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 1.10 2008/09/04 09:08:22 hasso Exp $ */
7 * Copyright (c) 2004 The DragonFly Project. All rights reserved.
9 * Copyright 2001 Niels Provos <provos@citi.umich.edu>
10 * All rights reserved.
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
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.
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.
34 #include "opt_inet6.h"
36 #include <sys/param.h>
37 #include <sys/systm.h>
39 #include <sys/filio.h>
40 #include <sys/fcntl.h>
41 #include <sys/socket.h>
42 #include <sys/kernel.h>
44 #include <vm/vm_zone.h>
47 #include <net/if_types.h>
49 #include <net/route.h>
50 #include <net/pf/if_pflog.h>
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>
63 #include <netinet/ip6.h>
66 #include <net/pf/pfvar.h>
70 * XXX: This should go to netinet/ip6.h (KAME)
74 #define PFFRAG_SEENLAST 0x0001 /* Seen the last fragment for this */
75 #define PFFRAG_NOBUFFER 0x0002 /* Non-buffering fragment cache */
76 #define PFFRAG_DROP 0x0004 /* Drop all fragments */
77 #define BUFFER_FRAGMENTS(fr) (!((fr)->fr_flags & PFFRAG_NOBUFFER))
80 TAILQ_HEAD(pf_fragqueue, pf_fragment) pf_fragqueue;
81 TAILQ_HEAD(pf_cachequeue, pf_fragment) pf_cachequeue;
83 static int pf_frag_compare(struct pf_fragment *,
84 struct pf_fragment *);
85 RB_HEAD(pf_frag_tree, pf_fragment) pf_frag_tree, pf_cache_tree;
86 RB_PROTOTYPE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
87 RB_GENERATE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
89 /* Private prototypes */
90 void pf_ip2key(struct pf_fragment *, struct ip *);
91 void pf_remove_fragment(struct pf_fragment *);
92 void pf_flush_fragments(void);
93 void pf_free_fragment(struct pf_fragment *);
94 struct pf_fragment *pf_find_fragment(struct ip *, struct pf_frag_tree *);
95 struct mbuf *pf_reassemble(struct mbuf **, struct pf_fragment **,
96 struct pf_frent *, int);
97 struct mbuf *pf_fragcache(struct mbuf **, struct ip*,
98 struct pf_fragment **, int, int, int *);
99 u_int16_t pf_cksum_fixup(u_int16_t, u_int16_t, u_int16_t);
100 int pf_normalize_tcpopt(struct pf_rule *, struct mbuf *,
101 struct tcphdr *, int);
103 #define DPFPRINTF(x) if (pf_status.debug >= PF_DEBUG_MISC) \
104 { kprintf("%s: ", __func__); kprintf x ;}
107 vm_zone_t pf_frent_pl, pf_frag_pl, pf_cache_pl, pf_cent_pl;
108 vm_zone_t pf_state_scrub_pl;
109 int pf_nfrents, pf_ncache;
112 pf_normalize_init(void)
115 pool_sethiwat(&pf_frag_pl, PFFRAG_FRAG_HIWAT);
116 pool_sethardlimit(&pf_frent_pl, PFFRAG_FRENT_HIWAT, NULL, 0);
117 pool_sethardlimit(&pf_cache_pl, PFFRAG_FRCACHE_HIWAT, NULL, 0);
118 pool_sethardlimit(&pf_cent_pl, PFFRAG_FRCENT_HIWAT, NULL, 0);
121 TAILQ_INIT(&pf_fragqueue);
122 TAILQ_INIT(&pf_cachequeue);
126 pf_frag_compare(struct pf_fragment *a, struct pf_fragment *b)
130 if ((diff = a->fr_id - b->fr_id))
132 else if ((diff = a->fr_p - b->fr_p))
134 else if (a->fr_src.s_addr < b->fr_src.s_addr)
136 else if (a->fr_src.s_addr > b->fr_src.s_addr)
138 else if (a->fr_dst.s_addr < b->fr_dst.s_addr)
140 else if (a->fr_dst.s_addr > b->fr_dst.s_addr)
146 pf_purge_expired_fragments(void)
148 struct pf_fragment *frag;
149 u_int32_t expire = time_second -
150 pf_default_rule.timeout[PFTM_FRAG];
152 while ((frag = TAILQ_LAST(&pf_fragqueue, pf_fragqueue)) != NULL) {
153 KASSERT((BUFFER_FRAGMENTS(frag)),
154 ("BUFFER_FRAGMENTS(frag) == 0: %s", __func__));
155 if (frag->fr_timeout > expire)
158 DPFPRINTF(("expiring %d(%p)\n", frag->fr_id, frag));
159 pf_free_fragment(frag);
162 while ((frag = TAILQ_LAST(&pf_cachequeue, pf_cachequeue)) != NULL) {
163 KASSERT((!BUFFER_FRAGMENTS(frag)),
164 ("BUFFER_FRAGMENTS(frag) != 0: %s", __func__));
165 if (frag->fr_timeout > expire)
168 DPFPRINTF(("expiring %d(%p)\n", frag->fr_id, frag));
169 pf_free_fragment(frag);
170 KASSERT((TAILQ_EMPTY(&pf_cachequeue) ||
