sys/net/pf/pf_norm.c

   1 /*      $OpenBSD: pf_norm.c,v 1.109 2007/05/28 17:16:39 henning Exp $ */
   2
   3 /*
   4  * Copyright (c) 2010 The DragonFly Project.  All rights reserved.
   5  *
   6  * Copyright 2001 Niels Provos <provos@citi.umich.edu>
   7  * All rights reserved.
   8  *
   9  * Redistribution and use in source and binary forms, with or without
  10  * modification, are permitted provided that the following conditions
  11  * are met:
  12  * 1. Redistributions of source code must retain the above copyright
  13  *    notice, this list of conditions and the following disclaimer.
  14  * 2. Redistributions in binary form must reproduce the above copyright
  15  *    notice, this list of conditions and the following disclaimer in the
  16  *    documentation and/or other materials provided with the distribution.
  17  *
  18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
  19  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
  20  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
  21  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
  22  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
  23  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  24  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  25  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  26  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
  27  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  28  */
  29
  30 #include "opt_inet.h"
  31 #include "opt_inet6.h"
  32
  33 #include <sys/param.h>
  34 #include <sys/systm.h>
  35 #include <sys/mbuf.h>
  36 #include <sys/filio.h>
  37 #include <sys/fcntl.h>
  38 #include <sys/socket.h>
  39 #include <sys/kernel.h>
  40 #include <sys/time.h>
  41 #include <vm/vm_zone.h>
  42
  43 #include <net/if.h>
  44 #include <net/if_types.h>
  45 #include <net/bpf.h>
  46 #include <net/route.h>
  47 #include <net/pf/if_pflog.h>
  48
  49 #include <netinet/in.h>
  50 #include <netinet/in_var.h>
  51 #include <netinet/in_systm.h>
  52 #include <netinet/ip.h>
  53 #include <netinet/ip_var.h>
  54 #include <netinet/tcp.h>
  55 #include <netinet/tcp_seq.h>
  56 #include <netinet/udp.h>
  57 #include <netinet/ip_icmp.h>
  58
  59 #ifdef INET6
  60 #include <netinet/ip6.h>
  61 #endif /* INET6 */
  62
  63 #include <net/pf/pfvar.h>
  64
  65 #define PFFRAG_SEENLAST 0x0001          /* Seen the last fragment for this */
  66 #define PFFRAG_NOBUFFER 0x0002          /* Non-buffering fragment cache */
  67 #define PFFRAG_DROP     0x0004          /* Drop all fragments */
  68 #define BUFFER_FRAGMENTS(fr)    (!((fr)->fr_flags & PFFRAG_NOBUFFER))
  69
  70
  71 TAILQ_HEAD(pf_fragqueue, pf_fragment)   pf_fragqueue;
  72 TAILQ_HEAD(pf_cachequeue, pf_fragment)  pf_cachequeue;
  73
  74 static __inline int      pf_frag_compare(struct pf_fragment *,
  75                             struct pf_fragment *);
  76 RB_HEAD(pf_frag_tree, pf_fragment)      pf_frag_tree, pf_cache_tree;
  77 RB_PROTOTYPE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
  78 RB_GENERATE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
  79
  80 /* Private prototypes */
  81 void                     pf_ip2key(struct pf_fragment *, struct ip *);
  82 void                     pf_remove_fragment(struct pf_fragment *);
  83 void                     pf_flush_fragments(void);
  84 void                     pf_free_fragment(struct pf_fragment *);
  85 struct pf_fragment      *pf_find_fragment(struct ip *, struct pf_frag_tree *);
  86 struct mbuf             *pf_reassemble(struct mbuf **, struct pf_fragment **,
  87                             struct pf_frent *, int);
  88 struct mbuf             *pf_fragcache(struct mbuf **, struct ip*,
  89                             struct pf_fragment **, int, int, int *);
  90 int                      pf_normalize_tcpopt(struct pf_rule *, struct mbuf *,
  91                             struct tcphdr *, int);
  92
  93 #define DPFPRINTF(x) do {                               \
  94         if (pf_status.debug >= PF_DEBUG_MISC) {         \
  95                 kprintf("%s: ", __func__);              \
  96                 kprintf x ;                             \
  97         }                                               \
  98 } while(0)
  99
 100 /* Globals */
 101 vm_zone_t                pf_frent_pl, pf_frag_pl, pf_cache_pl, pf_cent_pl;
 102 vm_zone_t                pf_state_scrub_pl;
 103 int                      pf_nfrents, pf_ncache;
 104
 105 void
 106 pf_normalize_init(void)
 107 {
 108         /* XXX
 109         pool_sethiwat(&pf_frag_pl, PFFRAG_FRAG_HIWAT);
 110         pool_sethardlimit(&pf_frent_pl, PFFRAG_FRENT_HIWAT, NULL, 0);
 111         pool_sethardlimit(&pf_cache_pl, PFFRAG_FRCACHE_HIWAT, NULL, 0);
 112         pool_sethardlimit(&pf_cent_pl, PFFRAG_FRCENT_HIWAT, NULL, 0);
 113         */
 114
 115         TAILQ_INIT(&pf_fragqueue);
 116         TAILQ_INIT(&pf_cachequeue);
 117 }
 118
 119 static __inline int
 120 pf_frag_compare(struct pf_fragment *a, struct pf_fragment *b)
 121 {
 122         int     diff;
 123
 124         if ((diff = a->fr_id - b->fr_id))
 125                 return (diff);
 126         else if ((diff = a->fr_p - b->fr_p))
 127                 return (diff);
 128         else if (a->fr_src.s_addr < b->fr_src.s_addr)
 129                 return (-1);
 130         else if (a->fr_src.s_addr > b->fr_src.s_addr)
 131                 return (1);
 132         else if (a->fr_dst.s_addr < b->fr_dst.s_addr)
 133                 return (-1);
 134         else if (a->fr_dst.s_addr > b->fr_dst.s_addr)
 135                 return (1);
 136         return (0);
 137 }
 138
 139 void
 140 pf_purge_expired_fragments(void)
 141 {
 142         struct pf_fragment      *frag;
 143         u_int32_t                expire = time_second -
 144                                     pf_default_rule.timeout[PFTM_FRAG];
 145
 146         while ((frag = TAILQ_LAST(&pf_fragqueue, pf_fragqueue)) != NULL) {
 147                 KASSERT((BUFFER_FRAGMENTS(frag)),
 148                         ("BUFFER_FRAGMENTS(frag) == 0: %s", __func__));
 149                 if (frag->fr_timeout > expire)
 150                         break;
 151
 152                 DPFPRINTF(("expiring %d(%p)\n", frag->fr_id, frag));
 153                 pf_free_fragment(frag);
 154         }
 155
 156         while ((frag = TAILQ_LAST(&pf_cachequeue, pf_cachequeue)) != NULL) {
 157                 KASSERT((!BUFFER_FRAGMENTS(frag)),
 158                         ("BUFFER_FRAGMENTS(frag) != 0: %s", __func__));
 159                 if (frag->fr_timeout > expire)
 160                         break;
 161
 162                 DPFPRINTF(("expiring %d(%p)\n", frag->fr_id, frag));
 163                 pf_free_fragment(frag);
 164                 KASSERT((TAILQ_EMPTY(&pf_cachequeue) ||
 165                     TAILQ_LAST(&pf_cachequeue, pf_cachequeue) != frag),
 166                     ("!(TAILQ_EMPTY() || TAILQ_LAST() == farg): %s",
 167                     __func__));
 168         }
 169 }
 170
 171 /*
 172  * Try to flush old fragments to make space for new ones
 173  */
 174
 175 void
 176 pf_flush_fragments(void)
 177 {
 178         struct pf_fragment      *frag;
 179         int                      goal;
 180
 181         goal = pf_nfrents * 9 / 10;
 182         DPFPRINTF(("trying to free > %d frents\n",
 183             pf_nfrents - goal));
 184         while (goal < pf_nfrents) {
 185                 frag = TAILQ_LAST(&pf_fragqueue, pf_fragqueue);
 186                 if (frag == NULL)
 187                         break;
 188                 pf_free_fragment(frag);
 189         }
 190
 191
 192         goal = pf_ncache * 9 / 10;
 193         DPFPRINTF(("trying to free > %d cache entries\n",
 194             pf_ncache - goal));
 195         while (goal < pf_ncache) {
 196                 frag = TAILQ_LAST(&pf_cachequeue, pf_cachequeue);
 197                 if (frag == NULL)
 198                         break;
 199                 pf_free_fragment(frag);
 200         }
 201 }
 202
 203 /* Frees the fragments and all associated entries */
 204
 205 void
 206 pf_free_fragment(struct pf_fragment *frag)
 207 {
 208         struct pf_frent         *frent;
 209         struct pf_frcache       *frcache;
 210
 211         /* Free all fragments */
 212         if (BUFFER_FRAGMENTS(frag)) {
 213                 for (frent = LIST_FIRST(&frag->fr_queue); frent;
 214                     frent = LIST_FIRST(&frag->fr_queue)) {
 215                         LIST_REMOVE(frent, fr_next);
 216
 217                         m_freem(frent->fr_m);
 218                         pool_put(&pf_frent_pl, frent);
 219                         pf_nfrents--;
 220                 }
 221         } else {
 222                 for (frcache = LIST_FIRST(&frag->fr_cache); frcache;
 223                     frcache = LIST_FIRST(&frag->fr_cache)) {
 224                         LIST_REMOVE(frcache, fr_next);
 225
 226                         KASSERT((LIST_EMPTY(&frag->fr_cache) ||
 227                             LIST_FIRST(&frag->fr_cache)->fr_off >
 228                             frcache->fr_end),
 229                             ("! (LIST_EMPTY() || LIST_FIRST()->fr_off >"
 230                              " frcache->fr_end): %s", __func__));
 231
 232                         pool_put(&pf_cent_pl, frcache);
 233                         pf_ncache--;
 234                 }
 235         }
 236
 237         pf_remove_fragment(frag);
 238 }
 239
 240 void
 241 pf_ip2key(struct pf_fragment *key, struct ip *ip)
 242 {
 243         key->fr_p = ip->ip_p;
 244         key->fr_id = ip->ip_id;
 245         key->fr_src.s_addr = ip->ip_src.s_addr;
 246         key->fr_dst.s_addr = ip->ip_dst.s_addr;
 247 }
 248
 249 struct pf_fragment *
 250 pf_find_fragment(struct ip *ip, struct pf_frag_tree *tree)
 251 {
 252         struct pf_fragment       key;
 253         struct pf_fragment      *frag;
 254
 255         pf_ip2key(&key, ip);
 256
 257         frag = RB_FIND(pf_frag_tree, tree, &key);
 258         if (frag != NULL) {
 259                 /* XXX Are we sure we want to update the timeout? */
 260                 frag->fr_timeout = time_second;
 261                 if (BUFFER_FRAGMENTS(frag)) {
 262                         TAILQ_REMOVE(&pf_fragqueue, frag, frag_next);
 263                         TAILQ_INSERT_HEAD(&pf_fragqueue, frag, frag_next);
 264                 } else {
 265                         TAILQ_REMOVE(&pf_cachequeue, frag, frag_next);
 266                         TAILQ_INSERT_HEAD(&pf_cachequeue, frag, frag_next);
