sys/netpfil/pf/pf_norm.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-2-Clause
   3  *
   4  * Copyright 2001 Niels Provos <provos@citi.umich.edu>
   5  * Copyright 2011 Alexander Bluhm <bluhm@openbsd.org>
   6  * All rights reserved.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  *
  17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
  18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
  19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
  20  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
  21  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
  22  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  23  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  24  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  25  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
  26  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  27  *
  28  *      $OpenBSD: pf_norm.c,v 1.114 2009/01/29 14:11:45 henning Exp $
  29  */
  30
  31 #include <sys/cdefs.h>
  32 __FBSDID("$FreeBSD$");
  33
  34 #include "opt_inet.h"
  35 #include "opt_inet6.h"
  36 #include "opt_pf.h"
  37
  38 #include <sys/param.h>
  39 #include <sys/kernel.h>
  40 #include <sys/lock.h>
  41 #include <sys/mbuf.h>
  42 #include <sys/mutex.h>
  43 #include <sys/refcount.h>
  44 #include <sys/rwlock.h>
  45 #include <sys/socket.h>
  46
  47 #include <net/if.h>
  48 #include <net/vnet.h>
  49 #include <net/pfvar.h>
  50 #include <net/if_pflog.h>
  51
  52 #include <netinet/in.h>
  53 #include <netinet/ip.h>
  54 #include <netinet/ip_var.h>
  55 #include <netinet6/ip6_var.h>
  56 #include <netinet/tcp.h>
  57 #include <netinet/tcp_fsm.h>
  58 #include <netinet/tcp_seq.h>
  59
  60 #ifdef INET6
  61 #include <netinet/ip6.h>
  62 #endif /* INET6 */
  63
  64 struct pf_frent {
  65         TAILQ_ENTRY(pf_frent)   fr_next;
  66         struct mbuf     *fe_m;
  67         uint16_t        fe_hdrlen;      /* ipv4 header length with ip options
  68                                            ipv6, extension, fragment header */
  69         uint16_t        fe_extoff;      /* last extension header offset or 0 */
  70         uint16_t        fe_len;         /* fragment length */
  71         uint16_t        fe_off;         /* fragment offset */
  72         uint16_t        fe_mff;         /* more fragment flag */
  73 };
  74
  75 struct pf_fragment_cmp {
  76         struct pf_addr  frc_src;
  77         struct pf_addr  frc_dst;
  78         uint32_t        frc_id;
  79         sa_family_t     frc_af;
  80         uint8_t         frc_proto;
  81 };
  82
  83 struct pf_fragment {
  84         struct pf_fragment_cmp  fr_key;
  85 #define fr_src  fr_key.frc_src
  86 #define fr_dst  fr_key.frc_dst
  87 #define fr_id   fr_key.frc_id
  88 #define fr_af   fr_key.frc_af
  89 #define fr_proto        fr_key.frc_proto
  90
  91         RB_ENTRY(pf_fragment) fr_entry;
  92         TAILQ_ENTRY(pf_fragment) frag_next;
  93         uint32_t        fr_timeout;
  94         uint16_t        fr_maxlen;      /* maximum length of single fragment */
  95         TAILQ_HEAD(pf_fragq, pf_frent) fr_queue;
  96 };
  97
  98 struct pf_fragment_tag {
  99         uint16_t        ft_hdrlen;      /* header length of reassembled pkt */
 100         uint16_t        ft_extoff;      /* last extension header offset or 0 */
 101         uint16_t        ft_maxlen;      /* maximum fragment payload length */
 102         uint32_t        ft_id;          /* fragment id */
 103 };
 104
 105 static struct mtx pf_frag_mtx;
 106 MTX_SYSINIT(pf_frag_mtx, &pf_frag_mtx, "pf fragments", MTX_DEF);
 107 #define PF_FRAG_LOCK()          mtx_lock(&pf_frag_mtx)
 108 #define PF_FRAG_UNLOCK()        mtx_unlock(&pf_frag_mtx)
 109 #define PF_FRAG_ASSERT()        mtx_assert(&pf_frag_mtx, MA_OWNED)
 110
 111 VNET_DEFINE(uma_zone_t, pf_state_scrub_z);      /* XXX: shared with pfsync */
 112
 113 static VNET_DEFINE(uma_zone_t, pf_frent_z);
 114 #define V_pf_frent_z    VNET(pf_frent_z)
 115 static VNET_DEFINE(uma_zone_t, pf_frag_z);
 116 #define V_pf_frag_z     VNET(pf_frag_z)
 117
 118 TAILQ_HEAD(pf_fragqueue, pf_fragment);
 119 TAILQ_HEAD(pf_cachequeue, pf_fragment);
 120 static VNET_DEFINE(struct pf_fragqueue, pf_fragqueue);
 121 #define V_pf_fragqueue                  VNET(pf_fragqueue)
 122 RB_HEAD(pf_frag_tree, pf_fragment);
 123 static VNET_DEFINE(struct pf_frag_tree, pf_frag_tree);
 124 #define V_pf_frag_tree                  VNET(pf_frag_tree)
 125 static int               pf_frag_compare(struct pf_fragment *,
 126                             struct pf_fragment *);
 127 static RB_PROTOTYPE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
 128 static RB_GENERATE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
 129
 130 static void     pf_flush_fragments(void);
 131 static void     pf_free_fragment(struct pf_fragment *);
 132 static void     pf_remove_fragment(struct pf_fragment *);
 133 static int      pf_normalize_tcpopt(struct pf_rule *, struct mbuf *,
 134                     struct tcphdr *, int, sa_family_t);
 135 static struct pf_frent *pf_create_fragment(u_short *);
 136 static struct pf_fragment *pf_find_fragment(struct pf_fragment_cmp *key,
 137                     struct pf_frag_tree *tree);
 138 static struct pf_fragment *pf_fillup_fragment(struct pf_fragment_cmp *,
 139                     struct pf_frent *, u_short *);
 140 static int      pf_isfull_fragment(struct pf_fragment *);
 141 static struct mbuf *pf_join_fragment(struct pf_fragment *);
 142 #ifdef INET
 143 static void     pf_scrub_ip(struct mbuf **, uint32_t, uint8_t, uint8_t);
 144 static int      pf_reassemble(struct mbuf **, struct ip *, int, u_short *);
 145 #endif  /* INET */
 146 #ifdef INET6
 147 static int      pf_reassemble6(struct mbuf **, struct ip6_hdr *,
 148                     struct ip6_frag *, uint16_t, uint16_t, u_short *);
 149 static void     pf_scrub_ip6(struct mbuf **, uint8_t);
 150 #endif  /* INET6 */
 151
 152 #define DPFPRINTF(x) do {                               \
 153         if (V_pf_status.debug >= PF_DEBUG_MISC) {       \
 154                 printf("%s: ", __func__);               \
 155                 printf x ;                              \
 156         }                                               \
 157 } while(0)
 158
 159 #ifdef INET
 160 static void
 161 pf_ip2key(struct ip *ip, int dir, struct pf_fragment_cmp *key)
 162 {
 163
 164         key->frc_src.v4 = ip->ip_src;
 165         key->frc_dst.v4 = ip->ip_dst;
 166         key->frc_af = AF_INET;
 167         key->frc_proto = ip->ip_p;
 168         key->frc_id = ip->ip_id;
 169 }
 170 #endif  /* INET */
 171
 172 void
 173 pf_normalize_init(void)
 174 {
 175
 176         V_pf_frag_z = uma_zcreate("pf frags", sizeof(struct pf_fragment),
 177             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
 178         V_pf_frent_z = uma_zcreate("pf frag entries", sizeof(struct pf_frent),
 179             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
 180         V_pf_state_scrub_z = uma_zcreate("pf state scrubs",
 181             sizeof(struct pf_state_scrub),  NULL, NULL, NULL, NULL,
 182             UMA_ALIGN_PTR, 0);
 183
 184         V_pf_limits[PF_LIMIT_FRAGS].zone = V_pf_frent_z;
 185         V_pf_limits[PF_LIMIT_FRAGS].limit = PFFRAG_FRENT_HIWAT;
 186         uma_zone_set_max(V_pf_frent_z, PFFRAG_FRENT_HIWAT);
 187         uma_zone_set_warning(V_pf_frent_z, "PF frag entries limit reached");
 188
 189         TAILQ_INIT(&V_pf_fragqueue);
 190 }
 191
 192 void
 193 pf_normalize_cleanup(void)
 194 {
 195
 196         uma_zdestroy(V_pf_state_scrub_z);
 197         uma_zdestroy(V_pf_frent_z);
 198         uma_zdestroy(V_pf_frag_z);
 199 }
 200
 201 static int
 202 pf_frag_compare(struct pf_fragment *a, struct pf_fragment *b)
 203 {
 204         int     diff;
 205
 206         if ((diff = a->fr_id - b->fr_id) != 0)
 207                 return (diff);
 208         if ((diff = a->fr_proto - b->fr_proto) != 0)
 209                 return (diff);
 210         if ((diff = a->fr_af - b->fr_af) != 0)
 211                 return (diff);
 212         if ((diff = pf_addr_cmp(&a->fr_src, &b->fr_src, a->fr_af)) != 0)
 213                 return (diff);
 214         if ((diff = pf_addr_cmp(&a->fr_dst, &b->fr_dst, a->fr_af)) != 0)
 215                 return (diff);
 216         return (0);
 217 }
 218
 219 void
 220 pf_purge_expired_fragments(void)
 221 {
 222         struct pf_fragment      *frag;
 223         u_int32_t                expire = time_uptime -
 224                                     V_pf_default_rule.timeout[PFTM_FRAG];
 225
 226         PF_FRAG_LOCK();
 227         while ((frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue)) != NULL) {
 228                 if (frag->fr_timeout > expire)
 229                         break;
 230
 231                 DPFPRINTF(("expiring %d(%p)\n", frag->fr_id, frag));
 232                 pf_free_fragment(frag);
 233         }
 234
 235         PF_FRAG_UNLOCK();
 236 }
 237
 238 /*
 239  * Try to flush old fragments to make space for new ones
 240  */
 241 static void
 242 pf_flush_fragments(void)
 243 {
 244         struct pf_fragment      *frag;
 245         int                      goal;
 246
 247         PF_FRAG_ASSERT();
 248
 249         goal = uma_zone_get_cur(V_pf_frent_z) * 9 / 10;
 250         DPFPRINTF(("trying to free %d frag entriess\n", goal));
 251         while (goal < uma_zone_get_cur(V_pf_frent_z)) {
 252                 frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue);
 253                 if (frag)
 254                         pf_free_fragment(frag);
 255                 else
 256                         break;
 257         }
 258 }
 259
 260 /* Frees the fragments and all associated entries */
 261 static void
 262 pf_free_fragment(struct pf_fragment *frag)
 263 {
 264         struct pf_frent         *frent;
 265
 266         PF_FRAG_ASSERT();