171 TAILQ_LAST(&pf_cachequeue, pf_cachequeue) != frag),
172 ("!(TAILQ_EMPTY() || TAILQ_LAST() == farg): %s",
178 * Try to flush old fragments to make space for new ones
182 pf_flush_fragments(void)
184 struct pf_fragment *frag;
187 goal = pf_nfrents * 9 / 10;
188 DPFPRINTF(("trying to free > %d frents\n",
190 while (goal < pf_nfrents) {
191 frag = TAILQ_LAST(&pf_fragqueue, pf_fragqueue);
194 pf_free_fragment(frag);
198 goal = pf_ncache * 9 / 10;
199 DPFPRINTF(("trying to free > %d cache entries\n",
201 while (goal < pf_ncache) {
202 frag = TAILQ_LAST(&pf_cachequeue, pf_cachequeue);
205 pf_free_fragment(frag);
209 /* Frees the fragments and all associated entries */
212 pf_free_fragment(struct pf_fragment *frag)
214 struct pf_frent *frent;
215 struct pf_frcache *frcache;
217 /* Free all fragments */
218 if (BUFFER_FRAGMENTS(frag)) {
219 for (frent = LIST_FIRST(&frag->fr_queue); frent;
220 frent = LIST_FIRST(&frag->fr_queue)) {
221 LIST_REMOVE(frent, fr_next);
223 m_freem(frent->fr_m);
224 pool_put(&pf_frent_pl, frent);
228 for (frcache = LIST_FIRST(&frag->fr_cache); frcache;
229 frcache = LIST_FIRST(&frag->fr_cache)) {
230 LIST_REMOVE(frcache, fr_next);
232 KASSERT((LIST_EMPTY(&frag->fr_cache) ||
233 LIST_FIRST(&frag->fr_cache)->fr_off >
235 ("! (LIST_EMPTY() || LIST_FIRST()->fr_off >"
236 " frcache->fr_end): %s", __func__));
238 pool_put(&pf_cent_pl, frcache);
243 pf_remove_fragment(frag);
247 pf_ip2key(struct pf_fragment *key, struct ip *ip)
249 key->fr_p = ip->ip_p;
250 key->fr_id = ip->ip_id;
251 key->fr_src.s_addr = ip->ip_src.s_addr;
252 key->fr_dst.s_addr = ip->ip_dst.s_addr;
256 pf_find_fragment(struct ip *ip, struct pf_frag_tree *tree)
258 struct pf_fragment key;
259 struct pf_fragment *frag;
263 frag = RB_FIND(pf_frag_tree, tree, &key);
265 /* XXX Are we sure we want to update the timeout? */
266 frag->fr_timeout = time_second;
267 if (BUFFER_FRAGMENTS(frag)) {
268 TAILQ_REMOVE(&pf_fragqueue, frag, frag_next);
269 TAILQ_INSERT_HEAD(&pf_fragqueue, frag, frag_next);
271 TAILQ_REMOVE(&pf_cachequeue, frag, frag_next);
272 TAILQ_INSERT_HEAD(&pf_cachequeue, frag, frag_next);
279 /* Removes a fragment from the fragment queue and frees the fragment */
282 pf_remove_fragment(struct pf_fragment *frag)
284 if (BUFFER_FRAGMENTS(frag)) {
285 RB_REMOVE(pf_frag_tree, &pf_frag_tree, frag);
286 TAILQ_REMOVE(&pf_fragqueue, frag, frag_next);
287 pool_put(&pf_frag_pl, frag);
289 RB_REMOVE(pf_frag_tree, &pf_cache_tree, frag);
290 TAILQ_REMOVE(&pf_cachequeue, frag, frag_next);
291 pool_put(&pf_cache_pl, frag);
295 #define FR_IP_OFF(fr) (((fr)->fr_ip->ip_off & IP_OFFMASK) << 3)
297 pf_reassemble(struct mbuf **m0, struct pf_fragment **frag,
298 struct pf_frent *frent, int mff)
300 struct mbuf *m = *m0, *m2;
301 struct pf_frent *frea, *next;
302 struct pf_frent *frep = NULL;
303 struct ip *ip = frent->fr_ip;
304 int hlen = ip->ip_hl << 2;
305 u_int16_t off = (ip->ip_off & IP_OFFMASK) << 3;
306 u_int16_t ip_len = ip->ip_len - ip->ip_hl * 4;
307 u_int16_t max = ip_len + off;
309 KASSERT((*frag == NULL || BUFFER_FRAGMENTS(*frag)),
310 ("! (*frag == NULL || BUFFER_FRAGMENTS(*frag)): %s", __func__));
312 /* Strip off ip header */
316 /* Create a new reassembly queue for this packet */
318 *frag = pool_get(&pf_frag_pl, PR_NOWAIT);
320 pf_flush_fragments();
321 *frag = pool_get(&pf_frag_pl, PR_NOWAIT);
326 (*frag)->fr_flags = 0;
328 (*frag)->fr_src = frent->fr_ip->ip_src;
329 (*frag)->fr_dst = frent->fr_ip->ip_dst;
330 (*frag)->fr_p = frent->fr_ip->ip_p;
331 (*frag)->fr_id = frent->fr_ip->ip_id;
332 (*frag)->fr_timeout = time_second;
333 LIST_INIT(&(*frag)->fr_queue);
335 RB_INSERT(pf_frag_tree, &pf_frag_tree, *frag);
336 TAILQ_INSERT_HEAD(&pf_fragqueue, *frag, frag_next);
338 /* We do not have a previous fragment */
344 * Find a fragment after the current one:
345 * - off contains the real shifted offset.
