NULL); 1077 while (r != NULL) { 1078 pf_counter_u64_add(&r->evaluations, 1); 1079 if (pfi_kkif_match(r->kif, kif) == r->ifnot) 1080 r = r->skip[PF_SKIP_IFP].ptr; 1081 else if (r->direction && r->direction != pd->dir) 1082 r = r->skip[PF_SKIP_DIR].ptr; 1083 else if (r->af && r->af != AF_INET) 1084 r = r->skip[PF_SKIP_AF].ptr; 1085 else if (r->proto && r->proto != h->ip_p) 1086 r = r->skip[PF_SKIP_PROTO].ptr; 1087 else if (PF_MISMATCHAW(&r->src.addr, 1088 (struct pf_addr *)&h->ip_src.s_addr, AF_INET, 1089 r->src.neg, kif, M_GETFIB(m))) 1090 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 1091 else if (PF_MISMATCHAW(&r->dst.addr, 1092 (struct pf_addr *)&h->ip_dst.s_addr, AF_INET, 1093 r->dst.neg, NULL, M_GETFIB(m))) 1094 r = r->skip[PF_SKIP_DST_ADDR].ptr; 1095 else if (r->match_tag && !pf_match_tag(m, r, &tag, 1096 pd->pf_mtag ? pd->pf_mtag->tag : 0)) 1097 r = TAILQ_NEXT(r, entries); 1098 else 1099 break; 1100 } 1101 1102 if (scrub_compat) { 1103 /* With scrub rules present IPv4 normalization happens only 1104 * if one of rules has matched and it's not a "no scrub" rule */ 1105 if (r == NULL || r->action == PF_NOSCRUB) 1106 return (PF_PASS); 1107 1108 pf_counter_u64_critical_enter(); 1109 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1); 1110 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len); 1111 pf_counter_u64_critical_exit(); 1112 pf_rule_to_actions(r, &pd->act); 1113 } 1114 1115 /* Check for illegal packets */ 1116 if (hlen < (int)sizeof(struct ip)) { 1117 REASON_SET(reason, PFRES_NORM); 1118 goto drop; 1119 } 1120 1121 if (hlen > ntohs(h->ip_len)) { 1122 REASON_SET(reason, PFRES_NORM); 1123 goto drop; 1124 } 1125 1126 /* Clear IP_DF if the rule uses the no-df option or we're in no-df mode */ 1127 if (((!scrub_compat && V_pf_status.reass & PF_REASS_NODF) || 1128 (r != NULL && r->rule_flag & PFRULE_NODF)) && 1129 (h->ip_off & htons(IP_DF)) 1130 ) { 1131 u_int16_t ip_off = h->ip_off; 1132 1133 h->ip_off &= htons(~IP_DF); 1134 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0); 1135 } 1136 1137 /* We will need other tests here */ 1138 if (!fragoff && !mff) 1139 goto no_fragment; 1140 1141 /* We're dealing with a fragment now. Don't allow fragments 1142 * with IP_DF to enter the cache. If the flag was cleared by 1143 * no-df above, fine. Otherwise drop it. 1144 */ 1145 if (h->ip_off & htons(IP_DF)) { 1146 DPFPRINTF(("IP_DF\n")); 1147 goto bad; 1148 } 1149 1150 ip_len = ntohs(h->ip_len) - hlen; 1151 1152 /* All fragments are 8 byte aligned */ 1153 if (mff && (ip_len & 0x7)) { 1154 DPFPRINTF(("mff and %d\n", ip_len)); 1155 goto bad; 1156 } 1157 1158 /* Respect maximum length */ 1159 if (fragoff + ip_len > IP_MAXPACKET) { 1160 DPFPRINTF(("max packet %d\n", fragoff + ip_len)); 1161 goto bad; 1162 } 1163 1164 if ((!scrub_compat && V_pf_status.reass) || 1165 (r != NULL && !(r->rule_flag & PFRULE_FRAGMENT_NOREASS)) 1166 ) { 1167 max = fragoff + ip_len; 1168 1169 /* Fully buffer all of the fragments 1170 * Might return a completely reassembled mbuf, or NULL */ 1171 PF_FRAG_LOCK(); 1172 DPFPRINTF(("reass frag %d @ %d-%d\n", h->ip_id, fragoff, max)); 1173 verdict = pf_reassemble(m0, h, pd->dir, reason); 1174 PF_FRAG_UNLOCK(); 1175 1176 if (verdict != PF_PASS) 1177 return (PF_DROP); 1178 1179 m = *m0; 1180 if (m == NULL) 1181 return (PF_DROP); 1182 1183 h = mtod(m, struct ip *); 1184 1185 no_fragment: 1186 /* At this point, only IP_DF is allowed in ip_off */ 1187 if (h->ip_off & ~htons(IP_DF)) { 1188 u_int16_t ip_off = h->ip_off; 1189 1190 h->ip_off &= htons(IP_DF); 1191 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0); 1192 } 1193 } 1194 1195 return (PF_PASS); 1196 1197 bad: 1198 DPFPRINTF(("dropping bad fragment\n")); 1199 REASON_SET(reason, PFRES_FRAG); 1200 drop: 1201 if (r != NULL && r->log) 1202 PFLOG_PACKET(kif, m, AF_INET, PF_DROP, *reason, r, NULL, NULL, pd, 1); 1203 1204 return (PF_DROP); 1205} 1206#endif 1207 1208#ifdef INET6 1209int 1210pf_normalize_ip6(struct mbuf **m0, struct pfi_kkif *kif, 1211 u_short *reason, struct pf_pdesc *pd) 1212{ 1213 struct mbuf *m = *m0; 1214 struct pf_krule *r; 1215 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 1216 int extoff; 1217 int off; 1218 struct ip6_ext ext; 1219 struct ip6_opt opt; 1220 struct ip6_frag frag; 1221 u_int32_t plen; 1222 int optend; 1223 int ooff; 1224 u_int8_t proto; 1225 int terminal; 1226 bool scrub_compat; 1227 1228 PF_RULES_RASSERT(); 1229 1230 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr); 1231 /* 1232 * Check if there are any scrub rules, matching or not. 1233 * Lack of scrub rules means: 1234 * - enforced packet normalization operation just like in OpenBSD 1235 * With scrub rules: 1236 * - packet normalization is performed if there is a matching scrub rule 1237 * XXX: Fragment reassembly always performed for IPv6! 