sys/net/pf/pf_norm.c

   1 /*      $OpenBSD: pf_norm.c,v 1.113 2008/05/07 07:07:29 markus Exp $ */
   2
   3 /*
   4  * Copyright (c) 2010 The DragonFly Project.  All rights reserved.
   5  *
   6  * Copyright 2001 Niels Provos <provos@citi.umich.edu>
   7  * All rights reserved.
   8  *
   9  * Redistribution and use in source and binary forms, with or without
  10  * modification, are permitted provided that the following conditions
  11  * are met:
  12  * 1. Redistributions of source code must retain the above copyright
  13  *    notice, this list of conditions and the following disclaimer.
  14  * 2. Redistributions in binary form must reproduce the above copyright
  15  *    notice, this list of conditions and the following disclaimer in the
  16  *    documentation and/or other materials provided with the distribution.
  17  *
  18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
  19  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
  20  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
  21  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
  22  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
  23  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  24  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  25  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  26  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
  27  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  28  */
  29
  30 #include "opt_inet.h"
  31 #include "opt_inet6.h"
  32
  33 #include <sys/param.h>
  34 #include <sys/systm.h>
  35 #include <sys/mbuf.h>
  36 #include <sys/filio.h>
  37 #include <sys/fcntl.h>
  38 #include <sys/socket.h>
  39 #include <sys/kernel.h>
  40 #include <sys/time.h>
  41
  42 #include <net/if.h>
  43 #include <net/if_types.h>
  44 #include <net/bpf.h>
  45 #include <net/route.h>
  46 #include <net/pf/if_pflog.h>
  47
  48 #include <netinet/in.h>
  49 #include <netinet/in_var.h>
  50 #include <netinet/in_systm.h>
  51 #include <netinet/ip.h>
  52 #include <netinet/ip_var.h>
  53 #include <netinet/tcp.h>
  54 #include <netinet/tcp_seq.h>
  55 #include <netinet/udp.h>
  56 #include <netinet/ip_icmp.h>
  57
  58 #ifdef INET6
  59 #include <netinet/ip6.h>
  60 #endif /* INET6 */
  61
  62 #include <net/pf/pfvar.h>
  63
  64 #define PFFRAG_SEENLAST 0x0001          /* Seen the last fragment for this */
  65 #define PFFRAG_NOBUFFER 0x0002          /* Non-buffering fragment cache */
  66 #define PFFRAG_DROP     0x0004          /* Drop all fragments */
  67 #define BUFFER_FRAGMENTS(fr)    (!((fr)->fr_flags & PFFRAG_NOBUFFER))
  68
  69
  70 TAILQ_HEAD(pf_fragqueue, pf_fragment)   pf_fragqueue;
  71 TAILQ_HEAD(pf_cachequeue, pf_fragment)  pf_cachequeue;
  72
  73 static __inline int      pf_frag_compare(struct pf_fragment *,
  74                             struct pf_fragment *);
  75 RB_HEAD(pf_frag_tree, pf_fragment)      pf_frag_tree, pf_cache_tree;
  76 RB_PROTOTYPE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
  77 RB_GENERATE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
  78
  79 /* Private prototypes */
  80 void                     pf_ip2key(struct pf_fragment *, struct ip *);
  81 void                     pf_remove_fragment(struct pf_fragment *);
  82 void                     pf_flush_fragments(void);
  83 void                     pf_free_fragment(struct pf_fragment *);
  84 struct pf_fragment      *pf_find_fragment(struct ip *, struct pf_frag_tree *);
  85 struct mbuf             *pf_reassemble(struct mbuf **, struct pf_fragment **,
  86                             struct pf_frent *, int);
  87 struct mbuf             *pf_fragcache(struct mbuf **, struct ip*,
  88                             struct pf_fragment **, int, int, int *);
  89 int                      pf_normalize_tcpopt(struct pf_rule *, struct mbuf *,
  90                             struct tcphdr *, int, sa_family_t);
  91
  92 #define DPFPRINTF(x) do {                               \
  93         if (pf_status.debug >= PF_DEBUG_MISC) {         \
  94                 kprintf("%s: ", __func__);              \
  95                 kprintf x ;                             \
  96         }                                               \
  97 } while(0)
  98
  99 static MALLOC_DEFINE(M_PFFRAGPL, "pffrag", "pf fragment pool list");
 100 static MALLOC_DEFINE(M_PFCACHEPL, "pffrcache", "pf fragment cache pool list");
 101 static MALLOC_DEFINE(M_PFFRENTPL, "pffrent", "pf frent pool list");
 102 static MALLOC_DEFINE(M_PFCENTPL, "pffrcent", "pf fragment cent pool list");
 103 static MALLOC_DEFINE(M_PFSTATESCRUBPL, "pfstatescrub", "pf state scrub pool list");
 104
 105 /* Globals */
 106 struct malloc_type       *pf_frent_pl, *pf_frag_pl, *pf_cache_pl, *pf_cent_pl;
 107 struct malloc_type       *pf_state_scrub_pl;
 108 int                      pf_nfrents, pf_ncache;
 109
 110 void
 111 pf_normalize_init(void)
 112 {
 113         /* XXX
 114         pool_sethiwat(&pf_frag_pl, PFFRAG_FRAG_HIWAT);
 115         pool_sethardlimit(&pf_frent_pl, PFFRAG_FRENT_HIWAT, NULL, 0);
 116         pool_sethardlimit(&pf_cache_pl, PFFRAG_FRCACHE_HIWAT, NULL, 0);
 117         pool_sethardlimit(&pf_cent_pl, PFFRAG_FRCENT_HIWAT, NULL, 0);
 118         */
 119
 120         TAILQ_INIT(&pf_fragqueue);
 121         TAILQ_INIT(&pf_cachequeue);
 122 }
 123
 124 static __inline int
 125 pf_frag_compare(struct pf_fragment *a, struct pf_fragment *b)
 126 {
 127         int     diff;
 128
 129         if ((diff = a->fr_id - b->fr_id))
 130                 return (diff);
 131         else if ((diff = a->fr_p - b->fr_p))
 132                 return (diff);
 133         else if (a->fr_src.s_addr < b->fr_src.s_addr)
 134                 return (-1);
 135         else if (a->fr_src.s_addr > b->fr_src.s_addr)
 136                 return (1);
 137         else if (a->fr_dst.s_addr < b->fr_dst.s_addr)
 138                 return (-1);
 139         else if (a->fr_dst.s_addr > b->fr_dst.s_addr)
 140                 return (1);
 141         return (0);
 142 }
 143
 144 void
 145 pf_purge_expired_fragments(void)
 146 {
 147         struct pf_fragment      *frag;
 148         u_int32_t                expire = time_second -
 149                                     pf_default_rule.timeout[PFTM_FRAG];
 150
 151         while ((frag = TAILQ_LAST(&pf_fragqueue, pf_fragqueue)) != NULL) {
 152                 KASSERT((BUFFER_FRAGMENTS(frag)),
 153                         ("BUFFER_FRAGMENTS(frag) == 0: %s", __func__));
 154                 if (frag->fr_timeout > expire)
 155                         break;
 156
 157                 DPFPRINTF(("expiring %d(%p)\n", frag->fr_id, frag));
 158                 pf_free_fragment(frag);
 159         }
 160
 161         while ((frag = TAILQ_LAST(&pf_cachequeue, pf_cachequeue)) != NULL) {
 162                 KASSERT((!BUFFER_FRAGMENTS(frag)),
 163                         ("BUFFER_FRAGMENTS(frag) != 0: %s", __func__));
 164                 if (frag->fr_timeout > expire)
 165                         break;
 166
 167                 DPFPRINTF(("expiring %d(%p)\n", frag->fr_id, frag));
 168                 pf_free_fragment(frag);
 169                 KASSERT((TAILQ_EMPTY(&pf_cachequeue) ||
 170                     TAILQ_LAST(&pf_cachequeue, pf_cachequeue) != frag),
 171                     ("!(TAILQ_EMPTY() || TAILQ_LAST() == farg): %s",
 172                     __func__));
 173         }
 174 }
 175
 176 /*
 177  * Try to flush old fragments to make space for new ones
 178  */
 179
 180 void
 181 pf_flush_fragments(void)
 182 {
 183         struct pf_fragment      *frag;
 184         int                      goal;
 185
 186         goal = pf_nfrents * 9 / 10;
 187         DPFPRINTF(("trying to free > %d frents\n",
 188             pf_nfrents - goal));
 189         while (goal < pf_nfrents) {
 190                 frag = TAILQ_LAST(&pf_fragqueue, pf_fragqueue);
 191                 if (frag == NULL)
 192                         break;
 193                 pf_free_fragment(frag);
 194         }
 195
 196
 197         goal = pf_ncache * 9 / 10;
 198         DPFPRINTF(("trying to free > %d cache entries\n",
 199             pf_ncache - goal));
 200         while (goal < pf_ncache) {
 201                 frag = TAILQ_LAST(&pf_cachequeue, pf_cachequeue);
 202                 if (frag == NULL)
 203                         break;
 204                 pf_free_fragment(frag);
 205         }
 206 }
 207
 208 /* Frees the fragments and all associated entries */
 209
 210 void
 211 pf_free_fragment(struct pf_fragment *frag)
 212 {
 213         struct pf_frent         *frent;
 214         struct pf_frcache       *frcache;
 215
 216         /* Free all fragments */
 217         if (BUFFER_FRAGMENTS(frag)) {
 218                 for (frent = LIST_FIRST(&frag->fr_queue); frent;
 219                     frent = LIST_FIRST(&frag->fr_queue)) {
 220                         LIST_REMOVE(frent, fr_next);
 221
 222                         m_freem(frent->fr_m);
 223                         kfree(frent, M_PFFRENTPL);
 224                         pf_nfrents--;
 225                 }
 226         } else {
 227                 for (frcache = LIST_FIRST(&frag->fr_cache); frcache;
 228                     frcache = LIST_FIRST(&frag->fr_cache)) {
 229                         LIST_REMOVE(frcache, fr_next);
 230
 231                         KASSERT((LIST_EMPTY(&frag->fr_cache) ||
 232                             LIST_FIRST(&frag->fr_cache)->fr_off >
 233                             frcache->fr_end),
 234                             ("! (LIST_EMPTY() || LIST_FIRST()->fr_off >"
 235                              " frcache->fr_end): %s", __func__));
 236
 237                         kfree(frcache, M_PFCENTPL);
 238                         pf_ncache--;
 239                 }
 240         }
 241
 242         pf_remove_fragment(frag);
 243 }
 244
 245 void
 246 pf_ip2key(struct pf_fragment *key, struct ip *ip)
 247 {
 248         key->fr_p = ip->ip_p;
 249         key->fr_id = ip->ip_id;
 250         key->fr_src.s_addr = ip->ip_src.s_addr;
 251         key->fr_dst.s_addr = ip->ip_dst.s_addr;
 252 }
 253
 254 struct pf_fragment *
 255 pf_find_fragment(struct ip *ip, struct pf_frag_tree *tree)
 256 {
 257         struct pf_fragment       key;
 258         struct pf_fragment      *frag;
 259
 260         pf_ip2key(&key, ip);
 261
 262         frag = RB_FIND(pf_frag_tree, tree, &key);
 263         if (frag != NULL) {
 264                 /* XXX Are we sure we want to update the timeout? */
 265                 frag->fr_timeout = time_second;
 266                 if (BUFFER_FRAGMENTS(frag)) {
 267                         TAILQ_REMOVE(&pf_fragqueue, frag, frag_next);
 268                         TAILQ_INSERT_HEAD(&pf_fragqueue, frag, frag_next);