347 LIST_FOREACH(frea, &(*frag)->fr_queue, fr_next) {
348 if (FR_IP_OFF(frea) > off)
353 KASSERT((frep != NULL || frea != NULL),
354 ("!(frep != NULL || frea != NULL): %s", __func__));
357 FR_IP_OFF(frep) + frep->fr_ip->ip_len - frep->fr_ip->ip_hl *
362 precut = FR_IP_OFF(frep) + frep->fr_ip->ip_len -
363 frep->fr_ip->ip_hl * 4 - off;
364 if (precut >= ip_len)
366 m_adj(frent->fr_m, precut);
367 DPFPRINTF(("overlap -%d\n", precut));
368 /* Enforce 8 byte boundaries */
369 ip->ip_off = ip->ip_off + (precut >> 3);
370 off = (ip->ip_off & IP_OFFMASK) << 3;
375 for (; frea != NULL && ip_len + off > FR_IP_OFF(frea);
380 aftercut = ip_len + off - FR_IP_OFF(frea);
381 DPFPRINTF(("adjust overlap %d\n", aftercut));
382 if (aftercut < frea->fr_ip->ip_len - frea->fr_ip->ip_hl
385 frea->fr_ip->ip_len =
386 frea->fr_ip->ip_len - aftercut;
387 frea->fr_ip->ip_off = frea->fr_ip->ip_off +
389 m_adj(frea->fr_m, aftercut);
393 /* This fragment is completely overlapped, loose it */
394 next = LIST_NEXT(frea, fr_next);
396 LIST_REMOVE(frea, fr_next);
397 pool_put(&pf_frent_pl, frea);
402 /* Update maximum data size */
403 if ((*frag)->fr_max < max)
404 (*frag)->fr_max = max;
405 /* This is the last segment */
407 (*frag)->fr_flags |= PFFRAG_SEENLAST;
410 LIST_INSERT_HEAD(&(*frag)->fr_queue, frent, fr_next);
412 LIST_INSERT_AFTER(frep, frent, fr_next);
414 /* Check if we are completely reassembled */
415 if (!((*frag)->fr_flags & PFFRAG_SEENLAST))
418 /* Check if we have all the data */
420 for (frep = LIST_FIRST(&(*frag)->fr_queue); frep; frep = next) {
421 next = LIST_NEXT(frep, fr_next);
423 off += frep->fr_ip->ip_len - frep->fr_ip->ip_hl * 4;
424 if (off < (*frag)->fr_max &&
425 (next == NULL || FR_IP_OFF(next) != off))
427 DPFPRINTF(("missing fragment at %d, next %d, max %d\n",
428 off, next == NULL ? -1 : FR_IP_OFF(next),
433 DPFPRINTF(("%d < %d?\n", off, (*frag)->fr_max));
434 if (off < (*frag)->fr_max)
437 /* We have all the data */
438 frent = LIST_FIRST(&(*frag)->fr_queue);
439 KASSERT((frent != NULL), ("frent == NULL: %s", __func__));
440 if ((frent->fr_ip->ip_hl << 2) + off > IP_MAXPACKET) {
441 DPFPRINTF(("drop: too big: %d\n", off));
442 pf_free_fragment(*frag);
446 next = LIST_NEXT(frent, fr_next);
448 /* Magic from ip_input */
454 pool_put(&pf_frent_pl, frent);
456 for (frent = next; frent != NULL; frent = next) {
457 next = LIST_NEXT(frent, fr_next);
460 pool_put(&pf_frent_pl, frent);
462 m->m_pkthdr.csum_flags &= m2->m_pkthdr.csum_flags;
463 m->m_pkthdr.csum_data += m2->m_pkthdr.csum_data;
468 * Note: this 1's complement optimization with <= 65535 fragments.
470 * Handle 1's complement carry for the 16 bit result. This can
471 * result in another carry which must also be handled.
473 m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) +
474 (m->m_pkthdr.csum_data >> 16);
475 if (m->m_pkthdr.csum_data > 0xFFFF)
476 m->m_pkthdr.csum_data -= 0xFFFF;
479 ip->ip_src = (*frag)->fr_src;
480 ip->ip_dst = (*frag)->fr_dst;
482 /* Remove from fragment queue */
483 pf_remove_fragment(*frag);
486 hlen = ip->ip_hl << 2;
487 ip->ip_len = off + hlen;
491 /* some debugging cruft by sklower, below, will go away soon */
492 /* XXX this should be done elsewhere */
493 if (m->m_flags & M_PKTHDR) {
495 for (m2 = m; m2; m2 = m2->m_next)
497 m->m_pkthdr.len = plen;
500 DPFPRINTF(("complete: %p(%d)\n", m, ip->ip_len));
504 /* Oops - fail safe - drop packet */
505 pool_put(&pf_frent_pl, frent);
512 pf_fragcache(struct mbuf **m0, struct ip *h, struct pf_fragment **frag, int mff,
513 int drop, int *nomem)
515 struct mbuf *m = *m0;
516 struct pf_frcache *frp, *fra, *cur = NULL;
517 int ip_len = h->ip_len - (h->ip_hl << 2);
518 u_int16_t off = h->ip_off << 3;
519 u_int16_t max = ip_len + off;
522 KASSERT((*frag == NULL || !BUFFER_FRAGMENTS(*frag)),
523 ("!(*frag == NULL || !BUFFER_FRAGMENTS(*frag)): %s", __func__));
525 /* Create a new range queue for this packet */
527 *frag = pool_get(&pf_cache_pl, PR_NOWAIT);
529 pf_flush_fragments();
530 *frag = pool_get(&pf_cache_pl, PR_NOWAIT);
535 /* Get an entry for the queue */
536 cur = pool_get(&pf_cent_pl, PR_NOWAIT);
538 pool_put(&pf_cache_pl, *frag);
544 (*frag)->fr_flags = PFFRAG_NOBUFFER;
546 (*frag)->fr_src = h->ip_src;
547 (*frag)->fr_dst = h->ip_dst;
548 (*frag)->fr_p = h->ip_p;
549 (*frag)->fr_id = h->ip_id;
550 (*frag)->fr_timeout = time_second;
554 LIST_INIT(&(*frag)->fr_cache);
555 LIST_INSERT_HEAD(&(*frag)->fr_cache, cur, fr_next);
557 RB_INSERT(pf_frag_tree, &pf_cache_tree, *frag);
558 TAILQ_INSERT_HEAD(&pf_cachequeue, *frag, frag_next);
560 DPFPRINTF(("fragcache[%d]: new %d-%d\n", h->ip_id, off, max));
566 * Find a fragment after the current one:
567 * - off contains the real shifted offset.