1238 */ 1239 scrub_compat = (r != NULL); 1240 while (r != NULL) { 1241 pf_counter_u64_add(&r->evaluations, 1); 1242 if (pfi_kkif_match(r->kif, kif) == r->ifnot) 1243 r = r->skip[PF_SKIP_IFP].ptr; 1244 else if (r->direction && r->direction != pd->dir) 1245 r = r->skip[PF_SKIP_DIR].ptr; 1246 else if (r->af && r->af != AF_INET6) 1247 r = r->skip[PF_SKIP_AF].ptr; 1248#if 0 /* header chain! */ 1249 else if (r->proto && r->proto != h->ip6_nxt) 1250 r = r->skip[PF_SKIP_PROTO].ptr; 1251#endif 1252 else if (PF_MISMATCHAW(&r->src.addr, 1253 (struct pf_addr *)&h->ip6_src, AF_INET6, 1254 r->src.neg, kif, M_GETFIB(m))) 1255 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 1256 else if (PF_MISMATCHAW(&r->dst.addr, 1257 (struct pf_addr *)&h->ip6_dst, AF_INET6, 1258 r->dst.neg, NULL, M_GETFIB(m))) 1259 r = r->skip[PF_SKIP_DST_ADDR].ptr; 1260 else 1261 break; 1262 } 1263 1264 if (scrub_compat) { 1265 /* With scrub rules present IPv6 normalization happens only 1266 * if one of rules has matched and it's not a "no scrub" rule */ 1267 if (r == NULL || r->action == PF_NOSCRUB) 1268 return (PF_PASS); 1269 1270 pf_counter_u64_critical_enter(); 1271 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1); 1272 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len); 1273 pf_counter_u64_critical_exit(); 1274 pf_rule_to_actions(r, &pd->act); 1275 } 1276 1277 /* Check for illegal packets */ 1278 if (sizeof(struct ip6_hdr) + IPV6_MAXPACKET < m->m_pkthdr.len) 1279 goto drop; 1280 1281again: 1282 h = mtod(m, struct ip6_hdr *); 1283 plen = ntohs(h->ip6_plen); 1284 /* jumbo payload option not supported */ 1285 if (plen == 0) 1286 goto drop; 1287 1288 extoff = 0; 1289 off = sizeof(struct ip6_hdr); 1290 proto = h->ip6_nxt; 1291 terminal = 0; 1292 do { 1293 switch (proto) { 1294 case IPPROTO_FRAGMENT: 1295 goto fragment; 1296 break; 1297 case IPPROTO_AH: 1298 case IPPROTO_ROUTING: 1299 case IPPROTO_DSTOPTS: 1300 if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL, 1301 NULL, AF_INET6)) 1302 goto shortpkt; 1303 extoff = off; 1304 if (proto == IPPROTO_AH) 1305 off += (ext.ip6e_len + 2) * 4; 1306 else 1307 off += (ext.ip6e_len + 1) * 8; 1308 proto = ext.ip6e_nxt; 1309 break; 1310 case IPPROTO_HOPOPTS: 1311 if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL, 1312 NULL, AF_INET6)) 1313 goto shortpkt; 1314 extoff = off; 1315 optend = off + (ext.ip6e_len + 1) * 8; 1316 ooff = off + sizeof(ext); 1317 do { 1318 if (!pf_pull_hdr(m, ooff, &opt.ip6o_type, 1319 sizeof(opt.ip6o_type), NULL, NULL, 1320 AF_INET6)) 1321 goto shortpkt; 1322 if (opt.ip6o_type == IP6OPT_PAD1) { 1323 ooff++; 1324 continue; 1325 } 1326 if (!pf_pull_hdr(m, ooff, &opt, sizeof(opt), 1327 NULL, NULL, AF_INET6)) 1328 goto shortpkt; 1329 if (ooff + sizeof(opt) + opt.ip6o_len > optend) 1330 goto drop; 1331 if (opt.ip6o_type == IP6OPT_JUMBO) 1332 goto drop; 1333 ooff += sizeof(opt) + opt.ip6o_len; 1334 } while (ooff < optend); 1335 1336 off = optend; 1337 proto = ext.ip6e_nxt; 1338 break; 1339 default: 1340 terminal = 1; 1341 break; 1342 } 1343 } while (!terminal); 1344 1345 if (sizeof(struct ip6_hdr) + plen > m->m_pkthdr.len) 1346 goto shortpkt; 1347 1348 return (PF_PASS); 1349 1350 fragment: 1351 if (pd->flags & PFDESC_IP_REAS) 1352 return (PF_DROP); 1353 if (sizeof(struct ip6_hdr) + plen > m->m_pkthdr.len) 1354 goto shortpkt; 1355 1356 if (!pf_pull_hdr(m, off, &frag, sizeof(frag), NULL, NULL, AF_INET6)) 1357 goto shortpkt; 1358 1359 /* Offset now points to data portion. */ 1360 off += sizeof(frag); 1361 1362 /* Returns PF_DROP or *m0 is NULL or completely reassembled mbuf. */ 1363 if (pf_reassemble6(m0, h, &frag, off, extoff, reason) != PF_PASS) 1364 return (PF_DROP); 1365 m = *m0; 1366 if (m == NULL) 1367 return (PF_DROP); 1368 1369 pd->flags |= PFDESC_IP_REAS; 1370 goto again; 1371 1372 shortpkt: 1373 REASON_SET(reason, PFRES_SHORT); 1374 if (r != NULL && r->log) 1375 PFLOG_PACKET(kif, m, AF_INET6, PF_DROP, *reason, r, NULL, NULL, pd, 1); 1376 return (PF_DROP); 1377 1378 drop: 1379 REASON_SET(reason, PFRES_NORM); 1380 if (r != NULL && r->log) 1381 PFLOG_PACKET(kif, m, AF_INET6, PF_DROP, *reason, r, NULL, NULL, pd, 1); 1382 return (PF_DROP); 1383} 1384#endif /* INET6 */ 1385 1386int 1387pf_normalize_tcp(struct pfi_kkif *kif, struct mbuf *m, int ipoff, 1388 int off, void *h, struct pf_pdesc *pd) 1389{ 1390 struct pf_krule *r, *rm = NULL; 1391 struct tcphdr *th = &pd->hdr.tcp; 1392 int rewrite = 0; 1393 u_short reason; 1394 u_int16_t flags; 1395 sa_family_t af = pd->af; 1396 int srs; 1397 1398 PF_RULES_RASSERT(); 1399 1400 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr); 1401 /* Check if there any scrub rules. Lack of scrub rules means enforced 1402 * packet normalization operation just like in OpenBSD. */ 1403 srs = (r != NULL); 1404 while (r != NULL) { 1405 pf_counter_u64_add(&r->evaluations, 1); 1406 if (pfi_kkif_match(r->kif, kif) == r->ifnot) 1407 r = r->skip[PF_SKIP_IFP].ptr; 1408 else if (r->direction && r->direction != pd->dir) 1409 r = r->skip[PF_SKIP_DIR].ptr; 1410 else if (r->af && r->af != af) 1411 r = r->skip[PF_SKIP_AF].ptr; 1412 else if (r->proto && r->proto != pd->proto) 1413 r = r->skip[PF_SKIP_PROTO].ptr; 