sys/netpfil/pf/pf_norm.c

   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("$FreeBSD$");
  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
  61 struct 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
  72 struct 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 };
  79
  80 struct pf_fragment {
  81         struct pf_fragment_cmp  fr_key;
  82 #define fr_src  fr_key.frc_src
  83 #define fr_dst  fr_key.frc_dst
  84 #define fr_id   fr_key.frc_id
  85 #define fr_af   fr_key.frc_af
  86 #define fr_proto        fr_key.frc_proto
  87
  88         RB_ENTRY(pf_fragment) fr_entry;
  89         TAILQ_ENTRY(pf_fragment) frag_next;
  90         uint8_t         fr_flags;       /* status flags */
  91 #define PFFRAG_SEENLAST         0x0001  /* Seen the last fragment for this */
  92 #define PFFRAG_NOBUFFER         0x0002  /* Non-buffering fragment cache */
  93 #define PFFRAG_DROP             0x0004  /* Drop all fragments */
  94 #define BUFFER_FRAGMENTS(fr)    (!((fr)->fr_flags & PFFRAG_NOBUFFER))
  95         uint16_t        fr_max;         /* fragment data max */
  96         uint32_t        fr_timeout;
  97         uint16_t        fr_maxlen;      /* maximum length of single fragment */
  98         uint16_t        fr_entries;     /* Total number of pf_fragment entries */
  99         TAILQ_HEAD(pf_fragq, pf_frent) fr_queue;
 100 };
 101 #define PF_MAX_FRENT_PER_FRAGMENT       64
 102
 103 struct pf_fragment_tag {
 104         uint16_t        ft_hdrlen;      /* header length of reassembled pkt */
 105         uint16_t        ft_extoff;      /* last extension header offset or 0 */
 106         uint16_t        ft_maxlen;      /* maximum fragment payload length */
 107         uint32_t        ft_id;          /* fragment id */
 108 };
 109
 110 static struct mtx pf_frag_mtx;
 111 #define PF_FRAG_LOCK()          mtx_lock(&pf_frag_mtx)
 112 #define PF_FRAG_UNLOCK()        mtx_unlock(&pf_frag_mtx)
 113 #define PF_FRAG_ASSERT()        mtx_assert(&pf_frag_mtx, MA_OWNED)
 114
 115 VNET_DEFINE(uma_zone_t, pf_state_scrub_z);      /* XXX: shared with pfsync */
 116
 117 static VNET_DEFINE(uma_zone_t, pf_frent_z);
 118 #define V_pf_frent_z    VNET(pf_frent_z)
 119 static VNET_DEFINE(uma_zone_t, pf_frag_z);
 120 #define V_pf_frag_z     VNET(pf_frag_z)
 121
 122 TAILQ_HEAD(pf_fragqueue, pf_fragment);
 123 TAILQ_HEAD(pf_cachequeue, pf_fragment);
 124 static VNET_DEFINE(struct pf_fragqueue, pf_fragqueue);
 125 #define V_pf_fragqueue                  VNET(pf_fragqueue)
 126 static VNET_DEFINE(struct pf_cachequeue,        pf_cachequeue);
 127 #define V_pf_cachequeue                 VNET(pf_cachequeue)
 128 RB_HEAD(pf_frag_tree, pf_fragment);
 129 static VNET_DEFINE(struct pf_frag_tree, pf_frag_tree);
 130 #define V_pf_frag_tree                  VNET(pf_frag_tree)
 131 static VNET_DEFINE(struct pf_frag_tree, pf_cache_tree);
 132 #define V_pf_cache_tree                 VNET(pf_cache_tree)
 133 static int               pf_frag_compare(struct pf_fragment *,
 134                             struct pf_fragment *);
 135 static RB_PROTOTYPE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
 136 static RB_GENERATE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
 137
 138 static void     pf_flush_fragments(void);
 139 static void     pf_free_fragment(struct pf_fragment *);
 140 static void     pf_remove_fragment(struct pf_fragment *);
 141 static int      pf_normalize_tcpopt(struct pf_rule *, struct mbuf *,
 142                     struct tcphdr *, int, sa_family_t);
 143 static struct pf_frent *pf_create_fragment(u_short *);
 144 static struct pf_fragment *pf_find_fragment(struct pf_fragment_cmp *key,
 145                     struct pf_frag_tree *tree);
 146 static struct pf_fragment *pf_fillup_fragment(struct pf_fragment_cmp *,
 147                     struct pf_frent *, u_short *);
 148 static int      pf_isfull_fragment(struct pf_fragment *);
 149 static struct mbuf *pf_join_fragment(struct pf_fragment *);
 150 #ifdef INET
 151 static void     pf_scrub_ip(struct mbuf **, uint32_t, uint8_t, uint8_t);
 152 static int      pf_reassemble(struct mbuf **, struct ip *, int, u_short *);
 153 static struct mbuf *pf_fragcache(struct mbuf **, struct ip*,
 154                     struct pf_fragment **, int, int, int *);
 155 #endif  /* INET */
 156 #ifdef INET6
 157 static int      pf_reassemble6(struct mbuf **, struct ip6_hdr *,
 158                     struct ip6_frag *, uint16_t, uint16_t, u_short *);
 159 static void     pf_scrub_ip6(struct mbuf **, uint8_t);
 160 #endif  /* INET6 */
 161
 162 #define DPFPRINTF(x) do {                               \
 163         if (V_pf_status.debug >= PF_DEBUG_MISC) {       \
 164                 printf("%s: ", __func__);               \
 165                 printf x ;                              \
 166         }                                               \
 167 } while(0)
 168
 169 #ifdef INET
 170 static void
 171 pf_ip2key(struct ip *ip, int dir, struct pf_fragment_cmp *key)
 172 {
 173
 174         key->frc_src.v4 = ip->ip_src;
 175         key->frc_dst.v4 = ip->ip_dst;
 176         key->frc_af = AF_INET;
 177         key->frc_proto = ip->ip_p;
 178         key->frc_id = ip->ip_id;
 179 }
 180 #endif  /* INET */
 181
 182 void
 183 pf_normalize_init(void)
 184 {
 185
 186         V_pf_frag_z = uma_zcreate("pf frags", sizeof(struct pf_fragment),
 187             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
 188         V_pf_frent_z = uma_zcreate("pf frag entries", sizeof(struct pf_frent),
 189             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
 190         V_pf_state_scrub_z = uma_zcreate("pf state scrubs",
 191             sizeof(struct pf_state_scrub),  NULL, NULL, NULL, NULL,
 192             UMA_ALIGN_PTR, 0);
 193
 194         V_pf_limits[PF_LIMIT_FRAGS].zone = V_pf_frent_z;
 195         V_pf_limits[PF_LIMIT_FRAGS].limit = PFFRAG_FRENT_HIWAT;
 196         uma_zone_set_max(V_pf_frent_z, PFFRAG_FRENT_HIWAT);
 197         uma_zone_set_warning(V_pf_frent_z, "PF frag entries limit reached");
 198
 199         mtx_init(&pf_frag_mtx, "pf fragments", NULL, MTX_DEF);
 200
 201         TAILQ_INIT(&V_pf_fragqueue);
 202         TAILQ_INIT(&V_pf_cachequeue);
 203 }
 204
 205 void
 206 pf_normalize_cleanup(void)
 207 {
 208
 209         uma_zdestroy(V_pf_state_scrub_z);
 210         uma_zdestroy(V_pf_frent_z);
 211         uma_zdestroy(V_pf_frag_z);
 212
 213         mtx_destroy(&pf_frag_mtx);
 214 }
 215
 216 static int
 217 pf_frag_compare(struct pf_fragment *a, struct pf_fragment *b)
 218 {
 219         int     diff;
 220
 221         if ((diff = a->fr_id - b->fr_id) != 0)
 222                 return (diff);
 223         if ((diff = a->fr_proto - b->fr_proto) != 0)
 224                 return (diff);
 225         if ((diff = a->fr_af - b->fr_af) != 0)
 226                 return (diff);
 227         if ((diff = pf_addr_cmp(&a->fr_src, &b->fr_src, a->fr_af)) != 0)
 228                 return (diff);
 229         if ((diff = pf_addr_cmp(&a->fr_dst, &b->fr_dst, a->fr_af)) != 0)
 230                 return (diff);
 231         return (0);
 232 }
 233
 234 void
 235 pf_purge_expired_fragments(void)
 236 {
 237         struct pf_fragment      *frag;
 238         u_int32_t                expire = time_uptime -
 239                                     V_pf_default_rule.timeout[PFTM_FRAG];
 240
 241         PF_FRAG_LOCK();
 242         while ((frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue)) != NULL) {
 243                 KASSERT((BUFFER_FRAGMENTS(frag)),
 244                     ("BUFFER_FRAGMENTS(frag) == 0: %s", __FUNCTION__));
 245                 if (frag->fr_timeout > expire)
 246                         break;
 247
 248                 DPFPRINTF(("expiring %d(%p)\n", frag->fr_id, frag));
 249                 pf_free_fragment(frag);
 250         }
 251
 252         while ((frag = TAILQ_LAST(&V_pf_cachequeue, pf_cachequeue)) != NULL) {
 253                 KASSERT((!BUFFER_FRAGMENTS(frag)),
 254                     ("BUFFER_FRAGMENTS(frag) != 0: %s", __FUNCTION__));
 255                 if (frag->fr_timeout > expire)
 256                         break;
 257
 258                 DPFPRINTF(("expiring %d(%p)\n", frag->fr_id, frag));
 259                 pf_free_fragment(frag);
 260                 KASSERT((TAILQ_EMPTY(&V_pf_cachequeue) ||
 261                     TAILQ_LAST(&V_pf_cachequeue, pf_cachequeue) != frag),
 262                     ("!(TAILQ_EMPTY() || TAILQ_LAST() == farg): %s",
 263                     __FUNCTION__));
 264         }
 265         PF_FRAG_UNLOCK();
 266 }
 267
 268 /*
 269  * Try to flush old fragments to make space for new ones
 270  */
 271 static void
 272 pf_flush_fragments(void)
 273 {
 274         struct pf_fragment      *frag, *cache;
 275         int                      goal;
 276
 277         PF_FRAG_ASSERT();
 278
 279         goal = uma_zone_get_cur(V_pf_frent_z) * 9 / 10;
 280         DPFPRINTF(("trying to free %d frag entriess\n", goal));
 281         while (goal < uma_zone_get_cur(V_pf_frent_z)) {
 282                 frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue);
 283                 if (frag)
 284                         pf_free_fragment(frag);
 285                 cache = TAILQ_LAST(&V_pf_cachequeue, pf_cachequeue);
 286                 if (cache)
 287                         pf_free_fragment(cache);
 288                 if (frag == NULL && cache == NULL)
 289                         break;
 290         }
 291 }
 292
 293 /* Frees the fragments and all associated entries */
 294 static void
 295 pf_free_fragment(struct pf_fragment *frag)
 296 {
 297         struct pf_frent         *frent;
 298
 299         PF_FRAG_ASSERT();
 300
 301         /* Free all fragments */
 302         if (BUFFER_FRAGMENTS(frag)) {
 303                 for (frent = TAILQ_FIRST(&frag->fr_queue); frent;
 304                     frent = TAILQ_FIRST(&frag->fr_queue)) {
 305                         TAILQ_REMOVE(&frag->fr_queue, frent, fr_next);
 306
 307                         m_freem(frent->fe_m);
 308                         uma_zfree(V_pf_frent_z, frent);
 309                 }
 310         } else {
 311                 for (frent = TAILQ_FIRST(&frag->fr_queue); frent;
 312                     frent = TAILQ_FIRST(&frag->fr_queue)) {
 313                         TAILQ_REMOVE(&frag->fr_queue, frent, fr_next);
 314
 315                         KASSERT((TAILQ_EMPTY(&frag->fr_queue) ||
 316                             TAILQ_FIRST(&frag->fr_queue)->fe_off >
 317                             frent->fe_len),
 318                             ("! (TAILQ_EMPTY() || TAILQ_FIRST()->fe_off >"
