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 <netinet/tcp.h> 56#include <netinet/tcp_fsm.h> 57#include <netinet/tcp_seq.h> 58 59#ifdef INET6 60#include <netinet/ip6.h> 61#endif /* INET6 */ 62 63struct pf_frent { 64 TAILQ_ENTRY(pf_frent) fr_next; 65 struct mbuf *fe_m; 66 uint16_t fe_hdrlen; /* ipv4 header length with ip options 67 ipv6, extension, fragment header */ 68 uint16_t fe_extoff; /* last extension header offset or 0 */ 69 uint16_t fe_len; /* fragment length */ 70 uint16_t fe_off; /* fragment offset */ 71 uint16_t fe_mff; /* more fragment flag */ 72}; 73 74struct pf_fragment_cmp { 75 struct pf_addr frc_src; 76 struct pf_addr frc_dst; 77 uint32_t frc_id; 78 sa_family_t frc_af; 79 uint8_t frc_proto; 80}; 81 82struct pf_fragment { 83 struct pf_fragment_cmp fr_key; 84#define fr_src fr_key.frc_src 85#define fr_dst fr_key.frc_dst 86#define fr_id fr_key.frc_id 87#define fr_af fr_key.frc_af 88#define fr_proto fr_key.frc_proto 89 90 /* pointers to queue element */ 91 struct pf_frent *fr_firstoff[PF_FRAG_ENTRY_POINTS]; 92 /* count entries between pointers */ 93 uint8_t fr_entries[PF_FRAG_ENTRY_POINTS]; 94 RB_ENTRY(pf_fragment) fr_entry; 95 TAILQ_ENTRY(pf_fragment) frag_next; 96 uint32_t fr_timeout; 97 uint16_t fr_maxlen; /* maximum length of single fragment */ 98 u_int16_t fr_holes; /* number of holes in the queue */ 99 TAILQ_HEAD(pf_fragq, pf_frent) fr_queue; 100}; 101 102struct pf_fragment_tag { 103 uint16_t ft_hdrlen; /* header length of reassembled pkt */ 104 uint16_t ft_extoff; /* last extension header offset or 0 */ 105 uint16_t ft_maxlen; /* maximum fragment payload length */ 106 uint32_t ft_id; /* fragment id */ 107}; 108 109static struct mtx pf_frag_mtx; 110MTX_SYSINIT(pf_frag_mtx, &pf_frag_mtx, "pf fragments", MTX_DEF); 111#define PF_FRAG_LOCK() mtx_lock(&pf_frag_mtx) 112#define PF_FRAG_UNLOCK() mtx_unlock(&pf_frag_mtx) 113#define PF_FRAG_ASSERT() mtx_assert(&pf_frag_mtx, MA_OWNED) 114 115VNET_DEFINE(uma_zone_t, pf_state_scrub_z); /* XXX: shared with pfsync */ 116 117VNET_DEFINE_STATIC(uma_zone_t, pf_frent_z); 118#define V_pf_frent_z VNET(pf_frent_z) 119VNET_DEFINE_STATIC(uma_zone_t, pf_frag_z); 120#define V_pf_frag_z VNET(pf_frag_z) 121 122TAILQ_HEAD(pf_fragqueue, pf_fragment); 123TAILQ_HEAD(pf_cachequeue, pf_fragment); 124VNET_DEFINE_STATIC(struct pf_fragqueue, pf_fragqueue); 125#define V_pf_fragqueue VNET(pf_fragqueue) 126RB_HEAD(pf_frag_tree, pf_fragment); 127VNET_DEFINE_STATIC(struct pf_frag_tree, pf_frag_tree); 128#define V_pf_frag_tree VNET(pf_frag_tree) 129static int pf_frag_compare(struct pf_fragment *, 130 struct pf_fragment *); 131static RB_PROTOTYPE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare); 132static RB_GENERATE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare); 133 134static void pf_flush_fragments(void); 135static void pf_free_fragment(struct pf_fragment *); 136static void pf_remove_fragment(struct pf_fragment *); 137static int pf_normalize_tcpopt(struct pf_krule *, struct mbuf *, 138 struct tcphdr *, int, sa_family_t); 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 void pf_scrub_ip(struct mbuf **, uint32_t, uint8_t, uint8_t); 155static int pf_reassemble(struct mbuf **, struct ip *, int, u_short *); 156#endif /* INET */ 157#ifdef INET6 158static int pf_reassemble6(struct mbuf **, struct ip6_hdr *, 159 struct ip6_frag *, uint16_t, uint16_t, u_short *); 160static void pf_scrub_ip6(struct mbuf **, uint8_t); 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 171static void 172pf_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 183void 184pf_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 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 212static int 213pf_frag_compare(struct pf_fragment *a, struct pf_fragment *b) 214{ 215 int diff; 216 217 if ((diff = a->fr_id - b->fr_id) != 0) 218 return (diff); 219 if ((diff = a->fr_proto - b->fr_proto) != 0) 220 return (diff); 221 if ((diff = a->fr_af - b->fr_af) != 0) 222 return (diff); 223 if ((diff = pf_addr_cmp(&a->fr_src, &b->fr_src, a->fr_af)) != 0) 224 return (diff); 225 if ((diff = pf_addr_cmp(&a->fr_dst, &b->fr_dst, a->fr_af)) != 0) 226 return (diff); 227 return (0); 228} 229 230void 231pf_purge_expired_fragments(void) 232{ 233 u_int32_t expire = time_uptime - 234 V_pf_default_rule.timeout[PFTM_FRAG]; 235 236 pf_purge_fragments(expire); 237} 238 239void 240pf_purge_fragments(uint32_t expire) 241{ 242 struct pf_fragment *frag; 243 244 PF_FRAG_LOCK(); 245 while ((frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue)) != NULL) { 246 if (frag->fr_timeout > expire) 247 break; 248 249 DPFPRINTF(("expiring %d(%p)\n", frag->fr_id, frag)); 250 pf_free_fragment(frag); 251 } 252 253 PF_FRAG_UNLOCK(); 254} 255 256/* 257 * Try to flush old fragments to make space for new ones 258 */ 259static void 260pf_flush_fragments(void) 261{ 262 struct pf_fragment *frag; 263 int goal; 264 265 PF_FRAG_ASSERT(); 266 267 goal = uma_zone_get_cur(V_pf_frent_z) * 9 / 10; 268 DPFPRINTF(("trying to free %d frag entriess\n", goal)); 269 while (goal < uma_zone_get_cur(V_pf_frent_z)) { 270 frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue); 271 if (frag) 272 pf_free_fragment(frag); 273 else 274 break; 275 } 276} 277 278/* Frees the fragments and all associated entries */ 279static void 280pf_free_fragment(struct pf_fragment *frag) 281{ 282 struct pf_frent *frent; 283 284 PF_FRAG_ASSERT(); 285 286 /* Free all fragments */ 287 for (frent = TAILQ_FIRST(&frag->fr_queue); frent; 288 frent = TAILQ_FIRST(&frag->fr_queue)) { 289 TAILQ_REMOVE(&frag->fr_queue, frent, fr_next); 290 291 m_freem(frent->fe_m); 292 uma_zfree(V_pf_frent_z, frent); 293 } 294 295 pf_remove_fragment(frag); 296} 297 298static struct pf_fragment * 299pf_find_fragment(struct pf_fragment_cmp *key, struct pf_frag_tree *tree) 300{ 301 struct pf_fragment *frag; 302 303 PF_FRAG_ASSERT(); 304 305 frag = RB_FIND(pf_frag_tree, tree, (struct pf_fragment *)key); 306 if (frag != NULL) { 307 /* XXX Are we sure we want to update the timeout? */ 308 frag->fr_timeout = time_uptime; 309 TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next); 310 TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next); 311 } 312 313 return (frag); 314} 315 316/* Removes a fragment from the fragment queue and frees the fragment */ 317static void 318pf_remove_fragment(struct pf_fragment *frag) 319{ 320 321 PF_FRAG_ASSERT(); 322 KASSERT(frag, ("frag != NULL")); 323 324 RB_REMOVE(pf_frag_tree, &V_pf_frag_tree, frag); 325 TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next); 326 uma_zfree(V_pf_frag_z, frag); 327} 328 329static struct pf_frent * 330pf_create_fragment(u_short *reason) 331{ 332 struct pf_frent *frent; 333 334 PF_FRAG_ASSERT(); 335 336 frent = uma_zalloc(V_pf_frent_z, M_NOWAIT); 337 if (frent == NULL) { 338 pf_flush_fragments(); 339 frent = uma_zalloc(V_pf_frent_z, M_NOWAIT); 340 if (frent == NULL) { 341 REASON_SET(reason, PFRES_MEMORY); 342 return (NULL); 343 } 344 } 345 346 return (frent); 347} 348 349/* 350 * Calculate the additional holes that were created in the fragment 351 * queue by inserting this fragment. A fragment in the middle 352 * creates one more hole by splitting. For each connected side, 353 * it loses one hole. 354 * Fragment entry must be in the queue when calling this function. 