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         TAILQ_HEAD(pf_fragq, pf_frent) fr_queue;
  99 };
 100
 101 struct pf_fragment_tag {
 102         uint16_t        ft_hdrlen;      /* header length of reassembled pkt */
 103         uint16_t        ft_extoff;      /* last extension header offset or 0 */
 104         uint16_t        ft_maxlen;      /* maximum fragment payload length */
 105         uint32_t        ft_id;          /* fragment id */
 106 };
 107
 108 static struct mtx pf_frag_mtx;
 109 #define PF_FRAG_LOCK()          mtx_lock(&pf_frag_mtx)
 110 #define PF_FRAG_UNLOCK()        mtx_unlock(&pf_frag_mtx)
 111 #define PF_FRAG_ASSERT()        mtx_assert(&pf_frag_mtx, MA_OWNED)
 112
 113 VNET_DEFINE(uma_zone_t, pf_state_scrub_z);      /* XXX: shared with pfsync */
 114
 115 static VNET_DEFINE(uma_zone_t, pf_frent_z);
 116 #define V_pf_frent_z    VNET(pf_frent_z)
 117 static VNET_DEFINE(uma_zone_t, pf_frag_z);
 118 #define V_pf_frag_z     VNET(pf_frag_z)
 119
 120 TAILQ_HEAD(pf_fragqueue, pf_fragment);
 121 TAILQ_HEAD(pf_cachequeue, pf_fragment);
 122 static VNET_DEFINE(struct pf_fragqueue, pf_fragqueue);
 123 #define V_pf_fragqueue                  VNET(pf_fragqueue)
 124 static VNET_DEFINE(struct pf_cachequeue,        pf_cachequeue);
 125 #define V_pf_cachequeue                 VNET(pf_cachequeue)
 126 RB_HEAD(pf_frag_tree, pf_fragment);
 127 static VNET_DEFINE(struct pf_frag_tree, pf_frag_tree);
 128 #define V_pf_frag_tree                  VNET(pf_frag_tree)
 129 static VNET_DEFINE(struct pf_frag_tree, pf_cache_tree);
 130 #define V_pf_cache_tree                 VNET(pf_cache_tree)
 131 static int               pf_frag_compare(struct pf_fragment *,
 132                             struct pf_fragment *);
 133 static RB_PROTOTYPE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
 134 static RB_GENERATE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
 135
 136 static void     pf_flush_fragments(void);
 137 static void     pf_free_fragment(struct pf_fragment *);
 138 static void     pf_remove_fragment(struct pf_fragment *);
 139 static int      pf_normalize_tcpopt(struct pf_rule *, struct mbuf *,
 140                     struct tcphdr *, int, sa_family_t);
 141 static struct pf_frent *pf_create_fragment(u_short *);
 142 static struct pf_fragment *pf_find_fragment(struct pf_fragment_cmp *key,
 143                     struct pf_frag_tree *tree);
 144 static struct pf_fragment *pf_fillup_fragment(struct pf_fragment_cmp *,
 145                     struct pf_frent *, u_short *);
 146 static int      pf_isfull_fragment(struct pf_fragment *);
 147 static struct mbuf *pf_join_fragment(struct pf_fragment *);
 148 #ifdef INET
 149 static void     pf_scrub_ip(struct mbuf **, uint32_t, uint8_t, uint8_t);
 150 static int      pf_reassemble(struct mbuf **, struct ip *, int, u_short *);
 151 static struct mbuf *pf_fragcache(struct mbuf **, struct ip*,
 152                     struct pf_fragment **, int, int, int *);
 153 #endif  /* INET */
 154 #ifdef INET6
 155 static int      pf_reassemble6(struct mbuf **, struct ip6_hdr *,
 156                     struct ip6_frag *, uint16_t, uint16_t, u_short *);
 157 static void     pf_scrub_ip6(struct mbuf **, uint8_t);
 158 #endif  /* INET6 */
 159
 160 #define DPFPRINTF(x) do {                               \
 161         if (V_pf_status.debug >= PF_DEBUG_MISC) {       \
 162                 printf("%s: ", __func__);               \
 163                 printf x ;                              \
 164         }                                               \
 165 } while(0)
 166
 167 #ifdef INET
 168 static void
 169 pf_ip2key(struct ip *ip, int dir, struct pf_fragment_cmp *key)
 170 {
 171
 172         key->frc_src.v4 = ip->ip_src;
 173         key->frc_dst.v4 = ip->ip_dst;
 174         key->frc_af = AF_INET;
 175         key->frc_proto = ip->ip_p;
 176         key->frc_id = ip->ip_id;
 177 }
 178 #endif  /* INET */
 179
 180 void
 181 pf_normalize_init(void)
 182 {
 183
 184         V_pf_frag_z = uma_zcreate("pf frags", sizeof(struct pf_fragment),
 185             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
 186         V_pf_frent_z = uma_zcreate("pf frag entries", sizeof(struct pf_frent),
 187             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
 188         V_pf_state_scrub_z = uma_zcreate("pf state scrubs",
 189             sizeof(struct pf_state_scrub),  NULL, NULL, NULL, NULL,
 190             UMA_ALIGN_PTR, 0);
 191
 192         V_pf_limits[PF_LIMIT_FRAGS].zone = V_pf_frent_z;
 193         V_pf_limits[PF_LIMIT_FRAGS].limit = PFFRAG_FRENT_HIWAT;
 194         uma_zone_set_max(V_pf_frent_z, PFFRAG_FRENT_HIWAT);
 195         uma_zone_set_warning(V_pf_frent_z, "PF frag entries limit reached");
 196
 197         mtx_init(&pf_frag_mtx, "pf fragments", NULL, MTX_DEF);
 198
 199         TAILQ_INIT(&V_pf_fragqueue);
 200         TAILQ_INIT(&V_pf_cachequeue);
 201 }
 202
 203 void
 204 pf_normalize_cleanup(void)
 205 {
 206
 207         uma_zdestroy(V_pf_state_scrub_z);
 208         uma_zdestroy(V_pf_frent_z);
 209         uma_zdestroy(V_pf_frag_z);
 210
 211         mtx_destroy(&pf_frag_mtx);
 212 }
 213
 214 static int
 215 pf_frag_compare(struct pf_fragment *a, struct pf_fragment *b)
 216 {
 217         int     diff;
 218
 219         if ((diff = a->fr_id - b->fr_id) != 0)
 220                 return (diff);
 221         if ((diff = a->fr_proto - b->fr_proto) != 0)
 222                 return (diff);
 223         if ((diff = a->fr_af - b->fr_af) != 0)
 224                 return (diff);
 225         if ((diff = pf_addr_cmp(&a->fr_src, &b->fr_src, a->fr_af)) != 0)
 226                 return (diff);
 227         if ((diff = pf_addr_cmp(&a->fr_dst, &b->fr_dst, a->fr_af)) != 0)
 228                 return (diff);
 229         return (0);
 230 }
 231
 232 void
 233 pf_purge_expired_fragments(void)
 234 {
 235         struct pf_fragment      *frag;
 236         u_int32_t                expire = time_uptime -
 237                                     V_pf_default_rule.timeout[PFTM_FRAG];
 238
 239         PF_FRAG_LOCK();
 240         while ((frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue)) != NULL) {
 241                 KASSERT((BUFFER_FRAGMENTS(frag)),
 242                     ("BUFFER_FRAGMENTS(frag) == 0: %s", __FUNCTION__));
 243                 if (frag->fr_timeout > expire)
 244                         break;
 245
 246                 DPFPRINTF(("expiring %d(%p)\n", frag->fr_id, frag));
 247                 pf_free_fragment(frag);
 248         }
 249
 250         while ((frag = TAILQ_LAST(&V_pf_cachequeue, pf_cachequeue)) != NULL) {
 251                 KASSERT((!BUFFER_FRAGMENTS(frag)),
 252                     ("BUFFER_FRAGMENTS(frag) != 0: %s", __FUNCTION__));
 253                 if (frag->fr_timeout > expire)
 254                         break;
 255
 256                 DPFPRINTF(("expiring %d(%p)\n", frag->fr_id, frag));
 257                 pf_free_fragment(frag);
 258                 KASSERT((TAILQ_EMPTY(&V_pf_cachequeue) ||
 259                     TAILQ_LAST(&V_pf_cachequeue, pf_cachequeue) != frag),
 260                     ("!(TAILQ_EMPTY() || TAILQ_LAST() == farg): %s",
 261                     __FUNCTION__));
 262         }
 263         PF_FRAG_UNLOCK();
 264 }
 265
 266 /*
 267  * Try to flush old fragments to make space for new ones
 268  */
 269 static void
 270 pf_flush_fragments(void)
 271 {
 272         struct pf_fragment      *frag, *cache;
 273         int                      goal;
 274
 275         PF_FRAG_ASSERT();
 276
 277         goal = uma_zone_get_cur(V_pf_frent_z) * 9 / 10;
 278         DPFPRINTF(("trying to free %d frag entriess\n", goal));
 279         while (goal < uma_zone_get_cur(V_pf_frent_z)) {
 280                 frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue);
 281                 if (frag)
 282                         pf_free_fragment(frag);
 283                 cache = TAILQ_LAST(&V_pf_cachequeue, pf_cachequeue);
 284                 if (cache)
 285                         pf_free_fragment(cache);
 286                 if (frag == NULL && cache == NULL)
 287                         break;
 288         }
 289 }
 290
 291 /* Frees the fragments and all associated entries */
 292 static void
 293 pf_free_fragment(struct pf_fragment *frag)
 294 {
 295         struct pf_frent         *frent;
 296
 297         PF_FRAG_ASSERT();
 298
 299         /* Free all fragments */
 300         if (BUFFER_FRAGMENTS(frag)) {
 301                 for (frent = TAILQ_FIRST(&frag->fr_queue); frent;
 302                     frent = TAILQ_FIRST(&frag->fr_queue)) {
 303                         TAILQ_REMOVE(&frag->fr_queue, frent, fr_next);
 304
 305                         m_freem(frent->fe_m);
 306                         uma_zfree(V_pf_frent_z, frent);
 307                 }
 308         } else {
 309                 for (frent = TAILQ_FIRST(&frag->fr_queue); frent;
 310                     frent = TAILQ_FIRST(&frag->fr_queue)) {
 311                         TAILQ_REMOVE(&frag->fr_queue, frent, fr_next);
 312
 313                         KASSERT((TAILQ_EMPTY(&frag->fr_queue) ||
 314                             TAILQ_FIRST(&frag->fr_queue)->fe_off >
 315                             frent->fe_len),
 316                             ("! (TAILQ_EMPTY() || TAILQ_FIRST()->fe_off >"
 317                             " frent->fe_len): %s", __func__));
 318
 319                         uma_zfree(V_pf_frent_z, frent);
 320                 }
 321         }
 322
 323         pf_remove_fragment(frag);
 324 }
 325
 326 static struct pf_fragment *
 327 pf_find_fragment(struct pf_fragment_cmp *key, struct pf_frag_tree *tree)
 328 {
 329         struct pf_fragment      *frag;
 330
 331         PF_FRAG_ASSERT();
 332
 333         frag = RB_FIND(pf_frag_tree, tree, (struct pf_fragment *)key);
 334         if (frag != NULL) {
 335                 /* XXX Are we sure we want to update the timeout? */
 336                 frag->fr_timeout = time_uptime;
 337                 if (BUFFER_FRAGMENTS(frag)) {
 338                         TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next);
 339                         TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next);
 340                 } else {
 341                         TAILQ_REMOVE(&V_pf_cachequeue, frag, frag_next);
 342                         TAILQ_INSERT_HEAD(&V_pf_cachequeue, frag, frag_next);
 343                 }
 344         }
 345
 346         return (frag);
 347 }
 348
 349 /* Removes a fragment from the fragment queue and frees the fragment */
 350 static void
 351 pf_remove_fragment(struct pf_fragment *frag)
 352 {
 353
 354         PF_FRAG_ASSERT();
 355
 356         if (BUFFER_FRAGMENTS(frag)) {
 357                 RB_REMOVE(pf_frag_tree, &V_pf_frag_tree, frag);
 358                 TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next);
 359                 uma_zfree(V_pf_frag_z, frag);
 360         } else {
 361                 RB_REMOVE(pf_frag_tree, &V_pf_cache_tree, frag);
 362                 TAILQ_REMOVE(&V_pf_cachequeue, frag, frag_next);
 363                 uma_zfree(V_pf_frag_z, frag);
 364         }
 365 }
 366
 367 static struct pf_frent *
 368 pf_create_fragment(u_short *reason)
 369 {
 370         struct pf_frent *frent;
 371
 372         PF_FRAG_ASSERT();
 373
 374         frent = uma_zalloc(V_pf_frent_z, M_NOWAIT);
 375         if (frent == NULL) {
 376                 pf_flush_fragments();
 377                 frent = uma_zalloc(V_pf_frent_z, M_NOWAIT);
 378                 if (frent == NULL) {
 379                         REASON_SET(reason, PFRES_MEMORY);
 380                         return (NULL);
 381                 }
 382         }
 383
 384         return (frent);
 385 }
 386
 387 struct pf_fragment *
 388 pf_fillup_fragment(struct pf_fragment_cmp *key, struct pf_frent *frent,
 389                 u_short *reason)
 390 {
 391         struct pf_frent         *after, *next, *prev;
 392         struct pf_fragment      *frag;
 393         uint16_t                total;
 394
 395         PF_FRAG_ASSERT();
 396
 397         /* No empty fragments. */
 398         if (frent->fe_len == 0) {
 399                 DPFPRINTF(("bad fragment: len 0"));
 400                 goto bad_fragment;
 401         }
 402
 403         /* All fragments are 8 byte aligned. */
 404         if (frent->fe_mff && (frent->fe_len & 0x7)) {
 405                 DPFPRINTF(("bad fragment: mff and len %d", frent->fe_len));
 406                 goto bad_fragment;
 407         }
 408
 409         /* Respect maximum length, IP_MAXPACKET == IPV6_MAXPACKET. */
 410         if (frent->fe_off + frent->fe_len > IP_MAXPACKET) {
 411                 DPFPRINTF(("bad fragment: max packet %d",
 412                     frent->fe_off + frent->fe_len));
 413                 goto bad_fragment;