 319                             " frent->fe_len): %s", __func__));
 320
 321                         uma_zfree(V_pf_frent_z, frent);
 322                 }
 323         }
 324
 325         pf_remove_fragment(frag);
 326 }
 327
 328 static struct pf_fragment *
 329 pf_find_fragment(struct pf_fragment_cmp *key, struct pf_frag_tree *tree)
 330 {
 331         struct pf_fragment      *frag;
 332
 333         PF_FRAG_ASSERT();
 334
 335         frag = RB_FIND(pf_frag_tree, tree, (struct pf_fragment *)key);
 336         if (frag != NULL) {
 337                 /* XXX Are we sure we want to update the timeout? */
 338                 frag->fr_timeout = time_uptime;
 339                 if (BUFFER_FRAGMENTS(frag)) {
 340                         TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next);
 341                         TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next);
 342                 } else {
 343                         TAILQ_REMOVE(&V_pf_cachequeue, frag, frag_next);
 344                         TAILQ_INSERT_HEAD(&V_pf_cachequeue, frag, frag_next);
 345                 }
 346         }
 347
 348         return (frag);
 349 }
 350
 351 /* Removes a fragment from the fragment queue and frees the fragment */
 352 static void
 353 pf_remove_fragment(struct pf_fragment *frag)
 354 {
 355
 356         PF_FRAG_ASSERT();
 357
 358         if (BUFFER_FRAGMENTS(frag)) {
 359                 RB_REMOVE(pf_frag_tree, &V_pf_frag_tree, frag);
 360                 TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next);
 361                 uma_zfree(V_pf_frag_z, frag);
 362         } else {
 363                 RB_REMOVE(pf_frag_tree, &V_pf_cache_tree, frag);
 364                 TAILQ_REMOVE(&V_pf_cachequeue, frag, frag_next);
 365                 uma_zfree(V_pf_frag_z, frag);
 366         }
 367 }
 368
 369 static struct pf_frent *
 370 pf_create_fragment(u_short *reason)
 371 {
 372         struct pf_frent *frent;
 373
 374         PF_FRAG_ASSERT();
 375
 376         frent = uma_zalloc(V_pf_frent_z, M_NOWAIT);
 377         if (frent == NULL) {
 378                 pf_flush_fragments();
 379                 frent = uma_zalloc(V_pf_frent_z, M_NOWAIT);
 380                 if (frent == NULL) {
 381                         REASON_SET(reason, PFRES_MEMORY);
 382                         return (NULL);
 383                 }
 384         }
 385
 386         return (frent);
 387 }
 388
 389 struct pf_fragment *
 390 pf_fillup_fragment(struct pf_fragment_cmp *key, struct pf_frent *frent,
 391                 u_short *reason)
 392 {
 393         struct pf_frent         *after, *next, *prev;
 394         struct pf_fragment      *frag;
 395         uint16_t                total;
 396
 397         PF_FRAG_ASSERT();
 398
 399         /* No empty fragments. */
 400         if (frent->fe_len == 0) {
 401                 DPFPRINTF(("bad fragment: len 0"));
 402                 goto bad_fragment;
 403         }
 404
 405         /* All fragments are 8 byte aligned. */
 406         if (frent->fe_mff && (frent->fe_len & 0x7)) {
 407                 DPFPRINTF(("bad fragment: mff and len %d", frent->fe_len));
 408                 goto bad_fragment;
 409         }
 410
 411         /* Respect maximum length, IP_MAXPACKET == IPV6_MAXPACKET. */
 412         if (frent->fe_off + frent->fe_len > IP_MAXPACKET) {
 413                 DPFPRINTF(("bad fragment: max packet %d",
 414                     frent->fe_off + frent->fe_len));
 415                 goto bad_fragment;
 416         }
 417
 418         DPFPRINTF((key->frc_af == AF_INET ?
 419             "reass frag %d @ %d-%d" : "reass frag %#08x @ %d-%d",
 420             key->frc_id, frent->fe_off, frent->fe_off + frent->fe_len));
 421
 422         /* Fully buffer all of the fragments in this fragment queue. */
 423         frag = pf_find_fragment(key, &V_pf_frag_tree);
 424
 425         /* Create a new reassembly queue for this packet. */
 426         if (frag == NULL) {
 427                 frag = uma_zalloc(V_pf_frag_z, M_NOWAIT);
 428                 if (frag == NULL) {
 429                         pf_flush_fragments();
 430                         frag = uma_zalloc(V_pf_frag_z, M_NOWAIT);
 431                         if (frag == NULL) {
 432                                 REASON_SET(reason, PFRES_MEMORY);
 433                                 goto drop_fragment;
 434                         }
 435                 }
 436
 437                 *(struct pf_fragment_cmp *)frag = *key;
 438                 frag->fr_flags = 0;
 439                 frag->fr_timeout = time_uptime;
 440                 frag->fr_maxlen = frent->fe_len;
 441                 frag->fr_entries = 0;
 442                 TAILQ_INIT(&frag->fr_queue);
 443
 444                 RB_INSERT(pf_frag_tree, &V_pf_frag_tree, frag);
 445                 TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next);
 446
 447                 /* We do not have a previous fragment. */
 448                 TAILQ_INSERT_HEAD(&frag->fr_queue, frent, fr_next);
 449
 450                 return (frag);
 451         }
 452
 453         if (frag->fr_entries >= PF_MAX_FRENT_PER_FRAGMENT)
 454                 goto bad_fragment;
 455
 456         KASSERT(!TAILQ_EMPTY(&frag->fr_queue), ("!TAILQ_EMPTY()->fr_queue"));
 457
 458         /* Remember maximum fragment len for refragmentation. */
 459         if (frent->fe_len > frag->fr_maxlen)
 460                 frag->fr_maxlen = frent->fe_len;
 461
 462         /* Maximum data we have seen already. */
 463         total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
 464                 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
 465
 466         /* Non terminal fragments must have more fragments flag. */
 467         if (frent->fe_off + frent->fe_len < total && !frent->fe_mff)
 468                 goto bad_fragment;
 469
 470         /* Check if we saw the last fragment already. */
 471         if (!TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_mff) {
 472                 if (frent->fe_off + frent->fe_len > total ||
 473                     (frent->fe_off + frent->fe_len == total && frent->fe_mff))
 474                         goto bad_fragment;
 475         } else {
 476                 if (frent->fe_off + frent->fe_len == total && !frent->fe_mff)
 477                         goto bad_fragment;
 478         }
 479
 480         /* Find a fragment after the current one. */
 481         prev = NULL;
 482         TAILQ_FOREACH(after, &frag->fr_queue, fr_next) {
 483                 if (after->fe_off > frent->fe_off)
 484                         break;
 485                 prev = after;
 486         }
 487
 488         KASSERT(prev != NULL || after != NULL,
 489             ("prev != NULL || after != NULL"));
 490
 491         if (prev != NULL && prev->fe_off + prev->fe_len > frent->fe_off) {
 492                 uint16_t precut;
 493
 494                 precut = prev->fe_off + prev->fe_len - frent->fe_off;
 495                 if (precut >= frent->fe_len)
 496                         goto bad_fragment;
 497                 DPFPRINTF(("overlap -%d", precut));
 498                 m_adj(frent->fe_m, precut);
 499                 frent->fe_off += precut;
 500                 frent->fe_len -= precut;
 501         }
 502
 503         for (; after != NULL && frent->fe_off + frent->fe_len > after->fe_off;
 504             after = next) {
 505                 uint16_t aftercut;
 506
 507                 aftercut = frent->fe_off + frent->fe_len - after->fe_off;
 508                 DPFPRINTF(("adjust overlap %d", aftercut));
 509                 if (aftercut < after->fe_len) {
 510                         m_adj(after->fe_m, aftercut);
 511                         after->fe_off += aftercut;
 512                         after->fe_len -= aftercut;
 513                         break;
 514                 }
 515
 516                 /* This fragment is completely overlapped, lose it. */
 517                 next = TAILQ_NEXT(after, fr_next);
 518                 m_freem(after->fe_m);
 519                 TAILQ_REMOVE(&frag->fr_queue, after, fr_next);
 520                 uma_zfree(V_pf_frent_z, after);
 521         }
 522
 523         if (prev == NULL)
 524                 TAILQ_INSERT_HEAD(&frag->fr_queue, frent, fr_next);
 525         else
 526                 TAILQ_INSERT_AFTER(&frag->fr_queue, prev, frent, fr_next);
 527
 528         frag->fr_entries++;
 529
 530         return (frag);
 531
 532 bad_fragment:
 533         REASON_SET(reason, PFRES_FRAG);
 534 drop_fragment:
 535         uma_zfree(V_pf_frent_z, frent);
 536         return (NULL);
 537 }
 538
 539 static int
 540 pf_isfull_fragment(struct pf_fragment *frag)
 541 {
 542         struct pf_frent *frent, *next;
 543         uint16_t off, total;
 544
 545         /* Check if we are completely reassembled */
 546         if (TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_mff)
 547                 return (0);
 548
 549         /* Maximum data we have seen already */
 550         total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
 551                 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
 552
 553         /* Check if we have all the data */
 554         off = 0;
 555         for (frent = TAILQ_FIRST(&frag->fr_queue); frent; frent = next) {
 556                 next = TAILQ_NEXT(frent, fr_next);
 557
 558                 off += frent->fe_len;
 559                 if (off < total && (next == NULL || next->fe_off != off)) {
 560                         DPFPRINTF(("missing fragment at %d, next %d, total %d",
 561                             off, next == NULL ? -1 : next->fe_off, total));
 562                         return (0);
 563                 }
 564         }
 565         DPFPRINTF(("%d < %d?", off, total));
 566         if (off < total)
 567                 return (0);
 568         KASSERT(off == total, ("off == total"));
 569
 570         return (1);
 571 }
 572
 573 static struct mbuf *
 574 pf_join_fragment(struct pf_fragment *frag)
 575 {
 576         struct mbuf *m, *m2;
 577         struct pf_frent *frent, *next;
 578
 579         frent = TAILQ_FIRST(&frag->fr_queue);
 580         next = TAILQ_NEXT(frent, fr_next);
 581
 582         m = frent->fe_m;
 583         m_adj(m, (frent->fe_hdrlen + frent->fe_len) - m->m_pkthdr.len);
 584         uma_zfree(V_pf_frent_z, frent);
 585         for (frent = next; frent != NULL; frent = next) {
 586                 next = TAILQ_NEXT(frent, fr_next);
 587
 588                 m2 = frent->fe_m;
 589                 /* Strip off ip header. */
 590                 m_adj(m2, frent->fe_hdrlen);