355 */ 356static int 357pf_frent_holes(struct pf_frent *frent) 358{ 359 struct pf_frent *prev = TAILQ_PREV(frent, pf_fragq, fr_next); 360 struct pf_frent *next = TAILQ_NEXT(frent, fr_next); 361 int holes = 1; 362 363 if (prev == NULL) { 364 if (frent->fe_off == 0) 365 holes--; 366 } else { 367 KASSERT(frent->fe_off != 0, ("frent->fe_off != 0")); 368 if (frent->fe_off == prev->fe_off + prev->fe_len) 369 holes--; 370 } 371 if (next == NULL) { 372 if (!frent->fe_mff) 373 holes--; 374 } else { 375 KASSERT(frent->fe_mff, ("frent->fe_mff")); 376 if (next->fe_off == frent->fe_off + frent->fe_len) 377 holes--; 378 } 379 return holes; 380} 381 382static inline int 383pf_frent_index(struct pf_frent *frent) 384{ 385 /* 386 * We have an array of 16 entry points to the queue. A full size 387 * 65535 octet IP packet can have 8192 fragments. So the queue 388 * traversal length is at most 512 and at most 16 entry points are 389 * checked. We need 128 additional bytes on a 64 bit architecture. 390 */ 391 CTASSERT(((u_int16_t)0xffff &~ 7) / (0x10000 / PF_FRAG_ENTRY_POINTS) == 392 16 - 1); 393 CTASSERT(((u_int16_t)0xffff >> 3) / PF_FRAG_ENTRY_POINTS == 512 - 1); 394 395 return frent->fe_off / (0x10000 / PF_FRAG_ENTRY_POINTS); 396} 397 398static int 399pf_frent_insert(struct pf_fragment *frag, struct pf_frent *frent, 400 struct pf_frent *prev) 401{ 402 int index; 403 404 CTASSERT(PF_FRAG_ENTRY_LIMIT <= 0xff); 405 406 /* 407 * A packet has at most 65536 octets. With 16 entry points, each one 408 * spawns 4096 octets. We limit these to 64 fragments each, which 409 * means on average every fragment must have at least 64 octets. 410 */ 411 index = pf_frent_index(frent); 412 if (frag->fr_entries[index] >= PF_FRAG_ENTRY_LIMIT) 413 return ENOBUFS; 414 frag->fr_entries[index]++; 415 416 if (prev == NULL) { 417 TAILQ_INSERT_HEAD(&frag->fr_queue, frent, fr_next); 418 } else { 419 KASSERT(prev->fe_off + prev->fe_len <= frent->fe_off, 420 ("overlapping fragment")); 421 TAILQ_INSERT_AFTER(&frag->fr_queue, prev, frent, fr_next); 422 } 423 424 if (frag->fr_firstoff[index] == NULL) { 425 KASSERT(prev == NULL || pf_frent_index(prev) < index, 426 ("prev == NULL || pf_frent_index(pref) < index")); 427 frag->fr_firstoff[index] = frent; 428 } else { 429 if (frent->fe_off < frag->fr_firstoff[index]->fe_off) { 430 KASSERT(prev == NULL || pf_frent_index(prev) < index, 431 ("prev == NULL || pf_frent_index(pref) < index")); 432 frag->fr_firstoff[index] = frent; 433 } else { 434 KASSERT(prev != NULL, ("prev != NULL")); 435 KASSERT(pf_frent_index(prev) == index, 436 ("pf_frent_index(prev) == index")); 437 } 438 } 439 440 frag->fr_holes += pf_frent_holes(frent); 441 442 return 0; 443} 444 445void 446pf_frent_remove(struct pf_fragment *frag, struct pf_frent *frent) 447{ 448#ifdef INVARIANTS 449 struct pf_frent *prev = TAILQ_PREV(frent, pf_fragq, fr_next); 450#endif 451 struct pf_frent *next = TAILQ_NEXT(frent, fr_next); 452 int index; 453 454 frag->fr_holes -= pf_frent_holes(frent); 455 456 index = pf_frent_index(frent); 457 KASSERT(frag->fr_firstoff[index] != NULL, ("frent not found")); 458 if (frag->fr_firstoff[index]->fe_off == frent->fe_off) { 459 if (next == NULL) { 460 frag->fr_firstoff[index] = NULL; 461 } else { 462 KASSERT(frent->fe_off + frent->fe_len <= next->fe_off, 463 ("overlapping fragment")); 464 if (pf_frent_index(next) == index) { 465 frag->fr_firstoff[index] = next; 466 } else { 467 frag->fr_firstoff[index] = NULL; 468 } 469 } 470 } else { 471 KASSERT(frag->fr_firstoff[index]->fe_off < frent->fe_off, 472 ("frag->fr_firstoff[index]->fe_off < frent->fe_off")); 473 KASSERT(prev != NULL, ("prev != NULL")); 474 KASSERT(prev->fe_off + prev->fe_len <= frent->fe_off, 475 ("overlapping fragment")); 476 KASSERT(pf_frent_index(prev) == index, 477 ("pf_frent_index(prev) == index")); 478 } 479 480 TAILQ_REMOVE(&frag->fr_queue, frent, fr_next); 481 482 KASSERT(frag->fr_entries[index] > 0, ("No fragments remaining")); 483 frag->fr_entries[index]--; 484} 485 486struct pf_frent * 487pf_frent_previous(struct pf_fragment *frag, struct pf_frent *frent) 488{ 489 struct pf_frent *prev, *next; 490 int index; 491 492 /* 493 * If there are no fragments after frag, take the final one. Assume 494 * that the global queue is not empty. 495 */ 496 prev = TAILQ_LAST(&frag->fr_queue, pf_fragq); 497 KASSERT(prev != NULL, ("prev != NULL")); 498 if (prev->fe_off <= frent->fe_off) 499 return prev; 500 /* 501 * We want to find a fragment entry that is before frag, but still 502 * close to it. Find the first fragment entry that is in the same 503 * entry point or in the first entry point after that. As we have 504 * already checked that there are entries behind frag, this will 505 * succeed. 506 */ 507 for (index = pf_frent_index(frent); index < PF_FRAG_ENTRY_POINTS; 508 index++) { 509 prev = frag->fr_firstoff[index]; 510 if (prev != NULL) 511 break; 512 } 513 KASSERT(prev != NULL, ("prev != NULL")); 514 /* 515 * In prev we may have a fragment from the same entry point that is 516 * before frent, or one that is just one position behind frent. 517 * In the latter case, we go back one step and have the predecessor. 518 * There may be none if the new fragment will be the first one. 519 */ 520 if (prev->fe_off > frent->fe_off) { 521 prev = TAILQ_PREV(prev, pf_fragq, fr_next); 522 if (prev == NULL) 523 return NULL; 524 KASSERT(prev->fe_off <= frent->fe_off, 525 ("prev->fe_off <= frent->fe_off")); 526 return prev; 527 } 528 /* 529 * In prev is the first fragment of the entry point. The offset 530 * of frag is behind it. Find the closest previous fragment. 531 */ 532 for (next = TAILQ_NEXT(prev, fr_next); next != NULL; 533 next = TAILQ_NEXT(next, fr_next)) { 534 if (next->fe_off > frent->fe_off) 535 break; 536 prev = next; 537 } 538 return prev; 539} 540 541static struct pf_fragment * 542pf_fillup_fragment(struct pf_fragment_cmp *key, struct pf_frent *frent, 543 u_short *reason) 544{ 545 struct pf_frent *after, *next, *prev; 546 struct pf_fragment *frag; 547 uint16_t total; 548 int old_index, new_index; 549 550 PF_FRAG_ASSERT(); 551 552 /* No empty fragments. */ 553 if (frent->fe_len == 0) { 554 DPFPRINTF(("bad fragment: len 0")); 555 goto bad_fragment; 556 } 557 558 /* All fragments are 8 byte aligned. */ 559 if (frent->fe_mff && (frent->fe_len & 0x7)) { 560 DPFPRINTF(("bad fragment: mff and len %d", frent->fe_len)); 561 goto bad_fragment; 562 } 563 564 /* Respect maximum length, IP_MAXPACKET == IPV6_MAXPACKET. */ 565 if (frent->fe_off + frent->fe_len > IP_MAXPACKET) { 566 DPFPRINTF(("bad fragment: max packet %d", 567 frent->fe_off + frent->fe_len)); 568 goto bad_fragment; 569 } 570 571 DPFPRINTF((key->frc_af == AF_INET ? 