1414 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af, 1415 r->src.neg, kif, M_GETFIB(m))) 1416 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 1417 else if (r->src.port_op && !pf_match_port(r->src.port_op, 1418 r->src.port[0], r->src.port[1], th->th_sport)) 1419 r = r->skip[PF_SKIP_SRC_PORT].ptr; 1420 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af, 1421 r->dst.neg, NULL, M_GETFIB(m))) 1422 r = r->skip[PF_SKIP_DST_ADDR].ptr; 1423 else if (r->dst.port_op && !pf_match_port(r->dst.port_op, 1424 r->dst.port[0], r->dst.port[1], th->th_dport)) 1425 r = r->skip[PF_SKIP_DST_PORT].ptr; 1426 else if (r->os_fingerprint != PF_OSFP_ANY && !pf_osfp_match( 1427 pf_osfp_fingerprint(pd, m, off, th), 1428 r->os_fingerprint)) 1429 r = TAILQ_NEXT(r, entries); 1430 else { 1431 rm = r; 1432 break; 1433 } 1434 } 1435 1436 if (srs) { 1437 /* With scrub rules present TCP normalization happens only 1438 * if one of rules has matched and it's not a "no scrub" rule */ 1439 if (rm == NULL || rm->action == PF_NOSCRUB) 1440 return (PF_PASS); 1441 1442 pf_counter_u64_critical_enter(); 1443 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1); 1444 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len); 1445 pf_counter_u64_critical_exit(); 1446 pf_rule_to_actions(rm, &pd->act); 1447 } 1448 1449 if (rm && rm->rule_flag & PFRULE_REASSEMBLE_TCP) 1450 pd->flags |= PFDESC_TCP_NORM; 1451 1452 flags = tcp_get_flags(th); 1453 if (flags & TH_SYN) { 1454 /* Illegal packet */ 1455 if (flags & TH_RST) 1456 goto tcp_drop; 1457 1458 if (flags & TH_FIN) 1459 goto tcp_drop; 1460 } else { 1461 /* Illegal packet */ 1462 if (!(flags & (TH_ACK|TH_RST))) 1463 goto tcp_drop; 1464 } 1465 1466 if (!(flags & TH_ACK)) { 1467 /* These flags are only valid if ACK is set */ 1468 if ((flags & TH_FIN) || (flags & TH_PUSH) || (flags & TH_URG)) 1469 goto tcp_drop; 1470 } 1471 1472 /* Check for illegal header length */ 1473 if (th->th_off < (sizeof(struct tcphdr) >> 2)) 1474 goto tcp_drop; 1475 1476 /* If flags changed, or reserved data set, then adjust */ 1477 if (flags != tcp_get_flags(th) || 1478 (tcp_get_flags(th) & (TH_RES1|TH_RES2|TH_RES2)) != 0) { 1479 u_int16_t ov, nv; 1480 1481 ov = *(u_int16_t *)(&th->th_ack + 1); 1482 flags &= ~(TH_RES1 | TH_RES2 | TH_RES3); 1483 tcp_set_flags(th, flags); 1484 nv = *(u_int16_t *)(&th->th_ack + 1); 1485 1486 th->th_sum = pf_proto_cksum_fixup(m, th->th_sum, ov, nv, 0); 1487 rewrite = 1; 1488 } 1489 1490 /* Remove urgent pointer, if TH_URG is not set */ 1491 if (!(flags & TH_URG) && th->th_urp) { 1492 th->th_sum = pf_proto_cksum_fixup(m, th->th_sum, th->th_urp, 1493 0, 0); 1494 th->th_urp = 0; 1495 rewrite = 1; 1496 } 1497 1498 /* copy back packet headers if we sanitized */ 1499 if (rewrite) 1500 m_copyback(m, off, sizeof(*th), (caddr_t)th); 1501 1502 return (PF_PASS); 1503 1504 tcp_drop: 1505 REASON_SET(&reason, PFRES_NORM); 1506 if (rm != NULL && r->log) 1507 PFLOG_PACKET(kif, m, AF_INET, PF_DROP, reason, r, NULL, NULL, pd, 1); 1508 return (PF_DROP); 1509} 1510 1511int 1512pf_normalize_tcp_init(struct mbuf *m, int off, struct pf_pdesc *pd, 1513 struct tcphdr *th, struct pf_state_peer *src, struct pf_state_peer *dst) 1514{ 1515 u_int32_t tsval, tsecr; 1516 u_int8_t hdr[60]; 1517 u_int8_t *opt; 1518 1519 KASSERT((src->scrub == NULL), 1520 ("pf_normalize_tcp_init: src->scrub != NULL")); 1521 1522 src->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT); 1523 if (src->scrub == NULL) 1524 return (1); 1525 1526 switch (pd->af) { 1527#ifdef INET 1528 case AF_INET: { 1529 struct ip *h = mtod(m, struct ip *); 1530 src->scrub->pfss_ttl = h->ip_ttl; 1531 break; 1532 } 1533#endif /* INET */ 1534#ifdef INET6 1535 case AF_INET6: { 1536 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 1537 src->scrub->pfss_ttl = h->ip6_hlim; 1538 break; 1539 } 1540#endif /* INET6 */ 1541 } 1542 1543 /* 1544 * All normalizations below are only begun if we see the start of 1545 * the connections. They must all set an enabled bit in pfss_flags 1546 */ 1547 if ((th->th_flags & TH_SYN) == 0) 1548 return (0); 1549 1550 if (th->th_off > (sizeof(struct tcphdr) >> 2) && src->scrub && 1551 pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) { 1552 /* Diddle with TCP options */ 1553 int hlen; 1554 opt = hdr + sizeof(struct tcphdr); 1555 hlen = (th->th_off << 2) - sizeof(struct tcphdr); 1556 while (hlen >= TCPOLEN_TIMESTAMP) { 1557 switch (*opt) { 1558 case TCPOPT_EOL: /* FALLTHROUGH */ 1559 case TCPOPT_NOP: 1560 opt++; 1561 hlen--; 1562 break; 1563 case TCPOPT_TIMESTAMP: 1564 if (opt[1] >= TCPOLEN_TIMESTAMP) { 1565 src->scrub->pfss_flags |= 1566 PFSS_TIMESTAMP; 1567 src->scrub->pfss_ts_mod = 1568 htonl(arc4random()); 1569 1570 /* note PFSS_PAWS not set yet */ 1571 memcpy(&tsval, &opt[2], 1572 sizeof(u_int32_t)); 1573 memcpy(&tsecr, &opt[6], 1574 sizeof(u_int32_t)); 1575 src->scrub->pfss_tsval0 = ntohl(tsval); 1576 src->scrub->pfss_tsval = ntohl(tsval); 1577 src->scrub->pfss_tsecr = ntohl(tsecr); 1578 