 414         }
 415
 416         DPFPRINTF((key->frc_af == AF_INET ?
 417             "reass frag %d @ %d-%d" : "reass frag %#08x @ %d-%d",
 418             key->frc_id, frent->fe_off, frent->fe_off + frent->fe_len));
 419
 420         /* Fully buffer all of the fragments in this fragment queue. */
 421         frag = pf_find_fragment(key, &V_pf_frag_tree);
 422
 423         /* Create a new reassembly queue for this packet. */
 424         if (frag == NULL) {
 425                 frag = uma_zalloc(V_pf_frag_z, M_NOWAIT);
 426                 if (frag == NULL) {
 427                         pf_flush_fragments();
 428                         frag = uma_zalloc(V_pf_frag_z, M_NOWAIT);
 429                         if (frag == NULL) {
 430                                 REASON_SET(reason, PFRES_MEMORY);
 431                                 goto drop_fragment;
 432                         }
 433                 }
 434
 435                 *(struct pf_fragment_cmp *)frag = *key;
 436                 frag->fr_flags = 0;
 437                 frag->fr_timeout = time_second;
 438                 frag->fr_maxlen = frent->fe_len;
 439                 TAILQ_INIT(&frag->fr_queue);
 440
 441                 RB_INSERT(pf_frag_tree, &V_pf_frag_tree, frag);
 442                 TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next);
 443
 444                 /* We do not have a previous fragment. */
 445                 TAILQ_INSERT_HEAD(&frag->fr_queue, frent, fr_next);
 446
 447                 return (frag);
 448         }
 449
 450         KASSERT(!TAILQ_EMPTY(&frag->fr_queue), ("!TAILQ_EMPTY()->fr_queue"));
 451
 452         /* Remember maximum fragment len for refragmentation. */
 453         if (frent->fe_len > frag->fr_maxlen)
 454                 frag->fr_maxlen = frent->fe_len;
 455
 456         /* Maximum data we have seen already. */
 457         total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
 458                 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
 459
 460         /* Non terminal fragments must have more fragments flag. */
 461         if (frent->fe_off + frent->fe_len < total && !frent->fe_mff)
 462                 goto bad_fragment;
 463
 464         /* Check if we saw the last fragment already. */
 465         if (!TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_mff) {
 466                 if (frent->fe_off + frent->fe_len > total ||
 467                     (frent->fe_off + frent->fe_len == total && frent->fe_mff))
 468                         goto bad_fragment;
 469         } else {
 470                 if (frent->fe_off + frent->fe_len == total && !frent->fe_mff)
 471                         goto bad_fragment;
 472         }
 473
 474         /* Find a fragment after the current one. */
 475         prev = NULL;
 476         TAILQ_FOREACH(after, &frag->fr_queue, fr_next) {
 477                 if (after->fe_off > frent->fe_off)
 478                         break;
 479                 prev = after;
 480         }
 481
 482         KASSERT(prev != NULL || after != NULL,
 483             ("prev != NULL || after != NULL"));
 484
 485         if (prev != NULL && prev->fe_off + prev->fe_len > frent->fe_off) {
 486                 uint16_t precut;
 487
 488                 precut = prev->fe_off + prev->fe_len - frent->fe_off;
 489                 if (precut >= frent->fe_len)
 490                         goto bad_fragment;
 491                 DPFPRINTF(("overlap -%d", precut));
 492                 m_adj(frent->fe_m, precut);
 493                 frent->fe_off += precut;
 494                 frent->fe_len -= precut;
 495         }
 496
 497         for (; after != NULL && frent->fe_off + frent->fe_len > after->fe_off;
 498             after = next) {
 499                 uint16_t aftercut;
 500
 501                 aftercut = frent->fe_off + frent->fe_len - after->fe_off;
 502                 DPFPRINTF(("adjust overlap %d", aftercut));
 503                 if (aftercut < after->fe_len) {
 504                         m_adj(after->fe_m, aftercut);
 505                         after->fe_off += aftercut;
 506                         after->fe_len -= aftercut;
 507                         break;
 508                 }
 509
 510                 /* This fragment is completely overlapped, lose it. */
 511                 next = TAILQ_NEXT(after, fr_next);
 512                 m_freem(after->fe_m);
 513                 TAILQ_REMOVE(&frag->fr_queue, after, fr_next);
 514                 uma_zfree(V_pf_frent_z, after);
 515         }
 516
 517         if (prev == NULL)
 518                 TAILQ_INSERT_HEAD(&frag->fr_queue, frent, fr_next);
 519         else
 520                 TAILQ_INSERT_AFTER(&frag->fr_queue, prev, frent, fr_next);
 521
 522         return (frag);
 523
 524 bad_fragment:
 525         REASON_SET(reason, PFRES_FRAG);
 526 drop_fragment:
 527         uma_zfree(V_pf_frent_z, frent);
 528         return (NULL);
 529 }
 530
 531 static int
 532 pf_isfull_fragment(struct pf_fragment *frag)
 533 {
 534         struct pf_frent *frent, *next;
 535         uint16_t off, total;
 536
 537         /* Check if we are completely reassembled */
 538         if (TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_mff)
 539                 return (0);
 540
 541         /* Maximum data we have seen already */
 542         total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
 543                 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
 544
 545         /* Check if we have all the data */
 546         off = 0;
 547         for (frent = TAILQ_FIRST(&frag->fr_queue); frent; frent = next) {
 548                 next = TAILQ_NEXT(frent, fr_next);
 549
 550                 off += frent->fe_len;
 551                 if (off < total && (next == NULL || next->fe_off != off)) {
 552                         DPFPRINTF(("missing fragment at %d, next %d, total %d",
 553                             off, next == NULL ? -1 : next->fe_off, total));
 554                         return (0);
 555                 }
 556         }
 557         DPFPRINTF(("%d < %d?", off, total));
 558         if (off < total)
 559                 return (0);
 560         KASSERT(off == total, ("off == total"));
 561
 562         return (1);
 563 }
 564
 565 static struct mbuf *
 566 pf_join_fragment(struct pf_fragment *frag)
 567 {
 568         struct mbuf *m, *m2;
 569         struct pf_frent *frent, *next;
 570
 571         frent = TAILQ_FIRST(&frag->fr_queue);
 572         next = TAILQ_NEXT(frent, fr_next);
 573
 574         m = frent->fe_m;
 575         m_adj(m, (frent->fe_hdrlen + frent->fe_len) - m->m_pkthdr.len);
 576         uma_zfree(V_pf_frent_z, frent);
 577         for (frent = next; frent != NULL; frent = next) {
 578                 next = TAILQ_NEXT(frent, fr_next);
 579
 580                 m2 = frent->fe_m;
 581                 /* Strip off ip header. */
 582                 m_adj(m2, frent->fe_hdrlen);
 583                 /* Strip off any trailing bytes. */
 584                 m_adj(m2, frent->fe_len - m2->m_pkthdr.len);
 585
 586                 uma_zfree(V_pf_frent_z, frent);
 587                 m_cat(m, m2);
 588         }
 589
 590         /* Remove from fragment queue. */
 591         pf_remove_fragment(frag);
 592
 593         return (m);
 594 }
 595
 596 #ifdef INET
 597 static int
 598 pf_reassemble(struct mbuf **m0, struct ip *ip, int dir, u_short *reason)
 599 {
 600         struct mbuf             *m = *m0;
 601         struct pf_frent         *frent;
 602         struct pf_fragment      *frag;
 603         struct pf_fragment_cmp  key;
 604         uint16_t                total, hdrlen;
 605
 606         /* Get an entry for the fragment queue */
 607         if ((frent = pf_create_fragment(reason)) == NULL)
 608                 return (PF_DROP);
 609
 610         frent->fe_m = m;
 611         frent->fe_hdrlen = ip->ip_hl << 2;
 612         frent->fe_extoff = 0;
 613         frent->fe_len = ntohs(ip->ip_len) - (ip->ip_hl << 2);
 614         frent->fe_off = (ntohs(ip->ip_off) & IP_OFFMASK) << 3;
 615         frent->fe_mff = ntohs(ip->ip_off) & IP_MF;
 616
 617         pf_ip2key(ip, dir, &key);
 618
 619         if ((frag = pf_fillup_fragment(&key, frent, reason)) == NULL)
 620                 return (PF_DROP);
 621
 622         /* The mbuf is part of the fragment entry, no direct free or access */
 623         m = *m0 = NULL;
 624
 625         if (!pf_isfull_fragment(frag))
 626                 return (PF_PASS);  /* drop because *m0 is NULL, no error */
 627
 628         /* We have all the data */
 629         frent = TAILQ_FIRST(&frag->fr_queue);
 630         KASSERT(frent != NULL, ("frent != NULL"));
 631         total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
 632                 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
 633         hdrlen = frent->fe_hdrlen;
 634
 635         m = *m0 = pf_join_fragment(frag);
 636         frag = NULL;
 637
 638         if (m->m_flags & M_PKTHDR) {
 639                 int plen = 0;
 640                 for (m = *m0; m; m = m->m_next)
 641                         plen += m->m_len;
 642                 m = *m0;
 643                 m->m_pkthdr.len = plen;
 644         }
 645
 646         ip = mtod(m, struct ip *);
 647         ip->ip_len = htons(hdrlen + total);
 648         ip->ip_off &= ~(IP_MF|IP_OFFMASK);
 649
 650         if (hdrlen + total > IP_MAXPACKET) {
 651                 DPFPRINTF(("drop: too big: %d", total));
 652                 ip->ip_len = 0;
 653                 REASON_SET(reason, PFRES_SHORT);
 654                 /* PF_DROP requires a valid mbuf *m0 in pf_test() */
 655                 return (PF_DROP);
 656         }
 657
 658         DPFPRINTF(("complete: %p(%d)\n", m, ntohs(ip->ip_len)));
 659         return (PF_PASS);
 660 }
 661 #endif  /* INET */
 662
 663 #ifdef INET6
 664 static int
 665 pf_reassemble6(struct mbuf **m0, struct ip6_hdr *ip6, struct ip6_frag *fraghdr,