 269                 } else {
 270                         TAILQ_REMOVE(&pf_cachequeue, frag, frag_next);
 271                         TAILQ_INSERT_HEAD(&pf_cachequeue, frag, frag_next);
 272                 }
 273         }
 274
 275         return (frag);
 276 }
 277
 278 /* Removes a fragment from the fragment queue and frees the fragment */
 279
 280 void
 281 pf_remove_fragment(struct pf_fragment *frag)
 282 {
 283         if (BUFFER_FRAGMENTS(frag)) {
 284                 RB_REMOVE(pf_frag_tree, &pf_frag_tree, frag);
 285                 TAILQ_REMOVE(&pf_fragqueue, frag, frag_next);
 286                 kfree(frag, M_PFFRAGPL);
 287         } else {
 288                 RB_REMOVE(pf_frag_tree, &pf_cache_tree, frag);
 289                 TAILQ_REMOVE(&pf_cachequeue, frag, frag_next);
 290                 kfree(frag, M_PFCACHEPL);
 291         }
 292 }
 293
 294 #define FR_IP_OFF(fr)   (((fr)->fr_ip->ip_off & IP_OFFMASK) << 3)
 295 struct mbuf *
 296 pf_reassemble(struct mbuf **m0, struct pf_fragment **frag,
 297     struct pf_frent *frent, int mff)
 298 {
 299         struct mbuf     *m = *m0, *m2;
 300         struct pf_frent *frea, *next;
 301         struct pf_frent *frep = NULL;
 302         struct ip       *ip = frent->fr_ip;
 303         int              hlen = ip->ip_hl << 2;
 304         u_int16_t        off = (ip->ip_off & IP_OFFMASK) << 3;
 305         u_int16_t        ip_len = ip->ip_len - ip->ip_hl * 4;
 306         u_int16_t        max = ip_len + off;
 307
 308         KASSERT((*frag == NULL || BUFFER_FRAGMENTS(*frag)),
 309             ("! (*frag == NULL || BUFFER_FRAGMENTS(*frag)): %s", __func__));
 310
 311         /* Strip off ip header */
 312         m->m_data += hlen;
 313         m->m_len -= hlen;
 314
 315         /* Create a new reassembly queue for this packet */
 316         if (*frag == NULL) {
 317                 *frag = kmalloc(sizeof(struct pf_fragment), M_PFFRAGPL, M_NOWAIT);
 318                 if (*frag == NULL) {
 319                         pf_flush_fragments();
 320                         *frag = kmalloc(sizeof(struct pf_fragment), M_PFFRAGPL, M_NOWAIT);
 321                         if (*frag == NULL)
 322                                 goto drop_fragment;
 323                 }
 324
 325                 (*frag)->fr_flags = 0;
 326                 (*frag)->fr_max = 0;
 327                 (*frag)->fr_src = frent->fr_ip->ip_src;
 328                 (*frag)->fr_dst = frent->fr_ip->ip_dst;
 329                 (*frag)->fr_p = frent->fr_ip->ip_p;
 330                 (*frag)->fr_id = frent->fr_ip->ip_id;
 331                 (*frag)->fr_timeout = time_second;
 332                 LIST_INIT(&(*frag)->fr_queue);
 333
 334                 RB_INSERT(pf_frag_tree, &pf_frag_tree, *frag);
 335                 TAILQ_INSERT_HEAD(&pf_fragqueue, *frag, frag_next);
 336
 337                 /* We do not have a previous fragment */
 338                 frep = NULL;
 339                 goto insert;
 340         }
 341
 342         /*
 343          * Find a fragment after the current one:
 344          *  - off contains the real shifted offset.
 345          */
 346         LIST_FOREACH(frea, &(*frag)->fr_queue, fr_next) {
 347                 if (FR_IP_OFF(frea) > off)
 348                         break;
 349                 frep = frea;
 350         }
 351
 352         KASSERT((frep != NULL || frea != NULL),
 353             ("!(frep != NULL || frea != NULL): %s", __func__));
 354
 355         if (frep != NULL &&
 356             FR_IP_OFF(frep) + frep->fr_ip->ip_len - frep->fr_ip->ip_hl *
 357             4 > off)
 358         {
 359                 u_int16_t       precut;
 360
 361                 precut = FR_IP_OFF(frep) + frep->fr_ip->ip_len -
 362                     frep->fr_ip->ip_hl * 4 - off;
 363                 if (precut >= ip_len)
 364                         goto drop_fragment;
 365                 m_adj(frent->fr_m, precut);
 366                 DPFPRINTF(("overlap -%d\n", precut));
 367                 /* Enforce 8 byte boundaries */
 368                 ip->ip_off = ip->ip_off + (precut >> 3);
 369                 off = (ip->ip_off & IP_OFFMASK) << 3;
 370                 ip_len -= precut;
 371                 ip->ip_len = ip_len;
 372         }
 373
 374         for (; frea != NULL && ip_len + off > FR_IP_OFF(frea);
 375             frea = next)
 376         {
 377                 u_int16_t       aftercut;
 378
 379                 aftercut = ip_len + off - FR_IP_OFF(frea);
 380                 DPFPRINTF(("adjust overlap %d\n", aftercut));
 381                 if (aftercut < frea->fr_ip->ip_len - frea->fr_ip->ip_hl
 382                     * 4)
 383                 {
 384                         frea->fr_ip->ip_len =
 385                             frea->fr_ip->ip_len - aftercut;
 386                         frea->fr_ip->ip_off = frea->fr_ip->ip_off +
 387                             (aftercut >> 3);
 388                         m_adj(frea->fr_m, aftercut);
 389                         break;
 390                 }
 391
 392                 /* This fragment is completely overlapped, lose it */
 393                 next = LIST_NEXT(frea, fr_next);
 394                 m_freem(frea->fr_m);
 395                 LIST_REMOVE(frea, fr_next);
 396                 kfree(frea, M_PFFRENTPL);
 397                 pf_nfrents--;
 398         }
 399
 400  insert:
 401         /* Update maximum data size */
 402         if ((*frag)->fr_max < max)
 403                 (*frag)->fr_max = max;
 404         /* This is the last segment */
 405         if (!mff)
 406                 (*frag)->fr_flags |= PFFRAG_SEENLAST;
 407
 408         if (frep == NULL)
 409                 LIST_INSERT_HEAD(&(*frag)->fr_queue, frent, fr_next);
 410         else
 411                 LIST_INSERT_AFTER(frep, frent, fr_next);
 412
 413         /* Check if we are completely reassembled */
 414         if (!((*frag)->fr_flags & PFFRAG_SEENLAST))
 415                 return (NULL);
 416
 417         /* Check if we have all the data */
 418         off = 0;
 419         for (frep = LIST_FIRST(&(*frag)->fr_queue); frep; frep = next) {
 420                 next = LIST_NEXT(frep, fr_next);
 421
 422                 off += frep->fr_ip->ip_len - frep->fr_ip->ip_hl * 4;
 423                 if (off < (*frag)->fr_max &&
 424                     (next == NULL || FR_IP_OFF(next) != off))
 425                 {
 426                         DPFPRINTF(("missing fragment at %d, next %d, max %d\n",
 427                             off, next == NULL ? -1 : FR_IP_OFF(next),
 428                             (*frag)->fr_max));
 429                         return (NULL);
 430                 }
 431         }
 432         DPFPRINTF(("%d < %d?\n", off, (*frag)->fr_max));
 433         if (off < (*frag)->fr_max)
 434                 return (NULL);
 435
 436         /* We have all the data */
 437         frent = LIST_FIRST(&(*frag)->fr_queue);
 438         KASSERT((frent != NULL), ("frent == NULL: %s", __func__));
 439         if ((frent->fr_ip->ip_hl << 2) + off > IP_MAXPACKET) {
 440                 DPFPRINTF(("drop: too big: %d\n", off));
 441                 pf_free_fragment(*frag);
 442                 *frag = NULL;
 443                 return (NULL);
 444         }
 445         next = LIST_NEXT(frent, fr_next);
 446
 447         /* Magic from ip_input */
 448         ip = frent->fr_ip;
 449         m = frent->fr_m;
 450         m2 = m->m_next;
 451         m->m_next = NULL;
 452         m_cat(m, m2);
 453         kfree(frent, M_PFFRENTPL);
 454         pf_nfrents--;
 455         for (frent = next; frent != NULL; frent = next) {
 456                 next = LIST_NEXT(frent, fr_next);
 457
 458                 m2 = frent->fr_m;
 459                 kfree(frent, M_PFFRENTPL);
 460                 pf_nfrents--;
 461                 m_cat(m, m2);
 462         }
 463
 464         ip->ip_src = (*frag)->fr_src;
 465         ip->ip_dst = (*frag)->fr_dst;
 466
 467         /* Remove from fragment queue */
 468         pf_remove_fragment(*frag);
 469         *frag = NULL;
 470
 471         hlen = ip->ip_hl << 2;
 472         ip->ip_len = off + hlen;
 473         m->m_len += hlen;
 474         m->m_data -= hlen;
 475
 476         /* some debugging cruft by sklower, below, will go away soon */
 477         /* XXX this should be done elsewhere */
 478         if (m->m_flags & M_PKTHDR) {
 479                 int plen = 0;
 480                 for (m2 = m; m2; m2 = m2->m_next)
 481                         plen += m2->m_len;
 482                 m->m_pkthdr.len = plen;
 483         }
 484
 485         DPFPRINTF(("complete: %p(%d)\n", m, ip->ip_len));
 486         return (m);
 487
 488  drop_fragment:
 489         /* Oops - fail safe - drop packet */
 490         kfree(frent, M_PFFRENTPL);
 491         pf_nfrents--;
 492         m_freem(m);
 493         return (NULL);
 494 }
 495
 496 struct mbuf *
 497 pf_fragcache(struct mbuf **m0, struct ip *h, struct pf_fragment **frag, int mff,
 498     int drop, int *nomem)
 499 {
 500         struct mbuf             *m = *m0;
 501         struct pf_frcache       *frp, *fra, *cur = NULL;
 502         int                      ip_len = h->ip_len - (h->ip_hl << 2);
 503         u_int16_t                off = h->ip_off << 3;
 504         u_int16_t                max = ip_len + off;
 505         int                      hosed = 0;
 506
 507         KASSERT((*frag == NULL || !BUFFER_FRAGMENTS(*frag)),
 508             ("!(*frag == NULL || !BUFFER_FRAGMENTS(*frag)): %s", __func__));
 509
 510         /* Create a new range queue for this packet */
 511         if (*frag == NULL) {
 512                 *frag = kmalloc(sizeof(struct pf_fragment), M_PFCACHEPL, M_NOWAIT);
 513                 if (*frag == NULL) {
 514                         pf_flush_fragments();
 515                         *frag = kmalloc(sizeof(struct pf_fragment), M_PFCACHEPL, M_NOWAIT);
 516                         if (*frag == NULL)
 517                                 goto no_mem;
 518                 }
 519
 520                 /* Get an entry for the queue */
 521                 cur = kmalloc(sizeof(struct pf_frcache), M_PFCENTPL, M_NOWAIT);
 522                 if (cur == NULL) {
 523                         kfree(*frag, M_PFCACHEPL);
 524                         *frag = NULL;
 525                         goto no_mem;
 526                 }
 527                 pf_ncache++;
 528
 529                 (*frag)->fr_flags = PFFRAG_NOBUFFER;
 530                 (*frag)->fr_max = 0;
 531                 (*frag)->fr_src = h->ip_src;
 532                 (*frag)->fr_dst = h->ip_dst;
 533                 (*frag)->fr_p = h->ip_p;
 534                 (*frag)->fr_id = h->ip_id;