 591                 /* Strip off any trailing bytes. */
 592                 m_adj(m2, frent->fe_len - m2->m_pkthdr.len);
 593
 594                 uma_zfree(V_pf_frent_z, frent);
 595                 m_cat(m, m2);
 596         }
 597
 598         /* Remove from fragment queue. */
 599         pf_remove_fragment(frag);
 600
 601         return (m);
 602 }
 603
 604 #ifdef INET
 605 static int
 606 pf_reassemble(struct mbuf **m0, struct ip *ip, int dir, u_short *reason)
 607 {
 608         struct mbuf             *m = *m0;
 609         struct pf_frent         *frent;
 610         struct pf_fragment      *frag;
 611         struct pf_fragment_cmp  key;
 612         uint16_t                total, hdrlen;
 613
 614         /* Get an entry for the fragment queue */
 615         if ((frent = pf_create_fragment(reason)) == NULL)
 616                 return (PF_DROP);
 617
 618         frent->fe_m = m;
 619         frent->fe_hdrlen = ip->ip_hl << 2;
 620         frent->fe_extoff = 0;
 621         frent->fe_len = ntohs(ip->ip_len) - (ip->ip_hl << 2);
 622         frent->fe_off = (ntohs(ip->ip_off) & IP_OFFMASK) << 3;
 623         frent->fe_mff = ntohs(ip->ip_off) & IP_MF;
 624
 625         pf_ip2key(ip, dir, &key);
 626
 627         if ((frag = pf_fillup_fragment(&key, frent, reason)) == NULL)
 628                 return (PF_DROP);
 629
 630         /* The mbuf is part of the fragment entry, no direct free or access */
 631         m = *m0 = NULL;
 632
 633         if (!pf_isfull_fragment(frag))
 634                 return (PF_PASS);  /* drop because *m0 is NULL, no error */
 635
 636         /* We have all the data */
 637         frent = TAILQ_FIRST(&frag->fr_queue);
 638         KASSERT(frent != NULL, ("frent != NULL"));
 639         total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
 640                 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
 641         hdrlen = frent->fe_hdrlen;
 642
 643         m = *m0 = pf_join_fragment(frag);
 644         frag = NULL;
 645
 646         if (m->m_flags & M_PKTHDR) {
 647                 int plen = 0;
 648                 for (m = *m0; m; m = m->m_next)
 649                         plen += m->m_len;
 650                 m = *m0;
 651                 m->m_pkthdr.len = plen;
 652         }
 653
 654         ip = mtod(m, struct ip *);
 655         ip->ip_len = htons(hdrlen + total);
 656         ip->ip_off &= ~(IP_MF|IP_OFFMASK);
 657
 658         if (hdrlen + total > IP_MAXPACKET) {
 659                 DPFPRINTF(("drop: too big: %d", total));
 660                 ip->ip_len = 0;
 661                 REASON_SET(reason, PFRES_SHORT);
 662                 /* PF_DROP requires a valid mbuf *m0 in pf_test() */
 663                 return (PF_DROP);
 664         }
 665
 666         DPFPRINTF(("complete: %p(%d)\n", m, ntohs(ip->ip_len)));
 667         return (PF_PASS);
 668 }
 669 #endif  /* INET */
 670
 671 #ifdef INET6
 672 static int
 673 pf_reassemble6(struct mbuf **m0, struct ip6_hdr *ip6, struct ip6_frag *fraghdr,
 674     uint16_t hdrlen, uint16_t extoff, u_short *reason)
 675 {
 676         struct mbuf             *m = *m0;
 677         struct pf_frent         *frent;
 678         struct pf_fragment      *frag;
 679         struct pf_fragment_cmp   key;
 680         struct m_tag            *mtag;
 681         struct pf_fragment_tag  *ftag;
 682         int                      off;
 683         uint32_t                 frag_id;
 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
 709         if ((frag = pf_fillup_fragment(&key, frent, reason)) == NULL) {
 710                 PF_FRAG_UNLOCK();
 711                 return (PF_DROP);
 712         }
 713
 714         /* The mbuf is part of the fragment entry, no direct free or access. */
 715         m = *m0 = NULL;
 716
 717         if (!pf_isfull_fragment(frag)) {
 718                 PF_FRAG_UNLOCK();
 719                 return (PF_PASS);  /* Drop because *m0 is NULL, no error. */
 720         }
 721
 722         /* We have all the data. */
 723         extoff = frent->fe_extoff;
 724         maxlen = frag->fr_maxlen;
 725         frag_id = frag->fr_id;
 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         ftag->ft_id = frag_id;
 763         m_tag_prepend(m, mtag);
 764
 765         ip6 = mtod(m, struct ip6_hdr *);
 766         ip6->ip6_plen = htons(hdrlen - sizeof(struct ip6_hdr) + total);
 767         if (extoff) {
 768                 /* Write protocol into next field of last extension header. */
 769                 m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt),
 770                     &off);
 771                 KASSERT(m, ("%s: short mbuf chain", __func__));
 772                 *(mtod(m, char *) + off) = proto;
 773                 m = *m0;
 774         } else
 775                 ip6->ip6_nxt = proto;
 776
 777         if (hdrlen - sizeof(struct ip6_hdr) + total > IPV6_MAXPACKET) {
 778                 DPFPRINTF(("drop: too big: %d", total));
 779                 ip6->ip6_plen = 0;
 780                 REASON_SET(reason, PFRES_SHORT);
 781                 /* PF_DROP requires a valid mbuf *m0 in pf_test6(). */
 782                 return (PF_DROP);
 783         }
 784
 785         DPFPRINTF(("complete: %p(%d)", m, ntohs(ip6->ip6_plen)));
 786         return (PF_PASS);
 787
 788 fail:
 789         REASON_SET(reason, PFRES_MEMORY);
 790         /* PF_DROP requires a valid mbuf *m0 in pf_test6(), will free later. */
 791         return (PF_DROP);
 792 }
 793 #endif  /* INET6 */
 794
 795 #ifdef INET
 796 static struct mbuf *
 797 pf_fragcache(struct mbuf **m0, struct ip *h, struct pf_fragment **frag, int mff,
 798     int drop, int *nomem)
 799 {
 800         struct mbuf             *m = *m0;
 801         struct pf_frent         *frp, *fra, *cur = NULL;
 802         int                      ip_len = ntohs(h->ip_len) - (h->ip_hl << 2);
 803         u_int16_t                off = ntohs(h->ip_off) << 3;
 804         u_int16_t                max = ip_len + off;
 805         int                      hosed = 0;
 806
 807         PF_FRAG_ASSERT();
 808         KASSERT((*frag == NULL || !BUFFER_FRAGMENTS(*frag)),
 809             ("!(*frag == NULL || !BUFFER_FRAGMENTS(*frag)): %s", __FUNCTION__));
 810
 811         /* Create a new range queue for this packet */
 812         if (*frag == NULL) {
 813                 *frag = uma_zalloc(V_pf_frag_z, M_NOWAIT);
 814                 if (*frag == NULL) {
 815                         pf_flush_fragments();
 816                         *frag = uma_zalloc(V_pf_frag_z, M_NOWAIT);
 817                         if (*frag == NULL)
 818                                 goto no_mem;
 819                 }
 820
 821                 /* Get an entry for the queue */
 822                 cur = uma_zalloc(V_pf_frent_z, M_NOWAIT);
 823                 if (cur == NULL) {
 824                         uma_zfree(V_pf_frag_z, *frag);
 825                         *frag = NULL;
 826                         goto no_mem;
 827                 }
 828
 829                 (*frag)->fr_flags = PFFRAG_NOBUFFER;
 830                 (*frag)->fr_max = 0;
 831                 (*frag)->fr_src.v4 = h->ip_src;
 832                 (*frag)->fr_dst.v4 = h->ip_dst;
 833                 (*frag)->fr_af = AF_INET;
 834                 (*frag)->fr_proto = h->ip_p;
 835                 (*frag)->fr_id = h->ip_id;
 836                 (*frag)->fr_timeout = time_uptime;
 837
 838                 cur->fe_off = off;
 839                 cur->fe_len = max; /* TODO: fe_len = max - off ? */
 840                 TAILQ_INIT(&(*frag)->fr_queue);
 841                 TAILQ_INSERT_HEAD(&(*frag)->fr_queue, cur, fr_next);
 842
 843                 RB_INSERT(pf_frag_tree, &V_pf_cache_tree, *frag);
 844                 TAILQ_INSERT_HEAD(&V_pf_cachequeue, *frag, frag_next);
 845
 846                 DPFPRINTF(("fragcache[%d]: new %d-%d\n", h->ip_id, off, max));
 847
 848                 goto pass;
 849         }
 850
 851         /*
 852          * Find a fragment after the current one:
 853          *  - off contains the real shifted offset.
 854          */
 855         frp = NULL;
 856         TAILQ_FOREACH(fra, &(*frag)->fr_queue, fr_next) {
 857                 if (fra->fe_off > off)
 858                         break;
 859                 frp = fra;
 860         }
 861
 862         KASSERT((frp != NULL || fra != NULL),
 863             ("!(frp != NULL || fra != NULL): %s", __FUNCTION__));
 864
 865         if (frp != NULL) {
 866                 int     precut;
 867
 868                 precut = frp->fe_len - off;
 869                 if (precut >= ip_len) {
 870                         /* Fragment is entirely a duplicate */
 871                         DPFPRINTF(("fragcache[%d]: dead (%d-%d) %d-%d\n",
 872                             h->ip_id, frp->fe_off, frp->fe_len, off, max));
 873                         goto drop_fragment;
 874                 }
 875                 if (precut == 0) {
 876                         /* They are adjacent.  Fixup cache entry */
 877                         DPFPRINTF(("fragcache[%d]: adjacent (%d-%d) %d-%d\n",
 878                             h->ip_id, frp->fe_off, frp->fe_len, off, max));
 879                         frp->fe_len = max;
 880                 } else if (precut > 0) {
 881                         /* The first part of this payload overlaps with a
 882                          * fragment that has already been passed.
 883                          * Need to trim off the first part of the payload.
 884                          * But to do so easily, we need to create another
 885                          * mbuf to throw the original header into.
 886                          */
 887
 888                         DPFPRINTF(("fragcache[%d]: chop %d (%d-%d) %d-%d\n",
 889                             h->ip_id, precut, frp->fe_off, frp->fe_len, off,
 890                             max));
 891
 892                         off += precut;
 893                         max -= precut;
 894                         /* Update the previous frag to encompass this one */
 895                         frp->fe_len = max;
 896
 897                         if (!drop) {
 898                                 /* XXX Optimization opportunity
 899                                  * This is a very heavy way to trim the payload.
 900                                  * we could do it much faster by diddling mbuf
 901                                  * internals but that would be even less legible
 902                                  * than this mbuf magic.  For my next trick,
 903                                  * I'll pull a rabbit out of my laptop.