 666     uint16_t hdrlen, uint16_t extoff, u_short *reason)
 667 {
 668         struct mbuf             *m = *m0;
 669         struct pf_frent         *frent;
 670         struct pf_fragment      *frag;
 671         struct pf_fragment_cmp   key;
 672         struct m_tag            *mtag;
 673         struct pf_fragment_tag  *ftag;
 674         int                      off;
 675         uint32_t                 frag_id;
 676         uint16_t                 total, maxlen;
 677         uint8_t                  proto;
 678
 679         PF_FRAG_LOCK();
 680
 681         /* Get an entry for the fragment queue. */
 682         if ((frent = pf_create_fragment(reason)) == NULL) {
 683                 PF_FRAG_UNLOCK();
 684                 return (PF_DROP);
 685         }
 686
 687         frent->fe_m = m;
 688         frent->fe_hdrlen = hdrlen;
 689         frent->fe_extoff = extoff;
 690         frent->fe_len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - hdrlen;
 691         frent->fe_off = ntohs(fraghdr->ip6f_offlg & IP6F_OFF_MASK);
 692         frent->fe_mff = fraghdr->ip6f_offlg & IP6F_MORE_FRAG;
 693
 694         key.frc_src.v6 = ip6->ip6_src;
 695         key.frc_dst.v6 = ip6->ip6_dst;
 696         key.frc_af = AF_INET6;
 697         /* Only the first fragment's protocol is relevant. */
 698         key.frc_proto = 0;
 699         key.frc_id = fraghdr->ip6f_ident;
 700
 701         if ((frag = pf_fillup_fragment(&key, frent, reason)) == NULL) {
 702                 PF_FRAG_UNLOCK();
 703                 return (PF_DROP);
 704         }
 705
 706         /* The mbuf is part of the fragment entry, no direct free or access. */
 707         m = *m0 = NULL;
 708
 709         if (!pf_isfull_fragment(frag)) {
 710                 PF_FRAG_UNLOCK();
 711                 return (PF_PASS);  /* Drop because *m0 is NULL, no error. */
 712         }
 713
 714         /* We have all the data. */
 715         extoff = frent->fe_extoff;
 716         maxlen = frag->fr_maxlen;
 717         frag_id = frag->fr_id;
 718         frent = TAILQ_FIRST(&frag->fr_queue);
 719         KASSERT(frent != NULL, ("frent != NULL"));
 720         total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
 721                 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
 722         hdrlen = frent->fe_hdrlen - sizeof(struct ip6_frag);
 723
 724         m = *m0 = pf_join_fragment(frag);
 725         frag = NULL;
 726
 727         PF_FRAG_UNLOCK();
 728
 729         /* Take protocol from first fragment header. */
 730         m = m_getptr(m, hdrlen + offsetof(struct ip6_frag, ip6f_nxt), &off);
 731         KASSERT(m, ("%s: short mbuf chain", __func__));
 732         proto = *(mtod(m, caddr_t) + off);
 733         m = *m0;
 734
 735         /* Delete frag6 header */
 736         if (ip6_deletefraghdr(m, hdrlen, M_NOWAIT) != 0)
 737                 goto fail;
 738
 739         if (m->m_flags & M_PKTHDR) {
 740                 int plen = 0;
 741                 for (m = *m0; m; m = m->m_next)
 742                         plen += m->m_len;
 743                 m = *m0;
 744                 m->m_pkthdr.len = plen;
 745         }
 746
 747         if ((mtag = m_tag_get(PF_REASSEMBLED, sizeof(struct pf_fragment_tag),
 748             M_NOWAIT)) == NULL)
 749                 goto fail;
 750         ftag = (struct pf_fragment_tag *)(mtag + 1);
 751         ftag->ft_hdrlen = hdrlen;
 752         ftag->ft_extoff = extoff;
 753         ftag->ft_maxlen = maxlen;
 754         ftag->ft_id = frag_id;
 755         m_tag_prepend(m, mtag);
 756
 757         ip6 = mtod(m, struct ip6_hdr *);
 758         ip6->ip6_plen = htons(hdrlen - sizeof(struct ip6_hdr) + total);
 759         if (extoff) {
 760                 /* Write protocol into next field of last extension header. */
 761                 m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt),
 762                     &off);
 763                 KASSERT(m, ("%s: short mbuf chain", __func__));
 764                 *(mtod(m, char *) + off) = proto;
 765                 m = *m0;
 766         } else
 767                 ip6->ip6_nxt = proto;
 768
 769         if (hdrlen - sizeof(struct ip6_hdr) + total > IPV6_MAXPACKET) {
 770                 DPFPRINTF(("drop: too big: %d", total));
 771                 ip6->ip6_plen = 0;
 772                 REASON_SET(reason, PFRES_SHORT);
 773                 /* PF_DROP requires a valid mbuf *m0 in pf_test6(). */
 774                 return (PF_DROP);
 775         }
 776
 777         DPFPRINTF(("complete: %p(%d)", m, ntohs(ip6->ip6_plen)));
 778         return (PF_PASS);
 779
 780 fail:
 781         REASON_SET(reason, PFRES_MEMORY);
 782         /* PF_DROP requires a valid mbuf *m0 in pf_test6(), will free later. */
 783         return (PF_DROP);
 784 }
 785 #endif  /* INET6 */
 786
 787 #ifdef INET
 788 static struct mbuf *
 789 pf_fragcache(struct mbuf **m0, struct ip *h, struct pf_fragment **frag, int mff,
 790     int drop, int *nomem)
 791 {
 792         struct mbuf             *m = *m0;
 793         struct pf_frent         *frp, *fra, *cur = NULL;
 794         int                      ip_len = ntohs(h->ip_len) - (h->ip_hl << 2);
 795         u_int16_t                off = ntohs(h->ip_off) << 3;
 796         u_int16_t                max = ip_len + off;
 797         int                      hosed = 0;
 798
 799         PF_FRAG_ASSERT();
 800         KASSERT((*frag == NULL || !BUFFER_FRAGMENTS(*frag)),
 801             ("!(*frag == NULL || !BUFFER_FRAGMENTS(*frag)): %s", __FUNCTION__));
 802
 803         /* Create a new range queue for this packet */
 804         if (*frag == NULL) {
 805                 *frag = uma_zalloc(V_pf_frag_z, M_NOWAIT);
 806                 if (*frag == NULL) {
 807                         pf_flush_fragments();
 808                         *frag = uma_zalloc(V_pf_frag_z, M_NOWAIT);
 809                         if (*frag == NULL)
 810                                 goto no_mem;
 811                 }
 812
 813                 /* Get an entry for the queue */
 814                 cur = uma_zalloc(V_pf_frent_z, M_NOWAIT);
 815                 if (cur == NULL) {
 816                         uma_zfree(V_pf_frag_z, *frag);
 817                         *frag = NULL;
 818                         goto no_mem;
 819                 }
 820
 821                 (*frag)->fr_flags = PFFRAG_NOBUFFER;
 822                 (*frag)->fr_max = 0;
 823                 (*frag)->fr_src.v4 = h->ip_src;
 824                 (*frag)->fr_dst.v4 = h->ip_dst;
 825                 (*frag)->fr_af = AF_INET;
 826                 (*frag)->fr_proto = h->ip_p;
 827                 (*frag)->fr_id = h->ip_id;
 828                 (*frag)->fr_timeout = time_uptime;
 829
 830                 cur->fe_off = off;
 831                 cur->fe_len = max; /* TODO: fe_len = max - off ? */
 832                 TAILQ_INIT(&(*frag)->fr_queue);
 833                 TAILQ_INSERT_HEAD(&(*frag)->fr_queue, cur, fr_next);
 834
 835                 RB_INSERT(pf_frag_tree, &V_pf_cache_tree, *frag);
 836                 TAILQ_INSERT_HEAD(&V_pf_cachequeue, *frag, frag_next);
 837
 838                 DPFPRINTF(("fragcache[%d]: new %d-%d\n", h->ip_id, off, max));
 839
 840                 goto pass;
 841         }
 842
 843         /*
 844          * Find a fragment after the current one:
 845          *  - off contains the real shifted offset.
 846          */
 847         frp = NULL;
 848         TAILQ_FOREACH(fra, &(*frag)->fr_queue, fr_next) {
 849                 if (fra->fe_off > off)
 850                         break;
 851                 frp = fra;
 852         }
 853
 854         KASSERT((frp != NULL || fra != NULL),
 855             ("!(frp != NULL || fra != NULL): %s", __FUNCTION__));
 856
 857         if (frp != NULL) {
 858                 int     precut;
 859
 860                 precut = frp->fe_len - off;
 861                 if (precut >= ip_len) {
 862                         /* Fragment is entirely a duplicate */
 863                         DPFPRINTF(("fragcache[%d]: dead (%d-%d) %d-%d\n",
 864                             h->ip_id, frp->fe_off, frp->fe_len, off, max));
 865                         goto drop_fragment;
 866                 }
 867                 if (precut == 0) {
 868                         /* They are adjacent.  Fixup cache entry */
 869                         DPFPRINTF(("fragcache[%d]: adjacent (%d-%d) %d-%d\n",
 870                             h->ip_id, frp->fe_off, frp->fe_len, off, max));
 871                         frp->fe_len = max;
 872                 } else if (precut > 0) {
 873                         /* The first part of this payload overlaps with a
 874                          * fragment that has already been passed.
 875                          * Need to trim off the first part of the payload.
 876                          * But to do so easily, we need to create another
 877                          * mbuf to throw the original header into.
 878                          */
 879
 880                         DPFPRINTF(("fragcache[%d]: chop %d (%d-%d) %d-%d\n",
 881                             h->ip_id, precut, frp->fe_off, frp->fe_len, off,
 882                             max));
 883
 884                         off += precut;
 885                         max -= precut;
 886                         /* Update the previous frag to encompass this one */
 887                         frp->fe_len = max;
 888
 889                         if (!drop) {
 890                                 /* XXX Optimization opportunity
 891                                  * This is a very heavy way to trim the payload.
 892                                  * we could do it much faster by diddling mbuf
 893                                  * internals but that would be even less legible
 894                                  * than this mbuf magic.  For my next trick,
 895                                  * I'll pull a rabbit out of my laptop.