 535                 (*frag)->fr_timeout = time_second;
 536
 537                 cur->fr_off = off;
 538                 cur->fr_end = max;
 539                 LIST_INIT(&(*frag)->fr_cache);
 540                 LIST_INSERT_HEAD(&(*frag)->fr_cache, cur, fr_next);
 541
 542                 RB_INSERT(pf_frag_tree, &pf_cache_tree, *frag);
 543                 TAILQ_INSERT_HEAD(&pf_cachequeue, *frag, frag_next);
 544
 545                 DPFPRINTF(("fragcache[%d]: new %d-%d\n", h->ip_id, off, max));
 546
 547                 goto pass;
 548         }
 549
 550         /*
 551          * Find a fragment after the current one:
 552          *  - off contains the real shifted offset.
 553          */
 554         frp = NULL;
 555         LIST_FOREACH(fra, &(*frag)->fr_cache, fr_next) {
 556                 if (fra->fr_off > off)
 557                         break;
 558                 frp = fra;
 559         }
 560
 561         KASSERT((frp != NULL || fra != NULL),
 562             ("!(frp != NULL || fra != NULL): %s", __func__));
 563
 564         if (frp != NULL) {
 565                 int     precut;
 566
 567                 precut = frp->fr_end - off;
 568                 if (precut >= ip_len) {
 569                         /* Fragment is entirely a duplicate */
 570                         DPFPRINTF(("fragcache[%d]: dead (%d-%d) %d-%d\n",
 571                             h->ip_id, frp->fr_off, frp->fr_end, off, max));
 572                         goto drop_fragment;
 573                 }
 574                 if (precut == 0) {
 575                         /* They are adjacent.  Fixup cache entry */
 576                         DPFPRINTF(("fragcache[%d]: adjacent (%d-%d) %d-%d\n",
 577                             h->ip_id, frp->fr_off, frp->fr_end, off, max));
 578                         frp->fr_end = max;
 579                 } else if (precut > 0) {
 580                         /* The first part of this payload overlaps with a
 581                          * fragment that has already been passed.
 582                          * Need to trim off the first part of the payload.
 583                          * But to do so easily, we need to create another
 584                          * mbuf to throw the original header into.
 585                          */
 586
 587                         DPFPRINTF(("fragcache[%d]: chop %d (%d-%d) %d-%d\n",
 588                             h->ip_id, precut, frp->fr_off, frp->fr_end, off,
 589                             max));
 590
 591                         off += precut;
 592                         max -= precut;
 593                         /* Update the previous frag to encompass this one */
 594                         frp->fr_end = max;
 595
 596                         if (!drop) {
 597                                 /* XXX Optimization opportunity
 598                                  * This is a very heavy way to trim the payload.
 599                                  * we could do it much faster by diddling mbuf
 600                                  * internals but that would be even less legible
 601                                  * than this mbuf magic.  For my next trick,
 602                                  * I'll pull a rabbit out of my laptop.
 603                                  */
 604                                 *m0 = m_dup(m, MB_DONTWAIT);
 605                                 /* From KAME Project : We have missed this! */
 606                                 m_adj(*m0, (h->ip_hl << 2) -
 607                                     (*m0)->m_pkthdr.len);
 608                                 if (*m0 == NULL)
 609                                         goto no_mem;
 610                                 KASSERT(((*m0)->m_next == NULL),
 611                                     ("(*m0)->m_next != NULL: %s",
 612                                     __func__));
 613                                 m_adj(m, precut + (h->ip_hl << 2));
 614                                 m_cat(*m0, m);
 615                                 m = *m0;
 616                                 if (m->m_flags & M_PKTHDR) {
 617                                         int plen = 0;
 618                                         struct mbuf *t;
 619                                         for (t = m; t; t = t->m_next)
 620                                                 plen += t->m_len;
 621                                         m->m_pkthdr.len = plen;
 622                                 }
 623
 624
 625                                 h = mtod(m, struct ip *);
 626
 627                                 KASSERT(((int)m->m_len ==
 628                                     h->ip_len - precut),
 629                                     ("m->m_len != h->ip_len - precut: %s",
 630                                     __func__));
 631                                 h->ip_off = h->ip_off +
 632                                     (precut >> 3);
 633                                 h->ip_len = h->ip_len - precut;
 634                         } else {
 635                                 hosed++;
 636                         }
 637                 } else {
 638                         /* There is a gap between fragments */
 639
 640                         DPFPRINTF(("fragcache[%d]: gap %d (%d-%d) %d-%d\n",
 641                             h->ip_id, -precut, frp->fr_off, frp->fr_end, off,
 642                             max));
 643
 644                         cur = kmalloc(sizeof(struct pf_frcache), M_PFCENTPL, M_NOWAIT);
 645                         if (cur == NULL)
 646                                 goto no_mem;
 647                         pf_ncache++;
 648
 649                         cur->fr_off = off;
 650                         cur->fr_end = max;
 651                         LIST_INSERT_AFTER(frp, cur, fr_next);
 652                 }
 653         }
 654
 655         if (fra != NULL) {
 656                 int     aftercut;
 657                 int     merge = 0;
 658
 659                 aftercut = max - fra->fr_off;
 660                 if (aftercut == 0) {
 661                         /* Adjacent fragments */
 662                         DPFPRINTF(("fragcache[%d]: adjacent %d-%d (%d-%d)\n",
 663                             h->ip_id, off, max, fra->fr_off, fra->fr_end));
 664                         fra->fr_off = off;
 665                         merge = 1;
 666                 } else if (aftercut > 0) {
 667                         /* Need to chop off the tail of this fragment */
 668                         DPFPRINTF(("fragcache[%d]: chop %d %d-%d (%d-%d)\n",
 669                             h->ip_id, aftercut, off, max, fra->fr_off,
 670                             fra->fr_end));
 671                         fra->fr_off = off;
 672                         max -= aftercut;
 673
 674                         merge = 1;
 675
 676                         if (!drop) {
 677                                 m_adj(m, -aftercut);
 678                                 if (m->m_flags & M_PKTHDR) {
 679                                         int plen = 0;
 680                                         struct mbuf *t;
 681                                         for (t = m; t; t = t->m_next)
 682                                                 plen += t->m_len;
 683                                         m->m_pkthdr.len = plen;
 684                                 }
 685                                 h = mtod(m, struct ip *);
 686                                 KASSERT(((int)m->m_len == h->ip_len - aftercut),
 687                                     ("m->m_len != h->ip_len - aftercut: %s",
 688                                     __func__));
 689                                 h->ip_len = h->ip_len - aftercut;
 690                         } else {
 691                                 hosed++;
 692                         }
 693                 } else if (frp == NULL) {
 694                         /* There is a gap between fragments */
 695                         DPFPRINTF(("fragcache[%d]: gap %d %d-%d (%d-%d)\n",
 696                             h->ip_id, -aftercut, off, max, fra->fr_off,
 697                             fra->fr_end));
 698
 699                         cur = kmalloc(sizeof(struct pf_frcache), M_PFCENTPL, M_NOWAIT);
 700                         if (cur == NULL)
 701                                 goto no_mem;
 702                         pf_ncache++;
 703
 704                         cur->fr_off = off;
 705                         cur->fr_end = max;
 706                         LIST_INSERT_BEFORE(fra, cur, fr_next);
 707                 }
 708
 709
 710                 /* Need to glue together two separate fragment descriptors */
 711                 if (merge) {
 712                         if (cur && fra->fr_off <= cur->fr_end) {
 713                                 /* Need to merge in a previous 'cur' */
 714                                 DPFPRINTF(("fragcache[%d]: adjacent(merge "
 715                                     "%d-%d) %d-%d (%d-%d)\n",
 716                                     h->ip_id, cur->fr_off, cur->fr_end, off,
 717                                     max, fra->fr_off, fra->fr_end));
 718                                 fra->fr_off = cur->fr_off;
 719                                 LIST_REMOVE(cur, fr_next);
 720                                 kfree(cur, M_PFCENTPL);
 721                                 pf_ncache--;
 722                                 cur = NULL;
 723
 724                         } else if (frp && fra->fr_off <= frp->fr_end) {
 725                                 /* Need to merge in a modified 'frp' */
 726                                 KASSERT((cur == NULL), ("cur != NULL: %s",
 727                                     __func__));
 728                                 DPFPRINTF(("fragcache[%d]: adjacent(merge "
 729                                     "%d-%d) %d-%d (%d-%d)\n",
 730                                     h->ip_id, frp->fr_off, frp->fr_end, off,
 731                                     max, fra->fr_off, fra->fr_end));
 732                                 fra->fr_off = frp->fr_off;
 733                                 LIST_REMOVE(frp, fr_next);
 734                                 kfree(frp, M_PFCENTPL);
 735                                 pf_ncache--;
 736                                 frp = NULL;
 737
 738                         }
 739                 }
 740         }
 741
 742         if (hosed) {
 743                 /*
 744                  * We must keep tracking the overall fragment even when
 745                  * we're going to drop it anyway so that we know when to
 746                  * free the overall descriptor.  Thus we drop the frag late.