 904                                  */
 905                                 *m0 = m_dup(m, M_NOWAIT);
 906                                 if (*m0 == NULL)
 907                                         goto no_mem;
 908                                 /* From KAME Project : We have missed this! */
 909                                 m_adj(*m0, (h->ip_hl << 2) -
 910                                     (*m0)->m_pkthdr.len);
 911
 912                                 KASSERT(((*m0)->m_next == NULL),
 913                                     ("(*m0)->m_next != NULL: %s",
 914                                     __FUNCTION__));
 915                                 m_adj(m, precut + (h->ip_hl << 2));
 916                                 m_cat(*m0, m);
 917                                 m = *m0;
 918                                 if (m->m_flags & M_PKTHDR) {
 919                                         int plen = 0;
 920                                         struct mbuf *t;
 921                                         for (t = m; t; t = t->m_next)
 922                                                 plen += t->m_len;
 923                                         m->m_pkthdr.len = plen;
 924                                 }
 925
 926
 927                                 h = mtod(m, struct ip *);
 928
 929                                 KASSERT(((int)m->m_len ==
 930                                     ntohs(h->ip_len) - precut),
 931                                     ("m->m_len != ntohs(h->ip_len) - precut: %s",
 932                                     __FUNCTION__));
 933                                 h->ip_off = htons(ntohs(h->ip_off) +
 934                                     (precut >> 3));
 935                                 h->ip_len = htons(ntohs(h->ip_len) - precut);
 936                         } else {
 937                                 hosed++;
 938                         }
 939                 } else {
 940                         /* There is a gap between fragments */
 941
 942                         DPFPRINTF(("fragcache[%d]: gap %d (%d-%d) %d-%d\n",
 943                             h->ip_id, -precut, frp->fe_off, frp->fe_len, off,
 944                             max));
 945
 946                         cur = uma_zalloc(V_pf_frent_z, M_NOWAIT);
 947                         if (cur == NULL)
 948                                 goto no_mem;
 949
 950                         cur->fe_off = off;
 951                         cur->fe_len = max;
 952                         TAILQ_INSERT_AFTER(&(*frag)->fr_queue, frp, cur, fr_next);
 953                 }
 954         }
 955
 956         if (fra != NULL) {
 957                 int     aftercut;
 958                 int     merge = 0;
 959
 960                 aftercut = max - fra->fe_off;
 961                 if (aftercut == 0) {
 962                         /* Adjacent fragments */
 963                         DPFPRINTF(("fragcache[%d]: adjacent %d-%d (%d-%d)\n",
 964                             h->ip_id, off, max, fra->fe_off, fra->fe_len));
 965                         fra->fe_off = off;
 966                         merge = 1;
 967                 } else if (aftercut > 0) {
 968                         /* Need to chop off the tail of this fragment */
 969                         DPFPRINTF(("fragcache[%d]: chop %d %d-%d (%d-%d)\n",
 970                             h->ip_id, aftercut, off, max, fra->fe_off,
 971                             fra->fe_len));
 972                         fra->fe_off = off;
 973                         max -= aftercut;
 974
 975                         merge = 1;
 976
 977                         if (!drop) {
 978                                 m_adj(m, -aftercut);
 979                                 if (m->m_flags & M_PKTHDR) {
 980                                         int plen = 0;
 981                                         struct mbuf *t;
 982                                         for (t = m; t; t = t->m_next)
 983                                                 plen += t->m_len;
 984                                         m->m_pkthdr.len = plen;
 985                                 }
 986                                 h = mtod(m, struct ip *);
 987                                 KASSERT(((int)m->m_len == ntohs(h->ip_len) - aftercut),
 988                                     ("m->m_len != ntohs(h->ip_len) - aftercut: %s",
 989                                     __FUNCTION__));
 990                                 h->ip_len = htons(ntohs(h->ip_len) - aftercut);
 991                         } else {
 992                                 hosed++;
 993                         }
 994                 } else if (frp == NULL) {
 995                         /* There is a gap between fragments */
 996                         DPFPRINTF(("fragcache[%d]: gap %d %d-%d (%d-%d)\n",
 997                             h->ip_id, -aftercut, off, max, fra->fe_off,
 998                             fra->fe_len));
 999
1000                         cur = uma_zalloc(V_pf_frent_z, M_NOWAIT);
1001                         if (cur == NULL)
1002                                 goto no_mem;
1003
1004                         cur->fe_off = off;
1005                         cur->fe_len = max;
1006                         TAILQ_INSERT_HEAD(&(*frag)->fr_queue, cur, fr_next);
1007                 }
1008
1009
1010                 /* Need to glue together two separate fragment descriptors */
1011                 if (merge) {
1012                         if (cur && fra->fe_off <= cur->fe_len) {
1013                                 /* Need to merge in a previous 'cur' */
1014                                 DPFPRINTF(("fragcache[%d]: adjacent(merge "
1015                                     "%d-%d) %d-%d (%d-%d)\n",
1016                                     h->ip_id, cur->fe_off, cur->fe_len, off,
1017                                     max, fra->fe_off, fra->fe_len));
1018                                 fra->fe_off = cur->fe_off;
1019                                 TAILQ_REMOVE(&(*frag)->fr_queue, cur, fr_next);
1020                                 uma_zfree(V_pf_frent_z, cur);
1021                                 cur = NULL;
1022
1023                         } else if (frp && fra->fe_off <= frp->fe_len) {
1024                                 /* Need to merge in a modified 'frp' */
1025                                 KASSERT((cur == NULL), ("cur != NULL: %s",
1026                                     __FUNCTION__));
1027                                 DPFPRINTF(("fragcache[%d]: adjacent(merge "
1028                                     "%d-%d) %d-%d (%d-%d)\n",
1029                                     h->ip_id, frp->fe_off, frp->fe_len, off,
1030                                     max, fra->fe_off, fra->fe_len));
1031                                 fra->fe_off = frp->fe_off;
1032                                 TAILQ_REMOVE(&(*frag)->fr_queue, frp, fr_next);
1033                                 uma_zfree(V_pf_frent_z, frp);
1034                                 frp = NULL;
1035
1036                         }
1037                 }
1038         }
1039
1040         if (hosed) {
1041                 /*
1042                  * We must keep tracking the overall fragment even when
1043                  * we're going to drop it anyway so that we know when to
1044                  * free the overall descriptor.  Thus we drop the frag late.
1045                  */
1046                 goto drop_fragment;
1047         }
1048
1049
1050  pass:
1051         /* Update maximum data size */
1052         if ((*frag)->fr_max < max)
1053                 (*frag)->fr_max = max;
1054
1055         /* This is the last segment */
1056         if (!mff)
1057                 (*frag)->fr_flags |= PFFRAG_SEENLAST;
1058
1059         /* Check if we are completely reassembled */
1060         if (((*frag)->fr_flags & PFFRAG_SEENLAST) &&
1061             TAILQ_FIRST(&(*frag)->fr_queue)->fe_off == 0 &&
1062             TAILQ_FIRST(&(*frag)->fr_queue)->fe_len == (*frag)->fr_max) {
1063                 /* Remove from fragment queue */
1064                 DPFPRINTF(("fragcache[%d]: done 0-%d\n", h->ip_id,
1065                     (*frag)->fr_max));
1066                 pf_free_fragment(*frag);
1067                 *frag = NULL;
1068         }
1069
1070         return (m);
1071
1072  no_mem:
1073         *nomem = 1;
1074
1075         /* Still need to pay attention to !IP_MF */
1076         if (!mff && *frag != NULL)
1077                 (*frag)->fr_flags |= PFFRAG_SEENLAST;
1078
1079         m_freem(m);
1080         return (NULL);
1081
1082  drop_fragment:
1083
1084         /* Still need to pay attention to !IP_MF */
1085         if (!mff && *frag != NULL)
1086                 (*frag)->fr_flags |= PFFRAG_SEENLAST;
1087
1088         if (drop) {
1089                 /* This fragment has been deemed bad.  Don't reass */
1090                 if (((*frag)->fr_flags & PFFRAG_DROP) == 0)
1091                         DPFPRINTF(("fragcache[%d]: dropping overall fragment\n",
1092                             h->ip_id));
1093                 (*frag)->fr_flags |= PFFRAG_DROP;
1094         }
1095
1096         m_freem(m);
1097         return (NULL);
1098 }
1099 #endif  /* INET */
1100
1101 #ifdef INET6
1102 int
1103 pf_refragment6(struct ifnet *ifp, struct mbuf **m0, struct m_tag *mtag)
1104 {
1105         struct mbuf             *m = *m0, *t;
1106         struct pf_fragment_tag  *ftag = (struct pf_fragment_tag *)(mtag + 1);
1107         struct pf_pdesc          pd;
1108         uint32_t                 frag_id;
1109         uint16_t                 hdrlen, extoff, maxlen;
1110         uint8_t                  proto;
1111         int                      error, action;
1112
1113         hdrlen = ftag->ft_hdrlen;
1114         extoff = ftag->ft_extoff;
1115         maxlen = ftag->ft_maxlen;
1116         frag_id = ftag->ft_id;
1117         m_tag_delete(m, mtag);
1118         mtag = NULL;
1119         ftag = NULL;
1120
1121         if (extoff) {
1122                 int off;
1123
1124                 /* Use protocol from next field of last extension header */
1125                 m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt),
1126                     &off);
1127                 KASSERT((m != NULL), ("pf_refragment6: short mbuf chain"));
1128                 proto = *(mtod(m, caddr_t) + off);
1129                 *(mtod(m, char *) + off) = IPPROTO_FRAGMENT;
1130                 m = *m0;
1131         } else {
1132                 struct ip6_hdr *hdr;
1133
1134                 hdr = mtod(m, struct ip6_hdr *);
1135                 proto = hdr->ip6_nxt;
1136                 hdr->ip6_nxt = IPPROTO_FRAGMENT;
1137         }
1138
1139         /* The MTU must be a multiple of 8 bytes, or we risk doing the
1140          * fragmentation wrong. */
1141         maxlen = maxlen & ~7;
1142
1143         /*
1144          * Maxlen may be less than 8 if there was only a single
1145          * fragment.  As it was fragmented before, add a fragment
1146          * header also for a single fragment.  If total or maxlen
1147          * is less than 8, ip6_fragment() will return EMSGSIZE and
1148          * we drop the packet.
1149          */
1150         error = ip6_fragment(ifp, m, hdrlen, proto, maxlen, frag_id);
1151         m = (*m0)->m_nextpkt;
1152         (*m0)->m_nextpkt = NULL;
1153         if (error == 0) {
1154                 /* The first mbuf contains the unfragmented packet. */
1155                 m_freem(*m0);
1156                 *m0 = NULL;
1157                 action = PF_PASS;
1158         } else {
1159                 /* Drop expects an mbuf to free. */
1160                 DPFPRINTF(("refragment error %d", error));
1161                 action = PF_DROP;
1162         }
1163         for (t = m; m; m = t) {
1164                 t = m->m_nextpkt;
1165                 m->m_nextpkt = NULL;
1166                 m->m_flags |= M_SKIP_FIREWALL;
1167                 memset(&pd, 0, sizeof(pd));
1168                 pd.pf_mtag = pf_find_mtag(m);
1169                 if (error == 0)
1170                         ip6_forward(m, 0);
1171                 else
1172                         m_freem(m);
1173         }
1174
1175         return (action);
1176 }
1177 #endif /* INET6 */
1178
1179 #ifdef INET
1180 int
1181 pf_normalize_ip(struct mbuf **m0, int dir, struct pfi_kif *kif, u_short *reason,
1182     struct pf_pdesc *pd)
1183 {
1184         struct mbuf             *m = *m0;
1185         struct pf_rule          *r;
1186         struct pf_fragment      *frag = NULL;
1187         struct pf_fragment_cmp  key;
1188         struct ip               *h = mtod(m, struct ip *);
1189         int                      mff = (ntohs(h->ip_off) & IP_MF);
1190         int                      hlen = h->ip_hl << 2;
1191         u_int16_t                fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
1192         u_int16_t                max;
1193         int                      ip_len;
1194         int                      ip_off;
1195         int                      tag = -1;
1196         int                      verdict;
1197
1198         PF_RULES_RASSERT();
1199
1200         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1201         while (r != NULL) {
1202                 r->evaluations++;
1203                 if (pfi_kif_match(r->kif, kif) == r->ifnot)
1204                         r = r->skip[PF_SKIP_IFP].ptr;
1205                 else if (r->direction && r->direction != dir)
1206                         r = r->skip[PF_SKIP_DIR].ptr;
1207                 else if (r->af && r->af != AF_INET)
1208                         r = r->skip[PF_SKIP_AF].ptr;
1209                 else if (r->proto && r->proto != h->ip_p)
1210                         r = r->skip[PF_SKIP_PROTO].ptr;
1211                 else if (PF_MISMATCHAW(&r->src.addr,
1212                     (struct pf_addr *)&h->ip_src.s_addr, AF_INET,
1213                     r->src.neg, kif, M_GETFIB(m)))
1214                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
1215                 else if (PF_MISMATCHAW(&r->dst.addr,
1216                     (struct pf_addr *)&h->ip_dst.s_addr, AF_INET,
1217                     r->dst.neg, NULL, M_GETFIB(m)))
1218                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
1219                 else if (r->match_tag && !pf_match_tag(m, r, &tag,
1220                     pd->pf_mtag ? pd->pf_mtag->tag : 0))
1221                         r = TAILQ_NEXT(r, entries);
1222                 else
1223                         break;
1224         }
1225
1226         if (r == NULL || r->action == PF_NOSCRUB)
1227                 return (PF_PASS);
1228         else {
1229                 r->packets[dir == PF_OUT]++;
1230                 r->bytes[dir == PF_OUT] += pd->tot_len;
1231         }
1232
1233         /* Check for illegal packets */
1234         if (hlen < (int)sizeof(struct ip))
1235                 goto drop;
1236
1237         if (hlen > ntohs(h->ip_len))
1238                 goto drop;
1239
1240         /* Clear IP_DF if the rule uses the no-df option */
1241         if (r->rule_flag & PFRULE_NODF && h->ip_off & htons(IP_DF)) {
1242                 u_int16_t ip_off = h->ip_off;
1243
1244                 h->ip_off &= htons(~IP_DF);
1245                 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
1246         }
1247
1248         /* We will need other tests here */
1249         if (!fragoff && !mff)
1250                 goto no_fragment;
1251
1252         /* We're dealing with a fragment now. Don't allow fragments
1253          * with IP_DF to enter the cache. If the flag was cleared by
1254          * no-df above, fine. Otherwise drop it.