getmicrouptime(&src->scrub->pfss_last); 1579 } 1580 /* FALLTHROUGH */ 1581 default: 1582 hlen -= MAX(opt[1], 2); 1583 opt += MAX(opt[1], 2); 1584 break; 1585 } 1586 } 1587 } 1588 1589 return (0); 1590} 1591 1592void 1593pf_normalize_tcp_cleanup(struct pf_kstate *state) 1594{ 1595 /* XXX Note: this also cleans up SCTP. */ 1596 uma_zfree(V_pf_state_scrub_z, state->src.scrub); 1597 uma_zfree(V_pf_state_scrub_z, state->dst.scrub); 1598 1599 /* Someday... flush the TCP segment reassembly descriptors. */ 1600} 1601int 1602pf_normalize_sctp_init(struct mbuf *m, int off, struct pf_pdesc *pd, 1603 struct pf_state_peer *src, struct pf_state_peer *dst) 1604{ 1605 src->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT); 1606 if (src->scrub == NULL) 1607 return (1); 1608 1609 dst->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT); 1610 if (dst->scrub == NULL) { 1611 uma_zfree(V_pf_state_scrub_z, src); 1612 return (1); 1613 } 1614 1615 dst->scrub->pfss_v_tag = pd->sctp_initiate_tag; 1616 1617 return (0); 1618} 1619 1620int 1621pf_normalize_tcp_stateful(struct mbuf *m, int off, struct pf_pdesc *pd, 1622 u_short *reason, struct tcphdr *th, struct pf_kstate *state, 1623 struct pf_state_peer *src, struct pf_state_peer *dst, int *writeback) 1624{ 1625 struct timeval uptime; 1626 u_int32_t tsval, tsecr; 1627 u_int tsval_from_last; 1628 u_int8_t hdr[60]; 1629 u_int8_t *opt; 1630 int copyback = 0; 1631 int got_ts = 0; 1632 size_t startoff; 1633 1634 KASSERT((src->scrub || dst->scrub), 1635 ("%s: src->scrub && dst->scrub!", __func__)); 1636 1637 /* 1638 * Enforce the minimum TTL seen for this connection. Negate a common 1639 * technique to evade an intrusion detection system and confuse 1640 * firewall state code. 1641 */ 1642 switch (pd->af) { 1643#ifdef INET 1644 case AF_INET: { 1645 if (src->scrub) { 1646 struct ip *h = mtod(m, struct ip *); 1647 if (h->ip_ttl > src->scrub->pfss_ttl) 1648 src->scrub->pfss_ttl = h->ip_ttl; 1649 h->ip_ttl = src->scrub->pfss_ttl; 1650 } 1651 break; 1652 } 1653#endif /* INET */ 1654#ifdef INET6 1655 case AF_INET6: { 1656 if (src->scrub) { 1657 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 1658 if (h->ip6_hlim > src->scrub->pfss_ttl) 1659 src->scrub->pfss_ttl = h->ip6_hlim; 1660 h->ip6_hlim = src->scrub->pfss_ttl; 1661 } 1662 break; 1663 } 1664#endif /* INET6 */ 1665 } 1666 1667 if (th->th_off > (sizeof(struct tcphdr) >> 2) && 1668 ((src->scrub && (src->scrub->pfss_flags & PFSS_TIMESTAMP)) || 1669 (dst->scrub && (dst->scrub->pfss_flags & PFSS_TIMESTAMP))) && 1670 pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) { 1671 /* Diddle with TCP options */ 1672 int hlen; 1673 opt = hdr + sizeof(struct tcphdr); 1674 hlen = (th->th_off << 2) - sizeof(struct tcphdr); 1675 while (hlen >= TCPOLEN_TIMESTAMP) { 1676 startoff = opt - (hdr + sizeof(struct tcphdr)); 1677 switch (*opt) { 1678 case TCPOPT_EOL: /* FALLTHROUGH */ 1679 case TCPOPT_NOP: 1680 opt++; 1681 hlen--; 1682 break; 1683 case TCPOPT_TIMESTAMP: 1684 /* Modulate the timestamps. Can be used for 1685 * NAT detection, OS uptime determination or 1686 * reboot detection. 1687 */ 1688 1689 if (got_ts) { 1690 /* Huh? Multiple timestamps!? */ 1691 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1692 DPFPRINTF(("multiple TS??\n")); 1693 pf_print_state(state); 1694 printf("\n"); 1695 } 1696 REASON_SET(reason, PFRES_TS); 1697 return (PF_DROP); 1698 } 1699 if (opt[1] >= TCPOLEN_TIMESTAMP) { 1700 memcpy(&tsval, &opt[2], 1701 sizeof(u_int32_t)); 1702 if (tsval && src->scrub && 1703 (src->scrub->pfss_flags & 1704 PFSS_TIMESTAMP)) { 1705 tsval = ntohl(tsval); 1706 pf_patch_32_unaligned(m, 1707 &th->th_sum, 1708 &opt[2], 1709 htonl(tsval + 1710 src->scrub->pfss_ts_mod), 1711 PF_ALGNMNT(startoff), 1712 0); 1713 copyback = 1; 1714 } 1715 1716 /* Modulate TS reply iff valid (!0) */ 1717 memcpy(&tsecr, &opt[6], 1718 sizeof(u_int32_t)); 1719 if (tsecr && dst->scrub && 1720 (dst->scrub->pfss_flags & 1721 PFSS_TIMESTAMP)) { 1722 tsecr = ntohl(tsecr) 1723 - dst->scrub->pfss_ts_mod; 1724 pf_patch_32_unaligned(m, 1725 &th->th_sum, 1726 &opt[6], 1727 htonl(tsecr), 1728 PF_ALGNMNT(startoff), 1729 0); 1730 copyback = 1; 1731 } 1732 got_ts = 1; 1733 } 1734 /* FALLTHROUGH */ 1735 default: 1736 hlen -= MAX(opt[1], 2); 1737 opt += MAX(opt[1], 2); 1738 break; 1739 } 1740 } 1741 if (copyback) { 1742 /* Copyback the options, caller copys back header */ 1743 *writeback = 1; 1744 m_copyback(m, off + sizeof(struct tcphdr), 1745 (th->th_off << 2) - sizeof(struct tcphdr), hdr + 1746 sizeof(struct tcphdr)); 1747 } 1748 } 1749 1750 /* 1751 * Must invalidate PAWS checks on connections idle for too long. 1752 * The fastest allowed timestamp clock is 1ms. That turns out to 1753 * be about 24 days before it wraps. XXX Right now our lowerbound 1754 * TS echo check only works for the first 12 days of a connection 1755 * when the TS has exhausted half its 32bit space 1756 */ 1757#define TS_MAX_IDLE (24*24*60*60) 1758#define TS_MAX_CONN (12*24*60*60) /* XXX remove when