sys/net/pf/pf_norm.c

   1 /*      $FreeBSD: src/sys/contrib/pf/net/pf_norm.c,v 1.10 2004/08/14 15:32:40 dwmalone Exp $    */
   2 /*      $OpenBSD: pf_norm.c,v 1.80.2.1 2004/04/30 21:46:33 brad Exp $ */
   3 /* add  $OpenBSD: pf_norm.c,v 1.87 2004/05/11 07:34:11 dhartmei Exp $ */
   4 /*      $DragonFly: src/sys/net/pf/pf_norm.c,v 1.9 2007/08/11 18:53:31 dillon Exp $ */
   5
   6 /*
   7  * Copyright (c) 2004 The DragonFly Project.  All rights reserved.
   8  *
   9  * Copyright 2001 Niels Provos <provos@citi.umich.edu>
  10  * All rights reserved.
  11  *
  12  * Redistribution and use in source and binary forms, with or without
  13  * modification, are permitted provided that the following conditions
  14  * are met:
  15  * 1. Redistributions of source code must retain the above copyright
  16  *    notice, this list of conditions and the following disclaimer.
  17  * 2. Redistributions in binary form must reproduce the above copyright
  18  *    notice, this list of conditions and the following disclaimer in the
  19  *    documentation and/or other materials provided with the distribution.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
  22  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
  23  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
  24  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
  25  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
  26  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  27  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  28  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  29  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
  30  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  31  */
  32
  33 #include "opt_inet.h"
  34 #include "opt_inet6.h"
  35
  36 #include <sys/param.h>
  37 #include <sys/systm.h>
  38 #include <sys/mbuf.h>
  39 #include <sys/filio.h>
  40 #include <sys/fcntl.h>
  41 #include <sys/socket.h>
  42 #include <sys/kernel.h>
  43 #include <sys/time.h>
  44 #include <vm/vm_zone.h>
  45
  46 #include <net/if.h>
  47 #include <net/if_types.h>
  48 #include <net/bpf.h>
  49 #include <net/route.h>
  50 #include <net/pf/if_pflog.h>
  51
  52 #include <netinet/in.h>
  53 #include <netinet/in_var.h>
  54 #include <netinet/in_systm.h>
  55 #include <netinet/ip.h>
  56 #include <netinet/ip_var.h>
  57 #include <netinet/tcp.h>
  58 #include <netinet/tcp_seq.h>
  59 #include <netinet/udp.h>
  60 #include <netinet/ip_icmp.h>
  61
  62 #ifdef INET6
  63 #include <netinet/ip6.h>
  64 #endif /* INET6 */
  65
  66 #include <net/pf/pfvar.h>
  67
  68 #ifdef INET6
  69 /*
  70  * XXX: This should go to netinet/ip6.h (KAME)
  71  */
  72 /* IPv6 options: common part */
  73 struct ip6_opt {
  74         u_int8_t ip6o_type;
  75         u_int8_t ip6o_len;
  76 } __packed;
  77
  78 /* Jumbo Payload Option */
  79 struct ip6_opt_jumbo {
  80         u_int8_t ip6oj_type;
  81         u_int8_t ip6oj_len;
  82         u_int8_t ip6oj_jumbo_len[4];
  83 } __packed;
  84
  85 /* NSAP Address Option */
  86 struct ip6_opt_nsap {
  87         u_int8_t ip6on_type;
  88         u_int8_t ip6on_len;
  89         u_int8_t ip6on_src_nsap_len;
  90         u_int8_t ip6on_dst_nsap_len;
  91         /* followed by source NSAP */
  92         /* followed by destination NSAP */
  93 } __packed;
  94
  95 /* Tunnel Limit Option */
  96 struct ip6_opt_tunnel {
  97         u_int8_t ip6ot_type;
  98         u_int8_t ip6ot_len;
  99         u_int8_t ip6ot_encap_limit;
 100 } __packed;
 101
 102 /* Router Alert Option */
 103 struct ip6_opt_router {
 104         u_int8_t ip6or_type;
 105         u_int8_t ip6or_len;
 106         u_int8_t ip6or_value[2];
 107 } __packed;
 108 #endif /* INET6 */
 109
 110 #define PFFRAG_SEENLAST 0x0001          /* Seen the last fragment for this */
 111 #define PFFRAG_NOBUFFER 0x0002          /* Non-buffering fragment cache */
 112 #define PFFRAG_DROP     0x0004          /* Drop all fragments */
 113 #define BUFFER_FRAGMENTS(fr)    (!((fr)->fr_flags & PFFRAG_NOBUFFER))
 114
 115
 116 TAILQ_HEAD(pf_fragqueue, pf_fragment)   pf_fragqueue;
 117 TAILQ_HEAD(pf_cachequeue, pf_fragment)  pf_cachequeue;
 118
 119 static int              pf_frag_compare(struct pf_fragment *,
 120                                         struct pf_fragment *);
 121 RB_HEAD(pf_frag_tree, pf_fragment)      pf_frag_tree, pf_cache_tree;
 122 RB_PROTOTYPE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
 123 RB_GENERATE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
 124
 125 /* Private prototypes */
 126 void                     pf_ip2key(struct pf_fragment *, struct ip *);
 127 void                     pf_remove_fragment(struct pf_fragment *);
 128 void                     pf_flush_fragments(void);
 129 void                     pf_free_fragment(struct pf_fragment *);
 130 struct pf_fragment      *pf_find_fragment(struct ip *, struct pf_frag_tree *);
 131 struct mbuf             *pf_reassemble(struct mbuf **, struct pf_fragment **,
 132                             struct pf_frent *, int);
 133 struct mbuf             *pf_fragcache(struct mbuf **, struct ip*,
 134                             struct pf_fragment **, int, int, int *);
 135 u_int16_t                pf_cksum_fixup(u_int16_t, u_int16_t, u_int16_t);
 136 int                      pf_normalize_tcpopt(struct pf_rule *, struct mbuf *,
 137                             struct tcphdr *, int);
 138
 139 #define DPFPRINTF(x)    if (pf_status.debug >= PF_DEBUG_MISC) \
 140                             { kprintf("%s: ", __func__); kprintf x ;}
 141
 142 /* Globals */
 143 vm_zone_t                pf_frent_pl, pf_frag_pl, pf_cache_pl, pf_cent_pl;
 144 vm_zone_t                pf_state_scrub_pl;
 145 int                      pf_nfrents, pf_ncache;
 146
 147 void
 148 pf_normalize_init(void)
 149 {
 150         /* XXX
 151         pool_sethiwat(&pf_frag_pl, PFFRAG_FRAG_HIWAT);
 152         pool_sethardlimit(&pf_frent_pl, PFFRAG_FRENT_HIWAT, NULL, 0);
 153         pool_sethardlimit(&pf_cache_pl, PFFRAG_FRCACHE_HIWAT, NULL, 0);
 154         pool_sethardlimit(&pf_cent_pl, PFFRAG_FRCENT_HIWAT, NULL, 0);
 155         */
 156
 157         TAILQ_INIT(&pf_fragqueue);
 158         TAILQ_INIT(&pf_cachequeue);
 159 }
 160
 161 static int
 162 pf_frag_compare(struct pf_fragment *a, struct pf_fragment *b)
 163 {
 164         int     diff;
 165
 166         if ((diff = a->fr_id - b->fr_id))
 167                 return (diff);
 168         else if ((diff = a->fr_p - b->fr_p))
 169                 return (diff);
 170         else if (a->fr_src.s_addr < b->fr_src.s_addr)
 171                 return (-1);
 172         else if (a->fr_src.s_addr > b->fr_src.s_addr)
 173                 return (1);
 174         else if (a->fr_dst.s_addr < b->fr_dst.s_addr)
 175                 return (-1);
 176         else if (a->fr_dst.s_addr > b->fr_dst.s_addr)
 177                 return (1);
 178         return (0);
 179 }
 180
 181 void
 182 pf_purge_expired_fragments(void)
 183 {
 184         struct pf_fragment      *frag;
 185         u_int32_t                expire = time_second -
 186                                     pf_default_rule.timeout[PFTM_FRAG];
 187
 188         while ((frag = TAILQ_LAST(&pf_fragqueue, pf_fragqueue)) != NULL) {
 189                 KASSERT((BUFFER_FRAGMENTS(frag)),
 190                         ("BUFFER_FRAGMENTS(frag) == 0: %s", __func__));
