pf_norm.c revision 284571
1/*- 2 * Copyright 2001 Niels Provos <provos@citi.umich.edu> 3 * Copyright 2011 Alexander Bluhm <bluhm@openbsd.org> 4 * All rights reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 16 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 17 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 18 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 19 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 20 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 21 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 22 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 23 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 24 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 25 * 26 * $OpenBSD: pf_norm.c,v 1.114 2009/01/29 14:11:45 henning Exp $ 27 */ 28 29#include <sys/cdefs.h> 30__FBSDID("$FreeBSD: stable/10/sys/netpfil/pf/pf_norm.c 284571 2015-06-18 20:34:39Z kp $"); 31 32#include "opt_inet.h" 33#include "opt_inet6.h" 34#include "opt_pf.h" 35 36#include <sys/param.h> 37#include <sys/lock.h> 38#include <sys/mbuf.h> 39#include <sys/mutex.h> 40#include <sys/refcount.h> 41#include <sys/rwlock.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 <netinet/tcp.h> 54#include <netinet/tcp_fsm.h> 55#include <netinet/tcp_seq.h> 56 57#ifdef INET6 58#include <netinet/ip6.h> 59#endif /* INET6 */ 60 61struct pf_frent { 62 TAILQ_ENTRY(pf_frent) fr_next; 63 struct mbuf *fe_m; 64 uint16_t fe_hdrlen; /* ipv4 header lenght with ip options 65 ipv6, extension, fragment header */ 66 uint16_t fe_extoff; /* last extension header offset or 0 */ 67 uint16_t fe_len; /* fragment length */ 68 uint16_t fe_off; /* fragment offset */ 69 uint16_t fe_mff; /* more fragment flag */ 70}; 71 72struct pf_fragment_cmp { 73 struct pf_addr frc_src; 74 struct pf_addr frc_dst; 75 uint32_t frc_id; 76 sa_family_t frc_af; 77 uint8_t frc_proto; 78 uint8_t frc_direction; 79}; 80 81struct pf_fragment { 82 struct pf_fragment_cmp fr_key; 83#define fr_src fr_key.frc_src 84#define fr_dst fr_key.frc_dst 85#define fr_id fr_key.frc_id 86#define fr_af fr_key.frc_af 87#define fr_proto fr_key.frc_proto 88#define fr_direction fr_key.frc_direction 89 90 RB_ENTRY(pf_fragment) fr_entry; 91 TAILQ_ENTRY(pf_fragment) frag_next; 92 uint8_t fr_flags; /* status flags */ 93#define PFFRAG_SEENLAST 0x0001 /* Seen the last fragment for this */ 94#define PFFRAG_NOBUFFER 0x0002 /* Non-buffering fragment cache */ 95#define PFFRAG_DROP 0x0004 /* Drop all fragments */ 96#define BUFFER_FRAGMENTS(fr) (!((fr)->fr_flags & PFFRAG_NOBUFFER)) 97 uint16_t fr_max; /* fragment data max */ 98 uint32_t fr_timeout; 99 uint16_t fr_maxlen; /* maximum length of single fragment */ 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}; 108 109static struct mtx pf_frag_mtx; 110#define PF_FRAG_LOCK() mtx_lock(&pf_frag_mtx) 111#define PF_FRAG_UNLOCK() mtx_unlock(&pf_frag_mtx) 112#define PF_FRAG_ASSERT() mtx_assert(&pf_frag_mtx, MA_OWNED) 113 114VNET_DEFINE(uma_zone_t, pf_state_scrub_z); /* XXX: shared with pfsync */ 115 116static VNET_DEFINE(uma_zone_t, pf_frent_z); 117#define V_pf_frent_z VNET(pf_frent_z) 118static VNET_DEFINE(uma_zone_t, pf_frag_z); 119#define V_pf_frag_z VNET(pf_frag_z) 120 121TAILQ_HEAD(pf_fragqueue, pf_fragment); 122TAILQ_HEAD(pf_cachequeue, pf_fragment); 123static VNET_DEFINE(struct pf_fragqueue, pf_fragqueue); 124#define V_pf_fragqueue VNET(pf_fragqueue) 125static VNET_DEFINE(struct pf_cachequeue, pf_cachequeue); 126#define V_pf_cachequeue VNET(pf_cachequeue) 127RB_HEAD(pf_frag_tree, pf_fragment); 128static VNET_DEFINE(struct pf_frag_tree, pf_frag_tree); 129#define V_pf_frag_tree VNET(pf_frag_tree) 130static VNET_DEFINE(struct pf_frag_tree, pf_cache_tree); 131#define V_pf_cache_tree VNET(pf_cache_tree) 132static int pf_frag_compare(struct pf_fragment *, 133 struct pf_fragment *); 134static RB_PROTOTYPE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare); 135static RB_GENERATE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare); 136 137/* Private prototypes */ 138static void pf_free_fragment(struct pf_fragment *); 139static void pf_remove_fragment(struct pf_fragment *); 140static int pf_normalize_tcpopt(struct pf_rule *, struct mbuf *, 141 struct tcphdr *, int, sa_family_t); 142#ifdef INET 143static void pf_scrub_ip(struct mbuf **, u_int32_t, u_int8_t, 144 u_int8_t); 145static void pf_flush_fragments(void); 146static struct pf_fragment *pf_find_fragment(struct pf_fragment_cmp *key, 147 struct pf_frag_tree *tree); 148struct pf_frent *pf_create_fragment(u_short *); 149static int pf_reassemble(struct mbuf **, struct ip *, int, 150 u_short *); 151int pf_reassemble6(struct mbuf **, struct ip6_hdr *, 152 struct ip6_frag *, uint16_t, uint16_t, int, 153 u_short *); 154static struct mbuf *pf_fragcache(struct mbuf **, struct ip*, 155 struct pf_fragment **, int, int, int *); 156static struct pf_fragment *pf_fillup_fragment(struct pf_fragment_cmp *, 157 struct pf_frent *, u_short *); 158int pf_isfull_fragment(struct pf_fragment *); 159struct mbuf *pf_join_fragment(struct pf_fragment *); 160 161 162#endif /* INET */ 163#ifdef INET6 164static void pf_scrub_ip6(struct mbuf **, u_int8_t); 165#endif 166#define DPFPRINTF(x) do { \ 167 if (V_pf_status.debug >= PF_DEBUG_MISC) { \ 168 printf("%s: ", __func__); \ 169 printf x ; \ 170 } \ 171} while(0) 172 173static void 174pf_ip2key(struct ip *ip, int dir, struct pf_fragment_cmp *key) 175{ 176 177 key->frc_src.v4 = ip->ip_src; 178 key->frc_dst.v4 = ip->ip_dst; 179 key->frc_af = AF_INET; 180 key->frc_proto = ip->ip_p; 181 key->frc_id = ip->ip_id; 182 key->frc_direction = dir; 183} 184 185void 186pf_normalize_init(void) 187{ 188 189 V_pf_frag_z = uma_zcreate("pf frags", sizeof(struct pf_fragment), 190 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 191 V_pf_frent_z = uma_zcreate("pf frag entries", sizeof(struct pf_frent), 192 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 193 V_pf_state_scrub_z = uma_zcreate("pf state scrubs", 194 sizeof(struct pf_state_scrub), NULL, NULL, NULL, NULL, 195 UMA_ALIGN_PTR, 0); 196 197 V_pf_limits[PF_LIMIT_FRAGS].zone = V_pf_frent_z; 198 V_pf_limits[PF_LIMIT_FRAGS].limit = PFFRAG_FRENT_HIWAT; 199 uma_zone_set_max(V_pf_frent_z, PFFRAG_FRENT_HIWAT); 200 uma_zone_set_warning(V_pf_frent_z, "PF frag entries limit reached"); 201 202 mtx_init(&pf_frag_mtx, "pf fragments", NULL, MTX_DEF); 203 204 TAILQ_INIT(&V_pf_fragqueue); 205 TAILQ_INIT(&V_pf_cachequeue); 206} 207 208void 209pf_normalize_cleanup(void) 210{ 211 212 uma_zdestroy(V_pf_state_scrub_z); 213 uma_zdestroy(V_pf_frent_z); 214 uma_zdestroy(V_pf_frag_z); 215 216 mtx_destroy(&pf_frag_mtx); 217} 218 219static int 220pf_frag_compare(struct pf_fragment *a, struct pf_fragment *b) 221{ 222 int diff; 223 224 if ((diff = a->fr_id - b->fr_id) != 0) 225 return (diff); 226 if ((diff = a->fr_proto - b->fr_proto) != 0) 227 return (diff); 228 if ((diff = a->fr_af - b->fr_af) != 0) 229 return (diff); 230 if ((diff = pf_addr_cmp(&a->fr_src, &b->fr_src, a->fr_af)) != 0) 231 return (diff); 232 if ((diff = pf_addr_cmp(&a->fr_dst, &b->fr_dst, a->fr_af)) != 0) 233 return (diff); 234 return (0); 235} 236 237void 238pf_purge_expired_fragments(void) 239{ 240 struct pf_fragment *frag; 241 u_int32_t expire = time_uptime - 242 V_pf_default_rule.timeout[PFTM_FRAG]; 243 244 PF_FRAG_LOCK(); 245 while ((frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue)) != NULL) { 246 KASSERT((BUFFER_FRAGMENTS(frag)), 247 ("BUFFER_FRAGMENTS(frag) == 0: %s", __FUNCTION__)); 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 while ((frag = TAILQ_LAST(&V_pf_cachequeue, pf_cachequeue)) != NULL) { 256 KASSERT((!BUFFER_FRAGMENTS(frag)), 257 ("BUFFER_FRAGMENTS(frag) != 0: %s", __FUNCTION__)); 258 if (frag->fr_timeout > expire) 259 break; 260 261 DPFPRINTF(("expiring %d(%p)\n", frag->fr_id, frag)); 262 pf_free_fragment(frag); 263 KASSERT((TAILQ_EMPTY(&V_pf_cachequeue) || 264 TAILQ_LAST(&V_pf_cachequeue, pf_cachequeue) != frag), 265 ("!