better tsecr check */ 1759 1760 getmicrouptime(&uptime); 1761 if (src->scrub && (src->scrub->pfss_flags & PFSS_PAWS) && 1762 (uptime.tv_sec - src->scrub->pfss_last.tv_sec > TS_MAX_IDLE || 1763 time_uptime - (state->creation / 1000) > TS_MAX_CONN)) { 1764 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1765 DPFPRINTF(("src idled out of PAWS\n")); 1766 pf_print_state(state); 1767 printf("\n"); 1768 } 1769 src->scrub->pfss_flags = (src->scrub->pfss_flags & ~PFSS_PAWS) 1770 | PFSS_PAWS_IDLED; 1771 } 1772 if (dst->scrub && (dst->scrub->pfss_flags & PFSS_PAWS) && 1773 uptime.tv_sec - dst->scrub->pfss_last.tv_sec > TS_MAX_IDLE) { 1774 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1775 DPFPRINTF(("dst idled out of PAWS\n")); 1776 pf_print_state(state); 1777 printf("\n"); 1778 } 1779 dst->scrub->pfss_flags = (dst->scrub->pfss_flags & ~PFSS_PAWS) 1780 | PFSS_PAWS_IDLED; 1781 } 1782 1783 if (got_ts && src->scrub && dst->scrub && 1784 (src->scrub->pfss_flags & PFSS_PAWS) && 1785 (dst->scrub->pfss_flags & PFSS_PAWS)) { 1786 /* Validate that the timestamps are "in-window". 1787 * RFC1323 describes TCP Timestamp options that allow 1788 * measurement of RTT (round trip time) and PAWS 1789 * (protection against wrapped sequence numbers). PAWS 1790 * gives us a set of rules for rejecting packets on 1791 * long fat pipes (packets that were somehow delayed 1792 * in transit longer than the time it took to send the 1793 * full TCP sequence space of 4Gb). We can use these 1794 * rules and infer a few others that will let us treat 1795 * the 32bit timestamp and the 32bit echoed timestamp 1796 * as sequence numbers to prevent a blind attacker from 1797 * inserting packets into a connection. 1798 * 1799 * RFC1323 tells us: 1800 * - The timestamp on this packet must be greater than 1801 * or equal to the last value echoed by the other 1802 * endpoint. The RFC says those will be discarded 1803 * since it is a dup that has already been acked. 1804 * This gives us a lowerbound on the timestamp. 1805 * timestamp >= other last echoed timestamp 1806 * - The timestamp will be less than or equal to 1807 * the last timestamp plus the time between the 1808 * last packet and now. The RFC defines the max 1809 * clock rate as 1ms. We will allow clocks to be 1810 * up to 10% fast and will allow a total difference 1811 * or 30 seconds due to a route change. And this 1812 * gives us an upperbound on the timestamp. 1813 * timestamp <= last timestamp + max ticks 1814 * We have to be careful here. Windows will send an 1815 * initial timestamp of zero and then initialize it 1816 * to a random value after the 3whs; presumably to 1817 * avoid a DoS by having to call an expensive RNG 1818 * during a SYN flood. Proof MS has at least one 1819 * good security geek. 1820 * 1821 * - The TCP timestamp option must also echo the other 1822 * endpoints timestamp. The timestamp echoed is the 1823 * one carried on the earliest unacknowledged segment 1824 * on the left edge of the sequence window. The RFC 1825 * states that the host will reject any echoed 1826 * timestamps that were larger than any ever sent. 1827 * This gives us an upperbound on the TS echo. 1828 * tescr <= largest_tsval 1829 * - The lowerbound on the TS echo is a little more 1830 * tricky to determine. The other endpoint's echoed 1831 * values will not decrease. But there may be 1832 * network conditions that re-order packets and 1833 * cause our view of them to decrease. For now the 1834 * only lowerbound we can safely determine is that 1835 * the TS echo will never be less than the original 1836 * TS. XXX There is probably a better lowerbound. 1837 * Remove TS_MAX_CONN with better lowerbound check. 1838 * tescr >= other original TS 1839 * 1840 * It is also important to note that the fastest 1841 * timestamp clock of 1ms will wrap its 32bit space in 1842 * 24 days. So we just disable TS checking after 24 1843 * days of idle time. We actually must use a 12d 1844 * connection limit until we can come up with a better 1845 * lowerbound to the TS echo check. 1846 */ 1847 struct timeval delta_ts; 1848 int ts_fudge; 1849 1850 /* 1851 * PFTM_TS_DIFF is how many seconds of leeway to allow 1852 * a host's timestamp. This can happen if the previous 1853 * packet got delayed in transit for much longer than 1854 * this packet. 1855 */ 1856 if ((ts_fudge = state->rule.ptr->timeout[PFTM_TS_DIFF]) == 0) 1857 ts_fudge = V_pf_default_rule.timeout[PFTM_TS_DIFF]; 1858 1859 /* Calculate max ticks since the last timestamp */ 1860#define TS_MAXFREQ 1100 /* RFC max TS freq of 1Khz + 10% skew */ 1861#define TS_MICROSECS 1000000 /* microseconds per second */ 1862 delta_ts = uptime; 1863 timevalsub(&delta_ts, &src->scrub->pfss_last); 1864 tsval_from_last = (delta_ts.tv_sec + ts_fudge) * TS_MAXFREQ; 1865 tsval_from_last += delta_ts.tv_usec / (TS_MICROSECS/TS_MAXFREQ); 1866 1867 if ((src->state >= TCPS_ESTABLISHED && 1868 dst->state >= TCPS_ESTABLISHED) && 1869 (SEQ_LT(tsval, dst->scrub->pfss_tsecr) || 1870 SEQ_GT(tsval, src->scrub->pfss_tsval + tsval_from_last) || 1871 (tsecr && (SEQ_GT(tsecr, dst->scrub->pfss_tsval) || 1872 SEQ_LT(tsecr, dst->scrub->pfss_tsval0))))) { 1873 /* Bad RFC1323 implementation or an insertion attack. 