 191                 if (frag->fr_timeout > expire)
 192                         break;
 193
 194                 DPFPRINTF(("expiring %d(%p)\n", frag->fr_id, frag));
 195                 pf_free_fragment(frag);
 196         }
 197
 198         while ((frag = TAILQ_LAST(&pf_cachequeue, pf_cachequeue)) != NULL) {
 199                 KASSERT((!BUFFER_FRAGMENTS(frag)),
 200                         ("BUFFER_FRAGMENTS(frag) != 0: %s", __func__));
 201                 if (frag->fr_timeout > expire)
 202                         break;
 203
 204                 DPFPRINTF(("expiring %d(%p)\n", frag->fr_id, frag));
 205                 pf_free_fragment(frag);
 206                 KASSERT((TAILQ_EMPTY(&pf_cachequeue) ||
 207                     TAILQ_LAST(&pf_cachequeue, pf_cachequeue) != frag),
 208                     ("!(TAILQ_EMPTY() || TAILQ_LAST() == farg): %s",
 209                     __func__));
 210         }
 211 }
 212
 213 /*
 214  * Try to flush old fragments to make space for new ones
 215  */
 216
 217 void
 218 pf_flush_fragments(void)
 219 {
 220         struct pf_fragment      *frag;
 221         int                      goal;
 222
 223         goal = pf_nfrents * 9 / 10;
 224         DPFPRINTF(("trying to free > %d frents\n",
 225             pf_nfrents - goal));
 226         while (goal < pf_nfrents) {
 227                 frag = TAILQ_LAST(&pf_fragqueue, pf_fragqueue);
 228                 if (frag == NULL)
 229                         break;
 230                 pf_free_fragment(frag);
 231         }
 232
 233
 234         goal = pf_ncache * 9 / 10;
 235         DPFPRINTF(("trying to free > %d cache entries\n",
 236             pf_ncache - goal));
 237         while (goal < pf_ncache) {
 238                 frag = TAILQ_LAST(&pf_cachequeue, pf_cachequeue);
 239                 if (frag == NULL)
 240                         break;
 241                 pf_free_fragment(frag);
 242         }
 243 }
 244
 245 /* Frees the fragments and all associated entries */
 246
 247 void
 248 pf_free_fragment(struct pf_fragment *frag)
 249 {
 250         struct pf_frent         *frent;
 251         struct pf_frcache       *frcache;
 252
 253         /* Free all fragments */
 254         if (BUFFER_FRAGMENTS(frag)) {
 255                 for (frent = LIST_FIRST(&frag->fr_queue); frent;
 256                     frent = LIST_FIRST(&frag->fr_queue)) {
 257                         LIST_REMOVE(frent, fr_next);
 258
 259                         m_freem(frent->fr_m);
 260                         pool_put(&pf_frent_pl, frent);
 261                         pf_nfrents--;
 262                 }
 263         } else {
 264                 for (frcache = LIST_FIRST(&frag->fr_cache); frcache;
 265                     frcache = LIST_FIRST(&frag->fr_cache)) {
 266                         LIST_REMOVE(frcache, fr_next);
 267
 268                         KASSERT((LIST_EMPTY(&frag->fr_cache) ||
 269                             LIST_FIRST(&frag->fr_cache)->fr_off >
 270                             frcache->fr_end),
 271                             ("! (LIST_EMPTY() || LIST_FIRST()->fr_off >"
 272                              " frcache->fr_end): %s", __func__));
 273
 274                         pool_put(&pf_cent_pl, frcache);
 275                         pf_ncache--;
 276                 }
 277         }
 278
 279         pf_remove_fragment(frag);
 280 }
 281
 282 void
 283 pf_ip2key(struct pf_fragment *key, struct ip *ip)
 284 {
 285         key->fr_p = ip->ip_p;
 286         key->fr_id = ip->ip_id;
 287         key->fr_src.s_addr = ip->ip_src.s_addr;
 288         key->fr_dst.s_addr = ip->ip_dst.s_addr;
 289 }
 290
 291 struct pf_fragment *
 292 pf_find_fragment(struct ip *ip, struct pf_frag_tree *tree)
 293 {
 294         struct pf_fragment       key;
 295         struct pf_fragment      *frag;
 296
 297         pf_ip2key(&key, ip);
 298
 299         frag = RB_FIND(pf_frag_tree, tree, &key);
 300         if (frag != NULL) {
 301                 /* XXX Are we sure we want to update the timeout? */
 302                 frag->fr_timeout = time_second;
 303                 if (BUFFER_FRAGMENTS(frag)) {
 304                         TAILQ_REMOVE(&pf_fragqueue, frag, frag_next);
 305                         TAILQ_INSERT_HEAD(&pf_fragqueue, frag, frag_next);
 306                 } else {
 307                         TAILQ_REMOVE(&pf_cachequeue, frag, frag_next);
 308                         TAILQ_INSERT_HEAD(&pf_cachequeue, frag, frag_next);
 309                 }
 310         }
 311
 312         return (frag);
 313 }
 314
 315 /* Removes a fragment from the fragment queue and frees the fragment */
 316
 317 void
 318 pf_remove_fragment(struct pf_fragment *frag)
 319 {
 320         if (BUFFER_FRAGMENTS(frag)) {
 321                 RB_REMOVE(pf_frag_tree, &pf_frag_tree, frag);
 322                 TAILQ_REMOVE(&pf_fragqueue, frag, frag_next);
 323                 pool_put(&pf_frag_pl, frag);
 324         } else {
 325                 RB_REMOVE(pf_frag_tree, &pf_cache_tree, frag);
 326                 TAILQ_REMOVE(&pf_cachequeue, frag, frag_next);
 327                 pool_put(&pf_cache_pl, frag);
 328         }
 329 }
 330
 331 #define FR_IP_OFF(fr)   (((fr)->fr_ip->ip_off & IP_OFFMASK) << 3)
 332 struct mbuf *
 333 pf_reassemble(struct mbuf **m0, struct pf_fragment **frag,
 334     struct pf_frent *frent, int mff)
 335 {
 336         struct mbuf     *m = *m0, *m2;
 337         struct pf_frent *frea, *next;
 338         struct pf_frent *frep = NULL;
 339         struct ip       *ip = frent->fr_ip;
 340         int              hlen = ip->ip_hl << 2;
 341         u_int16_t        off = (ip->ip_off & IP_OFFMASK) << 3;
 342         u_int16_t        ip_len = ip->ip_len - ip->ip_hl * 4;
 343         u_int16_t        max = ip_len + off;
 344
 345         KASSERT((*frag == NULL || BUFFER_FRAGMENTS(*frag)),
 346             ("! (*frag == NULL || BUFFER_FRAGMENTS(*frag)): %s", __func__));
 347
 348         /* Strip off ip header */
 349         m->m_data += hlen;
 350         m->m_len -= hlen;
 351
 352         /* Create a new reassembly queue for this packet */
 353         if (*frag == NULL) {
 354                 *frag = pool_get(&pf_frag_pl, PR_NOWAIT);
 355                 if (*frag == NULL) {
 356                         pf_flush_fragments();
 357                         *frag = pool_get(&pf_frag_pl, PR_NOWAIT);
 358                         if (*frag == NULL)
 359                                 goto drop_fragment;
 360                 }
 361
 362                 (*frag)->fr_flags = 0;
 363                 (*frag)->fr_max = 0;
 364                 (*frag)->fr_src = frent->fr_ip->ip_src;
 365                 (*frag)->fr_dst = frent->fr_ip->ip_dst;
 366                 (*frag)->fr_p = frent->fr_ip->ip_p;
 367                 (*frag)->fr_id = frent->fr_ip->ip_id;
 368                 (*frag)->fr_timeout = time_second;
 369                 LIST_INIT(&(*frag)->fr_queue);
 370
 371                 RB_INSERT(pf_frag_tree, &pf_frag_tree, *frag);
 372                 TAILQ_INSERT_HEAD(&pf_fragqueue, *frag, frag_next);
 373
 374                 /* We do not have a previous fragment */
 375                 frep = NULL;
 376                 goto insert;
 377         }
 378
 379         /*
 380          * Find a fragment after the current one:
 381          *  - off contains the real shifted offset.