570 LIST_FOREACH(fra, &(*frag)->fr_cache, fr_next) {
571 if (fra->fr_off > off)
576 KASSERT((frp != NULL || fra != NULL),
577 ("!(frp != NULL || fra != NULL): %s", __func__));
582 precut = frp->fr_end - off;
583 if (precut >= ip_len) {
584 /* Fragment is entirely a duplicate */
585 DPFPRINTF(("fragcache[%d]: dead (%d-%d) %d-%d\n",
586 h->ip_id, frp->fr_off, frp->fr_end, off, max));
590 /* They are adjacent. Fixup cache entry */
591 DPFPRINTF(("fragcache[%d]: adjacent (%d-%d) %d-%d\n",
592 h->ip_id, frp->fr_off, frp->fr_end, off, max));
594 } else if (precut > 0) {
595 /* The first part of this payload overlaps with a
596 * fragment that has already been passed.
597 * Need to trim off the first part of the payload.
598 * But to do so easily, we need to create another
599 * mbuf to throw the original header into.
602 DPFPRINTF(("fragcache[%d]: chop %d (%d-%d) %d-%d\n",
603 h->ip_id, precut, frp->fr_off, frp->fr_end, off,
608 /* Update the previous frag to encompass this one */
612 /* XXX Optimization opportunity
613 * This is a very heavy way to trim the payload.
614 * we could do it much faster by diddling mbuf
615 * internals but that would be even less legible
616 * than this mbuf magic. For my next trick,
617 * I'll pull a rabbit out of my laptop.
619 *m0 = m_dup(m, MB_DONTWAIT);
620 /* From KAME Project : We have missed this! */
621 m_adj(*m0, (h->ip_hl << 2) -
622 (*m0)->m_pkthdr.len);
625 KASSERT(((*m0)->m_next == NULL),
626 ("(*m0)->m_next != NULL: %s",
628 m_adj(m, precut + (h->ip_hl << 2));
631 if (m->m_flags & M_PKTHDR) {
634 for (t = m; t; t = t->m_next)
636 m->m_pkthdr.len = plen;
640 h = mtod(m, struct ip *);
642 KASSERT(((int)m->m_len ==
644 ("m->m_len != h->ip_len - precut: %s",
646 h->ip_off = h->ip_off +
648 h->ip_len = h->ip_len - precut;
653 /* There is a gap between fragments */
655 DPFPRINTF(("fragcache[%d]: gap %d (%d-%d) %d-%d\n",
656 h->ip_id, -precut, frp->fr_off, frp->fr_end, off,
659 cur = pool_get(&pf_cent_pl, PR_NOWAIT);
666 LIST_INSERT_AFTER(frp, cur, fr_next);
674 aftercut = max - fra->fr_off;
676 /* Adjacent fragments */
677 DPFPRINTF(("fragcache[%d]: adjacent %d-%d (%d-%d)\n",
678 h->ip_id, off, max, fra->fr_off, fra->fr_end));
681 } else if (aftercut > 0) {
682 /* Need to chop off the tail of this fragment */
683 DPFPRINTF(("fragcache[%d]: chop %d %d-%d (%d-%d)\n",
684 h->ip_id, aftercut, off, max, fra->fr_off,
693 if (m->m_flags & M_PKTHDR) {
696 for (t = m; t; t = t->m_next)
698 m->m_pkthdr.len = plen;
700 h = mtod(m, struct ip *);
701 KASSERT(((int)m->m_len == h->ip_len - aftercut),
702 ("m->m_len != h->ip_len - aftercut: %s",
704 h->ip_len = h->ip_len - aftercut;
708 } else if (frp == NULL) {
709 /* There is a gap between fragments */
710 DPFPRINTF(("fragcache[%d]: gap %d %d-%d (%d-%d)\n",
711 h->ip_id, -aftercut, off, max, fra->fr_off,
714 cur = pool_get(&pf_cent_pl, PR_NOWAIT);
721 LIST_INSERT_BEFORE(fra, cur, fr_next);
725 /* Need to glue together two separate fragment descriptors */
727 if (cur && fra->fr_off <= cur->fr_end) {
728 /* Need to merge in a previous 'cur' */
729 DPFPRINTF(("fragcache[%d]: adjacent(merge "
730 "%d-%d) %d-%d (%d-%d)\n",
731 h->ip_id, cur->fr_off, cur->fr_end, off,
732 max, fra->fr_off, fra->fr_end));
733 fra->fr_off = cur->fr_off;
734 LIST_REMOVE(cur, fr_next);
735 pool_put(&pf_cent_pl, cur);
739 } else if (frp && fra->fr_off <= frp->fr_end) {
740 /* Need to merge in a modified 'frp' */
741 KASSERT((cur == NULL), ("cur != NULL: %s",
743 DPFPRINTF(("fragcache[%d]: adjacent(merge "
744 "%d-%d) %d-%d (%d-%d)\n",
745 h->ip_id, frp->fr_off, frp->fr_end, off,
746 max, fra->fr_off, fra->fr_end));
747 fra->fr_off = frp->fr_off;
748 LIST_REMOVE(frp, fr_next);
749 pool_put(&pf_cent_pl, frp);
759 * We must keep tracking the overall fragment even when
760 * we're going to drop it anyway so that we know when to
761 * free the overall descriptor. Thus we drop the frag late.
