xref: /dragonfly/sys/net/pf/pf_norm.c (revision acea3a2a)
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 1.3 2005/02/17 13:59:59 joerg 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 			    { printf("%s: ", __func__); printf 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_cat(m, m2);
499 	}
500 
501 	ip->ip_src = (*frag)->fr_src;
502 	ip->ip_dst = (*frag)->fr_dst;
503 
504 	/* Remove from fragment queue */
505 	pf_remove_fragment(*frag);
506 	*frag = NULL;
507 
508 	hlen = ip->ip_hl << 2;
509 	ip->ip_len = off + hlen;
510 	m->m_len += hlen;
511 	m->m_data -= hlen;
512 
513 	/* some debugging cruft by sklower, below, will go away soon */
514 	/* XXX this should be done elsewhere */
515 	if (m->m_flags & M_PKTHDR) {
516 		int plen = 0;
517 		for (m2 = m; m2; m2 = m2->m_next)
518 			plen += m2->m_len;
519 		m->m_pkthdr.len = plen;
520 	}
521 
522 	DPFPRINTF(("complete: %p(%d)\n", m, ip->ip_len));
523 	return (m);
524 
525  drop_fragment:
526 	/* Oops - fail safe - drop packet */
527 	pool_put(&pf_frent_pl, frent);
528 	pf_nfrents--;
529 	m_freem(m);
530 	return (NULL);
531 }
532 
533 struct mbuf *
534 pf_fragcache(struct mbuf **m0, struct ip *h, struct pf_fragment **frag, int mff,
535     int drop, int *nomem)
536 {
537 	struct mbuf		*m = *m0;
538 	struct pf_frcache	*frp, *fra, *cur = NULL;
539 	int			 ip_len = h->ip_len - (h->ip_hl << 2);
540 	u_int16_t		 off = h->ip_off << 3;
541 	u_int16_t		 max = ip_len + off;
542 	int			 hosed = 0;
543 
544 	KASSERT((*frag == NULL || !BUFFER_FRAGMENTS(*frag)),
545 	    ("!(*frag == NULL || !BUFFER_FRAGMENTS(*frag)): %s", __func__));
546 
547 	/* Create a new range queue for this packet */
548 	if (*frag == NULL) {
549 		*frag = pool_get(&pf_cache_pl, PR_NOWAIT);
550 		if (*frag == NULL) {
551 			pf_flush_fragments();
552 			*frag = pool_get(&pf_cache_pl, PR_NOWAIT);
553 			if (*frag == NULL)
554 				goto no_mem;
555 		}
556 
557 		/* Get an entry for the queue */
558 		cur = pool_get(&pf_cent_pl, PR_NOWAIT);
559 		if (cur == NULL) {
560 			pool_put(&pf_cache_pl, *frag);
561 			*frag = NULL;
562 			goto no_mem;
563 		}
564 		pf_ncache++;
565 
566 		(*frag)->fr_flags = PFFRAG_NOBUFFER;
567 		(*frag)->fr_max = 0;
568 		(*frag)->fr_src = h->ip_src;
569 		(*frag)->fr_dst = h->ip_dst;
570 		(*frag)->fr_p = h->ip_p;
571 		(*frag)->fr_id = h->ip_id;
572 		(*frag)->fr_timeout = time_second;
573 
574 		cur->fr_off = off;
575 		cur->fr_end = max;
576 		LIST_INIT(&(*frag)->fr_cache);
577 		LIST_INSERT_HEAD(&(*frag)->fr_cache, cur, fr_next);
578 
579 		RB_INSERT(pf_frag_tree, &pf_cache_tree, *frag);
580 		TAILQ_INSERT_HEAD(&pf_cachequeue, *frag, frag_next);
581 
582 		DPFPRINTF(("fragcache[%d]: new %d-%d\n", h->ip_id, off, max));
583 
584 		goto pass;
585 	}
586 
587 	/*
588 	 * Find a fragment after the current one:
589 	 *  - off contains the real shifted offset.