 267
 268         /* Free all fragments */
 269         for (frent = TAILQ_FIRST(&frag->fr_queue); frent;
 270             frent = TAILQ_FIRST(&frag->fr_queue)) {
 271                 TAILQ_REMOVE(&frag->fr_queue, frent, fr_next);
 272
 273                 m_freem(frent->fe_m);
 274                 uma_zfree(V_pf_frent_z, frent);
 275         }
 276
 277         pf_remove_fragment(frag);
 278 }
 279
 280 static struct pf_fragment *
 281 pf_find_fragment(struct pf_fragment_cmp *key, struct pf_frag_tree *tree)
 282 {
 283         struct pf_fragment      *frag;
 284
 285         PF_FRAG_ASSERT();
 286
 287         frag = RB_FIND(pf_frag_tree, tree, (struct pf_fragment *)key);
 288         if (frag != NULL) {
 289                 /* XXX Are we sure we want to update the timeout? */
 290                 frag->fr_timeout = time_uptime;
 291                 TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next);
 292                 TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next);
 293         }
 294
 295         return (frag);
 296 }
 297
 298 /* Removes a fragment from the fragment queue and frees the fragment */
 299 static void
 300 pf_remove_fragment(struct pf_fragment *frag)
 301 {
 302
 303         PF_FRAG_ASSERT();
 304
 305         RB_REMOVE(pf_frag_tree, &V_pf_frag_tree, frag);
 306         TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next);
 307         uma_zfree(V_pf_frag_z, frag);
 308 }
 309
 310 static struct pf_frent *
 311 pf_create_fragment(u_short *reason)
 312 {
 313         struct pf_frent *frent;
 314
 315         PF_FRAG_ASSERT();
 316
 317         frent = uma_zalloc(V_pf_frent_z, M_NOWAIT);
 318         if (frent == NULL) {
 319                 pf_flush_fragments();
 320                 frent = uma_zalloc(V_pf_frent_z, M_NOWAIT);
 321                 if (frent == NULL) {
 322                         REASON_SET(reason, PFRES_MEMORY);
 323                         return (NULL);
 324                 }
 325         }
 326
 327         return (frent);
 328 }
 329
 330 static struct pf_fragment *
 331 pf_fillup_fragment(struct pf_fragment_cmp *key, struct pf_frent *frent,
 332                 u_short *reason)
 333 {
 334         struct pf_frent         *after, *next, *prev;
 335         struct pf_fragment      *frag;
 336         uint16_t                total;
 337
 338         PF_FRAG_ASSERT();
 339
 340         /* No empty fragments. */
 341         if (frent->fe_len == 0) {
 342                 DPFPRINTF(("bad fragment: len 0"));
 343                 goto bad_fragment;
 344         }
 345
 346         /* All fragments are 8 byte aligned. */
 347         if (frent->fe_mff && (frent->fe_len & 0x7)) {
 348                 DPFPRINTF(("bad fragment: mff and len %d", frent->fe_len));
 349                 goto bad_fragment;
 350         }
 351
 352         /* Respect maximum length, IP_MAXPACKET == IPV6_MAXPACKET. */
 353         if (frent->fe_off + frent->fe_len > IP_MAXPACKET) {
 354                 DPFPRINTF(("bad fragment: max packet %d",
 355                     frent->fe_off + frent->fe_len));
 356                 goto bad_fragment;
 357         }
 358
 359         DPFPRINTF((key->frc_af == AF_INET ?
 360             "reass frag %d @ %d-%d" : "reass frag %#08x @ %d-%d",
 361             key->frc_id, frent->fe_off, frent->fe_off + frent->fe_len));
 362
 363         /* Fully buffer all of the fragments in this fragment queue. */
 364         frag = pf_find_fragment(key, &V_pf_frag_tree);
 365
 366         /* Create a new reassembly queue for this packet. */
 367         if (frag == NULL) {
 368                 frag = uma_zalloc(V_pf_frag_z, M_NOWAIT);
 369                 if (frag == NULL) {
 370                         pf_flush_fragments();
 371                         frag = uma_zalloc(V_pf_frag_z, M_NOWAIT);
 372                         if (frag == NULL) {
 373                                 REASON_SET(reason, PFRES_MEMORY);
 374                                 goto drop_fragment;
 375                         }
 376                 }
 377
 378                 *(struct pf_fragment_cmp *)frag = *key;
 379                 frag->fr_timeout = time_uptime;
 380                 frag->fr_maxlen = frent->fe_len;
 381                 TAILQ_INIT(&frag->fr_queue);
 382
 383                 RB_INSERT(pf_frag_tree, &V_pf_frag_tree, frag);
 384                 TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next);
 385
 386                 /* We do not have a previous fragment. */
 387                 TAILQ_INSERT_HEAD(&frag->fr_queue, frent, fr_next);
 388
 389                 return (frag);
 390         }
 391
 392         KASSERT(!TAILQ_EMPTY(&frag->fr_queue), ("!TAILQ_EMPTY()->fr_queue"));
 393
 394         /* Remember maximum fragment len for refragmentation. */
 395         if (frent->fe_len > frag->fr_maxlen)
 396                 frag->fr_maxlen = frent->fe_len;
 397
 398         /* Maximum data we have seen already. */
 399         total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
 400                 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
 401
 402         /* Non terminal fragments must have more fragments flag. */
 403         if (frent->fe_off + frent->fe_len < total && !frent->fe_mff)
 404                 goto bad_fragment;
 405
 406         /* Check if we saw the last fragment already. */
 407         if (!TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_mff) {
 408                 if (frent->fe_off + frent->fe_len > total ||
 409                     (frent->fe_off + frent->fe_len == total && frent->fe_mff))
 410                         goto bad_fragment;
 411         } else {
 412                 if (frent->fe_off + frent->fe_len == total && !frent->fe_mff)
 413                         goto bad_fragment;
 414         }
 415
 416         /* Find a fragment after the current one. */
 417         prev = NULL;
 418         TAILQ_FOREACH(after, &frag->fr_queue, fr_next) {
 419                 if (after->fe_off > frent->fe_off)
 420                         break;
 421                 prev = after;
 422         }
 423
 424         KASSERT(prev != NULL || after != NULL,
 425             ("prev != NULL || after != NULL"));
 426
 427         if (prev != NULL && prev->fe_off + prev->fe_len > frent->fe_off) {
 428                 uint16_t precut;
 429
 430                 precut = prev->fe_off + prev->fe_len - frent->fe_off;
 431                 if (precut >= frent->fe_len)
 432                         goto bad_fragment;
 433                 DPFPRINTF(("overlap -%d", precut));
 434                 m_adj(frent->fe_m, precut);
 435                 frent->fe_off += precut;
 436                 frent->fe_len -= precut;
 437         }
 438
 439         for (; after != NULL && frent->fe_off + frent->fe_len > after->fe_off;
 440             after = next) {
 441                 uint16_t aftercut;
 442
 443                 aftercut = frent->fe_off + frent->fe_len - after->fe_off;
 444                 DPFPRINTF(("adjust overlap %d", aftercut));
 445                 if (aftercut < after->fe_len) {
 446                         m_adj(after->fe_m, aftercut);
 447                         after->fe_off += aftercut;
 448                         after->fe_len -= aftercut;
 449                         break;
 450                 }
 451
 452                 /* This fragment is completely overlapped, lose it. */
 453                 next = TAILQ_NEXT(after, fr_next);
 454                 m_freem(after->fe_m);
 455                 TAILQ_REMOVE(&frag->fr_queue, after, fr_next);
 456                 uma_zfree(V_pf_frent_z, after);
 457         }
 458
 459         if (prev == NULL)
 460                 TAILQ_INSERT_HEAD(&frag->fr_queue, frent, fr_next);
 461         else
 462                 TAILQ_INSERT_AFTER(&frag->fr_queue, prev, frent, fr_next);
 463
 464         return (frag);
 465
 466 bad_fragment:
 467         REASON_SET(reason, PFRES_FRAG);
 468 drop_fragment:
 469         uma_zfree(V_pf_frent_z, frent);
 470         return (NULL);
 471 }
 472
 473 static int
 474 pf_isfull_fragment(struct pf_fragment *frag)
 475 {
 476         struct pf_frent *frent, *next;
 477         uint16_t off, total;
 478
 479         /* Check if we are completely reassembled */
 480         if (TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_mff)
 481                 return (0);
 482
 483         /* Maximum data we have seen already */
 484         total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
 485                 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
 486
 487         /* Check if we have all the data */
 488         off = 0;
 489         for (frent = TAILQ_FIRST(&frag->fr_queue); frent; frent = next) {
 490                 next = TAILQ_NEXT(frent, fr_next);
 491
 492                 off += frent->fe_len;
 493                 if (off < total && (next == NULL || next->fe_off != off)) {
 494                         DPFPRINTF(("missing fragment at %d, next %d, total %d",
 495                             off, next == NULL ? -1 : next->fe_off, total));
 496                         return (0);
 497                 }
 498         }
 499         DPFPRINTF(("%d < %d?", off, total));
 500         if (off < total)
 501                 return (0);
 502         KASSERT(off == total, ("off == total"));
 503
 504         return (1);
 505 }
 506
 507 static struct mbuf *
 508 pf_join_fragment(struct pf_fragment *frag)
 509 {
 510         struct mbuf *m, *m2;
 511         struct pf_frent *frent, *next;
 512
 513         frent = TAILQ_FIRST(&frag->fr_queue);
 514         next = TAILQ_NEXT(frent, fr_next);
 515
 516         m = frent->fe_m;
 517         m_adj(m, (frent->fe_hdrlen + frent->fe_len) - m->m_pkthdr.len);
 518         uma_zfree(V_pf_frent_z, frent);
 519         for (frent = next; frent != NULL; frent = next) {
 520                 next = TAILQ_NEXT(frent, fr_next);
 521