768 /* Update maximum data size */
769 if ((*frag)->fr_max < max)
770 (*frag)->fr_max = max;
772 /* This is the last segment */
774 (*frag)->fr_flags |= PFFRAG_SEENLAST;
776 /* Check if we are completely reassembled */
777 if (((*frag)->fr_flags & PFFRAG_SEENLAST) &&
778 LIST_FIRST(&(*frag)->fr_cache)->fr_off == 0 &&
779 LIST_FIRST(&(*frag)->fr_cache)->fr_end == (*frag)->fr_max) {
780 /* Remove from fragment queue */
781 DPFPRINTF(("fragcache[%d]: done 0-%d\n", h->ip_id,
783 pf_free_fragment(*frag);
792 /* Still need to pay attention to !IP_MF */
793 if (!mff && *frag != NULL)
794 (*frag)->fr_flags |= PFFRAG_SEENLAST;
801 /* Still need to pay attention to !IP_MF */
802 if (!mff && *frag != NULL)
803 (*frag)->fr_flags |= PFFRAG_SEENLAST;
806 /* This fragment has been deemed bad. Don't reass */
807 if (((*frag)->fr_flags & PFFRAG_DROP) == 0)
808 DPFPRINTF(("fragcache[%d]: dropping overall fragment\n",
810 (*frag)->fr_flags |= PFFRAG_DROP;
818 pf_normalize_ip(struct mbuf **m0, int dir, struct pfi_kif *kif, u_short *reason)
820 struct mbuf *m = *m0;
822 struct pf_frent *frent;
823 struct pf_fragment *frag = NULL;
824 struct ip *h = mtod(m, struct ip *);
825 int mff = (h->ip_off & IP_MF);
826 int hlen = h->ip_hl << 2;
827 u_int16_t fragoff = (h->ip_off & IP_OFFMASK) << 3;
832 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
835 if (r->kif != NULL &&
836 (r->kif != kif && r->kif != kif->pfik_parent) == !r->ifnot)
837 r = r->skip[PF_SKIP_IFP].ptr;
838 else if (r->direction && r->direction != dir)
839 r = r->skip[PF_SKIP_DIR].ptr;
840 else if (r->af && r->af != AF_INET)
841 r = r->skip[PF_SKIP_AF].ptr;
842 else if (r->proto && r->proto != h->ip_p)
843 r = r->skip[PF_SKIP_PROTO].ptr;
844 else if (PF_MISMATCHAW(&r->src.addr,
845 (struct pf_addr *)&h->ip_src.s_addr, AF_INET, r->src.not))
846 r = r->skip[PF_SKIP_SRC_ADDR].ptr;
847 else if (PF_MISMATCHAW(&r->dst.addr,
848 (struct pf_addr *)&h->ip_dst.s_addr, AF_INET, r->dst.not))
849 r = r->skip[PF_SKIP_DST_ADDR].ptr;
859 /* Check for illegal packets */
860 if (hlen < (int)sizeof(struct ip))
863 if (hlen > h->ip_len)
866 /* Clear IP_DF if the rule uses the no-df option */
867 if (r->rule_flag & PFRULE_NODF)
870 /* We will need other tests here */
871 if (!fragoff && !mff)
874 /* We're dealing with a fragment now. Don't allow fragments
875 * with IP_DF to enter the cache. If the flag was cleared by
876 * no-df above, fine. Otherwise drop it.
878 if (h->ip_off & IP_DF) {
879 DPFPRINTF(("IP_DF\n"));
883 ip_len = h->ip_len - hlen;
884 ip_off = (h->ip_off & IP_OFFMASK) << 3;
886 /* All fragments are 8 byte aligned */
887 if (mff && (ip_len & 0x7)) {
888 DPFPRINTF(("mff and %d\n", ip_len));
892 /* Respect maximum length */
893 if (fragoff + ip_len > IP_MAXPACKET) {
894 DPFPRINTF(("max packet %d\n", fragoff + ip_len));
897 max = fragoff + ip_len;
899 if ((r->rule_flag & (PFRULE_FRAGCROP|PFRULE_FRAGDROP)) == 0) {
900 /* Fully buffer all of the fragments */
902 frag = pf_find_fragment(h, &pf_frag_tree);
904 /* Check if we saw the last fragment already */
905 if (frag != NULL && (frag->fr_flags & PFFRAG_SEENLAST) &&
909 /* Get an entry for the fragment queue */
910 frent = pool_get(&pf_frent_pl, PR_NOWAIT);
912 REASON_SET(reason, PFRES_MEMORY);
919 /* Might return a completely reassembled mbuf, or NULL */
920 DPFPRINTF(("reass frag %d @ %d-%d\n", h->ip_id, fragoff, max));
921 *m0 = m = pf_reassemble(m0, &frag, frent, mff);
926 if (frag != NULL && (frag->fr_flags & PFFRAG_DROP))
929 h = mtod(m, struct ip *);
931 /* non-buffering fragment cache (drops or masks overlaps) */
935 if (m->m_pkthdr.fw_flags & PF_MBUF_FRAGCACHE) {
936 /* Already passed the fragment cache in the
937 * input direction. If we continued, it would
938 * appear to be a dup and would be dropped.