 747                  */
 748                 goto drop_fragment;
 749         }
 750
 751
 752  pass:
 753         /* Update maximum data size */
 754         if ((*frag)->fr_max < max)
 755                 (*frag)->fr_max = max;
 756
 757         /* This is the last segment */
 758         if (!mff)
 759                 (*frag)->fr_flags |= PFFRAG_SEENLAST;
 760
 761         /* Check if we are completely reassembled */
 762         if (((*frag)->fr_flags & PFFRAG_SEENLAST) &&
 763             LIST_FIRST(&(*frag)->fr_cache)->fr_off == 0 &&
 764             LIST_FIRST(&(*frag)->fr_cache)->fr_end == (*frag)->fr_max) {
 765                 /* Remove from fragment queue */
 766                 DPFPRINTF(("fragcache[%d]: done 0-%d\n", h->ip_id,
 767                     (*frag)->fr_max));
 768                 pf_free_fragment(*frag);
 769                 *frag = NULL;
 770         }
 771
 772         return (m);
 773
 774  no_mem:
 775         *nomem = 1;
 776
 777         /* Still need to pay attention to !IP_MF */
 778         if (!mff && *frag != NULL)
 779                 (*frag)->fr_flags |= PFFRAG_SEENLAST;
 780
 781         m_freem(m);
 782         return (NULL);
 783
 784  drop_fragment:
 785
 786         /* Still need to pay attention to !IP_MF */
 787         if (!mff && *frag != NULL)
 788                 (*frag)->fr_flags |= PFFRAG_SEENLAST;
 789
 790         if (drop) {
 791                 /* This fragment has been deemed bad.  Don't reass */
 792                 if (((*frag)->fr_flags & PFFRAG_DROP) == 0)
 793                         DPFPRINTF(("fragcache[%d]: dropping overall fragment\n",
 794                             h->ip_id));
 795                 (*frag)->fr_flags |= PFFRAG_DROP;
 796         }
 797
 798         m_freem(m);
 799         return (NULL);
 800 }
 801
 802 int
 803 pf_normalize_ip(struct mbuf **m0, int dir, struct pfi_kif *kif, u_short *reason,
 804     struct pf_pdesc *pd)
 805 {
 806         struct mbuf             *m = *m0;
 807         struct pf_rule          *r;
 808         struct pf_frent         *frent;
 809         struct pf_fragment      *frag = NULL;
 810         struct ip               *h = mtod(m, struct ip *);
 811         int                      mff = (h->ip_off & IP_MF);
 812         int                      hlen = h->ip_hl << 2;
 813         u_int16_t                fragoff = (h->ip_off & IP_OFFMASK) << 3;
 814         u_int16_t                max;
 815         int                      ip_len;
 816         int                      tag = -1;
 817
 818         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
 819         while (r != NULL) {
 820                 r->evaluations++;
 821                 if (pfi_kif_match(r->kif, kif) == r->ifnot)
 822                         r = r->skip[PF_SKIP_IFP].ptr;
 823                 else if (r->direction && r->direction != dir)
 824                         r = r->skip[PF_SKIP_DIR].ptr;
 825                 else if (r->af && r->af != AF_INET)
 826                         r = r->skip[PF_SKIP_AF].ptr;
 827                 else if (r->proto && r->proto != h->ip_p)
 828                         r = r->skip[PF_SKIP_PROTO].ptr;
 829                 else if (PF_MISMATCHAW(&r->src.addr,
 830                     (struct pf_addr *)&h->ip_src.s_addr, AF_INET,
 831                     r->src.neg, kif))
 832                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
 833                 else if (PF_MISMATCHAW(&r->dst.addr,
 834                     (struct pf_addr *)&h->ip_dst.s_addr, AF_INET,
 835                     r->dst.neg, NULL))
 836                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
 837                 else if (r->match_tag && !pf_match_tag(m, r, &tag))
 838                         r = TAILQ_NEXT(r, entries);
 839                 else
 840                         break;
 841         }
 842
 843         if (r == NULL || r->action == PF_NOSCRUB)
 844                 return (PF_PASS);
 845         else {
 846                 r->packets[dir == PF_OUT]++;
 847                 r->bytes[dir == PF_OUT] += pd->tot_len;
 848         }
 849
 850         /* Check for illegal packets */
 851         if (hlen < (int)sizeof(struct ip))
 852                 goto drop;
 853
 854         if (hlen > h->ip_len)
 855                 goto drop;
 856
 857         /* Clear IP_DF if the rule uses the no-df option */
 858         if (r->rule_flag & PFRULE_NODF && h->ip_off & IP_DF) {
 859                 u_int16_t ip_off = h->ip_off;
 860
 861                 h->ip_off &= ~IP_DF;
 862                 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
 863         }
 864
 865         /* We will need other tests here */
 866         if (!fragoff && !mff)
 867                 goto no_fragment;
 868
 869         /* We're dealing with a fragment now. Don't allow fragments
 870          * with IP_DF to enter the cache. If the flag was cleared by
 871          * no-df above, fine. Otherwise drop it.
 872          */
 873         if (h->ip_off & IP_DF) {
 874                 DPFPRINTF(("IP_DF\n"));
 875                 goto bad;
 876         }
 877
 878         ip_len = h->ip_len - hlen;
 879
 880         /* All fragments are 8 byte aligned */
 881         if (mff && (ip_len & 0x7)) {
 882                 DPFPRINTF(("mff and %d\n", ip_len));
 883                 goto bad;
 884         }
 885
 886         /* Respect maximum length */
 887         if (fragoff + ip_len > IP_MAXPACKET) {
 888                 DPFPRINTF(("max packet %d\n", fragoff + ip_len));
 889                 goto bad;
 890         }
 891         max = fragoff + ip_len;
 892
 893         if ((r->rule_flag & (PFRULE_FRAGCROP|PFRULE_FRAGDROP)) == 0) {
 894                 /* Fully buffer all of the fragments */
 895
 896                 frag = pf_find_fragment(h, &pf_frag_tree);
 897
 898                 /* Check if we saw the last fragment already */
 899                 if (frag != NULL && (frag->fr_flags & PFFRAG_SEENLAST) &&
 900                     max > frag->fr_max)
 901                         goto bad;
 902
 903                 /* Get an entry for the fragment queue */
 904                 frent = kmalloc(sizeof(struct pf_frent), M_PFFRENTPL, M_NOWAIT);
 905                 if (frent == NULL) {
 906                         REASON_SET(reason, PFRES_MEMORY);
 907                         return (PF_DROP);
 908                 }
 909                 pf_nfrents++;
 910                 frent->fr_ip = h;
 911                 frent->fr_m = m;
 912
 913                 /* Might return a completely reassembled mbuf, or NULL */
 914                 DPFPRINTF(("reass frag %d @ %d-%d\n", h->ip_id, fragoff, max));
 915                 *m0 = m = pf_reassemble(m0, &frag, frent, mff);
 916
 917                 if (m == NULL)
 918                         return (PF_DROP);
 919
 920                 if (frag != NULL && (frag->fr_flags & PFFRAG_DROP))
 921                         goto drop;
 922
 923                 h = mtod(m, struct ip *);
 924         } else {
 925                 /* non-buffering fragment cache (drops or masks overlaps) */
 926                 int     nomem = 0;
 927
 928                 if (dir == PF_OUT && m->m_pkthdr.pf.flags & PF_TAG_FRAGCACHE) {
 929                         /*
 930                          * Already passed the fragment cache in the
 931                          * input direction.  If we continued, it would
 932                          * appear to be a dup and would be dropped.