 267                 }
 268         }
 269
 270         return (frag);
 271 }
 272
 273 /* Removes a fragment from the fragment queue and frees the fragment */
 274
 275 void
 276 pf_remove_fragment(struct pf_fragment *frag)
 277 {
 278         if (BUFFER_FRAGMENTS(frag)) {
 279                 RB_REMOVE(pf_frag_tree, &pf_frag_tree, frag);
 280                 TAILQ_REMOVE(&pf_fragqueue, frag, frag_next);
 281                 pool_put(&pf_frag_pl, frag);
 282         } else {
 283                 RB_REMOVE(pf_frag_tree, &pf_cache_tree, frag);
 284                 TAILQ_REMOVE(&pf_cachequeue, frag, frag_next);
 285                 pool_put(&pf_cache_pl, frag);
 286         }
 287 }
 288
 289 #define FR_IP_OFF(fr)   (((fr)->fr_ip->ip_off & IP_OFFMASK) << 3)
 290 struct mbuf *
 291 pf_reassemble(struct mbuf **m0, struct pf_fragment **frag,
 292     struct pf_frent *frent, int mff)
 293 {
 294         struct mbuf     *m = *m0, *m2;
 295         struct pf_frent *frea, *next;
 296         struct pf_frent *frep = NULL;
 297         struct ip       *ip = frent->fr_ip;
 298         int              hlen = ip->ip_hl << 2;
 299         u_int16_t        off = (ip->ip_off & IP_OFFMASK) << 3;
 300         u_int16_t        ip_len = ip->ip_len - ip->ip_hl * 4;
 301         u_int16_t        max = ip_len + off;
 302
 303         KASSERT((*frag == NULL || BUFFER_FRAGMENTS(*frag)),
 304             ("! (*frag == NULL || BUFFER_FRAGMENTS(*frag)): %s", __func__));
 305
 306         /* Strip off ip header */
 307         m->m_data += hlen;
 308         m->m_len -= hlen;
 309
 310         /* Create a new reassembly queue for this packet */
 311         if (*frag == NULL) {
 312                 *frag = pool_get(&pf_frag_pl, PR_NOWAIT);
 313                 if (*frag == NULL) {
 314                         pf_flush_fragments();
 315                         *frag = pool_get(&pf_frag_pl, PR_NOWAIT);
 316                         if (*frag == NULL)
 317                                 goto drop_fragment;
 318                 }
 319
 320                 (*frag)->fr_flags = 0;
 321                 (*frag)->fr_max = 0;
 322                 (*frag)->fr_src = frent->fr_ip->ip_src;
 323                 (*frag)->fr_dst = frent->fr_ip->ip_dst;
 324                 (*frag)->fr_p = frent->fr_ip->ip_p;
 325                 (*frag)->fr_id = frent->fr_ip->ip_id;
 326                 (*frag)->fr_timeout = time_second;
 327                 LIST_INIT(&(*frag)->fr_queue);
 328
 329                 RB_INSERT(pf_frag_tree, &pf_frag_tree, *frag);
 330                 TAILQ_INSERT_HEAD(&pf_fragqueue, *frag, frag_next);
 331
 332                 /* We do not have a previous fragment */
 333                 frep = NULL;
 334                 goto insert;
 335         }
 336
 337         /*
 338          * Find a fragment after the current one:
 339          *  - off contains the real shifted offset.
 340          */
 341         LIST_FOREACH(frea, &(*frag)->fr_queue, fr_next) {
 342                 if (FR_IP_OFF(frea) > off)
 343                         break;
 344                 frep = frea;
 345         }
 346
 347         KASSERT((frep != NULL || frea != NULL),
 348             ("!(frep != NULL || frea != NULL): %s", __func__));
 349
 350         if (frep != NULL &&
 351             FR_IP_OFF(frep) + frep->fr_ip->ip_len - frep->fr_ip->ip_hl *
 352             4 > off)
 353         {
 354                 u_int16_t       precut;
 355
 356                 precut = FR_IP_OFF(frep) + frep->fr_ip->ip_len -
 357                     frep->fr_ip->ip_hl * 4 - off;
 358                 if (precut >= ip_len)
 359                         goto drop_fragment;
 360                 m_adj(frent->fr_m, precut);
 361                 DPFPRINTF(("overlap -%d\n", precut));
 362                 /* Enforce 8 byte boundaries */
 363                 ip->ip_off = ip->ip_off + (precut >> 3);
 364                 off = (ip->ip_off & IP_OFFMASK) << 3;
 365                 ip_len -= precut;
 366                 ip->ip_len = ip_len;
 367         }
 368
 369         for (; frea != NULL && ip_len + off > FR_IP_OFF(frea);
 370             frea = next)
 371         {
 372                 u_int16_t       aftercut;
 373
 374                 aftercut = ip_len + off - FR_IP_OFF(frea);
 375                 DPFPRINTF(("adjust overlap %d\n", aftercut));
 376                 if (aftercut < frea->fr_ip->ip_len - frea->fr_ip->ip_hl
 377                     * 4)
 378                 {
 379                         frea->fr_ip->ip_len =
 380                             frea->fr_ip->ip_len - aftercut;
 381                         frea->fr_ip->ip_off = frea->fr_ip->ip_off +
 382                             (aftercut >> 3);
 383                         m_adj(frea->fr_m, aftercut);
 384                         break;
 385                 }
 386
 387                 /* This fragment is completely overlapped, lose it */
 388                 next = LIST_NEXT(frea, fr_next);
 389                 m_freem(frea->fr_m);
 390                 LIST_REMOVE(frea, fr_next);
 391                 pool_put(&pf_frent_pl, frea);
 392                 pf_nfrents--;
 393         }
 394
 395  insert:
 396         /* Update maximum data size */
 397         if ((*frag)->fr_max < max)
 398                 (*frag)->fr_max = max;
 399         /* This is the last segment */
 400         if (!mff)
 401                 (*frag)->fr_flags |= PFFRAG_SEENLAST;
 402
 403         if (frep == NULL)
 404                 LIST_INSERT_HEAD(&(*frag)->fr_queue, frent, fr_next);
 405         else
 406                 LIST_INSERT_AFTER(frep, frent, fr_next);
 407
 408         /* Check if we are completely reassembled */
 409         if (!((*frag)->fr_flags & PFFRAG_SEENLAST))
 410                 return (NULL);
 411
 412         /* Check if we have all the data */
 413         off = 0;
 414         for (frep = LIST_FIRST(&(*frag)->fr_queue); frep; frep = next) {
 415                 next = LIST_NEXT(frep, fr_next);
 416
 417                 off += frep->fr_ip->ip_len - frep->fr_ip->ip_hl * 4;
 418                 if (off < (*frag)->fr_max &&
 419                     (next == NULL || FR_IP_OFF(next) != off))
 420                 {
 421                         DPFPRINTF(("missing fragment at %d, next %d, max %d\n",
 422                             off, next == NULL ? -1 : FR_IP_OFF(next),
 423                             (*frag)->fr_max));
 424                         return (NULL);
 425                 }
 426         }
 427         DPFPRINTF(("%d < %d?\n", off, (*frag)->fr_max));
 428         if (off < (*frag)->fr_max)
 429                 return (NULL);
 430
 431         /* We have all the data */
 432         frent = LIST_FIRST(&(*frag)->fr_queue);
 433         KASSERT((frent != NULL), ("frent == NULL: %s", __func__));
 434         if ((frent->fr_ip->ip_hl << 2) + off > IP_MAXPACKET) {
 435                 DPFPRINTF(("drop: too big: %d\n", off));
 436                 pf_free_fragment(*frag);
 437                 *frag = NULL;
 438                 return (NULL);
 439         }
 440         next = LIST_NEXT(frent, fr_next);
 441
 442         /* Magic from ip_input */
 443         ip = frent->fr_ip;
 444         m = frent->fr_m;
 445         m2 = m->m_next;
 446         m->m_next = NULL;
 447         m_cat(m, m2);
 448         pool_put(&pf_frent_pl, frent);
 449         pf_nfrents--;
 450         for (frent = next; frent != NULL; frent = next) {
 451                 next = LIST_NEXT(frent, fr_next);
 452
 453                 m2 = frent->fr_m;
 454                 pool_put(&pf_frent_pl, frent);
 455                 pf_nfrents--;
 456                 m_cat(m, m2);
 457         }
 458
 459         ip->ip_src = (*frag)->fr_src;
 460         ip->ip_dst = (*frag)->fr_dst;
 461
 462         /* Remove from fragment queue */
 463         pf_remove_fragment(*frag);
 464         *frag = NULL;
 465
 466         hlen = ip->ip_hl << 2;
 467         ip->ip_len = off + hlen;
 468         m->m_len += hlen;
 469         m->m_data -= hlen;
 470
 471         /* some debugging cruft by sklower, below, will go away soon */
 472         /* XXX this should be done elsewhere */
 473         if (m->m_flags & M_PKTHDR) {
 474                 int plen = 0;
 475                 for (m2 = m; m2; m2 = m2->m_next)
 476                         plen += m2->m_len;
 477                 m->m_pkthdr.len = plen;
 478         }
 479
 480         DPFPRINTF(("complete: %p(%d)\n", m, ip->ip_len));
 481         return (m);
 482
 483  drop_fragment:
 484         /* Oops - fail safe - drop packet */
 485         pool_put(&pf_frent_pl, frent);
 486         pf_nfrents--;
 487         m_freem(m);
 488         return (NULL);
 489 }
 490
 491 struct mbuf *
 492 pf_fragcache(struct mbuf **m0, struct ip *h, struct pf_fragment **frag, int mff,
 493     int drop, int *nomem)
 494 {
 495         struct mbuf             *m = *m0;
 496         struct pf_frcache       *frp, *fra, *cur = NULL;
 497         int                      ip_len = h->ip_len - (h->ip_hl << 2);
 498         u_int16_t                off = h->ip_off << 3;
 499         u_int16_t                max = ip_len + off;
 500         int                      hosed = 0;
 501
 502         KASSERT((*frag == NULL || !BUFFER_FRAGMENTS(*frag)),
 503             ("!(*frag == NULL || !BUFFER_FRAGMENTS(*frag)): %s", __func__));
 504
 505         /* Create a new range queue for this packet */
 506         if (*frag == NULL) {
 507                 *frag = pool_get(&pf_cache_pl, PR_NOWAIT);
 508                 if (*frag == NULL) {
 509                         pf_flush_fragments();
 510                         *frag = pool_get(&pf_cache_pl, PR_NOWAIT);
 511                         if (*frag == NULL)
 512                                 goto no_mem;
 513                 }
 514
 515                 /* Get an entry for the queue */
 516                 cur = pool_get(&pf_cent_pl, PR_NOWAIT);
 517                 if (cur == NULL) {
 518                         pool_put(&pf_cache_pl, *frag);
 519                         *frag = NULL;
 520                         goto no_mem;
 521                 }
 522                 pf_ncache++;
 523
 524                 (*frag)->fr_flags = PFFRAG_NOBUFFER;
 525                 (*frag)->fr_max = 0;
 526                 (*frag)->fr_src = h->ip_src;
 527                 (*frag)->fr_dst = h->ip_dst;
 528                 (*frag)->fr_p = h->ip_p;
 529                 (*frag)->fr_id = h->ip_id;
 530                 (*frag)->fr_timeout = time_second;
 531
 532                 cur->fr_off = off;
 533                 cur->fr_end = max;