 382          */
 383         LIST_FOREACH(frea, &(*frag)->fr_queue, fr_next) {
 384                 if (FR_IP_OFF(frea) > off)
 385                         break;
 386                 frep = frea;
 387         }
 388
 389         KASSERT((frep != NULL || frea != NULL),
 390             ("!(frep != NULL || frea != NULL): %s", __func__));
 391
 392         if (frep != NULL &&
 393             FR_IP_OFF(frep) + frep->fr_ip->ip_len - frep->fr_ip->ip_hl *
 394             4 > off)
 395         {
 396                 u_int16_t       precut;
 397
 398                 precut = FR_IP_OFF(frep) + frep->fr_ip->ip_len -
 399                     frep->fr_ip->ip_hl * 4 - off;
 400                 if (precut >= ip_len)
 401                         goto drop_fragment;
 402                 m_adj(frent->fr_m, precut);
 403                 DPFPRINTF(("overlap -%d\n", precut));
 404                 /* Enforce 8 byte boundaries */
 405                 ip->ip_off = ip->ip_off + (precut >> 3);
 406                 off = (ip->ip_off & IP_OFFMASK) << 3;
 407                 ip_len -= precut;
 408                 ip->ip_len = ip_len;
 409         }
 410
 411         for (; frea != NULL && ip_len + off > FR_IP_OFF(frea);
 412             frea = next)
 413         {
 414                 u_int16_t       aftercut;
 415
 416                 aftercut = ip_len + off - FR_IP_OFF(frea);
 417                 DPFPRINTF(("adjust overlap %d\n", aftercut));
 418                 if (aftercut < frea->fr_ip->ip_len - frea->fr_ip->ip_hl
 419                     * 4)
 420                 {
 421                         frea->fr_ip->ip_len =
 422                             frea->fr_ip->ip_len - aftercut;
 423                         frea->fr_ip->ip_off = frea->fr_ip->ip_off +
 424                             (aftercut >> 3);
 425                         m_adj(frea->fr_m, aftercut);
 426                         break;
 427                 }
 428
 429                 /* This fragment is completely overlapped, loose it */
 430                 next = LIST_NEXT(frea, fr_next);
 431                 m_freem(frea->fr_m);
 432                 LIST_REMOVE(frea, fr_next);
 433                 pool_put(&pf_frent_pl, frea);
 434                 pf_nfrents--;
 435         }
 436
 437  insert:
 438         /* Update maximum data size */
 439         if ((*frag)->fr_max < max)
 440                 (*frag)->fr_max = max;
 441         /* This is the last segment */
 442         if (!mff)
 443                 (*frag)->fr_flags |= PFFRAG_SEENLAST;
 444
 445         if (frep == NULL)
 446                 LIST_INSERT_HEAD(&(*frag)->fr_queue, frent, fr_next);
 447         else
 448                 LIST_INSERT_AFTER(frep, frent, fr_next);
 449
 450         /* Check if we are completely reassembled */
 451         if (!((*frag)->fr_flags & PFFRAG_SEENLAST))
 452                 return (NULL);
 453
 454         /* Check if we have all the data */
 455         off = 0;
 456         for (frep = LIST_FIRST(&(*frag)->fr_queue); frep; frep = next) {
 457                 next = LIST_NEXT(frep, fr_next);
 458
 459                 off += frep->fr_ip->ip_len - frep->fr_ip->ip_hl * 4;
 460                 if (off < (*frag)->fr_max &&
 461                     (next == NULL || FR_IP_OFF(next) != off))
 462                 {
 463                         DPFPRINTF(("missing fragment at %d, next %d, max %d\n",
 464                             off, next == NULL ? -1 : FR_IP_OFF(next),
 465                             (*frag)->fr_max));
 466                         return (NULL);
 467                 }
 468         }
 469         DPFPRINTF(("%d < %d?\n", off, (*frag)->fr_max));
 470         if (off < (*frag)->fr_max)
 471                 return (NULL);
 472
 473         /* We have all the data */
 474         frent = LIST_FIRST(&(*frag)->fr_queue);
 475         KASSERT((frent != NULL), ("frent == NULL: %s", __func__));
 476         if ((frent->fr_ip->ip_hl << 2) + off > IP_MAXPACKET) {
 477                 DPFPRINTF(("drop: too big: %d\n", off));
 478                 pf_free_fragment(*frag);
 479                 *frag = NULL;
 480                 return (NULL);
 481         }
 482         next = LIST_NEXT(frent, fr_next);
 483
 484         /* Magic from ip_input */
 485         ip = frent->fr_ip;
 486         m = frent->fr_m;
 487         m2 = m->m_next;
 488         m->m_next = NULL;
 489         m_cat(m, m2);
 490         pool_put(&pf_frent_pl, frent);
 491         pf_nfrents--;
 492         for (frent = next; frent != NULL; frent = next) {
 493                 next = LIST_NEXT(frent, fr_next);
 494
 495                 m2 = frent->fr_m;
 496                 pool_put(&pf_frent_pl, frent);
 497                 pf_nfrents--;
 498                 m->m_pkthdr.csum_flags &= m2->m_pkthdr.csum_flags;
 499                 m->m_pkthdr.csum_data += m2->m_pkthdr.csum_data;
 500                 m_cat(m, m2);
 501         }
 502
 503         /*
 504          * Note: this 1's complement optimization with <= 65535 fragments.
 505          *
 506          * Handle 1's complement carry for the 16 bit result.  This can
 507          * result in another carry which must also be handled.
 508          */
 509         m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) +
 510                                 (m->m_pkthdr.csum_data >> 16);
 511         if (m->m_pkthdr.csum_data > 0xFFFF)
 512                 m->m_pkthdr.csum_data -= 0xFFFF;
 513
 514
 515         ip->ip_src = (*frag)->fr_src;
 516         ip->ip_dst = (*frag)->fr_dst;
 517
 518         /* Remove from fragment queue */
 519         pf_remove_fragment(*frag);
 520         *frag = NULL;
 521
 522         hlen = ip->ip_hl << 2;
 523         ip->ip_len = off + hlen;
 524         m->m_len += hlen;
 525         m->m_data -= hlen;
 526
 527         /* some debugging cruft by sklower, below, will go away soon */
 528         /* XXX this should be done elsewhere */
 529         if (m->m_flags & M_PKTHDR) {
 530                 int plen = 0;
 531                 for (m2 = m; m2; m2 = m2->m_next)
 532                         plen += m2->m_len;
 533                 m->m_pkthdr.len = plen;
 534         }
 535
 536         DPFPRINTF(("complete: %p(%d)\n", m, ip->ip_len));
 537         return (m);
 538
 539  drop_fragment:
 540         /* Oops - fail safe - drop packet */
 541         pool_put(&pf_frent_pl, frent);
 542         pf_nfrents--;
 543         m_freem(m);
 544         return (NULL);
 545 }
 546
 547 struct mbuf *
 548 pf_fragcache(struct mbuf **m0, struct ip *h, struct pf_fragment **frag, int mff,
 549     int drop, int *nomem)
 550 {
 551         struct mbuf             *m = *m0;
 552         struct pf_frcache       *frp, *fra, *cur = NULL;
 553         int                      ip_len = h->ip_len - (h->ip_hl << 2);
 554         u_int16_t                off = h->ip_off << 3;
 555         u_int16_t                max = ip_len + off;
 556         int                      hosed = 0;
 557
 558         KASSERT((*frag == NULL || !BUFFER_FRAGMENTS(*frag)),
 559             ("!(*frag == NULL || !BUFFER_FRAGMENTS(*frag)): %s", __func__));
 560
 561         /* Create a new range queue for this packet */
 562         if (*frag == NULL) {
 563                 *frag = pool_get(&pf_cache_pl, PR_NOWAIT);
 564                 if (*frag == NULL) {
 565                         pf_flush_fragments();
 566                         *frag = pool_get(&pf_cache_pl, PR_NOWAIT);
 567                         if (*frag == NULL)
 568                                 goto no_mem;
 569                 }
 570
 571                 /* Get an entry for the queue */
 572                 cur = pool_get(&pf_cent_pl, PR_NOWAIT);
 573                 if (cur == NULL) {
 574                         pool_put(&pf_cache_pl, *frag);
 575                         *frag = NULL;
 576                         goto no_mem;
 577                 }
 578                 pf_ncache++;
 579
 580                 (*frag)->fr_flags = PFFRAG_NOBUFFER;
 581                 (*frag)->fr_max = 0;
 582                 (*frag)->fr_src = h->ip_src;
 583                 (*frag)->fr_dst = h->ip_dst;
 584                 (*frag)->fr_p = h->ip_p;
 585                 (*frag)->fr_id = h->ip_id;
 586                 (*frag)->fr_timeout = time_second;
 587
 588                 cur->fr_off = off;
 589                 cur->fr_end = max;
 590                 LIST_INIT(&(*frag)->fr_cache);
 591                 LIST_INSERT_HEAD(&(*frag)->fr_cache, cur, fr_next);
 592
 593                 RB_INSERT(pf_frag_tree, &pf_cache_tree, *frag);
 594                 TAILQ_INSERT_HEAD(&pf_cachequeue, *frag, frag_next);
 595
 596                 DPFPRINTF(("fragcache[%d]: new %d-%d\n", h->ip_id, off, max));
 597
 598                 goto pass;
 599         }
 600
 601         /*
 602          * Find a fragment after the current one:
 603          *  - off contains the real shifted offset.