(TAILQ_EMPTY() || TAILQ_LAST() == farg): %s", 266 __FUNCTION__)); 267 } 268 PF_FRAG_UNLOCK(); 269} 270 271#ifdef INET 272/* 273 * Try to flush old fragments to make space for new ones 274 */ 275static void 276pf_flush_fragments(void) 277{ 278 struct pf_fragment *frag, *cache; 279 int goal; 280 281 PF_FRAG_ASSERT(); 282 283 goal = uma_zone_get_cur(V_pf_frent_z) * 9 / 10; 284 DPFPRINTF(("trying to free %d frag entriess\n", goal)); 285 while (goal < uma_zone_get_cur(V_pf_frent_z)) { 286 frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue); 287 if (frag) 288 pf_free_fragment(frag); 289 cache = TAILQ_LAST(&V_pf_cachequeue, pf_cachequeue); 290 if (cache) 291 pf_free_fragment(cache); 292 if (frag == NULL && cache == NULL) 293 break; 294 } 295} 296#endif /* INET */ 297 298/* Frees the fragments and all associated entries */ 299static void 300pf_free_fragment(struct pf_fragment *frag) 301{ 302 struct pf_frent *frent; 303 304 PF_FRAG_ASSERT(); 305 306 /* Free all fragments */ 307 if (BUFFER_FRAGMENTS(frag)) { 308 for (frent = TAILQ_FIRST(&frag->fr_queue); frent; 309 frent = TAILQ_FIRST(&frag->fr_queue)) { 310 TAILQ_REMOVE(&frag->fr_queue, frent, fr_next); 311 312 m_freem(frent->fe_m); 313 uma_zfree(V_pf_frent_z, frent); 314 } 315 } else { 316 for (frent = TAILQ_FIRST(&frag->fr_queue); frent; 317 frent = TAILQ_FIRST(&frag->fr_queue)) { 318 TAILQ_REMOVE(&frag->fr_queue, frent, fr_next); 319 320 KASSERT((TAILQ_EMPTY(&frag->fr_queue) || 321 TAILQ_FIRST(&frag->fr_queue)->fe_off > 322 frent->fe_len), 323 ("! (TAILQ_EMPTY() || TAILQ_FIRST()->fe_off >" 324 " frent->fe_len): %s", __func__)); 325 326 uma_zfree(V_pf_frent_z, frent); 327 } 328 } 329 330 pf_remove_fragment(frag); 331} 332 333#ifdef INET 334static struct pf_fragment * 335pf_find_fragment(struct pf_fragment_cmp *key, struct pf_frag_tree *tree) 336{ 337 struct pf_fragment *frag; 338 339 PF_FRAG_ASSERT(); 340 341 frag = RB_FIND(pf_frag_tree, tree, (struct pf_fragment *)key); 342 if (frag != NULL) { 343 /* XXX Are we sure we want to update the timeout? */ 344 frag->fr_timeout = time_uptime; 345 if (BUFFER_FRAGMENTS(frag)) { 346 TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next); 347 TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next); 348 } else { 349 TAILQ_REMOVE(&V_pf_cachequeue, frag, frag_next); 350 TAILQ_INSERT_HEAD(&V_pf_cachequeue, frag, frag_next); 351 } 352 } 353 354 return (frag); 355} 356#endif /* INET */ 357 358/* Removes a fragment from the fragment queue and frees the fragment */ 359 360static void 361pf_remove_fragment(struct pf_fragment *frag) 362{ 363 364 PF_FRAG_ASSERT(); 365 366 if (BUFFER_FRAGMENTS(frag)) { 367 RB_REMOVE(pf_frag_tree, &V_pf_frag_tree, frag); 368 TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next); 369 uma_zfree(V_pf_frag_z, frag); 370 } else { 371 RB_REMOVE(pf_frag_tree, &V_pf_cache_tree, frag); 372 TAILQ_REMOVE(&V_pf_cachequeue, frag, frag_next); 373 uma_zfree(V_pf_frag_z, frag); 374 } 375} 376 377#ifdef INET 378struct pf_frent * 379pf_create_fragment(u_short *reason) 380{ 381 struct pf_frent *frent; 382 383 PF_FRAG_ASSERT(); 384 385 frent = uma_zalloc(V_pf_frent_z, M_NOWAIT); 386 if (frent == NULL) { 387 pf_flush_fragments(); 388 frent = uma_zalloc(V_pf_frent_z, M_NOWAIT); 389 if (frent == NULL) { 390 REASON_SET(reason, PFRES_MEMORY); 391 return (NULL); 392 } 393 } 394 395 return (frent); 396} 397 398struct pf_fragment * 399pf_fillup_fragment(struct pf_fragment_cmp *key, struct pf_frent *frent, 400 u_short *reason) 401{ 402 struct pf_frent *after, *next, *prev; 403 struct pf_fragment *frag; 404 uint16_t total; 405 406 PF_FRAG_ASSERT(); 407 408 /* No empty fragments. */ 409 if (frent->fe_len == 0) { 410 DPFPRINTF(("bad fragment: len 0")); 411 goto bad_fragment; 412 } 413 414 /* All fragments are 8 byte aligned. */ 415 if (frent->fe_mff && (frent->fe_len & 0x7)) { 416 DPFPRINTF(("bad fragment: mff and len %d", frent->fe_len)); 417 goto bad_fragment; 418 } 419 420 /* Respect maximum length, IP_MAXPACKET == IPV6_MAXPACKET. */ 421 if (frent->fe_off + frent->fe_len > IP_MAXPACKET) { 422 DPFPRINTF(("bad fragment: max packet %d", 423 frent->fe_off + frent->fe_len)); 424 goto bad_fragment; 425 } 426 427 DPFPRINTF((key->frc_af == AF_INET ? 428 "reass frag %d @ %d-%d" : "reass frag %#08x @ %d-%d", 429 key->frc_id, frent->fe_off, frent->fe_off + frent->fe_len)); 430 431 /* Fully buffer all of the fragments in this fragment queue. */ 432 frag = pf_find_fragment(key, &V_pf_frag_tree); 433 434 /* Create a new reassembly queue for this packet. */ 435 if (frag == NULL) { 436 frag = uma_zalloc(V_pf_frag_z, M_NOWAIT); 437 if (frag == NULL) { 438 pf_flush_fragments(); 439 frag = uma_zalloc(V_pf_frag_z, M_NOWAIT); 440 if (frag == NULL) { 441 REASON_SET(reason, PFRES_MEMORY); 442 goto drop_fragment; 443 } 444 } 445 446 *(struct pf_fragment_cmp *)frag = *key; 447 frag->fr_timeout = time_second; 448 frag->fr_maxlen = frent->fe_len; 449 TAILQ_INIT(&frag->fr_queue); 450 451 RB_INSERT(pf_frag_tree, &V_pf_frag_tree, frag); 452 TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next); 453 454 /* We do not have a previous fragment. */ 455 TAILQ_INSERT_HEAD(&frag->fr_queue, frent, fr_next); 456 457 return (frag); 458 } 459 460 KASSERT(!TAILQ_EMPTY(&frag->fr_queue), ("!TAILQ_EMPTY()->fr_queue")); 461 462 /* Remember maximum fragment len for refragmentation. */ 463 if (frent->fe_len > frag->fr_maxlen) 464 frag->fr_maxlen = frent->fe_len; 465 466 /* Maximum data we have seen already. */ 467 total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off + 468 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len; 469 470 /* Non terminal fragments must have more fragments flag. */ 471 if (frent->fe_off + frent->fe_len < total && !frent->fe_mff) 472 goto bad_fragment; 473 474 /* Check if we saw the last fragment already. */ 475 if (!TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_mff) { 476 if (frent->fe_off + frent->fe_len > total || 477 (frent->fe_off + frent->fe_len == total && frent->fe_mff)) 478 goto bad_fragment; 479 } else { 480 if (frent->fe_off + frent->fe_len == total && !frent->fe_mff) 481 goto bad_fragment; 482 } 483 484 /* Find a fragment after the current one. */ 485 prev = NULL; 486 TAILQ_FOREACH(after, &frag->fr_queue, fr_next) { 487 if (after->fe_off > frent->fe_off) 488 break; 489 prev = after; 490 } 491 492 KASSERT(prev != NULL || after != NULL, 493 ("prev != NULL || after != NULL")); 494 495 if (prev != NULL && prev->fe_off + prev->fe_len > frent->fe_off) { 496 uint16_t precut; 497 498 precut = prev->fe_off + prev->fe_len - frent->fe_off; 499 if (precut >= frent->fe_len) 500 goto bad_fragment; 501 DPFPRINTF(("overlap -%d", precut)); 502 m_adj(frent->fe_m, precut); 503 frent->fe_off += precut; 504 frent->fe_len -= precut; 505 } 506 507 for (; after != NULL && frent->fe_off + frent->fe_len > after->fe_off; 508 after = next) { 509 uint16_t aftercut; 510 511 aftercut = frent->fe_off + frent->fe_len - after->fe_off; 512 DPFPRINTF(("adjust overlap %d", aftercut)); 513 if (aftercut < after->fe_len) { 514 m_adj(after->fe_m, aftercut); 515 after->fe_off += aftercut; 516 after->fe_len -= aftercut; 517 break; 518 } 519 520 /* This fragment is completely overlapped, lose it. */ 521 next = TAILQ_NEXT(after, fr_next); 522 m_freem(after->fe_m); 523 TAILQ_REMOVE(&frag->fr_queue, after, fr_next); 524 uma_zfree(V_pf_frent_z, after); 525 } 526 527 if (prev == NULL) 528 TAILQ_INSERT_HEAD(&frag->fr_queue, frent, fr_next); 529 else 530 TAILQ_INSERT_AFTER(&frag->fr_queue, prev, frent, fr_next); 531 532 return (frag); 533 534bad_fragment: 535 REASON_SET(reason, PFRES_FRAG); 536drop_fragment: 537 uma_zfree(V_pf_frent_z, frent); 538 return (NULL); 539} 540 541int 542pf_isfull_fragment(struct pf_fragment *frag) 543{ 544 struct pf_frent *frent, *next; 545 uint16_t off, total; 546 547 /* Check if we are completely reassembled */ 548 if (TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_mff) 549 return (0); 550 551 /* Maximum data we have seen already */ 552 total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off + 553 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len; 554 555 /* Check if we have all the data */ 556 off = 0; 557 for (frent = TAILQ_FIRST(&frag->fr_queue); frent; frent = next) { 558 next = TAILQ_NEXT(frent, fr_next); 559 560 off += frent->fe_len; 561 if (off < total && (next == NULL || next->fe_off != off)) { 562 DPFPRINTF(("missing fragment at %d, next %d, total %d", 563 off, next == NULL ? -1 : next->fe_off, total)); 564 return (0); 565 } 566 } 567 DPFPRINTF(("%d < %d?", off, total)); 568 if (off < total) 569 return (0); 570 KASSERT(off == total, ("off == total")); 571 572 return (1); 573} 574 575struct mbuf * 576pf_join_fragment(struct pf_fragment *frag) 577{ 578 struct mbuf *m, *m2; 579 struct pf_frent *frent, *next; 580 581 frent = TAILQ_FIRST(&frag->fr_queue); 582 next = TAILQ_NEXT(frent, fr_next); 583 584 /* Magic from ip_input. */ 585 m = frent->fe_m; 586 m2 = m->m_next; 587 m->m_next = NULL; 588 m_cat(m, m2); 589 uma_zfree(V_pf_frent_z, frent); 590 for (frent = next; frent != NULL; frent = next) { 591 next = TAILQ_NEXT(frent, fr_next); 592 593 m2 = frent->fe_m; 594 /* Strip off ip header. */ 595 m_adj(m2, frent->fe_hdrlen); 596 uma_zfree(V_pf_frent_z, frent); 597 m_cat(m, m2); 598 } 599 600 /* Remove from fragment queue. */ 601 pf_remove_fragment(frag); 602 603 return (m); 604} 605 606#define FR_IP_OFF(fr) ((ntohs((fr)->fr_ip->ip_off) & IP_OFFMASK) << 3) 607static int 608pf_reassemble(struct mbuf **m0, struct ip *ip, int dir, u_short *reason) 609{ 610 struct mbuf *m = *m0; 611 struct pf_frent *frent; 612 struct pf_fragment *frag; 613 struct pf_fragment_cmp key; 614 uint16_t total, hdrlen; 615 616 /* Get an entry for the fragment queue */ 617 if ((frent = pf_create_fragment(reason)) == NULL) 618 return (PF_DROP); 619 620 frent->fe_m = m; 621 frent->fe_hdrlen = ip->ip_hl << 2; 622 frent->fe_extoff = 0; 623 frent->fe_len = ntohs(ip->ip_len) - (ip->ip_hl << 2); 624 frent->fe_off = (ntohs(ip->ip_off) & IP_OFFMASK) << 3; 625 frent->fe_mff = ntohs(ip->ip_off) & IP_MF; 626 627 pf_ip2key(ip, dir, &key); 628 629 if ((frag = pf_fillup_fragment(&key, frent, reason)) == NULL) 630 return (PF_DROP); 631 632 /* The mbuf is part of the fragment entry, no direct free or access */ 633 m = *m0 = NULL; 634 635 if (!pf_isfull_fragment(frag)) 636 return (PF_PASS); /* drop because *m0 is NULL, no error */ 637 638 /* We have all the data */ 639 frent = TAILQ_FIRST(&frag->fr_queue); 640 KASSERT(frent != NULL, ("frent != NULL")); 641 total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off + 642 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len; 643 hdrlen = frent->fe_hdrlen; 644 645 m = *m0 = pf_join_fragment(frag); 646 frag = NULL; 647 648 if (m->m_flags & M_PKTHDR) { 649 int plen = 0; 650 for (m = *m0; m; m = m->m_next) 651 plen += m->m_len; 652 m = *m0; 653 m->m_pkthdr.len = plen; 654 } 655 656 ip = mtod(m, struct ip *); 657 ip->ip_len = htons(hdrlen + total); 658 ip->ip_off &= ~(IP_MF|IP_OFFMASK); 659 660 if (hdrlen + total > IP_MAXPACKET) { 661 DPFPRINTF(("drop: too big: %d", total)); 662 ip->ip_len = 0; 663 REASON_SET(reason, PFRES_SHORT); 664 /* PF_DROP requires a valid mbuf *m0 in pf_test() */ 665 return (PF_DROP); 666 } 667 668 DPFPRINTF(("complete: %p(%d)\n", m, ntohs(ip->ip_len))); 669 return (PF_PASS); 670} 671 672#ifdef INET6 673int 674pf_reassemble6(struct mbuf **m0, struct ip6_hdr *ip6, struct ip6_frag *fraghdr, 675 uint16_t hdrlen, uint16_t extoff, int dir, u_short *reason) 676{ 677 struct mbuf *m = *m0; 678 struct pf_frent *frent; 679 struct pf_fragment *frag; 680 struct pf_fragment_cmp key; 681 struct m_tag *mtag; 682 struct pf_fragment_tag *ftag; 683 int off; 684 uint16_t total, maxlen; 685 uint8_t proto; 686 687 PF_FRAG_LOCK(); 688 689 /* Get an entry for the fragment queue. */ 690 if ((frent = pf_create_fragment(reason)) == NULL) { 691 PF_FRAG_UNLOCK(); 692 return (PF_DROP); 693 } 694 695 frent->fe_m = m; 696 frent->fe_hdrlen = hdrlen; 697 frent->fe_extoff = extoff; 698 frent->fe_len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - hdrlen; 699 frent->fe_off = ntohs(fraghdr->ip6f_offlg & IP6F_OFF_MASK); 700 frent->fe_mff = fraghdr->ip6f_offlg & IP6F_MORE_FRAG; 701 702 key.frc_src.v6 = ip6->ip6_src; 703 key.frc_dst.v6 = ip6->ip6_dst; 704 key.frc_af = AF_INET6; 705 /* Only the first fragment's protocol is relevant. */ 706 key.frc_proto = 0; 707 key.frc_id = fraghdr->ip6f_ident; 708 key.frc_direction = dir; 709 710 if ((frag = pf_fillup_fragment(&key, frent, reason)) == NULL) { 711 PF_FRAG_UNLOCK(); 712 return (PF_DROP); 713 } 714 715 /* The mbuf is part of the fragment entry, no direct free or access. */ 716 m = *m0 = NULL; 717 718 if (!pf_isfull_fragment(frag)) { 719 PF_FRAG_UNLOCK(); 720 return (PF_PASS); /* Drop because *m0 is NULL, no error. */ 721 } 722 723 /* We have all the data. */ 724 extoff = frent->fe_extoff; 725 maxlen = frag->fr_maxlen; 726 frent = TAILQ_FIRST(&frag->fr_queue); 727 KASSERT(frent != NULL, ("frent != NULL")); 728 total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off + 729 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len; 730 hdrlen = frent->fe_hdrlen - sizeof(struct ip6_frag); 731 732 m = *m0 = pf_join_fragment(frag); 733 frag = NULL; 734 735 PF_FRAG_UNLOCK(); 736 737 /* Take protocol from first fragment header. */ 738 m = m_getptr(m, hdrlen + offsetof(struct ip6_frag, ip6f_nxt), &off); 739 KASSERT(m, ("%s: short mbuf chain", __func__)); 740 proto = *(mtod(m, caddr_t) + off); 741 m = *m0; 742 743 /* Delete frag6 header */ 744 if (ip6_deletefraghdr(m, hdrlen, M_NOWAIT) != 0) 745 goto fail; 746 747 if (m->m_flags & M_PKTHDR) { 748 int plen = 0; 749 for (m = *m0; m; m = m->m_next) 750 plen += m->m_len; 751 m = *m0; 752 m->m_pkthdr.len = plen; 753 } 754 755 if ((mtag = m_tag_get(PF_REASSEMBLED, sizeof(struct pf_fragment_tag), 756 M_NOWAIT)) == NULL) 757 goto fail; 758 ftag = (struct pf_fragment_tag *)(mtag + 1); 759 ftag->ft_hdrlen = hdrlen; 760 ftag->ft_extoff = extoff; 761 ftag->ft_maxlen = maxlen; 762 m_tag_prepend(m, mtag); 763 764 ip6 = mtod(m, struct ip6_hdr *); 765 ip6->ip6_plen = htons(hdrlen - sizeof(struct ip6_hdr) + total); 766 if (extoff) { 767 /* Write protocol into next field of last extension header. */ 768 m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt), 769 &off); 770 KASSERT(m, ("%s: short mbuf chain", __func__)); 771 *(mtod(m, char *) + off) = proto; 772 m = *m0; 773 } else 774 ip6->ip6_nxt = proto; 775 776 if (hdrlen - sizeof(struct ip6_hdr) + total > IPV6_MAXPACKET) { 777 DPFPRINTF(("drop: too big: %d", total)); 778 ip6->ip6_plen = 0; 779 REASON_SET(reason, PFRES_SHORT); 780 /* PF_DROP requires a valid mbuf *m0 in pf_test6(). */ 781 return (PF_DROP); 782 } 783 784 DPFPRINTF(("complete: %p(%d)", m, ntohs(ip6->ip6_plen))); 785 return (PF_PASS); 786 787fail: 788 REASON_SET(reason, PFRES_MEMORY); 789 /* PF_DROP requires a valid mbuf *m0 in pf_test6(), will free later. */ 790 return (PF_DROP); 791} 792 793#endif 794 795static struct mbuf * 796pf_fragcache(struct mbuf **m0, struct ip *h, struct pf_fragment **frag, int mff, 797 int drop, int *nomem) 798{ 799 struct mbuf *m = *m0; 800 struct pf_frent *frp, *fra, *cur = NULL; 801 int ip_len = ntohs(h->ip_len) - (h->ip_hl << 2); 802 u_int16_t off = ntohs(h->ip_off) << 3; 803 u_int16_t max = ip_len + off; 804 int hosed = 0; 805 806 PF_FRAG_ASSERT(); 807 KASSERT((*frag == NULL || !BUFFER_FRAGMENTS(*frag)), 808 ("!(*frag == NULL || !BUFFER_FRAGMENTS(*frag)): %s", __FUNCTION__)); 809 810 /* Create a new range queue for this packet */ 811 if (*frag == NULL) { 812 *frag = uma_zalloc(V_pf_frag_z, M_NOWAIT); 813 if (*frag == NULL) { 814 pf_flush_fragments(); 815 *frag = uma_zalloc(V_pf_frag_z, M_NOWAIT); 816 if (*frag == NULL) 817 goto no_mem; 818 } 819 820 /* Get an entry for the queue */ 821 cur = uma_zalloc(V_pf_frent_z, M_NOWAIT); 822 if (cur == NULL) { 823 uma_zfree(V_pf_frag_z, *frag); 824 *frag = NULL; 825 goto no_mem; 826 } 827 828 (*frag)->fr_flags = PFFRAG_NOBUFFER; 829 (*frag)->fr_max = 0; 830 (*frag)->fr_src.v4 = h->ip_src; 831 (*frag)->fr_dst.v4 = h->ip_dst; 832 (*frag)->fr_id = h->ip_id; 833 (*frag)->fr_timeout = time_uptime; 834 835 cur->fe_off = off; 836 cur->fe_len = max; /* TODO: fe_len = max - off ? */ 837 TAILQ_INIT(&(*frag)->fr_queue); 838 TAILQ_INSERT_HEAD(&(*frag)->fr_queue, cur, fr_next); 839 840 RB_INSERT(pf_frag_tree, &V_pf_cache_tree, *frag); 841 TAILQ_INSERT_HEAD(&V_pf_cachequeue, *frag, frag_next); 842 843 DPFPRINTF(("fragcache[%d]: new %d-%d\n", h->ip_id, off, max)); 844 845 goto pass; 846 } 847 848 /* 849 * Find a fragment after the current one: 850 * - off contains the real shifted offset. 851 */ 852 frp = NULL; 853 TAILQ_FOREACH(fra, &(*frag)->fr_queue, fr_next) { 854 if (fra->fe_off > off) 855 break; 856 frp = fra; 857 } 858 859 KASSERT((frp != NULL || fra != NULL), 860 ("!(frp != NULL || fra != NULL): %s", __FUNCTION__)); 861 862 if (frp != NULL) { 863 int precut; 864 865 precut = frp->fe_len - off; 866 if (precut >= ip_len) { 867 /* Fragment is entirely a duplicate */ 868 DPFPRINTF(("fragcache[%d]: dead (%d-%d) %d-%d\n", 869 h->ip_id, frp->fe_off, frp->fe_len, off, max)); 870 goto drop_fragment; 871 } 872 if (precut == 0) { 873 /* They are adjacent. Fixup cache entry */ 874 DPFPRINTF(("fragcache[%d]: adjacent (%d-%d) %d-%d\n", 875 h->ip_id, frp->fe_off, frp->fe_len, off, max)); 876 frp->fe_len = max; 877 } else if (precut > 0) { 878 /* The first part of this payload overlaps with a 879 * fragment that has already been passed. 880 * Need to trim off the first part of the payload. 