 896                                  */
 897                                 *m0 = m_dup(m, M_NOWAIT);
 898                                 if (*m0 == NULL)
 899                                         goto no_mem;
 900                                 /* From KAME Project : We have missed this! */
 901                                 m_adj(*m0, (h->ip_hl << 2) -
 902                                     (*m0)->m_pkthdr.len);
 903
 904                                 KASSERT(((*m0)->m_next == NULL),
 905                                     ("(*m0)->m_next != NULL: %s",
 906                                     __FUNCTION__));
 907                                 m_adj(m, precut + (h->ip_hl << 2));
 908                                 m_cat(*m0, m);
 909                                 m = *m0;
 910                                 if (m->m_flags & M_PKTHDR) {
 911                                         int plen = 0;
 912                                         struct mbuf *t;
 913                                         for (t = m; t; t = t->m_next)
 914                                                 plen += t->m_len;
 915                                         m->m_pkthdr.len = plen;
 916                                 }
 917
 918
 919                                 h = mtod(m, struct ip *);
 920
 921                                 KASSERT(((int)m->m_len ==
 922                                     ntohs(h->ip_len) - precut),
 923                                     ("m->m_len != ntohs(h->ip_len) - precut: %s",
 924                                     __FUNCTION__));
 925                                 h->ip_off = htons(ntohs(h->ip_off) +
 926                                     (precut >> 3));
 927                                 h->ip_len = htons(ntohs(h->ip_len) - precut);
 928                         } else {
 929                                 hosed++;
 930                         }
 931                 } else {
 932                         /* There is a gap between fragments */
 933
 934                         DPFPRINTF(("fragcache[%d]: gap %d (%d-%d) %d-%d\n",
 935                             h->ip_id, -precut, frp->fe_off, frp->fe_len, off,
 936                             max));
 937
 938                         cur = uma_zalloc(V_pf_frent_z, M_NOWAIT);
 939                         if (cur == NULL)
 940                                 goto no_mem;
 941
 942                         cur->fe_off = off;
 943                         cur->fe_len = max;
 944                         TAILQ_INSERT_AFTER(&(*frag)->fr_queue, frp, cur, fr_next);
 945                 }
 946         }
 947
 948         if (fra != NULL) {
 949                 int     aftercut;
 950                 int     merge = 0;
 951
 952                 aftercut = max - fra->fe_off;
 953                 if (aftercut == 0) {
 954                         /* Adjacent fragments */
 955                         DPFPRINTF(("fragcache[%d]: adjacent %d-%d (%d-%d)\n",
 956                             h->ip_id, off, max, fra->fe_off, fra->fe_len));
 957                         fra->fe_off = off;
 958                         merge = 1;
 959                 } else if (aftercut > 0) {
 960                         /* Need to chop off the tail of this fragment */
 961                         DPFPRINTF(("fragcache[%d]: chop %d %d-%d (%d-%d)\n",
 962                             h->ip_id, aftercut, off, max, fra->fe_off,
 963                             fra->fe_len));
 964                         fra->fe_off = off;
 965                         max -= aftercut;
 966
 967                         merge = 1;
 968
 969                         if (!drop) {
 970                                 m_adj(m, -aftercut);
 971                                 if (m->m_flags & M_PKTHDR) {
 972                                         int plen = 0;
 973                                         struct mbuf *t;
 974                                         for (t = m; t; t = t->m_next)
 975                                                 plen += t->m_len;
 976                                         m->m_pkthdr.len = plen;
 977                                 }
 978                                 h = mtod(m, struct ip *);
 979                                 KASSERT(((int)m->m_len == ntohs(h->ip_len) - aftercut),
 980                                     ("m->m_len != ntohs(h->ip_len) - aftercut: %s",
 981                                     __FUNCTION__));
 982                                 h->ip_len = htons(ntohs(h->ip_len) - aftercut);
 983                         } else {
 984                                 hosed++;
 985                         }
 986                 } else if (frp == NULL) {
 987                         /* There is a gap between fragments */
 988                         DPFPRINTF(("fragcache[%d]: gap %d %d-%d (%d-%d)\n",
 989                             h->ip_id, -aftercut, off, max, fra->fe_off,
 990                             fra->fe_len));
 991
 992                         cur = uma_zalloc(V_pf_frent_z, M_NOWAIT);
 993                         if (cur == NULL)
 994                                 goto no_mem;
 995
 996                         cur->fe_off = off;
 997                         cur->fe_len = max;
 998                         TAILQ_INSERT_HEAD(&(*frag)->fr_queue, cur, fr_next);
 999                 }
1000
1001
1002                 /* Need to glue together two separate fragment descriptors */
1003                 if (merge) {
1004                         if (cur && fra->fe_off <= cur->fe_len) {
1005                                 /* Need to merge in a previous 'cur' */
1006                                 DPFPRINTF(("fragcache[%d]: adjacent(merge "
1007                                     "%d-%d) %d-%d (%d-%d)\n",
1008                                     h->ip_id, cur->fe_off, cur->fe_len, off,
1009                                     max, fra->fe_off, fra->fe_len));
1010                                 fra->fe_off = cur->fe_off;
1011                                 TAILQ_REMOVE(&(*frag)->fr_queue, cur, fr_next);
1012                                 uma_zfree(V_pf_frent_z, cur);
1013                                 cur = NULL;
1014
1015                         } else if (frp && fra->fe_off <= frp->fe_len) {
1016                                 /* Need to merge in a modified 'frp' */
1017                                 KASSERT((cur == NULL), ("cur != NULL: %s",
1018                                     __FUNCTION__));
1019                                 DPFPRINTF(("fragcache[%d]: adjacent(merge "
1020                                     "%d-%d) %d-%d (%d-%d)\n",
1021                                     h->ip_id, frp->fe_off, frp->fe_len, off,
1022                                     max, fra->fe_off, fra->fe_len));
1023                                 fra->fe_off = frp->fe_off;
1024                                 TAILQ_REMOVE(&(*frag)->fr_queue, frp, fr_next);
1025                                 uma_zfree(V_pf_frent_z, frp);
1026                                 frp = NULL;
1027
1028                         }
1029                 }
1030         }
1031
1032         if (hosed) {
1033                 /*
1034                  * We must keep tracking the overall fragment even when
1035                  * we're going to drop it anyway so that we know when to
1036                  * free the overall descriptor.  Thus we drop the frag late.
1037                  */
1038                 goto drop_fragment;
1039         }
1040
1041
1042  pass:
1043         /* Update maximum data size */
1044         if ((*frag)->fr_max < max)
1045                 (*frag)->fr_max = max;
1046
1047         /* This is the last segment */
1048         if (!mff)
1049                 (*frag)->fr_flags |= PFFRAG_SEENLAST;
1050
1051         /* Check if we are completely reassembled */
1052         if (((*frag)->fr_flags & PFFRAG_SEENLAST) &&
1053             TAILQ_FIRST(&(*frag)->fr_queue)->fe_off == 0 &&
1054             TAILQ_FIRST(&(*frag)->fr_queue)->fe_len == (*frag)->fr_max) {
1055                 /* Remove from fragment queue */
1056                 DPFPRINTF(("fragcache[%d]: done 0-%d\n", h->ip_id,
1057                     (*frag)->fr_max));
1058                 pf_free_fragment(*frag);
1059                 *frag = NULL;
1060         }
1061
1062         return (m);
1063
1064  no_mem:
1065         *nomem = 1;
1066
1067         /* Still need to pay attention to !IP_MF */
1068         if (!mff && *frag != NULL)
1069                 (*frag)->fr_flags |= PFFRAG_SEENLAST;
1070
1071         m_freem(m);
1072         return (NULL);
1073
1074  drop_fragment:
1075
1076         /* Still need to pay attention to !IP_MF */
1077         if (!mff && *frag != NULL)
1078                 (*frag)->fr_flags |= PFFRAG_SEENLAST;
1079
1080         if (drop) {
1081                 /* This fragment has been deemed bad.  Don't reass */
1082                 if (((*frag)->fr_flags & PFFRAG_DROP) == 0)
1083                         DPFPRINTF(("fragcache[%d]: dropping overall fragment\n",
1084                             h->ip_id));
1085                 (*frag)->fr_flags |= PFFRAG_DROP;
1086         }
1087
1088         m_freem(m);
1089         return (NULL);
1090 }
1091 #endif  /* INET */
1092
1093 #ifdef INET6
1094 int
1095 pf_refragment6(struct ifnet *ifp, struct mbuf **m0, struct m_tag *mtag)
1096 {
1097         struct mbuf             *m = *m0, *t;
1098         struct pf_fragment_tag  *ftag = (struct pf_fragment_tag *)(mtag + 1);
1099         struct pf_pdesc          pd;
1100         uint32_t                 frag_id;
1101         uint16_t                 hdrlen, extoff, maxlen;
1102         uint8_t                  proto;
1103         int                      error, action;
1104
1105         hdrlen = ftag->ft_hdrlen;
1106         extoff = ftag->ft_extoff;
1107         maxlen = ftag->ft_maxlen;
1108         frag_id = ftag->ft_id;
1109         m_tag_delete(m, mtag);
1110         mtag = NULL;
1111         ftag = NULL;
1112
1113         if (extoff) {
1114                 int off;
1115
1116                 /* Use protocol from next field of last extension header */
1117                 m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt),
1118                     &off);
1119                 KASSERT((m != NULL), ("pf_refragment6: short mbuf chain"));
1120                 proto = *(mtod(m, caddr_t) + off);
1121                 *(mtod(m, char *) + off) = IPPROTO_FRAGMENT;
1122                 m = *m0;
1123         } else {
1124                 struct ip6_hdr *hdr;
1125
1126                 hdr = mtod(m, struct ip6_hdr *);
1127                 proto = hdr->ip6_nxt;
1128                 hdr->ip6_nxt = IPPROTO_FRAGMENT;
1129         }
1130
1131         /*
1132          * Maxlen may be less than 8 if there was only a single
1133          * fragment.  As it was fragmented before, add a fragment
1134          * header also for a single fragment.  If total or maxlen
1135          * is less than 8, ip6_fragment() will return EMSGSIZE and
1136          * we drop the packet.
1137          */
1138         error = ip6_fragment(ifp, m, hdrlen, proto, maxlen, frag_id);
1139         m = (*m0)->m_nextpkt;
1140         (*m0)->m_nextpkt = NULL;
1141         if (error == 0) {
1142                 /* The first mbuf contains the unfragmented packet. */
1143                 m_freem(*m0);
1144                 *m0 = NULL;
1145                 action = PF_PASS;
1146         } else {
1147                 /* Drop expects an mbuf to free. */
1148                 DPFPRINTF(("refragment error %d", error));
1149                 action = PF_DROP;
1150         }
1151         for (t = m; m; m = t) {
1152                 t = m->m_nextpkt;
1153                 m->m_nextpkt = NULL;
1154                 m->m_flags |= M_SKIP_FIREWALL;
1155                 memset(&pd, 0, sizeof(pd));
1156                 pd.pf_mtag = pf_find_mtag(m);
1157                 if (error == 0)
1158                         ip6_forward(m, 0);
1159                 else
1160                         m_freem(m);
1161         }
1162
1163         return (action);
1164 }
1165 #endif /* INET6 */
1166
1167 #ifdef INET
1168 int
1169 pf_normalize_ip(struct mbuf **m0, int dir, struct pfi_kif *kif, u_short *reason,
1170     struct pf_pdesc *pd)
1171 {
1172         struct mbuf             *m = *m0;
1173         struct pf_rule          *r;
1174         struct pf_fragment      *frag = NULL;
1175         struct pf_fragment_cmp  key;
1176         struct ip               *h = mtod(m, struct ip *);
1177         int                      mff = (ntohs(h->ip_off) & IP_MF);
1178         int                      hlen = h->ip_hl << 2;
1179         u_int16_t                fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
1180         u_int16_t                max;
1181         int                      ip_len;
1182         int                      ip_off;
1183         int                      tag = -1;
1184         int                      verdict;
1185
1186         PF_RULES_RASSERT();
1187
1188         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1189         while (r != NULL) {
1190                 r->evaluations++;
1191                 if (pfi_kif_match(r->kif, kif) == r->ifnot)
1192                         r = r->skip[PF_SKIP_IFP].ptr;
1193                 else if (r->direction && r->direction != dir)
1194                         r = r->skip[PF_SKIP_DIR].ptr;
1195                 else if (r->af && r->af != AF_INET)
1196                         r = r->skip[PF_SKIP_AF].ptr;
1197                 else if (r->proto && r->proto != h->ip_p)
1198                         r = r->skip[PF_SKIP_PROTO].ptr;
1199                 else if (PF_MISMATCHAW(&r->src.addr,
1200                     (struct pf_addr *)&h->ip_src.s_addr, AF_INET,
1201                     r->src.neg, kif, M_GETFIB(m)))
1202                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
1203                 else if (PF_MISMATCHAW(&r->dst.addr,
1204                     (struct pf_addr *)&h->ip_dst.s_addr, AF_INET,
1205                     r->dst.neg, NULL, M_GETFIB(m)))
1206                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
1207                 else if (r->match_tag && !pf_match_tag(m, r, &tag,
1208                     pd->pf_mtag ? pd->pf_mtag->tag : 0))
1209                         r = TAILQ_NEXT(r, entries);
1210                 else
1211                         break;
1212         }
1213
1214         if (r == NULL || r->action == PF_NOSCRUB)
1215                 return (PF_PASS);
1216         else {
1217                 r->packets[dir == PF_OUT]++;
1218                 r->bytes[dir == PF_OUT] += pd->tot_len;
1219         }
1220
1221         /* Check for illegal packets */
1222         if (hlen < (int)sizeof(struct ip))
1223                 goto drop;
1224
1225         if (hlen > ntohs(h->ip_len))
1226                 goto drop;
1227
1228         /* Clear IP_DF if the rule uses the no-df option */
1229         if (r->rule_flag & PFRULE_NODF && h->ip_off & htons(IP_DF)) {
1230                 u_int16_t ip_off = h->ip_off;
1231
1232                 h->ip_off &= htons(~IP_DF);
1233                 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
1234         }
1235
1236         /* We will need other tests here */
1237         if (!fragoff && !mff)
1238                 goto no_fragment;
1239
1240         /* We're dealing with a fragment now. Don't allow fragments
1241          * with IP_DF to enter the cache. If the flag was cleared by
1242          * no-df above, fine. Otherwise drop it.