 604          */
 605         frp = NULL;
 606         LIST_FOREACH(fra, &(*frag)->fr_cache, fr_next) {
 607                 if (fra->fr_off > off)
 608                         break;
 609                 frp = fra;
 610         }
 611
 612         KASSERT((frp != NULL || fra != NULL),
 613             ("!(frp != NULL || fra != NULL): %s", __func__));
 614
 615         if (frp != NULL) {
 616                 int     precut;
 617
 618                 precut = frp->fr_end - off;
 619                 if (precut >= ip_len) {
 620                         /* Fragment is entirely a duplicate */
 621                         DPFPRINTF(("fragcache[%d]: dead (%d-%d) %d-%d\n",
 622                             h->ip_id, frp->fr_off, frp->fr_end, off, max));
 623                         goto drop_fragment;
 624                 }
 625                 if (precut == 0) {
 626                         /* They are adjacent.  Fixup cache entry */
 627                         DPFPRINTF(("fragcache[%d]: adjacent (%d-%d) %d-%d\n",
 628                             h->ip_id, frp->fr_off, frp->fr_end, off, max));
 629                         frp->fr_end = max;
 630                 } else if (precut > 0) {
 631                         /* The first part of this payload overlaps with a
 632                          * fragment that has already been passed.
 633                          * Need to trim off the first part of the payload.
 634                          * But to do so easily, we need to create another
 635                          * mbuf to throw the original header into.
 636                          */
 637
 638                         DPFPRINTF(("fragcache[%d]: chop %d (%d-%d) %d-%d\n",
 639                             h->ip_id, precut, frp->fr_off, frp->fr_end, off,
 640                             max));
 641
 642                         off += precut;
 643                         max -= precut;
 644                         /* Update the previous frag to encompass this one */
 645                         frp->fr_end = max;
 646
 647                         if (!drop) {
 648                                 /* XXX Optimization opportunity
 649                                  * This is a very heavy way to trim the payload.
 650                                  * we could do it much faster by diddling mbuf
 651                                  * internals but that would be even less legible
 652                                  * than this mbuf magic.  For my next trick,
 653                                  * I'll pull a rabbit out of my laptop.
 654                                  */
 655                                 *m0 = m_dup(m, MB_DONTWAIT);
 656                                 /* From KAME Project : We have missed this! */
 657                                 m_adj(*m0, (h->ip_hl << 2) -
 658                                     (*m0)->m_pkthdr.len);
 659                                 if (*m0 == NULL)
 660                                         goto no_mem;
 661                                 KASSERT(((*m0)->m_next == NULL),
 662                                     ("(*m0)->m_next != NULL: %s",
 663                                     __func__));
 664                                 m_adj(m, precut + (h->ip_hl << 2));
 665                                 m_cat(*m0, m);
 666                                 m = *m0;
 667                                 if (m->m_flags & M_PKTHDR) {
 668                                         int plen = 0;
 669                                         struct mbuf *t;
 670                                         for (t = m; t; t = t->m_next)
 671                                                 plen += t->m_len;
 672                                         m->m_pkthdr.len = plen;
 673                                 }
 674
 675
 676                                 h = mtod(m, struct ip *);
 677
 678                                 KASSERT(((int)m->m_len ==
 679                                     h->ip_len - precut),
 680                                     ("m->m_len != h->ip_len - precut: %s",
 681                                     __func__));
 682                                 h->ip_off = h->ip_off +
 683                                     (precut >> 3);
 684                                 h->ip_len = h->ip_len - precut;
 685                         } else {
 686                                 hosed++;
 687                         }
 688                 } else {
 689                         /* There is a gap between fragments */
 690
 691                         DPFPRINTF(("fragcache[%d]: gap %d (%d-%d) %d-%d\n",
 692                             h->ip_id, -precut, frp->fr_off, frp->fr_end, off,
 693                             max));
 694
 695                         cur = pool_get(&pf_cent_pl, PR_NOWAIT);
 696                         if (cur == NULL)
 697                                 goto no_mem;
 698                         pf_ncache++;
 699
 700                         cur->fr_off = off;
 701                         cur->fr_end = max;
 702                         LIST_INSERT_AFTER(frp, cur, fr_next);
 703                 }
 704         }
 705
 706         if (fra != NULL) {
 707                 int     aftercut;
 708                 int     merge = 0;
 709
 710                 aftercut = max - fra->fr_off;
 711                 if (aftercut == 0) {
 712                         /* Adjacent fragments */
 713                         DPFPRINTF(("fragcache[%d]: adjacent %d-%d (%d-%d)\n",
 714                             h->ip_id, off, max, fra->fr_off, fra->fr_end));
 715                         fra->fr_off = off;
 716                         merge = 1;
 717                 } else if (aftercut > 0) {
 718                         /* Need to chop off the tail of this fragment */
 719                         DPFPRINTF(("fragcache[%d]: chop %d %d-%d (%d-%d)\n",
 720                             h->ip_id, aftercut, off, max, fra->fr_off,
 721                             fra->fr_end));
 722                         fra->fr_off = off;
 723                         max -= aftercut;
 724
 725                         merge = 1;
 726
 727                         if (!drop) {
 728                                 m_adj(m, -aftercut);
 729                                 if (m->m_flags & M_PKTHDR) {
 730                                         int plen = 0;
 731                                         struct mbuf *t;
 732                                         for (t = m; t; t = t->m_next)
 733                                                 plen += t->m_len;
 734                                         m->m_pkthdr.len = plen;
 735                                 }
 736                                 h = mtod(m, struct ip *);
 737                                 KASSERT(((int)m->m_len == h->ip_len - aftercut),
 738                                     ("m->m_len != h->ip_len - aftercut: %s",
 739                                     __func__));
 740                                 h->ip_len = h->ip_len - aftercut;
 741                         } else {
 742                                 hosed++;
 743                         }
 744                 } else if (frp == NULL) {
 745                         /* There is a gap between fragments */
 746                         DPFPRINTF(("fragcache[%d]: gap %d %d-%d (%d-%d)\n",
 747                             h->ip_id, -aftercut, off, max, fra->fr_off,
 748                             fra->fr_end));
 749
 750                         cur = pool_get(&pf_cent_pl, PR_NOWAIT);
 751                         if (cur == NULL)
 752                                 goto no_mem;
 753                         pf_ncache++;
 754
 755                         cur->fr_off = off;
 756                         cur->fr_end = max;
 757                         LIST_INSERT_BEFORE(fra, cur, fr_next);
 758                 }
 759
 760
 761                 /* Need to glue together two separate fragment descriptors */
 762                 if (merge) {
 763                         if (cur && fra->fr_off <= cur->fr_end) {
 764                                 /* Need to merge in a previous 'cur' */
 765                                 DPFPRINTF(("fragcache[%d]: adjacent(merge "
 766                                     "%d-%d) %d-%d (%d-%d)\n",
 767                                     h->ip_id, cur->fr_off, cur->fr_end, off,
 768                                     max, fra->fr_off, fra->fr_end));
 769                                 fra->fr_off = cur->fr_off;
 770                                 LIST_REMOVE(cur, fr_next);
 771                                 pool_put(&pf_cent_pl, cur);
 772                                 pf_ncache--;
 773                                 cur = NULL;
 774
 775                         } else if (frp && fra->fr_off <= frp->fr_end) {
 776                                 /* Need to merge in a modified 'frp' */
 777                                 KASSERT((cur == NULL), ("cur != NULL: %s",
 778                                     __func__));
 779                                 DPFPRINTF(("fragcache[%d]: adjacent(merge "
 780                                     "%d-%d) %d-%d (%d-%d)\n",
 781                                     h->ip_id, frp->fr_off, frp->fr_end, off,
 782                                     max, fra->fr_off, fra->fr_end));
 783                                 fra->fr_off = frp->fr_off;
 784                                 LIST_REMOVE(frp, fr_next);
 785                                 pool_put(&pf_cent_pl, frp);
 786                                 pf_ncache--;
 787                                 frp = NULL;
 788
 789                         }
 790                 }
 791         }
 792
 793         if (hosed) {
 794                 /*
 795                  * We must keep tracking the overall fragment even when
 796                  * we're going to drop it anyway so that we know when to
 797                  * free the overall descriptor.  Thus we drop the frag late.