1255          */
1256         if (h->ip_off & htons(IP_DF)) {
1257                 DPFPRINTF(("IP_DF\n"));
1258                 goto bad;
1259         }
1260
1261         ip_len = ntohs(h->ip_len) - hlen;
1262         ip_off = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
1263
1264         /* All fragments are 8 byte aligned */
1265         if (mff && (ip_len & 0x7)) {
1266                 DPFPRINTF(("mff and %d\n", ip_len));
1267                 goto bad;
1268         }
1269
1270         /* Respect maximum length */
1271         if (fragoff + ip_len > IP_MAXPACKET) {
1272                 DPFPRINTF(("max packet %d\n", fragoff + ip_len));
1273                 goto bad;
1274         }
1275         max = fragoff + ip_len;
1276
1277         if ((r->rule_flag & (PFRULE_FRAGCROP|PFRULE_FRAGDROP)) == 0) {
1278
1279                 /* Fully buffer all of the fragments */
1280                 PF_FRAG_LOCK();
1281
1282                 pf_ip2key(h, dir, &key);
1283                 frag = pf_find_fragment(&key, &V_pf_frag_tree);
1284
1285                 /* Check if we saw the last fragment already */
1286                 if (frag != NULL && (frag->fr_flags & PFFRAG_SEENLAST) &&
1287                     max > frag->fr_max)
1288                         goto bad;
1289
1290                 /* Might return a completely reassembled mbuf, or NULL */
1291                 DPFPRINTF(("reass frag %d @ %d-%d\n", h->ip_id, fragoff, max));
1292                 verdict = pf_reassemble(m0, h, dir, reason);
1293                 PF_FRAG_UNLOCK();
1294
1295                 if (verdict != PF_PASS)
1296                         return (PF_DROP);
1297
1298                 m = *m0;
1299                 if (m == NULL)
1300                         return (PF_DROP);
1301
1302                 /* use mtag from concatenated mbuf chain */
1303                 pd->pf_mtag = pf_find_mtag(m);
1304 #ifdef DIAGNOSTIC
1305                 if (pd->pf_mtag == NULL) {
1306                         printf("%s: pf_find_mtag returned NULL(1)\n", __func__);
1307                         if ((pd->pf_mtag = pf_get_mtag(m)) == NULL) {
1308                                 m_freem(m);
1309                                 *m0 = NULL;
1310                                 goto no_mem;
1311                         }
1312                 }
1313 #endif
1314                 h = mtod(m, struct ip *);
1315         } else {
1316                 /* non-buffering fragment cache (drops or masks overlaps) */
1317                 int     nomem = 0;
1318
1319                 if (dir == PF_OUT && pd->pf_mtag->flags & PF_TAG_FRAGCACHE) {
1320                         /*
1321                          * Already passed the fragment cache in the
1322                          * input direction.  If we continued, it would
1323                          * appear to be a dup and would be dropped.
1324                          */
1325                         goto fragment_pass;
1326                 }
1327
1328                 PF_FRAG_LOCK();
1329                 pf_ip2key(h, dir, &key);
1330                 frag = pf_find_fragment(&key, &V_pf_cache_tree);
1331
1332                 /* Check if we saw the last fragment already */
1333                 if (frag != NULL && (frag->fr_flags & PFFRAG_SEENLAST) &&
1334                     max > frag->fr_max) {
1335                         if (r->rule_flag & PFRULE_FRAGDROP)
1336                                 frag->fr_flags |= PFFRAG_DROP;
1337                         goto bad;
1338                 }
1339
1340                 *m0 = m = pf_fragcache(m0, h, &frag, mff,
1341                     (r->rule_flag & PFRULE_FRAGDROP) ? 1 : 0, &nomem);
1342                 PF_FRAG_UNLOCK();
1343                 if (m == NULL) {
1344                         if (nomem)
1345                                 goto no_mem;
1346                         goto drop;
1347                 }
1348
1349                 /* use mtag from copied and trimmed mbuf chain */
1350                 pd->pf_mtag = pf_find_mtag(m);
1351 #ifdef DIAGNOSTIC
1352                 if (pd->pf_mtag == NULL) {
1353                         printf("%s: pf_find_mtag returned NULL(2)\n", __func__);
1354                         if ((pd->pf_mtag = pf_get_mtag(m)) == NULL) {
1355                                 m_freem(m);
1356                                 *m0 = NULL;
1357                                 goto no_mem;
1358                         }
1359                 }
1360 #endif
1361                 if (dir == PF_IN)
1362                         pd->pf_mtag->flags |= PF_TAG_FRAGCACHE;
1363
1364                 if (frag != NULL && (frag->fr_flags & PFFRAG_DROP))
1365                         goto drop;
1366                 goto fragment_pass;
1367         }
1368
1369  no_fragment:
1370         /* At this point, only IP_DF is allowed in ip_off */
1371         if (h->ip_off & ~htons(IP_DF)) {
1372                 u_int16_t ip_off = h->ip_off;
1373
1374                 h->ip_off &= htons(IP_DF);
1375                 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
1376         }
1377
1378         /* not missing a return here */
1379
1380  fragment_pass:
1381         pf_scrub_ip(&m, r->rule_flag, r->min_ttl, r->set_tos);
1382
1383         if ((r->rule_flag & (PFRULE_FRAGCROP|PFRULE_FRAGDROP)) == 0)
1384                 pd->flags |= PFDESC_IP_REAS;
1385         return (PF_PASS);
1386
1387  no_mem:
1388         REASON_SET(reason, PFRES_MEMORY);
1389         if (r != NULL && r->log)
1390                 PFLOG_PACKET(kif, m, AF_INET, dir, *reason, r, NULL, NULL, pd,
1391                     1);
1392         return (PF_DROP);
1393
1394  drop:
1395         REASON_SET(reason, PFRES_NORM);
1396         if (r != NULL && r->log)
1397                 PFLOG_PACKET(kif, m, AF_INET, dir, *reason, r, NULL, NULL, pd,
1398                     1);
1399         return (PF_DROP);
1400
1401  bad:
1402         DPFPRINTF(("dropping bad fragment\n"));
1403
1404         /* Free associated fragments */
1405         if (frag != NULL) {
1406                 pf_free_fragment(frag);
1407                 PF_FRAG_UNLOCK();
1408         }
1409
1410         REASON_SET(reason, PFRES_FRAG);
1411         if (r != NULL && r->log)
1412                 PFLOG_PACKET(kif, m, AF_INET, dir, *reason, r, NULL, NULL, pd,
1413                     1);
1414
1415         return (PF_DROP);
1416 }
1417 #endif
1418
1419 #ifdef INET6
1420 int
1421 pf_normalize_ip6(struct mbuf **m0, int dir, struct pfi_kif *kif,
1422     u_short *reason, struct pf_pdesc *pd)
1423 {
1424         struct mbuf             *m = *m0;
1425         struct pf_rule          *r;
1426         struct ip6_hdr          *h = mtod(m, struct ip6_hdr *);
1427         int                      extoff;
1428         int                      off;
1429         struct ip6_ext           ext;
1430         struct ip6_opt           opt;
1431         struct ip6_opt_jumbo     jumbo;
1432         struct ip6_frag          frag;
1433         u_int32_t                jumbolen = 0, plen;
1434         int                      optend;
1435         int                      ooff;
1436         u_int8_t                 proto;
1437         int                      terminal;
1438
1439         PF_RULES_RASSERT();
1440
1441         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1442         while (r != NULL) {
1443                 r->evaluations++;
1444                 if (pfi_kif_match(r->kif, kif) == r->ifnot)
1445                         r = r->skip[PF_SKIP_IFP].ptr;
1446                 else if (r->direction && r->direction != dir)
1447                         r = r->skip[PF_SKIP_DIR].ptr;
1448                 else if (r->af && r->af != AF_INET6)
1449                         r = r->skip[PF_SKIP_AF].ptr;
1450 #if 0 /* header chain! */
1451                 else if (r->proto && r->proto != h->ip6_nxt)
1452                         r = r->skip[PF_SKIP_PROTO].ptr;
1453 #endif
1454                 else if (PF_MISMATCHAW(&r->src.addr,
1455                     (struct pf_addr *)&h->ip6_src, AF_INET6,
1456                     r->src.neg, kif, M_GETFIB(m)))
1457                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
1458                 else if (PF_MISMATCHAW(&r->dst.addr,
1459                     (struct pf_addr *)&h->ip6_dst, AF_INET6,
1460                     r->dst.neg, NULL, M_GETFIB(m)))
1461                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
1462                 else
1463                         break;
1464         }
1465
1466         if (r == NULL || r->action == PF_NOSCRUB)
1467                 return (PF_PASS);
1468         else {
1469                 r->packets[dir == PF_OUT]++;
1470                 r->bytes[dir == PF_OUT] += pd->tot_len;
1471         }
1472
1473         /* Check for illegal packets */
1474         if (sizeof(struct ip6_hdr) + IPV6_MAXPACKET < m->m_pkthdr.len)
1475                 goto drop;
1476
1477         extoff = 0;
1478         off = sizeof(struct ip6_hdr);
1479         proto = h->ip6_nxt;
1480         terminal = 0;
1481         do {
1482                 switch (proto) {
1483                 case IPPROTO_FRAGMENT:
1484                         goto fragment;
1485                         break;
1486                 case IPPROTO_AH:
1487                 case IPPROTO_ROUTING:
1488                 case IPPROTO_DSTOPTS:
1489                         if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL,
1490                             NULL, AF_INET6))
1491                                 goto shortpkt;
1492                         extoff = off;
1493                         if (proto == IPPROTO_AH)
1494                                 off += (ext.ip6e_len + 2) * 4;
1495                         else
1496                                 off += (ext.ip6e_len + 1) * 8;
1497                         proto = ext.ip6e_nxt;
1498                         break;
1499                 case IPPROTO_HOPOPTS:
1500                         if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL,
1501                             NULL, AF_INET6))
1502                                 goto shortpkt;
1503                         extoff = off;
1504                         optend = off + (ext.ip6e_len + 1) * 8;
1505                         ooff = off + sizeof(ext);
1506                         do {
1507                                 if (!pf_pull_hdr(m, ooff, &opt.ip6o_type,
1508                                     sizeof(opt.ip6o_type), NULL, NULL,
1509                                     AF_INET6))
1510                                         goto shortpkt;
1511                                 if (opt.ip6o_type == IP6OPT_PAD1) {
1512                                         ooff++;
1513                                         continue;
1514                                 }
1515                                 if (!pf_pull_hdr(m, ooff, &opt, sizeof(opt),
1516                                     NULL, NULL, AF_INET6))
1517                                         goto shortpkt;
1518                                 if (ooff + sizeof(opt) + opt.ip6o_len > optend)
1519                                         goto drop;
1520                                 switch (opt.ip6o_type) {
1521                                 case IP6OPT_JUMBO:
1522                                         if (h->ip6_plen != 0)
1523                                                 goto drop;
1524                                         if (!pf_pull_hdr(m, ooff, &jumbo,
1525                                             sizeof(jumbo), NULL, NULL,
1526                                             AF_INET6))
1527                                                 goto shortpkt;
1528                                         memcpy(&jumbolen, jumbo.ip6oj_jumbo_len,
1529                                             sizeof(jumbolen));
1530                                         jumbolen = ntohl(jumbolen);
1531                                         if (jumbolen <= IPV6_MAXPACKET)
1532                                                 goto drop;
1533                                         if (sizeof(struct ip6_hdr) + jumbolen !=
1534                                             m->m_pkthdr.len)
1535                                                 goto drop;