 522                 m2 = frent->fe_m;
 523                 /* Strip off ip header. */
 524                 m_adj(m2, frent->fe_hdrlen);
 525                 /* Strip off any trailing bytes. */
 526                 m_adj(m2, frent->fe_len - m2->m_pkthdr.len);
 527
 528                 uma_zfree(V_pf_frent_z, frent);
 529                 m_cat(m, m2);
 530         }
 531
 532         /* Remove from fragment queue. */
 533         pf_remove_fragment(frag);
 534
 535         return (m);
 536 }
 537
 538 #ifdef INET
 539 static int
 540 pf_reassemble(struct mbuf **m0, struct ip *ip, int dir, u_short *reason)
 541 {
 542         struct mbuf             *m = *m0;
 543         struct pf_frent         *frent;
 544         struct pf_fragment      *frag;
 545         struct pf_fragment_cmp  key;
 546         uint16_t                total, hdrlen;
 547
 548         /* Get an entry for the fragment queue */
 549         if ((frent = pf_create_fragment(reason)) == NULL)
 550                 return (PF_DROP);
 551
 552         frent->fe_m = m;
 553         frent->fe_hdrlen = ip->ip_hl << 2;
 554         frent->fe_extoff = 0;
 555         frent->fe_len = ntohs(ip->ip_len) - (ip->ip_hl << 2);
 556         frent->fe_off = (ntohs(ip->ip_off) & IP_OFFMASK) << 3;
 557         frent->fe_mff = ntohs(ip->ip_off) & IP_MF;
 558
 559         pf_ip2key(ip, dir, &key);
 560
 561         if ((frag = pf_fillup_fragment(&key, frent, reason)) == NULL)
 562                 return (PF_DROP);
 563
 564         /* The mbuf is part of the fragment entry, no direct free or access */
 565         m = *m0 = NULL;
 566
 567         if (!pf_isfull_fragment(frag))
 568                 return (PF_PASS);  /* drop because *m0 is NULL, no error */
 569
 570         /* We have all the data */
 571         frent = TAILQ_FIRST(&frag->fr_queue);
 572         KASSERT(frent != NULL, ("frent != NULL"));
 573         total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
 574                 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
 575         hdrlen = frent->fe_hdrlen;
 576
 577         m = *m0 = pf_join_fragment(frag);
 578         frag = NULL;
 579
 580         if (m->m_flags & M_PKTHDR) {
 581                 int plen = 0;
 582                 for (m = *m0; m; m = m->m_next)
 583                         plen += m->m_len;
 584                 m = *m0;
 585                 m->m_pkthdr.len = plen;
 586         }
 587
 588         ip = mtod(m, struct ip *);
 589         ip->ip_len = htons(hdrlen + total);
 590         ip->ip_off &= ~(IP_MF|IP_OFFMASK);
 591
 592         if (hdrlen + total > IP_MAXPACKET) {
 593                 DPFPRINTF(("drop: too big: %d", total));
 594                 ip->ip_len = 0;
 595                 REASON_SET(reason, PFRES_SHORT);
 596                 /* PF_DROP requires a valid mbuf *m0 in pf_test() */
 597                 return (PF_DROP);
 598         }
 599
 600         DPFPRINTF(("complete: %p(%d)\n", m, ntohs(ip->ip_len)));
 601         return (PF_PASS);
 602 }
 603 #endif  /* INET */
 604
 605 #ifdef INET6
 606 static int
 607 pf_reassemble6(struct mbuf **m0, struct ip6_hdr *ip6, struct ip6_frag *fraghdr,
 608     uint16_t hdrlen, uint16_t extoff, u_short *reason)
 609 {
 610         struct mbuf             *m = *m0;
 611         struct pf_frent         *frent;
 612         struct pf_fragment      *frag;
 613         struct pf_fragment_cmp   key;
 614         struct m_tag            *mtag;
 615         struct pf_fragment_tag  *ftag;
 616         int                      off;
 617         uint32_t                 frag_id;
 618         uint16_t                 total, maxlen;
 619         uint8_t                  proto;
 620
 621         PF_FRAG_LOCK();
 622
 623         /* Get an entry for the fragment queue. */
 624         if ((frent = pf_create_fragment(reason)) == NULL) {
 625                 PF_FRAG_UNLOCK();
 626                 return (PF_DROP);
 627         }
 628
 629         frent->fe_m = m;
 630         frent->fe_hdrlen = hdrlen;
 631         frent->fe_extoff = extoff;
 632         frent->fe_len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - hdrlen;
 633         frent->fe_off = ntohs(fraghdr->ip6f_offlg & IP6F_OFF_MASK);
 634         frent->fe_mff = fraghdr->ip6f_offlg & IP6F_MORE_FRAG;
 635
 636         key.frc_src.v6 = ip6->ip6_src;
 637         key.frc_dst.v6 = ip6->ip6_dst;
 638         key.frc_af = AF_INET6;
 639         /* Only the first fragment's protocol is relevant. */
 640         key.frc_proto = 0;
 641         key.frc_id = fraghdr->ip6f_ident;
 642
 643         if ((frag = pf_fillup_fragment(&key, frent, reason)) == NULL) {
 644                 PF_FRAG_UNLOCK();
 645                 return (PF_DROP);
 646         }
 647
 648         /* The mbuf is part of the fragment entry, no direct free or access. */
 649         m = *m0 = NULL;
 650
 651         if (!pf_isfull_fragment(frag)) {
 652                 PF_FRAG_UNLOCK();
 653                 return (PF_PASS);  /* Drop because *m0 is NULL, no error. */
 654         }
 655
 656         /* We have all the data. */
 657         extoff = frent->fe_extoff;
 658         maxlen = frag->fr_maxlen;
 659         frag_id = frag->fr_id;
 660         frent = TAILQ_FIRST(&frag->fr_queue);
 661         KASSERT(frent != NULL, ("frent != NULL"));
 662         total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
 663                 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
 664         hdrlen = frent->fe_hdrlen - sizeof(struct ip6_frag);
 665
 666         m = *m0 = pf_join_fragment(frag);
 667         frag = NULL;
 668
 669         PF_FRAG_UNLOCK();
 670
 671         /* Take protocol from first fragment header. */
 672         m = m_getptr(m, hdrlen + offsetof(struct ip6_frag, ip6f_nxt), &off);
 673         KASSERT(m, ("%s: short mbuf chain", __func__));
 674         proto = *(mtod(m, caddr_t) + off);
 675         m = *m0;
 676
 677         /* Delete frag6 header */
 678         if (ip6_deletefraghdr(m, hdrlen, M_NOWAIT) != 0)
 679                 goto fail;
 680
 681         if (m->m_flags & M_PKTHDR) {
 682                 int plen = 0;
 683                 for (m = *m0; m; m = m->m_next)
 684                         plen += m->m_len;
 685                 m = *m0;
 686                 m->m_pkthdr.len = plen;
 687         }
 688
 689         if ((mtag = m_tag_get(PF_REASSEMBLED, sizeof(struct pf_fragment_tag),
 690             M_NOWAIT)) == NULL)
 691                 goto fail;
 692         ftag = (struct pf_fragment_tag *)(mtag + 1);
 693         ftag->ft_hdrlen = hdrlen;
 694         ftag->ft_extoff = extoff;
 695         ftag->ft_maxlen = maxlen;
 696         ftag->ft_id = frag_id;
 697         m_tag_prepend(m, mtag);
 698
 699         ip6 = mtod(m, struct ip6_hdr *);
 700         ip6->ip6_plen = htons(hdrlen - sizeof(struct ip6_hdr) + total);
 701         if (extoff) {
 702                 /* Write protocol into next field of last extension header. */
 703                 m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt),
 704                     &off);
 705                 KASSERT(m, ("%s: short mbuf chain", __func__));
 706                 *(mtod(m, char *) + off) = proto;
 707                 m = *m0;
 708         } else
 709                 ip6->ip6_nxt = proto;
 710
 711         if (hdrlen - sizeof(struct ip6_hdr) + total > IPV6_MAXPACKET) {
 712                 DPFPRINTF(("drop: too big: %d", total));
 713                 ip6->ip6_plen = 0;
 714                 REASON_SET(reason, PFRES_SHORT);
 715                 /* PF_DROP requires a valid mbuf *m0 in pf_test6(). */
 716                 return (PF_DROP);
 717         }
 718
 719         DPFPRINTF(("complete: %p(%d)", m, ntohs(ip6->ip6_plen)));
 720         return (PF_PASS);
 721
 722 fail:
 723         REASON_SET(reason, PFRES_MEMORY);
 724         /* PF_DROP requires a valid mbuf *m0 in pf_test6(), will free later. */
 725         return (PF_DROP);
 726 }
 727 #endif  /* INET6 */
 728
 729 #ifdef INET6
 730 int
 731 pf_refragment6(struct ifnet *ifp, struct mbuf **m0, struct m_tag *mtag)
 732 {
 733         struct mbuf             *m = *m0, *t;
 734         struct pf_fragment_tag  *ftag = (struct pf_fragment_tag *)(mtag + 1);
 735         struct pf_pdesc          pd;
 736         uint32_t                 frag_id;
 737         uint16_t                 hdrlen, extoff, maxlen;
 738         uint8_t                  proto;
 739         int                      error, action;
 740
 741         hdrlen = ftag->ft_hdrlen;
 742         extoff = ftag->ft_extoff;
 743         maxlen = ftag->ft_maxlen;
 744         frag_id = ftag->ft_id;
 745         m_tag_delete(m, mtag);
 746         mtag = NULL;
 747         ftag = NULL;
 748
 749         if (extoff) {
 750                 int off;
 751
 752                 /* Use protocol from next field of last extension header */
 753                 m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt),
 754                     &off);
 755                 KASSERT((m != NULL), ("pf_refragment6: short mbuf chain"));
 756                 proto = *(mtod(m, caddr_t) + off);
 757                 *(mtod(m, char *) + off) = IPPROTO_FRAGMENT;
 758                 m = *m0;
 759         } else {
 760                 struct ip6_hdr *hdr;
 761
 762                 hdr = mtod(m, struct ip6_hdr *);
 763                 proto = hdr->ip6_nxt;
 764                 hdr->ip6_nxt = IPPROTO_FRAGMENT;
 765         }
 766
 767         /* The MTU must be a multiple of 8 bytes, or we risk doing the
 768          * fragmentation wrong. */
 769         maxlen = maxlen & ~7;
 770
 771         /*
 772          * Maxlen may be less than 8 if there was only a single
 773          * fragment.  As it was fragmented before, add a fragment
 774          * header also for a single fragment.  If total or maxlen
 775          * is less than 8, ip6_fragment() will return EMSGSIZE and
 776          * we drop the packet.