 798                  */
 799                 goto drop_fragment;
 800         }
 801
 802
 803  pass:
 804         /* Update maximum data size */
 805         if ((*frag)->fr_max < max)
 806                 (*frag)->fr_max = max;
 807
 808         /* This is the last segment */
 809         if (!mff)
 810                 (*frag)->fr_flags |= PFFRAG_SEENLAST;
 811
 812         /* Check if we are completely reassembled */
 813         if (((*frag)->fr_flags & PFFRAG_SEENLAST) &&
 814             LIST_FIRST(&(*frag)->fr_cache)->fr_off == 0 &&
 815             LIST_FIRST(&(*frag)->fr_cache)->fr_end == (*frag)->fr_max) {
 816                 /* Remove from fragment queue */
 817                 DPFPRINTF(("fragcache[%d]: done 0-%d\n", h->ip_id,
 818                     (*frag)->fr_max));
 819                 pf_free_fragment(*frag);
 820                 *frag = NULL;
 821         }
 822
 823         return (m);
 824
 825  no_mem:
 826         *nomem = 1;
 827
 828         /* Still need to pay attention to !IP_MF */
 829         if (!mff && *frag != NULL)
 830                 (*frag)->fr_flags |= PFFRAG_SEENLAST;
 831
 832         m_freem(m);
 833         return (NULL);
 834
 835  drop_fragment:
 836
 837         /* Still need to pay attention to !IP_MF */
 838         if (!mff && *frag != NULL)
 839                 (*frag)->fr_flags |= PFFRAG_SEENLAST;
 840
 841         if (drop) {
 842                 /* This fragment has been deemed bad.  Don't reass */
 843                 if (((*frag)->fr_flags & PFFRAG_DROP) == 0)
 844                         DPFPRINTF(("fragcache[%d]: dropping overall fragment\n",
 845                             h->ip_id));
 846                 (*frag)->fr_flags |= PFFRAG_DROP;
 847         }
 848
 849         m_freem(m);
 850         return (NULL);
 851 }
 852
 853 int
 854 pf_normalize_ip(struct mbuf **m0, int dir, struct pfi_kif *kif, u_short *reason)
 855 {
 856         struct mbuf             *m = *m0;
 857         struct pf_rule          *r;
 858         struct pf_frent         *frent;
 859         struct pf_fragment      *frag = NULL;
 860         struct ip               *h = mtod(m, struct ip *);
 861         int                      mff = (h->ip_off & IP_MF);
 862         int                      hlen = h->ip_hl << 2;
 863         u_int16_t                fragoff = (h->ip_off & IP_OFFMASK) << 3;
 864         u_int16_t                max;
 865         int                      ip_len;
 866         int                      ip_off;
 867
 868         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
 869         while (r != NULL) {
 870                 r->evaluations++;
 871                 if (r->kif != NULL &&
 872                     (r->kif != kif && r->kif != kif->pfik_parent) == !r->ifnot)
 873                         r = r->skip[PF_SKIP_IFP].ptr;
 874                 else if (r->direction && r->direction != dir)
 875                         r = r->skip[PF_SKIP_DIR].ptr;
 876                 else if (r->af && r->af != AF_INET)
 877                         r = r->skip[PF_SKIP_AF].ptr;
 878                 else if (r->proto && r->proto != h->ip_p)
 879                         r = r->skip[PF_SKIP_PROTO].ptr;
 880                 else if (PF_MISMATCHAW(&r->src.addr,
 881                     (struct pf_addr *)&h->ip_src.s_addr, AF_INET, r->src.not))
 882                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
 883                 else if (PF_MISMATCHAW(&r->dst.addr,
 884                     (struct pf_addr *)&h->ip_dst.s_addr, AF_INET, r->dst.not))
 885                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
 886                 else
 887                         break;
 888         }
 889
 890         if (r == NULL)
 891                 return (PF_PASS);
 892         else
 893                 r->packets++;
 894
 895         /* Check for illegal packets */
 896         if (hlen < (int)sizeof(struct ip))
 897                 goto drop;
 898
 899         if (hlen > h->ip_len)
 900                 goto drop;
 901
 902         /* Clear IP_DF if the rule uses the no-df option */
 903         if (r->rule_flag & PFRULE_NODF)
 904                 h->ip_off &= ~IP_DF;
 905
 906         /* We will need other tests here */
 907         if (!fragoff && !mff)
 908                 goto no_fragment;
 909
 910         /* We're dealing with a fragment now. Don't allow fragments
 911          * with IP_DF to enter the cache. If the flag was cleared by
 912          * no-df above, fine. Otherwise drop it.
 913          */
 914         if (h->ip_off & IP_DF) {
 915                 DPFPRINTF(("IP_DF\n"));
 916                 goto bad;
 917         }
 918
 919         ip_len = h->ip_len - hlen;
 920         ip_off = (h->ip_off & IP_OFFMASK) << 3;
 921
 922         /* All fragments are 8 byte aligned */
 923         if (mff && (ip_len & 0x7)) {
 924                 DPFPRINTF(("mff and %d\n", ip_len));
 925                 goto bad;
 926         }
 927
 928         /* Respect maximum length */
 929         if (fragoff + ip_len > IP_MAXPACKET) {
 930                 DPFPRINTF(("max packet %d\n", fragoff + ip_len));
 931                 goto bad;
 932         }
 933         max = fragoff + ip_len;
 934
 935         if ((r->rule_flag & (PFRULE_FRAGCROP|PFRULE_FRAGDROP)) == 0) {
 936                 /* Fully buffer all of the fragments */
 937
 938                 frag = pf_find_fragment(h, &pf_frag_tree);
 939
 940                 /* Check if we saw the last fragment already */
 941                 if (frag != NULL && (frag->fr_flags & PFFRAG_SEENLAST) &&
 942                     max > frag->fr_max)
 943                         goto bad;
 944
 945                 /* Get an entry for the fragment queue */
 946                 frent = pool_get(&pf_frent_pl, PR_NOWAIT);
 947                 if (frent == NULL) {
 948                         REASON_SET(reason, PFRES_MEMORY);
 949                         return (PF_DROP);
 950                 }
 951                 pf_nfrents++;
 952                 frent->fr_ip = h;
 953                 frent->fr_m = m;
 954
 955                 /* Might return a completely reassembled mbuf, or NULL */
 956                 DPFPRINTF(("reass frag %d @ %d-%d\n", h->ip_id, fragoff, max));
 957                 *m0 = m = pf_reassemble(m0, &frag, frent, mff);
 958
 959                 if (m == NULL)
 960                         return (PF_DROP);
 961
 962                 if (frag != NULL && (frag->fr_flags & PFFRAG_DROP))
 963                         goto drop;
 964
 965                 h = mtod(m, struct ip *);
 966         } else {
 967                 /* non-buffering fragment cache (drops or masks overlaps) */
 968                 int     nomem = 0;
 969
 970                 if (dir == PF_OUT) {
 971                         if (m->m_pkthdr.fw_flags & PF_MBUF_FRAGCACHE) {
 972                                 /* Already passed the fragment cache in the
 973                                  * input direction.  If we continued, it would
 974                                  * appear to be a dup and would be dropped.