944 frag = pf_find_fragment(h, &pf_cache_tree);
946 /* Check if we saw the last fragment already */
947 if (frag != NULL && (frag->fr_flags & PFFRAG_SEENLAST) &&
948 max > frag->fr_max) {
949 if (r->rule_flag & PFRULE_FRAGDROP)
950 frag->fr_flags |= PFFRAG_DROP;
954 *m0 = m = pf_fragcache(m0, h, &frag, mff,
955 (r->rule_flag & PFRULE_FRAGDROP) ? 1 : 0, &nomem);
963 m->m_pkthdr.fw_flags |= PF_MBUF_FRAGCACHE;
965 if (frag != NULL && (frag->fr_flags & PFFRAG_DROP))
971 /* At this point, only IP_DF is allowed in ip_off */
974 /* Enforce a minimum ttl, may cause endless packet loops */
975 if (r->min_ttl && h->ip_ttl < r->min_ttl)
976 h->ip_ttl = r->min_ttl;
978 if (r->rule_flag & PFRULE_RANDOMID) {
980 h->ip_id = ip_randomid();
982 h->ip_id = htons(ip_id++);
989 /* Enforce a minimum ttl, may cause endless packet loops */
990 if (r->min_ttl && h->ip_ttl < r->min_ttl)
991 h->ip_ttl = r->min_ttl;
996 REASON_SET(reason, PFRES_MEMORY);
997 if (r != NULL && r->log)
998 PFLOG_PACKET(kif, h, m, AF_INET, dir, *reason, r, NULL, NULL);
1002 REASON_SET(reason, PFRES_NORM);
1003 if (r != NULL && r->log)
1004 PFLOG_PACKET(kif, h, m, AF_INET, dir, *reason, r, NULL, NULL);
1008 DPFPRINTF(("dropping bad fragment\n"));
1010 /* Free associated fragments */
1012 pf_free_fragment(frag);
1014 REASON_SET(reason, PFRES_FRAG);
1015 if (r != NULL && r->log)
1016 PFLOG_PACKET(kif, h, m, AF_INET, dir, *reason, r, NULL, NULL);
1023 pf_normalize_ip6(struct mbuf **m0, int dir, struct pfi_kif *kif,
1026 struct mbuf *m = *m0;
1028 struct ip6_hdr *h = mtod(m, struct ip6_hdr *);
1032 struct ip6_opt_jumbo jumbo;
1033 struct ip6_frag frag;
1034 u_int32_t jumbolen = 0, plen;
1035 u_int16_t fragoff = 0;
1041 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1044 if (r->kif != NULL &&
1045 (r->kif != kif && r->kif != kif->pfik_parent) == !r->ifnot)
1046 r = r->skip[PF_SKIP_IFP].ptr;
1047 else if (r->direction && r->direction != dir)
1048 r = r->skip[PF_SKIP_DIR].ptr;
1049 else if (r->af && r->af != AF_INET6)
1050 r = r->skip[PF_SKIP_AF].ptr;
1051 #if 0 /* header chain! */
1052 else if (r->proto && r->proto != h->ip6_nxt)
1053 r = r->skip[PF_SKIP_PROTO].ptr;
1055 else if (PF_MISMATCHAW(&r->src.addr,
1056 (struct pf_addr *)&h->ip6_src, AF_INET6, r->src.not))
1057 r = r->skip[PF_SKIP_SRC_ADDR].ptr;
1058 else if (PF_MISMATCHAW(&r->dst.addr,
1059 (struct pf_addr *)&h->ip6_dst, AF_INET6, r->dst.not))
1060 r = r->skip[PF_SKIP_DST_ADDR].ptr;
1070 /* Check for illegal packets */
1071 if (sizeof(struct ip6_hdr) + IPV6_MAXPACKET < m->m_pkthdr.len)
1074 off = sizeof(struct ip6_hdr);
1079 case IPPROTO_FRAGMENT:
1083 case IPPROTO_ROUTING:
1084 case IPPROTO_DSTOPTS:
1085 if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL,
1088 if (proto == IPPROTO_AH)
1089 off += (ext.ip6e_len + 2) * 4;
1091 off += (ext.ip6e_len + 1) * 8;
1092 proto = ext.ip6e_nxt;
1094 case IPPROTO_HOPOPTS:
1095 if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL,
1098 optend = off + (ext.ip6e_len + 1) * 8;
1099 ooff = off + sizeof(ext);
1101 if (!pf_pull_hdr(m, ooff, &opt.ip6o_type,
1102 sizeof(opt.ip6o_type), NULL, NULL,
1105 if (opt.ip6o_type == IP6OPT_PAD1) {
1109 if (!pf_pull_hdr(m, ooff, &opt, sizeof(opt),
1110 NULL, NULL, AF_INET6))
1112 if (ooff + sizeof(opt) + opt.ip6o_len > optend)
1114 switch (opt.ip6o_type) {
1116 if (h->ip6_plen != 0)
1118 if (!pf_pull_hdr(m, ooff, &jumbo,
1119 sizeof(jumbo), NULL, NULL,
1122 memcpy(&jumbolen, jumbo.ip6oj_jumbo_len,
1124 jumbolen = ntohl(jumbolen);
1125 if (jumbolen <= IPV6_MAXPACKET)
1127 if (sizeof(struct ip6_hdr) + jumbolen !=
1134 ooff += sizeof(opt) + opt.ip6o_len;
1135 } while (ooff < optend);
1138 proto = ext.ip6e_nxt;
1144 } while (!terminal);
1146 /* jumbo payload option must be present, or plen > 0 */
1147 if (ntohs(h->ip6_plen) == 0)
1150 plen = ntohs(h->ip6_plen);
1153 if (sizeof(struct ip6_hdr) + plen > m->m_pkthdr.len)