572 "reass frag %d @ %d-%d" : "reass frag %#08x @ %d-%d", 573 key->frc_id, frent->fe_off, frent->fe_off + frent->fe_len)); 574 575 /* Fully buffer all of the fragments in this fragment queue. */ 576 frag = pf_find_fragment(key, &V_pf_frag_tree); 577 578 /* Create a new reassembly queue for this packet. */ 579 if (frag == NULL) { 580 frag = uma_zalloc(V_pf_frag_z, M_NOWAIT); 581 if (frag == NULL) { 582 pf_flush_fragments(); 583 frag = uma_zalloc(V_pf_frag_z, M_NOWAIT); 584 if (frag == NULL) { 585 REASON_SET(reason, PFRES_MEMORY); 586 goto drop_fragment; 587 } 588 } 589 590 *(struct pf_fragment_cmp *)frag = *key; 591 memset(frag->fr_firstoff, 0, sizeof(frag->fr_firstoff)); 592 memset(frag->fr_entries, 0, sizeof(frag->fr_entries)); 593 frag->fr_timeout = time_uptime; 594 frag->fr_maxlen = frent->fe_len; 595 frag->fr_holes = 1; 596 TAILQ_INIT(&frag->fr_queue); 597 598 RB_INSERT(pf_frag_tree, &V_pf_frag_tree, frag); 599 TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next); 600 601 /* We do not have a previous fragment, cannot fail. */ 602 pf_frent_insert(frag, frent, NULL); 603 604 return (frag); 605 } 606 607 KASSERT(!TAILQ_EMPTY(&frag->fr_queue), ("!TAILQ_EMPTY()->fr_queue")); 608 609 /* Remember maximum fragment len for refragmentation. */ 610 if (frent->fe_len > frag->fr_maxlen) 611 frag->fr_maxlen = frent->fe_len; 612 613 /* Maximum data we have seen already. */ 614 total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off + 615 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len; 616 617 /* Non terminal fragments must have more fragments flag. */ 618 if (frent->fe_off + frent->fe_len < total && !frent->fe_mff) 619 goto bad_fragment; 620 621 /* Check if we saw the last fragment already. */ 622 if (!TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_mff) { 623 if (frent->fe_off + frent->fe_len > total || 624 (frent->fe_off + frent->fe_len == total && frent->fe_mff)) 625 goto bad_fragment; 626 } else { 627 if (frent->fe_off + frent->fe_len == total && !frent->fe_mff) 628 goto bad_fragment; 629 } 630 631 /* Find neighbors for newly inserted fragment */ 632 prev = pf_frent_previous(frag, frent); 633 if (prev == NULL) { 634 after = TAILQ_FIRST(&frag->fr_queue); 635 KASSERT(after != NULL, ("after != NULL")); 636 } else { 637 after = TAILQ_NEXT(prev, fr_next); 638 } 639 640 if (prev != NULL && prev->fe_off + prev->fe_len > frent->fe_off) { 641 uint16_t precut; 642 643 precut = prev->fe_off + prev->fe_len - frent->fe_off; 644 if (precut >= frent->fe_len) 645 goto bad_fragment; 646 DPFPRINTF(("overlap -%d", precut)); 647 m_adj(frent->fe_m, precut); 648 frent->fe_off += precut; 649 frent->fe_len -= precut; 650 } 651 652 for (; after != NULL && frent->fe_off + frent->fe_len > after->fe_off; 653 after = next) { 654 uint16_t aftercut; 655 656 aftercut = frent->fe_off + frent->fe_len - after->fe_off; 657 DPFPRINTF(("adjust overlap %d", aftercut)); 658 if (aftercut < after->fe_len) { 659 m_adj(after->fe_m, aftercut); 660 old_index = pf_frent_index(after); 661 after->fe_off += aftercut; 662 after->fe_len -= aftercut; 663 new_index = pf_frent_index(after); 664 if (old_index != new_index) { 665 DPFPRINTF(("frag index %d, new %d", 666 old_index, new_index)); 667 /* Fragment switched queue as fe_off changed */ 668 after->fe_off -= aftercut; 669 after->fe_len += aftercut; 670 /* Remove restored fragment from old queue */ 671 pf_frent_remove(frag, after); 672 after->fe_off += aftercut; 673 after->fe_len -= aftercut; 674 /* Insert into correct queue */ 675 if (pf_frent_insert(frag, after, prev)) { 676 DPFPRINTF( 677 ("fragment requeue limit exceeded")); 678 m_freem(after->fe_m); 679 uma_zfree(V_pf_frent_z, after); 680 /* There is not way to recover */ 681 goto bad_fragment; 682 } 683 } 684 break; 685 } 686 687 /* This fragment is completely overlapped, lose it. */ 688 next = TAILQ_NEXT(after, fr_next); 689 pf_frent_remove(frag, after); 690 m_freem(after->fe_m); 691 uma_zfree(V_pf_frent_z, after); 692 } 693 694 /* If part of the queue gets too long, there is not way to recover. */ 695 if (pf_frent_insert(frag, frent, prev)) { 696 DPFPRINTF(("fragment queue limit exceeded")); 697 goto bad_fragment; 698 } 699 700 return (frag); 701 702bad_fragment: 703 REASON_SET(reason, PFRES_FRAG); 704drop_fragment: 705 uma_zfree(V_pf_frent_z, frent); 706 return (NULL); 707} 708 709static struct mbuf * 710pf_join_fragment(struct pf_fragment *frag) 711{ 712 struct mbuf *m, *m2; 713 struct pf_frent *frent, *next; 714 715 frent = TAILQ_FIRST(&frag->fr_queue); 716 next = TAILQ_NEXT(frent, fr_next); 717 718 m = frent->fe_m; 719 m_adj(m, (frent->fe_hdrlen + frent->fe_len) - m->m_pkthdr.len); 720 uma_zfree(V_pf_frent_z, frent); 721 for (frent = next; frent != NULL; frent = next) { 722 next = TAILQ_NEXT(frent, fr_next); 723 724 m2 = frent->fe_m; 725 /* Strip off ip header. */ 726 m_adj(m2, frent->fe_hdrlen); 727 /* Strip off any trailing bytes. */ 728 m_adj(m2, frent->fe_len - m2->m_pkthdr.len); 729 730 uma_zfree(V_pf_frent_z, frent); 731 m_cat(m, m2); 732 } 733 734 /* Remove from fragment queue. */ 735 pf_remove_fragment(frag); 736 737 return (m); 738} 739 740#ifdef INET 741static int 742pf_reassemble(struct mbuf **m0, struct ip *ip, int dir, u_short *reason) 743{ 744 struct mbuf *m = *m0; 745 struct pf_frent *frent; 746 struct pf_fragment *frag; 747 struct pf_fragment_cmp key; 748 uint16_t total, hdrlen; 749 750 /* Get an entry for the fragment queue */ 751 if ((frent = pf_create_fragment(reason)) == NULL) 752 return (PF_DROP); 753 754 frent->fe_m = m; 755 frent->fe_hdrlen = ip->ip_hl << 2; 756 frent->fe_extoff = 0; 757 frent->fe_len = ntohs(ip->ip_len) - (ip->ip_hl << 2); 758 frent->fe_off = (ntohs(ip->ip_off) & IP_OFFMASK) << 3; 759 frent->fe_mff = ntohs(ip->ip_off) & IP_MF; 760 761 pf_ip2key(ip, dir, &key); 762 763 if ((frag = pf_fillup_fragment(&key, frent, reason)) == NULL) 764 return (PF_DROP); 765 766 /* The mbuf is part of the fragment entry, no direct free or access */ 767 m = *m0 = NULL; 768 769 if (frag->fr_holes) { 770 DPFPRINTF(("frag %d, holes %d", frag->fr_id, frag->fr_holes)); 771 return (PF_PASS); /* drop because *m0 is NULL, no error */ 772 } 773 774 /* We have all the data */ 775 frent = TAILQ_FIRST(&frag->fr_queue); 776 KASSERT(frent != NULL, ("frent != NULL")); 777 total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off + 778 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len; 779 hdrlen = frent->fe_hdrlen; 780 781 m = *m0 = pf_join_fragment(frag); 782 frag = NULL; 783 784 if (m->m_flags & M_PKTHDR) { 785 int plen = 0; 786 for (m = *m0; m; m = m->m_next) 787 plen += m->m_len; 788 m = *m0; 789 m->m_pkthdr.len = plen; 790 } 791 792 ip = mtod(m, struct ip *); 793 ip->ip_sum = pf_cksum_fixup(ip->ip_sum, ip->ip_len, 794 htons(hdrlen + total), 0); 795 ip->ip_len = htons(hdrlen + total); 796 ip->ip_sum = pf_cksum_fixup(ip->ip_sum, ip->ip_off, 797 ip->ip_off & ~(IP_MF|IP_OFFMASK), 0); 798 ip->ip_off &= ~(IP_MF|IP_OFFMASK); 799 800 if (hdrlen + total > IP_MAXPACKET) { 801 DPFPRINTF(("drop: too big: %d", total)); 802 ip->ip_len = 0; 803 REASON_SET(reason, PFRES_SHORT); 804 /* PF_DROP requires a valid mbuf *m0 in pf_test() */ 805 return (PF_DROP); 806 } 807 808 DPFPRINTF(("complete: %p(%d)\n", m, ntohs(ip->ip_len))); 809 return (PF_PASS); 810} 811#endif /* INET */ 812 813#ifdef INET6 814static int 815pf_reassemble6(struct mbuf **m0, struct ip6_hdr *ip6, struct ip6_frag *fraghdr, 816 uint16_t hdrlen, uint16_t extoff, u_short *reason) 817{ 818 struct mbuf *m = *m0; 819 struct pf_frent *frent; 820 struct pf_fragment *frag; 821 struct pf_fragment_cmp key; 822 struct m_tag *mtag; 823 struct pf_fragment_tag *ftag; 824 int off; 825 uint32_t frag_id; 826 uint16_t total, maxlen; 827 uint8_t proto; 828 829 PF_FRAG_LOCK(); 830 831 /* Get an entry for the fragment queue. */ 832 if ((frent = pf_create_fragment(reason)) == NULL) { 833 PF_FRAG_UNLOCK(); 834 return (PF_DROP); 835 } 836 837 frent->fe_m = m; 838 frent->fe_hdrlen = hdrlen; 839 frent->fe_extoff = extoff; 840 frent->fe_len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - hdrlen; 841 frent->fe_off = ntohs(fraghdr->ip6f_offlg & IP6F_OFF_MASK); 842 frent->fe_mff = fraghdr->ip6f_offlg & IP6F_MORE_FRAG; 843 844 key.frc_src.v6 = ip6->ip6_src; 845 key.frc_dst.v6 = ip6->ip6_dst; 846 key.frc_af = AF_INET6; 847 /* Only the first fragment's protocol is relevant. */ 848 key.frc_proto = 0; 849 key.frc_id = fraghdr->ip6f_ident; 850 851 if ((frag = pf_fillup_fragment(&key, frent, reason)) == NULL) { 852 PF_FRAG_UNLOCK(); 853 return (PF_DROP); 854 } 855 856 /* The mbuf is part of the fragment entry, no direct free or access. */ 857 m = *m0 = NULL; 858 859 if (frag->fr_holes) { 860 DPFPRINTF(("frag %d, holes %d", frag->fr_id, frag->fr_holes)); 861 PF_FRAG_UNLOCK(); 862 return (PF_PASS); /* Drop because *m0 is NULL, no error. */ 863 } 864 865 /* We have all the data. */ 866 frent = TAILQ_FIRST(&frag->fr_queue); 867 KASSERT(frent != NULL, ("frent != NULL")); 868 extoff = frent->fe_extoff; 869 maxlen = frag->fr_maxlen; 870 frag_id = frag->fr_id; 871 total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off + 872 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len; 873 hdrlen = frent->fe_hdrlen - sizeof(struct ip6_frag); 874 875 m = *m0 = pf_join_fragment(frag); 876 frag = NULL; 877 878 PF_FRAG_UNLOCK(); 879 880 /* Take protocol from first fragment header. */ 881 m = m_getptr(m, hdrlen + offsetof(struct ip6_frag, ip6f_nxt), &off); 882 KASSERT(m, ("%s: short mbuf chain", __func__)); 883 proto = *(mtod(m, caddr_t) + off); 884 m = *m0; 885 886 /* Delete frag6 header */ 887 if (ip6_deletefraghdr(m, hdrlen, M_NOWAIT) != 0) 888 goto fail; 889 890 if (m->m_flags & M_PKTHDR) { 891 int plen = 0; 892 for (m = *m0; m; m = m->m_next) 893 plen += m->m_len; 894 m = *m0; 895 m->m_pkthdr.len = plen; 896 } 897 898 if ((mtag = m_tag_get(PF_REASSEMBLED, sizeof(struct pf_fragment_tag), 899 M_NOWAIT)) == NULL) 900 goto fail; 901 ftag = (struct pf_fragment_tag *)(mtag + 1); 902 ftag->ft_hdrlen = hdrlen; 903 ftag->ft_extoff = extoff; 904 ftag->ft_maxlen = maxlen; 905 ftag->ft_id = frag_id; 906 m_tag_prepend(m, mtag); 907 908 ip6 = mtod(m, struct ip6_hdr *); 909 ip6->ip6_plen = htons(hdrlen - sizeof(struct ip6_hdr) + total); 910 if (extoff) { 911 /* Write protocol into next field of last extension header. */ 912 m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt), 913 &off); 914 KASSERT(m, ("%s: short mbuf chain", __func__)); 915 *(mtod(m, char *) + off) = proto; 916 m = *m0; 917 } else 918 ip6->ip6_nxt = proto; 919 920 if (hdrlen - sizeof(struct ip6_hdr) + total > IPV6_MAXPACKET) { 921 DPFPRINTF(("drop: too big: %d", total)); 922 ip6->ip6_plen = 0; 923 REASON_SET(reason, PFRES_SHORT); 924 /* PF_DROP requires a valid mbuf *m0 in pf_test6(). */ 925 return (PF_DROP); 926 } 927 928 DPFPRINTF(("complete: %p(%d)", m, ntohs(ip6->ip6_plen))); 929 return (PF_PASS); 930 931fail: 932 REASON_SET(reason, PFRES_MEMORY); 933 /* PF_DROP requires a valid mbuf *m0 in pf_test6(), will free later. */ 934 return (PF_DROP); 935} 936#endif /* INET6 */ 937 938#ifdef INET6 939int 940pf_refragment6(struct ifnet *ifp, struct mbuf **m0, struct m_tag *mtag) 941{ 942 struct mbuf *m = *m0, *t; 943 struct pf_fragment_tag *ftag = (struct pf_fragment_tag *)(mtag + 1); 944 struct pf_pdesc pd; 945 uint32_t frag_id; 946 uint16_t hdrlen, extoff, maxlen; 947 uint8_t proto; 948 int error, action; 949 950 hdrlen = ftag->ft_hdrlen; 951 extoff = ftag->ft_extoff; 952 maxlen = ftag->ft_maxlen; 953 frag_id = ftag->ft_id; 954 m_tag_delete(m, mtag); 955 mtag = NULL; 956 ftag = NULL; 957 958 if (extoff) { 959 int off; 960 961 /* Use protocol from next field of last extension header */ 962 m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt), 963 &off); 964 KASSERT((m != NULL), ("pf_refragment6: short mbuf chain")); 965 proto = *(mtod(m, caddr_t) + off); 966 *(mtod(m, char *) + off) = IPPROTO_FRAGMENT; 967 m = *m0; 968 } else { 969 struct ip6_hdr *hdr; 970 971 hdr = mtod(m, struct ip6_hdr *); 972 proto = hdr->ip6_nxt; 973 hdr->ip6_nxt = IPPROTO_FRAGMENT; 974 } 975 976 /* The MTU must be a multiple of 8 bytes, or we risk doing the 977 * fragmentation wrong. */ 978 maxlen = maxlen & ~7; 979 980 /* 981 * Maxlen may be less than 8 if there was only a single 982 * fragment. As it was fragmented before, add a fragment 983 * header also for a single fragment. If total or maxlen 984 * is less than 8, ip6_fragment() will return EMSGSIZE and 985 * we drop the packet. 986 */ 987 error = ip6_fragment(ifp, m, hdrlen, proto, maxlen, frag_id); 988 m = (*m0)->m_nextpkt; 989 (*m0)->m_nextpkt = NULL; 990 if (error == 0) { 991 /* The first mbuf contains the unfragmented packet. */ 992 m_freem(*m0); 993 *m0 = NULL; 994 action = PF_PASS; 995 } else { 996 /* Drop expects an mbuf to free. */ 997 DPFPRINTF(("refragment error %d", error)); 998 action = PF_DROP; 999 } 1000 for (t = m; m; m = t) { 1001 t = m->m_nextpkt; 1002 m->m_nextpkt = NULL; 1003 m->m_flags |= M_SKIP_FIREWALL; 1004 memset(&pd, 0, sizeof(pd)); 1005 pd.pf_mtag = pf_find_mtag(m); 1006 if (error == 0) 1007 ip6_forward(m, 0); 1008 else 1009 m_freem(m); 1010 } 1011 1012 return (action); 1013} 1014#endif /* INET6 */ 1015 1016#ifdef INET 1017int 1018pf_normalize_ip(struct mbuf **m0, int dir, struct pfi_kkif *kif, u_short *reason, 1019 struct pf_pdesc *pd) 1020{ 1021 struct mbuf *m = *m0; 1022 struct pf_krule *r; 1023 struct ip *h = mtod(m, struct ip *); 1024 int mff = (ntohs(h->ip_off) & IP_MF); 1025 int hlen = h->ip_hl << 2; 1026 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3; 1027 u_int16_t max; 1028 int ip_len; 1029 int ip_off; 1030 int tag = -1; 1031 int verdict; 1032 1033 PF_RULES_RASSERT(); 1034 1035 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr); 1036 while (r != NULL) { 1037 counter_u64_add(r->evaluations, 1); 1038 if (pfi_kkif_match(r->kif, kif) == r->ifnot) 1039 r = r->skip[PF_SKIP_IFP].ptr; 1040 else if (r->direction && r->direction != dir) 1041 r = r->skip[PF_SKIP_DIR].ptr; 1042 else if (r->af && r->af != AF_INET) 1043 r = r->skip[PF_SKIP_AF].ptr; 1044 else if (r->proto && r->proto != h->ip_p) 1045 r = r->skip[PF_SKIP_PROTO].ptr; 1046 else if (PF_MISMATCHAW(&r->src.addr, 1047 (struct pf_addr *)&h->ip_src.s_addr, AF_INET, 1048 r->src.neg, kif, M_GETFIB(m))) 1049 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 1050 else if (PF_MISMATCHAW(&r->dst.addr, 1051 (struct pf_addr *)&h->ip_dst.s_addr, AF_INET, 1052 r->dst.neg, NULL, M_GETFIB(m))) 1053 r = r->skip[PF_SKIP_DST_ADDR].ptr; 1054 else if (r->match_tag && !pf_match_tag(m, r, &tag, 1055 pd->pf_mtag ? pd->pf_mtag->tag : 0)) 1056 r = TAILQ_NEXT(r, entries); 1057 else 1058 break; 1059 } 1060 1061 if (r == NULL || r->action == PF_NOSCRUB) 1062 return (PF_PASS); 1063 else { 1064 counter_u64_add(r->packets[dir == PF_OUT], 1); 1065 counter_u64_add(r->bytes[dir == PF_OUT], pd->tot_len); 1066 } 1067 1068 /* Check for illegal packets */ 1069 if (hlen < (int)sizeof(struct ip)) { 1070 REASON_SET(reason, PFRES_NORM); 1071 goto drop; 1072 } 1073 1074 if (hlen > ntohs(h->ip_len)) { 1075 REASON_SET(reason, PFRES_NORM); 1076 goto drop; 1077 } 1078 1079 /* Clear IP_DF if the rule uses the no-df option */ 1080 if (r->rule_flag & PFRULE_NODF && h->ip_off & htons(IP_DF)) { 1081 u_int16_t ip_off = h->ip_off; 1082 1083 h->ip_off &= htons(~IP_DF); 1084 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0); 1085 } 1086 1087 /* We will need other tests here */ 1088 if (!fragoff && !mff) 1089 goto no_fragment; 1090 1091 /* We're dealing with a fragment now. Don't allow fragments 1092 * with IP_DF to enter the cache. If the flag was cleared by 1093 * no-df above, fine. Otherwise drop it. 1094 */ 1095 if (h->ip_off & htons(IP_DF)) { 1096 DPFPRINTF(("IP_DF\n")); 1097 goto bad; 1098 } 1099 1100 ip_len = ntohs(h->ip_len) - hlen; 1101 ip_off = (ntohs(h->ip_off) & IP_OFFMASK) << 3; 1102 1103 /* All