 534                 LIST_INIT(&(*frag)->fr_cache);
 535                 LIST_INSERT_HEAD(&(*frag)->fr_cache, cur, fr_next);
 536
 537                 RB_INSERT(pf_frag_tree, &pf_cache_tree, *frag);
 538                 TAILQ_INSERT_HEAD(&pf_cachequeue, *frag, frag_next);
 539
 540                 DPFPRINTF(("fragcache[%d]: new %d-%d\n", h->ip_id, off, max));
 541
 542                 goto pass;
 543         }
 544
 545         /*
 546          * Find a fragment after the current one:
 547          *  - off contains the real shifted offset.
 548          */
 549         frp = NULL;
 550         LIST_FOREACH(fra, &(*frag)->fr_cache, fr_next) {
 551                 if (fra->fr_off > off)
 552                         break;
 553                 frp = fra;
 554         }
 555
 556         KASSERT((frp != NULL || fra != NULL),
 557             ("!(frp != NULL || fra != NULL): %s", __func__));
 558
 559         if (frp != NULL) {
 560                 int     precut;
 561
 562                 precut = frp->fr_end - off;
 563                 if (precut >= ip_len) {
 564                         /* Fragment is entirely a duplicate */
 565                         DPFPRINTF(("fragcache[%d]: dead (%d-%d) %d-%d\n",
 566                             h->ip_id, frp->fr_off, frp->fr_end, off, max));
 567                         goto drop_fragment;
 568                 }
 569                 if (precut == 0) {
 570                         /* They are adjacent.  Fixup cache entry */
 571                         DPFPRINTF(("fragcache[%d]: adjacent (%d-%d) %d-%d\n",
 572                             h->ip_id, frp->fr_off, frp->fr_end, off, max));
 573                         frp->fr_end = max;
 574                 } else if (precut > 0) {
 575                         /* The first part of this payload overlaps with a
 576                          * fragment that has already been passed.
 577                          * Need to trim off the first part of the payload.
 578                          * But to do so easily, we need to create another
 579                          * mbuf to throw the original header into.
 580                          */
 581
 582                         DPFPRINTF(("fragcache[%d]: chop %d (%d-%d) %d-%d\n",
 583                             h->ip_id, precut, frp->fr_off, frp->fr_end, off,
 584                             max));
 585
 586                         off += precut;
 587                         max -= precut;
 588                         /* Update the previous frag to encompass this one */
 589                         frp->fr_end = max;
 590
 591                         if (!drop) {
 592                                 /* XXX Optimization opportunity
 593                                  * This is a very heavy way to trim the payload.
 594                                  * we could do it much faster by diddling mbuf
 595                                  * internals but that would be even less legible
 596                                  * than this mbuf magic.  For my next trick,
 597                                  * I'll pull a rabbit out of my laptop.
 598                                  */
 599                                 *m0 = m_dup(m, MB_DONTWAIT);
 600                                 /* From KAME Project : We have missed this! */
 601                                 m_adj(*m0, (h->ip_hl << 2) -
 602                                     (*m0)->m_pkthdr.len);
 603                                 if (*m0 == NULL)
 604                                         goto no_mem;
 605                                 KASSERT(((*m0)->m_next == NULL),
 606                                     ("(*m0)->m_next != NULL: %s",
 607                                     __func__));
 608                                 m_adj(m, precut + (h->ip_hl << 2));
 609                                 m_cat(*m0, m);
 610                                 m = *m0;
 611                                 if (m->m_flags & M_PKTHDR) {
 612                                         int plen = 0;
 613                                         struct mbuf *t;
 614                                         for (t = m; t; t = t->m_next)
 615                                                 plen += t->m_len;
 616                                         m->m_pkthdr.len = plen;
 617                                 }
 618
 619
 620                                 h = mtod(m, struct ip *);
 621
 622                                 KASSERT(((int)m->m_len ==
 623                                     h->ip_len - precut),
 624                                     ("m->m_len != h->ip_len - precut: %s",
 625                                     __func__));
 626                                 h->ip_off = h->ip_off +
 627                                     (precut >> 3);
 628                                 h->ip_len = h->ip_len - precut;
 629                         } else {
 630                                 hosed++;
 631                         }
 632                 } else {
 633                         /* There is a gap between fragments */
 634
 635                         DPFPRINTF(("fragcache[%d]: gap %d (%d-%d) %d-%d\n",
 636                             h->ip_id, -precut, frp->fr_off, frp->fr_end, off,
 637                             max));
 638
 639                         cur = pool_get(&pf_cent_pl, PR_NOWAIT);
 640                         if (cur == NULL)
 641                                 goto no_mem;
 642                         pf_ncache++;
 643
 644                         cur->fr_off = off;
 645                         cur->fr_end = max;
 646                         LIST_INSERT_AFTER(frp, cur, fr_next);
 647                 }
 648         }
 649
 650         if (fra != NULL) {
 651                 int     aftercut;
 652                 int     merge = 0;
 653
 654                 aftercut = max - fra->fr_off;
 655                 if (aftercut == 0) {
 656                         /* Adjacent fragments */
 657                         DPFPRINTF(("fragcache[%d]: adjacent %d-%d (%d-%d)\n",
 658                             h->ip_id, off, max, fra->fr_off, fra->fr_end));
 659                         fra->fr_off = off;
 660                         merge = 1;
 661                 } else if (aftercut > 0) {
 662                         /* Need to chop off the tail of this fragment */
 663                         DPFPRINTF(("fragcache[%d]: chop %d %d-%d (%d-%d)\n",
 664                             h->ip_id, aftercut, off, max, fra->fr_off,
 665                             fra->fr_end));
 666                         fra->fr_off = off;
 667                         max -= aftercut;
 668
 669                         merge = 1;
 670
 671                         if (!drop) {
 672                                 m_adj(m, -aftercut);
 673                                 if (m->m_flags & M_PKTHDR) {
 674                                         int plen = 0;
 675                                         struct mbuf *t;
 676                                         for (t = m; t; t = t->m_next)
 677                                                 plen += t->m_len;
 678                                         m->m_pkthdr.len = plen;
 679                                 }
 680                                 h = mtod(m, struct ip *);
 681                                 KASSERT(((int)m->m_len == h->ip_len - aftercut),
 682                                     ("m->m_len != h->ip_len - aftercut: %s",
 683                                     __func__));
 684                                 h->ip_len = h->ip_len - aftercut;
 685                         } else {
 686                                 hosed++;
 687                         }
 688                 } else if (frp == NULL) {
 689                         /* There is a gap between fragments */
 690                         DPFPRINTF(("fragcache[%d]: gap %d %d-%d (%d-%d)\n",
 691                             h->ip_id, -aftercut, off, max, fra->fr_off,
 692                             fra->fr_end));
 693
 694                         cur = pool_get(&pf_cent_pl, PR_NOWAIT);
 695                         if (cur == NULL)
 696                                 goto no_mem;
 697                         pf_ncache++;
 698
 699                         cur->fr_off = off;
 700                         cur->fr_end = max;
 701                         LIST_INSERT_BEFORE(fra, cur, fr_next);
 702                 }
 703
 704
 705                 /* Need to glue together two separate fragment descriptors */
 706                 if (merge) {
 707                         if (cur && fra->fr_off <= cur->fr_end) {
 708                                 /* Need to merge in a previous 'cur' */
 709                                 DPFPRINTF(("fragcache[%d]: adjacent(merge "
 710                                     "%d-%d) %d-%d (%d-%d)\n",
 711                                     h->ip_id, cur->fr_off, cur->fr_end, off,
 712                                     max, fra->fr_off, fra->fr_end));
 713                                 fra->fr_off = cur->fr_off;
 714                                 LIST_REMOVE(cur, fr_next);
 715                                 pool_put(&pf_cent_pl, cur);
 716                                 pf_ncache--;
 717                                 cur = NULL;
 718
 719                         } else if (frp && fra->fr_off <= frp->fr_end) {
 720                                 /* Need to merge in a modified 'frp' */
 721                                 KASSERT((cur == NULL), ("cur != NULL: %s",
 722                                     __func__));
 723                                 DPFPRINTF(("fragcache[%d]: adjacent(merge "
 724                                     "%d-%d) %d-%d (%d-%d)\n",
 725                                     h->ip_id, frp->fr_off, frp->fr_end, off,
 726                                     max, fra->fr_off, fra->fr_end));
 727                                 fra->fr_off = frp->fr_off;
 728                                 LIST_REMOVE(frp, fr_next);
 729                                 pool_put(&pf_cent_pl, frp);
 730                                 pf_ncache--;
 731                                 frp = NULL;
 732
 733                         }
 734                 }
 735         }
 736
 737         if (hosed) {
 738                 /*
 739                  * We must keep tracking the overall fragment even when
 740                  * we're going to drop it anyway so that we know when to
 741                  * free the overall descriptor.  Thus we drop the frag late.