1156 /* Enforce a minimum ttl, may cause endless packet loops */
1157 if (r->min_ttl && h->ip6_hlim < r->min_ttl)
1158 h->ip6_hlim = r->min_ttl;
1163 if (ntohs(h->ip6_plen) == 0 || jumbolen)
1165 plen = ntohs(h->ip6_plen);
1167 if (!pf_pull_hdr(m, off, &frag, sizeof(frag), NULL, NULL, AF_INET6))
1169 fragoff = ntohs(frag.ip6f_offlg & IP6F_OFF_MASK);
1170 if (fragoff + (plen - off - sizeof(frag)) > IPV6_MAXPACKET)
1173 /* do something about it */
1177 REASON_SET(reason, PFRES_SHORT);
1178 if (r != NULL && r->log)
1179 PFLOG_PACKET(kif, h, m, AF_INET6, dir, *reason, r, NULL, NULL);
1183 REASON_SET(reason, PFRES_NORM);
1184 if (r != NULL && r->log)
1185 PFLOG_PACKET(kif, h, m, AF_INET6, dir, *reason, r, NULL, NULL);
1189 REASON_SET(reason, PFRES_FRAG);
1190 if (r != NULL && r->log)
1191 PFLOG_PACKET(kif, h, m, AF_INET6, dir, *reason, r, NULL, NULL);
1197 pf_normalize_tcp(int dir, struct pfi_kif *kif, struct mbuf *m, int ipoff,
1198 int off, void *h, struct pf_pdesc *pd)
1200 struct pf_rule *r, *rm = NULL;
1201 struct tcphdr *th = pd->hdr.tcp;
1205 sa_family_t af = pd->af;
1207 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1210 if (r->kif != NULL &&
1211 (r->kif != kif && r->kif != kif->pfik_parent) == !r->ifnot)
1212 r = r->skip[PF_SKIP_IFP].ptr;
1213 else if (r->direction && r->direction != dir)
1214 r = r->skip[PF_SKIP_DIR].ptr;
1215 else if (r->af && r->af != af)
1216 r = r->skip[PF_SKIP_AF].ptr;
1217 else if (r->proto && r->proto != pd->proto)
1218 r = r->skip[PF_SKIP_PROTO].ptr;
1219 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af, r->src.not))
1220 r = r->skip[PF_SKIP_SRC_ADDR].ptr;
1221 else if (r->src.port_op && !pf_match_port(r->src.port_op,
1222 r->src.port[0], r->src.port[1], th->th_sport))
1223 r = r->skip[PF_SKIP_SRC_PORT].ptr;
1224 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af, r->dst.not))
1225 r = r->skip[PF_SKIP_DST_ADDR].ptr;
1226 else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
1227 r->dst.port[0], r->dst.port[1], th->th_dport))
1228 r = r->skip[PF_SKIP_DST_PORT].ptr;
1229 else if (r->os_fingerprint != PF_OSFP_ANY && !pf_osfp_match(
1230 pf_osfp_fingerprint(pd, m, off, th),
1232 r = TAILQ_NEXT(r, entries);
1244 if (rm->rule_flag & PFRULE_REASSEMBLE_TCP)
1245 pd->flags |= PFDESC_TCP_NORM;
1247 flags = th->th_flags;
1248 if (flags & TH_SYN) {
1249 /* Illegal packet */
1256 /* Illegal packet */
1257 if (!(flags & (TH_ACK|TH_RST)))
1261 if (!(flags & TH_ACK)) {
1262 /* These flags are only valid if ACK is set */
1263 if ((flags & TH_FIN) || (flags & TH_PUSH) || (flags & TH_URG))
1267 /* Check for illegal header length */
1268 if (th->th_off < (sizeof(struct tcphdr) >> 2))
1271 /* If flags changed, or reserved data set, then adjust */
1272 if (flags != th->th_flags || th->th_x2 != 0) {
1275 ov = *(u_int16_t *)(&th->th_ack + 1);
1276 th->th_flags = flags;
1278 nv = *(u_int16_t *)(&th->th_ack + 1);
1280 th->th_sum = pf_cksum_fixup(th->th_sum, ov, nv);
1284 /* Remove urgent pointer, if TH_URG is not set */
1285 if (!(flags & TH_URG) && th->th_urp) {
1286 th->th_sum = pf_cksum_fixup(th->th_sum, th->th_urp, 0);
1291 /* Process options */
1292 if (r->max_mss && pf_normalize_tcpopt(r, m, th, off))
1295 /* copy back packet headers if we sanitized */
1297 m_copyback(m, off, sizeof(*th), (caddr_t)th);
1302 REASON_SET(&reason, PFRES_NORM);
1303 if (rm != NULL && r->log)
1304 PFLOG_PACKET(kif, h, m, AF_INET, dir, reason, r, NULL, NULL);
1309 pf_normalize_tcp_init(struct mbuf *m, int off, struct pf_pdesc *pd,
1310 struct tcphdr *th, struct pf_state_peer *src, struct pf_state_peer *dst)
1315 KASSERT((src->scrub == NULL),
1316 ("pf_normalize_tcp_init: src->scrub != NULL"));
1318 src->scrub = pool_get(&pf_state_scrub_pl, PR_NOWAIT);
1319 if (src->scrub == NULL)
1321 bzero(src->scrub, sizeof(*src->scrub));
1326 struct ip *h = mtod(m, struct ip *);
1327 src->scrub->pfss_ttl = h->ip_ttl;