1874 * 1875 * - Solaris 2.6 and 2.7 are known to send another ACK 1876 * after the FIN,FIN|ACK,ACK closing that carries 1877 * an old timestamp. 1878 */ 1879 1880 DPFPRINTF(("Timestamp failed %c%c%c%c\n", 1881 SEQ_LT(tsval, dst->scrub->pfss_tsecr) ? '0' : ' ', 1882 SEQ_GT(tsval, src->scrub->pfss_tsval + 1883 tsval_from_last) ? '1' : ' ', 1884 SEQ_GT(tsecr, dst->scrub->pfss_tsval) ? '2' : ' ', 1885 SEQ_LT(tsecr, dst->scrub->pfss_tsval0)? '3' : ' ')); 1886 DPFPRINTF((" tsval: %u tsecr: %u +ticks: %u " 1887 "idle: %jus %lums\n", 1888 tsval, tsecr, tsval_from_last, 1889 (uintmax_t)delta_ts.tv_sec, 1890 delta_ts.tv_usec / 1000)); 1891 DPFPRINTF((" src->tsval: %u tsecr: %u\n", 1892 src->scrub->pfss_tsval, src->scrub->pfss_tsecr)); 1893 DPFPRINTF((" dst->tsval: %u tsecr: %u tsval0: %u" 1894 "\n", dst->scrub->pfss_tsval, 1895 dst->scrub->pfss_tsecr, dst->scrub->pfss_tsval0)); 1896 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1897 pf_print_state(state); 1898 pf_print_flags(th->th_flags); 1899 printf("\n"); 1900 } 1901 REASON_SET(reason, PFRES_TS); 1902 return (PF_DROP); 1903 } 1904 1905 /* XXX I'd really like to require tsecr but it's optional */ 1906 1907 } else if (!got_ts && (th->th_flags & TH_RST) == 0 && 1908 ((src->state == TCPS_ESTABLISHED && dst->state == TCPS_ESTABLISHED) 1909 || pd->p_len > 0 || (th->th_flags & TH_SYN)) && 1910 src->scrub && dst->scrub && 1911 (src->scrub->pfss_flags & PFSS_PAWS) && 1912 (dst->scrub->pfss_flags & PFSS_PAWS)) { 1913 /* Didn't send a timestamp. Timestamps aren't really useful 1914 * when: 1915 * - connection opening or closing (often not even sent). 1916 * but we must not let an attacker to put a FIN on a 1917 * data packet to sneak it through our ESTABLISHED check. 1918 * - on a TCP reset. RFC suggests not even looking at TS. 1919 * - on an empty ACK. The TS will not be echoed so it will 1920 * probably not help keep the RTT calculation in sync and 1921 * there isn't as much danger when the sequence numbers 1922 * got wrapped. So some stacks don't include TS on empty 1923 * ACKs :-( 1924 * 1925 * To minimize the disruption to mostly RFC1323 conformant 1926 * stacks, we will only require timestamps on data packets. 1927 * 1928 * And what do ya know, we cannot require timestamps on data 1929 * packets. There appear to be devices that do legitimate 1930 * TCP connection hijacking. There are HTTP devices that allow 1931 * a 3whs (with timestamps) and then buffer the HTTP request. 1932 * If the intermediate device has the HTTP response cache, it 1933 * will spoof the response but not bother timestamping its 1934 * packets. So we can look for the presence of a timestamp in 1935 * the first data packet and if there, require it in all future 1936 * packets. 1937 */ 1938 1939 if (pd->p_len > 0 && (src->scrub->pfss_flags & PFSS_DATA_TS)) { 1940 /* 1941 * Hey! Someone tried to sneak a packet in. Or the 1942 * stack changed its RFC1323 behavior?!?! 1943 */ 1944 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1945 DPFPRINTF(("Did not receive expected RFC1323 " 1946 "timestamp\n")); 1947 pf_print_state(state); 1948 pf_print_flags(th->th_flags); 1949 printf("\n"); 1950 } 1951 REASON_SET(reason, PFRES_TS); 1952 return (PF_DROP); 1953 } 1954 } 1955 1956 /* 1957 * We will note if a host sends his data packets with or without 1958 * timestamps. And require all data packets to contain a timestamp 1959 * if the first does. PAWS implicitly requires that all data packets be 1960 * timestamped. But I think there are middle-man devices that hijack 1961 * TCP streams immediately after the 3whs and don't timestamp their 1962 * packets (seen in a WWW accelerator or cache). 1963 */ 1964 if (pd->p_len > 0 && src->scrub && (src->scrub->pfss_flags & 1965 (PFSS_TIMESTAMP|PFSS_DATA_TS|PFSS_DATA_NOTS)) == PFSS_TIMESTAMP) { 1966 if (got_ts) 1967 src->scrub->pfss_flags |= PFSS_DATA_TS; 1968 else { 1969 src->scrub->pfss_flags |= PFSS_DATA_NOTS; 1970 if (V_pf_status.debug >= PF_DEBUG_MISC && dst->scrub && 1971 (dst->scrub->pfss_flags & PFSS_TIMESTAMP)) { 1972 /* Don't warn if other host rejected RFC1323 */ 1973 DPFPRINTF(("Broken RFC1323 stack did not " 1974 "timestamp data packet. Disabled PAWS " 1975 "security.