 777          */
 778         error = ip6_fragment(ifp, m, hdrlen, proto, maxlen, frag_id);
 779         m = (*m0)->m_nextpkt;
 780         (*m0)->m_nextpkt = NULL;
 781         if (error == 0) {
 782                 /* The first mbuf contains the unfragmented packet. */
 783                 m_freem(*m0);
 784                 *m0 = NULL;
 785                 action = PF_PASS;
 786         } else {
 787                 /* Drop expects an mbuf to free. */
 788                 DPFPRINTF(("refragment error %d", error));
 789                 action = PF_DROP;
 790         }
 791         for (t = m; m; m = t) {
 792                 t = m->m_nextpkt;
 793                 m->m_nextpkt = NULL;
 794                 m->m_flags |= M_SKIP_FIREWALL;
 795                 memset(&pd, 0, sizeof(pd));
 796                 pd.pf_mtag = pf_find_mtag(m);
 797                 if (error == 0)
 798                         ip6_forward(m, 0);
 799                 else
 800                         m_freem(m);
 801         }
 802
 803         return (action);
 804 }
 805 #endif /* INET6 */
 806
 807 #ifdef INET
 808 int
 809 pf_normalize_ip(struct mbuf **m0, int dir, struct pfi_kif *kif, u_short *reason,
 810     struct pf_pdesc *pd)
 811 {
 812         struct mbuf             *m = *m0;
 813         struct pf_rule          *r;
 814         struct ip               *h = mtod(m, struct ip *);
 815         int                      mff = (ntohs(h->ip_off) & IP_MF);
 816         int                      hlen = h->ip_hl << 2;
 817         u_int16_t                fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
 818         u_int16_t                max;
 819         int                      ip_len;
 820         int                      ip_off;
 821         int                      tag = -1;
 822         int                      verdict;
 823
 824         PF_RULES_RASSERT();
 825
 826         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
 827         while (r != NULL) {
 828                 r->evaluations++;
 829                 if (pfi_kif_match(r->kif, kif) == r->ifnot)
 830                         r = r->skip[PF_SKIP_IFP].ptr;
 831                 else if (r->direction && r->direction != dir)
 832                         r = r->skip[PF_SKIP_DIR].ptr;
 833                 else if (r->af && r->af != AF_INET)
 834                         r = r->skip[PF_SKIP_AF].ptr;
 835                 else if (r->proto && r->proto != h->ip_p)
 836                         r = r->skip[PF_SKIP_PROTO].ptr;
 837                 else if (PF_MISMATCHAW(&r->src.addr,
 838                     (struct pf_addr *)&h->ip_src.s_addr, AF_INET,
 839                     r->src.neg, kif, M_GETFIB(m)))
 840                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
 841                 else if (PF_MISMATCHAW(&r->dst.addr,
 842                     (struct pf_addr *)&h->ip_dst.s_addr, AF_INET,
 843                     r->dst.neg, NULL, M_GETFIB(m)))
 844                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
 845                 else if (r->match_tag && !pf_match_tag(m, r, &tag,
 846                     pd->pf_mtag ? pd->pf_mtag->tag : 0))
 847                         r = TAILQ_NEXT(r, entries);
 848                 else
 849                         break;
 850         }
 851
 852         if (r == NULL || r->action == PF_NOSCRUB)
 853                 return (PF_PASS);
 854         else {
 855                 r->packets[dir == PF_OUT]++;
 856                 r->bytes[dir == PF_OUT] += pd->tot_len;
 857         }
 858
 859         /* Check for illegal packets */
 860         if (hlen < (int)sizeof(struct ip)) {
 861                 REASON_SET(reason, PFRES_NORM);
 862                 goto drop;
 863         }
 864
 865         if (hlen > ntohs(h->ip_len)) {
 866                 REASON_SET(reason, PFRES_NORM);
 867                 goto drop;
 868         }
 869
 870         /* Clear IP_DF if the rule uses the no-df option */
 871         if (r->rule_flag & PFRULE_NODF && h->ip_off & htons(IP_DF)) {
 872                 u_int16_t ip_off = h->ip_off;
 873
 874                 h->ip_off &= htons(~IP_DF);
 875                 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
 876         }
 877
 878         /* We will need other tests here */
 879         if (!fragoff && !mff)
 880                 goto no_fragment;
 881
 882         /* We're dealing with a fragment now. Don't allow fragments
 883          * with IP_DF to enter the cache. If the flag was cleared by
 884          * no-df above, fine. Otherwise drop it.
 885          */
 886         if (h->ip_off & htons(IP_DF)) {
 887                 DPFPRINTF(("IP_DF\n"));
 888                 goto bad;
 889         }
 890
 891         ip_len = ntohs(h->ip_len) - hlen;
 892         ip_off = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
 893
 894         /* All fragments are 8 byte aligned */
 895         if (mff && (ip_len & 0x7)) {
 896                 DPFPRINTF(("mff and %d\n", ip_len));
 897                 goto bad;
 898         }
 899
 900         /* Respect maximum length */
 901         if (fragoff + ip_len > IP_MAXPACKET) {
 902                 DPFPRINTF(("max packet %d\n", fragoff + ip_len));
 903                 goto bad;
 904         }
 905         max = fragoff + ip_len;
 906
 907         /* Fully buffer all of the fragments
 908          * Might return a completely reassembled mbuf, or NULL */
 909         PF_FRAG_LOCK();
 910         DPFPRINTF(("reass frag %d @ %d-%d\n", h->ip_id, fragoff, max));
 911         verdict = pf_reassemble(m0, h, dir, reason);
 912         PF_FRAG_UNLOCK();
 913
 914         if (verdict != PF_PASS)
 915                 return (PF_DROP);
 916
 917         m = *m0;
 918         if (m == NULL)
 919                 return (PF_DROP);
 920
 921         h = mtod(m, struct ip *);
 922
 923  no_fragment:
 924         /* At this point, only IP_DF is allowed in ip_off */
 925         if (h->ip_off & ~htons(IP_DF)) {
 926                 u_int16_t ip_off = h->ip_off;
 927
 928                 h->ip_off &= htons(IP_DF);
 929                 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
 930         }
 931
 932         pf_scrub_ip(&m, r->rule_flag, r->min_ttl, r->set_tos);
 933
 934         return (PF_PASS);
 935
 936  bad:
 937         DPFPRINTF(("dropping bad fragment\n"));
 938         REASON_SET(reason, PFRES_FRAG);
 939  drop:
 940         if (r != NULL && r->log)
 941                 PFLOG_PACKET(kif, m, AF_INET, dir, *reason, r, NULL, NULL, pd,
 942                     1);
 943
 944         return (PF_DROP);
 945 }
 946 #endif
 947
 948 #ifdef INET6
 949 int
 950 pf_normalize_ip6(struct mbuf **m0, int dir, struct pfi_kif *kif,
 951     u_short *reason, struct pf_pdesc *pd)
 952 {
 953         struct mbuf             *m = *m0;
 954         struct pf_rule          *r;
 955         struct ip6_hdr          *h = mtod(m, struct ip6_hdr *);
 956         int                      extoff;
 957         int                      off;
 958         struct ip6_ext           ext;
 959         struct ip6_opt           opt;
 960         struct ip6_opt_jumbo     jumbo;
 961         struct ip6_frag          frag;
 962         u_int32_t                jumbolen = 0, plen;
 963         int                      optend;
 964         int                      ooff;
 965         u_int8_t                 proto;
 966         int                      terminal;
 967
 968         PF_RULES_RASSERT();
 969
 970         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
 971         while (r != NULL) {
 972                 r->evaluations++;
 973                 if (pfi_kif_match(r->kif, kif) == r->ifnot)
 974                         r = r->skip[PF_SKIP_IFP].ptr;
 975                 else if (r->direction && r->direction != dir)
 976                         r = r->skip[PF_SKIP_DIR].ptr;
 977                 else if (r->af && r->af != AF_INET6)
 978                         r = r->skip[PF_SKIP_AF].ptr;
 979 #if 0 /* header chain! */
 980                 else if (r->proto && r->proto != h->ip6_nxt)
 981                         r = r->skip[PF_SKIP_PROTO].ptr;
 982 #endif
 983                 else if (PF_MISMATCHAW(&r->src.addr,
 984                     (struct pf_addr *)&h->ip6_src, AF_INET6,
 985                     r->src.neg, kif, M_GETFIB(m)))
 986                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
 987                 else if (PF_MISMATCHAW(&r->dst.addr,
 988                     (struct pf_addr *)&h->ip6_dst, AF_INET6,
 989                     r->dst.neg, NULL, M_GETFIB(m)))
 990                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
 991                 else
 992                         break;
 993         }
 994
 995         if (r == NULL || r->action == PF_NOSCRUB)
 996                 return (PF_PASS);
 997         else {
 998                 r->packets[dir == PF_OUT]++;
 999                 r->bytes[dir == PF_OUT] += pd->tot_len;
1000         }
1001
1002         /* Check for illegal packets */
1003         if (sizeof(struct ip6_hdr) + IPV6_MAXPACKET < m->m_pkthdr.len)
1004                 goto drop;
1005
1006         extoff = 0;
1007         off = sizeof(struct ip6_hdr);
1008         proto = h->ip6_nxt;
1009         terminal = 0;
1010         do {
1011                 switch (proto) {
1012                 case IPPROTO_FRAGMENT:
1013                         goto fragment;
1014                         break;
1015                 case IPPROTO_AH:
1016                 case IPPROTO_ROUTING:
1017                 case IPPROTO_DSTOPTS:
1018                         if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL,
1019                             NULL, AF_INET6))
1020                                 goto shortpkt;
1021                         extoff = off;
1022                         if (proto == IPPROTO_AH)
1023                                 off += (ext.ip6e_len + 2) * 4;
1024                         else
1025                                 off += (ext.ip6e_len + 1) * 8;
1026                         proto = ext.ip6e_nxt;
1027                         break;