fragments are 8 byte aligned */ 1104 if (mff && (ip_len & 0x7)) { 1105 DPFPRINTF(("mff and %d\n", ip_len)); 1106 goto bad; 1107 } 1108 1109 /* Respect maximum length */ 1110 if (fragoff + ip_len > IP_MAXPACKET) { 1111 DPFPRINTF(("max packet %d\n", fragoff + ip_len)); 1112 goto bad; 1113 } 1114 max = fragoff + ip_len; 1115 1116 /* Fully buffer all of the fragments 1117 * Might return a completely reassembled mbuf, or NULL */ 1118 PF_FRAG_LOCK(); 1119 DPFPRINTF(("reass frag %d @ %d-%d\n", h->ip_id, fragoff, max)); 1120 verdict = pf_reassemble(m0, h, dir, reason); 1121 PF_FRAG_UNLOCK(); 1122 1123 if (verdict != PF_PASS) 1124 return (PF_DROP); 1125 1126 m = *m0; 1127 if (m == NULL) 1128 return (PF_DROP); 1129 1130 h = mtod(m, struct ip *); 1131 1132 no_fragment: 1133 /* At this point, only IP_DF is allowed in ip_off */ 1134 if (h->ip_off & ~htons(IP_DF)) { 1135 u_int16_t ip_off = h->ip_off; 1136 1137 h->ip_off &= htons(IP_DF); 1138 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0); 1139 } 1140 1141 pf_scrub_ip(&m, r->rule_flag, r->min_ttl, r->set_tos); 1142 1143 return (PF_PASS); 1144 1145 bad: 1146 DPFPRINTF(("dropping bad fragment\n")); 1147 REASON_SET(reason, PFRES_FRAG); 1148 drop: 1149 if (r != NULL && r->log) 1150 PFLOG_PACKET(kif, m, AF_INET, dir, *reason, r, NULL, NULL, pd, 1151 1); 1152 1153 return (PF_DROP); 1154} 1155#endif 1156 1157#ifdef INET6 1158int 1159pf_normalize_ip6(struct mbuf **m0, int dir, struct pfi_kkif *kif, 1160 u_short *reason, struct pf_pdesc *pd) 1161{ 1162 struct mbuf *m = *m0; 1163 struct pf_krule *r; 1164 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 1165 int extoff; 1166 int off; 1167 struct ip6_ext ext; 1168 struct ip6_opt opt; 1169 struct ip6_opt_jumbo jumbo; 1170 struct ip6_frag frag; 1171 u_int32_t jumbolen = 0, plen; 1172 int optend; 1173 int ooff; 1174 u_int8_t proto; 1175 int terminal; 1176 1177 PF_RULES_RASSERT(); 1178 1179 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr); 1180 while (r != NULL) { 1181 counter_u64_add(r->evaluations, 1); 1182 if (pfi_kkif_match(r->kif, kif) == r->ifnot) 1183 r = r->skip[PF_SKIP_IFP].ptr; 1184 else if (r->direction && r->direction != dir) 1185 r = r->skip[PF_SKIP_DIR].ptr; 1186 else if (r->af && r->af != AF_INET6) 1187 r = r->skip[PF_SKIP_AF].ptr; 1188#if 0 /* header chain! */ 1189 else if (r->proto && r->proto != h->ip6_nxt) 1190 r = r->skip[PF_SKIP_PROTO].ptr; 1191#endif 1192 else if (PF_MISMATCHAW(&r->src.addr, 1193 (struct pf_addr *)&h->ip6_src, AF_INET6, 1194 r->src.neg, kif, M_GETFIB(m))) 1195 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 1196 else if (PF_MISMATCHAW(&r->dst.addr, 1197 (struct pf_addr *)&h->ip6_dst, AF_INET6, 1198 r->dst.neg, NULL, M_GETFIB(m))) 1199 r = r->skip[PF_SKIP_DST_ADDR].ptr; 1200 else 1201 break; 1202 } 1203 1204 if (r == NULL || r->action == PF_NOSCRUB) 1205 return (PF_PASS); 1206 else { 1207 counter_u64_add(r->packets[dir == PF_OUT], 1); 1208 counter_u64_add(r->bytes[dir == PF_OUT], pd->tot_len); 1209 } 1210 1211 /* Check for illegal packets */ 1212 if (sizeof(struct ip6_hdr) + IPV6_MAXPACKET < m->m_pkthdr.len) 1213 goto drop; 1214 1215 extoff = 0; 1216 off = sizeof(struct ip6_hdr); 1217 proto = h->ip6_nxt; 1218 terminal = 0; 1219 do { 1220 switch (proto) { 1221 case IPPROTO_FRAGMENT: 1222 goto fragment; 1223 break; 1224 case IPPROTO_AH: 1225 case IPPROTO_ROUTING: 1226 case IPPROTO_DSTOPTS: 1227 if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL, 1228 NULL, AF_INET6)) 1229 goto shortpkt; 1230 extoff = off; 1231 if (proto == IPPROTO_AH) 1232 off += (ext.ip6e_len + 2) * 4; 1233 else 1234 off += (ext.ip6e_len + 1) * 8; 1235 proto = ext.ip6e_nxt; 1236 break; 1237 case IPPROTO_HOPOPTS: 1238 if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL, 1239 NULL, AF_INET6)) 1240 goto shortpkt; 1241 extoff = off; 1242 optend = off + (ext.ip6e_len + 1) * 8; 1243 ooff = off + sizeof(ext); 1244 do { 1245 if (!pf_pull_hdr(m, ooff, &opt.ip6o_type, 1246 sizeof(opt.ip6o_type), NULL, NULL, 1247 AF_INET6)) 1248 goto shortpkt; 1249 if (opt.ip6o_type == IP6OPT_PAD1) { 1250 ooff++; 1251 continue; 1252 } 1253 if (!pf_pull_hdr(m, ooff, &opt, sizeof(opt), 1254 NULL, NULL, AF_INET6)) 1255 goto shortpkt; 1256 if (ooff + sizeof(opt) + opt.ip6o_len > optend) 1257 goto drop; 1258 switch (opt.ip6o_type) { 1259 case IP6OPT_JUMBO: 1260 if (h->ip6_plen != 0) 1261 goto drop; 1262 if (!pf_pull_hdr(m, ooff, &jumbo, 1263 sizeof(jumbo), NULL, NULL, 1264 AF_INET6)) 1265 goto shortpkt; 1266 memcpy(&jumbolen, jumbo.ip6oj_jumbo_len, 1267 sizeof(jumbolen)); 1268 jumbolen = ntohl(jumbolen); 1269 if (jumbolen <= IPV6_MAXPACKET) 1270 goto drop; 1271 if (sizeof(struct ip6_hdr) + jumbolen != 1272 m->m_pkthdr.len) 1273 goto drop; 1274 break; 1275 default: 1276 break; 1277 } 1278 ooff += sizeof(opt) + opt.ip6o_len; 1279 } while (ooff < optend); 1280 1281 off = optend; 1282 proto = ext.ip6e_nxt; 1283 break; 1284 default: 1285 terminal = 1; 1286 break; 1287 } 1288 } while (!terminal); 1289 1290 /* jumbo payload option must be present, or plen > 0 */ 1291 if (ntohs(h->ip6_plen) == 0) 1292 plen = jumbolen; 1293 else 1294 plen = ntohs(h->ip6_plen); 1295 if (plen == 0) 1296 goto drop; 1297 if (sizeof(struct ip6_hdr) + plen > m->m_pkthdr.len) 1298 goto shortpkt; 1299 1300 pf_scrub_ip6(&m, r->min_ttl); 1301 1302 return (PF_PASS); 1303 1304 fragment: 1305 /* Jumbo payload packets cannot be fragmented. */ 1306 plen = ntohs(h->ip6_plen); 1307 if (plen == 0 || jumbolen) 1308 goto drop; 1309 if (sizeof(struct ip6_hdr) + plen > m->m_pkthdr.len) 1310 goto shortpkt; 1311 1312 if (!pf_pull_hdr(m, off, &frag, sizeof(frag), NULL, NULL, AF_INET6)) 1313 goto shortpkt; 1314 1315 /* Offset now points to data portion. */ 1316 off += sizeof(frag); 1317 1318 /* Returns PF_DROP or *m0 is NULL or completely reassembled mbuf. */ 1319 if (pf_reassemble6(m0, h, &frag, off, extoff, reason) != PF_PASS) 1320 return (PF_DROP); 1321 m = *m0; 1322 if (m == NULL) 1323 return (PF_DROP); 1324 1325 pd->flags |= PFDESC_IP_REAS; 1326 return (PF_PASS); 1327 1328 shortpkt: 1329 REASON_SET(reason, PFRES_SHORT); 1330 if (r != NULL && r->log) 1331 PFLOG_PACKET(kif, m, AF_INET6, dir, *reason, r, NULL, NULL, pd, 1332 1); 1333 return (PF_DROP); 1334 1335 drop: 1336 REASON_SET(reason, PFRES_NORM); 1337 if (r != NULL && r->log) 1338 PFLOG_PACKET(kif, m, AF_INET6, dir, *reason, r, NULL, NULL, pd, 1339 1); 1340 return (PF_DROP); 1341} 1342#endif /* INET6 */ 1343 1344int 1345pf_normalize_tcp(int dir, struct pfi_kkif *kif, struct mbuf *m, int ipoff, 1346 int off, void *h, struct pf_pdesc *pd) 1347{ 1348 struct pf_krule *r, *rm = NULL; 1349 struct tcphdr *th = pd->hdr.tcp; 1350 int rewrite = 0; 1351 u_short reason; 1352 u_int8_t flags; 1353 sa_family_t af = pd->af; 1354 1355 PF_RULES_RASSERT(); 1356 1357 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr); 1358 while (r != NULL) { 1359 counter_u64_add(r->evaluations, 1); 1360 if (pfi_kkif_match(r->kif, kif) == r->ifnot) 1361 r = r->skip[PF_SKIP_IFP].ptr; 1362 else if (r->direction && r->direction != dir) 1363 r = r->skip[PF_SKIP_DIR].ptr; 1364 else if (r->af && r->af != af) 1365 r = r->skip[PF_SKIP_AF].ptr; 1366 else if (r->proto && r->proto != pd->proto) 1367 r = r->skip[PF_SKIP_PROTO].ptr; 1368 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af, 1369 r->src.neg, kif, M_GETFIB(m))) 1370 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 1371 else if (r->src.port_op && !pf_match_port(r->src.port_op, 1372 r->src.port[0], r->src.port[1], th->th_sport)) 1373 r = r->skip[PF_SKIP_SRC_PORT].ptr; 1374 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af, 1375 r->dst.neg, NULL, M_GETFIB(m))) 1376 r = r->skip[PF_SKIP_DST_ADDR].ptr; 1377 else if (r->dst.port_op && !pf_match_port(r->dst.port_op, 1378 r->dst.port[0], r->dst.port[1], th->th_dport)) 1379 r = r->skip[PF_SKIP_DST_PORT].ptr; 1380 else if (r->os_fingerprint != PF_OSFP_ANY && !pf_osfp_match( 1381 pf_osfp_fingerprint(pd, m, off, th), 1382 r->os_fingerprint)) 1383 r = TAILQ_NEXT(r, entries); 1384 else { 1385 rm = r; 1386 break; 1387 } 1388 } 1389 1390 if (rm == NULL || rm->action == PF_NOSCRUB) 1391 return (PF_PASS); 1392 else { 1393 counter_u64_add(r->packets[dir == PF_OUT], 1); 1394 counter_u64_add(r->bytes[dir == PF_OUT], pd->tot_len); 1395 } 1396 1397 if (rm->rule_flag & PFRULE_REASSEMBLE_TCP) 1398 pd->flags |= PFDESC_TCP_NORM; 1399 1400 flags = th->th_flags; 1401 if (flags & TH_SYN) { 1402 /* Illegal packet */ 1403 if (flags & TH_RST) 1404 goto tcp_drop; 1405 1406 if (flags & TH_FIN) 1407 goto tcp_drop; 1408 } else { 1409 /* Illegal packet */ 1410 if (!