\n")); 1976 pf_print_state(state); 1977 pf_print_flags(th->th_flags); 1978 printf("\n"); 1979 } 1980 } 1981 } 1982 1983 /* 1984 * Update PAWS values 1985 */ 1986 if (got_ts && src->scrub && PFSS_TIMESTAMP == (src->scrub->pfss_flags & 1987 (PFSS_PAWS_IDLED|PFSS_TIMESTAMP))) { 1988 getmicrouptime(&src->scrub->pfss_last); 1989 if (SEQ_GEQ(tsval, src->scrub->pfss_tsval) || 1990 (src->scrub->pfss_flags & PFSS_PAWS) == 0) 1991 src->scrub->pfss_tsval = tsval; 1992 1993 if (tsecr) { 1994 if (SEQ_GEQ(tsecr, src->scrub->pfss_tsecr) || 1995 (src->scrub->pfss_flags & PFSS_PAWS) == 0) 1996 src->scrub->pfss_tsecr = tsecr; 1997 1998 if ((src->scrub->pfss_flags & PFSS_PAWS) == 0 && 1999 (SEQ_LT(tsval, src->scrub->pfss_tsval0) || 2000 src->scrub->pfss_tsval0 == 0)) { 2001 /* tsval0 MUST be the lowest timestamp */ 2002 src->scrub->pfss_tsval0 = tsval; 2003 } 2004 2005 /* Only fully initialized after a TS gets echoed */ 2006 if ((src->scrub->pfss_flags & PFSS_PAWS) == 0) 2007 src->scrub->pfss_flags |= PFSS_PAWS; 2008 } 2009 } 2010 2011 /* I have a dream.... TCP segment reassembly.... */ 2012 return (0); 2013} 2014 2015int 2016pf_normalize_mss(struct mbuf *m, int off, struct pf_pdesc *pd) 2017{ 2018 struct tcphdr *th = &pd->hdr.tcp; 2019 u_int16_t *mss; 2020 int thoff; 2021 int opt, cnt, optlen = 0; 2022 u_char opts[TCP_MAXOLEN]; 2023 u_char *optp = opts; 2024 size_t startoff; 2025 2026 thoff = th->th_off << 2; 2027 cnt = thoff - sizeof(struct tcphdr); 2028 2029 if (cnt > 0 && !pf_pull_hdr(m, off + sizeof(*th), opts, cnt, 2030 NULL, NULL, pd->af)) 2031 return (0); 2032 2033 for (; cnt > 0; cnt -= optlen, optp += optlen) { 2034 startoff = optp - opts; 2035 opt = optp[0]; 2036 if (opt == TCPOPT_EOL) 2037 break; 2038 if (opt == TCPOPT_NOP) 2039 optlen = 1; 2040 else { 2041 if (cnt < 2) 2042 break; 2043 optlen = optp[1]; 2044 if (optlen < 2 || optlen > cnt) 2045 break; 2046 } 2047 switch (opt) { 2048 case TCPOPT_MAXSEG: 2049 mss = (u_int16_t *)(optp + 2); 2050 if ((ntohs(*mss)) > pd->act.max_mss) { 2051 pf_patch_16_unaligned(m, 2052 &th->th_sum, 2053 mss, htons(pd->act.max_mss), 2054 PF_ALGNMNT(startoff), 2055 0); 2056 m_copyback(m, off + sizeof(*th), 2057 thoff - sizeof(*th), opts); 2058 m_copyback(m, off, sizeof(*th), (caddr_t)th); 2059 } 2060 break; 2061 default: 2062 break; 2063 } 2064 } 2065 2066 return (0); 2067} 2068 2069static int 2070pf_scan_sctp(struct mbuf *m, int ipoff, int off, struct pf_pdesc *pd, 2071 struct pfi_kkif *kif) 2072{ 2073 struct sctp_chunkhdr ch = { }; 2074 int chunk_off = sizeof(struct sctphdr); 2075 int chunk_start; 2076 int ret; 2077 2078 while (off + chunk_off < pd->tot_len) { 2079 if (!pf_pull_hdr(m, off + chunk_off, &ch, sizeof(ch), NULL, 2080 NULL, pd->af)) 2081 return (PF_DROP); 2082 2083 /* Length includes the header, this must be at least 4. */ 2084 if (ntohs(ch.chunk_length) < 4) 2085 return (PF_DROP); 2086 2087 chunk_start = chunk_off; 2088 chunk_off += roundup(ntohs(ch.chunk_length), 4); 2089 2090 switch (ch.chunk_type) { 2091 case SCTP_INITIATION: 2092 case SCTP_INITIATION_ACK: { 2093 struct sctp_init_chunk init; 2094 2095 if (!pf_pull_hdr(m, off + chunk_start, &init, 2096 sizeof(init), NULL, NULL, pd->af)) 2097 return (PF_DROP); 2098 2099 /* 2100 * RFC 9620, Section 3.3.2, "The Initiate Tag is allowed to have 2101 * any value except 0." 2102 */ 2103 if (init.init.initiate_tag == 0) 2104 return (PF_DROP); 2105 if (init.init.num_inbound_streams == 0) 2106 return (PF_DROP); 2107 if (init.init.num_outbound_streams == 0) 2108 return (PF_DROP); 2109 if (ntohl(init.init.a_rwnd) < SCTP_MIN_RWND) 2110 return (PF_DROP); 2111 2112 /* 2113 * RFC 9260, Section 3.1, INIT chunks MUST have zero 2114 * verification tag. 2115 */ 2116 if (ch.chunk_type == SCTP_INITIATION && 2117 pd->hdr.sctp.v_tag != 0) 2118 return (PF_DROP); 2119 2120 pd->sctp_initiate_tag = init.init.initiate_tag; 2121 2122 if (ch.chunk_type == SCTP_INITIATION) 2123 pd->sctp_flags |= PFDESC_SCTP_INIT; 2124 else 2125 pd->sctp_flags |= PFDESC_SCTP_INIT_ACK; 2126 2127 ret = pf_multihome_scan_init(m, off + chunk_start, 2128 ntohs(init.ch.chunk_length), pd, kif); 2129 if (ret != PF_PASS) 2130 return (ret); 2131 2132 break; 2133 } 2134 case SCTP_ABORT_ASSOCIATION: 2135 pd->sctp_flags |= PFDESC_SCTP_ABORT; 2136 break; 2137 case SCTP_SHUTDOWN: 2138 case SCTP_SHUTDOWN_ACK: 2139 pd->sctp_flags |= PFDESC_SCTP_SHUTDOWN; 2140 break; 2141 case SCTP_SHUTDOWN_COMPLETE: 2142 pd->sctp_flags |= PFDESC_SCTP_SHUTDOWN_COMPLETE; 2143 break; 2144 case SCTP_COOKIE_ECHO: 2145 pd->sctp_flags |= PFDESC_SCTP_COOKIE; 2146 break; 2147 case SCTP_COOKIE_ACK: 2148 pd->sctp_flags |= PFDESC_SCTP_COOKIE_ACK; 2149 break; 2150 case SCTP_DATA: 2151 pd->sctp_flags |= PFDESC_SCTP_DATA; 2152 break; 2153 case SCTP_HEARTBEAT_REQUEST: 2154 pd->sctp_flags |= PFDESC_SCTP_HEARTBEAT; 2155 break; 2156 case SCTP_HEARTBEAT_ACK: 2157 pd->sctp_flags |= PFDESC_SCTP_HEARTBEAT_ACK; 2158 break; 2159 case SCTP_ASCONF: 2160 pd->sctp_flags |= PFDESC_SCTP_ASCONF; 2161 2162 ret = pf_multihome_scan_asconf(m, off + chunk_start, 2163 ntohs(ch.chunk_length), pd, kif); 2164 if (ret != PF_PASS) 2165 return (ret); 2166 break; 2167 default: 2168 pd->sctp_flags |= PFDESC_SCTP_OTHER; 2169 break; 2170 } 2171 } 2172 2173 /* Validate chunk lengths vs. packet length. */ 2174 if (off + chunk_off != pd->tot_len) 2175 return (PF_DROP); 2176 2177 /* 2178 * INIT, INIT_ACK or SHUTDOWN_COMPLETE chunks must always be the only 2179 * one in a packet. 