 742                  */
 743                 goto drop_fragment;
 744         }
 745
 746
 747  pass:
 748         /* Update maximum data size */
 749         if ((*frag)->fr_max < max)
 750                 (*frag)->fr_max = max;
 751
 752         /* This is the last segment */
 753         if (!mff)
 754                 (*frag)->fr_flags |= PFFRAG_SEENLAST;
 755
 756         /* Check if we are completely reassembled */
 757         if (((*frag)->fr_flags & PFFRAG_SEENLAST) &&
 758             LIST_FIRST(&(*frag)->fr_cache)->fr_off == 0 &&
 759             LIST_FIRST(&(*frag)->fr_cache)->fr_end == (*frag)->fr_max) {
 760                 /* Remove from fragment queue */
 761                 DPFPRINTF(("fragcache[%d]: done 0-%d\n", h->ip_id,
 762                     (*frag)->fr_max));
 763                 pf_free_fragment(*frag);
 764                 *frag = NULL;
 765         }
 766
 767         return (m);
 768
 769  no_mem:
 770         *nomem = 1;
 771
 772         /* Still need to pay attention to !IP_MF */
 773         if (!mff && *frag != NULL)
 774                 (*frag)->fr_flags |= PFFRAG_SEENLAST;
 775
 776         m_freem(m);
 777         return (NULL);
 778
 779  drop_fragment:
 780
 781         /* Still need to pay attention to !IP_MF */
 782         if (!mff && *frag != NULL)
 783                 (*frag)->fr_flags |= PFFRAG_SEENLAST;
 784
 785         if (drop) {
 786                 /* This fragment has been deemed bad.  Don't reass */
 787                 if (((*frag)->fr_flags & PFFRAG_DROP) == 0)
 788                         DPFPRINTF(("fragcache[%d]: dropping overall fragment\n",
 789                             h->ip_id));
 790                 (*frag)->fr_flags |= PFFRAG_DROP;
 791         }
 792
 793         m_freem(m);
 794         return (NULL);
 795 }
 796
 797 int
 798 pf_normalize_ip(struct mbuf **m0, int dir, struct pfi_kif *kif, u_short *reason,
 799     struct pf_pdesc *pd)
 800 {
 801         struct mbuf             *m = *m0;
 802         struct pf_rule          *r;
 803         struct pf_frent         *frent;
 804         struct pf_fragment      *frag = NULL;
 805         struct ip               *h = mtod(m, struct ip *);
 806         int                      mff = (h->ip_off & IP_MF);
 807         int                      hlen = h->ip_hl << 2;
 808         u_int16_t                fragoff = (h->ip_off & IP_OFFMASK) << 3;
 809         u_int16_t                max;
 810         int                      ip_len;
 811         int                      ip_off;
 812
 813         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
 814         while (r != NULL) {
 815                 r->evaluations++;
 816                 if (pfi_kif_match(r->kif, kif) == r->ifnot)
 817                         r = r->skip[PF_SKIP_IFP].ptr;
 818                 else if (r->direction && r->direction != dir)
 819                         r = r->skip[PF_SKIP_DIR].ptr;
 820                 else if (r->af && r->af != AF_INET)
 821                         r = r->skip[PF_SKIP_AF].ptr;
 822                 else if (r->proto && r->proto != h->ip_p)
 823                         r = r->skip[PF_SKIP_PROTO].ptr;
 824                 else if (PF_MISMATCHAW(&r->src.addr,
 825                     (struct pf_addr *)&h->ip_src.s_addr, AF_INET,
 826                     r->src.neg, kif))
 827                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
 828                 else if (PF_MISMATCHAW(&r->dst.addr,
 829                     (struct pf_addr *)&h->ip_dst.s_addr, AF_INET,
 830                     r->dst.neg, NULL))
 831                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
 832                 else
 833                         break;
 834         }
 835
 836         if (r == NULL || r->action == PF_NOSCRUB)
 837                 return (PF_PASS);
 838         else {
 839                 r->packets[dir == PF_OUT]++;
 840                 r->bytes[dir == PF_OUT] += pd->tot_len;
 841         }
 842
 843         /* Check for illegal packets */
 844         if (hlen < (int)sizeof(struct ip))
 845                 goto drop;
 846
 847         if (hlen > h->ip_len)
 848                 goto drop;
 849
 850         /* Clear IP_DF if the rule uses the no-df option */
 851         if (r->rule_flag & PFRULE_NODF && h->ip_off & IP_DF) {
 852                 u_int16_t ip_off = h->ip_off;
 853
 854                 h->ip_off &= ~IP_DF;
 855                 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
 856         }
 857
 858         /* We will need other tests here */
 859         if (!fragoff && !mff)
 860                 goto no_fragment;
 861
 862         /* We're dealing with a fragment now. Don't allow fragments
 863          * with IP_DF to enter the cache. If the flag was cleared by
 864          * no-df above, fine. Otherwise drop it.
 865          */
 866         if (h->ip_off & IP_DF) {
 867                 DPFPRINTF(("IP_DF\n"));
 868                 goto bad;
 869         }
 870
 871         ip_len = h->ip_len - hlen;
 872         ip_off = (h->ip_off & IP_OFFMASK) << 3;
 873
 874         /* All fragments are 8 byte aligned */
 875         if (mff && (ip_len & 0x7)) {
 876                 DPFPRINTF(("mff and %d\n", ip_len));
 877                 goto bad;
 878         }
 879
 880         /* Respect maximum length */
 881         if (fragoff + ip_len > IP_MAXPACKET) {
 882                 DPFPRINTF(("max packet %d\n", fragoff + ip_len));
 883                 goto bad;
 884         }
 885         max = fragoff + ip_len;
 886
 887         if ((r->rule_flag & (PFRULE_FRAGCROP|PFRULE_FRAGDROP)) == 0) {
 888                 /* Fully buffer all of the fragments */
 889
 890                 frag = pf_find_fragment(h, &pf_frag_tree);
 891
 892                 /* Check if we saw the last fragment already */
 893                 if (frag != NULL && (frag->fr_flags & PFFRAG_SEENLAST) &&
 894                     max > frag->fr_max)
 895                         goto bad;
 896
 897                 /* Get an entry for the fragment queue */
 898                 frent = pool_get(&pf_frent_pl, PR_NOWAIT);
 899                 if (frent == NULL) {
 900                         REASON_SET(reason, PFRES_MEMORY);
 901                         return (PF_DROP);
 902                 }
 903                 pf_nfrents++;
 904                 frent->fr_ip = h;
 905                 frent->fr_m = m;
 906
 907                 /* Might return a completely reassembled mbuf, or NULL */
 908                 DPFPRINTF(("reass frag %d @ %d-%d\n", h->ip_id, fragoff, max));
 909                 *m0 = m = pf_reassemble(m0, &frag, frent, mff);
 910
 911                 if (m == NULL)
 912                         return (PF_DROP);
 913
 914                 if (frag != NULL && (frag->fr_flags & PFFRAG_DROP))
 915                         goto drop;
 916
 917                 h = mtod(m, struct ip *);
 918         } else {
 919                 /* non-buffering fragment cache (drops or masks overlaps) */
 920                 int     nomem = 0;
 921
 922                 if (dir == PF_OUT && m->m_pkthdr.pf.flags & PF_TAG_FRAGCACHE) {
 923                         /*
 924                          * Already passed the fragment cache in the
 925                          * input direction.  If we continued, it would
 926                          * appear to be a dup and would be dropped.