1028                 case IPPROTO_HOPOPTS:
1029                         if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL,
1030                             NULL, AF_INET6))
1031                                 goto shortpkt;
1032                         extoff = off;
1033                         optend = off + (ext.ip6e_len + 1) * 8;
1034                         ooff = off + sizeof(ext);
1035                         do {
1036                                 if (!pf_pull_hdr(m, ooff, &opt.ip6o_type,
1037                                     sizeof(opt.ip6o_type), NULL, NULL,
1038                                     AF_INET6))
1039                                         goto shortpkt;
1040                                 if (opt.ip6o_type == IP6OPT_PAD1) {
1041                                         ooff++;
1042                                         continue;
1043                                 }
1044                                 if (!pf_pull_hdr(m, ooff, &opt, sizeof(opt),
1045                                     NULL, NULL, AF_INET6))
1046                                         goto shortpkt;
1047                                 if (ooff + sizeof(opt) + opt.ip6o_len > optend)
1048                                         goto drop;
1049                                 switch (opt.ip6o_type) {
1050                                 case IP6OPT_JUMBO:
1051                                         if (h->ip6_plen != 0)
1052                                                 goto drop;
1053                                         if (!pf_pull_hdr(m, ooff, &jumbo,
1054                                             sizeof(jumbo), NULL, NULL,
1055                                             AF_INET6))
1056                                                 goto shortpkt;
1057                                         memcpy(&jumbolen, jumbo.ip6oj_jumbo_len,
1058                                             sizeof(jumbolen));
1059                                         jumbolen = ntohl(jumbolen);
1060                                         if (jumbolen <= IPV6_MAXPACKET)
1061                                                 goto drop;
1062                                         if (sizeof(struct ip6_hdr) + jumbolen !=
1063                                             m->m_pkthdr.len)
1064                                                 goto drop;
1065                                         break;
1066                                 default:
1067                                         break;
1068                                 }
1069                                 ooff += sizeof(opt) + opt.ip6o_len;
1070                         } while (ooff < optend);
1071
1072                         off = optend;
1073                         proto = ext.ip6e_nxt;
1074                         break;
1075                 default:
1076                         terminal = 1;
1077                         break;
1078                 }
1079         } while (!terminal);
1080
1081         /* jumbo payload option must be present, or plen > 0 */
1082         if (ntohs(h->ip6_plen) == 0)
1083                 plen = jumbolen;
1084         else
1085                 plen = ntohs(h->ip6_plen);
1086         if (plen == 0)
1087                 goto drop;
1088         if (sizeof(struct ip6_hdr) + plen > m->m_pkthdr.len)
1089                 goto shortpkt;
1090
1091         pf_scrub_ip6(&m, r->min_ttl);
1092
1093         return (PF_PASS);
1094
1095  fragment:
1096         /* Jumbo payload packets cannot be fragmented. */
1097         plen = ntohs(h->ip6_plen);
1098         if (plen == 0 || jumbolen)
1099                 goto drop;
1100         if (sizeof(struct ip6_hdr) + plen > m->m_pkthdr.len)
1101                 goto shortpkt;
1102
1103         if (!pf_pull_hdr(m, off, &frag, sizeof(frag), NULL, NULL, AF_INET6))
1104                 goto shortpkt;
1105
1106         /* Offset now points to data portion. */
1107         off += sizeof(frag);
1108
1109         /* Returns PF_DROP or *m0 is NULL or completely reassembled mbuf. */
1110         if (pf_reassemble6(m0, h, &frag, off, extoff, reason) != PF_PASS)
1111                 return (PF_DROP);
1112         m = *m0;
1113         if (m == NULL)
1114                 return (PF_DROP);
1115
1116         pd->flags |= PFDESC_IP_REAS;
1117         return (PF_PASS);
1118
1119  shortpkt:
1120         REASON_SET(reason, PFRES_SHORT);
1121         if (r != NULL && r->log)
1122                 PFLOG_PACKET(kif, m, AF_INET6, dir, *reason, r, NULL, NULL, pd,
1123                     1);
1124         return (PF_DROP);
1125
1126  drop:
1127         REASON_SET(reason, PFRES_NORM);
1128         if (r != NULL && r->log)
1129                 PFLOG_PACKET(kif, m, AF_INET6, dir, *reason, r, NULL, NULL, pd,
1130                     1);
1131         return (PF_DROP);
1132 }
1133 #endif /* INET6 */
1134
1135 int
1136 pf_normalize_tcp(int dir, struct pfi_kif *kif, struct mbuf *m, int ipoff,
1137     int off, void *h, struct pf_pdesc *pd)
1138 {
1139         struct pf_rule  *r, *rm = NULL;
1140         struct tcphdr   *th = pd->hdr.tcp;
1141         int              rewrite = 0;
1142         u_short          reason;
1143         u_int8_t         flags;
1144         sa_family_t      af = pd->af;
1145
1146         PF_RULES_RASSERT();
1147
1148         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1149         while (r != NULL) {
1150                 r->evaluations++;
1151                 if (pfi_kif_match(r->kif, kif) == r->ifnot)
1152                         r = r->skip[PF_SKIP_IFP].ptr;
1153                 else if (r->direction && r->direction != dir)
1154                         r = r->skip[PF_SKIP_DIR].ptr;
1155                 else if (r->af && r->af != af)
1156                         r = r->skip[PF_SKIP_AF].ptr;
1157                 else if (r->proto && r->proto != pd->proto)
1158                         r = r->skip[PF_SKIP_PROTO].ptr;
1159                 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
1160                     r->src.neg, kif, M_GETFIB(m)))
1161                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
1162                 else if (r->src.port_op && !pf_match_port(r->src.port_op,
1163                             r->src.port[0], r->src.port[1], th->th_sport))
1164                         r = r->skip[PF_SKIP_SRC_PORT].ptr;
1165                 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
1166                     r->dst.neg, NULL, M_GETFIB(m)))
1167                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
1168                 else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
1169                             r->dst.port[0], r->dst.port[1], th->th_dport))
1170                         r = r->skip[PF_SKIP_DST_PORT].ptr;
1171                 else if (r->os_fingerprint != PF_OSFP_ANY && !pf_osfp_match(
1172                             pf_osfp_fingerprint(pd, m, off, th),
1173                             r->os_fingerprint))
1174                         r = TAILQ_NEXT(r, entries);
1175                 else {
1176                         rm = r;
1177                         break;
1178                 }
1179         }
1180
1181         if (rm == NULL || rm->action == PF_NOSCRUB)
1182                 return (PF_PASS);
1183         else {
1184                 r->packets[dir == PF_OUT]++;
1185                 r->bytes[dir == PF_OUT] += pd->tot_len;
1186         }
1187
1188         if (rm->rule_flag & PFRULE_REASSEMBLE_TCP)
1189                 pd->flags |= PFDESC_TCP_NORM;
1190
1191         flags = th->th_flags;
1192         if (flags & TH_SYN) {
1193                 /* Illegal packet */
1194                 if (flags & TH_RST)
1195                         goto tcp_drop;
1196
1197                 if (flags & TH_FIN)
1198                         goto tcp_drop;
1199         } else {
1200                 /* Illegal packet */
1201                 if (!(flags & (TH_ACK|TH_RST)))
1202                         goto tcp_drop;
1203         }
1204
1205         if (!(flags & TH_ACK)) {
1206                 /* These flags are only valid if ACK is set */
1207                 if ((flags & TH_FIN) || (flags & TH_PUSH) || (flags & TH_URG))
1208                         goto tcp_drop;
1209         }
1210
1211         /* Check for illegal header length */
1212         if (th->th_off < (sizeof(struct tcphdr) >> 2))
1213                 goto tcp_drop;
1214
1215         /* If flags changed, or reserved data set, then adjust */
1216         if (flags != th->th_flags || th->th_x2 != 0) {
1217                 u_int16_t       ov, nv;
1218
1219                 ov = *(u_int16_t *)(&th->th_ack + 1);
1220                 th->th_flags = flags;
1221                 th->th_x2 = 0;
1222                 nv = *(u_int16_t *)(&th->th_ack + 1);
1223
1224                 th->th_sum = pf_proto_cksum_fixup(m, th->th_sum, ov, nv, 0);
1225                 rewrite = 1;
1226         }
1227
1228         /* Remove urgent pointer, if TH_URG is not set */
1229         if (!(flags & TH_URG) && th->th_urp) {
1230                 th->th_sum = pf_proto_cksum_fixup(m, th->th_sum, th->th_urp,
1231                     0, 0);
1232                 th->th_urp = 0;
1233                 rewrite = 1;
1234         }
1235
1236         /* Process options */
1237         if (r->max_mss && pf_normalize_tcpopt(r, m, th, off, pd->af))
1238                 rewrite = 1;
1239
1240         /* copy back packet headers if we sanitized */
1241         if (rewrite)
1242                 m_copyback(m, off, sizeof(*th), (caddr_t)th);
1243
1244         return (PF_PASS);
1245
1246  tcp_drop:
1247         REASON_SET(&reason, PFRES_NORM);
1248         if (rm != NULL && r->log)
1249                 PFLOG_PACKET(kif, m, AF_INET, dir, reason, r, NULL, NULL, pd,
1250                     1);
1251         return (PF_DROP);
1252 }
1253
1254 int
1255 pf_normalize_tcp_init(struct mbuf *m, int off, struct pf_pdesc *pd,
1256     struct tcphdr *th, struct pf_state_peer *src, struct pf_state_peer *dst)
1257 {
1258         u_int32_t tsval, tsecr;
1259         u_int8_t hdr[60];
1260         u_int8_t *opt;