1333 struct ip6_hdr *h = mtod(m, struct ip6_hdr *);
1334 src->scrub->pfss_ttl = h->ip6_hlim;
1342 * All normalizations below are only begun if we see the start of
1343 * the connections. They must all set an enabled bit in pfss_flags
1345 if ((th->th_flags & TH_SYN) == 0)
1349 if (th->th_off > (sizeof(struct tcphdr) >> 2) && src->scrub &&
1350 pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) {
1351 /* Diddle with TCP options */
1353 opt = hdr + sizeof(struct tcphdr);
1354 hlen = (th->th_off << 2) - sizeof(struct tcphdr);
1355 while (hlen >= TCPOLEN_TIMESTAMP) {
1357 case TCPOPT_EOL: /* FALLTHROUGH */
1362 case TCPOPT_TIMESTAMP:
1363 if (opt[1] >= TCPOLEN_TIMESTAMP) {
1364 src->scrub->pfss_flags |=
1366 src->scrub->pfss_ts_mod = karc4random();
1370 hlen -= MAX(opt[1], 2);
1371 opt += MAX(opt[1], 2);
1381 pf_normalize_tcp_cleanup(struct pf_state *state)
1383 if (state->src.scrub)
1384 pool_put(&pf_state_scrub_pl, state->src.scrub);
1385 if (state->dst.scrub)
1386 pool_put(&pf_state_scrub_pl, state->dst.scrub);
1388 /* Someday... flush the TCP segment reassembly descriptors. */
1392 pf_normalize_tcp_stateful(struct mbuf *m, int off, struct pf_pdesc *pd,
1393 u_short *reason, struct tcphdr *th, struct pf_state_peer *src,
1394 struct pf_state_peer *dst, int *writeback)
1400 KASSERT((src->scrub || dst->scrub),
1401 ("pf_normalize_tcp_statefull: src->scrub && dst->scrub!"));
1404 * Enforce the minimum TTL seen for this connection. Negate a common
1405 * technique to evade an intrusion detection system and confuse
1406 * firewall state code.
1412 struct ip *h = mtod(m, struct ip *);
1413 if (h->ip_ttl > src->scrub->pfss_ttl)
1414 src->scrub->pfss_ttl = h->ip_ttl;
1415 h->ip_ttl = src->scrub->pfss_ttl;
1423 struct ip6_hdr *h = mtod(m, struct ip6_hdr *);
1424 if (h->ip6_hlim > src->scrub->pfss_ttl)
1425 src->scrub->pfss_ttl = h->ip6_hlim;
1426 h->ip6_hlim = src->scrub->pfss_ttl;
1433 if (th->th_off > (sizeof(struct tcphdr) >> 2) &&
1434 ((src->scrub && (src->scrub->pfss_flags & PFSS_TIMESTAMP)) ||
1435 (dst->scrub && (dst->scrub->pfss_flags & PFSS_TIMESTAMP))) &&
1436 pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) {
1437 /* Diddle with TCP options */
1439 opt = hdr + sizeof(struct tcphdr);
1440 hlen = (th->th_off << 2) - sizeof(struct tcphdr);
1441 while (hlen >= TCPOLEN_TIMESTAMP) {
1443 case TCPOPT_EOL: /* FALLTHROUGH */
1448 case TCPOPT_TIMESTAMP:
1449 /* Modulate the timestamps. Can be used for
1450 * NAT detection, OS uptime determination or
1453 if (opt[1] >= TCPOLEN_TIMESTAMP) {
1456 (src->scrub->pfss_flags &
1458 memcpy(&ts_value, &opt[2],
1460 ts_value = htonl(ntohl(ts_value)
1461 + src->scrub->pfss_ts_mod);
1462 pf_change_a(&opt[2],
1463 &th->th_sum, ts_value, 0);
1467 /* Modulate TS reply iff valid (!0) */
1468 memcpy(&ts_value, &opt[6],
1470 if (ts_value && dst->scrub &&
1471 (dst->scrub->pfss_flags &
1473 ts_value = htonl(ntohl(ts_value)
1474 - dst->scrub->pfss_ts_mod);
1475 pf_change_a(&opt[6],
1476 &th->th_sum, ts_value, 0);
1482 hlen -= MAX(opt[1], 2);
1483 opt += MAX(opt[1], 2);
1488 /* Copyback the options, caller copys back header */
1490 m_copyback(m, off + sizeof(struct tcphdr),
1491 (th->th_off << 2) - sizeof(struct tcphdr), hdr +
1492 sizeof(struct tcphdr));
1497 /* I have a dream.... TCP segment reassembly.... */
1502 pf_normalize_tcpopt(struct pf_rule *r, struct mbuf *m, struct tcphdr *th,
1507 int opt, cnt, optlen = 0;
1511 thoff = th->th_off << 2;
1512 cnt = thoff - sizeof(struct tcphdr);
1513 optp = mtod(m, caddr_t) + off + sizeof(struct tcphdr);
1515 for (; cnt > 0; cnt -= optlen, optp += optlen) {
1517 if (opt == TCPOPT_EOL)
1519 if (opt == TCPOPT_NOP)
1525 if (optlen < 2 || optlen > cnt)
1530 mss = (u_int16_t *)(optp + 2);
1531 if ((ntohs(*mss)) > r->max_mss) {
1532 th->th_sum = pf_cksum_fixup(th->th_sum,
1533 *mss, htons(r->max_mss));
1534 *mss = htons(r->max_mss);