1536                                         break;
1537                                 default:
1538                                         break;
1539                                 }
1540                                 ooff += sizeof(opt) + opt.ip6o_len;
1541                         } while (ooff < optend);
1542
1543                         off = optend;
1544                         proto = ext.ip6e_nxt;
1545                         break;
1546                 default:
1547                         terminal = 1;
1548                         break;
1549                 }
1550         } while (!terminal);
1551
1552         /* jumbo payload option must be present, or plen > 0 */
1553         if (ntohs(h->ip6_plen) == 0)
1554                 plen = jumbolen;
1555         else
1556                 plen = ntohs(h->ip6_plen);
1557         if (plen == 0)
1558                 goto drop;
1559         if (sizeof(struct ip6_hdr) + plen > m->m_pkthdr.len)
1560                 goto shortpkt;
1561
1562         pf_scrub_ip6(&m, r->min_ttl);
1563
1564         return (PF_PASS);
1565
1566  fragment:
1567         /* Jumbo payload packets cannot be fragmented. */
1568         plen = ntohs(h->ip6_plen);
1569         if (plen == 0 || jumbolen)
1570                 goto drop;
1571         if (sizeof(struct ip6_hdr) + plen > m->m_pkthdr.len)
1572                 goto shortpkt;
1573
1574         if (!pf_pull_hdr(m, off, &frag, sizeof(frag), NULL, NULL, AF_INET6))
1575                 goto shortpkt;
1576
1577         /* Offset now points to data portion. */
1578         off += sizeof(frag);
1579
1580         /* Returns PF_DROP or *m0 is NULL or completely reassembled mbuf. */
1581         if (pf_reassemble6(m0, h, &frag, off, extoff, reason) != PF_PASS)
1582                 return (PF_DROP);
1583         m = *m0;
1584         if (m == NULL)
1585                 return (PF_DROP);
1586
1587         pd->flags |= PFDESC_IP_REAS;
1588         return (PF_PASS);
1589
1590  shortpkt:
1591         REASON_SET(reason, PFRES_SHORT);
1592         if (r != NULL && r->log)
1593                 PFLOG_PACKET(kif, m, AF_INET6, dir, *reason, r, NULL, NULL, pd,
1594                     1);
1595         return (PF_DROP);
1596
1597  drop:
1598         REASON_SET(reason, PFRES_NORM);
1599         if (r != NULL && r->log)
1600                 PFLOG_PACKET(kif, m, AF_INET6, dir, *reason, r, NULL, NULL, pd,
1601                     1);
1602         return (PF_DROP);
1603 }
1604 #endif /* INET6 */
1605
1606 int
1607 pf_normalize_tcp(int dir, struct pfi_kif *kif, struct mbuf *m, int ipoff,
1608     int off, void *h, struct pf_pdesc *pd)
1609 {
1610         struct pf_rule  *r, *rm = NULL;
1611         struct tcphdr   *th = pd->hdr.tcp;
1612         int              rewrite = 0;
1613         u_short          reason;
1614         u_int8_t         flags;
1615         sa_family_t      af = pd->af;
1616
1617         PF_RULES_RASSERT();
1618
1619         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1620         while (r != NULL) {
1621                 r->evaluations++;
1622                 if (pfi_kif_match(r->kif, kif) == r->ifnot)
1623                         r = r->skip[PF_SKIP_IFP].ptr;
1624                 else if (r->direction && r->direction != dir)
1625                         r = r->skip[PF_SKIP_DIR].ptr;
1626                 else if (r->af && r->af != af)
1627                         r = r->skip[PF_SKIP_AF].ptr;
1628                 else if (r->proto && r->proto != pd->proto)
1629                         r = r->skip[PF_SKIP_PROTO].ptr;
1630                 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
1631                     r->src.neg, kif, M_GETFIB(m)))
1632                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
1633                 else if (r->src.port_op && !pf_match_port(r->src.port_op,
1634                             r->src.port[0], r->src.port[1], th->th_sport))
1635                         r = r->skip[PF_SKIP_SRC_PORT].ptr;
1636                 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
1637                     r->dst.neg, NULL, M_GETFIB(m)))
1638                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
1639                 else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
1640                             r->dst.port[0], r->dst.port[1], th->th_dport))
1641                         r = r->skip[PF_SKIP_DST_PORT].ptr;
1642                 else if (r->os_fingerprint != PF_OSFP_ANY && !pf_osfp_match(
1643                             pf_osfp_fingerprint(pd, m, off, th),
1644                             r->os_fingerprint))
1645                         r = TAILQ_NEXT(r, entries);
1646                 else {
1647                         rm = r;
1648                         break;
1649                 }
1650         }
1651
1652         if (rm == NULL || rm->action == PF_NOSCRUB)
1653                 return (PF_PASS);
1654         else {
1655                 r->packets[dir == PF_OUT]++;
1656                 r->bytes[dir == PF_OUT] += pd->tot_len;
1657         }
1658
1659         if (rm->rule_flag & PFRULE_REASSEMBLE_TCP)
1660                 pd->flags |= PFDESC_TCP_NORM;
1661
1662         flags = th->th_flags;
1663         if (flags & TH_SYN) {
1664                 /* Illegal packet */
1665                 if (flags & TH_RST)
1666                         goto tcp_drop;
1667
1668                 if (flags & TH_FIN)
1669                         goto tcp_drop;
1670         } else {
1671                 /* Illegal packet */
1672                 if (!(flags & (TH_ACK|TH_RST)))
1673                         goto tcp_drop;
1674         }
1675
1676         if (!(flags & TH_ACK)) {
1677                 /* These flags are only valid if ACK is set */
1678                 if ((flags & TH_FIN) || (flags & TH_PUSH) || (flags & TH_URG))
1679                         goto tcp_drop;
1680         }
1681
1682         /* Check for illegal header length */
1683         if (th->th_off < (sizeof(struct tcphdr) >> 2))
1684                 goto tcp_drop;
1685
1686         /* If flags changed, or reserved data set, then adjust */
1687         if (flags != th->th_flags || th->th_x2 != 0) {
1688                 u_int16_t       ov, nv;
1689
1690                 ov = *(u_int16_t *)(&th->th_ack + 1);
1691                 th->th_flags = flags;
1692                 th->th_x2 = 0;
1693                 nv = *(u_int16_t *)(&th->th_ack + 1);
1694
1695                 th->th_sum = pf_proto_cksum_fixup(m, th->th_sum, ov, nv, 0);
1696                 rewrite = 1;
1697         }
1698
1699         /* Remove urgent pointer, if TH_URG is not set */
1700         if (!(flags & TH_URG) && th->th_urp) {
1701                 th->th_sum = pf_proto_cksum_fixup(m, th->th_sum, th->th_urp,
1702                     0, 0);
1703                 th->th_urp = 0;
1704                 rewrite = 1;
1705         }
1706
1707         /* Process options */
1708         if (r->max_mss && pf_normalize_tcpopt(r, m, th, off, pd->af))
1709                 rewrite = 1;
1710
1711         /* copy back packet headers if we sanitized */
1712         if (rewrite)
1713                 m_copyback(m, off, sizeof(*th), (caddr_t)th);
1714
1715         return (PF_PASS);
1716
1717  tcp_drop:
1718         REASON_SET(&reason, PFRES_NORM);
1719         if (rm != NULL && r->log)
1720                 PFLOG_PACKET(kif, m, AF_INET, dir, reason, r, NULL, NULL, pd,
1721                     1);
1722         return (PF_DROP);
1723 }
1724
1725 int
1726 pf_normalize_tcp_init(struct mbuf *m, int off, struct pf_pdesc *pd,
1727     struct tcphdr *th, struct pf_state_peer *src, struct pf_state_peer *dst)
1728 {
1729         u_int32_t tsval, tsecr;
1730         u_int8_t hdr[60];
1731         u_int8_t *opt;
1732
1733         KASSERT((src->scrub == NULL),
1734             ("pf_normalize_tcp_init: src->scrub != NULL"));
1735
1736         src->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT);
1737         if (src->scrub == NULL)
1738                 return (1);
1739
1740         switch (pd->af) {
1741 #ifdef INET
1742         case AF_INET: {
1743                 struct ip *h = mtod(m, struct ip *);
1744                 src->scrub->pfss_ttl = h->ip_ttl;
1745                 break;
1746         }
1747 #endif /* INET */
1748 #ifdef INET6
1749         case AF_INET6: {
1750                 struct ip6_hdr *h = mtod(m, struct ip6_hdr *);
1751                 src->scrub->pfss_ttl = h->ip6_hlim;
1752                 break;
1753         }
1754 #endif /* INET6 */
1755         }
1756
1757
1758         /*
1759          * All normalizations below are only begun if we see the start of
1760          * the connections.  They must all set an enabled bit in pfss_flags
1761          */
1762         if ((th->th_flags & TH_SYN) == 0)
1763                 return (0);
1764
1765
1766         if (th->th_off > (sizeof(struct tcphdr) >> 2) && src->scrub &&
1767             pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) {
1768                 /* Diddle with TCP options */
1769                 int hlen;
1770                 opt = hdr + sizeof(struct tcphdr);
1771                 hlen = (th->th_off << 2) - sizeof(struct tcphdr);
1772                 while (hlen >= TCPOLEN_TIMESTAMP) {
1773                         switch (*opt) {
1774                         case TCPOPT_EOL:        /* FALLTHROUGH */
1775                         case TCPOPT_NOP:
1776                                 opt++;
1777                                 hlen--;
1778                                 break;
1779                         case TCPOPT_TIMESTAMP:
1780                                 if (opt[1] >= TCPOLEN_TIMESTAMP) {
1781                                         src->scrub->pfss_flags |=
1782                                             PFSS_TIMESTAMP;
1783                                         src->scrub->pfss_ts_mod =
1784                                             htonl(arc4random());
1785
1786                                         /* note PFSS_PAWS not set yet */
1787                                         memcpy(&tsval, &opt[2],
1788                                             sizeof(u_int32_t));
1789                                         memcpy(&tsecr, &opt[6],
1790                                             sizeof(u_int32_t));
1791                                         src->scrub->pfss_tsval0 = ntohl(tsval);
1792                                         src->scrub->pfss_tsval = ntohl(tsval);
1793                                         src->scrub->pfss_tsecr = ntohl(tsecr);
1794                                         getmicrouptime(&src->scrub->pfss_last);
1795                                 }
1796                                 /* FALLTHROUGH */
1797                         default:
1798                                 hlen -= MAX(opt[1], 2);
1799                                 opt += MAX(opt[1], 2);
1800                                 break;
1801                         }
1802                 }
1803         }
1804
1805         return (0);
1806 }
1807
1808 void
1809 pf_normalize_tcp_cleanup(struct pf_state *state)
1810 {
1811         if (state->src.scrub)
1812                 uma_zfree(V_pf_state_scrub_z, state->src.scrub);
1813         if (state->dst.scrub)
1814                 uma_zfree(V_pf_state_scrub_z, state->dst.scrub);
1815
1816         /* Someday... flush the TCP segment reassembly descriptors. */
1817 }
1818
1819 int
1820 pf_normalize_tcp_stateful(struct mbuf *m, int off, struct pf_pdesc *pd,
1821     u_short *reason, struct tcphdr *th, struct pf_state *state,
1822     struct pf_state_peer *src, struct pf_state_peer *dst, int *writeback)
1823 {
1824         struct timeval uptime;
1825         u_int32_t tsval, tsecr;
1826         u_int tsval_from_last;
1827         u_int8_t hdr[60];
1828         u_int8_t *opt;
1829         int copyback = 0;
1830         int got_ts = 0;
1831
1832         KASSERT((src->scrub || dst->scrub),
1833             ("%s: src->scrub && dst->scrub!", __func__));
1834
1835         /*
1836          * Enforce the minimum TTL seen for this connection.  Negate a common
1837          * technique to evade an intrusion detection system and confuse
1838          * firewall state code.