1261
1262         KASSERT((src->scrub == NULL),
1263             ("pf_normalize_tcp_init: src->scrub != NULL"));
1264
1265         src->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT);
1266         if (src->scrub == NULL)
1267                 return (1);
1268
1269         switch (pd->af) {
1270 #ifdef INET
1271         case AF_INET: {
1272                 struct ip *h = mtod(m, struct ip *);
1273                 src->scrub->pfss_ttl = h->ip_ttl;
1274                 break;
1275         }
1276 #endif /* INET */
1277 #ifdef INET6
1278         case AF_INET6: {
1279                 struct ip6_hdr *h = mtod(m, struct ip6_hdr *);
1280                 src->scrub->pfss_ttl = h->ip6_hlim;
1281                 break;
1282         }
1283 #endif /* INET6 */
1284         }
1285
1286
1287         /*
1288          * All normalizations below are only begun if we see the start of
1289          * the connections.  They must all set an enabled bit in pfss_flags
1290          */
1291         if ((th->th_flags & TH_SYN) == 0)
1292                 return (0);
1293
1294
1295         if (th->th_off > (sizeof(struct tcphdr) >> 2) && src->scrub &&
1296             pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) {
1297                 /* Diddle with TCP options */
1298                 int hlen;
1299                 opt = hdr + sizeof(struct tcphdr);
1300                 hlen = (th->th_off << 2) - sizeof(struct tcphdr);
1301                 while (hlen >= TCPOLEN_TIMESTAMP) {
1302                         switch (*opt) {
1303                         case TCPOPT_EOL:        /* FALLTHROUGH */
1304                         case TCPOPT_NOP:
1305                                 opt++;
1306                                 hlen--;
1307                                 break;
1308                         case TCPOPT_TIMESTAMP:
1309                                 if (opt[1] >= TCPOLEN_TIMESTAMP) {
1310                                         src->scrub->pfss_flags |=
1311                                             PFSS_TIMESTAMP;
1312                                         src->scrub->pfss_ts_mod =
1313                                             htonl(arc4random());
1314
1315                                         /* note PFSS_PAWS not set yet */
1316                                         memcpy(&tsval, &opt[2],
1317                                             sizeof(u_int32_t));
1318                                         memcpy(&tsecr, &opt[6],
1319                                             sizeof(u_int32_t));
1320                                         src->scrub->pfss_tsval0 = ntohl(tsval);
1321                                         src->scrub->pfss_tsval = ntohl(tsval);
1322                                         src->scrub->pfss_tsecr = ntohl(tsecr);
1323                                         getmicrouptime(&src->scrub->pfss_last);
1324                                 }
1325                                 /* FALLTHROUGH */
1326                         default:
1327                                 hlen -= MAX(opt[1], 2);
1328                                 opt += MAX(opt[1], 2);
1329                                 break;
1330                         }
1331                 }
1332         }
1333
1334         return (0);
1335 }
1336
1337 void
1338 pf_normalize_tcp_cleanup(struct pf_state *state)
1339 {
1340         if (state->src.scrub)
1341                 uma_zfree(V_pf_state_scrub_z, state->src.scrub);
1342         if (state->dst.scrub)
1343                 uma_zfree(V_pf_state_scrub_z, state->dst.scrub);
1344
1345         /* Someday... flush the TCP segment reassembly descriptors. */
1346 }
1347
1348 int
1349 pf_normalize_tcp_stateful(struct mbuf *m, int off, struct pf_pdesc *pd,
1350     u_short *reason, struct tcphdr *th, struct pf_state *state,
1351     struct pf_state_peer *src, struct pf_state_peer *dst, int *writeback)
1352 {
1353         struct timeval uptime;
1354         u_int32_t tsval, tsecr;
1355         u_int tsval_from_last;
1356         u_int8_t hdr[60];
1357         u_int8_t *opt;
1358         int copyback = 0;
1359         int got_ts = 0;
1360
1361         KASSERT((src->scrub || dst->scrub),
1362             ("%s: src->scrub && dst->scrub!", __func__));
1363
1364         /*
1365          * Enforce the minimum TTL seen for this connection.  Negate a common
1366          * technique to evade an intrusion detection system and confuse
1367          * firewall state code.
1368          */
1369         switch (pd->af) {
1370 #ifdef INET
1371         case AF_INET: {
1372                 if (src->scrub) {
1373                         struct ip *h = mtod(m, struct ip *);
1374                         if (h->ip_ttl > src->scrub->pfss_ttl)
1375                                 src->scrub->pfss_ttl = h->ip_ttl;
1376                         h->ip_ttl = src->scrub->pfss_ttl;
1377                 }
1378                 break;
1379         }
1380 #endif /* INET */
1381 #ifdef INET6
1382         case AF_INET6: {
1383                 if (src->scrub) {
1384                         struct ip6_hdr *h = mtod(m, struct ip6_hdr *);
1385                         if (h->ip6_hlim > src->scrub->pfss_ttl)
1386                                 src->scrub->pfss_ttl = h->ip6_hlim;
1387                         h->ip6_hlim = src->scrub->pfss_ttl;
1388                 }
1389                 break;
1390         }
1391 #endif /* INET6 */
1392         }
1393
1394         if (th->th_off > (sizeof(struct tcphdr) >> 2) &&
1395             ((src->scrub && (src->scrub->pfss_flags & PFSS_TIMESTAMP)) ||
1396             (dst->scrub && (dst->scrub->pfss_flags & PFSS_TIMESTAMP))) &&
1397             pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) {
1398                 /* Diddle with TCP options */
1399                 int hlen;
1400                 opt = hdr + sizeof(struct tcphdr);
1401                 hlen = (th->th_off << 2) - sizeof(struct tcphdr);
1402                 while (hlen >= TCPOLEN_TIMESTAMP) {
1403                         switch (*opt) {
1404                         case TCPOPT_EOL:        /* FALLTHROUGH */
1405                         case TCPOPT_NOP:
1406                                 opt++;
1407                                 hlen--;
1408                                 break;
1409                         case TCPOPT_TIMESTAMP:
1410                                 /* Modulate the timestamps.  Can be used for
1411                                  * NAT detection, OS uptime determination or
1412                                  * reboot detection.
1413                                  */
1414
1415                                 if (got_ts) {
1416                                         /* Huh?  Multiple timestamps!? */
1417                                         if (V_pf_status.debug >= PF_DEBUG_MISC) {
1418                                                 DPFPRINTF(("multiple TS??"));
1419                                                 pf_print_state(state);
1420                                                 printf("\n");
1421                                         }
1422                                         REASON_SET(reason, PFRES_TS);
1423                                         return (PF_DROP);
1424                                 }
1425                                 if (opt[1] >= TCPOLEN_TIMESTAMP) {
1426                                         memcpy(&tsval, &opt[2],
1427                                             sizeof(u_int32_t));
1428                                         if (tsval && src->scrub &&
1429                                             (src->scrub->pfss_flags &
1430                                             PFSS_TIMESTAMP)) {
1431                                                 tsval = ntohl(tsval);
1432                                                 pf_change_proto_a(m, &opt[2],
1433                                                     &th->th_sum,
1434                                                     htonl(tsval +
1435                                                     src->scrub->pfss_ts_mod),
1436                                                     0);
1437                                                 copyback = 1;
1438                                         }
1439
1440                                         /* Modulate TS reply iff valid (!0) */
1441                                         memcpy(&tsecr, &opt[6],
1442                                             sizeof(u_int32_t));
1443                                         if (tsecr && dst->scrub &&
1444                                             (dst->scrub->pfss_flags &
1445                                             PFSS_TIMESTAMP)) {
1446                                                 tsecr = ntohl(tsecr)
1447                                                     - dst->scrub->pfss_ts_mod;
1448                                                 pf_change_proto_a(m, &opt[6],
1449                                                     &th->th_sum, htonl(tsecr),
1450                                                     0);
1451                                                 copyback = 1;
1452                                         }
1453                                         got_ts = 1;
1454                                 }
1455                                 /* FALLTHROUGH */
1456                         default:
1457                                 hlen -= MAX(opt[1], 2);
1458                                 opt += MAX(opt[1], 2);
1459                                 break;
1460                         }
1461                 }
1462                 if (copyback) {
1463                         /* Copyback the options, caller copys back header */
1464                         *writeback = 1;
1465                         m_copyback(m, off + sizeof(struct tcphdr),
1466                             (th->th_off << 2) - sizeof(struct tcphdr), hdr +
1467                             sizeof(struct tcphdr));
1468                 }
1469         }
1470
1471
1472         /*
1473          * Must invalidate PAWS checks on connections idle for too long.