 975                                  */
 976                                 goto fragment_pass;
 977                         }
 978                 }
 979
 980                 frag = pf_find_fragment(h, &pf_cache_tree);
 981
 982                 /* Check if we saw the last fragment already */
 983                 if (frag != NULL && (frag->fr_flags & PFFRAG_SEENLAST) &&
 984                     max > frag->fr_max) {
 985                         if (r->rule_flag & PFRULE_FRAGDROP)
 986                                 frag->fr_flags |= PFFRAG_DROP;
 987                         goto bad;
 988                 }
 989
 990                 *m0 = m = pf_fragcache(m0, h, &frag, mff,
 991                     (r->rule_flag & PFRULE_FRAGDROP) ? 1 : 0, &nomem);
 992                 if (m == NULL) {
 993                         if (nomem)
 994                                 goto no_mem;
 995                         goto drop;
 996                 }
 997
 998                 if (dir == PF_IN)
 999                         m->m_pkthdr.fw_flags |= PF_MBUF_FRAGCACHE;
1000
1001                 if (frag != NULL && (frag->fr_flags & PFFRAG_DROP))
1002                         goto drop;
1003                 goto fragment_pass;
1004         }
1005
1006  no_fragment:
1007         /* At this point, only IP_DF is allowed in ip_off */
1008         h->ip_off &= IP_DF;
1009
1010         /* Enforce a minimum ttl, may cause endless packet loops */
1011         if (r->min_ttl && h->ip_ttl < r->min_ttl)
1012                 h->ip_ttl = r->min_ttl;
1013
1014         if (r->rule_flag & PFRULE_RANDOMID) {
1015 #ifdef RANDOM_IP_ID
1016                 h->ip_id = ip_randomid();
1017 #else
1018                 h->ip_id = htons(ip_id++);
1019 #endif
1020         }
1021
1022         return (PF_PASS);
1023
1024  fragment_pass:
1025         /* Enforce a minimum ttl, may cause endless packet loops */
1026         if (r->min_ttl && h->ip_ttl < r->min_ttl)
1027                 h->ip_ttl = r->min_ttl;
1028
1029         return (PF_PASS);
1030
1031  no_mem:
1032         REASON_SET(reason, PFRES_MEMORY);
1033         if (r != NULL && r->log)
1034                 PFLOG_PACKET(kif, h, m, AF_INET, dir, *reason, r, NULL, NULL);
1035         return (PF_DROP);
1036
1037  drop:
1038         REASON_SET(reason, PFRES_NORM);
1039         if (r != NULL && r->log)
1040                 PFLOG_PACKET(kif, h, m, AF_INET, dir, *reason, r, NULL, NULL);
1041         return (PF_DROP);
1042
1043  bad:
1044         DPFPRINTF(("dropping bad fragment\n"));
1045
1046         /* Free associated fragments */
1047         if (frag != NULL)
1048                 pf_free_fragment(frag);
1049
1050         REASON_SET(reason, PFRES_FRAG);
1051         if (r != NULL && r->log)
1052                 PFLOG_PACKET(kif, h, m, AF_INET, dir, *reason, r, NULL, NULL);
1053
1054         return (PF_DROP);
1055 }
1056
1057 #ifdef INET6
1058 int
1059 pf_normalize_ip6(struct mbuf **m0, int dir, struct pfi_kif *kif,
1060     u_short *reason)
1061 {
1062         struct mbuf             *m = *m0;
1063         struct pf_rule          *r;
1064         struct ip6_hdr          *h = mtod(m, struct ip6_hdr *);
1065         int                      off;
1066         struct ip6_ext           ext;
1067         struct ip6_opt           opt;
1068         struct ip6_opt_jumbo     jumbo;
1069         struct ip6_frag          frag;
1070         u_int32_t                jumbolen = 0, plen;
1071         u_int16_t                fragoff = 0;
1072         int                      optend;
1073         int                      ooff;
1074         u_int8_t                 proto;
1075         int                      terminal;
1076
1077         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1078         while (r != NULL) {
1079                 r->evaluations++;
1080                 if (r->kif != NULL &&
1081                     (r->kif != kif && r->kif != kif->pfik_parent) == !r->ifnot)
1082                         r = r->skip[PF_SKIP_IFP].ptr;
1083                 else if (r->direction && r->direction != dir)
1084                         r = r->skip[PF_SKIP_DIR].ptr;
1085                 else if (r->af && r->af != AF_INET6)
1086                         r = r->skip[PF_SKIP_AF].ptr;
1087 #if 0 /* header chain! */
1088                 else if (r->proto && r->proto != h->ip6_nxt)
1089                         r = r->skip[PF_SKIP_PROTO].ptr;
1090 #endif
1091                 else if (PF_MISMATCHAW(&r->src.addr,
1092                     (struct pf_addr *)&h->ip6_src, AF_INET6, r->src.not))
1093                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
1094                 else if (PF_MISMATCHAW(&r->dst.addr,
1095                     (struct pf_addr *)&h->ip6_dst, AF_INET6, r->dst.not))
1096                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
1097                 else
1098                         break;
1099         }
1100
1101         if (r == NULL)
1102                 return (PF_PASS);
1103         else
1104                 r->packets++;
1105
1106         /* Check for illegal packets */
1107         if (sizeof(struct ip6_hdr) + IPV6_MAXPACKET < m->m_pkthdr.len)
1108                 goto drop;
1109
1110         off = sizeof(struct ip6_hdr);
1111         proto = h->ip6_nxt;
1112         terminal = 0;
1113         do {
1114                 switch (proto) {
1115                 case IPPROTO_FRAGMENT:
1116                         goto fragment;
1117                         break;
1118                 case IPPROTO_AH:
1119                 case IPPROTO_ROUTING:
1120                 case IPPROTO_DSTOPTS:
1121                         if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL,
1122                             NULL, AF_INET6))
1123                                 goto shortpkt;
1124                         if (proto == IPPROTO_AH)
1125                                 off += (ext.ip6e_len + 2) * 4;
1126                         else
1127                                 off += (ext.ip6e_len + 1) * 8;
1128                         proto = ext.ip6e_nxt;
1129                         break;
1130                 case IPPROTO_HOPOPTS:
1131                         if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL,
1132                             NULL, AF_INET6))
1133                                 goto shortpkt;
1134                         optend = off + (ext.ip6e_len + 1) * 8;
1135                         ooff = off + sizeof(ext);
1136                         do {
1137                                 if (!pf_pull_hdr(m, ooff, &opt.ip6o_type,
1138                                     sizeof(opt.ip6o_type), NULL, NULL,
1139                                     AF_INET6))
1140                                         goto shortpkt;
1141                                 if (opt.ip6o_type == IP6OPT_PAD1) {
1142                                         ooff++;
1143                                         continue;
1144                                 }
1145                                 if (!pf_pull_hdr(m, ooff, &opt, sizeof(opt),
1146                                     NULL, NULL, AF_INET6))
1147                                         goto shortpkt;
1148                                 if (ooff + sizeof(opt) + opt.ip6o_len > optend)
1149                                         goto drop;
1150                                 switch (opt.ip6o_type) {
1151                                 case IP6OPT_JUMBO:
1152                                         if (h->ip6_plen != 0)
1153                                                 goto drop;
1154                                         if (!pf_pull_hdr(m, ooff, &jumbo,
1155                                             sizeof(jumbo), NULL, NULL,
1156                                             AF_INET6))
1157                                                 goto shortpkt;
1158                                         memcpy(&jumbolen, jumbo.ip6oj_jumbo_len,
1159                                             sizeof(jumbolen));
1160                                         jumbolen = ntohl(jumbolen);
1161                                         if (jumbolen <= IPV6_MAXPACKET)
1162                                                 goto drop;
1163                                         if (sizeof(struct ip6_hdr) + jumbolen !=
1164                                             m->m_pkthdr.len)
1165                                                 goto drop;
1166                                         break;
1167                                 default:
1168                                         break;
1169                                 }