 927                          */
 928                         goto fragment_pass;
 929                 }
 930
 931                 frag = pf_find_fragment(h, &pf_cache_tree);
 932
 933                 /* Check if we saw the last fragment already */
 934                 if (frag != NULL && (frag->fr_flags & PFFRAG_SEENLAST) &&
 935                     max > frag->fr_max) {
 936                         if (r->rule_flag & PFRULE_FRAGDROP)
 937                                 frag->fr_flags |= PFFRAG_DROP;
 938                         goto bad;
 939                 }
 940
 941                 *m0 = m = pf_fragcache(m0, h, &frag, mff,
 942                     (r->rule_flag & PFRULE_FRAGDROP) ? 1 : 0, &nomem);
 943                 if (m == NULL) {
 944                         if (nomem)
 945                                 goto no_mem;
 946                         goto drop;
 947                 }
 948
 949                 if (dir == PF_IN)
 950                         m->m_pkthdr.pf.flags |= PF_TAG_FRAGCACHE;
 951
 952                 if (frag != NULL && (frag->fr_flags & PFFRAG_DROP))
 953                         goto drop;
 954                 goto fragment_pass;
 955         }
 956
 957  no_fragment:
 958         /* At this point, only IP_DF is allowed in ip_off */
 959         if (h->ip_off & ~IP_DF) {
 960                 u_int16_t ip_off = h->ip_off;
 961
 962                 h->ip_off &= IP_DF;
 963                 h->ip_sum = pf_cksum_fixup(h->ip_sum, htons(ip_off), htons(h->ip_off), 0);
 964         }
 965
 966         /* Enforce a minimum ttl, may cause endless packet loops */
 967         if (r->min_ttl && h->ip_ttl < r->min_ttl) {
 968                 u_int16_t ip_ttl = h->ip_ttl;
 969
 970                 h->ip_ttl = r->min_ttl;
 971                 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_ttl, h->ip_ttl, 0);
 972         }
 973
 974         if (r->rule_flag & PFRULE_RANDOMID) {
 975                 u_int16_t ip_id = h->ip_id;
 976
 977                 h->ip_id = ip_randomid();
 978                 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_id, h->ip_id, 0);
 979         }
 980         if ((r->rule_flag & (PFRULE_FRAGCROP|PFRULE_FRAGDROP)) == 0)
 981                 pd->flags |= PFDESC_IP_REAS;
 982
 983         return (PF_PASS);
 984
 985  fragment_pass:
 986         /* Enforce a minimum ttl, may cause endless packet loops */
 987         if (r->min_ttl && h->ip_ttl < r->min_ttl) {
 988                 u_int16_t ip_ttl = h->ip_ttl;
 989
 990                 h->ip_ttl = r->min_ttl;
 991                 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_ttl, h->ip_ttl, 0);
 992         }
 993         if ((r->rule_flag & (PFRULE_FRAGCROP|PFRULE_FRAGDROP)) == 0)
 994                 pd->flags |= PFDESC_IP_REAS;
 995         return (PF_PASS);
 996
 997  no_mem:
 998         REASON_SET(reason, PFRES_MEMORY);
 999         if (r != NULL && r->log)
1000                 PFLOG_PACKET(kif, h, m, AF_INET, dir, *reason, r, NULL, NULL, pd);
1001         return (PF_DROP);
1002
1003  drop:
1004         REASON_SET(reason, PFRES_NORM);
1005         if (r != NULL && r->log)
1006                 PFLOG_PACKET(kif, h, m, AF_INET, dir, *reason, r, NULL, NULL, pd);
1007         return (PF_DROP);
1008
1009  bad:
1010         DPFPRINTF(("dropping bad fragment\n"));
1011
1012         /* Free associated fragments */
1013         if (frag != NULL)
1014                 pf_free_fragment(frag);
1015
1016         REASON_SET(reason, PFRES_FRAG);
1017         if (r != NULL && r->log)
1018                 PFLOG_PACKET(kif, h, m, AF_INET, dir, *reason, r, NULL, NULL, pd);
1019
1020         return (PF_DROP);
1021 }
1022
1023 #ifdef INET6
1024 int
1025 pf_normalize_ip6(struct mbuf **m0, int dir, struct pfi_kif *kif,
1026     u_short *reason, struct pf_pdesc *pd)
1027 {
1028         struct mbuf             *m = *m0;
1029         struct pf_rule          *r;
1030         struct ip6_hdr          *h = mtod(m, struct ip6_hdr *);
1031         int                      off;
1032         struct ip6_ext           ext;
1033         struct ip6_opt           opt;
1034         struct ip6_opt_jumbo     jumbo;
1035         struct ip6_frag          frag;
1036         u_int32_t                jumbolen = 0, plen;
1037         u_int16_t                fragoff = 0;
1038         int                      optend;
1039         int                      ooff;
1040         u_int8_t                 proto;
1041         int                      terminal;
1042
1043         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1044         while (r != NULL) {
1045                 r->evaluations++;
1046                 if (pfi_kif_match(r->kif, kif) == r->ifnot)
1047                         r = r->skip[PF_SKIP_IFP].ptr;
1048                 else if (r->direction && r->direction != dir)
1049                         r = r->skip[PF_SKIP_DIR].ptr;
1050                 else if (r->af && r->af != AF_INET6)
1051                         r = r->skip[PF_SKIP_AF].ptr;
1052 #if 0 /* header chain! */
1053                 else if (r->proto && r->proto != h->ip6_nxt)
1054                         r = r->skip[PF_SKIP_PROTO].ptr;
1055 #endif
1056                 else if (PF_MISMATCHAW(&r->src.addr,
1057                     (struct pf_addr *)&h->ip6_src, AF_INET6,
1058                     r->src.neg, kif))
1059                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
1060                 else if (PF_MISMATCHAW(&r->dst.addr,
1061                     (struct pf_addr *)&h->ip6_dst, AF_INET6,
1062                     r->dst.neg, NULL))
1063                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
1064                 else
1065                         break;
1066         }
1067
1068         if (r == NULL || r->action == PF_NOSCRUB)
1069                 return (PF_PASS);
1070         else {
1071                 r->packets[dir == PF_OUT]++;
1072                 r->bytes[dir == PF_OUT] += pd->tot_len;
1073         }
1074
1075         /* Check for illegal packets */
1076         if (sizeof(struct ip6_hdr) + IPV6_MAXPACKET < m->m_pkthdr.len)
1077                 goto drop;
1078
1079         off = sizeof(struct ip6_hdr);
1080         proto = h->ip6_nxt;
1081         terminal = 0;
1082         do {
1083                 switch (proto) {
1084                 case IPPROTO_FRAGMENT:
1085                         goto fragment;
1086                         break;
1087                 case IPPROTO_AH:
1088                 case IPPROTO_ROUTING:
1089                 case IPPROTO_DSTOPTS:
1090                         if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL,
1091                             NULL, AF_INET6))
1092                                 goto shortpkt;
1093                         if (proto == IPPROTO_AH)
1094                                 off += (ext.ip6e_len + 2) * 4;
1095                         else
1096                                 off += (ext.ip6e_len + 1) * 8;
1097                         proto = ext.ip6e_nxt;
1098                         break;
1099                 case IPPROTO_HOPOPTS:
1100                         if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL,
1101                             NULL, AF_INET6))
1102                                 goto shortpkt;
1103                         optend = off + (ext.ip6e_len + 1) * 8;
1104                         ooff = off + sizeof(ext);
1105                         do {
1106                                 if (!pf_pull_hdr(m, ooff, &opt.ip6o_type,
1107                                     sizeof(opt.ip6o_type), NULL, NULL,
1108                                     AF_INET6))
1109                                         goto shortpkt;
1110                                 if (opt.ip6o_type == IP6OPT_PAD1) {
1111                                         ooff++;
1112                                         continue;
1113                                 }
1114                                 if (!pf_pull_hdr(m, ooff, &opt, sizeof(opt),
1115                                     NULL, NULL, AF_INET6))
1116                                         goto shortpkt;
1117                                 if (ooff + sizeof(opt) + opt.ip6o_len > optend)
1118                                         goto drop;
1119                                 switch (opt.ip6o_type) {
1120                                 case IP6OPT_JUMBO:
1121                                         if (h->ip6_plen != 0)
1122                                                 goto drop;
1123                                         if (!pf_pull_hdr(m, ooff, &jumbo,
1124                                             sizeof(jumbo), NULL, NULL,
1125                                             AF_INET6))
1126                                                 goto shortpkt;
1127                                         memcpy(&jumbolen, jumbo.ip6oj_jumbo_len,
1128                                             sizeof(jumbolen));
1129                                         jumbolen = ntohl(jumbolen);
1130                                         if (jumbolen <= IPV6_MAXPACKET)
1131                                                 goto drop;
1132                                         if (sizeof(struct ip6_hdr) + jumbolen !=
1133                                             m->m_pkthdr.len)
1134                                                 goto drop;
1135                                         break;
1136                                 default:
1137                                         break;
1138                                 }
1139                                 ooff += sizeof(opt) + opt.ip6o_len;
1140                         } while (ooff < optend);
1141
1142                         off = optend;
1143                         proto = ext.ip6e_nxt;
1144                         break;
1145                 default:
1146                         terminal = 1;
1147                         break;
1148                 }
1149         } while (!terminal);
1150
1151         /* jumbo payload option must be present, or plen > 0 */
1152         if (ntohs(h->ip6_plen) == 0)
1153                 plen = jumbolen;
1154         else
1155                 plen = ntohs(h->ip6_plen);
1156         if (plen == 0)
1157                 goto drop;
1158         if (sizeof(struct ip6_hdr) + plen > m->m_pkthdr.len)
1159                 goto shortpkt;
1160
1161         /* Enforce a minimum ttl, may cause endless packet loops */
1162         if (r->min_ttl && h->ip6_hlim < r->min_ttl)
1163                 h->ip6_hlim = r->min_ttl;
1164
1165         return (PF_PASS);
1166
1167  fragment:
1168         if (ntohs(h->ip6_plen) == 0 || jumbolen)
1169                 goto drop;
1170         plen = ntohs(h->ip6_plen);