1243          */
1244         if (h->ip_off & htons(IP_DF)) {
1245                 DPFPRINTF(("IP_DF\n"));
1246                 goto bad;
1247         }
1248
1249         ip_len = ntohs(h->ip_len) - hlen;
1250         ip_off = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
1251
1252         /* All fragments are 8 byte aligned */
1253         if (mff && (ip_len & 0x7)) {
1254                 DPFPRINTF(("mff and %d\n", ip_len));
1255                 goto bad;
1256         }
1257
1258         /* Respect maximum length */
1259         if (fragoff + ip_len > IP_MAXPACKET) {
1260                 DPFPRINTF(("max packet %d\n", fragoff + ip_len));
1261                 goto bad;
1262         }
1263         max = fragoff + ip_len;
1264
1265         if ((r->rule_flag & (PFRULE_FRAGCROP|PFRULE_FRAGDROP)) == 0) {
1266
1267                 /* Fully buffer all of the fragments */
1268                 PF_FRAG_LOCK();
1269
1270                 pf_ip2key(h, dir, &key);
1271                 frag = pf_find_fragment(&key, &V_pf_frag_tree);
1272
1273                 /* Check if we saw the last fragment already */
1274                 if (frag != NULL && (frag->fr_flags & PFFRAG_SEENLAST) &&
1275                     max > frag->fr_max)
1276                         goto bad;
1277
1278                 /* Might return a completely reassembled mbuf, or NULL */
1279                 DPFPRINTF(("reass frag %d @ %d-%d\n", h->ip_id, fragoff, max));
1280                 verdict = pf_reassemble(m0, h, dir, reason);
1281                 PF_FRAG_UNLOCK();
1282
1283                 if (verdict != PF_PASS)
1284                         return (PF_DROP);
1285
1286                 m = *m0;
1287                 if (m == NULL)
1288                         return (PF_DROP);
1289
1290                 /* use mtag from concatenated mbuf chain */
1291                 pd->pf_mtag = pf_find_mtag(m);
1292 #ifdef DIAGNOSTIC
1293                 if (pd->pf_mtag == NULL) {
1294                         printf("%s: pf_find_mtag returned NULL(1)\n", __func__);
1295                         if ((pd->pf_mtag = pf_get_mtag(m)) == NULL) {
1296                                 m_freem(m);
1297                                 *m0 = NULL;
1298                                 goto no_mem;
1299                         }
1300                 }
1301 #endif
1302                 h = mtod(m, struct ip *);
1303         } else {
1304                 /* non-buffering fragment cache (drops or masks overlaps) */
1305                 int     nomem = 0;
1306
1307                 if (dir == PF_OUT && pd->pf_mtag->flags & PF_TAG_FRAGCACHE) {
1308                         /*
1309                          * Already passed the fragment cache in the
1310                          * input direction.  If we continued, it would
1311                          * appear to be a dup and would be dropped.
1312                          */
1313                         goto fragment_pass;
1314                 }
1315
1316                 PF_FRAG_LOCK();
1317                 pf_ip2key(h, dir, &key);
1318                 frag = pf_find_fragment(&key, &V_pf_cache_tree);
1319
1320                 /* Check if we saw the last fragment already */
1321                 if (frag != NULL && (frag->fr_flags & PFFRAG_SEENLAST) &&
1322                     max > frag->fr_max) {
1323                         if (r->rule_flag & PFRULE_FRAGDROP)
1324                                 frag->fr_flags |= PFFRAG_DROP;
1325                         goto bad;
1326                 }
1327
1328                 *m0 = m = pf_fragcache(m0, h, &frag, mff,
1329                     (r->rule_flag & PFRULE_FRAGDROP) ? 1 : 0, &nomem);
1330                 PF_FRAG_UNLOCK();
1331                 if (m == NULL) {
1332                         if (nomem)
1333                                 goto no_mem;
1334                         goto drop;
1335                 }
1336
1337                 /* use mtag from copied and trimmed mbuf chain */
1338                 pd->pf_mtag = pf_find_mtag(m);
1339 #ifdef DIAGNOSTIC
1340                 if (pd->pf_mtag == NULL) {
1341                         printf("%s: pf_find_mtag returned NULL(2)\n", __func__);
1342                         if ((pd->pf_mtag = pf_get_mtag(m)) == NULL) {
1343                                 m_freem(m);
1344                                 *m0 = NULL;
1345                                 goto no_mem;
1346                         }
1347                 }
1348 #endif
1349                 if (dir == PF_IN)
1350                         pd->pf_mtag->flags |= PF_TAG_FRAGCACHE;
1351
1352                 if (frag != NULL && (frag->fr_flags & PFFRAG_DROP))
1353                         goto drop;
1354                 goto fragment_pass;
1355         }
1356
1357  no_fragment:
1358         /* At this point, only IP_DF is allowed in ip_off */
1359         if (h->ip_off & ~htons(IP_DF)) {
1360                 u_int16_t ip_off = h->ip_off;
1361
1362                 h->ip_off &= htons(IP_DF);
1363                 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
1364         }
1365
1366         /* not missing a return here */
1367
1368  fragment_pass:
1369         pf_scrub_ip(&m, r->rule_flag, r->min_ttl, r->set_tos);
1370
1371         if ((r->rule_flag & (PFRULE_FRAGCROP|PFRULE_FRAGDROP)) == 0)
1372                 pd->flags |= PFDESC_IP_REAS;
1373         return (PF_PASS);
1374
1375  no_mem:
1376         REASON_SET(reason, PFRES_MEMORY);
1377         if (r != NULL && r->log)
1378                 PFLOG_PACKET(kif, m, AF_INET, dir, *reason, r, NULL, NULL, pd,
1379                     1);
1380         return (PF_DROP);
1381
1382  drop:
1383         REASON_SET(reason, PFRES_NORM);
1384         if (r != NULL && r->log)
1385                 PFLOG_PACKET(kif, m, AF_INET, dir, *reason, r, NULL, NULL, pd,
1386                     1);
1387         return (PF_DROP);
1388
1389  bad:
1390         DPFPRINTF(("dropping bad fragment\n"));
1391
1392         /* Free associated fragments */
1393         if (frag != NULL) {
1394                 pf_free_fragment(frag);
1395                 PF_FRAG_UNLOCK();
1396         }
1397
1398         REASON_SET(reason, PFRES_FRAG);
1399         if (r != NULL && r->log)
1400                 PFLOG_PACKET(kif, m, AF_INET, dir, *reason, r, NULL, NULL, pd,
1401                     1);
1402
1403         return (PF_DROP);
1404 }
1405 #endif
1406
1407 #ifdef INET6
1408 int
1409 pf_normalize_ip6(struct mbuf **m0, int dir, struct pfi_kif *kif,
1410     u_short *reason, struct pf_pdesc *pd)
1411 {
1412         struct mbuf             *m = *m0;
1413         struct pf_rule          *r;
1414         struct ip6_hdr          *h = mtod(m, struct ip6_hdr *);
1415         int                      extoff;
1416         int                      off;
1417         struct ip6_ext           ext;
1418         struct ip6_opt           opt;
1419         struct ip6_opt_jumbo     jumbo;
1420         struct ip6_frag          frag;
1421         u_int32_t                jumbolen = 0, plen;
1422         int                      optend;
1423         int                      ooff;
1424         u_int8_t                 proto;
1425         int                      terminal;
1426
1427         PF_RULES_RASSERT();
1428
1429         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1430         while (r != NULL) {
1431                 r->evaluations++;
1432                 if (pfi_kif_match(r->kif, kif) == r->ifnot)
1433                         r = r->skip[PF_SKIP_IFP].ptr;
1434                 else if (r->direction && r->direction != dir)
1435                         r = r->skip[PF_SKIP_DIR].ptr;
1436                 else if (r->af && r->af != AF_INET6)
1437                         r = r->skip[PF_SKIP_AF].ptr;
1438 #if 0 /* header chain! */
1439                 else if (r->proto && r->proto != h->ip6_nxt)
1440                         r = r->skip[PF_SKIP_PROTO].ptr;
1441 #endif
1442                 else if (PF_MISMATCHAW(&r->src.addr,
1443                     (struct pf_addr *)&h->ip6_src, AF_INET6,
1444                     r->src.neg, kif, M_GETFIB(m)))
1445                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
1446                 else if (PF_MISMATCHAW(&r->dst.addr,
1447                     (struct pf_addr *)&h->ip6_dst, AF_INET6,
1448                     r->dst.neg, NULL, M_GETFIB(m)))
1449                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
1450                 else
1451                         break;
1452         }
1453
1454         if (r == NULL || r->action == PF_NOSCRUB)
1455                 return (PF_PASS);
1456         else {
1457                 r->packets[dir == PF_OUT]++;
1458                 r->bytes[dir == PF_OUT] += pd->tot_len;
1459         }
1460
1461         /* Check for illegal packets */
1462         if (sizeof(struct ip6_hdr) + IPV6_MAXPACKET < m->m_pkthdr.len)
1463                 goto drop;
1464
1465         extoff = 0;
1466         off = sizeof(struct ip6_hdr);
1467         proto = h->ip6_nxt;
1468         terminal = 0;
1469         do {
1470                 switch (proto) {
1471                 case IPPROTO_FRAGMENT:
1472                         goto fragment;
1473                         break;
1474                 case IPPROTO_AH:
1475                 case IPPROTO_ROUTING:
1476                 case IPPROTO_DSTOPTS:
1477                         if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL,
1478                             NULL, AF_INET6))
1479                                 goto shortpkt;
1480                         extoff = off;
1481                         if (proto == IPPROTO_AH)
1482                                 off += (ext.ip6e_len + 2) * 4;
1483                         else
1484                                 off += (ext.ip6e_len + 1) * 8;
1485                         proto = ext.ip6e_nxt;
1486                         break;
1487                 case IPPROTO_HOPOPTS:
1488                         if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL,
1489                             NULL, AF_INET6))
1490                                 goto shortpkt;
1491                         extoff = off;
1492                         optend = off + (ext.ip6e_len + 1) * 8;
1493                         ooff = off + sizeof(ext);
1494                         do {
1495                                 if (!pf_pull_hdr(m, ooff, &opt.ip6o_type,
1496                                     sizeof(opt.ip6o_type), NULL, NULL,
1497                                     AF_INET6))
1498                                         goto shortpkt;
1499                                 if (opt.ip6o_type == IP6OPT_PAD1) {
1500                                         ooff++;
1501                                         continue;
1502                                 }
1503                                 if (!pf_pull_hdr(m, ooff, &opt, sizeof(opt),
1504                                     NULL, NULL, AF_INET6))
1505                                         goto shortpkt;
1506                                 if (ooff + sizeof(opt) + opt.ip6o_len > optend)
1507                                         goto drop;
1508                                 switch (opt.ip6o_type) {
1509                                 case IP6OPT_JUMBO:
1510                                         if (h->ip6_plen != 0)
1511                                                 goto drop;
1512                                         if (!pf_pull_hdr(m, ooff, &jumbo,
1513                                             sizeof(jumbo), NULL, NULL,
1514                                             AF_INET6))
1515                                                 goto shortpkt;
1516                                         memcpy(&jumbolen, jumbo.ip6oj_jumbo_len,
1517                                             sizeof(jumbolen));
1518                                         jumbolen = ntohl(jumbolen);
1519                                         if (jumbolen <= IPV6_MAXPACKET)
1520                                                 goto drop;
1521                                         if (sizeof(struct ip6_hdr) + jumbolen !=
1522                                             m->m_pkthdr.len)