 933                          */
 934                         goto fragment_pass;
 935                 }
 936
 937                 frag = pf_find_fragment(h, &pf_cache_tree);
 938
 939                 /* Check if we saw the last fragment already */
 940                 if (frag != NULL && (frag->fr_flags & PFFRAG_SEENLAST) &&
 941                     max > frag->fr_max) {
 942                         if (r->rule_flag & PFRULE_FRAGDROP)
 943                                 frag->fr_flags |= PFFRAG_DROP;
 944                         goto bad;
 945                 }
 946
 947                 *m0 = m = pf_fragcache(m0, h, &frag, mff,
 948                     (r->rule_flag & PFRULE_FRAGDROP) ? 1 : 0, &nomem);
 949                 if (m == NULL) {
 950                         if (nomem)
 951                                 goto no_mem;
 952                         goto drop;
 953                 }
 954
 955                 if (dir == PF_IN)
 956                         m->m_pkthdr.pf.flags |= PF_TAG_FRAGCACHE;
 957
 958                 if (frag != NULL && (frag->fr_flags & PFFRAG_DROP))
 959                         goto drop;
 960                 goto fragment_pass;
 961         }
 962
 963  no_fragment:
 964         /* At this point, only IP_DF is allowed in ip_off */
 965         if (h->ip_off & ~IP_DF) {
 966                 u_int16_t ip_off = h->ip_off;
 967
 968                 h->ip_off &= IP_DF;
 969                 h->ip_sum = pf_cksum_fixup(h->ip_sum, htons(ip_off), htons(h->ip_off), 0);
 970         }
 971
 972         /* Enforce a minimum ttl, may cause endless packet loops */
 973         if (r->min_ttl && h->ip_ttl < r->min_ttl) {
 974                 u_int16_t ip_ttl = h->ip_ttl;
 975
 976                 h->ip_ttl = r->min_ttl;
 977                 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_ttl, h->ip_ttl, 0);
 978         }
 979
 980         /* Enforce tos */
 981         if (r->rule_flag & PFRULE_SET_TOS) {
 982                 u_int16_t       ov, nv;
 983
 984                 ov = *(u_int16_t *)h;
 985                 h->ip_tos = r->set_tos;
 986                 nv = *(u_int16_t *)h;
 987
 988                 h->ip_sum = pf_cksum_fixup(h->ip_sum, ov, nv, 0);
 989         }
 990
 991         if (r->rule_flag & PFRULE_RANDOMID) {
 992                 u_int16_t ip_id = h->ip_id;
 993
 994                 h->ip_id = ip_randomid();
 995                 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_id, h->ip_id, 0);
 996         }
 997         if ((r->rule_flag & (PFRULE_FRAGCROP|PFRULE_FRAGDROP)) == 0)
 998                 pd->flags |= PFDESC_IP_REAS;
 999
1000         return (PF_PASS);
1001
1002  fragment_pass:
1003         /* Enforce a minimum ttl, may cause endless packet loops */
1004         if (r->min_ttl && h->ip_ttl < r->min_ttl) {
1005                 u_int16_t ip_ttl = h->ip_ttl;
1006
1007                 h->ip_ttl = r->min_ttl;
1008                 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_ttl, h->ip_ttl, 0);
1009         }
1010         /* Enforce tos */
1011         if (r->rule_flag & PFRULE_SET_TOS) {
1012                 u_int16_t       ov, nv;
1013
1014                 ov = *(u_int16_t *)h;
1015                 h->ip_tos = r->set_tos;
1016                 nv = *(u_int16_t *)h;
1017
1018                 h->ip_sum = pf_cksum_fixup(h->ip_sum, ov, nv, 0);
1019         }
1020         if ((r->rule_flag & (PFRULE_FRAGCROP|PFRULE_FRAGDROP)) == 0)
1021                 pd->flags |= PFDESC_IP_REAS;
1022         return (PF_PASS);
1023
1024  no_mem:
1025         REASON_SET(reason, PFRES_MEMORY);
1026         if (r != NULL && r->log)
1027                 PFLOG_PACKET(kif, h, m, AF_INET, dir, *reason, r, NULL, NULL, pd);
1028         return (PF_DROP);
1029
1030  drop:
1031         REASON_SET(reason, PFRES_NORM);
1032         if (r != NULL && r->log)
1033                 PFLOG_PACKET(kif, h, m, AF_INET, dir, *reason, r, NULL, NULL, pd);
1034         return (PF_DROP);
1035
1036  bad:
1037         DPFPRINTF(("dropping bad fragment\n"));
1038
1039         /* Free associated fragments */
1040         if (frag != NULL)
1041                 pf_free_fragment(frag);
1042
1043         REASON_SET(reason, PFRES_FRAG);
1044         if (r != NULL && r->log)
1045                 PFLOG_PACKET(kif, h, m, AF_INET, dir, *reason, r, NULL, NULL, pd);
1046
1047         return (PF_DROP);
1048 }
1049
1050 #ifdef INET6
1051 int
1052 pf_normalize_ip6(struct mbuf **m0, int dir, struct pfi_kif *kif,
1053     u_short *reason, struct pf_pdesc *pd)
1054 {
1055         struct mbuf             *m = *m0;
1056         struct pf_rule          *r;
1057         struct ip6_hdr          *h = mtod(m, struct ip6_hdr *);
1058         int                      off;
1059         struct ip6_ext           ext;
1060         struct ip6_opt           opt;
1061         struct ip6_opt_jumbo     jumbo;
1062         struct ip6_frag          frag;
1063         u_int32_t                jumbolen = 0, plen;
1064         u_int16_t                fragoff = 0;
1065         int                      optend;
1066         int                      ooff;
1067         u_int8_t                 proto;
1068         int                      terminal;
1069
1070         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1071         while (r != NULL) {
1072                 r->evaluations++;
1073                 if (pfi_kif_match(r->kif, kif) == r->ifnot)
1074                         r = r->skip[PF_SKIP_IFP].ptr;
1075                 else if (r->direction && r->direction != dir)
1076                         r = r->skip[PF_SKIP_DIR].ptr;
1077                 else if (r->af && r->af != AF_INET6)
1078                         r = r->skip[PF_SKIP_AF].ptr;
1079 #if 0 /* header chain! */
1080                 else if (r->proto && r->proto != h->ip6_nxt)
1081                         r = r->skip[PF_SKIP_PROTO].ptr;
1082 #endif
1083                 else if (PF_MISMATCHAW(&r->src.addr,
1084                     (struct pf_addr *)&h->ip6_src, AF_INET6,
1085                     r->src.neg, kif))
1086                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
1087                 else if (PF_MISMATCHAW(&r->dst.addr,
1088                     (struct pf_addr *)&h->ip6_dst, AF_INET6,
1089                     r->dst.neg, NULL))
1090                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
1091                 else
1092                         break;
1093         }
1094
1095         if (r == NULL || r->action == PF_NOSCRUB)
1096                 return (PF_PASS);
1097         else {
1098                 r->packets[dir == PF_OUT]++;
1099                 r->bytes[dir == PF_OUT] += pd->tot_len;
1100         }
1101
1102         /* Check for illegal packets */
1103         if (sizeof(struct ip6_hdr) + IPV6_MAXPACKET < m->m_pkthdr.len)
1104                 goto drop;
1105
1106         off = sizeof(struct ip6_hdr);
1107         proto = h->ip6_nxt;
1108         terminal = 0;
1109         do {
1110                 switch (proto) {
1111                 case IPPROTO_FRAGMENT:
1112                         goto fragment;
1113                         break;
1114                 case IPPROTO_AH:
1115                 case IPPROTO_ROUTING:
1116                 case IPPROTO_DSTOPTS:
1117                         if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL,
1118                             NULL, AF_INET6))
1119                                 goto shortpkt;
1120                         if (proto == IPPROTO_AH)
1121                                 off += (ext.ip6e_len + 2) * 4;
1122                         else
1123                                 off += (ext.ip6e_len + 1) * 8;
1124                         proto = ext.ip6e_nxt;
1125                         break;
1126                 case IPPROTO_HOPOPTS:
1127                         if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL,
1128                             NULL, AF_INET6))
1129                                 goto shortpkt;
1130                         optend = off + (ext.ip6e_len + 1) * 8;
1131                         ooff = off + sizeof(ext);
1132                         do {
1133                                 if (!pf_pull_hdr(m, ooff, &opt.ip6o_type,
1134                                     sizeof(opt.ip6o_type), NULL, NULL,
1135                                     AF_INET6))
1136                                         goto shortpkt;
1137                                 if (opt.ip6o_type == IP6OPT_PAD1) {
1138                                         ooff++;
1139                                         continue;
1140                                 }
1141                                 if (!pf_pull_hdr(m, ooff, &opt, sizeof(opt),
1142                                     NULL, NULL, AF_INET6))
1143                                         goto shortpkt;
1144                                 if (ooff + sizeof(opt) + opt.ip6o_len > optend)
1145                                         goto drop;
1146                                 switch (opt.ip6o_type) {
1147                                 case IP6OPT_JUMBO:
1148                                         if (h->ip6_plen != 0)
1149                                                 goto drop;
1150                                         if (!pf_pull_hdr(m, ooff, &jumbo,
1151                                             sizeof(jumbo), NULL, NULL,
1152                                             AF_INET6))
1153                                                 goto shortpkt;
1154                                         memcpy(&jumbolen, jumbo.ip6oj_jumbo_len,
1155                                             sizeof(jumbolen));
1156                                         jumbolen = ntohl(jumbolen);
1157                                         if (jumbolen <= IPV6_MAXPACKET)
1158                                                 goto drop;
1159                                         if (sizeof(struct ip6_hdr) + jumbolen !=
1160                                             m->m_pkthdr.len)
1161                                                 goto drop;
1162                                         break;
1163                                 default:
1164                                         break;
1165                                 }
1166                                 ooff += sizeof(opt) + opt.ip6o_len;
1167                         } while (ooff < optend);
1168
1169                         off = optend;
1170                         proto = ext.ip6e_nxt;
1171                         break;
1172                 default:
1173                         terminal = 1;
1174                         break;
1175                 }
1176         } while (!terminal);
1177
1178         /* jumbo payload option must be present, or plen > 0 */
1179         if (ntohs(h->ip6_plen) == 0)
1180                 plen = jumbolen;
1181         else
1182                 plen = ntohs(h->ip6_plen);
1183         if (plen == 0)
1184                 goto drop;
1185         if (sizeof(struct ip6_hdr) + plen > m->m_pkthdr.len)
1186                 goto shortpkt;