1171
1172         if (!pf_pull_hdr(m, off, &frag, sizeof(frag), NULL, NULL, AF_INET6))
1173                 goto shortpkt;
1174         fragoff = ntohs(frag.ip6f_offlg & IP6F_OFF_MASK);
1175         if (fragoff + (plen - off - sizeof(frag)) > IPV6_MAXPACKET)
1176                 goto badfrag;
1177
1178         /* do something about it */
1179         /* remember to set pd->flags |= PFDESC_IP_REAS */
1180         return (PF_PASS);
1181
1182  shortpkt:
1183         REASON_SET(reason, PFRES_SHORT);
1184         if (r != NULL && r->log)
1185                 PFLOG_PACKET(kif, h, m, AF_INET6, dir, *reason, r, NULL, NULL, pd);
1186         return (PF_DROP);
1187
1188  drop:
1189         REASON_SET(reason, PFRES_NORM);
1190         if (r != NULL && r->log)
1191                 PFLOG_PACKET(kif, h, m, AF_INET6, dir, *reason, r, NULL, NULL, pd);
1192         return (PF_DROP);
1193
1194  badfrag:
1195         REASON_SET(reason, PFRES_FRAG);
1196         if (r != NULL && r->log)
1197                 PFLOG_PACKET(kif, h, m, AF_INET6, dir, *reason, r, NULL, NULL, pd);
1198         return (PF_DROP);
1199 }
1200 #endif /* INET6 */
1201
1202 int
1203 pf_normalize_tcp(int dir, struct pfi_kif *kif, struct mbuf *m, int ipoff,
1204     int off, void *h, struct pf_pdesc *pd)
1205 {
1206         struct pf_rule  *r, *rm = NULL;
1207         struct tcphdr   *th = pd->hdr.tcp;
1208         int              rewrite = 0;
1209         u_short          reason;
1210         u_int8_t         flags;
1211         sa_family_t      af = pd->af;
1212
1213         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1214         while (r != NULL) {
1215                 r->evaluations++;
1216                 if (pfi_kif_match(r->kif, kif) == r->ifnot)
1217                         r = r->skip[PF_SKIP_IFP].ptr;
1218                 else if (r->direction && r->direction != dir)
1219                         r = r->skip[PF_SKIP_DIR].ptr;
1220                 else if (r->af && r->af != af)
1221                         r = r->skip[PF_SKIP_AF].ptr;
1222                 else if (r->proto && r->proto != pd->proto)
1223                         r = r->skip[PF_SKIP_PROTO].ptr;
1224                 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
1225                     r->src.neg, kif))
1226                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
1227                 else if (r->src.port_op && !pf_match_port(r->src.port_op,
1228                             r->src.port[0], r->src.port[1], th->th_sport))
1229                         r = r->skip[PF_SKIP_SRC_PORT].ptr;
1230                 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
1231                     r->dst.neg, NULL))
1232                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
1233                 else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
1234                             r->dst.port[0], r->dst.port[1], th->th_dport))
1235                         r = r->skip[PF_SKIP_DST_PORT].ptr;
1236                 else if (r->os_fingerprint != PF_OSFP_ANY && !pf_osfp_match(
1237                             pf_osfp_fingerprint(pd, m, off, th),
1238                             r->os_fingerprint))
1239                         r = TAILQ_NEXT(r, entries);
1240                 else {
1241                         rm = r;
1242                         break;
1243                 }
1244         }
1245
1246         if (rm == NULL || rm->action == PF_NOSCRUB)
1247                 return (PF_PASS);
1248         else {
1249                 r->packets[dir == PF_OUT]++;
1250                 r->bytes[dir == PF_OUT] += pd->tot_len;
1251         }
1252
1253         if (rm->rule_flag & PFRULE_REASSEMBLE_TCP)
1254                 pd->flags |= PFDESC_TCP_NORM;
1255
1256         flags = th->th_flags;
1257         if (flags & TH_SYN) {
1258                 /* Illegal packet */
1259                 if (flags & TH_RST)
1260                         goto tcp_drop;
1261
1262                 if (flags & TH_FIN)
1263                         flags &= ~TH_FIN;
1264         } else {
1265                 /* Illegal packet */
1266                 if (!(flags & (TH_ACK|TH_RST)))
1267                         goto tcp_drop;
1268         }
1269
1270         if (!(flags & TH_ACK)) {
1271                 /* These flags are only valid if ACK is set */
1272                 if ((flags & TH_FIN) || (flags & TH_PUSH) || (flags & TH_URG))
1273                         goto tcp_drop;
1274         }
1275
1276         /* Check for illegal header length */
1277         if (th->th_off < (sizeof(struct tcphdr) >> 2))
1278                 goto tcp_drop;
1279
1280         /* If flags changed, or reserved data set, then adjust */
1281         if (flags != th->th_flags || th->th_x2 != 0) {
1282                 u_int16_t       ov, nv;
1283
1284                 ov = *(u_int16_t *)(&th->th_ack + 1);
1285                 th->th_flags = flags;
1286                 th->th_x2 = 0;
1287                 nv = *(u_int16_t *)(&th->th_ack + 1);
1288
1289                 th->th_sum = pf_cksum_fixup(th->th_sum, ov, nv, 0);
1290                 rewrite = 1;
1291         }
1292
1293         /* Remove urgent pointer, if TH_URG is not set */
1294         if (!(flags & TH_URG) && th->th_urp) {
1295                 th->th_sum = pf_cksum_fixup(th->th_sum, th->th_urp, 0, 0);
1296                 th->th_urp = 0;
1297                 rewrite = 1;
1298         }
1299
1300         /* Process options */
1301         if (r->max_mss && pf_normalize_tcpopt(r, m, th, off))
1302                 rewrite = 1;
1303
1304         /* copy back packet headers if we sanitized */
1305         if (rewrite)
1306                 m_copyback(m, off, sizeof(*th), (caddr_t)th);
1307
1308         return (PF_PASS);
1309
1310  tcp_drop:
1311         REASON_SET(&reason, PFRES_NORM);
1312         if (rm != NULL && r->log)
1313                 PFLOG_PACKET(kif, h, m, AF_INET, dir, reason, r, NULL, NULL, pd);
1314         return (PF_DROP);
1315 }
1316
1317 int
1318 pf_normalize_tcp_init(struct mbuf *m, int off, struct pf_pdesc *pd,
1319     struct tcphdr *th, struct pf_state_peer *src, struct pf_state_peer *dst)
1320 {
1321         u_int32_t tsval, tsecr;
1322         u_int8_t hdr[60];
1323         u_int8_t *opt;
1324
1325         KASSERT((src->scrub == NULL),
1326             ("pf_normalize_tcp_init: src->scrub != NULL"));
1327
1328         src->scrub = pool_get(&pf_state_scrub_pl, PR_NOWAIT);
1329         if (src->scrub == NULL)
1330                 return (1);
1331         bzero(src->scrub, sizeof(*src->scrub));
1332
1333         switch (pd->af) {
1334 #ifdef INET
1335         case AF_INET: {
1336                 struct ip *h = mtod(m, struct ip *);
1337                 src->scrub->pfss_ttl = h->ip_ttl;
1338                 break;
1339         }
1340 #endif /* INET */
1341 #ifdef INET6
1342         case AF_INET6: {
1343                 struct ip6_hdr *h = mtod(m, struct ip6_hdr *);
1344                 src->scrub->pfss_ttl = h->ip6_hlim;
1345                 break;
1346         }
1347 #endif /* INET6 */
1348         }
1349
1350
1351         /*
1352          * All normalizations below are only begun if we see the start of
1353          * the connections.  They must all set an enabled bit in pfss_flags
1354          */
1355         if ((th->th_flags & TH_SYN) == 0)
1356                 return (0);
1357
1358
1359         if (th->th_off > (sizeof(struct tcphdr) >> 2) && src->scrub &&
1360             pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) {
1361                 /* Diddle with TCP options */
1362                 int hlen;
1363                 opt = hdr + sizeof(struct tcphdr);
1364                 hlen = (th->th_off << 2) - sizeof(struct tcphdr);
1365                 while (hlen >= TCPOLEN_TIMESTAMP) {
1366                         switch (*opt) {
1367                         case TCPOPT_EOL:        /* FALLTHROUGH */
1368                         case TCPOPT_NOP:
1369                                 opt++;
1370                                 hlen--;
1371                                 break;
1372                         case TCPOPT_TIMESTAMP:
1373                                 if (opt[1] >= TCPOLEN_TIMESTAMP) {
1374                                         src->scrub->pfss_flags |=
1375                                             PFSS_TIMESTAMP;
1376                                         src->scrub->pfss_ts_mod = karc4random();
1377
1378                                         /* note PFSS_PAWS not set yet */
1379                                         memcpy(&tsval, &opt[2],
1380                                             sizeof(u_int32_t));
1381                                         memcpy(&tsecr, &opt[6],
1382                                             sizeof(u_int32_t));
1383                                         src->scrub->pfss_tsval0 = ntohl(tsval);
1384                                         src->scrub->pfss_tsval = ntohl(tsval);
1385                                         src->scrub->pfss_tsecr = ntohl(tsecr);
1386                                         getmicrouptime(&src->scrub->pfss_last);
1387                                 }
1388                                 /* FALLTHROUGH */
1389                         default:
1390                                 hlen -= MAX(opt[1], 2);
1391                                 opt += MAX(opt[1], 2);
1392                                 break;
1393                         }
1394                 }
1395         }
1396
1397         return (0);
1398 }
1399
1400 void
1401 pf_normalize_tcp_cleanup(struct pf_state *state)
1402 {
1403         if (state->src.scrub)
1404                 pool_put(&pf_state_scrub_pl, state->src.scrub);
1405         if (state->dst.scrub)
1406                 pool_put(&pf_state_scrub_pl, state->dst.scrub);
1407
1408         /* Someday... flush the TCP segment reassembly descriptors. */
1409 }
1410
1411 int
1412 pf_normalize_tcp_stateful(struct mbuf *m, int off, struct pf_pdesc *pd,
1413     u_short *reason, struct tcphdr *th, struct pf_state *state,
1414     struct pf_state_peer *src, struct pf_state_peer *dst, int *writeback)
1415 {
1416         struct timeval uptime;
1417         u_int32_t tsval, tsecr;
1418         u_int tsval_from_last;
1419         u_int8_t hdr[60];
1420         u_int8_t *opt;
1421         int copyback = 0;
1422         int got_ts = 0;
1423
1424         KASSERT((src->scrub || dst->scrub),
1425             ("pf_normalize_tcp_statefull: src->scrub && dst->scrub!"));
1426
1427         /*
1428          * Enforce the minimum TTL seen for this connection.  Negate a common
1429          * technique to evade an intrusion detection system and confuse
1430          * firewall state code.