1474          * The fastest allowed timestamp clock is 1ms.  That turns out to
1475          * be about 24 days before it wraps.  XXX Right now our lowerbound
1476          * TS echo check only works for the first 12 days of a connection
1477          * when the TS has exhausted half its 32bit space
1478          */
1479 #define TS_MAX_IDLE     (24*24*60*60)
1480 #define TS_MAX_CONN     (12*24*60*60)   /* XXX remove when better tsecr check */
1481
1482         getmicrouptime(&uptime);
1483         if (src->scrub && (src->scrub->pfss_flags & PFSS_PAWS) &&
1484             (uptime.tv_sec - src->scrub->pfss_last.tv_sec > TS_MAX_IDLE ||
1485             time_uptime - state->creation > TS_MAX_CONN))  {
1486                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1487                         DPFPRINTF(("src idled out of PAWS\n"));
1488                         pf_print_state(state);
1489                         printf("\n");
1490                 }
1491                 src->scrub->pfss_flags = (src->scrub->pfss_flags & ~PFSS_PAWS)
1492                     | PFSS_PAWS_IDLED;
1493         }
1494         if (dst->scrub && (dst->scrub->pfss_flags & PFSS_PAWS) &&
1495             uptime.tv_sec - dst->scrub->pfss_last.tv_sec > TS_MAX_IDLE) {
1496                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1497                         DPFPRINTF(("dst idled out of PAWS\n"));
1498                         pf_print_state(state);
1499                         printf("\n");
1500                 }
1501                 dst->scrub->pfss_flags = (dst->scrub->pfss_flags & ~PFSS_PAWS)
1502                     | PFSS_PAWS_IDLED;
1503         }
1504
1505         if (got_ts && src->scrub && dst->scrub &&
1506             (src->scrub->pfss_flags & PFSS_PAWS) &&
1507             (dst->scrub->pfss_flags & PFSS_PAWS)) {
1508                 /* Validate that the timestamps are "in-window".
1509                  * RFC1323 describes TCP Timestamp options that allow
1510                  * measurement of RTT (round trip time) and PAWS
1511                  * (protection against wrapped sequence numbers).  PAWS
1512                  * gives us a set of rules for rejecting packets on
1513                  * long fat pipes (packets that were somehow delayed
1514                  * in transit longer than the time it took to send the
1515                  * full TCP sequence space of 4Gb).  We can use these
1516                  * rules and infer a few others that will let us treat
1517                  * the 32bit timestamp and the 32bit echoed timestamp
1518                  * as sequence numbers to prevent a blind attacker from
1519                  * inserting packets into a connection.
1520                  *
1521                  * RFC1323 tells us:
1522                  *  - The timestamp on this packet must be greater than
1523                  *    or equal to the last value echoed by the other
1524                  *    endpoint.  The RFC says those will be discarded
1525                  *    since it is a dup that has already been acked.
1526                  *    This gives us a lowerbound on the timestamp.
1527                  *        timestamp >= other last echoed timestamp
1528                  *  - The timestamp will be less than or equal to
1529                  *    the last timestamp plus the time between the
1530                  *    last packet and now.  The RFC defines the max
1531                  *    clock rate as 1ms.  We will allow clocks to be
1532                  *    up to 10% fast and will allow a total difference
1533                  *    or 30 seconds due to a route change.  And this
1534                  *    gives us an upperbound on the timestamp.
1535                  *        timestamp <= last timestamp + max ticks
1536                  *    We have to be careful here.  Windows will send an
1537                  *    initial timestamp of zero and then initialize it
1538                  *    to a random value after the 3whs; presumably to
1539                  *    avoid a DoS by having to call an expensive RNG
1540                  *    during a SYN flood.  Proof MS has at least one
1541                  *    good security geek.
1542                  *
1543                  *  - The TCP timestamp option must also echo the other
1544                  *    endpoints timestamp.  The timestamp echoed is the
1545                  *    one carried on the earliest unacknowledged segment
1546                  *    on the left edge of the sequence window.  The RFC
1547                  *    states that the host will reject any echoed
1548                  *    timestamps that were larger than any ever sent.
1549                  *    This gives us an upperbound on the TS echo.
1550                  *        tescr <= largest_tsval
1551                  *  - The lowerbound on the TS echo is a little more
1552                  *    tricky to determine.  The other endpoint's echoed
1553                  *    values will not decrease.  But there may be
1554                  *    network conditions that re-order packets and
1555                  *    cause our view of them to decrease.  For now the
1556                  *    only lowerbound we can safely determine is that
1557                  *    the TS echo will never be less than the original
1558                  *    TS.  XXX There is probably a better lowerbound.
1559                  *    Remove TS_MAX_CONN with better lowerbound check.
1560                  *        tescr >= other original TS
1561                  *
1562                  * It is also important to note that the fastest
1563                  * timestamp clock of 1ms will wrap its 32bit space in
1564                  * 24 days.  So we just disable TS checking after 24
1565                  * days of idle time.  We actually must use a 12d
1566                  * connection limit until we can come up with a better
1567                  * lowerbound to the TS echo check.
1568                  */
1569                 struct timeval delta_ts;
1570                 int ts_fudge;
1571
1572
1573                 /*
1574                  * PFTM_TS_DIFF is how many seconds of leeway to allow
1575                  * a host's timestamp.  This can happen if the previous
1576                  * packet got delayed in transit for much longer than
1577                  * this packet.
1578                  */
1579                 if ((ts_fudge = state->rule.ptr->timeout[PFTM_TS_DIFF]) == 0)
1580                         ts_fudge = V_pf_default_rule.timeout[PFTM_TS_DIFF];
1581
1582                 /* Calculate max ticks since the last timestamp */
1583 #define TS_MAXFREQ      1100            /* RFC max TS freq of 1Khz + 10% skew */
1584 #define TS_MICROSECS    1000000         /* microseconds per second */
1585                 delta_ts = uptime;
1586                 timevalsub(&delta_ts, &src->scrub->pfss_last);
1587                 tsval_from_last = (delta_ts.tv_sec + ts_fudge) * TS_MAXFREQ;
1588                 tsval_from_last += delta_ts.tv_usec / (TS_MICROSECS/TS_MAXFREQ);
1589
1590                 if ((src->state >= TCPS_ESTABLISHED &&
1591                     dst->state >= TCPS_ESTABLISHED) &&
1592                     (SEQ_LT(tsval, dst->scrub->pfss_tsecr) ||
1593                     SEQ_GT(tsval, src->scrub->pfss_tsval + tsval_from_last) ||
1594                     (tsecr && (SEQ_GT(tsecr, dst->scrub->pfss_tsval) ||
1595                     SEQ_LT(tsecr, dst->scrub->pfss_tsval0))))) {
1596                         /* Bad RFC1323 implementation or an insertion attack.
1597                          *
1598                          * - Solaris 2.6 and 2.7 are known to send another ACK
1599                          *   after the FIN,FIN|ACK,ACK closing that carries
1600                          *   an old timestamp.
1601                          */
1602
1603                         DPFPRINTF(("Timestamp failed %c%c%c%c\n",
1604                             SEQ_LT(tsval, dst->scrub->pfss_tsecr) ? '0' : ' ',
1605                             SEQ_GT(tsval, src->scrub->pfss_tsval +
1606                             tsval_from_last) ? '1' : ' ',
1607                             SEQ_GT(tsecr, dst->scrub->pfss_tsval) ? '2' : ' ',
1608                             SEQ_LT(tsecr, dst->scrub->pfss_tsval0)? '3' : ' '));
1609                         DPFPRINTF((" tsval: %u  tsecr: %u  +ticks: %u  "
1610                             "idle: %jus %lums\n",
1611                             tsval, tsecr, tsval_from_last,
1612                             (uintmax_t)delta_ts.tv_sec,
1613                             delta_ts.tv_usec / 1000));
1614                         DPFPRINTF((" src->tsval: %u  tsecr: %u\n",
1615                             src->scrub->pfss_tsval, src->scrub->pfss_tsecr));
1616                         DPFPRINTF((" dst->tsval: %u  tsecr: %u  tsval0: %u"
1617                             "\n", dst->scrub->pfss_tsval,
1618                             dst->scrub->pfss_tsecr, dst->scrub->pfss_tsval0));
1619                         if (V_pf_status.debug >= PF_DEBUG_MISC) {
1620                                 pf_print_state(state);
1621                                 pf_print_flags(th->th_flags);
1622                                 printf("\n");
1623                         }
1624                         REASON_SET(reason, PFRES_TS);
1625                         return (PF_DROP);
1626                 }
1627
1628                 /* XXX I'd really like to require tsecr but it's optional */
1629
1630         } else if (!got_ts && (th->th_flags & TH_RST) == 0 &&
1631             ((src->state == TCPS_ESTABLISHED && dst->state == TCPS_ESTABLISHED)
1632             || pd->p_len > 0 || (th->th_flags & TH_SYN)) &&
1633             src->scrub && dst->scrub &&
1634             (src->scrub->pfss_flags & PFSS_PAWS) &&
1635             (dst->scrub->pfss_flags & PFSS_PAWS)) {
1636                 /* Didn't send a timestamp.  Timestamps aren't really useful
1637                  * when:
1638                  *  - connection opening or closing (often not even sent).