1187
1188         /* Enforce a minimum ttl, may cause endless packet loops */
1189         if (r->min_ttl && h->ip6_hlim < r->min_ttl)
1190                 h->ip6_hlim = r->min_ttl;
1191
1192         return (PF_PASS);
1193
1194  fragment:
1195         if (ntohs(h->ip6_plen) == 0 || jumbolen)
1196                 goto drop;
1197         plen = ntohs(h->ip6_plen);
1198
1199         if (!pf_pull_hdr(m, off, &frag, sizeof(frag), NULL, NULL, AF_INET6))
1200                 goto shortpkt;
1201         fragoff = ntohs(frag.ip6f_offlg & IP6F_OFF_MASK);
1202         if (fragoff + (plen - off - sizeof(frag)) > IPV6_MAXPACKET)
1203                 goto badfrag;
1204
1205         /* do something about it */
1206         /* remember to set pd->flags |= PFDESC_IP_REAS */
1207         return (PF_PASS);
1208
1209  shortpkt:
1210         REASON_SET(reason, PFRES_SHORT);
1211         if (r != NULL && r->log)
1212                 PFLOG_PACKET(kif, h, m, AF_INET6, dir, *reason, r, NULL, NULL, pd);
1213         return (PF_DROP);
1214
1215  drop:
1216         REASON_SET(reason, PFRES_NORM);
1217         if (r != NULL && r->log)
1218                 PFLOG_PACKET(kif, h, m, AF_INET6, dir, *reason, r, NULL, NULL, pd);
1219         return (PF_DROP);
1220
1221  badfrag:
1222         REASON_SET(reason, PFRES_FRAG);
1223         if (r != NULL && r->log)
1224                 PFLOG_PACKET(kif, h, m, AF_INET6, dir, *reason, r, NULL, NULL, pd);
1225         return (PF_DROP);
1226 }
1227 #endif /* INET6 */
1228
1229 int
1230 pf_normalize_tcp(int dir, struct pfi_kif *kif, struct mbuf *m, int ipoff,
1231     int off, void *h, struct pf_pdesc *pd)
1232 {
1233         struct pf_rule  *r, *rm = NULL;
1234         struct tcphdr   *th = pd->hdr.tcp;
1235         int              rewrite = 0;
1236         u_short          reason;
1237         u_int8_t         flags;
1238         sa_family_t      af = pd->af;
1239
1240         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1241         while (r != NULL) {
1242                 r->evaluations++;
1243                 if (pfi_kif_match(r->kif, kif) == r->ifnot)
1244                         r = r->skip[PF_SKIP_IFP].ptr;
1245                 else if (r->direction && r->direction != dir)
1246                         r = r->skip[PF_SKIP_DIR].ptr;
1247                 else if (r->af && r->af != af)
1248                         r = r->skip[PF_SKIP_AF].ptr;
1249                 else if (r->proto && r->proto != pd->proto)
1250                         r = r->skip[PF_SKIP_PROTO].ptr;
1251                 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
1252                     r->src.neg, kif))
1253                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
1254                 else if (r->src.port_op && !pf_match_port(r->src.port_op,
1255                             r->src.port[0], r->src.port[1], th->th_sport))
1256                         r = r->skip[PF_SKIP_SRC_PORT].ptr;
1257                 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
1258                     r->dst.neg, NULL))
1259                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
1260                 else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
1261                             r->dst.port[0], r->dst.port[1], th->th_dport))
1262                         r = r->skip[PF_SKIP_DST_PORT].ptr;
1263                 else if (r->os_fingerprint != PF_OSFP_ANY && !pf_osfp_match(
1264                             pf_osfp_fingerprint(pd, m, off, th),
1265                             r->os_fingerprint))
1266                         r = TAILQ_NEXT(r, entries);
1267                 else {
1268                         rm = r;
1269                         break;
1270                 }
1271         }
1272
1273         if (rm == NULL || rm->action == PF_NOSCRUB)
1274                 return (PF_PASS);
1275         else {
1276                 r->packets[dir == PF_OUT]++;
1277                 r->bytes[dir == PF_OUT] += pd->tot_len;
1278         }
1279
1280         if (rm->rule_flag & PFRULE_REASSEMBLE_TCP)
1281                 pd->flags |= PFDESC_TCP_NORM;
1282
1283         flags = th->th_flags;
1284         if (flags & TH_SYN) {
1285                 /* Illegal packet */
1286                 if (flags & TH_RST)
1287                         goto tcp_drop;
1288
1289                 if (flags & TH_FIN)
1290                         flags &= ~TH_FIN;
1291         } else {
1292                 /* Illegal packet */
1293                 if (!(flags & (TH_ACK|TH_RST)))
1294                         goto tcp_drop;
1295         }
1296
1297         if (!(flags & TH_ACK)) {
1298                 /* These flags are only valid if ACK is set */
1299                 if ((flags & TH_FIN) || (flags & TH_PUSH) || (flags & TH_URG))
1300                         goto tcp_drop;
1301         }
1302
1303         /* Check for illegal header length */
1304         if (th->th_off < (sizeof(struct tcphdr) >> 2))
1305                 goto tcp_drop;
1306
1307         /* If flags changed, or reserved data set, then adjust */
1308         if (flags != th->th_flags || th->th_x2 != 0) {
1309                 u_int16_t       ov, nv;
1310
1311                 ov = *(u_int16_t *)(&th->th_ack + 1);
1312                 th->th_flags = flags;
1313                 th->th_x2 = 0;
1314                 nv = *(u_int16_t *)(&th->th_ack + 1);
1315
1316                 th->th_sum = pf_cksum_fixup(th->th_sum, ov, nv, 0);
1317                 rewrite = 1;
1318         }
1319
1320         /* Remove urgent pointer, if TH_URG is not set */
1321         if (!(flags & TH_URG) && th->th_urp) {
1322                 th->th_sum = pf_cksum_fixup(th->th_sum, th->th_urp, 0, 0);
1323                 th->th_urp = 0;
1324                 rewrite = 1;
1325         }
1326
1327         /* Process options */
1328         if (r->max_mss && pf_normalize_tcpopt(r, m, th, off, pd->af))
1329                 rewrite = 1;
1330
1331         /* copy back packet headers if we sanitized */
1332         if (rewrite)
1333                 m_copyback(m, off, sizeof(*th), (caddr_t)th);
1334
1335         return (PF_PASS);
1336
1337  tcp_drop:
1338         REASON_SET(&reason, PFRES_NORM);
1339         if (rm != NULL && r->log)
1340                 PFLOG_PACKET(kif, h, m, AF_INET, dir, reason, r, NULL, NULL, pd);
1341         return (PF_DROP);
1342 }
1343
1344 int
1345 pf_normalize_tcp_init(struct mbuf *m, int off, struct pf_pdesc *pd,
1346     struct tcphdr *th, struct pf_state_peer *src, struct pf_state_peer *dst)
1347 {
1348         u_int32_t tsval, tsecr;
1349         u_int8_t hdr[60];
1350         u_int8_t *opt;
1351
1352         KASSERT((src->scrub == NULL),
1353             ("pf_normalize_tcp_init: src->scrub != NULL"));
1354
1355         src->scrub = kmalloc(sizeof(struct pf_state_scrub), M_PFSTATESCRUBPL,
1356             M_NOWAIT | M_ZERO);
1357         if (src->scrub == NULL)
1358                 return (1);
1359
1360         switch (pd->af) {
1361 #ifdef INET
1362         case AF_INET: {
1363                 struct ip *h = mtod(m, struct ip *);
1364                 src->scrub->pfss_ttl = h->ip_ttl;
1365                 break;
1366         }
1367 #endif /* INET */
1368 #ifdef INET6
1369         case AF_INET6: {
1370                 struct ip6_hdr *h = mtod(m, struct ip6_hdr *);
1371                 src->scrub->pfss_ttl = h->ip6_hlim;
1372                 break;
1373         }
1374 #endif /* INET6 */
1375         }
1376
1377
1378         /*
1379          * All normalizations below are only begun if we see the start of
1380          * the connections.  They must all set an enabled bit in pfss_flags
1381          */
1382         if ((th->th_flags & TH_SYN) == 0)
1383                 return (0);
1384
1385
1386         if (th->th_off > (sizeof(struct tcphdr) >> 2) && src->scrub &&
1387             pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) {
1388                 /* Diddle with TCP options */
1389                 int hlen;
1390                 opt = hdr + sizeof(struct tcphdr);
1391                 hlen = (th->th_off << 2) - sizeof(struct tcphdr);
1392                 while (hlen >= TCPOLEN_TIMESTAMP) {
1393                         switch (*opt) {
1394                         case TCPOPT_EOL:        /* FALLTHROUGH */
1395                         case TCPOPT_NOP:
1396                                 opt++;
1397                                 hlen--;
1398                                 break;
1399                         case TCPOPT_TIMESTAMP:
1400                                 if (opt[1] >= TCPOLEN_TIMESTAMP) {
1401                                         src->scrub->pfss_flags |=
1402                                             PFSS_TIMESTAMP;
1403                                         src->scrub->pfss_ts_mod = karc4random();
1404
1405                                         /* note PFSS_PAWS not set yet */
1406                                         memcpy(&tsval, &opt[2],
1407                                             sizeof(u_int32_t));
1408                                         memcpy(&tsecr, &opt[6],
1409                                             sizeof(u_int32_t));
1410                                         src->scrub->pfss_tsval0 = ntohl(tsval);
1411                                         src->scrub->pfss_tsval = ntohl(tsval);
1412                                         src->scrub->pfss_tsecr = ntohl(tsecr);
1413                                         getmicrouptime(&src->scrub->pfss_last);
1414                                 }
1415                                 /* FALLTHROUGH */
1416                         default:
1417                                 hlen -= MAX(opt[1], 2);
1418                                 opt += MAX(opt[1], 2);
1419                                 break;
1420                         }
1421                 }
1422         }
1423
1424         return (0);
1425 }
1426
1427 void
1428 pf_normalize_tcp_cleanup(struct pf_state *state)
1429 {
1430         if (state->src.scrub)
1431                 kfree(state->src.scrub, M_PFSTATESCRUBPL);
1432         if (state->dst.scrub)
1433                 kfree(state->dst.scrub, M_PFSTATESCRUBPL);
1434
1435         /* Someday... flush the TCP segment reassembly descriptors. */
1436 }
1437
1438 int
1439 pf_normalize_tcp_stateful(struct mbuf *m, int off, struct pf_pdesc *pd,
1440     u_short *reason, struct tcphdr *th, struct pf_state *state,
1441     struct pf_state_peer *src, struct pf_state_peer *dst, int *writeback)
1442 {
1443         struct timeval uptime;
1444         u_int32_t tsval, tsecr;
1445         u_int tsval_from_last;
1446         u_int8_t hdr[60];
1447         u_int8_t *opt;
1448         int copyback = 0;
1449         int got_ts = 0;
1450
1451         KASSERT((src->scrub || dst->scrub),
1452             ("pf_normalize_tcp_statefull: src->scrub && dst->scrub!"));
1453
1454         /*
1455          * Enforce the minimum TTL seen for this connection.  Negate a common
1456          * technique to evade an intrusion detection system and confuse
1457          * firewall state code.