881 * But to do so easily, we need to create another 882 * mbuf to throw the original header into. 883 */ 884 885 DPFPRINTF(("fragcache[%d]: chop %d (%d-%d) %d-%d\n", 886 h->ip_id, precut, frp->fe_off, frp->fe_len, off, 887 max)); 888 889 off += precut; 890 max -= precut; 891 /* Update the previous frag to encompass this one */ 892 frp->fe_len = max; 893 894 if (!drop) { 895 /* XXX Optimization opportunity 896 * This is a very heavy way to trim the payload. 897 * we could do it much faster by diddling mbuf 898 * internals but that would be even less legible 899 * than this mbuf magic. For my next trick, 900 * I'll pull a rabbit out of my laptop. 901 */ 902 *m0 = m_dup(m, M_NOWAIT); 903 if (*m0 == NULL) 904 goto no_mem; 905 /* From KAME Project : We have missed this! */ 906 m_adj(*m0, (h->ip_hl << 2) - 907 (*m0)->m_pkthdr.len); 908 909 KASSERT(((*m0)->m_next == NULL), 910 ("(*m0)->m_next != NULL: %s", 911 __FUNCTION__)); 912 m_adj(m, precut + (h->ip_hl << 2)); 913 m_cat(*m0, m); 914 m = *m0; 915 if (m->m_flags & M_PKTHDR) { 916 int plen = 0; 917 struct mbuf *t; 918 for (t = m; t; t = t->m_next) 919 plen += t->m_len; 920 m->m_pkthdr.len = plen; 921 } 922 923 924 h = mtod(m, struct ip *); 925 926 KASSERT(((int)m->m_len == 927 ntohs(h->ip_len) - precut), 928 ("m->m_len != ntohs(h->ip_len) - precut: %s", 929 __FUNCTION__)); 930 h->ip_off = htons(ntohs(h->ip_off) + 931 (precut >> 3)); 932 h->ip_len = htons(ntohs(h->ip_len) - precut); 933 } else { 934 hosed++; 935 } 936 } else { 937 /* There is a gap between fragments */ 938 939 DPFPRINTF(("fragcache[%d]: gap %d (%d-%d) %d-%d\n", 940 h->ip_id, -precut, frp->fe_off, frp->fe_len, off, 941 max)); 942 943 cur = uma_zalloc(V_pf_frent_z, M_NOWAIT); 944 if (cur == NULL) 945 goto no_mem; 946 947 cur->fe_off = off; 948 cur->fe_len = max; 949 TAILQ_INSERT_AFTER(&(*frag)->fr_queue, frp, cur, fr_next); 950 } 951 } 952 953 if (fra != NULL) { 954 int aftercut; 955 int merge = 0; 956 957 aftercut = max - fra->fe_off; 958 if (aftercut == 0) { 959 /* Adjacent fragments */ 960 DPFPRINTF(("fragcache[%d]: adjacent %d-%d (%d-%d)\n", 961 h->ip_id, off, max, fra->fe_off, fra->fe_len)); 962 fra->fe_off = off; 963 merge = 1; 964 } else if (aftercut > 0) { 965 /* Need to chop off the tail of this fragment */ 966 DPFPRINTF(("fragcache[%d]: chop %d %d-%d (%d-%d)\n", 967 h->ip_id, aftercut, off, max, fra->fe_off, 968 fra->fe_len)); 969 fra->fe_off = off; 970 max -= aftercut; 971 972 merge = 1; 973 974 if (!drop) { 975 m_adj(m, -aftercut); 976 if (m->m_flags & M_PKTHDR) { 977 int plen = 0; 978 struct mbuf *t; 979 for (t = m; t; t = t->m_next) 980 plen += t->m_len; 981 m->m_pkthdr.len = plen; 982 } 983 h = mtod(m, struct ip *); 984 KASSERT(((int)m->m_len == ntohs(h->ip_len) - aftercut), 985 ("m->m_len != ntohs(h->ip_len) - aftercut: %s", 986 __FUNCTION__)); 987 h->ip_len = htons(ntohs(h->ip_len) - aftercut); 988 } else { 989 hosed++; 990 } 991 } else if (frp == NULL) { 992 /* There is a gap between fragments */ 993 DPFPRINTF(("fragcache[%d]: gap %d %d-%d (%d-%d)\n", 994 h->ip_id, -aftercut, off, max, fra->fe_off, 995 fra->fe_len)); 996 997 cur = uma_zalloc(V_pf_frent_z, M_NOWAIT); 998 if (cur == NULL) 999 goto no_mem; 1000 1001 cur->fe_off = off; 1002 cur->fe_len = max; 1003 TAILQ_INSERT_HEAD(&(*frag)->fr_queue, cur, fr_next); 1004 } 1005 1006 1007 /* Need to glue together two separate fragment descriptors */ 1008 if (merge) { 1009 if (cur && fra->fe_off <= cur->fe_len) { 1010 /* Need to merge in a previous 'cur' */ 1011 DPFPRINTF(("fragcache[%d]: adjacent(merge " 1012 "%d-%d) %d-%d (%d-%d)\n", 1013 h->ip_id, cur->fe_off, cur->fe_len, off, 1014 max, fra->fe_off, fra->fe_len)); 1015 fra->fe_off = cur->fe_off; 1016 TAILQ_REMOVE(&(*frag)->fr_queue, cur, fr_next); 1017 uma_zfree(V_pf_frent_z, cur); 1018 cur = NULL; 1019 1020 } else if (frp && fra->fe_off <= frp->fe_len) { 1021 /* Need to merge in a modified 'frp' */ 1022 KASSERT((cur == NULL), ("cur != NULL: %s", 1023 __FUNCTION__)); 1024 DPFPRINTF(("fragcache[%d]: adjacent(merge " 1025 "%d-%d) %d-%d (%d-%d)\n", 1026 h->ip_id, frp->fe_off, frp->fe_len, off, 1027 max, fra->fe_off, fra->fe_len)); 1028 fra->fe_off = frp->fe_off; 1029 TAILQ_REMOVE(&(*frag)->fr_queue, frp, fr_next); 1030 uma_zfree(V_pf_frent_z, frp); 1031 frp = NULL; 1032 1033 } 1034 } 1035 } 1036 1037 if (hosed) { 1038 /* 1039 * We must keep tracking the overall fragment even when 1040 * we're going to drop it anyway so that we know when to 1041 * free the overall descriptor. Thus we drop the frag late. 1042 */ 1043 goto drop_fragment; 1044 } 1045 1046 1047 pass: 1048 /* Update maximum data size */ 1049 if ((*frag)->fr_max < max) 1050 (*frag)->fr_max = max; 1051 1052 /* This is the last segment */ 1053 if (!mff) 1054 (*frag)->fr_flags |= PFFRAG_SEENLAST; 1055 1056 /* Check if we are completely reassembled */ 1057 if (((*frag)->fr_flags & PFFRAG_SEENLAST) && 1058 TAILQ_FIRST(&(*frag)->fr_queue)->fe_off == 0 && 1059 TAILQ_FIRST(&(*frag)->fr_queue)->fe_len == (*frag)->fr_max) { 1060 /* Remove from fragment queue */ 1061 DPFPRINTF(("fragcache[%d]: done 0-%d\n", h->ip_id, 1062 (*frag)->fr_max)); 1063 pf_free_fragment(*frag); 1064 *frag = NULL; 1065 } 1066 1067 return (m); 1068 1069 no_mem: 1070 *nomem = 1; 1071 1072 /* Still need to pay attention to !IP_MF */ 1073 if (!mff && *frag != NULL) 1074 (*frag)->fr_flags |= PFFRAG_SEENLAST; 1075 1076 m_freem(m); 1077 return (NULL); 1078 1079 drop_fragment: 1080 1081 /* Still need to pay attention to !IP_MF */ 1082 if (!mff && *frag != NULL) 1083 (*frag)->fr_flags |= PFFRAG_SEENLAST; 1084 1085 if (drop) { 1086 /* This fragment has been deemed bad. Don't reass */ 1087 if (((*frag)->fr_flags & PFFRAG_DROP) == 0) 1088 DPFPRINTF(("fragcache[%d]: dropping overall fragment\n", 1089 h->ip_id)); 1090 (*frag)->fr_flags |= PFFRAG_DROP; 1091 } 1092 1093 m_freem(m); 1094 return (NULL); 1095} 1096 1097int 1098pf_refragment6(struct ifnet *ifp, struct mbuf **m0, struct m_tag *mtag) 1099{ 1100 struct mbuf *m = *m0, *t; 1101 struct pf_fragment_tag *ftag = (struct pf_fragment_tag *)(mtag + 1); 1102 struct pf_pdesc pd; 1103 uint16_t hdrlen, extoff, maxlen; 1104 uint8_t proto; 1105 int error, action; 1106 1107 hdrlen = ftag->ft_hdrlen; 1108 extoff = ftag->ft_extoff; 1109 maxlen = ftag->ft_maxlen; 1110 m_tag_delete(m, mtag); 1111 mtag = NULL; 1112 ftag = NULL; 1113 1114 if (extoff) { 1115 int off; 1116 1117 /* Use protocol from next field of last extension header */ 1118 m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt), 1119 &off); 1120 KASSERT((m != NULL), ("pf_refragment6: short mbuf chain")); 1121 proto = *(mtod(m, caddr_t) + off); 1122 *(mtod(m, char *) + off) = IPPROTO_FRAGMENT; 1123 m = *m0; 1124 } else { 1125 struct ip6_hdr *hdr; 1126 1127 hdr = mtod(m, struct ip6_hdr *); 1128 proto = hdr->ip6_nxt; 1129 hdr->ip6_nxt = IPPROTO_FRAGMENT; 1130 } 1131 1132 /* 1133 * Maxlen may be less than 8 if there was only a single 1134 * fragment. As it was fragmented before, add a fragment 1135 * header also for a single fragment. If total or maxlen 1136 * is less than 8, ip6_fragment() will return EMSGSIZE and 1137 * we drop the packet. 1138 */ 1139 error = ip6_fragment(ifp, m, hdrlen, proto, maxlen); 1140 m = (*m0)->m_nextpkt; 1141 (*m0)->m_nextpkt = NULL; 1142 if (error == 0) { 1143 /* The first mbuf contains the unfragmented packet. */ 1144 m_freem(*m0); 1145 *m0 = NULL; 1146 action = PF_PASS; 1147 } else { 1148 /* Drop expects an mbuf to free. */ 1149 DPFPRINTF(("refragment error %d", error)); 1150 action = PF_DROP; 1151 } 1152 for (t = m; m; m = t) { 1153 t = m->m_nextpkt; 1154 m->m_nextpkt = NULL; 1155 memset(&pd, 0, sizeof(pd)); 1156 pd.pf_mtag = pf_find_mtag(m); 1157 if (error == 0) 1158 ip6_forward(m, 0); 1159 else 1160 m_freem(m); 1161 } 1162 1163 return (action); 1164} 1165 1166int 1167pf_normalize_ip(struct mbuf **m0, int dir, struct pfi_kif *kif, u_short *reason, 1168 struct pf_pdesc *pd) 1169{ 1170 struct mbuf *m = *m0; 1171 struct pf_rule *r; 1172 struct pf_fragment *frag = NULL; 1173 struct pf_fragment_cmp key; 1174 struct ip *h = mtod(m, struct ip *); 1175 int mff = (ntohs(h->ip_off) & IP_MF); 1176 int hlen = h->ip_hl << 2; 1177 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3; 1178 u_int16_t max; 1179 int ip_len; 1180 int ip_off; 1181 int tag = -1; 1182 int verdict; 1183 1184 PF_RULES_RASSERT(); 1185 1186 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr); 1187 while (r != NULL) { 1188 r->evaluations++; 1189 if (pfi_kif_match(r->kif, kif) == r->ifnot) 1190 r = r->skip[PF_SKIP_IFP].ptr; 1191 else if (r->direction && r->direction != dir) 1192 r = r->skip[PF_SKIP_DIR].ptr; 1193 else if (r->af && r->af != AF_INET) 1194 r = r->skip[PF_SKIP_AF].ptr; 1195 else if (r->proto && r->proto != h->ip_p) 1196 r = r->skip[PF_SKIP_PROTO].ptr; 1197 else if (PF_MISMATCHAW(&r->src.addr, 1198 (struct pf_addr *)&h->ip_src.s_addr, AF_INET, 1199 r->src.neg, kif, M_GETFIB(m))) 1200 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 1201 else if (PF_MISMATCHAW(&r->dst.addr, 1202 (struct pf_addr *)&h->ip_dst.s_addr, AF_INET, 1203 r->dst.neg, NULL, M_GETFIB(m))) 1204 r = r->skip[PF_SKIP_DST_ADDR].ptr; 1205 else if (r->match_tag && !pf_match_tag(m, r, &tag, 1206 pd->pf_mtag ? pd->pf_mtag->tag : 0)) 1207 r = TAILQ_NEXT(r, entries); 1208 else 1209 break; 1210 } 1211 1212 if (r == NULL || r->action == PF_NOSCRUB) 1213 return (PF_PASS); 1214 else { 1215 r->packets[dir == PF_OUT]++; 1216 r->bytes[dir == PF_OUT] += pd->tot_len; 1217 } 1218 1219 /* Check for illegal packets */ 1220 if (hlen < (int)sizeof(struct ip)) 1221 goto drop; 1222 1223 if (hlen > ntohs(h->ip_len)) 1224 goto drop; 1225 1226 /* Clear IP_DF if the rule uses the no-df option */ 1227 if (r->rule_flag & PFRULE_NODF && h->ip_off & htons(IP_DF)) { 1228 u_int16_t ip_off = h->ip_off; 1229 1230 h->ip_off &= htons(~IP_DF); 1231 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0); 1232 } 1233 1234 /* We will need other tests here */ 1235 if (!fragoff && !mff) 1236 goto no_fragment; 1237 1238 /* We're dealing with a fragment now. Don't allow fragments 1239 * with IP_DF to enter the cache. If the flag was cleared by 1240 * no-df above, fine. Otherwise drop it. 1241 */ 1242 if (h->ip_off & htons(IP_DF)) { 1243 DPFPRINTF(("IP_DF\n")); 1244 goto bad; 1245 } 1246 1247 ip_len = ntohs(h->ip_len) - hlen; 1248 ip_off = (ntohs(h->ip_off) & IP_OFFMASK) << 3; 1249 1250 /* All fragments are 8 byte aligned */ 1251 if (mff && (ip_len & 0x7)) { 1252 DPFPRINTF(("mff and %d\n", ip_len)); 1253 goto bad; 1254 } 1255 1256 /* Respect maximum length */ 1257 if (fragoff + ip_len > IP_MAXPACKET) { 1258 DPFPRINTF(("max packet %d\n", fragoff + ip_len)); 1259 goto bad; 1260 } 1261 max = fragoff + ip_len; 1262 1263 if ((r->rule_flag & (PFRULE_FRAGCROP|PFRULE_FRAGDROP)) == 0) { 1264 1265 /* Fully buffer all of the fragments */ 1266 PF_FRAG_LOCK(); 1267 1268 pf_ip2key(h, dir, &key); 1269 frag = pf_find_fragment(&key, &V_pf_frag_tree); 1270 1271 /* Check if we saw the last fragment already */ 1272 if (frag != NULL && (frag->fr_flags & PFFRAG_SEENLAST) && 1273 max > frag->fr_max) 1274 goto bad; 1275 1276 /* Might return a completely reassembled mbuf, or NULL */ 1277 DPFPRINTF(("reass frag %d @ %d-%d\n", h->ip_id, fragoff, max)); 1278 verdict = pf_reassemble(m0, h, dir, reason); 1279 PF_FRAG_UNLOCK(); 1280 1281 if (verdict != PF_PASS) 1282 return (PF_DROP); 1283 1284 m = *m0; 1285 if (m == NULL) 1286 return (PF_DROP); 1287 1288 /* use mtag from concatenated mbuf chain */ 1289 pd->pf_mtag = pf_find_mtag(m); 1290#ifdef DIAGNOSTIC 1291 if (pd->pf_mtag == NULL) { 1292 printf("%s: pf_find_mtag returned NULL(1)\n", __func__); 1293 if ((pd->pf_mtag = pf_get_mtag(m)) == NULL) { 1294 m_freem(m); 1295 *m0 = NULL; 1296 goto no_mem; 1297 } 1298 } 1299#endif 1300 if (frag != NULL && (frag->fr_flags & PFFRAG_DROP)) 1301 goto drop; 1302 1303 h = mtod(m, struct ip *); 1304 } else { 1305 /* non-buffering fragment cache (drops or masks overlaps) */ 1306 int nomem = 0; 1307 1308 if (dir == PF_OUT && pd->pf_mtag->flags & PF_TAG_FRAGCACHE) { 1309 /* 1310 * Already passed the fragment cache in the 1311 * input direction. If we continued, it would 1312 * appear to be a dup and would be dropped. 1313 */ 1314 goto fragment_pass; 1315 } 1316 1317 PF_FRAG_LOCK(); 1318 pf_ip2key(h, dir, &key); 1319 frag = pf_find_fragment(&key, &V_pf_cache_tree); 1320 1321 /* Check if we saw the last fragment already */ 1322 if (frag != NULL && (frag->fr_flags & PFFRAG_SEENLAST) && 1323 max > frag->fr_max) { 1324 if (r->rule_flag & PFRULE_FRAGDROP) 1325 frag->fr_flags |= PFFRAG_DROP; 1326 goto bad; 1327 } 1328 1329 *m0 = m = pf_fragcache(m0, h, &frag, mff, 1330 (r->rule_flag & PFRULE_FRAGDROP) ? 1 : 0, &nomem); 1331 PF_FRAG_UNLOCK(); 1332 if (m == NULL) { 1333 if (nomem) 1334 goto no_mem; 1335 goto drop; 1336 } 1337 1338 /* use mtag from copied and trimmed mbuf chain */ 1339 pd->pf_mtag = pf_find_mtag(m); 1340#ifdef DIAGNOSTIC 1341 if (pd->pf_mtag == NULL) { 1342 printf("%s: pf_find_mtag returned NULL(2)\n", __func__); 1343 if ((pd->pf_mtag = pf_get_mtag(m)) == NULL) { 1344 m_freem(m); 1345 *m0 = NULL; 1346 goto no_mem; 1347 } 1348 } 1349#endif 1350 if (dir == PF_IN) 1351 pd->pf_mtag->flags |= PF_TAG_FRAGCACHE; 1352 1353 if (frag != NULL && (frag->fr_flags & PFFRAG_DROP)) 1354 goto drop; 1355 goto fragment_pass; 1356 } 1357 1358 no_fragment: 1359 /* At this point, only IP_DF is allowed in ip_off */ 1360 if (h->ip_off & ~htons(IP_DF)) { 1361 u_int16_t ip_off = h->ip_off; 1362 1363 h->ip_off &= htons(IP_DF); 1364 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0); 1365 } 1366 1367 /* not missing a return here */ 1368 1369 fragment_pass: 1370 pf_scrub_ip(&m, r->rule_flag, r->min_ttl, r->set_tos); 1371 1372 if ((r->rule_flag & (PFRULE_FRAGCROP|PFRULE_FRAGDROP)) == 0) 1373 pd->flags |= PFDESC_IP_REAS; 1374 return (PF_PASS); 1375 1376 no_mem: 1377 REASON_SET(reason, PFRES_MEMORY); 1378 if (r != NULL && r->log) 1379 PFLOG_PACKET(kif, m, AF_INET, dir, *reason, r, NULL, NULL, pd, 1380 1); 1381 return (PF_DROP); 1382 1383 drop: 1384 REASON_SET(reason, PFRES_NORM); 1385 if (r != NULL && r->log) 1386 PFLOG_PACKET(kif, m, AF_INET, dir, *reason, r, NULL, NULL, pd, 1387 1); 1388 return (PF_DROP); 1389 1390 bad: 1391 DPFPRINTF(("dropping bad fragment\n")); 1392 1393 /* Free associated fragments */ 1394 if (frag != NULL) { 1395 pf_free_fragment(frag); 1396 PF_FRAG_UNLOCK(); 1397 } 1398 1399 REASON_SET(reason, PFRES_FRAG); 1400 if (r != NULL && r->log) 1401 PFLOG_PACKET(kif, m, AF_INET, dir, *reason, r, NULL, NULL, pd, 1402 1); 1403 1404 return (PF_DROP); 1405} 1406#endif 1407 1408#ifdef INET6 1409int 1410pf_normalize_ip6(struct mbuf **m0, int dir, struct pfi_kif *kif, 1411 u_short *reason, struct pf_pdesc *pd) 1412{ 1413 struct mbuf *m = *m0; 1414 struct pf_rule *r; 1415 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 1416 int extoff; 1417 int off; 1418 struct ip6_ext ext; 1419 struct ip6_opt opt; 1420 struct ip6_opt_jumbo jumbo; 1421 struct ip6_frag frag; 1422 u_int32_t jumbolen = 0, plen; 1423 int optend; 1424 int ooff; 1425 u_int8_t proto; 1426 int terminal; 1427 1428 PF_RULES_RASSERT(); 1429 1430 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr); 1431 while (r != NULL) { 1432 r->evaluations++; 1433 if (pfi_kif_match(r->kif, kif) == r->ifnot) 1434 r = r->skip[PF_SKIP_IFP].ptr; 1435 else if (r->direction && r->direction != dir) 1436 r = r->skip[PF_SKIP_DIR].ptr; 1437 else if (r->af && r->af != AF_INET6) 1438 r = r->skip[PF_SKIP_AF].ptr; 1439#if 0 /* header chain! */ 1440 else if (r->proto && r->proto != h->ip6_nxt) 1441 r = r->skip[PF_SKIP_PROTO].ptr; 1442#endif 1443 else if (PF_MISMATCHAW(&r->src.addr, 1444 (struct pf_addr *)&h->ip6_src, AF_INET6, 1445 r->src.neg, kif, M_GETFIB(m))) 1446 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 1447 else if (PF_MISMATCHAW(&r->dst.addr, 1448 (struct pf_addr *)&h->ip6_dst, AF_INET6, 1449 r->dst.neg, NULL, M_GETFIB(m))) 1450 r = r->skip[PF_SKIP_DST_ADDR].ptr; 1451 else 1452 break; 1453 } 1454 1455 if (r == NULL || r->action == PF_NOSCRUB) 1456 return (PF_PASS); 1457 else { 1458 r->packets[dir == PF_OUT]++; 1459 r->bytes[dir == PF_OUT] += pd->tot_len; 1460 } 1461 1462 /* Check for illegal packets */ 1463 if (sizeof(struct ip6_hdr) + IPV6_MAXPACKET < m->m_pkthdr.len) 1464 goto drop; 1465 1466 extoff = 0; 1467 off = sizeof(struct ip6_hdr); 1468 proto = h->ip6_nxt; 1469 terminal = 0; 1470 do { 1471 switch (proto) { 1472 case IPPROTO_FRAGMENT: 1473 goto fragment; 1474 break; 1475 case IPPROTO_AH: 1476 case IPPROTO_ROUTING: 1477 case IPPROTO_DSTOPTS: 1478 if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL, 1479 NULL, AF_INET6)) 1480 goto shortpkt; 1481 extoff = off; 1482 if (proto == IPPROTO_AH) 1483 off += (ext.ip6e_len + 2) * 4; 1484 else 1485 off += (ext.ip6e_len + 1) * 8; 1486 proto = ext.ip6e_nxt; 1487 break; 1488 case IPPROTO_HOPOPTS: 1489 if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL, 1490 NULL, AF_INET6)) 1491 goto shortpkt; 1492 extoff = off; 1493 optend = off + (ext.ip6e_len + 1) * 8; 1494 ooff = off + sizeof(ext); 1495 do { 1496 if (!pf_pull_hdr(m, ooff, &opt.ip6o_type, 