590 	 */
591 	frp = NULL;
592 	LIST_FOREACH(fra, &(*frag)->fr_cache, fr_next) {
593 		if (fra->fr_off > off)
594 			break;
595 		frp = fra;
596 	}
597 
598 	KASSERT((frp != NULL || fra != NULL),
599 	    ("!(frp != NULL || fra != NULL): %s", __func__));
600 
601 	if (frp != NULL) {
602 		int	precut;
603 
604 		precut = frp->fr_end - off;
605 		if (precut >= ip_len) {
606 			/* Fragment is entirely a duplicate */
607 			DPFPRINTF(("fragcache[%d]: dead (%d-%d) %d-%d\n",
608 			    h->ip_id, frp->fr_off, frp->fr_end, off, max));
609 			goto drop_fragment;
610 		}
611 		if (precut == 0) {
612 			/* They are adjacent.  Fixup cache entry */
613 			DPFPRINTF(("fragcache[%d]: adjacent (%d-%d) %d-%d\n",
614 			    h->ip_id, frp->fr_off, frp->fr_end, off, max));
615 			frp->fr_end = max;
616 		} else if (precut > 0) {
617 			/* The first part of this payload overlaps with a
618 			 * fragment that has already been passed.
619 			 * Need to trim off the first part of the payload.
620 			 * But to do so easily, we need to create another
621 			 * mbuf to throw the original header into.
622 			 */
623 
624 			DPFPRINTF(("fragcache[%d]: chop %d (%d-%d) %d-%d\n",
625 			    h->ip_id, precut, frp->fr_off, frp->fr_end, off,
626 			    max));
627 
628 			off += precut;
629 			max -= precut;
630 			/* Update the previous frag to encompass this one */
631 			frp->fr_end = max;
632 
633 			if (!drop) {
634 				/* XXX Optimization opportunity
635 				 * This is a very heavy way to trim the payload.
636 				 * we could do it much faster by diddling mbuf
637 				 * internals but that would be even less legible
638 				 * than this mbuf magic.  For my next trick,
639 				 * I'll pull a rabbit out of my laptop.
640 				 */
641 				*m0 = m_dup(m, MB_DONTWAIT);
642 				/* From KAME Project : We have missed this! */
643 				m_adj(*m0, (h->ip_hl << 2) -
644 				    (*m0)->m_pkthdr.len);
645 				if (*m0 == NULL)
646 					goto no_mem;
647 				KASSERT(((*m0)->m_next == NULL),
648 				    ("(*m0)->m_next != NULL: %s",
649 				    __func__));
650 				m_adj(m, precut + (h->ip_hl << 2));
651 				m_cat(*m0, m);
652 				m = *m0;
653 				if (m->m_flags & M_PKTHDR) {
654 					int plen = 0;
655 					struct mbuf *t;
656 					for (t = m; t; t = t->m_next)
657 						plen += t->m_len;
658 					m->m_pkthdr.len = plen;
659 				}
660 
661 
662 				h = mtod(m, struct ip *);
663 
664 				KASSERT(((int)m->m_len ==
665 				    h->ip_len - precut),
666 				    ("m->m_len != h->ip_len - precut: %s",
667 				    __func__));
668 				h->ip_off = h->ip_off +
669 				    (precut >> 3);
670 				h->ip_len = h->ip_len - precut;
671 			} else {
672 				hosed++;
673 			}
674 		} else {
675 			/* There is a gap between fragments */
676 
677 			DPFPRINTF(("fragcache[%d]: gap %d (%d-%d) %d-%d\n",
678 			    h->ip_id, -precut, frp->fr_off, frp->fr_end, off,
679 			    max));
680 
681 			cur = pool_get(&pf_cent_pl, PR_NOWAIT);
682 			if (cur == NULL)