1170                                 ooff += sizeof(opt) + opt.ip6o_len;
1171                         } while (ooff < optend);
1172
1173                         off = optend;
1174                         proto = ext.ip6e_nxt;
1175                         break;
1176                 default:
1177                         terminal = 1;
1178                         break;
1179                 }
1180         } while (!terminal);
1181
1182         /* jumbo payload option must be present, or plen > 0 */
1183         if (ntohs(h->ip6_plen) == 0)
1184                 plen = jumbolen;
1185         else
1186                 plen = ntohs(h->ip6_plen);
1187         if (plen == 0)
1188                 goto drop;
1189         if (sizeof(struct ip6_hdr) + plen > m->m_pkthdr.len)
1190                 goto shortpkt;
1191
1192         /* Enforce a minimum ttl, may cause endless packet loops */
1193         if (r->min_ttl && h->ip6_hlim < r->min_ttl)
1194                 h->ip6_hlim = r->min_ttl;
1195
1196         return (PF_PASS);
1197
1198  fragment:
1199         if (ntohs(h->ip6_plen) == 0 || jumbolen)
1200                 goto drop;
1201         plen = ntohs(h->ip6_plen);
1202
1203         if (!pf_pull_hdr(m, off, &frag, sizeof(frag), NULL, NULL, AF_INET6))
1204                 goto shortpkt;
1205         fragoff = ntohs(frag.ip6f_offlg & IP6F_OFF_MASK);
1206         if (fragoff + (plen - off - sizeof(frag)) > IPV6_MAXPACKET)
1207                 goto badfrag;
1208
1209         /* do something about it */
1210         return (PF_PASS);
1211
1212  shortpkt:
1213         REASON_SET(reason, PFRES_SHORT);
1214         if (r != NULL && r->log)
1215                 PFLOG_PACKET(kif, h, m, AF_INET6, dir, *reason, r, NULL, NULL);
1216         return (PF_DROP);
1217
1218  drop:
1219         REASON_SET(reason, PFRES_NORM);
1220         if (r != NULL && r->log)
1221                 PFLOG_PACKET(kif, h, m, AF_INET6, dir, *reason, r, NULL, NULL);
1222         return (PF_DROP);
1223
1224  badfrag:
1225         REASON_SET(reason, PFRES_FRAG);
1226         if (r != NULL && r->log)
1227                 PFLOG_PACKET(kif, h, m, AF_INET6, dir, *reason, r, NULL, NULL);
1228         return (PF_DROP);
1229 }
1230 #endif
1231
1232 int
1233 pf_normalize_tcp(int dir, struct pfi_kif *kif, struct mbuf *m, int ipoff,
1234     int off, void *h, struct pf_pdesc *pd)
1235 {
1236         struct pf_rule  *r, *rm = NULL;
1237         struct tcphdr   *th = pd->hdr.tcp;
1238         int              rewrite = 0;
1239         u_short          reason;
1240         u_int8_t         flags;
1241         sa_family_t      af = pd->af;
1242
1243         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1244         while (r != NULL) {
1245                 r->evaluations++;
1246                 if (r->kif != NULL &&
1247                     (r->kif != kif && r->kif != kif->pfik_parent) == !r->ifnot)
1248                         r = r->skip[PF_SKIP_IFP].ptr;
1249                 else if (r->direction && r->direction != dir)
1250                         r = r->skip[PF_SKIP_DIR].ptr;
1251                 else if (r->af && r->af != af)
1252                         r = r->skip[PF_SKIP_AF].ptr;
1253                 else if (r->proto && r->proto != pd->proto)
1254                         r = r->skip[PF_SKIP_PROTO].ptr;
1255                 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af, r->src.not))
1256                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
1257                 else if (r->src.port_op && !pf_match_port(r->src.port_op,
1258                             r->src.port[0], r->src.port[1], th->th_sport))
1259                         r = r->skip[PF_SKIP_SRC_PORT].ptr;
1260                 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af, r->dst.not))
1261                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
1262                 else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
1263                             r->dst.port[0], r->dst.port[1], th->th_dport))
1264                         r = r->skip[PF_SKIP_DST_PORT].ptr;
1265                 else if (r->os_fingerprint != PF_OSFP_ANY && !pf_osfp_match(
1266                             pf_osfp_fingerprint(pd, m, off, th),
1267                             r->os_fingerprint))
1268                         r = TAILQ_NEXT(r, entries);
1269                 else {
1270                         rm = r;
1271                         break;
1272                 }
1273         }
1274
1275         if (rm == NULL)
1276                 return (PF_PASS);
1277         else
1278                 r->packets++;
1279
1280         if (rm->rule_flag & PFRULE_REASSEMBLE_TCP)
1281                 pd->flags |= PFDESC_TCP_NORM;
1282
1283         flags = th->th_flags;
1284         if (flags & TH_SYN) {
1285                 /* Illegal packet */
1286                 if (flags & TH_RST)
1287                         goto tcp_drop;
1288
1289                 if (flags & TH_FIN)
1290                         flags &= ~TH_FIN;
1291         } else {
1292                 /* Illegal packet */
1293                 if (!(flags & (TH_ACK|TH_RST)))
1294                         goto tcp_drop;
1295         }
1296
1297         if (!(flags & TH_ACK)) {
1298                 /* These flags are only valid if ACK is set */
1299                 if ((flags & TH_FIN) || (flags & TH_PUSH) || (flags & TH_URG))
1300                         goto tcp_drop;
1301         }
1302
1303         /* Check for illegal header length */
1304         if (th->th_off < (sizeof(struct tcphdr) >> 2))
1305                 goto tcp_drop;
1306
1307         /* If flags changed, or reserved data set, then adjust */
1308         if (flags != th->th_flags || th->th_x2 != 0) {
1309                 u_int16_t       ov, nv;
1310
1311                 ov = *(u_int16_t *)(&th->th_ack + 1);
1312                 th->th_flags = flags;
1313                 th->th_x2 = 0;
1314                 nv = *(u_int16_t *)(&th->th_ack + 1);
1315
1316                 th->th_sum = pf_cksum_fixup(th->th_sum, ov, nv);
1317                 rewrite = 1;
1318         }
1319
1320         /* Remove urgent pointer, if TH_URG is not set */
1321         if (!(flags & TH_URG) && th->th_urp) {
1322                 th->th_sum = pf_cksum_fixup(th->th_sum, th->th_urp, 0);
1323                 th->th_urp = 0;
1324                 rewrite = 1;
1325         }
1326
1327         /* Process options */
1328         if (r->max_mss && pf_normalize_tcpopt(r, m, th, off))
1329                 rewrite = 1;
1330
1331         /* copy back packet headers if we sanitized */
1332         if (rewrite)
1333                 m_copyback(m, off, sizeof(*th), (caddr_t)th);
1334
1335         return (PF_PASS);
1336
1337  tcp_drop:
1338         REASON_SET(&reason, PFRES_NORM);
1339         if (rm != NULL && r->log)
1340                 PFLOG_PACKET(kif, h, m, AF_INET, dir, reason, r, NULL, NULL);
1341         return (PF_DROP);
1342 }
1343
1344 int
1345 pf_normalize_tcp_init(struct mbuf *m, int off, struct pf_pdesc *pd,
1346     struct tcphdr *th, struct pf_state_peer *src, struct pf_state_peer *dst)
1347 {
1348         u_int8_t hdr[60];
1349         u_int8_t *opt;
1350
1351         KASSERT((src->scrub == NULL),
1352             ("pf_normalize_tcp_init: src->scrub != NULL"));
1353
1354         src->scrub = pool_get(&pf_state_scrub_pl, PR_NOWAIT);
1355         if (src->scrub == NULL)
1356                 return (1);
1357         bzero(src->scrub, sizeof(*src->scrub));
1358
1359         switch (pd->af) {
1360 #ifdef INET
1361         case AF_INET: {
1362                 struct ip *h = mtod(m, struct ip *);
1363                 src->scrub->pfss_ttl = h->ip_ttl;
1364                 break;
1365         }
1366 #endif /* INET */
1367 #ifdef INET6
1368         case AF_INET6: {
1369                 struct ip6_hdr *h = mtod(m, struct ip6_hdr *);
1370                 src->scrub->pfss_ttl = h->ip6_hlim;
1371                 break;
1372         }
1373 #endif /* INET6 */
1374         }
1375
1376
1377         /*
1378          * All normalizations below are only begun if we see the start of
1379          * the connections.  They must all set an enabled bit in pfss_flags
1380          */
1381         if ((th->th_flags & TH_SYN) == 0)
1382                 return (0);
1383
1384
1385         if (th->th_off > (sizeof(struct tcphdr) >> 2) && src->scrub &&
1386             pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) {
1387                 /* Diddle with TCP options */
1388                 int hlen;
1389                 opt = hdr + sizeof(struct tcphdr);
1390                 hlen = (th->th_off << 2) - sizeof(struct tcphdr);