1839          */
1840         switch (pd->af) {
1841 #ifdef INET
1842         case AF_INET: {
1843                 if (src->scrub) {
1844                         struct ip *h = mtod(m, struct ip *);
1845                         if (h->ip_ttl > src->scrub->pfss_ttl)
1846                                 src->scrub->pfss_ttl = h->ip_ttl;
1847                         h->ip_ttl = src->scrub->pfss_ttl;
1848                 }
1849                 break;
1850         }
1851 #endif /* INET */
1852 #ifdef INET6
1853         case AF_INET6: {
1854                 if (src->scrub) {
1855                         struct ip6_hdr *h = mtod(m, struct ip6_hdr *);
1856                         if (h->ip6_hlim > src->scrub->pfss_ttl)
1857                                 src->scrub->pfss_ttl = h->ip6_hlim;
1858                         h->ip6_hlim = src->scrub->pfss_ttl;
1859                 }
1860                 break;
1861         }
1862 #endif /* INET6 */
1863         }
1864
1865         if (th->th_off > (sizeof(struct tcphdr) >> 2) &&
1866             ((src->scrub && (src->scrub->pfss_flags & PFSS_TIMESTAMP)) ||
1867             (dst->scrub && (dst->scrub->pfss_flags & PFSS_TIMESTAMP))) &&
1868             pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) {
1869                 /* Diddle with TCP options */
1870                 int hlen;
1871                 opt = hdr + sizeof(struct tcphdr);
1872                 hlen = (th->th_off << 2) - sizeof(struct tcphdr);
1873                 while (hlen >= TCPOLEN_TIMESTAMP) {
1874                         switch (*opt) {
1875                         case TCPOPT_EOL:        /* FALLTHROUGH */
1876                         case TCPOPT_NOP:
1877                                 opt++;
1878                                 hlen--;
1879                                 break;
1880                         case TCPOPT_TIMESTAMP:
1881                                 /* Modulate the timestamps.  Can be used for
1882                                  * NAT detection, OS uptime determination or
1883                                  * reboot detection.
1884                                  */
1885
1886                                 if (got_ts) {
1887                                         /* Huh?  Multiple timestamps!? */
1888                                         if (V_pf_status.debug >= PF_DEBUG_MISC) {
1889                                                 DPFPRINTF(("multiple TS??"));
1890                                                 pf_print_state(state);
1891                                                 printf("\n");
1892                                         }
1893                                         REASON_SET(reason, PFRES_TS);
1894                                         return (PF_DROP);
1895                                 }
1896                                 if (opt[1] >= TCPOLEN_TIMESTAMP) {
1897                                         memcpy(&tsval, &opt[2],
1898                                             sizeof(u_int32_t));
1899                                         if (tsval && src->scrub &&
1900                                             (src->scrub->pfss_flags &
1901                                             PFSS_TIMESTAMP)) {
1902                                                 tsval = ntohl(tsval);
1903                                                 pf_change_proto_a(m, &opt[2],
1904                                                     &th->th_sum,
1905                                                     htonl(tsval +
1906                                                     src->scrub->pfss_ts_mod),
1907                                                     0);
1908                                                 copyback = 1;
1909                                         }
1910
1911                                         /* Modulate TS reply iff valid (!0) */
1912                                         memcpy(&tsecr, &opt[6],
1913                                             sizeof(u_int32_t));
1914                                         if (tsecr && dst->scrub &&
1915                                             (dst->scrub->pfss_flags &
1916                                             PFSS_TIMESTAMP)) {
1917                                                 tsecr = ntohl(tsecr)
1918                                                     - dst->scrub->pfss_ts_mod;
1919                                                 pf_change_proto_a(m, &opt[6],
1920                                                     &th->th_sum, htonl(tsecr),
1921                                                     0);
1922                                                 copyback = 1;
1923                                         }
1924                                         got_ts = 1;
1925                                 }
1926                                 /* FALLTHROUGH */
1927                         default:
1928                                 hlen -= MAX(opt[1], 2);
1929                                 opt += MAX(opt[1], 2);
1930                                 break;
1931                         }
1932                 }
1933                 if (copyback) {
1934                         /* Copyback the options, caller copys back header */
1935                         *writeback = 1;
1936                         m_copyback(m, off + sizeof(struct tcphdr),
1937                             (th->th_off << 2) - sizeof(struct tcphdr), hdr +
1938                             sizeof(struct tcphdr));
1939                 }
1940         }
1941
1942
1943         /*
1944          * Must invalidate PAWS checks on connections idle for too long.
1945          * The fastest allowed timestamp clock is 1ms.  That turns out to
1946          * be about 24 days before it wraps.  XXX Right now our lowerbound
1947          * TS echo check only works for the first 12 days of a connection
1948          * when the TS has exhausted half its 32bit space
1949          */
1950 #define TS_MAX_IDLE     (24*24*60*60)
1951 #define TS_MAX_CONN     (12*24*60*60)   /* XXX remove when better tsecr check */
1952
1953         getmicrouptime(&uptime);
1954         if (src->scrub && (src->scrub->pfss_flags & PFSS_PAWS) &&
1955             (uptime.tv_sec - src->scrub->pfss_last.tv_sec > TS_MAX_IDLE ||
1956             time_uptime - state->creation > TS_MAX_CONN))  {
1957                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1958                         DPFPRINTF(("src idled out of PAWS\n"));
1959                         pf_print_state(state);
1960                         printf("\n");
1961                 }
1962                 src->scrub->pfss_flags = (src->scrub->pfss_flags & ~PFSS_PAWS)
1963                     | PFSS_PAWS_IDLED;
1964         }
1965         if (dst->scrub && (dst->scrub->pfss_flags & PFSS_PAWS) &&
1966             uptime.tv_sec - dst->scrub->pfss_last.tv_sec > TS_MAX_IDLE) {
1967                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1968                         DPFPRINTF(("dst idled out of PAWS\n"));
1969                         pf_print_state(state);
1970                         printf("\n");
1971                 }
1972                 dst->scrub->pfss_flags = (dst->scrub->pfss_flags & ~PFSS_PAWS)
1973                     | PFSS_PAWS_IDLED;
1974         }
1975
1976         if (got_ts && src->scrub && dst->scrub &&
1977             (src->scrub->pfss_flags & PFSS_PAWS) &&
1978             (dst->scrub->pfss_flags & PFSS_PAWS)) {
1979                 /* Validate that the timestamps are "in-window".
1980                  * RFC1323 describes TCP Timestamp options that allow
1981                  * measurement of RTT (round trip time) and PAWS
1982                  * (protection against wrapped sequence numbers).  PAWS
1983                  * gives us a set of rules for rejecting packets on
1984                  * long fat pipes (packets that were somehow delayed
1985                  * in transit longer than the time it took to send the
1986                  * full TCP sequence space of 4Gb).  We can use these
1987                  * rules and infer a few others that will let us treat
1988                  * the 32bit timestamp and the 32bit echoed timestamp
1989                  * as sequence numbers to prevent a blind attacker from
1990                  * inserting packets into a connection.
1991                  *
1992                  * RFC1323 tells us:
1993                  *  - The timestamp on this packet must be greater than
1994                  *    or equal to the last value echoed by the other
1995                  *    endpoint.  The RFC says those will be discarded
1996                  *    since it is a dup that has already been acked.
1997                  *    This gives us a lowerbound on the timestamp.
1998                  *        timestamp >= other last echoed timestamp
1999                  *  - The timestamp will be less than or equal to
2000                  *    the last timestamp plus the time between the
2001                  *    last packet and now.  The RFC defines the max
2002                  *    clock rate as 1ms.  We will allow clocks to be
2003                  *    up to 10% fast and will allow a total difference
2004                  *    or 30 seconds due to a route change.  And this
2005                  *    gives us an upperbound on the timestamp.
2006                  *        timestamp <= last timestamp + max ticks
2007                  *    We have to be careful here.  Windows will send an
2008                  *    initial timestamp of zero and then initialize it
2009                  *    to a random value after the 3whs; presumably to
2010                  *    avoid a DoS by having to call an expensive RNG
2011                  *    during a SYN flood.  Proof MS has at least one
2012                  *    good security geek.
2013                  *
2014                  *  - The TCP timestamp option must also echo the other
2015                  *    endpoints timestamp.  The timestamp echoed is the
2016                  *    one carried on the earliest unacknowledged segment
2017                  *    on the left edge of the sequence window.  The RFC
2018                  *    states that the host will reject any echoed
2019                  *    timestamps that were larger than any ever sent.
2020                  *    This gives us an upperbound on the TS echo.
2021                  *        tescr <= largest_tsval
2022                  *  - The lowerbound on the TS echo is a little more
2023                  *    tricky to determine.  The other endpoint's echoed
2024                  *    values will not decrease.  But there may be
2025                  *    network conditions that re-order packets and
2026                  *    cause our view of them to decrease.  For now the
2027                  *    only lowerbound we can safely determine is that
2028                  *    the TS echo will never be less than the original
2029                  *    TS.  XXX There is probably a better lowerbound.
2030                  *    Remove TS_MAX_CONN with better lowerbound check.
2031                  *        tescr >= other original TS
2032                  *
2033                  * It is also important to note that the fastest
2034                  * timestamp clock of 1ms will wrap its 32bit space in
2035                  * 24 days.  So we just disable TS checking after 24
2036                  * days of idle time.  We actually must use a 12d
2037                  * connection limit until we can come up with a better
2038                  * lowerbound to the TS echo check.
2039                  */
2040                 struct timeval delta_ts;
2041                 int ts_fudge;
2042
2043
2044                 /*
2045                  * PFTM_TS_DIFF is how many seconds of leeway to allow
2046                  * a host's timestamp.  This can happen if the previous
2047                  * packet got delayed in transit for much longer than
2048                  * this packet.
2049                  */
2050                 if ((ts_fudge = state->rule.ptr->timeout[PFTM_TS_DIFF]) == 0)
2051                         ts_fudge = V_pf_default_rule.timeout[PFTM_TS_DIFF];
2052
2053                 /* Calculate max ticks since the last timestamp */
2054 #define TS_MAXFREQ      1100            /* RFC max TS freq of 1Khz + 10% skew */
2055 #define TS_MICROSECS    1000000         /* microseconds per second */
2056                 delta_ts = uptime;
2057                 timevalsub(&delta_ts, &src->scrub->pfss_last);
2058                 tsval_from_last = (delta_ts.tv_sec + ts_fudge) * TS_MAXFREQ;
2059                 tsval_from_last += delta_ts.tv_usec / (TS_MICROSECS/TS_MAXFREQ);
2060
2061                 if ((src->state >= TCPS_ESTABLISHED &&
2062                     dst->state >= TCPS_ESTABLISHED) &&
2063                     (SEQ_LT(tsval, dst->scrub->pfss_tsecr) ||
2064                     SEQ_GT(tsval, src->scrub->pfss_tsval + tsval_from_last) ||
2065                     (tsecr && (SEQ_GT(tsecr, dst->scrub->pfss_tsval) ||
2066                     SEQ_LT(tsecr, dst->scrub->pfss_tsval0))))) {
2067                         /* Bad RFC1323 implementation or an insertion attack.