1523                                                 goto drop;
1524                                         break;
1525                                 default:
1526                                         break;
1527                                 }
1528                                 ooff += sizeof(opt) + opt.ip6o_len;
1529                         } while (ooff < optend);
1530
1531                         off = optend;
1532                         proto = ext.ip6e_nxt;
1533                         break;
1534                 default:
1535                         terminal = 1;
1536                         break;
1537                 }
1538         } while (!terminal);
1539
1540         /* jumbo payload option must be present, or plen > 0 */
1541         if (ntohs(h->ip6_plen) == 0)
1542                 plen = jumbolen;
1543         else
1544                 plen = ntohs(h->ip6_plen);
1545         if (plen == 0)
1546                 goto drop;
1547         if (sizeof(struct ip6_hdr) + plen > m->m_pkthdr.len)
1548                 goto shortpkt;
1549
1550         pf_scrub_ip6(&m, r->min_ttl);
1551
1552         return (PF_PASS);
1553
1554  fragment:
1555         /* Jumbo payload packets cannot be fragmented. */
1556         plen = ntohs(h->ip6_plen);
1557         if (plen == 0 || jumbolen)
1558                 goto drop;
1559         if (sizeof(struct ip6_hdr) + plen > m->m_pkthdr.len)
1560                 goto shortpkt;
1561
1562         if (!pf_pull_hdr(m, off, &frag, sizeof(frag), NULL, NULL, AF_INET6))
1563                 goto shortpkt;
1564
1565         /* Offset now points to data portion. */
1566         off += sizeof(frag);
1567
1568         /* Returns PF_DROP or *m0 is NULL or completely reassembled mbuf. */
1569         if (pf_reassemble6(m0, h, &frag, off, extoff, reason) != PF_PASS)
1570                 return (PF_DROP);
1571         m = *m0;
1572         if (m == NULL)
1573                 return (PF_DROP);
1574
1575         pd->flags |= PFDESC_IP_REAS;
1576         return (PF_PASS);
1577
1578  shortpkt:
1579         REASON_SET(reason, PFRES_SHORT);
1580         if (r != NULL && r->log)
1581                 PFLOG_PACKET(kif, m, AF_INET6, dir, *reason, r, NULL, NULL, pd,
1582                     1);
1583         return (PF_DROP);
1584
1585  drop:
1586         REASON_SET(reason, PFRES_NORM);
1587         if (r != NULL && r->log)
1588                 PFLOG_PACKET(kif, m, AF_INET6, dir, *reason, r, NULL, NULL, pd,
1589                     1);
1590         return (PF_DROP);
1591 }
1592 #endif /* INET6 */
1593
1594 int
1595 pf_normalize_tcp(int dir, struct pfi_kif *kif, struct mbuf *m, int ipoff,
1596     int off, void *h, struct pf_pdesc *pd)
1597 {
1598         struct pf_rule  *r, *rm = NULL;
1599         struct tcphdr   *th = pd->hdr.tcp;
1600         int              rewrite = 0;
1601         u_short          reason;
1602         u_int8_t         flags;
1603         sa_family_t      af = pd->af;
1604
1605         PF_RULES_RASSERT();
1606
1607         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1608         while (r != NULL) {
1609                 r->evaluations++;
1610                 if (pfi_kif_match(r->kif, kif) == r->ifnot)
1611                         r = r->skip[PF_SKIP_IFP].ptr;
1612                 else if (r->direction && r->direction != dir)
1613                         r = r->skip[PF_SKIP_DIR].ptr;
1614                 else if (r->af && r->af != af)
1615                         r = r->skip[PF_SKIP_AF].ptr;
1616                 else if (r->proto && r->proto != pd->proto)
1617                         r = r->skip[PF_SKIP_PROTO].ptr;
1618                 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
1619                     r->src.neg, kif, M_GETFIB(m)))
1620                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
1621                 else if (r->src.port_op && !pf_match_port(r->src.port_op,
1622                             r->src.port[0], r->src.port[1], th->th_sport))
1623                         r = r->skip[PF_SKIP_SRC_PORT].ptr;
1624                 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
1625                     r->dst.neg, NULL, M_GETFIB(m)))
1626                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
1627                 else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
1628                             r->dst.port[0], r->dst.port[1], th->th_dport))
1629                         r = r->skip[PF_SKIP_DST_PORT].ptr;
1630                 else if (r->os_fingerprint != PF_OSFP_ANY && !pf_osfp_match(
1631                             pf_osfp_fingerprint(pd, m, off, th),
1632                             r->os_fingerprint))
1633                         r = TAILQ_NEXT(r, entries);
1634                 else {
1635                         rm = r;
1636                         break;
1637                 }
1638         }
1639
1640         if (rm == NULL || rm->action == PF_NOSCRUB)
1641                 return (PF_PASS);
1642         else {
1643                 r->packets[dir == PF_OUT]++;
1644                 r->bytes[dir == PF_OUT] += pd->tot_len;
1645         }
1646
1647         if (rm->rule_flag & PFRULE_REASSEMBLE_TCP)
1648                 pd->flags |= PFDESC_TCP_NORM;
1649
1650         flags = th->th_flags;
1651         if (flags & TH_SYN) {
1652                 /* Illegal packet */
1653                 if (flags & TH_RST)
1654                         goto tcp_drop;
1655
1656                 if (flags & TH_FIN)
1657                         goto tcp_drop;
1658         } else {
1659                 /* Illegal packet */
1660                 if (!(flags & (TH_ACK|TH_RST)))
1661                         goto tcp_drop;
1662         }
1663
1664         if (!(flags & TH_ACK)) {
1665                 /* These flags are only valid if ACK is set */
1666                 if ((flags & TH_FIN) || (flags & TH_PUSH) || (flags & TH_URG))
1667                         goto tcp_drop;
1668         }
1669
1670         /* Check for illegal header length */
1671         if (th->th_off < (sizeof(struct tcphdr) >> 2))
1672                 goto tcp_drop;
1673
1674         /* If flags changed, or reserved data set, then adjust */
1675         if (flags != th->th_flags || th->th_x2 != 0) {
1676                 u_int16_t       ov, nv;
1677
1678                 ov = *(u_int16_t *)(&th->th_ack + 1);
1679                 th->th_flags = flags;
1680                 th->th_x2 = 0;
1681                 nv = *(u_int16_t *)(&th->th_ack + 1);
1682
1683                 th->th_sum = pf_cksum_fixup(th->th_sum, ov, nv, 0);
1684                 rewrite = 1;
1685         }
1686
1687         /* Remove urgent pointer, if TH_URG is not set */
1688         if (!(flags & TH_URG) && th->th_urp) {
1689                 th->th_sum = pf_cksum_fixup(th->th_sum, th->th_urp, 0, 0);
1690                 th->th_urp = 0;
1691                 rewrite = 1;
1692         }
1693
1694         /* Process options */
1695         if (r->max_mss && pf_normalize_tcpopt(r, m, th, off, pd->af))
1696                 rewrite = 1;
1697
1698         /* copy back packet headers if we sanitized */
1699         if (rewrite)
1700                 m_copyback(m, off, sizeof(*th), (caddr_t)th);
1701
1702         return (PF_PASS);
1703
1704  tcp_drop:
1705         REASON_SET(&reason, PFRES_NORM);
1706         if (rm != NULL && r->log)
1707                 PFLOG_PACKET(kif, m, AF_INET, dir, reason, r, NULL, NULL, pd,
1708                     1);
1709         return (PF_DROP);
1710 }
1711
1712 int
1713 pf_normalize_tcp_init(struct mbuf *m, int off, struct pf_pdesc *pd,
1714     struct tcphdr *th, struct pf_state_peer *src, struct pf_state_peer *dst)
1715 {
1716         u_int32_t tsval, tsecr;
1717         u_int8_t hdr[60];
1718         u_int8_t *opt;
1719
1720         KASSERT((src->scrub == NULL),
1721             ("pf_normalize_tcp_init: src->scrub != NULL"));
1722
1723         src->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT);
1724         if (src->scrub == NULL)
1725                 return (1);
1726
1727         switch (pd->af) {
1728 #ifdef INET
1729         case AF_INET: {
1730                 struct ip *h = mtod(m, struct ip *);
1731                 src->scrub->pfss_ttl = h->ip_ttl;
1732                 break;
1733         }
1734 #endif /* INET */
1735 #ifdef INET6
1736         case AF_INET6: {
1737                 struct ip6_hdr *h = mtod(m, struct ip6_hdr *);
1738                 src->scrub->pfss_ttl = h->ip6_hlim;
1739                 break;
1740         }
1741 #endif /* INET6 */
1742         }
1743
1744
1745         /*
1746          * All normalizations below are only begun if we see the start of
1747          * the connections.  They must all set an enabled bit in pfss_flags
1748          */
1749         if ((th->th_flags & TH_SYN) == 0)
1750                 return (0);
1751
1752
1753         if (th->th_off > (sizeof(struct tcphdr) >> 2) && src->scrub &&
1754             pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) {
1755                 /* Diddle with TCP options */
1756                 int hlen;
1757                 opt = hdr + sizeof(struct tcphdr);
1758                 hlen = (th->th_off << 2) - sizeof(struct tcphdr);
1759                 while (hlen >= TCPOLEN_TIMESTAMP) {
1760                         switch (*opt) {
1761                         case TCPOPT_EOL:        /* FALLTHROUGH */
1762                         case TCPOPT_NOP:
1763                                 opt++;
1764                                 hlen--;
1765                                 break;
1766                         case TCPOPT_TIMESTAMP:
1767                                 if (opt[1] >= TCPOLEN_TIMESTAMP) {
1768                                         src->scrub->pfss_flags |=
1769                                             PFSS_TIMESTAMP;
1770                                         src->scrub->pfss_ts_mod =
1771                                             htonl(arc4random());
1772
1773                                         /* note PFSS_PAWS not set yet */
1774                                         memcpy(&tsval, &opt[2],
1775                                             sizeof(u_int32_t));
1776                                         memcpy(&tsecr, &opt[6],
1777                                             sizeof(u_int32_t));
1778                                         src->scrub->pfss_tsval0 = ntohl(tsval);
1779                                         src->scrub->pfss_tsval = ntohl(tsval);
1780                                         src->scrub->pfss_tsecr = ntohl(tsecr);
1781                                         getmicrouptime(&src->scrub->pfss_last);
1782                                 }
1783                                 /* FALLTHROUGH */
1784                         default:
1785                                 hlen -= MAX(opt[1], 2);
1786                                 opt += MAX(opt[1], 2);
1787                                 break;
1788                         }
1789                 }
1790         }
1791
1792         return (0);
1793 }
1794
1795 void
1796 pf_normalize_tcp_cleanup(struct pf_state *state)
1797 {
1798         if (state->src.scrub)
1799                 uma_zfree(V_pf_state_scrub_z, state->src.scrub);
1800         if (state->dst.scrub)
1801                 uma_zfree(V_pf_state_scrub_z, state->dst.scrub);
1802
1803         /* Someday... flush the TCP segment reassembly descriptors. */
1804 }
1805
1806 int
1807 pf_normalize_tcp_stateful(struct mbuf *m, int off, struct pf_pdesc *pd,
1808     u_short *reason, struct tcphdr *th, struct pf_state *state,
1809     struct pf_state_peer *src, struct pf_state_peer *dst, int *writeback)
1810 {
1811         struct timeval uptime;
1812         u_int32_t tsval, tsecr;
1813         u_int tsval_from_last;
1814         u_int8_t hdr[60];
1815         u_int8_t *opt;
1816         int copyback = 0;
1817         int got_ts = 0;
1818
1819         KASSERT((src->scrub || dst->scrub),
1820             ("%s: src->scrub && dst->scrub!", __func__));
1821
1822         /*
1823          * Enforce the minimum TTL seen for this connection.  Negate a common
1824          * technique to evade an intrusion detection system and confuse
1825          * firewall state code.