1391                 while (hlen >= TCPOLEN_TIMESTAMP) {
1392                         switch (*opt) {
1393                         case TCPOPT_EOL:        /* FALLTHROUGH */
1394                         case TCPOPT_NOP:
1395                                 opt++;
1396                                 hlen--;
1397                                 break;
1398                         case TCPOPT_TIMESTAMP:
1399                                 if (opt[1] >= TCPOLEN_TIMESTAMP) {
1400                                         src->scrub->pfss_flags |=
1401                                             PFSS_TIMESTAMP;
1402                                         src->scrub->pfss_ts_mod = karc4random();
1403                                 }
1404                                 /* FALLTHROUGH */
1405                         default:
1406                                 hlen -= MAX(opt[1], 2);
1407                                 opt += MAX(opt[1], 2);
1408                                 break;
1409                         }
1410                 }
1411         }
1412
1413         return (0);
1414 }
1415
1416 void
1417 pf_normalize_tcp_cleanup(struct pf_state *state)
1418 {
1419         if (state->src.scrub)
1420                 pool_put(&pf_state_scrub_pl, state->src.scrub);
1421         if (state->dst.scrub)
1422                 pool_put(&pf_state_scrub_pl, state->dst.scrub);
1423
1424         /* Someday... flush the TCP segment reassembly descriptors. */
1425 }
1426
1427 int
1428 pf_normalize_tcp_stateful(struct mbuf *m, int off, struct pf_pdesc *pd,
1429     u_short *reason, struct tcphdr *th, struct pf_state_peer *src,
1430     struct pf_state_peer *dst, int *writeback)
1431 {
1432         u_int8_t hdr[60];
1433         u_int8_t *opt;
1434         int copyback = 0;
1435
1436         KASSERT((src->scrub || dst->scrub),
1437             ("pf_normalize_tcp_statefull: src->scrub && dst->scrub!"));
1438
1439         /*
1440          * Enforce the minimum TTL seen for this connection.  Negate a common
1441          * technique to evade an intrusion detection system and confuse
1442          * firewall state code.
1443          */
1444         switch (pd->af) {
1445 #ifdef INET
1446         case AF_INET: {
1447                 if (src->scrub) {
1448                         struct ip *h = mtod(m, struct ip *);
1449                         if (h->ip_ttl > src->scrub->pfss_ttl)
1450                                 src->scrub->pfss_ttl = h->ip_ttl;
1451                         h->ip_ttl = src->scrub->pfss_ttl;
1452                 }
1453                 break;
1454         }
1455 #endif /* INET */
1456 #ifdef INET6
1457         case AF_INET6: {
1458                 if (src->scrub) {
1459                         struct ip6_hdr *h = mtod(m, struct ip6_hdr *);
1460                         if (h->ip6_hlim > src->scrub->pfss_ttl)
1461                                 src->scrub->pfss_ttl = h->ip6_hlim;
1462                         h->ip6_hlim = src->scrub->pfss_ttl;
1463                 }
1464                 break;
1465         }
1466 #endif /* INET6 */
1467         }
1468
1469         if (th->th_off > (sizeof(struct tcphdr) >> 2) &&
1470             ((src->scrub && (src->scrub->pfss_flags & PFSS_TIMESTAMP)) ||
1471             (dst->scrub && (dst->scrub->pfss_flags & PFSS_TIMESTAMP))) &&
1472             pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) {
1473                 /* Diddle with TCP options */
1474                 int hlen;
1475                 opt = hdr + sizeof(struct tcphdr);
1476                 hlen = (th->th_off << 2) - sizeof(struct tcphdr);
1477                 while (hlen >= TCPOLEN_TIMESTAMP) {
1478                         switch (*opt) {
1479                         case TCPOPT_EOL:        /* FALLTHROUGH */
1480                         case TCPOPT_NOP:
1481                                 opt++;
1482                                 hlen--;
1483                                 break;
1484                         case TCPOPT_TIMESTAMP:
1485                                 /* Modulate the timestamps.  Can be used for
1486                                  * NAT detection, OS uptime determination or
1487                                  * reboot detection.
1488                                  */
1489                                 if (opt[1] >= TCPOLEN_TIMESTAMP) {
1490                                         u_int32_t ts_value;
1491                                         if (src->scrub &&
1492                                             (src->scrub->pfss_flags &
1493                                             PFSS_TIMESTAMP)) {
1494                                                 memcpy(&ts_value, &opt[2],
1495                                                     sizeof(u_int32_t));
1496                                                 ts_value = htonl(ntohl(ts_value)
1497                                                     + src->scrub->pfss_ts_mod);
1498                                                 pf_change_a(&opt[2],
1499                                                     &th->th_sum, ts_value, 0);
1500                                                 copyback = 1;
1501                                         }
1502
1503                                         /* Modulate TS reply iff valid (!0) */
1504                                         memcpy(&ts_value, &opt[6],
1505                                             sizeof(u_int32_t));
1506                                         if (ts_value && dst->scrub &&
1507                                             (dst->scrub->pfss_flags &
1508                                             PFSS_TIMESTAMP)) {
1509                                                 ts_value = htonl(ntohl(ts_value)
1510                                                     - dst->scrub->pfss_ts_mod);
1511                                                 pf_change_a(&opt[6],
1512                                                     &th->th_sum, ts_value, 0);
1513                                                 copyback = 1;
1514                                         }
1515                                 }
1516                                 /* FALLTHROUGH */
1517                         default:
1518                                 hlen -= MAX(opt[1], 2);
1519                                 opt += MAX(opt[1], 2);
1520                                 break;
1521                         }
1522                 }
1523                 if (copyback) {
1524                         /* Copyback the options, caller copys back header */
1525                         *writeback = 1;
1526                         m_copyback(m, off + sizeof(struct tcphdr),
1527                             (th->th_off << 2) - sizeof(struct tcphdr), hdr +
1528                             sizeof(struct tcphdr));
1529                 }
1530         }
1531
1532
1533         /* I have a dream....  TCP segment reassembly.... */
1534         return (0);
1535 }
1536
1537 int
1538 pf_normalize_tcpopt(struct pf_rule *r, struct mbuf *m, struct tcphdr *th,
1539     int off)
1540 {
1541         u_int16_t       *mss;
1542         int              thoff;
1543         int              opt, cnt, optlen = 0;
1544         int              rewrite = 0;
1545         u_char          *optp;
1546
1547         thoff = th->th_off << 2;
1548         cnt = thoff - sizeof(struct tcphdr);
1549         optp = mtod(m, caddr_t) + off + sizeof(struct tcphdr);
1550
1551         for (; cnt > 0; cnt -= optlen, optp += optlen) {
1552                 opt = optp[0];
1553                 if (opt == TCPOPT_EOL)
1554                         break;
1555                 if (opt == TCPOPT_NOP)
1556                         optlen = 1;
1557                 else {
1558                         if (cnt < 2)
1559                                 break;
1560                         optlen = optp[1];
1561                         if (optlen < 2 || optlen > cnt)
1562                                 break;
1563                 }
1564                 switch (opt) {
1565                 case TCPOPT_MAXSEG:
1566                         mss = (u_int16_t *)(optp + 2);
1567                         if ((ntohs(*mss)) > r->max_mss) {
1568                                 th->th_sum = pf_cksum_fixup(th->th_sum,
1569                                     *mss, htons(r->max_mss));
1570                                 *mss = htons(r->max_mss);
1571                                 rewrite = 1;
1572                         }
1573                         break;
1574                 default:
1575                         break;
1576                 }
1577         }
1578
1579         return (rewrite);
1580 }