1458          */
1459         switch (pd->af) {
1460 #ifdef INET
1461         case AF_INET: {
1462                 if (src->scrub) {
1463                         struct ip *h = mtod(m, struct ip *);
1464                         if (h->ip_ttl > src->scrub->pfss_ttl)
1465                                 src->scrub->pfss_ttl = h->ip_ttl;
1466                         h->ip_ttl = src->scrub->pfss_ttl;
1467                 }
1468                 break;
1469         }
1470 #endif /* INET */
1471 #ifdef INET6
1472         case AF_INET6: {
1473                 if (src->scrub) {
1474                         struct ip6_hdr *h = mtod(m, struct ip6_hdr *);
1475                         if (h->ip6_hlim > src->scrub->pfss_ttl)
1476                                 src->scrub->pfss_ttl = h->ip6_hlim;
1477                         h->ip6_hlim = src->scrub->pfss_ttl;
1478                 }
1479                 break;
1480         }
1481 #endif /* INET6 */
1482         }
1483
1484         if (th->th_off > (sizeof(struct tcphdr) >> 2) &&
1485             ((src->scrub && (src->scrub->pfss_flags & PFSS_TIMESTAMP)) ||
1486             (dst->scrub && (dst->scrub->pfss_flags & PFSS_TIMESTAMP))) &&
1487             pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) {
1488                 /* Diddle with TCP options */
1489                 int hlen;
1490                 opt = hdr + sizeof(struct tcphdr);
1491                 hlen = (th->th_off << 2) - sizeof(struct tcphdr);
1492                 while (hlen >= TCPOLEN_TIMESTAMP) {
1493                         switch (*opt) {
1494                         case TCPOPT_EOL:        /* FALLTHROUGH */
1495                         case TCPOPT_NOP:
1496                                 opt++;
1497                                 hlen--;
1498                                 break;
1499                         case TCPOPT_TIMESTAMP:
1500                                 /* Modulate the timestamps.  Can be used for
1501                                  * NAT detection, OS uptime determination or
1502                                  * reboot detection.
1503                                  */
1504
1505                                 if (got_ts) {
1506                                         /* Huh?  Multiple timestamps!? */
1507                                         if (pf_status.debug >= PF_DEBUG_MISC) {
1508                                                 DPFPRINTF(("multiple TS??"));
1509                                                 pf_print_state(state);
1510                                                 kprintf("\n");
1511                                         }
1512                                         REASON_SET(reason, PFRES_TS);
1513                                         return (PF_DROP);
1514                                 }
1515                                 if (opt[1] >= TCPOLEN_TIMESTAMP) {
1516                                         memcpy(&tsval, &opt[2],
1517                                             sizeof(u_int32_t));
1518                                         if (tsval && src->scrub &&
1519                                             (src->scrub->pfss_flags &
1520                                             PFSS_TIMESTAMP)) {
1521                                                 tsval = ntohl(tsval);
1522                                                 pf_change_a(&opt[2],
1523                                                     &th->th_sum,
1524                                                     htonl(tsval +
1525                                                     src->scrub->pfss_ts_mod),
1526                                                     0);
1527                                                 copyback = 1;
1528                                         }
1529
1530                                         /* Modulate TS reply iff valid (!0) */
1531                                         memcpy(&tsecr, &opt[6],
1532                                             sizeof(u_int32_t));
1533                                         if (tsecr && dst->scrub &&
1534                                             (dst->scrub->pfss_flags &
1535                                             PFSS_TIMESTAMP)) {
1536                                                 tsecr = ntohl(tsecr)
1537                                                     - dst->scrub->pfss_ts_mod;
1538                                                 pf_change_a(&opt[6],
1539                                                     &th->th_sum, htonl(tsecr),
1540                                                     0);
1541                                                 copyback = 1;
1542                                         }
1543                                         got_ts = 1;
1544                                 }
1545                                 /* FALLTHROUGH */
1546                         default:
1547                                 hlen -= MAX(opt[1], 2);
1548                                 opt += MAX(opt[1], 2);
1549                                 break;
1550                         }
1551                 }
1552                 if (copyback) {
1553                         /* Copyback the options, caller copys back header */
1554                         *writeback = 1;
1555                         m_copyback(m, off + sizeof(struct tcphdr),
1556                             (th->th_off << 2) - sizeof(struct tcphdr), hdr +
1557                             sizeof(struct tcphdr));
1558                 }
1559         }
1560
1561
1562         /*
1563          * Must invalidate PAWS checks on connections idle for too long.
1564          * The fastest allowed timestamp clock is 1ms.  That turns out to
1565          * be about 24 days before it wraps.  XXX Right now our lowerbound
1566          * TS echo check only works for the first 12 days of a connection
1567          * when the TS has exhausted half its 32bit space
1568          */
1569 #define TS_MAX_IDLE     (24*24*60*60)
1570 #define TS_MAX_CONN     (12*24*60*60)   /* XXX remove when better tsecr check */
1571
1572         getmicrouptime(&uptime);
1573         if (src->scrub && (src->scrub->pfss_flags & PFSS_PAWS) &&
1574             (uptime.tv_sec - src->scrub->pfss_last.tv_sec > TS_MAX_IDLE ||
1575             time_second - state->creation > TS_MAX_CONN))  {
1576                 if (pf_status.debug >= PF_DEBUG_MISC) {
1577                         DPFPRINTF(("src idled out of PAWS\n"));
1578                         pf_print_state(state);
1579                         kprintf("\n");
1580                 }
1581                 src->scrub->pfss_flags = (src->scrub->pfss_flags & ~PFSS_PAWS)
1582                     | PFSS_PAWS_IDLED;
1583         }
1584         if (dst->scrub && (dst->scrub->pfss_flags & PFSS_PAWS) &&
1585             uptime.tv_sec - dst->scrub->pfss_last.tv_sec > TS_MAX_IDLE) {
1586                 if (pf_status.debug >= PF_DEBUG_MISC) {
1587                         DPFPRINTF(("dst idled out of PAWS\n"));
1588                         pf_print_state(state);
1589                         kprintf("\n");
1590                 }
1591                 dst->scrub->pfss_flags = (dst->scrub->pfss_flags & ~PFSS_PAWS)
1592                     | PFSS_PAWS_IDLED;
1593         }
1594
1595         if (got_ts && src->scrub && dst->scrub &&
1596             (src->scrub->pfss_flags & PFSS_PAWS) &&
1597             (dst->scrub->pfss_flags & PFSS_PAWS)) {
1598                 /* Validate that the timestamps are "in-window".
1599                  * RFC1323 describes TCP Timestamp options that allow
1600                  * measurement of RTT (round trip time) and PAWS
1601                  * (protection against wrapped sequence numbers).  PAWS
1602                  * gives us a set of rules for rejecting packets on
1603                  * long fat pipes (packets that were somehow delayed
1604                  * in transit longer than the time it took to send the
1605                  * full TCP sequence space of 4Gb).  We can use these
1606                  * rules and infer a few others that will let us treat
1607                  * the 32bit timestamp and the 32bit echoed timestamp
1608                  * as sequence numbers to prevent a blind attacker from
1609                  * inserting packets into a connection.
1610                  *
1611                  * RFC1323 tells us:
1612                  *  - The timestamp on this packet must be greater than
1613                  *    or equal to the last value echoed by the other
1614                  *    endpoint.  The RFC says those will be discarded
1615                  *    since it is a dup that has already been acked.
1616                  *    This gives us a lowerbound on the timestamp.
1617                  *        timestamp >= other last echoed timestamp
1618                  *  - The timestamp will be less than or equal to
1619                  *    the last timestamp plus the time between the
1620                  *    last packet and now.  The RFC defines the max
1621                  *    clock rate as 1ms.  We will allow clocks to be
1622                  *    up to 10% fast and will allow a total difference
1623                  *    or 30 seconds due to a route change.  And this
1624                  *    gives us an upperbound on the timestamp.
1625                  *        timestamp <= last timestamp + max ticks
1626                  *    We have to be careful here.  Windows will send an
1627                  *    initial timestamp of zero and then initialize it
1628                  *    to a random value after the 3whs; presumably to
1629                  *    avoid a DoS by having to call an expensive RNG
1630                  *    during a SYN flood.  Proof MS has at least one
1631                  *    good security geek.
1632                  *
1633                  *  - The TCP timestamp option must also echo the other
1634                  *    endpoints timestamp.  The timestamp echoed is the
1635                  *    one carried on the earliest unacknowledged segment
1636                  *    on the left edge of the sequence window.  The RFC
1637                  *    states that the host will reject any echoed
1638                  *    timestamps that were larger than any ever sent.
1639                  *    This gives us an upperbound on the TS echo.
1640                  *        tescr <= largest_tsval
1641                  *  - The lowerbound on the TS echo is a little more
1642                  *    tricky to determine.  The other endpoint's echoed
1643                  *    values will not decrease.  But there may be
1644                  *    network conditions that re-order packets and
1645                  *    cause our view of them to decrease.  For now the
1646                  *    only lowerbound we can safely determine is that
1647                  *    the TS echo will never be less than the original
1648                  *    TS.  XXX There is probably a better lowerbound.
1649                  *    Remove TS_MAX_CONN with better lowerbound check.
1650                  *        tescr >= other original TS
1651                  *
1652                  * It is also important to note that the fastest
1653                  * timestamp clock of 1ms will wrap its 32bit space in
1654                  * 24 days.  So we just disable TS checking after 24
1655                  * days of idle time.  We actually must use a 12d
1656                  * connection limit until we can come up with a better
1657                  * lowerbound to the TS echo check.
1658                  */
1659                 struct timeval delta_ts;
1660                 int ts_fudge;
1661
1662
1663                 /*
1664                  * PFTM_TS_DIFF is how many seconds of leeway to allow
1665                  * a host's timestamp.  This can happen if the previous
1666                  * packet got delayed in transit for much longer than
1667                  * this packet.