1826          */
1827         switch (pd->af) {
1828 #ifdef INET
1829         case AF_INET: {
1830                 if (src->scrub) {
1831                         struct ip *h = mtod(m, struct ip *);
1832                         if (h->ip_ttl > src->scrub->pfss_ttl)
1833                                 src->scrub->pfss_ttl = h->ip_ttl;
1834                         h->ip_ttl = src->scrub->pfss_ttl;
1835                 }
1836                 break;
1837         }
1838 #endif /* INET */
1839 #ifdef INET6
1840         case AF_INET6: {
1841                 if (src->scrub) {
1842                         struct ip6_hdr *h = mtod(m, struct ip6_hdr *);
1843                         if (h->ip6_hlim > src->scrub->pfss_ttl)
1844                                 src->scrub->pfss_ttl = h->ip6_hlim;
1845                         h->ip6_hlim = src->scrub->pfss_ttl;
1846                 }
1847                 break;
1848         }
1849 #endif /* INET6 */
1850         }
1851
1852         if (th->th_off > (sizeof(struct tcphdr) >> 2) &&
1853             ((src->scrub && (src->scrub->pfss_flags & PFSS_TIMESTAMP)) ||
1854             (dst->scrub && (dst->scrub->pfss_flags & PFSS_TIMESTAMP))) &&
1855             pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) {
1856                 /* Diddle with TCP options */
1857                 int hlen;
1858                 opt = hdr + sizeof(struct tcphdr);
1859                 hlen = (th->th_off << 2) - sizeof(struct tcphdr);
1860                 while (hlen >= TCPOLEN_TIMESTAMP) {
1861                         switch (*opt) {
1862                         case TCPOPT_EOL:        /* FALLTHROUGH */
1863                         case TCPOPT_NOP:
1864                                 opt++;
1865                                 hlen--;
1866                                 break;
1867                         case TCPOPT_TIMESTAMP:
1868                                 /* Modulate the timestamps.  Can be used for
1869                                  * NAT detection, OS uptime determination or
1870                                  * reboot detection.
1871                                  */
1872
1873                                 if (got_ts) {
1874                                         /* Huh?  Multiple timestamps!? */
1875                                         if (V_pf_status.debug >= PF_DEBUG_MISC) {
1876                                                 DPFPRINTF(("multiple TS??"));
1877                                                 pf_print_state(state);
1878                                                 printf("\n");
1879                                         }
1880                                         REASON_SET(reason, PFRES_TS);
1881                                         return (PF_DROP);
1882                                 }
1883                                 if (opt[1] >= TCPOLEN_TIMESTAMP) {
1884                                         memcpy(&tsval, &opt[2],
1885                                             sizeof(u_int32_t));
1886                                         if (tsval && src->scrub &&
1887                                             (src->scrub->pfss_flags &
1888                                             PFSS_TIMESTAMP)) {
1889                                                 tsval = ntohl(tsval);
1890                                                 pf_change_a(&opt[2],
1891                                                     &th->th_sum,
1892                                                     htonl(tsval +
1893                                                     src->scrub->pfss_ts_mod),
1894                                                     0);
1895                                                 copyback = 1;
1896                                         }
1897
1898                                         /* Modulate TS reply iff valid (!0) */
1899                                         memcpy(&tsecr, &opt[6],
1900                                             sizeof(u_int32_t));
1901                                         if (tsecr && dst->scrub &&
1902                                             (dst->scrub->pfss_flags &
1903                                             PFSS_TIMESTAMP)) {
1904                                                 tsecr = ntohl(tsecr)
1905                                                     - dst->scrub->pfss_ts_mod;
1906                                                 pf_change_a(&opt[6],
1907                                                     &th->th_sum, htonl(tsecr),
1908                                                     0);
1909                                                 copyback = 1;
1910                                         }
1911                                         got_ts = 1;
1912                                 }
1913                                 /* FALLTHROUGH */
1914                         default:
1915                                 hlen -= MAX(opt[1], 2);
1916                                 opt += MAX(opt[1], 2);
1917                                 break;
1918                         }
1919                 }
1920                 if (copyback) {
1921                         /* Copyback the options, caller copys back header */
1922                         *writeback = 1;
1923                         m_copyback(m, off + sizeof(struct tcphdr),
1924                             (th->th_off << 2) - sizeof(struct tcphdr), hdr +
1925                             sizeof(struct tcphdr));
1926                 }
1927         }
1928
1929
1930         /*
1931          * Must invalidate PAWS checks on connections idle for too long.
1932          * The fastest allowed timestamp clock is 1ms.  That turns out to
1933          * be about 24 days before it wraps.  XXX Right now our lowerbound
1934          * TS echo check only works for the first 12 days of a connection
1935          * when the TS has exhausted half its 32bit space
1936          */
1937 #define TS_MAX_IDLE     (24*24*60*60)
1938 #define TS_MAX_CONN     (12*24*60*60)   /* XXX remove when better tsecr check */
1939
1940         getmicrouptime(&uptime);
1941         if (src->scrub && (src->scrub->pfss_flags & PFSS_PAWS) &&
1942             (uptime.tv_sec - src->scrub->pfss_last.tv_sec > TS_MAX_IDLE ||
1943             time_uptime - state->creation > TS_MAX_CONN))  {
1944                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1945                         DPFPRINTF(("src idled out of PAWS\n"));
1946                         pf_print_state(state);
1947                         printf("\n");
1948                 }
1949                 src->scrub->pfss_flags = (src->scrub->pfss_flags & ~PFSS_PAWS)
1950                     | PFSS_PAWS_IDLED;
1951         }
1952         if (dst->scrub && (dst->scrub->pfss_flags & PFSS_PAWS) &&
1953             uptime.tv_sec - dst->scrub->pfss_last.tv_sec > TS_MAX_IDLE) {
1954                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1955                         DPFPRINTF(("dst idled out of PAWS\n"));
1956                         pf_print_state(state);
1957                         printf("\n");
1958                 }
1959                 dst->scrub->pfss_flags = (dst->scrub->pfss_flags & ~PFSS_PAWS)
1960                     | PFSS_PAWS_IDLED;
1961         }
1962
1963         if (got_ts && src->scrub && dst->scrub &&
1964             (src->scrub->pfss_flags & PFSS_PAWS) &&
1965             (dst->scrub->pfss_flags & PFSS_PAWS)) {
1966                 /* Validate that the timestamps are "in-window".
1967                  * RFC1323 describes TCP Timestamp options that allow
1968                  * measurement of RTT (round trip time) and PAWS
1969                  * (protection against wrapped sequence numbers).  PAWS
1970                  * gives us a set of rules for rejecting packets on
1971                  * long fat pipes (packets that were somehow delayed
1972                  * in transit longer than the time it took to send the
1973                  * full TCP sequence space of 4Gb).  We can use these
1974                  * rules and infer a few others that will let us treat
1975                  * the 32bit timestamp and the 32bit echoed timestamp
1976                  * as sequence numbers to prevent a blind attacker from
1977                  * inserting packets into a connection.
1978                  *
1979                  * RFC1323 tells us:
1980                  *  - The timestamp on this packet must be greater than
1981                  *    or equal to the last value echoed by the other
1982                  *    endpoint.  The RFC says those will be discarded
1983                  *    since it is a dup that has already been acked.
1984                  *    This gives us a lowerbound on the timestamp.
1985                  *        timestamp >= other last echoed timestamp
1986                  *  - The timestamp will be less than or equal to
1987                  *    the last timestamp plus the time between the
1988                  *    last packet and now.  The RFC defines the max
1989                  *    clock rate as 1ms.  We will allow clocks to be
1990                  *    up to 10% fast and will allow a total difference
1991                  *    or 30 seconds due to a route change.  And this
1992                  *    gives us an upperbound on the timestamp.
1993                  *        timestamp <= last timestamp + max ticks
1994                  *    We have to be careful here.  Windows will send an
1995                  *    initial timestamp of zero and then initialize it
1996                  *    to a random value after the 3whs; presumably to
1997                  *    avoid a DoS by having to call an expensive RNG
1998                  *    during a SYN flood.  Proof MS has at least one
1999                  *    good security geek.
2000                  *
2001                  *  - The TCP timestamp option must also echo the other
2002                  *    endpoints timestamp.  The timestamp echoed is the
2003                  *    one carried on the earliest unacknowledged segment
2004                  *    on the left edge of the sequence window.  The RFC
2005                  *    states that the host will reject any echoed
2006                  *    timestamps that were larger than any ever sent.
2007                  *    This gives us an upperbound on the TS echo.
2008                  *        tescr <= largest_tsval
2009                  *  - The lowerbound on the TS echo is a little more
2010                  *    tricky to determine.  The other endpoint's echoed
2011                  *    values will not decrease.  But there may be
2012                  *    network conditions that re-order packets and
2013                  *    cause our view of them to decrease.  For now the
2014                  *    only lowerbound we can safely determine is that
2015                  *    the TS echo will never be less than the original
2016                  *    TS.  XXX There is probably a better lowerbound.
2017                  *    Remove TS_MAX_CONN with better lowerbound check.
2018                  *        tescr >= other original TS
2019                  *
2020                  * It is also important to note that the fastest
2021                  * timestamp clock of 1ms will wrap its 32bit space in
2022                  * 24 days.  So we just disable TS checking after 24
2023                  * days of idle time.  We actually must use a 12d
2024                  * connection limit until we can come up with a better
2025                  * lowerbound to the TS echo check.
2026                  */
2027                 struct timeval delta_ts;
2028                 int ts_fudge;
2029
2030
2031                 /*
2032                  * PFTM_TS_DIFF is how many seconds of leeway to allow
2033                  * a host's timestamp.  This can happen if the previous
2034                  * packet got delayed in transit for much longer than
2035                  * this packet.
2036                  */
2037                 if ((ts_fudge = state->rule.ptr->timeout[PFTM_TS_DIFF]) == 0)
2038                         ts_fudge = V_pf_default_rule.timeout[PFTM_TS_DIFF];
2039
2040                 /* Calculate max ticks since the last timestamp */
2041 #define TS_MAXFREQ      1100            /* RFC max TS freq of 1Khz + 10% skew */
2042 #define TS_MICROSECS    1000000         /* microseconds per second */
2043                 delta_ts = uptime;
2044                 timevalsub(&delta_ts, &src->scrub->pfss_last);
2045                 tsval_from_last = (delta_ts.tv_sec + ts_fudge) * TS_MAXFREQ;
2046                 tsval_from_last += delta_ts.tv_usec / (TS_MICROSECS/TS_MAXFREQ);
2047
2048                 if ((src->state >= TCPS_ESTABLISHED &&
2049                     dst->state >= TCPS_ESTABLISHED) &&
2050                     (SEQ_LT(tsval, dst->scrub->pfss_tsecr) ||
2051                     SEQ_GT(tsval, src->scrub->pfss_tsval + tsval_from_last) ||
2052                     (tsecr && (SEQ_GT(tsecr, dst->scrub->pfss_tsval) ||
2053                     SEQ_LT(tsecr, dst->scrub->pfss_tsval0))))) {
2054                         /* Bad RFC1323 implementation or an insertion attack.