2068                          *
2069                          * - Solaris 2.6 and 2.7 are known to send another ACK
2070                          *   after the FIN,FIN|ACK,ACK closing that carries
2071                          *   an old timestamp.
2072                          */
2073
2074                         DPFPRINTF(("Timestamp failed %c%c%c%c\n",
2075                             SEQ_LT(tsval, dst->scrub->pfss_tsecr) ? '0' : ' ',
2076                             SEQ_GT(tsval, src->scrub->pfss_tsval +
2077                             tsval_from_last) ? '1' : ' ',
2078                             SEQ_GT(tsecr, dst->scrub->pfss_tsval) ? '2' : ' ',
2079                             SEQ_LT(tsecr, dst->scrub->pfss_tsval0)? '3' : ' '));
2080                         DPFPRINTF((" tsval: %u  tsecr: %u  +ticks: %u  "
2081                             "idle: %jus %lums\n",
2082                             tsval, tsecr, tsval_from_last,
2083                             (uintmax_t)delta_ts.tv_sec,
2084                             delta_ts.tv_usec / 1000));
2085                         DPFPRINTF((" src->tsval: %u  tsecr: %u\n",
2086                             src->scrub->pfss_tsval, src->scrub->pfss_tsecr));
2087                         DPFPRINTF((" dst->tsval: %u  tsecr: %u  tsval0: %u"
2088                             "\n", dst->scrub->pfss_tsval,
2089                             dst->scrub->pfss_tsecr, dst->scrub->pfss_tsval0));
2090                         if (V_pf_status.debug >= PF_DEBUG_MISC) {
2091                                 pf_print_state(state);
2092                                 pf_print_flags(th->th_flags);
2093                                 printf("\n");
2094                         }
2095                         REASON_SET(reason, PFRES_TS);
2096                         return (PF_DROP);
2097                 }
2098
2099                 /* XXX I'd really like to require tsecr but it's optional */
2100
2101         } else if (!got_ts && (th->th_flags & TH_RST) == 0 &&
2102             ((src->state == TCPS_ESTABLISHED && dst->state == TCPS_ESTABLISHED)
2103             || pd->p_len > 0 || (th->th_flags & TH_SYN)) &&
2104             src->scrub && dst->scrub &&
2105             (src->scrub->pfss_flags & PFSS_PAWS) &&
2106             (dst->scrub->pfss_flags & PFSS_PAWS)) {
2107                 /* Didn't send a timestamp.  Timestamps aren't really useful
2108                  * when:
2109                  *  - connection opening or closing (often not even sent).
2110                  *    but we must not let an attacker to put a FIN on a
2111                  *    data packet to sneak it through our ESTABLISHED check.
2112                  *  - on a TCP reset.  RFC suggests not even looking at TS.
2113                  *  - on an empty ACK.  The TS will not be echoed so it will
2114                  *    probably not help keep the RTT calculation in sync and
2115                  *    there isn't as much danger when the sequence numbers
2116                  *    got wrapped.  So some stacks don't include TS on empty
2117                  *    ACKs :-(
2118                  *
2119                  * To minimize the disruption to mostly RFC1323 conformant
2120                  * stacks, we will only require timestamps on data packets.
2121                  *
2122                  * And what do ya know, we cannot require timestamps on data
2123                  * packets.  There appear to be devices that do legitimate
2124                  * TCP connection hijacking.  There are HTTP devices that allow
2125                  * a 3whs (with timestamps) and then buffer the HTTP request.
2126                  * If the intermediate device has the HTTP response cache, it
2127                  * will spoof the response but not bother timestamping its
2128                  * packets.  So we can look for the presence of a timestamp in
2129                  * the first data packet and if there, require it in all future
2130                  * packets.
2131                  */
2132
2133                 if (pd->p_len > 0 && (src->scrub->pfss_flags & PFSS_DATA_TS)) {
2134                         /*
2135                          * Hey!  Someone tried to sneak a packet in.  Or the
2136                          * stack changed its RFC1323 behavior?!?!
2137                          */
2138                         if (V_pf_status.debug >= PF_DEBUG_MISC) {
2139                                 DPFPRINTF(("Did not receive expected RFC1323 "
2140                                     "timestamp\n"));
2141                                 pf_print_state(state);
2142                                 pf_print_flags(th->th_flags);
2143                                 printf("\n");
2144                         }
2145                         REASON_SET(reason, PFRES_TS);
2146                         return (PF_DROP);
2147                 }
2148         }
2149
2150
2151         /*
2152          * We will note if a host sends his data packets with or without
2153          * timestamps.  And require all data packets to contain a timestamp
2154          * if the first does.  PAWS implicitly requires that all data packets be
2155          * timestamped.  But I think there are middle-man devices that hijack
2156          * TCP streams immediately after the 3whs and don't timestamp their
2157          * packets (seen in a WWW accelerator or cache).
2158          */
2159         if (pd->p_len > 0 && src->scrub && (src->scrub->pfss_flags &
2160             (PFSS_TIMESTAMP|PFSS_DATA_TS|PFSS_DATA_NOTS)) == PFSS_TIMESTAMP) {
2161                 if (got_ts)
2162                         src->scrub->pfss_flags |= PFSS_DATA_TS;
2163                 else {
2164                         src->scrub->pfss_flags |= PFSS_DATA_NOTS;
2165                         if (V_pf_status.debug >= PF_DEBUG_MISC && dst->scrub &&
2166                             (dst->scrub->pfss_flags & PFSS_TIMESTAMP)) {
2167                                 /* Don't warn if other host rejected RFC1323 */
2168                                 DPFPRINTF(("Broken RFC1323 stack did not "
2169                                     "timestamp data packet. Disabled PAWS "
2170                                     "security.\n"));
2171                                 pf_print_state(state);
2172                                 pf_print_flags(th->th_flags);
2173                                 printf("\n");
2174                         }
2175                 }
2176         }
2177
2178
2179         /*
2180          * Update PAWS values
2181          */
2182         if (got_ts && src->scrub && PFSS_TIMESTAMP == (src->scrub->pfss_flags &
2183             (PFSS_PAWS_IDLED|PFSS_TIMESTAMP))) {
2184                 getmicrouptime(&src->scrub->pfss_last);
2185                 if (SEQ_GEQ(tsval, src->scrub->pfss_tsval) ||
2186                     (src->scrub->pfss_flags & PFSS_PAWS) == 0)
2187                         src->scrub->pfss_tsval = tsval;
2188
2189                 if (tsecr) {
2190                         if (SEQ_GEQ(tsecr, src->scrub->pfss_tsecr) ||
2191                             (src->scrub->pfss_flags & PFSS_PAWS) == 0)
2192                                 src->scrub->pfss_tsecr = tsecr;
2193
2194                         if ((src->scrub->pfss_flags & PFSS_PAWS) == 0 &&
2195                             (SEQ_LT(tsval, src->scrub->pfss_tsval0) ||
2196                             src->scrub->pfss_tsval0 == 0)) {
2197                                 /* tsval0 MUST be the lowest timestamp */
2198                                 src->scrub->pfss_tsval0 = tsval;
2199                         }
2200
2201                         /* Only fully initialized after a TS gets echoed */
2202                         if ((src->scrub->pfss_flags & PFSS_PAWS) == 0)
2203                                 src->scrub->pfss_flags |= PFSS_PAWS;
2204                 }
2205         }
2206
2207         /* I have a dream....  TCP segment reassembly.... */
2208         return (0);
2209 }
2210
2211 static int
2212 pf_normalize_tcpopt(struct pf_rule *r, struct mbuf *m, struct tcphdr *th,
2213     int off, sa_family_t af)
2214 {
2215         u_int16_t       *mss;
2216         int              thoff;
2217         int              opt, cnt, optlen = 0;
2218         int              rewrite = 0;
2219         u_char           opts[TCP_MAXOLEN];
2220         u_char          *optp = opts;
2221
2222         thoff = th->th_off << 2;
2223         cnt = thoff - sizeof(struct tcphdr);
2224
2225         if (cnt > 0 && !pf_pull_hdr(m, off + sizeof(*th), opts, cnt,
2226             NULL, NULL, af))
2227                 return (rewrite);
2228
2229         for (; cnt > 0; cnt -= optlen, optp += optlen) {
2230                 opt = optp[0];
2231                 if (opt == TCPOPT_EOL)
2232                         break;
2233                 if (opt == TCPOPT_NOP)
2234                         optlen = 1;
2235                 else {
2236                         if (cnt < 2)
2237                                 break;
2238                         optlen = optp[1];
2239                         if (optlen < 2 || optlen > cnt)
2240                                 break;
2241                 }
2242                 switch (opt) {
2243                 case TCPOPT_MAXSEG:
2244                         mss = (u_int16_t *)(optp + 2);
2245                         if ((ntohs(*mss)) > r->max_mss) {
2246                                 th->th_sum = pf_proto_cksum_fixup(m,
2247                                     th->th_sum, *mss, htons(r->max_mss), 0);
2248                                 *mss = htons(r->max_mss);
2249                                 rewrite = 1;
2250                         }
2251                         break;
2252                 default:
2253                         break;
2254                 }
2255         }
2256
2257         if (rewrite)
2258                 m_copyback(m, off + sizeof(*th), thoff - sizeof(*th), opts);
2259
2260         return (rewrite);
2261 }
2262
2263 #ifdef INET
2264 static void
2265 pf_scrub_ip(struct mbuf **m0, u_int32_t flags, u_int8_t min_ttl, u_int8_t tos)
2266 {
2267         struct mbuf             *m = *m0;
2268         struct ip               *h = mtod(m, struct ip *);
2269
2270         /* Clear IP_DF if no-df was requested */
2271         if (flags & PFRULE_NODF && h->ip_off & htons(IP_DF)) {
2272                 u_int16_t ip_off = h->ip_off;
2273
2274                 h->ip_off &= htons(~IP_DF);
2275                 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
2276         }
2277
2278         /* Enforce a minimum ttl, may cause endless packet loops */
2279         if (min_ttl && h->ip_ttl < min_ttl) {
2280                 u_int16_t ip_ttl = h->ip_ttl;
2281
2282                 h->ip_ttl = min_ttl;
2283                 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_ttl, h->ip_ttl, 0);
2284         }
2285
2286         /* Enforce tos */
2287         if (flags & PFRULE_SET_TOS) {
2288                 u_int16_t       ov, nv;
2289
2290                 ov = *(u_int16_t *)h;
2291                 h->ip_tos = tos;
2292                 nv = *(u_int16_t *)h;
2293
2294                 h->ip_sum = pf_cksum_fixup(h->ip_sum, ov, nv, 0);
2295         }
2296
2297         /* random-id, but not for fragments */
2298         if (flags & PFRULE_RANDOMID && !(h->ip_off & ~htons(IP_DF))) {
2299                 u_int16_t ip_id = h->ip_id;
2300
2301                 h->ip_id = ip_randomid();
2302                 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_id, h->ip_id, 0);
2303         }
2304 }
2305 #endif /* INET */
2306
2307 #ifdef INET6
2308 static void
2309 pf_scrub_ip6(struct mbuf **m0, u_int8_t min_ttl)
2310 {
2311         struct mbuf             *m = *m0;
2312         struct ip6_hdr          *h = mtod(m, struct ip6_hdr *);
2313
2314         /* Enforce a minimum ttl, may cause endless packet loops */
2315         if (min_ttl && h->ip6_hlim < min_ttl)
2316                 h->ip6_hlim = min_ttl;
2317 }
2318 #endif