1431          */
1432         switch (pd->af) {
1433 #ifdef INET
1434         case AF_INET: {
1435                 if (src->scrub) {
1436                         struct ip *h = mtod(m, struct ip *);
1437                         if (h->ip_ttl > src->scrub->pfss_ttl)
1438                                 src->scrub->pfss_ttl = h->ip_ttl;
1439                         h->ip_ttl = src->scrub->pfss_ttl;
1440                 }
1441                 break;
1442         }
1443 #endif /* INET */
1444 #ifdef INET6
1445         case AF_INET6: {
1446                 if (src->scrub) {
1447                         struct ip6_hdr *h = mtod(m, struct ip6_hdr *);
1448                         if (h->ip6_hlim > src->scrub->pfss_ttl)
1449                                 src->scrub->pfss_ttl = h->ip6_hlim;
1450                         h->ip6_hlim = src->scrub->pfss_ttl;
1451                 }
1452                 break;
1453         }
1454 #endif /* INET6 */
1455         }
1456
1457         if (th->th_off > (sizeof(struct tcphdr) >> 2) &&
1458             ((src->scrub && (src->scrub->pfss_flags & PFSS_TIMESTAMP)) ||
1459             (dst->scrub && (dst->scrub->pfss_flags & PFSS_TIMESTAMP))) &&
1460             pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) {
1461                 /* Diddle with TCP options */
1462                 int hlen;
1463                 opt = hdr + sizeof(struct tcphdr);
1464                 hlen = (th->th_off << 2) - sizeof(struct tcphdr);
1465                 while (hlen >= TCPOLEN_TIMESTAMP) {
1466                         switch (*opt) {
1467                         case TCPOPT_EOL:        /* FALLTHROUGH */
1468                         case TCPOPT_NOP:
1469                                 opt++;
1470                                 hlen--;
1471                                 break;
1472                         case TCPOPT_TIMESTAMP:
1473                                 /* Modulate the timestamps.  Can be used for
1474                                  * NAT detection, OS uptime determination or
1475                                  * reboot detection.
1476                                  */
1477
1478                                 if (got_ts) {
1479                                         /* Huh?  Multiple timestamps!? */
1480                                         if (pf_status.debug >= PF_DEBUG_MISC) {
1481                                                 DPFPRINTF(("multiple TS??"));
1482                                                 pf_print_state(state);
1483                                                 kprintf("\n");
1484                                         }
1485                                         REASON_SET(reason, PFRES_TS);
1486                                         return (PF_DROP);
1487                                 }
1488                                 if (opt[1] >= TCPOLEN_TIMESTAMP) {
1489                                         memcpy(&tsval, &opt[2],
1490                                             sizeof(u_int32_t));
1491                                         if (tsval && src->scrub &&
1492                                             (src->scrub->pfss_flags &
1493                                             PFSS_TIMESTAMP)) {
1494                                                 tsval = ntohl(tsval);
1495                                                 pf_change_a(&opt[2],
1496                                                     &th->th_sum,
1497                                                     htonl(tsval +
1498                                                     src->scrub->pfss_ts_mod),
1499                                                     0);
1500                                                 copyback = 1;
1501                                         }
1502
1503                                         /* Modulate TS reply iff valid (!0) */
1504                                         memcpy(&tsecr, &opt[6],
1505                                             sizeof(u_int32_t));
1506                                         if (tsecr && dst->scrub &&
1507                                             (dst->scrub->pfss_flags &
1508                                             PFSS_TIMESTAMP)) {
1509                                                 tsecr = ntohl(tsecr)
1510                                                     - dst->scrub->pfss_ts_mod;
1511                                                 pf_change_a(&opt[6],
1512                                                     &th->th_sum, htonl(tsecr),
1513                                                     0);
1514                                                 copyback = 1;
1515                                         }
1516                                         got_ts = 1;
1517                                 }
1518                                 /* FALLTHROUGH */
1519                         default:
1520                                 hlen -= MAX(opt[1], 2);
1521                                 opt += MAX(opt[1], 2);
1522                                 break;
1523                         }
1524                 }
1525                 if (copyback) {
1526                         /* Copyback the options, caller copys back header */
1527                         *writeback = 1;
1528                         m_copyback(m, off + sizeof(struct tcphdr),
1529                             (th->th_off << 2) - sizeof(struct tcphdr), hdr +
1530                             sizeof(struct tcphdr));
1531                 }
1532         }
1533
1534
1535         /*
1536          * Must invalidate PAWS checks on connections idle for too long.
1537          * The fastest allowed timestamp clock is 1ms.  That turns out to
1538          * be about 24 days before it wraps.  XXX Right now our lowerbound
1539          * TS echo check only works for the first 12 days of a connection
1540          * when the TS has exhausted half its 32bit space
1541          */
1542 #define TS_MAX_IDLE     (24*24*60*60)
1543 #define TS_MAX_CONN     (12*24*60*60)   /* XXX remove when better tsecr check */
1544
1545         getmicrouptime(&uptime);
1546         if (src->scrub && (src->scrub->pfss_flags & PFSS_PAWS) &&
1547             (uptime.tv_sec - src->scrub->pfss_last.tv_sec > TS_MAX_IDLE ||
1548             time_second - state->creation > TS_MAX_CONN))  {
1549                 if (pf_status.debug >= PF_DEBUG_MISC) {
1550                         DPFPRINTF(("src idled out of PAWS\n"));
1551                         pf_print_state(state);
1552                         kprintf("\n");
1553                 }
1554                 src->scrub->pfss_flags = (src->scrub->pfss_flags & ~PFSS_PAWS)
1555                     | PFSS_PAWS_IDLED;
1556         }
1557         if (dst->scrub && (dst->scrub->pfss_flags & PFSS_PAWS) &&
1558             uptime.tv_sec - dst->scrub->pfss_last.tv_sec > TS_MAX_IDLE) {
1559                 if (pf_status.debug >= PF_DEBUG_MISC) {
1560                         DPFPRINTF(("dst idled out of PAWS\n"));
1561                         pf_print_state(state);
1562                         kprintf("\n");
1563                 }
1564                 dst->scrub->pfss_flags = (dst->scrub->pfss_flags & ~PFSS_PAWS)
1565                     | PFSS_PAWS_IDLED;
1566         }
1567
1568         if (got_ts && src->scrub && dst->scrub &&
1569             (src->scrub->pfss_flags & PFSS_PAWS) &&
1570             (dst->scrub->pfss_flags & PFSS_PAWS)) {
1571                 /* Validate that the timestamps are "in-window".
1572                  * RFC1323 describes TCP Timestamp options that allow
1573                  * measurement of RTT (round trip time) and PAWS
1574                  * (protection against wrapped sequence numbers).  PAWS
1575                  * gives us a set of rules for rejecting packets on
1576                  * long fat pipes (packets that were somehow delayed
1577                  * in transit longer than the time it took to send the
1578                  * full TCP sequence space of 4Gb).  We can use these
1579                  * rules and infer a few others that will let us treat
1580                  * the 32bit timestamp and the 32bit echoed timestamp
1581                  * as sequence numbers to prevent a blind attacker from
1582                  * inserting packets into a connection.
1583                  *
1584                  * RFC1323 tells us:
1585                  *  - The timestamp on this packet must be greater than
1586                  *    or equal to the last value echoed by the other
1587                  *    endpoint.  The RFC says those will be discarded
1588                  *    since it is a dup that has already been acked.
1589                  *    This gives us a lowerbound on the timestamp.
1590                  *        timestamp >= other last echoed timestamp
1591                  *  - The timestamp will be less than or equal to
1592                  *    the last timestamp plus the time between the
1593                  *    last packet and now.  The RFC defines the max
1594                  *    clock rate as 1ms.  We will allow clocks to be
1595                  *    up to 10% fast and will allow a total difference
1596                  *    or 30 seconds due to a route change.  And this
1597                  *    gives us an upperbound on the timestamp.
1598                  *        timestamp <= last timestamp + max ticks
1599                  *    We have to be careful here.  Windows will send an
1600                  *    initial timestamp of zero and then initialize it
1601                  *    to a random value after the 3whs; presumably to
1602                  *    avoid a DoS by having to call an expensive RNG
1603                  *    during a SYN flood.  Proof MS has at least one
1604                  *    good security geek.
1605                  *
1606                  *  - The TCP timestamp option must also echo the other
1607                  *    endpoints timestamp.  The timestamp echoed is the
1608                  *    one carried on the earliest unacknowledged segment
1609                  *    on the left edge of the sequence window.  The RFC
1610                  *    states that the host will reject any echoed
1611                  *    timestamps that were larger than any ever sent.
1612                  *    This gives us an upperbound on the TS echo.
1613                  *        tescr <= largest_tsval
1614                  *  - The lowerbound on the TS echo is a little more
1615                  *    tricky to determine.  The other endpoint's echoed
1616                  *    values will not decrease.  But there may be
1617                  *    network conditions that re-order packets and
1618                  *    cause our view of them to decrease.  For now the
1619                  *    only lowerbound we can safely determine is that
1620                  *    the TS echo will never be less than the original
1621                  *    TS.  XXX There is probably a better lowerbound.
1622                  *    Remove TS_MAX_CONN with better lowerbound check.
1623                  *        tescr >= other original TS
1624                  *
1625                  * It is also important to note that the fastest
1626                  * timestamp clock of 1ms will wrap its 32bit space in
1627                  * 24 days.  So we just disable TS checking after 24
1628                  * days of idle time.  We actually must use a 12d
1629                  * connection limit until we can come up with a better
1630                  * lowerbound to the TS echo check.
1631                  */
1632                 struct timeval delta_ts;
1633                 int ts_fudge;
1634
1635
1636                 /*
1637                  * PFTM_TS_DIFF is how many seconds of leeway to allow
1638                  * a host's timestamp.  This can happen if the previous
1639                  * packet got delayed in transit for much longer than
1640                  * this packet.