2055                          *
2056                          * - Solaris 2.6 and 2.7 are known to send another ACK
2057                          *   after the FIN,FIN|ACK,ACK closing that carries
2058                          *   an old timestamp.
2059                          */
2060
2061                         DPFPRINTF(("Timestamp failed %c%c%c%c\n",
2062                             SEQ_LT(tsval, dst->scrub->pfss_tsecr) ? '0' : ' ',
2063                             SEQ_GT(tsval, src->scrub->pfss_tsval +
2064                             tsval_from_last) ? '1' : ' ',
2065                             SEQ_GT(tsecr, dst->scrub->pfss_tsval) ? '2' : ' ',
2066                             SEQ_LT(tsecr, dst->scrub->pfss_tsval0)? '3' : ' '));
2067                         DPFPRINTF((" tsval: %u  tsecr: %u  +ticks: %u  "
2068                             "idle: %jus %lums\n",
2069                             tsval, tsecr, tsval_from_last,
2070                             (uintmax_t)delta_ts.tv_sec,
2071                             delta_ts.tv_usec / 1000));
2072                         DPFPRINTF((" src->tsval: %u  tsecr: %u\n",
2073                             src->scrub->pfss_tsval, src->scrub->pfss_tsecr));
2074                         DPFPRINTF((" dst->tsval: %u  tsecr: %u  tsval0: %u"
2075                             "\n", dst->scrub->pfss_tsval,
2076                             dst->scrub->pfss_tsecr, dst->scrub->pfss_tsval0));
2077                         if (V_pf_status.debug >= PF_DEBUG_MISC) {
2078                                 pf_print_state(state);
2079                                 pf_print_flags(th->th_flags);
2080                                 printf("\n");
2081                         }
2082                         REASON_SET(reason, PFRES_TS);
2083                         return (PF_DROP);
2084                 }
2085
2086                 /* XXX I'd really like to require tsecr but it's optional */
2087
2088         } else if (!got_ts && (th->th_flags & TH_RST) == 0 &&
2089             ((src->state == TCPS_ESTABLISHED && dst->state == TCPS_ESTABLISHED)
2090             || pd->p_len > 0 || (th->th_flags & TH_SYN)) &&
2091             src->scrub && dst->scrub &&
2092             (src->scrub->pfss_flags & PFSS_PAWS) &&
2093             (dst->scrub->pfss_flags & PFSS_PAWS)) {
2094                 /* Didn't send a timestamp.  Timestamps aren't really useful
2095                  * when:
2096                  *  - connection opening or closing (often not even sent).
2097                  *    but we must not let an attacker to put a FIN on a
2098                  *    data packet to sneak it through our ESTABLISHED check.
2099                  *  - on a TCP reset.  RFC suggests not even looking at TS.
2100                  *  - on an empty ACK.  The TS will not be echoed so it will
2101                  *    probably not help keep the RTT calculation in sync and
2102                  *    there isn't as much danger when the sequence numbers
2103                  *    got wrapped.  So some stacks don't include TS on empty
2104                  *    ACKs :-(
2105                  *
2106                  * To minimize the disruption to mostly RFC1323 conformant
2107                  * stacks, we will only require timestamps on data packets.
2108                  *
2109                  * And what do ya know, we cannot require timestamps on data
2110                  * packets.  There appear to be devices that do legitimate
2111                  * TCP connection hijacking.  There are HTTP devices that allow
2112                  * a 3whs (with timestamps) and then buffer the HTTP request.
2113                  * If the intermediate device has the HTTP response cache, it
2114                  * will spoof the response but not bother timestamping its
2115                  * packets.  So we can look for the presence of a timestamp in
2116                  * the first data packet and if there, require it in all future
2117                  * packets.
2118                  */
2119
2120                 if (pd->p_len > 0 && (src->scrub->pfss_flags & PFSS_DATA_TS)) {
2121                         /*
2122                          * Hey!  Someone tried to sneak a packet in.  Or the
2123                          * stack changed its RFC1323 behavior?!?!
2124                          */
2125                         if (V_pf_status.debug >= PF_DEBUG_MISC) {
2126                                 DPFPRINTF(("Did not receive expected RFC1323 "
2127                                     "timestamp\n"));
2128                                 pf_print_state(state);
2129                                 pf_print_flags(th->th_flags);
2130                                 printf("\n");
2131                         }
2132                         REASON_SET(reason, PFRES_TS);
2133                         return (PF_DROP);
2134                 }
2135         }
2136
2137
2138         /*
2139          * We will note if a host sends his data packets with or without
2140          * timestamps.  And require all data packets to contain a timestamp
2141          * if the first does.  PAWS implicitly requires that all data packets be
2142          * timestamped.  But I think there are middle-man devices that hijack
2143          * TCP streams immediately after the 3whs and don't timestamp their
2144          * packets (seen in a WWW accelerator or cache).
2145          */
2146         if (pd->p_len > 0 && src->scrub && (src->scrub->pfss_flags &
2147             (PFSS_TIMESTAMP|PFSS_DATA_TS|PFSS_DATA_NOTS)) == PFSS_TIMESTAMP) {
2148                 if (got_ts)
2149                         src->scrub->pfss_flags |= PFSS_DATA_TS;
2150                 else {
2151                         src->scrub->pfss_flags |= PFSS_DATA_NOTS;
2152                         if (V_pf_status.debug >= PF_DEBUG_MISC && dst->scrub &&
2153                             (dst->scrub->pfss_flags & PFSS_TIMESTAMP)) {
2154                                 /* Don't warn if other host rejected RFC1323 */
2155                                 DPFPRINTF(("Broken RFC1323 stack did not "
2156                                     "timestamp data packet. Disabled PAWS "
2157                                     "security.\n"));
2158                                 pf_print_state(state);
2159                                 pf_print_flags(th->th_flags);
2160                                 printf("\n");
2161                         }
2162                 }
2163         }
2164
2165
2166         /*
2167          * Update PAWS values
2168          */
2169         if (got_ts && src->scrub && PFSS_TIMESTAMP == (src->scrub->pfss_flags &
2170             (PFSS_PAWS_IDLED|PFSS_TIMESTAMP))) {
2171                 getmicrouptime(&src->scrub->pfss_last);
2172                 if (SEQ_GEQ(tsval, src->scrub->pfss_tsval) ||
2173                     (src->scrub->pfss_flags & PFSS_PAWS) == 0)
2174                         src->scrub->pfss_tsval = tsval;
2175
2176                 if (tsecr) {
2177                         if (SEQ_GEQ(tsecr, src->scrub->pfss_tsecr) ||
2178                             (src->scrub->pfss_flags & PFSS_PAWS) == 0)
2179                                 src->scrub->pfss_tsecr = tsecr;
2180
2181                         if ((src->scrub->pfss_flags & PFSS_PAWS) == 0 &&
2182                             (SEQ_LT(tsval, src->scrub->pfss_tsval0) ||
2183                             src->scrub->pfss_tsval0 == 0)) {
2184                                 /* tsval0 MUST be the lowest timestamp */
2185                                 src->scrub->pfss_tsval0 = tsval;
2186                         }
2187
2188                         /* Only fully initialized after a TS gets echoed */
2189                         if ((src->scrub->pfss_flags & PFSS_PAWS) == 0)
2190                                 src->scrub->pfss_flags |= PFSS_PAWS;
2191                 }
2192         }
2193
2194         /* I have a dream....  TCP segment reassembly.... */
2195         return (0);
2196 }
2197
2198 static int
2199 pf_normalize_tcpopt(struct pf_rule *r, struct mbuf *m, struct tcphdr *th,
2200     int off, sa_family_t af)
2201 {
2202         u_int16_t       *mss;
2203         int              thoff;
2204         int              opt, cnt, optlen = 0;
2205         int              rewrite = 0;
2206         u_char           opts[TCP_MAXOLEN];
2207         u_char          *optp = opts;
2208
2209         thoff = th->th_off << 2;
2210         cnt = thoff - sizeof(struct tcphdr);
2211
2212         if (cnt > 0 && !pf_pull_hdr(m, off + sizeof(*th), opts, cnt,
2213             NULL, NULL, af))
2214                 return (rewrite);
2215
2216         for (; cnt > 0; cnt -= optlen, optp += optlen) {
2217                 opt = optp[0];
2218                 if (opt == TCPOPT_EOL)
2219                         break;
2220                 if (opt == TCPOPT_NOP)
2221                         optlen = 1;
2222                 else {
2223                         if (cnt < 2)
2224                                 break;
2225                         optlen = optp[1];
2226                         if (optlen < 2 || optlen > cnt)
2227                                 break;
2228                 }
2229                 switch (opt) {
2230                 case TCPOPT_MAXSEG:
2231                         mss = (u_int16_t *)(optp + 2);
2232                         if ((ntohs(*mss)) > r->max_mss) {
2233                                 th->th_sum = pf_cksum_fixup(th->th_sum,
2234                                     *mss, htons(r->max_mss), 0);
2235                                 *mss = htons(r->max_mss);
2236                                 rewrite = 1;
2237                         }
2238                         break;
2239                 default:
2240                         break;
2241                 }
2242         }
2243
2244         if (rewrite)
2245                 m_copyback(m, off + sizeof(*th), thoff - sizeof(*th), opts);
2246
2247         return (rewrite);
2248 }
2249
2250 #ifdef INET
2251 static void
2252 pf_scrub_ip(struct mbuf **m0, u_int32_t flags, u_int8_t min_ttl, u_int8_t tos)
2253 {
2254         struct mbuf             *m = *m0;
2255         struct ip               *h = mtod(m, struct ip *);
2256
2257         /* Clear IP_DF if no-df was requested */
2258         if (flags & PFRULE_NODF && h->ip_off & htons(IP_DF)) {
2259                 u_int16_t ip_off = h->ip_off;
2260
2261                 h->ip_off &= htons(~IP_DF);
2262                 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
2263         }
2264
2265         /* Enforce a minimum ttl, may cause endless packet loops */
2266         if (min_ttl && h->ip_ttl < min_ttl) {
2267                 u_int16_t ip_ttl = h->ip_ttl;
2268
2269                 h->ip_ttl = min_ttl;
2270                 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_ttl, h->ip_ttl, 0);
2271         }
2272
2273         /* Enforce tos */
2274         if (flags & PFRULE_SET_TOS) {
2275                 u_int16_t       ov, nv;
2276
2277                 ov = *(u_int16_t *)h;
2278                 h->ip_tos = tos;
2279                 nv = *(u_int16_t *)h;
2280
2281                 h->ip_sum = pf_cksum_fixup(h->ip_sum, ov, nv, 0);
2282         }
2283
2284         /* random-id, but not for fragments */
2285         if (flags & PFRULE_RANDOMID && !(h->ip_off & ~htons(IP_DF))) {
2286                 u_int16_t ip_id = h->ip_id;
2287
2288                 h->ip_id = ip_randomid();
2289                 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_id, h->ip_id, 0);
2290         }
2291 }
2292 #endif /* INET */
2293
2294 #ifdef INET6
2295 static void
2296 pf_scrub_ip6(struct mbuf **m0, u_int8_t min_ttl)
2297 {
2298         struct mbuf             *m = *m0;
2299         struct ip6_hdr          *h = mtod(m, struct ip6_hdr *);
2300
2301         /* Enforce a minimum ttl, may cause endless packet loops */
2302         if (min_ttl && h->ip6_hlim < min_ttl)
2303                 h->ip6_hlim = min_ttl;
2304 }
2305 #endif