(flags & (TH_ACK|TH_RST))) 1411 goto tcp_drop; 1412 } 1413 1414 if (!(flags & TH_ACK)) { 1415 /* These flags are only valid if ACK is set */ 1416 if ((flags & TH_FIN) || (flags & TH_PUSH) || (flags & TH_URG)) 1417 goto tcp_drop; 1418 } 1419 1420 /* Check for illegal header length */ 1421 if (th->th_off < (sizeof(struct tcphdr) >> 2)) 1422 goto tcp_drop; 1423 1424 /* If flags changed, or reserved data set, then adjust */ 1425 if (flags != th->th_flags || th->th_x2 != 0) { 1426 u_int16_t ov, nv; 1427 1428 ov = *(u_int16_t *)(&th->th_ack + 1); 1429 th->th_flags = flags; 1430 th->th_x2 = 0; 1431 nv = *(u_int16_t *)(&th->th_ack + 1); 1432 1433 th->th_sum = pf_proto_cksum_fixup(m, th->th_sum, ov, nv, 0); 1434 rewrite = 1; 1435 } 1436 1437 /* Remove urgent pointer, if TH_URG is not set */ 1438 if (!(flags & TH_URG) && th->th_urp) { 1439 th->th_sum = pf_proto_cksum_fixup(m, th->th_sum, th->th_urp, 1440 0, 0); 1441 th->th_urp = 0; 1442 rewrite = 1; 1443 } 1444 1445 /* Process options */ 1446 if (r->max_mss && pf_normalize_tcpopt(r, m, th, off, pd->af)) 1447 rewrite = 1; 1448 1449 /* copy back packet headers if we sanitized */ 1450 if (rewrite) 1451 m_copyback(m, off, sizeof(*th), (caddr_t)th); 1452 1453 return (PF_PASS); 1454 1455 tcp_drop: 1456 REASON_SET(&reason, PFRES_NORM); 1457 if (rm != NULL && r->log) 1458 PFLOG_PACKET(kif, m, AF_INET, dir, reason, r, NULL, NULL, pd, 1459 1); 1460 return (PF_DROP); 1461} 1462 1463int 1464pf_normalize_tcp_init(struct mbuf *m, int off, struct pf_pdesc *pd, 1465 struct tcphdr *th, struct pf_state_peer *src, struct pf_state_peer *dst) 1466{ 1467 u_int32_t tsval, tsecr; 1468 u_int8_t hdr[60]; 1469 u_int8_t *opt; 1470 1471 KASSERT((src->scrub == NULL), 1472 ("pf_normalize_tcp_init: src->scrub != NULL")); 1473 1474 src->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT); 1475 if (src->scrub == NULL) 1476 return (1); 1477 1478 switch (pd->af) { 1479#ifdef INET 1480 case AF_INET: { 1481 struct ip *h = mtod(m, struct ip *); 1482 src->scrub->pfss_ttl = h->ip_ttl; 1483 break; 1484 } 1485#endif /* INET */ 1486#ifdef INET6 1487 case AF_INET6: { 1488 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 1489 src->scrub->pfss_ttl = h->ip6_hlim; 1490 break; 1491 } 1492#endif /* INET6 */ 1493 } 1494 1495 1496 /* 1497 * All normalizations below are only begun if we see the start of 1498 * the connections. They must all set an enabled bit in pfss_flags 1499 */ 1500 if ((th->th_flags & TH_SYN) == 0) 1501 return (0); 1502 1503 1504 if (th->th_off > (sizeof(struct tcphdr) >> 2) && src->scrub && 1505 pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) { 1506 /* Diddle with TCP options */ 1507 int hlen; 1508 opt = hdr + sizeof(struct tcphdr); 1509 hlen = (th->th_off << 2) - sizeof(struct tcphdr); 1510 while (hlen >= TCPOLEN_TIMESTAMP) { 1511 switch (*opt) { 1512 case TCPOPT_EOL: /* FALLTHROUGH */ 1513 case TCPOPT_NOP: 1514 opt++; 1515 hlen--; 1516 break; 1517 case TCPOPT_TIMESTAMP: 1518 if (opt[1] >= TCPOLEN_TIMESTAMP) { 1519 src->scrub->pfss_flags |= 1520 PFSS_TIMESTAMP; 1521 src->scrub->pfss_ts_mod = 1522 htonl(arc4random()); 1523 1524 /* note PFSS_PAWS not set yet */ 1525 memcpy(&tsval, &opt[2], 1526 sizeof(u_int32_t)); 1527 memcpy(&tsecr, &opt[6], 1528 sizeof(u_int32_t)); 1529 src->scrub->pfss_tsval0 = ntohl(tsval); 1530 src->scrub->pfss_tsval = ntohl(tsval); 1531 src->scrub->pfss_tsecr = ntohl(tsecr); 1532 getmicrouptime(&src->scrub->pfss_last); 1533 } 1534 /* FALLTHROUGH */ 1535 default: 1536 hlen -= MAX(opt[1], 2); 1537 opt += MAX(opt[1], 2); 1538 break; 1539 } 1540 } 1541 } 1542 1543 return (0); 1544} 1545 1546void 1547pf_normalize_tcp_cleanup(struct pf_state *state) 1548{ 1549 if (state->src.scrub) 1550 uma_zfree(V_pf_state_scrub_z, state->src.scrub); 1551 if (state->dst.scrub) 1552 uma_zfree(V_pf_state_scrub_z, state->dst.scrub); 1553 1554 /* Someday... flush the TCP segment reassembly descriptors. */ 1555} 1556 1557int 1558pf_normalize_tcp_stateful(struct mbuf *m, int off, struct pf_pdesc *pd, 1559 u_short *reason, struct tcphdr *th, struct pf_state *state, 1560 struct pf_state_peer *src, struct pf_state_peer *dst, int *writeback) 1561{ 1562 struct timeval uptime; 1563 u_int32_t tsval, tsecr; 1564 u_int tsval_from_last; 1565 u_int8_t hdr[60]; 1566 u_int8_t *opt; 1567 int copyback = 0; 1568 int got_ts = 0; 1569 size_t startoff; 1570 1571 KASSERT((src->scrub || dst->scrub), 1572 ("%s: src->scrub && dst->scrub!", __func__)); 1573 1574 /* 1575 * Enforce the minimum TTL seen for this connection. Negate a common 1576 * technique to evade an intrusion detection system and confuse 1577 * firewall state code. 1578 */ 1579 switch (pd->af) { 1580#ifdef INET 1581 case AF_INET: { 1582 if (src->scrub) { 1583 struct ip *h = mtod(m, struct ip *); 1584 if (h->ip_ttl > src->scrub->pfss_ttl) 1585 src->scrub->pfss_ttl = h->ip_ttl; 1586 h->ip_ttl = src->scrub->pfss_ttl; 1587 } 1588 break; 1589 } 1590#endif /* INET */ 1591#ifdef INET6 1592 case AF_INET6: { 1593 if (src->scrub) { 1594 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 1595 if (h->ip6_hlim > src->scrub->pfss_ttl) 1596 src->scrub->pfss_ttl = h->ip6_hlim; 1597 h->ip6_hlim = src->scrub->pfss_ttl; 1598 } 1599 break; 1600 } 1601#endif /* INET6 */ 1602 } 1603 1604 if (th->th_off > (sizeof(struct tcphdr) >> 2) && 1605 ((src->scrub && (src->scrub->pfss_flags & PFSS_TIMESTAMP)) || 1606 (dst->scrub && (dst->scrub->pfss_flags & PFSS_TIMESTAMP))) && 1607 pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) { 1608 /* Diddle with TCP options */ 1609 int hlen; 1610 opt = hdr + sizeof(struct tcphdr); 1611 hlen = (th->th_off << 2) - sizeof(struct tcphdr); 1612 while (hlen >= TCPOLEN_TIMESTAMP) { 1613 startoff = opt - (hdr + sizeof(struct tcphdr)); 1614 switch (*opt) { 1615 case TCPOPT_EOL: /* FALLTHROUGH */ 1616 case TCPOPT_NOP: 1617 opt++; 1618 hlen--; 1619 break; 1620 case TCPOPT_TIMESTAMP: 1621 /* Modulate the timestamps. Can be used for 1622 * NAT detection, OS uptime determination or 1623 * reboot detection. 