2180 */ 2181 if ((pd->sctp_flags & PFDESC_SCTP_INIT) && 2182 (pd->sctp_flags & ~PFDESC_SCTP_INIT)) 2183 return (PF_DROP); 2184 if ((pd->sctp_flags & PFDESC_SCTP_INIT_ACK) && 2185 (pd->sctp_flags & ~PFDESC_SCTP_INIT_ACK)) 2186 return (PF_DROP); 2187 if ((pd->sctp_flags & PFDESC_SCTP_SHUTDOWN_COMPLETE) && 2188 (pd->sctp_flags & ~PFDESC_SCTP_SHUTDOWN_COMPLETE)) 2189 return (PF_DROP); 2190 2191 return (PF_PASS); 2192} 2193 2194int 2195pf_normalize_sctp(int dir, struct pfi_kkif *kif, struct mbuf *m, int ipoff, 2196 int off, void *h, struct pf_pdesc *pd) 2197{ 2198 struct pf_krule *r, *rm = NULL; 2199 struct sctphdr *sh = &pd->hdr.sctp; 2200 u_short reason; 2201 sa_family_t af = pd->af; 2202 int srs; 2203 2204 PF_RULES_RASSERT(); 2205 2206 /* Unconditionally scan the SCTP packet, because we need to look for 2207 * things like shutdown and asconf chunks. */ 2208 if (pf_scan_sctp(m, ipoff, off, pd, kif) != PF_PASS) 2209 goto sctp_drop; 2210 2211 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr); 2212 /* Check if there any scrub rules. Lack of scrub rules means enforced 2213 * packet normalization operation just like in OpenBSD. */ 2214 srs = (r != NULL); 2215 while (r != NULL) { 2216 pf_counter_u64_add(&r->evaluations, 1); 2217 if (pfi_kkif_match(r->kif, kif) == r->ifnot) 2218 r = r->skip[PF_SKIP_IFP].ptr; 2219 else if (r->direction && r->direction != dir) 2220 r = r->skip[PF_SKIP_DIR].ptr; 2221 else if (r->af && r->af != af) 2222 r = r->skip[PF_SKIP_AF].ptr; 2223 else if (r->proto && r->proto != pd->proto) 2224 r = r->skip[PF_SKIP_PROTO].ptr; 2225 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af, 2226 r->src.neg, kif, M_GETFIB(m))) 2227 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 2228 else if (r->src.port_op && !pf_match_port(r->src.port_op, 2229 r->src.port[0], r->src.port[1], sh->src_port)) 2230 r = r->skip[PF_SKIP_SRC_PORT].ptr; 2231 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af, 2232 r->dst.neg, NULL, M_GETFIB(m))) 2233 r = r->skip[PF_SKIP_DST_ADDR].ptr; 2234 else if (r->dst.port_op && !pf_match_port(r->dst.port_op, 2235 r->dst.port[0], r->dst.port[1], sh->dest_port)) 2236 r = r->skip[PF_SKIP_DST_PORT].ptr; 2237 else { 2238 rm = r; 2239 break; 2240 } 2241 } 2242 2243 if (srs) { 2244 /* With scrub rules present SCTP normalization happens only 2245 * if one of rules has matched and it's not a "no scrub" rule */ 2246 if (rm == NULL || rm->action == PF_NOSCRUB) 2247 return (PF_PASS); 2248 2249 pf_counter_u64_critical_enter(); 2250 pf_counter_u64_add_protected(&r->packets[dir == PF_OUT], 1); 2251 pf_counter_u64_add_protected(&r->bytes[dir == PF_OUT], pd->tot_len); 2252 pf_counter_u64_critical_exit(); 2253 } 2254 2255 /* Verify we're a multiple of 4 bytes long */ 2256 if ((pd->tot_len - off - sizeof(struct sctphdr)) % 4) 2257 goto sctp_drop; 2258 2259 /* INIT chunk needs to be the only chunk */ 2260 if (pd->sctp_flags & PFDESC_SCTP_INIT) 2261 if (pd->sctp_flags & ~PFDESC_SCTP_INIT) 2262 goto sctp_drop; 2263 2264 return (PF_PASS); 2265 2266sctp_drop: 2267 REASON_SET(&reason, PFRES_NORM); 2268 if (rm != NULL && r->log) 2269 PFLOG_PACKET(kif, m, AF_INET, PF_DROP, reason, r, NULL, NULL, pd, 2270 1); 2271 2272 return (PF_DROP); 2273} 2274 2275#ifdef INET 2276void 2277pf_scrub_ip(struct mbuf **m0, struct pf_pdesc *pd) 2278{ 2279 struct mbuf *m = *m0; 2280 struct ip *h = mtod(m, struct ip *); 2281 2282 /* Clear IP_DF if no-df was requested */ 2283 if (pd->act.flags & PFSTATE_NODF && h->ip_off & htons(IP_DF)) { 2284 u_int16_t ip_off = h->ip_off; 2285 2286 h->ip_off &= htons(~IP_DF); 2287 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0); 2288 } 2289 2290 /* Enforce a minimum ttl, may cause endless packet loops */ 2291 if (pd->act.min_ttl && h->ip_ttl < pd->act.min_ttl) { 2292 u_int16_t ip_ttl = h->ip_ttl; 2293 2294 h->ip_ttl = pd->act.min_ttl; 2295 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_ttl, h->ip_ttl, 0); 2296 } 2297 2298 /* Enforce tos */ 2299 if (pd->act.flags & PFSTATE_SETTOS) { 2300 u_int16_t ov, nv; 2301 2302 ov = *(u_int16_t *)h; 2303 h->ip_tos = pd->act.set_tos | (h->ip_tos & IPTOS_ECN_MASK); 2304 nv = *(u_int16_t *)h; 2305 2306 h->ip_sum = pf_cksum_fixup(h->ip_sum, ov, nv, 0); 2307 } 2308 2309 /* random-id, but not for fragments */ 2310 if (pd->act.flags & PFSTATE_RANDOMID && !(h->ip_off & ~htons(IP_DF))) { 2311 uint16_t ip_id = h->ip_id; 2312 2313 ip_fillid(h); 2314 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_id, h->ip_id, 0); 2315 } 2316} 2317#endif /* INET */ 2318 2319#ifdef INET6 2320void 2321pf_scrub_ip6(struct mbuf **m0, struct pf_pdesc *pd) 2322{ 2323 struct mbuf *m = *m0; 2324 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 2325 2326 /* Enforce a minimum ttl, may cause endless packet loops */ 2327 if (pd->act.min_ttl && h->ip6_hlim < pd->act.min_ttl) 2328 h->ip6_hlim = pd->act.min_ttl; 2329 2330 /* Enforce tos. Set traffic class bits */ 2331 if (pd->act.flags & PFSTATE_SETTOS) { 2332 h->ip6_flow &= IPV6_FLOWLABEL_MASK | IPV6_VERSION_MASK; 2333 h->ip6_flow |= htonl((pd->act.set_tos | IPV6_ECN(h)) << 20); 2334 } 2335} 2336#endif 2337