1497 sizeof(opt.ip6o_type), NULL, NULL, 1498 AF_INET6)) 1499 goto shortpkt; 1500 if (opt.ip6o_type == IP6OPT_PAD1) { 1501 ooff++; 1502 continue; 1503 } 1504 if (!pf_pull_hdr(m, ooff, &opt, sizeof(opt), 1505 NULL, NULL, AF_INET6)) 1506 goto shortpkt; 1507 if (ooff + sizeof(opt) + opt.ip6o_len > optend) 1508 goto drop; 1509 switch (opt.ip6o_type) { 1510 case IP6OPT_JUMBO: 1511 if (h->ip6_plen != 0) 1512 goto drop; 1513 if (!pf_pull_hdr(m, ooff, &jumbo, 1514 sizeof(jumbo), NULL, NULL, 1515 AF_INET6)) 1516 goto shortpkt; 1517 memcpy(&jumbolen, jumbo.ip6oj_jumbo_len, 1518 sizeof(jumbolen)); 1519 jumbolen = ntohl(jumbolen); 1520 if (jumbolen <= IPV6_MAXPACKET) 1521 goto drop; 1522 if (sizeof(struct ip6_hdr) + jumbolen != 1523 m->m_pkthdr.len) 1524 goto drop; 1525 break; 1526 default: 1527 break; 1528 } 1529 ooff += sizeof(opt) + opt.ip6o_len; 1530 } while (ooff < optend); 1531 1532 off = optend; 1533 proto = ext.ip6e_nxt; 1534 break; 1535 default: 1536 terminal = 1; 1537 break; 1538 } 1539 } while (!terminal); 1540 1541 /* jumbo payload option must be present, or plen > 0 */ 1542 if (ntohs(h->ip6_plen) == 0) 1543 plen = jumbolen; 1544 else 1545 plen = ntohs(h->ip6_plen); 1546 if (plen == 0) 1547 goto drop; 1548 if (sizeof(struct ip6_hdr) + plen > m->m_pkthdr.len) 1549 goto shortpkt; 1550 1551 pf_scrub_ip6(&m, r->min_ttl); 1552 1553 return (PF_PASS); 1554 1555 fragment: 1556 /* Jumbo payload packets cannot be fragmented. */ 1557 plen = ntohs(h->ip6_plen); 1558 if (plen == 0 || jumbolen) 1559 goto drop; 1560 if (sizeof(struct ip6_hdr) + plen > m->m_pkthdr.len) 1561 goto shortpkt; 1562 1563 if (!pf_pull_hdr(m, off, &frag, sizeof(frag), NULL, NULL, AF_INET6)) 1564 goto shortpkt; 1565 1566 /* Offset now points to data portion. */ 1567 off += sizeof(frag); 1568 1569 /* Returns PF_DROP or *m0 is NULL or completely reassembled mbuf. */ 1570 if (pf_reassemble6(m0, h, &frag, off, extoff, dir, reason) != PF_PASS) 1571 return (PF_DROP); 1572 m = *m0; 1573 if (m == NULL) 1574 return (PF_DROP); 1575 1576 pd->flags |= PFDESC_IP_REAS; 1577 return (PF_PASS); 1578 1579 shortpkt: 1580 REASON_SET(reason, PFRES_SHORT); 1581 if (r != NULL && r->log) 1582 PFLOG_PACKET(kif, m, AF_INET6, dir, *reason, r, NULL, NULL, pd, 1583 1); 1584 return (PF_DROP); 1585 1586 drop: 1587 REASON_SET(reason, PFRES_NORM); 1588 if (r != NULL && r->log) 1589 PFLOG_PACKET(kif, m, AF_INET6, dir, *reason, r, NULL, NULL, pd, 1590 1); 1591 return (PF_DROP); 1592} 1593#endif /* INET6 */ 1594 1595int 1596pf_normalize_tcp(int dir, struct pfi_kif *kif, struct mbuf *m, int ipoff, 1597 int off, void *h, struct pf_pdesc *pd) 1598{ 1599 struct pf_rule *r, *rm = NULL; 1600 struct tcphdr *th = pd->hdr.tcp; 1601 int rewrite = 0; 1602 u_short reason; 1603 u_int8_t flags; 1604 sa_family_t af = pd->af; 1605 1606 PF_RULES_RASSERT(); 1607 1608 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr); 1609 while (r != NULL) { 1610 r->evaluations++; 1611 if (pfi_kif_match(r->kif, kif) == r->ifnot) 1612 r = r->skip[PF_SKIP_IFP].ptr; 1613 else if (r->direction && r->direction != dir) 1614 r = r->skip[PF_SKIP_DIR].ptr; 1615 else if (r->af && r->af != af) 1616 r = r->skip[PF_SKIP_AF].ptr; 1617 else if (r->proto && r->proto != pd->proto) 1618 r = r->skip[PF_SKIP_PROTO].ptr; 1619 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af, 1620 r->src.neg, kif, M_GETFIB(m))) 1621 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 1622 else if (r->src.port_op && !pf_match_port(r->src.port_op, 1623 r->src.port[0], r->src.port[1], th->th_sport)) 1624 r = r->skip[PF_SKIP_SRC_PORT].ptr; 1625 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af, 1626 r->dst.neg, NULL, M_GETFIB(m))) 1627 r = r->skip[PF_SKIP_DST_ADDR].ptr; 1628 else if (r->dst.port_op && !pf_match_port(r->dst.port_op, 1629 r->dst.port[0], r->dst.port[1], th->th_dport)) 1630 r = r->skip[PF_SKIP_DST_PORT].ptr; 1631 else if (r->os_fingerprint != PF_OSFP_ANY && !pf_osfp_match( 1632 pf_osfp_fingerprint(pd, m, off, th), 1633 r->os_fingerprint)) 1634 r = TAILQ_NEXT(r, entries); 1635 else { 1636 rm = r; 1637 break; 1638 } 1639 } 1640 1641 if (rm == NULL || rm->action == PF_NOSCRUB) 1642 return (PF_PASS); 1643 else { 1644 r->packets[dir == PF_OUT]++; 1645 r->bytes[dir == PF_OUT] += pd->tot_len; 1646 } 1647 1648 if (rm->rule_flag & PFRULE_REASSEMBLE_TCP) 1649 pd->flags |= PFDESC_TCP_NORM; 1650 1651 flags = th->th_flags; 1652 if (flags & TH_SYN) { 1653 /* Illegal packet */ 1654 if (flags & TH_RST) 1655 goto tcp_drop; 1656 1657 if (flags & TH_FIN) 1658 goto tcp_drop; 1659 } else { 1660 /* Illegal packet */ 1661 if (!(flags & (TH_ACK|TH_RST))) 1662 goto tcp_drop; 1663 } 1664 1665 if (!(flags & TH_ACK)) { 1666 /* These flags are only valid if ACK is set */ 1667 if ((flags & TH_FIN) || (flags & TH_PUSH) || (flags & TH_URG)) 1668 goto tcp_drop; 1669 } 1670 1671 /* Check for illegal header length */ 1672 if (th->th_off < (sizeof(struct tcphdr) >> 2)) 1673 goto tcp_drop; 1674 1675 /* If flags changed, or reserved data set, then adjust */ 1676 if (flags != th->th_flags || th->th_x2 != 0) { 1677 u_int16_t ov, nv; 1678 1679 ov = *(u_int16_t *)(&th->th_ack + 1); 1680 th->th_flags = flags; 1681 th->th_x2 = 0; 1682 nv = *(u_int16_t *)(&th->th_ack + 1); 1683 1684 th->th_sum = pf_cksum_fixup(th->th_sum, ov, nv, 0); 1685 rewrite = 1; 1686 } 1687 1688 /* Remove urgent pointer, if TH_URG is not set */ 1689 if (!(flags & TH_URG) && th->th_urp) { 1690 th->th_sum = pf_cksum_fixup(th->th_sum, th->th_urp, 0, 0); 1691 th->th_urp = 0; 1692 rewrite = 1; 1693 } 1694 1695 /* Process options */ 1696 if (r->max_mss && pf_normalize_tcpopt(r, m, th, off, pd->af)) 1697 rewrite = 1; 1698 1699 /* copy back packet headers if we sanitized */ 1700 if (rewrite) 1701 m_copyback(m, off, sizeof(*th), (caddr_t)th); 1702 1703 return (PF_PASS); 1704 1705 tcp_drop: 1706 REASON_SET(&reason, PFRES_NORM); 1707 if (rm != NULL && r->log) 1708 PFLOG_PACKET(kif, m, AF_INET, dir, reason, r, NULL, NULL, pd, 1709 1); 1710 return (PF_DROP); 1711} 1712 1713int 1714pf_normalize_tcp_init(struct mbuf *m, int off, struct pf_pdesc *pd, 1715 struct tcphdr *th, struct pf_state_peer *src, struct pf_state_peer *dst) 1716{ 1717 u_int32_t tsval, tsecr; 1718 u_int8_t hdr[60]; 1719 u_int8_t *opt; 1720 1721 KASSERT((src->scrub == NULL), 1722 ("pf_normalize_tcp_init: src->scrub != NULL")); 1723 1724 src->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT); 1725 if (src->scrub == NULL) 1726 return (1); 1727 1728 switch (pd->af) { 1729#ifdef INET 1730 case AF_INET: { 1731 struct ip *h = mtod(m, struct ip *); 1732 src->scrub->pfss_ttl = h->ip_ttl; 1733 break; 1734 } 1735#endif /* INET */ 1736#ifdef INET6 1737 case AF_INET6: { 1738 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 1739 src->scrub->pfss_ttl = h->ip6_hlim; 1740 break; 1741 } 1742#endif /* INET6 */ 1743 } 1744 1745 1746 /* 1747 * All normalizations below are only begun if we see the start of 1748 * the connections. They must all set an enabled bit in pfss_flags 1749 */ 1750 if ((th->th_flags & TH_SYN) == 0) 1751 return (0); 1752 1753 1754 if (th->th_off > (sizeof(struct tcphdr) >> 2) && src->scrub && 1755 pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) { 1756 /* Diddle with TCP options */ 1757 int hlen; 1758 opt = hdr + sizeof(struct tcphdr); 1759 hlen = (th->th_off << 2) - sizeof(struct tcphdr); 1760 while (hlen >= TCPOLEN_TIMESTAMP) { 1761 switch (*opt) { 1762 case TCPOPT_EOL: /* FALLTHROUGH */ 1763 case TCPOPT_NOP: 1764 opt++; 1765 hlen--; 1766 break; 1767 case TCPOPT_TIMESTAMP: 1768 if (opt[1] >= TCPOLEN_TIMESTAMP) { 1769 src->scrub->pfss_flags |= 1770 PFSS_TIMESTAMP; 1771 src->scrub->pfss_ts_mod = 1772 htonl(arc4random()); 1773 1774 /* note PFSS_PAWS not set yet */ 1775 memcpy(&tsval, &opt[2], 1776 sizeof(u_int32_t)); 1777 memcpy(&tsecr, &opt[6], 1778 sizeof(u_int32_t)); 1779 src->scrub->pfss_tsval0 = ntohl(tsval); 1780 src->scrub->pfss_tsval = ntohl(tsval); 1781 src->scrub->pfss_tsecr = ntohl(tsecr); 1782 getmicrouptime(&src->scrub->pfss_last); 1783 } 1784 /* FALLTHROUGH */ 1785 default: 1786 hlen -= MAX(opt[1], 2); 1787 opt += MAX(opt[1], 2); 1788 break; 1789 } 1790 } 1791 } 1792 1793 return (0); 1794} 1795 1796void 1797pf_normalize_tcp_cleanup(struct pf_state *state) 1798{ 1799 if (state->src.scrub) 1800 uma_zfree(V_pf_state_scrub_z, state->src.scrub); 1801 if (state->dst.scrub) 1802 uma_zfree(V_pf_state_scrub_z, state->dst.scrub); 1803 1804 /* Someday... flush the TCP segment reassembly descriptors. */ 1805} 1806 1807int 1808pf_normalize_tcp_stateful(struct mbuf *m, int off, struct pf_pdesc *pd, 1809 u_short *reason, struct tcphdr *th, struct pf_state *state, 1810 struct pf_state_peer *src, struct pf_state_peer *dst, int *writeback) 1811{ 1812 struct timeval uptime; 1813 u_int32_t tsval, tsecr; 1814 u_int tsval_from_last; 1815 u_int8_t hdr[60]; 1816 u_int8_t *opt; 1817 int copyback = 0; 1818 int got_ts = 0; 1819 1820 KASSERT((src->scrub || dst->scrub), 1821 ("%s: src->scrub && dst->scrub!", __func__)); 1822 1823 /* 1824 * Enforce the minimum TTL seen for this connection. Negate a common 1825 * technique to evade an intrusion detection system and confuse 1826 * firewall state code. 1827 */ 1828 switch (pd->af) { 1829#ifdef INET 1830 case AF_INET: { 1831 if (src->scrub) { 1832 struct ip *h = mtod(m, struct ip *); 1833 if (h->ip_ttl > src->scrub->pfss_ttl) 1834 src->scrub->pfss_ttl = h->ip_ttl; 1835 h->ip_ttl = src->scrub->pfss_ttl; 1836 } 1837 break; 1838 } 1839#endif /* INET */ 1840#ifdef INET6 1841 case AF_INET6: { 1842 if (src->scrub) { 1843 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 1844 if (h->ip6_hlim > src->scrub->pfss_ttl) 1845 src->scrub->pfss_ttl = h->ip6_hlim; 1846 h->ip6_hlim = src->scrub->pfss_ttl; 1847 } 1848 break; 1849 } 1850#endif /* INET6 */ 1851 } 1852 1853 if (th->th_off > (sizeof(struct tcphdr) >> 2) && 1854 ((src->scrub && (src->scrub->pfss_flags & PFSS_TIMESTAMP)) || 1855 (dst->scrub && (dst->scrub->pfss_flags & PFSS_TIMESTAMP))) && 1856 pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) { 1857 /* Diddle with TCP options */ 1858 int hlen; 1859 opt = hdr + sizeof(struct tcphdr); 1860 hlen = (th->th_off << 2) - sizeof(struct tcphdr); 1861 while (hlen >= TCPOLEN_TIMESTAMP) { 1862 switch (*opt) { 1863 case TCPOPT_EOL: /* FALLTHROUGH */ 1864 case TCPOPT_NOP: 1865 opt++; 1866 hlen--; 1867 break; 1868 case TCPOPT_TIMESTAMP: 1869 /* Modulate the timestamps. Can be used for 1870 * NAT detection, OS uptime determination or 1871 * reboot detection. 1872 */ 1873 1874 if (got_ts) { 1875 /* Huh? Multiple timestamps!? */ 1876 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1877 DPFPRINTF(("multiple TS??")); 1878 pf_print_state(state); 1879 printf("\n"); 1880 } 1881 REASON_SET(reason, PFRES_TS); 1882 return (PF_DROP); 1883 } 1884 if (opt[1] >= TCPOLEN_TIMESTAMP) { 1885 memcpy(&tsval, &opt[2], 1886 sizeof(u_int32_t)); 1887 if (tsval && src->scrub && 1888 (src->scrub->pfss_flags & 1889 PFSS_TIMESTAMP)) { 1890 tsval = ntohl(tsval); 1891 pf_change_a(&opt[2], 1892 &th->th_sum, 1893 htonl(tsval + 1894 src->scrub->pfss_ts_mod), 1895 0); 1896 copyback = 1; 1897 } 1898 1899 /* Modulate TS reply iff valid (!0) */ 1900 memcpy(&tsecr, &opt[6], 1901 sizeof(u_int32_t)); 1902 if (tsecr && dst->scrub && 1903 (dst->scrub->pfss_flags & 1904 PFSS_TIMESTAMP)) { 1905 tsecr = ntohl(tsecr) 1906 - dst->scrub->pfss_ts_mod; 1907 pf_change_a(&opt[6], 1908 &th->th_sum, htonl(tsecr), 1909 0); 1910 copyback = 1; 1911 } 1912 got_ts = 1; 1913 } 1914 /* FALLTHROUGH */ 1915 default: 1916 hlen -= MAX(opt[1], 2); 1917 opt += MAX(opt[1], 2); 1918 break; 1919 } 1920 } 1921 if (copyback) { 1922 /* Copyback the options, caller copys back header */ 1923 *writeback = 1; 1924 m_copyback(m, off + sizeof(struct tcphdr), 1925 (th->th_off << 2) - sizeof(struct tcphdr), hdr + 1926 sizeof(struct tcphdr)); 1927 } 1928 } 1929 1930 1931 /* 1932 * Must invalidate PAWS checks on connections idle for too long. 1933 * The fastest allowed timestamp clock is 1ms. That turns out to 1934 * be about 24 days before it wraps. XXX Right now our lowerbound 1935 * TS echo check only works for the first 12 days of a connection 1936 * when the TS has exhausted half its 32bit space 1937 */ 1938#define TS_MAX_IDLE (24*24*60*60) 1939#define TS_MAX_CONN (12*24*60*60) /* XXX remove when better tsecr check */ 1940 1941 getmicrouptime(&uptime); 1942 if (src->scrub && (src->scrub->pfss_flags & PFSS_PAWS) && 1943 (uptime.tv_sec - src->scrub->pfss_last.tv_sec > TS_MAX_IDLE || 1944 time_uptime - state->creation > TS_MAX_CONN)) { 1945 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1946 DPFPRINTF(("src idled out of PAWS\n")); 1947 pf_print_state(state); 1948 printf("\n"); 1949 } 1950 src->scrub->pfss_flags = (src->scrub->pfss_flags & ~PFSS_PAWS) 1951 | PFSS_PAWS_IDLED; 1952 } 1953 if (dst->scrub && (dst->scrub->pfss_flags & PFSS_PAWS) && 1954 uptime.tv_sec - dst->scrub->pfss_last.tv_sec > TS_MAX_IDLE) { 1955 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1956 DPFPRINTF(("dst idled out of PAWS\n")); 1957 pf_print_state(state); 1958 printf("\n"); 1959 } 1960 dst->scrub->pfss_flags = (dst->scrub->pfss_flags & ~PFSS_PAWS) 1961 | PFSS_PAWS_IDLED; 1962 } 1963 1964 if (got_ts && src->scrub && dst->scrub && 1965 (src->scrub->pfss_flags & PFSS_PAWS) && 1966 (dst->scrub->pfss_flags & PFSS_PAWS)) { 1967 /* Validate that the timestamps are "in-window". 1968 * RFC1323 describes TCP Timestamp options that allow 1969 * measurement of RTT (round trip time) and PAWS 1970 * (protection against wrapped sequence numbers). PAWS 1971 * gives us a set of rules for rejecting packets on 1972 * long fat pipes (packets that were somehow delayed 1973 * in transit longer than the time it took to send the 1974 * full TCP sequence space of 4Gb). We can use these 1975 * rules and infer a few others that will let us treat 1976 * the 32bit timestamp and the 32bit echoed timestamp 1977 * as sequence numbers to prevent a blind attacker from 1978 * inserting packets into a connection. 1979 * 1980 * RFC1323 tells us: 1981 * - The timestamp on this packet must be greater than 1982 * or equal to the last value echoed by the other 1983 * endpoint. The RFC says those will be discarded 1984 * since it is a dup that has already been acked. 1985 * This gives us a lowerbound on the timestamp. 1986 * timestamp >= other last echoed timestamp 1987 * - The timestamp will be less than or equal to 1988 * the last timestamp plus the time between the 1989 * last packet and now. The RFC defines the max 1990 * clock rate as 1ms. We will allow clocks to be 1991 * up to 10% fast and will allow a total difference 1992 * or 30 seconds due to a route change. And this 1993 * gives us an upperbound on the timestamp. 1994 * timestamp <= last timestamp + max ticks 1995 * We have to be careful here. Windows will send an 1996 * initial timestamp of zero and then initialize it 1997 * to a random value after the 3whs; presumably to 1998 * avoid a DoS by having to call an expensive RNG 1999 * during a SYN flood. Proof MS has at least one 2000 * good security geek. 2001 * 2002 * - The TCP timestamp option must also echo the other 2003 * endpoints timestamp. The timestamp echoed is the 2004 * one carried on the earliest unacknowledged segment 2005 * on the left edge of the sequence window. The RFC 2006 * states that the host will reject any echoed 2007 * timestamps that were larger than any ever sent. 2008 * This gives us an upperbound on the TS echo. 2009 * tescr <= largest_tsval 2010 * - The lowerbound on the TS echo is a little more 2011 * tricky to determine. The other endpoint's echoed 2012 * values will not decrease. But there may be 2013 * network conditions that re-order packets and 2014 * cause our view of them to decrease. For now the 2015 * only lowerbound we can safely determine is that 2016 * the TS echo will never be less than the original 2017 * TS. XXX There is probably a better lowerbound. 2018 * Remove TS_MAX_CONN with better lowerbound check. 2019 * tescr >= other original TS 2020 * 2021 * It is also important to note that the fastest 2022 * timestamp clock of 1ms will wrap its 32bit space in 2023 * 24 days. So we just disable TS checking after 24 2024 * days of idle time. We actually must use a 12d 2025 * connection limit until we can come up with a better 2026 * lowerbound to the TS echo check. 2027 */ 2028 struct timeval delta_ts; 2029 int ts_fudge; 2030 2031 2032 /* 2033 * PFTM_TS_DIFF is how many seconds of leeway to allow 2034 * a host's timestamp. This can happen if the previous 2035 * packet got delayed in transit for much longer than 2036 * this packet. 2037 */ 2038 if ((ts_fudge = state->rule.ptr->timeout[PFTM_TS_DIFF]) == 0) 2039 ts_fudge = V_pf_default_rule.timeout[PFTM_TS_DIFF]; 2040 2041 /* Calculate max ticks since the last timestamp */ 2042#define TS_MAXFREQ 1100 /* RFC max TS freq of 1Khz + 10% skew */ 2043#define TS_MICROSECS 1000000 /* microseconds per second */ 2044 delta_ts = uptime; 2045 timevalsub(&delta_ts, &src->scrub->pfss_last); 2046 tsval_from_last = (delta_ts.tv_sec + ts_fudge) * TS_MAXFREQ; 2047 tsval_from_last += delta_ts.tv_usec / (TS_MICROSECS/TS_MAXFREQ); 2048 2049 if ((src->state >= TCPS_ESTABLISHED && 2050 dst->state >= TCPS_ESTABLISHED) && 2051 (SEQ_LT(tsval, dst->scrub->pfss_tsecr) || 2052 SEQ_GT(tsval, src->scrub->pfss_tsval + tsval_from_last) || 2053 (tsecr && (SEQ_GT(tsecr, dst->scrub->pfss_tsval) || 2054 SEQ_LT(tsecr, dst->scrub->pfss_tsval0))))) { 2055 /* Bad RFC1323 implementation or an insertion attack. 