1624 */ 1625 1626 if (got_ts) { 1627 /* Huh? Multiple timestamps!? */ 1628 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1629 DPFPRINTF(("multiple TS??")); 1630 pf_print_state(state); 1631 printf("\n"); 1632 } 1633 REASON_SET(reason, PFRES_TS); 1634 return (PF_DROP); 1635 } 1636 if (opt[1] >= TCPOLEN_TIMESTAMP) { 1637 memcpy(&tsval, &opt[2], 1638 sizeof(u_int32_t)); 1639 if (tsval && src->scrub && 1640 (src->scrub->pfss_flags & 1641 PFSS_TIMESTAMP)) { 1642 tsval = ntohl(tsval); 1643 pf_patch_32_unaligned(m, 1644 &th->th_sum, 1645 &opt[2], 1646 htonl(tsval + 1647 src->scrub->pfss_ts_mod), 1648 PF_ALGNMNT(startoff), 1649 0); 1650 copyback = 1; 1651 } 1652 1653 /* Modulate TS reply iff valid (!0) */ 1654 memcpy(&tsecr, &opt[6], 1655 sizeof(u_int32_t)); 1656 if (tsecr && dst->scrub && 1657 (dst->scrub->pfss_flags & 1658 PFSS_TIMESTAMP)) { 1659 tsecr = ntohl(tsecr) 1660 - dst->scrub->pfss_ts_mod; 1661 pf_patch_32_unaligned(m, 1662 &th->th_sum, 1663 &opt[6], 1664 htonl(tsecr), 1665 PF_ALGNMNT(startoff), 1666 0); 1667 copyback = 1; 1668 } 1669 got_ts = 1; 1670 } 1671 /* FALLTHROUGH */ 1672 default: 1673 hlen -= MAX(opt[1], 2); 1674 opt += MAX(opt[1], 2); 1675 break; 1676 } 1677 } 1678 if (copyback) { 1679 /* Copyback the options, caller copys back header */ 1680 *writeback = 1; 1681 m_copyback(m, off + sizeof(struct tcphdr), 1682 (th->th_off << 2) - sizeof(struct tcphdr), hdr + 1683 sizeof(struct tcphdr)); 1684 } 1685 } 1686 1687 1688 /* 1689 * Must invalidate PAWS checks on connections idle for too long. 1690 * The fastest allowed timestamp clock is 1ms. That turns out to 1691 * be about 24 days before it wraps. XXX Right now our lowerbound 1692 * TS echo check only works for the first 12 days of a connection 1693 * when the TS has exhausted half its 32bit space 1694 */ 1695#define TS_MAX_IDLE (24*24*60*60) 1696#define TS_MAX_CONN (12*24*60*60) /* XXX remove when better tsecr check */ 1697 1698 getmicrouptime(&uptime); 1699 if (src->scrub && (src->scrub->pfss_flags & PFSS_PAWS) && 1700 (uptime.tv_sec - src->scrub->pfss_last.tv_sec > TS_MAX_IDLE || 1701 time_uptime - state->creation > TS_MAX_CONN)) { 1702 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1703 DPFPRINTF(("src idled out of PAWS\n")); 1704 pf_print_state(state); 1705 printf("\n"); 1706 } 1707 src->scrub->pfss_flags = (src->scrub->pfss_flags & ~PFSS_PAWS) 1708 | PFSS_PAWS_IDLED; 1709 } 1710 if (dst->scrub && (dst->scrub->pfss_flags & PFSS_PAWS) && 1711 uptime.tv_sec - dst->scrub->pfss_last.tv_sec > TS_MAX_IDLE) { 1712 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1713 DPFPRINTF(("dst idled out of PAWS\n")); 1714 pf_print_state(state); 1715 printf("\n"); 1716 } 1717 dst->scrub->pfss_flags = (dst->scrub->pfss_flags & ~PFSS_PAWS) 1718 | PFSS_PAWS_IDLED; 1719 } 1720 1721 if (got_ts && src->scrub && dst->scrub && 1722 (src->scrub->pfss_flags & PFSS_PAWS) && 1723 (dst->scrub->pfss_flags & PFSS_PAWS)) { 1724 /* Validate that the timestamps are "in-window". 1725 * RFC1323 describes TCP Timestamp options that allow 1726 * measurement of RTT (round trip time) and PAWS 1727 * (protection against wrapped sequence numbers). PAWS 1728 * gives us a set of rules for rejecting packets on 1729 * long fat pipes (packets that were somehow delayed 1730 * in transit longer than the time it took to send the 1731 * full TCP sequence space of 4Gb). We can use these 1732 * rules and infer a few others that will let us treat 1733 * the 32bit timestamp and the 32bit echoed timestamp 1734 * as sequence numbers to prevent a blind attacker from 1735 * inserting packets into a connection. 1736 * 1737 * RFC1323 tells us: 1738 * - The timestamp on this packet must be greater than 1739 * or equal to the last value echoed by the other 1740 * endpoint. The RFC says those will be discarded 1741 * since it is a dup that has already been acked. 1742 * This gives us a lowerbound on the timestamp. 1743 * timestamp >= other last echoed timestamp 1744 * - The timestamp will be less than or equal to 1745 * the last timestamp plus the time between the 1746 * last packet and now. The RFC defines the max 1747 * clock rate as 1ms. We will allow clocks to be 1748 * up to 10% fast and will allow a total difference 1749 * or 30 seconds due to a route change. And this 1750 * gives us an upperbound on the timestamp. 1751 * timestamp <= last timestamp + max ticks 1752 * We have to be careful here. Windows will send an 1753 * initial timestamp of zero and then initialize it 1754 * to a random value after the 3whs; presumably to 1755 * avoid a DoS by having to call an expensive RNG 1756 * during a SYN flood. Proof MS has at least one 1757 * good security geek. 1758 * 1759 * - The TCP timestamp option must also echo the other 1760 * endpoints timestamp. The timestamp echoed is the 1761 * one carried on the earliest unacknowledged segment 1762 * on the left edge of the sequence window. The RFC 1763 * states that the host will reject any echoed 1764 * timestamps that were larger than any ever sent. 1765 * This gives us an upperbound on the TS echo. 1766 * tescr <= largest_tsval 1767 * - The lowerbound on the TS echo is a little more 1768 * tricky to determine. The other endpoint's echoed 1769 * values will not decrease. But there may be 1770 * network conditions that re-order packets and 1771 * cause our view of them to decrease. For now the 1772 * only lowerbound we can safely determine is that 1773 * the TS echo will never be less than the original 1774 * TS. XXX There is probably a better lowerbound. 1775 * Remove TS_MAX_CONN with better lowerbound check. 1776 * tescr >= other original TS 1777 * 1778 * It is also important to note that the fastest 1779 * timestamp clock of 1ms will wrap its 32bit space in 1780 * 24 days. So we just disable TS checking after 24 1781 * days of idle time. We actually must use a 12d 1782 * connection limit until we can come up with a better 1783 * lowerbound to the TS echo check. 1784 */ 1785 struct timeval delta_ts; 1786 int ts_fudge; 1787 1788 1789 /* 1790 * PFTM_TS_DIFF is how many seconds of leeway to allow 1791 * a host's timestamp. This can happen if the previous 1792 * packet got delayed in transit for much longer than 1793 * this packet. 1794 */ 1795 if ((ts_fudge = state->rule.ptr->timeout[PFTM_TS_DIFF]) == 0) 1796 ts_fudge = V_pf_default_rule.timeout[PFTM_TS_DIFF]; 1797 1798 /* Calculate max ticks since the last timestamp */ 1799#define TS_MAXFREQ 1100 /* RFC max TS freq of 1Khz + 10% skew */ 1800#define TS_MICROSECS 1000000 /* microseconds per second */ 1801 delta_ts = uptime; 1802 timevalsub(&delta_ts, &src->scrub->pfss_last); 1803 tsval_from_last = (delta_ts.tv_sec + ts_fudge) * TS_MAXFREQ; 1804 tsval_from_last += delta_ts.tv_usec / (TS_MICROSECS/TS_MAXFREQ); 1805 1806 if ((src->state >= TCPS_ESTABLISHED && 1807 dst->state >= TCPS_ESTABLISHED) && 1808 (SEQ_LT(tsval, dst->scrub->pfss_tsecr) || 1809 SEQ_GT(tsval, src->scrub->pfss_tsval + tsval_from_last) || 1810 (tsecr && (SEQ_GT(tsecr, dst->scrub->pfss_tsval) || 1811 SEQ_LT(tsecr, dst->scrub->pfss_tsval0))))) { 1812 /* Bad RFC1323 implementation or an insertion attack. 1813 * 1814 * - Solaris 2.6 and 2.7 are known to send another ACK 1815 * after the FIN,FIN|ACK,ACK closing that carries 1816 * an old timestamp. 