1641                  */
1642                 if ((ts_fudge = state->rule.ptr->timeout[PFTM_TS_DIFF]) == 0)
1643                         ts_fudge = pf_default_rule.timeout[PFTM_TS_DIFF];
1644
1645
1646                 /* Calculate max ticks since the last timestamp */
1647 #define TS_MAXFREQ      1100            /* RFC max TS freq of 1Khz + 10% skew */
1648 #define TS_MICROSECS    1000000         /* microseconds per second */
1649 #ifndef timersub
1650 #define timersub(tvp, uvp, vvp)                                         \
1651         do {                                                            \
1652                 (vvp)->tv_sec = (tvp)->tv_sec - (uvp)->tv_sec;          \
1653                 (vvp)->tv_usec = (tvp)->tv_usec - (uvp)->tv_usec;       \
1654                 if ((vvp)->tv_usec < 0) {                               \
1655                         (vvp)->tv_sec--;                                \
1656                         (vvp)->tv_usec += 1000000;                      \
1657                 }                                                       \
1658         } while (0)
1659 #endif
1660
1661                 timersub(&uptime, &src->scrub->pfss_last, &delta_ts);
1662                 tsval_from_last = (delta_ts.tv_sec + ts_fudge) * TS_MAXFREQ;
1663                 tsval_from_last += delta_ts.tv_usec / (TS_MICROSECS/TS_MAXFREQ);
1664
1665
1666                 if ((src->state >= TCPS_ESTABLISHED &&
1667                     dst->state >= TCPS_ESTABLISHED) &&
1668                     (SEQ_LT(tsval, dst->scrub->pfss_tsecr) ||
1669                     SEQ_GT(tsval, src->scrub->pfss_tsval + tsval_from_last) ||
1670                     (tsecr && (SEQ_GT(tsecr, dst->scrub->pfss_tsval) ||
1671                     SEQ_LT(tsecr, dst->scrub->pfss_tsval0))))) {
1672                         /* Bad RFC1323 implementation or an insertion attack.
1673                          *
1674                          * - Solaris 2.6 and 2.7 are known to send another ACK
1675                          *   after the FIN,FIN|ACK,ACK closing that carries
1676                          *   an old timestamp.
1677                          */
1678
1679                         DPFPRINTF(("Timestamp failed %c%c%c%c\n",
1680                             SEQ_LT(tsval, dst->scrub->pfss_tsecr) ? '0' : ' ',
1681                             SEQ_GT(tsval, src->scrub->pfss_tsval +
1682                             tsval_from_last) ? '1' : ' ',
1683                             SEQ_GT(tsecr, dst->scrub->pfss_tsval) ? '2' : ' ',
1684                             SEQ_LT(tsecr, dst->scrub->pfss_tsval0)? '3' : ' '));
1685                         DPFPRINTF((" tsval: %u  tsecr: %u  +ticks: %u  "
1686                             "idle: %lus %lums\n",
1687                             tsval, tsecr, tsval_from_last, delta_ts.tv_sec,
1688                             delta_ts.tv_usec / 1000));
1689                         DPFPRINTF((" src->tsval: %u  tsecr: %u\n",
1690                             src->scrub->pfss_tsval, src->scrub->pfss_tsecr));
1691                         DPFPRINTF((" dst->tsval: %u  tsecr: %u  tsval0: %u"
1692                             "\n", dst->scrub->pfss_tsval,
1693                             dst->scrub->pfss_tsecr, dst->scrub->pfss_tsval0));
1694                         if (pf_status.debug >= PF_DEBUG_MISC) {
1695                                 pf_print_state(state);
1696                                 pf_print_flags(th->th_flags);
1697                                 kprintf("\n");
1698                         }
1699                         REASON_SET(reason, PFRES_TS);
1700                         return (PF_DROP);
1701                 }
1702
1703                 /* XXX I'd really like to require tsecr but it's optional */
1704
1705         } else if (!got_ts && (th->th_flags & TH_RST) == 0 &&
1706             ((src->state == TCPS_ESTABLISHED && dst->state == TCPS_ESTABLISHED)
1707             || pd->p_len > 0 || (th->th_flags & TH_SYN)) &&
1708             src->scrub && dst->scrub &&
1709             (src->scrub->pfss_flags & PFSS_PAWS) &&
1710             (dst->scrub->pfss_flags & PFSS_PAWS)) {
1711                 /* Didn't send a timestamp.  Timestamps aren't really useful
1712                  * when:
1713                  *  - connection opening or closing (often not even sent).
1714                  *    but we must not let an attacker to put a FIN on a
1715                  *    data packet to sneak it through our ESTABLISHED check.
1716                  *  - on a TCP reset.  RFC suggests not even looking at TS.
1717                  *  - on an empty ACK.  The TS will not be echoed so it will
1718                  *    probably not help keep the RTT calculation in sync and
1719                  *    there isn't as much danger when the sequence numbers
1720                  *    got wrapped.  So some stacks don't include TS on empty
1721                  *    ACKs :-(
1722                  *
1723                  * To minimize the disruption to mostly RFC1323 conformant
1724                  * stacks, we will only require timestamps on data packets.
1725                  *
1726                  * And what do ya know, we cannot require timestamps on data
1727                  * packets.  There appear to be devices that do legitimate
1728                  * TCP connection hijacking.  There are HTTP devices that allow
1729                  * a 3whs (with timestamps) and then buffer the HTTP request.
1730                  * If the intermediate device has the HTTP response cache, it
1731                  * will spoof the response but not bother timestamping its
1732                  * packets.  So we can look for the presence of a timestamp in
1733                  * the first data packet and if there, require it in all future
1734                  * packets.
1735                  */
1736
1737                 if (pd->p_len > 0 && (src->scrub->pfss_flags & PFSS_DATA_TS)) {
1738                         /*
1739                          * Hey!  Someone tried to sneak a packet in.  Or the
1740                          * stack changed its RFC1323 behavior?!?!
1741                          */
1742                         if (pf_status.debug >= PF_DEBUG_MISC) {
1743                                 DPFPRINTF(("Did not receive expected RFC1323 "
1744                                     "timestamp\n"));
1745                                 pf_print_state(state);
1746                                 pf_print_flags(th->th_flags);
1747                                 kprintf("\n");
1748                         }
1749                         REASON_SET(reason, PFRES_TS);
1750                         return (PF_DROP);
1751                 }
1752         }
1753
1754
1755         /*
1756          * We will note if a host sends his data packets with or without
1757          * timestamps.  And require all data packets to contain a timestamp
1758          * if the first does.  PAWS implicitly requires that all data packets be
1759          * timestamped.  But I think there are middle-man devices that hijack
1760          * TCP streams immediately after the 3whs and don't timestamp their
1761          * packets (seen in a WWW accelerator or cache).
1762          */
1763         if (pd->p_len > 0 && src->scrub && (src->scrub->pfss_flags &
1764             (PFSS_TIMESTAMP|PFSS_DATA_TS|PFSS_DATA_NOTS)) == PFSS_TIMESTAMP) {
1765                 if (got_ts)
1766                         src->scrub->pfss_flags |= PFSS_DATA_TS;
1767                 else {
1768                         src->scrub->pfss_flags |= PFSS_DATA_NOTS;
1769                         if (pf_status.debug >= PF_DEBUG_MISC && dst->scrub &&
1770                             (dst->scrub->pfss_flags & PFSS_TIMESTAMP)) {
1771                                 /* Don't warn if other host rejected RFC1323 */
1772                                 DPFPRINTF(("Broken RFC1323 stack did not "
1773                                     "timestamp data packet. Disabled PAWS "
1774                                     "security.\n"));
1775                                 pf_print_state(state);
1776                                 pf_print_flags(th->th_flags);
1777                                 kprintf("\n");
1778                         }
1779                 }
1780         }
1781
1782
1783         /*
1784          * Update PAWS values
1785          */
1786         if (got_ts && src->scrub && PFSS_TIMESTAMP == (src->scrub->pfss_flags &
1787             (PFSS_PAWS_IDLED|PFSS_TIMESTAMP))) {
1788                 getmicrouptime(&src->scrub->pfss_last);
1789                 if (SEQ_GEQ(tsval, src->scrub->pfss_tsval) ||
1790                     (src->scrub->pfss_flags & PFSS_PAWS) == 0)
1791                         src->scrub->pfss_tsval = tsval;
1792
1793                 if (tsecr) {
1794                         if (SEQ_GEQ(tsecr, src->scrub->pfss_tsecr) ||
1795                             (src->scrub->pfss_flags & PFSS_PAWS) == 0)
1796                                 src->scrub->pfss_tsecr = tsecr;
1797
1798                         if ((src->scrub->pfss_flags & PFSS_PAWS) == 0 &&
1799                             (SEQ_LT(tsval, src->scrub->pfss_tsval0) ||
1800                             src->scrub->pfss_tsval0 == 0)) {
1801                                 /* tsval0 MUST be the lowest timestamp */
1802                                 src->scrub->pfss_tsval0 = tsval;
1803                         }
1804
1805                         /* Only fully initialized after a TS gets echoed */
1806                         if ((src->scrub->pfss_flags & PFSS_PAWS) == 0)
1807                                 src->scrub->pfss_flags |= PFSS_PAWS;
1808                 }
1809         }
1810
1811         /* I have a dream....  TCP segment reassembly.... */
1812         return (0);
1813 }
1814
1815 int
1816 pf_normalize_tcpopt(struct pf_rule *r, struct mbuf *m, struct tcphdr *th,
1817     int off)
1818 {
1819         u_int16_t       *mss;
1820         int              thoff;
1821         int              opt, cnt, optlen = 0;
1822         int              rewrite = 0;
1823         u_char          *optp;
1824
1825         thoff = th->th_off << 2;
1826         cnt = thoff - sizeof(struct tcphdr);
1827         optp = mtod(m, caddr_t) + off + sizeof(struct tcphdr);
1828
1829         for (; cnt > 0; cnt -= optlen, optp += optlen) {
1830                 opt = optp[0];
1831                 if (opt == TCPOPT_EOL)
1832                         break;
1833                 if (opt == TCPOPT_NOP)
1834                         optlen = 1;
1835                 else {
1836                         if (cnt < 2)
1837                                 break;
1838                         optlen = optp[1];
1839                         if (optlen < 2 || optlen > cnt)
1840                                 break;
1841                 }
1842                 switch (opt) {
1843                 case TCPOPT_MAXSEG:
1844                         mss = (u_int16_t *)(optp + 2);
1845                         if ((ntohs(*mss)) > r->max_mss) {
1846                                 th->th_sum = pf_cksum_fixup(th->th_sum,
1847                                     *mss, htons(r->max_mss), 0);
1848                                 *mss = htons(r->max_mss);
1849                                 rewrite = 1;
1850                         }
1851                         break;
1852                 default:
1853                         break;
1854                 }
1855         }
1856
1857         return (rewrite);
1858 }