1668                  */
1669                 if ((ts_fudge = state->rule.ptr->timeout[PFTM_TS_DIFF]) == 0)
1670                         ts_fudge = pf_default_rule.timeout[PFTM_TS_DIFF];
1671
1672
1673                 /* Calculate max ticks since the last timestamp */
1674 #define TS_MAXFREQ      1100            /* RFC max TS freq of 1Khz + 10% skew */
1675 #define TS_MICROSECS    1000000         /* microseconds per second */
1676 #ifndef timersub
1677 #define timersub(tvp, uvp, vvp)                                         \
1678         do {                                                            \
1679                 (vvp)->tv_sec = (tvp)->tv_sec - (uvp)->tv_sec;          \
1680                 (vvp)->tv_usec = (tvp)->tv_usec - (uvp)->tv_usec;       \
1681                 if ((vvp)->tv_usec < 0) {                               \
1682                         (vvp)->tv_sec--;                                \
1683                         (vvp)->tv_usec += 1000000;                      \
1684                 }                                                       \
1685         } while (0)
1686 #endif
1687
1688                 timersub(&uptime, &src->scrub->pfss_last, &delta_ts);
1689                 tsval_from_last = (delta_ts.tv_sec + ts_fudge) * TS_MAXFREQ;
1690                 tsval_from_last += delta_ts.tv_usec / (TS_MICROSECS/TS_MAXFREQ);
1691
1692
1693                 if ((src->state >= TCPS_ESTABLISHED &&
1694                     dst->state >= TCPS_ESTABLISHED) &&
1695                     (SEQ_LT(tsval, dst->scrub->pfss_tsecr) ||
1696                     SEQ_GT(tsval, src->scrub->pfss_tsval + tsval_from_last) ||
1697                     (tsecr && (SEQ_GT(tsecr, dst->scrub->pfss_tsval) ||
1698                     SEQ_LT(tsecr, dst->scrub->pfss_tsval0))))) {
1699                         /* Bad RFC1323 implementation or an insertion attack.
1700                          *
1701                          * - Solaris 2.6 and 2.7 are known to send another ACK
1702                          *   after the FIN,FIN|ACK,ACK closing that carries
1703                          *   an old timestamp.
1704                          */
1705
1706                         DPFPRINTF(("Timestamp failed %c%c%c%c\n",
1707                             SEQ_LT(tsval, dst->scrub->pfss_tsecr) ? '0' : ' ',
1708                             SEQ_GT(tsval, src->scrub->pfss_tsval +
1709                             tsval_from_last) ? '1' : ' ',
1710                             SEQ_GT(tsecr, dst->scrub->pfss_tsval) ? '2' : ' ',
1711                             SEQ_LT(tsecr, dst->scrub->pfss_tsval0)? '3' : ' '));
1712                         DPFPRINTF((" tsval: %u  tsecr: %u  +ticks: %u  "
1713                             "idle: %lus %lums\n",
1714                             tsval, tsecr, tsval_from_last, delta_ts.tv_sec,
1715                             delta_ts.tv_usec / 1000));
1716                         DPFPRINTF((" src->tsval: %u  tsecr: %u\n",
1717                             src->scrub->pfss_tsval, src->scrub->pfss_tsecr));
1718                         DPFPRINTF((" dst->tsval: %u  tsecr: %u  tsval0: %u"
1719                             "\n", dst->scrub->pfss_tsval,
1720                             dst->scrub->pfss_tsecr, dst->scrub->pfss_tsval0));
1721                         if (pf_status.debug >= PF_DEBUG_MISC) {
1722                                 pf_print_state(state);
1723                                 pf_print_flags(th->th_flags);
1724                                 kprintf("\n");
1725                         }
1726                         REASON_SET(reason, PFRES_TS);
1727                         return (PF_DROP);
1728                 }
1729
1730                 /* XXX I'd really like to require tsecr but it's optional */
1731
1732         } else if (!got_ts && (th->th_flags & TH_RST) == 0 &&
1733             ((src->state == TCPS_ESTABLISHED && dst->state == TCPS_ESTABLISHED)
1734             || pd->p_len > 0 || (th->th_flags & TH_SYN)) &&
1735             src->scrub && dst->scrub &&
1736             (src->scrub->pfss_flags & PFSS_PAWS) &&
1737             (dst->scrub->pfss_flags & PFSS_PAWS)) {
1738                 /* Didn't send a timestamp.  Timestamps aren't really useful
1739                  * when:
1740                  *  - connection opening or closing (often not even sent).
1741                  *    but we must not let an attacker to put a FIN on a
1742                  *    data packet to sneak it through our ESTABLISHED check.
1743                  *  - on a TCP reset.  RFC suggests not even looking at TS.
1744                  *  - on an empty ACK.  The TS will not be echoed so it will
1745                  *    probably not help keep the RTT calculation in sync and
1746                  *    there isn't as much danger when the sequence numbers
1747                  *    got wrapped.  So some stacks don't include TS on empty
1748                  *    ACKs :-(
1749                  *
1750                  * To minimize the disruption to mostly RFC1323 conformant
1751                  * stacks, we will only require timestamps on data packets.
1752                  *
1753                  * And what do ya know, we cannot require timestamps on data
1754                  * packets.  There appear to be devices that do legitimate
1755                  * TCP connection hijacking.  There are HTTP devices that allow
1756                  * a 3whs (with timestamps) and then buffer the HTTP request.
1757                  * If the intermediate device has the HTTP response cache, it
1758                  * will spoof the response but not bother timestamping its
1759                  * packets.  So we can look for the presence of a timestamp in
1760                  * the first data packet and if there, require it in all future
1761                  * packets.
1762                  */
1763
1764                 if (pd->p_len > 0 && (src->scrub->pfss_flags & PFSS_DATA_TS)) {
1765                         /*
1766                          * Hey!  Someone tried to sneak a packet in.  Or the
1767                          * stack changed its RFC1323 behavior?!?!
1768                          */
1769                         if (pf_status.debug >= PF_DEBUG_MISC) {
1770                                 DPFPRINTF(("Did not receive expected RFC1323 "
1771                                     "timestamp\n"));
1772                                 pf_print_state(state);
1773                                 pf_print_flags(th->th_flags);
1774                                 kprintf("\n");
1775                         }
1776                         REASON_SET(reason, PFRES_TS);
1777                         return (PF_DROP);
1778                 }
1779         }
1780
1781
1782         /*
1783          * We will note if a host sends his data packets with or without
1784          * timestamps.  And require all data packets to contain a timestamp
1785          * if the first does.  PAWS implicitly requires that all data packets be
1786          * timestamped.  But I think there are middle-man devices that hijack
1787          * TCP streams immediately after the 3whs and don't timestamp their
1788          * packets (seen in a WWW accelerator or cache).
1789          */
1790         if (pd->p_len > 0 && src->scrub && (src->scrub->pfss_flags &
1791             (PFSS_TIMESTAMP|PFSS_DATA_TS|PFSS_DATA_NOTS)) == PFSS_TIMESTAMP) {
1792                 if (got_ts)
1793                         src->scrub->pfss_flags |= PFSS_DATA_TS;
1794                 else {
1795                         src->scrub->pfss_flags |= PFSS_DATA_NOTS;
1796                         if (pf_status.debug >= PF_DEBUG_MISC && dst->scrub &&
1797                             (dst->scrub->pfss_flags & PFSS_TIMESTAMP)) {
1798                                 /* Don't warn if other host rejected RFC1323 */
1799                                 DPFPRINTF(("Broken RFC1323 stack did not "
1800                                     "timestamp data packet. Disabled PAWS "
1801                                     "security.\n"));
1802                                 pf_print_state(state);
1803                                 pf_print_flags(th->th_flags);
1804                                 kprintf("\n");
1805                         }
1806                 }
1807         }
1808
1809
1810         /*
1811          * Update PAWS values
1812          */
1813         if (got_ts && src->scrub && PFSS_TIMESTAMP == (src->scrub->pfss_flags &
1814             (PFSS_PAWS_IDLED|PFSS_TIMESTAMP))) {
1815                 getmicrouptime(&src->scrub->pfss_last);
1816                 if (SEQ_GEQ(tsval, src->scrub->pfss_tsval) ||
1817                     (src->scrub->pfss_flags & PFSS_PAWS) == 0)
1818                         src->scrub->pfss_tsval = tsval;
1819
1820                 if (tsecr) {
1821                         if (SEQ_GEQ(tsecr, src->scrub->pfss_tsecr) ||
1822                             (src->scrub->pfss_flags & PFSS_PAWS) == 0)
1823                                 src->scrub->pfss_tsecr = tsecr;
1824
1825                         if ((src->scrub->pfss_flags & PFSS_PAWS) == 0 &&
1826                             (SEQ_LT(tsval, src->scrub->pfss_tsval0) ||
1827                             src->scrub->pfss_tsval0 == 0)) {
1828                                 /* tsval0 MUST be the lowest timestamp */
1829                                 src->scrub->pfss_tsval0 = tsval;
1830                         }
1831
1832                         /* Only fully initialized after a TS gets echoed */
1833                         if ((src->scrub->pfss_flags & PFSS_PAWS) == 0)
1834                                 src->scrub->pfss_flags |= PFSS_PAWS;
1835                 }
1836         }
1837
1838         /* I have a dream....  TCP segment reassembly.... */
1839         return (0);
1840 }
1841
1842 int
1843 pf_normalize_tcpopt(struct pf_rule *r, struct mbuf *m, struct tcphdr *th,
1844     int off, sa_family_t af)
1845 {
1846         u_int16_t       *mss;
1847         int              thoff;
1848         int              opt, cnt, optlen = 0;
1849         int              rewrite = 0;
1850         u_char           opts[TCP_MAXOLEN];
1851         u_char          *optp = opts;
1852
1853         thoff = th->th_off << 2;
1854         cnt = thoff - sizeof(struct tcphdr);
1855
1856         if (cnt > 0 && !pf_pull_hdr(m, off + sizeof(*th), opts, cnt,
1857             NULL, NULL, af))
1858                 return (rewrite);
1859
1860         for (; cnt > 0; cnt -= optlen, optp += optlen) {
1861                 opt = optp[0];
1862                 if (opt == TCPOPT_EOL)
1863                         break;
1864                 if (opt == TCPOPT_NOP)
1865                         optlen = 1;
1866                 else {
1867                         if (cnt < 2)
1868                                 break;
1869                         optlen = optp[1];
1870                         if (optlen < 2 || optlen > cnt)
1871                                 break;
1872                 }
1873                 switch (opt) {
1874                 case TCPOPT_MAXSEG:
1875                         mss = (u_int16_t *)(optp + 2);
1876                         if ((ntohs(*mss)) > r->max_mss) {
1877                                 th->th_sum = pf_cksum_fixup(th->th_sum,
1878                                     *mss, htons(r->max_mss), 0);
1879                                 *mss = htons(r->max_mss);
1880                                 rewrite = 1;
1881                         }
1882                         break;
1883                 default:
1884                         break;
1885                 }
1886         }
1887
1888         if (rewrite)
1889                 m_copyback(m, off + sizeof(*th), thoff - sizeof(*th), opts);
1890
1891         return (rewrite);
1892 }