1639                  *    but we must not let an attacker to put a FIN on a
1640                  *    data packet to sneak it through our ESTABLISHED check.
1641                  *  - on a TCP reset.  RFC suggests not even looking at TS.
1642                  *  - on an empty ACK.  The TS will not be echoed so it will
1643                  *    probably not help keep the RTT calculation in sync and
1644                  *    there isn't as much danger when the sequence numbers
1645                  *    got wrapped.  So some stacks don't include TS on empty
1646                  *    ACKs :-(
1647                  *
1648                  * To minimize the disruption to mostly RFC1323 conformant
1649                  * stacks, we will only require timestamps on data packets.
1650                  *
1651                  * And what do ya know, we cannot require timestamps on data
1652                  * packets.  There appear to be devices that do legitimate
1653                  * TCP connection hijacking.  There are HTTP devices that allow
1654                  * a 3whs (with timestamps) and then buffer the HTTP request.
1655                  * If the intermediate device has the HTTP response cache, it
1656                  * will spoof the response but not bother timestamping its
1657                  * packets.  So we can look for the presence of a timestamp in
1658                  * the first data packet and if there, require it in all future
1659                  * packets.
1660                  */
1661
1662                 if (pd->p_len > 0 && (src->scrub->pfss_flags & PFSS_DATA_TS)) {
1663                         /*
1664                          * Hey!  Someone tried to sneak a packet in.  Or the
1665                          * stack changed its RFC1323 behavior?!?!
1666                          */
1667                         if (V_pf_status.debug >= PF_DEBUG_MISC) {
1668                                 DPFPRINTF(("Did not receive expected RFC1323 "
1669                                     "timestamp\n"));
1670                                 pf_print_state(state);
1671                                 pf_print_flags(th->th_flags);
1672                                 printf("\n");
1673                         }
1674                         REASON_SET(reason, PFRES_TS);
1675                         return (PF_DROP);
1676                 }
1677         }
1678
1679
1680         /*
1681          * We will note if a host sends his data packets with or without
1682          * timestamps.  And require all data packets to contain a timestamp
1683          * if the first does.  PAWS implicitly requires that all data packets be
1684          * timestamped.  But I think there are middle-man devices that hijack
1685          * TCP streams immediately after the 3whs and don't timestamp their
1686          * packets (seen in a WWW accelerator or cache).
1687          */
1688         if (pd->p_len > 0 && src->scrub && (src->scrub->pfss_flags &
1689             (PFSS_TIMESTAMP|PFSS_DATA_TS|PFSS_DATA_NOTS)) == PFSS_TIMESTAMP) {
1690                 if (got_ts)
1691                         src->scrub->pfss_flags |= PFSS_DATA_TS;
1692                 else {
1693                         src->scrub->pfss_flags |= PFSS_DATA_NOTS;
1694                         if (V_pf_status.debug >= PF_DEBUG_MISC && dst->scrub &&
1695                             (dst->scrub->pfss_flags & PFSS_TIMESTAMP)) {
1696                                 /* Don't warn if other host rejected RFC1323 */
1697                                 DPFPRINTF(("Broken RFC1323 stack did not "
1698                                     "timestamp data packet. Disabled PAWS "
1699                                     "security.\n"));
1700                                 pf_print_state(state);
1701                                 pf_print_flags(th->th_flags);
1702                                 printf("\n");
1703                         }
1704                 }
1705         }
1706
1707
1708         /*
1709          * Update PAWS values
1710          */
1711         if (got_ts && src->scrub && PFSS_TIMESTAMP == (src->scrub->pfss_flags &
1712             (PFSS_PAWS_IDLED|PFSS_TIMESTAMP))) {
1713                 getmicrouptime(&src->scrub->pfss_last);
1714                 if (SEQ_GEQ(tsval, src->scrub->pfss_tsval) ||
1715                     (src->scrub->pfss_flags & PFSS_PAWS) == 0)
1716                         src->scrub->pfss_tsval = tsval;
1717
1718                 if (tsecr) {
1719                         if (SEQ_GEQ(tsecr, src->scrub->pfss_tsecr) ||
1720                             (src->scrub->pfss_flags & PFSS_PAWS) == 0)
1721                                 src->scrub->pfss_tsecr = tsecr;
1722
1723                         if ((src->scrub->pfss_flags & PFSS_PAWS) == 0 &&
1724                             (SEQ_LT(tsval, src->scrub->pfss_tsval0) ||
1725                             src->scrub->pfss_tsval0 == 0)) {
1726                                 /* tsval0 MUST be the lowest timestamp */
1727                                 src->scrub->pfss_tsval0 = tsval;
1728                         }
1729
1730                         /* Only fully initialized after a TS gets echoed */
1731                         if ((src->scrub->pfss_flags & PFSS_PAWS) == 0)
1732                                 src->scrub->pfss_flags |= PFSS_PAWS;
1733                 }
1734         }
1735
1736         /* I have a dream....  TCP segment reassembly.... */
1737         return (0);
1738 }
1739
1740 static int
1741 pf_normalize_tcpopt(struct pf_rule *r, struct mbuf *m, struct tcphdr *th,
1742     int off, sa_family_t af)
1743 {
1744         u_int16_t       *mss;
1745         int              thoff;
1746         int              opt, cnt, optlen = 0;
1747         int              rewrite = 0;
1748         u_char           opts[TCP_MAXOLEN];
1749         u_char          *optp = opts;
1750
1751         thoff = th->th_off << 2;
1752         cnt = thoff - sizeof(struct tcphdr);
1753
1754         if (cnt > 0 && !pf_pull_hdr(m, off + sizeof(*th), opts, cnt,
1755             NULL, NULL, af))
1756                 return (rewrite);
1757
1758         for (; cnt > 0; cnt -= optlen, optp += optlen) {
1759                 opt = optp[0];
1760                 if (opt == TCPOPT_EOL)
1761                         break;
1762                 if (opt == TCPOPT_NOP)
1763                         optlen = 1;
1764                 else {
1765                         if (cnt < 2)
1766                                 break;
1767                         optlen = optp[1];
1768                         if (optlen < 2 || optlen > cnt)
1769                                 break;
1770                 }
1771                 switch (opt) {
1772                 case TCPOPT_MAXSEG:
1773                         mss = (u_int16_t *)(optp + 2);
1774                         if ((ntohs(*mss)) > r->max_mss) {
1775                                 th->th_sum = pf_proto_cksum_fixup(m,
1776                                     th->th_sum, *mss, htons(r->max_mss), 0);
1777                                 *mss = htons(r->max_mss);
1778                                 rewrite = 1;
1779                         }
1780                         break;
1781                 default:
1782                         break;
1783                 }
1784         }
1785
1786         if (rewrite)
1787                 m_copyback(m, off + sizeof(*th), thoff - sizeof(*th), opts);
1788
1789         return (rewrite);
1790 }
1791
1792 #ifdef INET
1793 static void
1794 pf_scrub_ip(struct mbuf **m0, u_int32_t flags, u_int8_t min_ttl, u_int8_t tos)
1795 {
1796         struct mbuf             *m = *m0;
1797         struct ip               *h = mtod(m, struct ip *);
1798
1799         /* Clear IP_DF if no-df was requested */
1800         if (flags & PFRULE_NODF && h->ip_off & htons(IP_DF)) {
1801                 u_int16_t ip_off = h->ip_off;
1802
1803                 h->ip_off &= htons(~IP_DF);
1804                 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
1805         }
1806
1807         /* Enforce a minimum ttl, may cause endless packet loops */
1808         if (min_ttl && h->ip_ttl < min_ttl) {
1809                 u_int16_t ip_ttl = h->ip_ttl;
1810
1811                 h->ip_ttl = min_ttl;
1812                 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_ttl, h->ip_ttl, 0);
1813         }
1814
1815         /* Enforce tos */
1816         if (flags & PFRULE_SET_TOS) {
1817                 u_int16_t       ov, nv;
1818
1819                 ov = *(u_int16_t *)h;
1820                 h->ip_tos = tos | (h->ip_tos & IPTOS_ECN_MASK);
1821                 nv = *(u_int16_t *)h;
1822
1823                 h->ip_sum = pf_cksum_fixup(h->ip_sum, ov, nv, 0);
1824         }
1825
1826         /* random-id, but not for fragments */
1827         if (flags & PFRULE_RANDOMID && !(h->ip_off & ~htons(IP_DF))) {
1828                 uint16_t ip_id = h->ip_id;
1829
1830                 ip_fillid(h);
1831                 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_id, h->ip_id, 0);
1832         }
1833 }
1834 #endif /* INET */
1835
1836 #ifdef INET6
1837 static void
1838 pf_scrub_ip6(struct mbuf **m0, u_int8_t min_ttl)
1839 {
1840         struct mbuf             *m = *m0;
1841         struct ip6_hdr          *h = mtod(m, struct ip6_hdr *);
1842
1843         /* Enforce a minimum ttl, may cause endless packet loops */
1844         if (min_ttl && h->ip6_hlim < min_ttl)
1845                 h->ip6_hlim = min_ttl;
1846 }
1847 #endif