1817 */ 1818 1819 DPFPRINTF(("Timestamp failed %c%c%c%c\n", 1820 SEQ_LT(tsval, dst->scrub->pfss_tsecr) ? '0' : ' ', 1821 SEQ_GT(tsval, src->scrub->pfss_tsval + 1822 tsval_from_last) ? '1' : ' ', 1823 SEQ_GT(tsecr, dst->scrub->pfss_tsval) ? '2' : ' ', 1824 SEQ_LT(tsecr, dst->scrub->pfss_tsval0)? '3' : ' ')); 1825 DPFPRINTF((" tsval: %u tsecr: %u +ticks: %u " 1826 "idle: %jus %lums\n", 1827 tsval, tsecr, tsval_from_last, 1828 (uintmax_t)delta_ts.tv_sec, 1829 delta_ts.tv_usec / 1000)); 1830 DPFPRINTF((" src->tsval: %u tsecr: %u\n", 1831 src->scrub->pfss_tsval, src->scrub->pfss_tsecr)); 1832 DPFPRINTF((" dst->tsval: %u tsecr: %u tsval0: %u" 1833 "\n", dst->scrub->pfss_tsval, 1834 dst->scrub->pfss_tsecr, dst->scrub->pfss_tsval0)); 1835 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1836 pf_print_state(state); 1837 pf_print_flags(th->th_flags); 1838 printf("\n"); 1839 } 1840 REASON_SET(reason, PFRES_TS); 1841 return (PF_DROP); 1842 } 1843 1844 /* XXX I'd really like to require tsecr but it's optional */ 1845 1846 } else if (!got_ts && (th->th_flags & TH_RST) == 0 && 1847 ((src->state == TCPS_ESTABLISHED && dst->state == TCPS_ESTABLISHED) 1848 || pd->p_len > 0 || (th->th_flags & TH_SYN)) && 1849 src->scrub && dst->scrub && 1850 (src->scrub->pfss_flags & PFSS_PAWS) && 1851 (dst->scrub->pfss_flags & PFSS_PAWS)) { 1852 /* Didn't send a timestamp. Timestamps aren't really useful 1853 * when: 1854 * - connection opening or closing (often not even sent). 1855 * but we must not let an attacker to put a FIN on a 1856 * data packet to sneak it through our ESTABLISHED check. 1857 * - on a TCP reset. RFC suggests not even looking at TS. 1858 * - on an empty ACK. The TS will not be echoed so it will 1859 * probably not help keep the RTT calculation in sync and 1860 * there isn't as much danger when the sequence numbers 1861 * got wrapped. So some stacks don't include TS on empty 1862 * ACKs :-( 1863 * 1864 * To minimize the disruption to mostly RFC1323 conformant 1865 * stacks, we will only require timestamps on data packets. 1866 * 1867 * And what do ya know, we cannot require timestamps on data 1868 * packets. There appear to be devices that do legitimate 1869 * TCP connection hijacking. There are HTTP devices that allow 1870 * a 3whs (with timestamps) and then buffer the HTTP request. 1871 * If the intermediate device has the HTTP response cache, it 1872 * will spoof the response but not bother timestamping its 1873 * packets. So we can look for the presence of a timestamp in 1874 * the first data packet and if there, require it in all future 1875 * packets. 1876 */ 1877 1878 if (pd->p_len > 0 && (src->scrub->pfss_flags & PFSS_DATA_TS)) { 1879 /* 1880 * Hey! Someone tried to sneak a packet in. Or the 1881 * stack changed its RFC1323 behavior?!?! 1882 */ 1883 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1884 DPFPRINTF(("Did not receive expected RFC1323 " 1885 "timestamp\n")); 1886 pf_print_state(state); 1887 pf_print_flags(th->th_flags); 1888 printf("\n"); 1889 } 1890 REASON_SET(reason, PFRES_TS); 1891 return (PF_DROP); 1892 } 1893 } 1894 1895 1896 /* 1897 * We will note if a host sends his data packets with or without 1898 * timestamps. And require all data packets to contain a timestamp 1899 * if the first does. PAWS implicitly requires that all data packets be 1900 * timestamped. But I think there are middle-man devices that hijack 1901 * TCP streams immediately after the 3whs and don't timestamp their 1902 * packets (seen in a WWW accelerator or cache). 1903 */ 1904 if (pd->p_len > 0 && src->scrub && (src->scrub->pfss_flags & 1905 (PFSS_TIMESTAMP|PFSS_DATA_TS|PFSS_DATA_NOTS)) == PFSS_TIMESTAMP) { 1906 if (got_ts) 1907 src->scrub->pfss_flags |= PFSS_DATA_TS; 1908 else { 1909 src->scrub->pfss_flags |= PFSS_DATA_NOTS; 1910 if (V_pf_status.debug >= PF_DEBUG_MISC && dst->scrub && 1911 (dst->scrub->pfss_flags & PFSS_TIMESTAMP)) { 1912 /* Don't warn if other host rejected RFC1323 */ 1913 DPFPRINTF(("Broken RFC1323 stack did not " 1914 "timestamp data packet. Disabled PAWS " 1915 "security.\n")); 1916 pf_print_state(state); 1917 pf_print_flags(th->th_flags); 1918 printf("\n"); 1919 } 1920 } 1921 } 1922 1923 1924 /* 1925 * Update PAWS values 1926 */ 1927 if (got_ts && src->scrub && PFSS_TIMESTAMP == (src->scrub->pfss_flags & 1928 (PFSS_PAWS_IDLED|PFSS_TIMESTAMP))) { 1929 getmicrouptime(&src->scrub->pfss_last); 1930 if (SEQ_GEQ(tsval, src->scrub->pfss_tsval) || 1931 (src->scrub->pfss_flags & PFSS_PAWS) == 0) 1932 src->scrub->pfss_tsval = tsval; 1933 1934 if (tsecr) { 1935 if (SEQ_GEQ(tsecr, src->scrub->pfss_tsecr) || 1936 (src->scrub->pfss_flags & PFSS_PAWS) == 0) 1937 src->scrub->pfss_tsecr = tsecr; 1938 1939 if ((src->scrub->pfss_flags & PFSS_PAWS) == 0 && 1940 (SEQ_LT(tsval, src->scrub->pfss_tsval0) || 1941 src->scrub->pfss_tsval0 == 0)) { 1942 /* tsval0 MUST be the lowest timestamp */ 1943 src->scrub->pfss_tsval0 = tsval; 1944 } 1945 1946 /* Only fully initialized after a TS gets echoed */ 1947 if ((src->scrub->pfss_flags & PFSS_PAWS) == 0) 1948 src->scrub->pfss_flags |= PFSS_PAWS; 1949 } 1950 } 1951 1952 /* I have a dream.... TCP segment reassembly.... */ 1953 return (0); 1954} 1955 1956static int 1957pf_normalize_tcpopt(struct pf_krule *r, struct mbuf *m, struct tcphdr *th, 1958 int off, sa_family_t af) 1959{ 1960 u_int16_t *mss; 1961 int thoff; 1962 int opt, cnt, optlen = 0; 1963 int rewrite = 0; 1964 u_char opts[TCP_MAXOLEN]; 1965 u_char *optp = opts; 1966 size_t startoff; 1967 1968 thoff = th->th_off << 2; 1969 cnt = thoff - sizeof(struct tcphdr); 1970 1971 if (cnt > 0 && !pf_pull_hdr(m, off + sizeof(*th), opts, cnt, 1972 NULL, NULL, af)) 1973 return (rewrite); 1974 1975 for (; cnt > 0; cnt -= optlen, optp += optlen) { 1976 startoff = optp - opts; 1977 opt = optp[0]; 1978 if (opt == TCPOPT_EOL) 1979 break; 1980 if (opt == TCPOPT_NOP) 1981 optlen = 1; 1982 else { 1983 if (cnt < 2) 1984 break; 1985 optlen = optp[1]; 1986 if (optlen < 2 || optlen > cnt) 1987 break; 1988 } 1989 switch (opt) { 1990 case TCPOPT_MAXSEG: 1991 mss = (u_int16_t *)(optp + 2); 1992 if ((ntohs(*mss)) > r->max_mss) { 1993 pf_patch_16_unaligned(m, 1994 &th->th_sum, 1995 mss, htons(r->max_mss), 1996 PF_ALGNMNT(startoff), 1997 0); 1998 rewrite = 1; 1999 } 2000 break; 2001 default: 2002 break; 2003 } 2004 } 2005 2006 if (rewrite) 2007 m_copyback(m, off + sizeof(*th), thoff - sizeof(*th), opts); 2008 2009 return (rewrite); 2010} 2011 2012#ifdef INET 2013static void 2014pf_scrub_ip(struct mbuf **m0, u_int32_t flags, u_int8_t min_ttl, u_int8_t tos) 2015{ 2016 struct mbuf *m = *m0; 2017 struct ip *h = mtod(m, struct ip *); 2018 2019 /* Clear IP_DF if no-df was requested */ 2020 if (flags & PFRULE_NODF && h->ip_off & htons(IP_DF)) { 2021 u_int16_t ip_off = h->ip_off; 2022 2023 h->ip_off &= htons(~IP_DF); 2024 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0); 2025 } 2026 2027 /* Enforce a minimum ttl, may cause endless packet loops */ 2028 if (min_ttl && h->ip_ttl < min_ttl) { 2029 u_int16_t ip_ttl = h->ip_ttl; 2030 2031 h->ip_ttl = min_ttl; 2032 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_ttl, h->ip_ttl, 0); 2033 } 2034 2035 /* Enforce tos */ 2036 if (flags & PFRULE_SET_TOS) { 2037 u_int16_t ov, nv; 2038 2039 ov = *(u_int16_t *)h; 2040 h->ip_tos = tos | (h->ip_tos & IPTOS_ECN_MASK); 2041 nv = *(u_int16_t *)h; 2042 2043 h->ip_sum = pf_cksum_fixup(h->ip_sum, ov, nv, 0); 2044 } 2045 2046 /* random-id, but not for fragments */ 2047 if (flags & PFRULE_RANDOMID && !(h->ip_off & ~htons(IP_DF))) { 2048 uint16_t ip_id = h->ip_id; 2049 2050 ip_fillid(h); 2051 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_id, h->ip_id, 0); 2052 } 2053} 2054#endif /* INET */ 2055 2056#ifdef INET6 2057static void 2058pf_scrub_ip6(struct mbuf **m0, u_int8_t min_ttl) 2059{ 2060 struct mbuf *m = *m0; 2061 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 2062 2063 /* Enforce a minimum ttl, may cause endless packet loops */ 2064 if (min_ttl && h->ip6_hlim < min_ttl) 2065 h->ip6_hlim = min_ttl; 2066} 2067#endif 2068