2056 * 2057 * - Solaris 2.6 and 2.7 are known to send another ACK 2058 * after the FIN,FIN|ACK,ACK closing that carries 2059 * an old timestamp. 2060 */ 2061 2062 DPFPRINTF(("Timestamp failed %c%c%c%c\n", 2063 SEQ_LT(tsval, dst->scrub->pfss_tsecr) ? '0' : ' ', 2064 SEQ_GT(tsval, src->scrub->pfss_tsval + 2065 tsval_from_last) ? '1' : ' ', 2066 SEQ_GT(tsecr, dst->scrub->pfss_tsval) ? '2' : ' ', 2067 SEQ_LT(tsecr, dst->scrub->pfss_tsval0)? '3' : ' ')); 2068 DPFPRINTF((" tsval: %u tsecr: %u +ticks: %u " 2069 "idle: %jus %lums\n", 2070 tsval, tsecr, tsval_from_last, 2071 (uintmax_t)delta_ts.tv_sec, 2072 delta_ts.tv_usec / 1000)); 2073 DPFPRINTF((" src->tsval: %u tsecr: %u\n", 2074 src->scrub->pfss_tsval, src->scrub->pfss_tsecr)); 2075 DPFPRINTF((" dst->tsval: %u tsecr: %u tsval0: %u" 2076 "\n", dst->scrub->pfss_tsval, 2077 dst->scrub->pfss_tsecr, dst->scrub->pfss_tsval0)); 2078 if (V_pf_status.debug >= PF_DEBUG_MISC) { 2079 pf_print_state(state); 2080 pf_print_flags(th->th_flags); 2081 printf("\n"); 2082 } 2083 REASON_SET(reason, PFRES_TS); 2084 return (PF_DROP); 2085 } 2086 2087 /* XXX I'd really like to require tsecr but it's optional */ 2088 2089 } else if (!got_ts && (th->th_flags & TH_RST) == 0 && 2090 ((src->state == TCPS_ESTABLISHED && dst->state == TCPS_ESTABLISHED) 2091 || pd->p_len > 0 || (th->th_flags & TH_SYN)) && 2092 src->scrub && dst->scrub && 2093 (src->scrub->pfss_flags & PFSS_PAWS) && 2094 (dst->scrub->pfss_flags & PFSS_PAWS)) { 2095 /* Didn't send a timestamp. Timestamps aren't really useful 2096 * when: 2097 * - connection opening or closing (often not even sent). 2098 * but we must not let an attacker to put a FIN on a 2099 * data packet to sneak it through our ESTABLISHED check. 2100 * - on a TCP reset. RFC suggests not even looking at TS. 2101 * - on an empty ACK. The TS will not be echoed so it will 2102 * probably not help keep the RTT calculation in sync and 2103 * there isn't as much danger when the sequence numbers 2104 * got wrapped. So some stacks don't include TS on empty 2105 * ACKs :-( 2106 * 2107 * To minimize the disruption to mostly RFC1323 conformant 2108 * stacks, we will only require timestamps on data packets. 2109 * 2110 * And what do ya know, we cannot require timestamps on data 2111 * packets. There appear to be devices that do legitimate 2112 * TCP connection hijacking. There are HTTP devices that allow 2113 * a 3whs (with timestamps) and then buffer the HTTP request. 2114 * If the intermediate device has the HTTP response cache, it 2115 * will spoof the response but not bother timestamping its 2116 * packets. So we can look for the presence of a timestamp in 2117 * the first data packet and if there, require it in all future 2118 * packets. 2119 */ 2120 2121 if (pd->p_len > 0 && (src->scrub->pfss_flags & PFSS_DATA_TS)) { 2122 /* 2123 * Hey! Someone tried to sneak a packet in. Or the 2124 * stack changed its RFC1323 behavior?!?! 2125 */ 2126 if (V_pf_status.debug >= PF_DEBUG_MISC) { 2127 DPFPRINTF(("Did not receive expected RFC1323 " 2128 "timestamp\n")); 2129 pf_print_state(state); 2130 pf_print_flags(th->th_flags); 2131 printf("\n"); 2132 } 2133 REASON_SET(reason, PFRES_TS); 2134 return (PF_DROP); 2135 } 2136 } 2137 2138 2139 /* 2140 * We will note if a host sends his data packets with or without 2141 * timestamps. And require all data packets to contain a timestamp 2142 * if the first does. PAWS implicitly requires that all data packets be 2143 * timestamped. But I think there are middle-man devices that hijack 2144 * TCP streams immediately after the 3whs and don't timestamp their 2145 * packets (seen in a WWW accelerator or cache). 2146 */ 2147 if (pd->p_len > 0 && src->scrub && (src->scrub->pfss_flags & 2148 (PFSS_TIMESTAMP|PFSS_DATA_TS|PFSS_DATA_NOTS)) == PFSS_TIMESTAMP) { 2149 if (got_ts) 2150 src->scrub->pfss_flags |= PFSS_DATA_TS; 2151 else { 2152 src->scrub->pfss_flags |= PFSS_DATA_NOTS; 2153 if (V_pf_status.debug >= PF_DEBUG_MISC && dst->scrub && 2154 (dst->scrub->pfss_flags & PFSS_TIMESTAMP)) { 2155 /* Don't warn if other host rejected RFC1323 */ 2156 DPFPRINTF(("Broken RFC1323 stack did not " 2157 "timestamp data packet. Disabled PAWS " 2158 "security.\n")); 2159 pf_print_state(state); 2160 pf_print_flags(th->th_flags); 2161 printf("\n"); 2162 } 2163 } 2164 } 2165 2166 2167 /* 2168 * Update PAWS values 2169 */ 2170 if (got_ts && src->scrub && PFSS_TIMESTAMP == (src->scrub->pfss_flags & 2171 (PFSS_PAWS_IDLED|PFSS_TIMESTAMP))) { 2172 getmicrouptime(&src->scrub->pfss_last); 2173 if (SEQ_GEQ(tsval, src->scrub->pfss_tsval) || 2174 (src->scrub->pfss_flags & PFSS_PAWS) == 0) 2175 src->scrub->pfss_tsval = tsval; 2176 2177 if (tsecr) { 2178 if (SEQ_GEQ(tsecr, src->scrub->pfss_tsecr) || 2179 (src->scrub->pfss_flags & PFSS_PAWS) == 0) 2180 src->scrub->pfss_tsecr = tsecr; 2181 2182 if ((src->scrub->pfss_flags & PFSS_PAWS) == 0 && 2183 (SEQ_LT(tsval, src->scrub->pfss_tsval0) || 2184 src->scrub->pfss_tsval0 == 0)) { 2185 /* tsval0 MUST be the lowest timestamp */ 2186 src->scrub->pfss_tsval0 = tsval; 2187 } 2188 2189 /* Only fully initialized after a TS gets echoed */ 2190 if ((src->scrub->pfss_flags & PFSS_PAWS) == 0) 2191 src->scrub->pfss_flags |= PFSS_PAWS; 2192 } 2193 } 2194 2195 /* I have a dream.... TCP segment reassembly.... */ 2196 return (0); 2197} 2198 2199static int 2200pf_normalize_tcpopt(struct pf_rule *r, struct mbuf *m, struct tcphdr *th, 2201 int off, sa_family_t af) 2202{ 2203 u_int16_t *mss; 2204 int thoff; 2205 int opt, cnt, optlen = 0; 2206 int rewrite = 0; 2207 u_char opts[TCP_MAXOLEN]; 2208 u_char *optp = opts; 2209 2210 thoff = th->th_off << 2; 2211 cnt = thoff - sizeof(struct tcphdr); 2212 2213 if (cnt > 0 && !pf_pull_hdr(m, off + sizeof(*th), opts, cnt, 2214 NULL, NULL, af)) 2215 return (rewrite); 2216 2217 for (; cnt > 0; cnt -= optlen, optp += optlen) { 2218 opt = optp[0]; 2219 if (opt == TCPOPT_EOL) 2220 break; 2221 if (opt == TCPOPT_NOP) 2222 optlen = 1; 2223 else { 2224 if (cnt < 2) 2225 break; 2226 optlen = optp[1]; 2227 if (optlen < 2 || optlen > cnt) 2228 break; 2229 } 2230 switch (opt) { 2231 case TCPOPT_MAXSEG: 2232 mss = (u_int16_t *)(optp + 2); 2233 if ((ntohs(*mss)) > r->max_mss) { 2234 th->th_sum = pf_cksum_fixup(th->th_sum, 2235 *mss, htons(r->max_mss), 0); 2236 *mss = htons(r->max_mss); 2237 rewrite = 1; 2238 } 2239 break; 2240 default: 2241 break; 2242 } 2243 } 2244 2245 if (rewrite) 2246 m_copyback(m, off + sizeof(*th), thoff - sizeof(*th), opts); 2247 2248 return (rewrite); 2249} 2250 2251#ifdef INET 2252static void 2253pf_scrub_ip(struct mbuf **m0, u_int32_t flags, u_int8_t min_ttl, u_int8_t tos) 2254{ 2255 struct mbuf *m = *m0; 2256 struct ip *h = mtod(m, struct ip *); 2257 2258 /* Clear IP_DF if no-df was requested */ 2259 if (flags & PFRULE_NODF && h->ip_off & htons(IP_DF)) { 2260 u_int16_t ip_off = h->ip_off; 2261 2262 h->ip_off &= htons(~IP_DF); 2263 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0); 2264 } 2265 2266 /* Enforce a minimum ttl, may cause endless packet loops */ 2267 if (min_ttl && h->ip_ttl < min_ttl) { 2268 u_int16_t ip_ttl = h->ip_ttl; 2269 2270 h->ip_ttl = min_ttl; 2271 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_ttl, h->ip_ttl, 0); 2272 } 2273 2274 /* Enforce tos */ 2275 if (flags & PFRULE_SET_TOS) { 2276 u_int16_t ov, nv; 2277 2278 ov = *(u_int16_t *)h; 2279 h->ip_tos = tos; 2280 nv = *(u_int16_t *)h; 2281 2282 h->ip_sum = pf_cksum_fixup(h->ip_sum, ov, nv, 0); 2283 } 2284 2285 /* random-id, but not for fragments */ 2286 if (flags & PFRULE_RANDOMID && !(h->ip_off & ~htons(IP_DF))) { 2287 u_int16_t ip_id = h->ip_id; 2288 2289 h->ip_id = ip_randomid(); 2290 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_id, h->ip_id, 0); 2291 } 2292} 2293#endif /* INET */ 2294 2295#ifdef INET6 2296static void 2297pf_scrub_ip6(struct mbuf **m0, u_int8_t min_ttl) 2298{ 2299 struct mbuf *m = *m0; 2300 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 2301 2302 /* Enforce a minimum ttl, may cause endless packet loops */ 2303 if (min_ttl && h->ip6_hlim < min_ttl) 2304 h->ip6_hlim = min_ttl; 2305} 2306#endif 2307