683 				goto no_mem;
684 			pf_ncache++;
685 
686 			cur->fr_off = off;
687 			cur->fr_end = max;
688 			LIST_INSERT_AFTER(frp, cur, fr_next);
689 		}
690 	}
691 
692 	if (fra != NULL) {
693 		int	aftercut;
694 		int	merge = 0;
695 
696 		aftercut = max - fra->fr_off;
697 		if (aftercut == 0) {
698 			/* Adjacent fragments */
699 			DPFPRINTF(("fragcache[%d]: adjacent %d-%d (%d-%d)\n",
700 			    h->ip_id, off, max, fra->fr_off, fra->fr_end));
701 			fra->fr_off = off;
702 			merge = 1;
703 		} else if (aftercut > 0) {
704 			/* Need to chop off the tail of this fragment */
705 			DPFPRINTF(("fragcache[%d]: chop %d %d-%d (%d-%d)\n",
706 			    h->ip_id, aftercut, off, max, fra->fr_off,
707 			    fra->fr_end));
708 			fra->fr_off = off;
709 			max -= aftercut;
710 
711 			merge = 1;
712 
713 			if (!drop) {
714 				m_adj(m, -aftercut);
715 				if (m->m_flags & M_PKTHDR) {
716 					int plen = 0;
717 					struct mbuf *t;
718 					for (t = m; t; t = t->m_next)
719 						plen += t->m_len;
720 					m->m_pkthdr.len = plen;
721 				}
722 				h = mtod(m, struct ip *);
723 				KASSERT(((int)m->m_len == h->ip_len - aftercut),
724 				    ("m->m_len != h->ip_len - aftercut: %s",
725 				    __func__));
726 				h->ip_len = h->ip_len - aftercut;
727 			} else {
728 				hosed++;
729 			}
730 		} else {
731 			/* There is a gap between fragments */
732 			DPFPRINTF(("fragcache[%d]: gap %d %d-%d (%d-%d)\n",
733 			    h->ip_id, -aftercut, off, max, fra->fr_off,
734 			    fra->fr_end));
735 
736 			cur = pool_get(&pf_cent_pl, PR_NOWAIT);
737 			if (cur == NULL)
738 				goto no_mem;
739 			pf_ncache++;
740 
741 			cur->fr_off = off;
742 			cur->fr_end = max;
743 			LIST_INSERT_BEFORE(fra, cur, fr_next);
744 		}
745 
746 
747 		/* Need to glue together two separate fragment descriptors */
748 		if (merge) {
749 			if (cur && fra->fr_off <= cur->fr_end) {
750 				/* Need to merge in a previous 'cur' */
751 				DPFPRINTF(("fragcache[%d]: adjacent(merge "
752 				    "%d-%d) %d-%d (%d-%d)\n",
753 				    h->ip_id, cur->fr_off, cur->fr_end, off,
754 				    max, fra->fr_off, fra->fr_end));
755 				fra->fr_off = cur->fr_off;
756 				LIST_REMOVE(cur, fr_next);
757 				pool_put(&pf_cent_pl, cur);
758 				pf_ncache--;
759 				cur = NULL;
760 
761 			} else if (frp && fra->fr_off <= frp->fr_end) {
762 				/* Need to merge in a modified 'frp' */
763 				KASSERT((cur == NULL), ("cur != NULL: %s",
764 				    __func__));
765 				DPFPRINTF(("fragcache[%d]: adjacent(merge "
766 				    "%d-%d) %d-%d (%d-%d)\n",
767 				    h->ip_id, frp->fr_off, frp->fr_end, off,
768 				    max, fra->fr_off, fra->fr_end));
769 				fra->fr_off = frp->fr_off;
770 				LIST_REMOVE(frp, fr_next);
771 				pool_put(&pf_cent_pl, frp);
772 				pf_ncache--;
773 				frp = NULL;
774 
775 			}
776 		}
777 	}
778 
779 	if (hosed) {
780 		/*
781 		 * We must keep tracking the overall fragment even when
782 		 * we're going to drop it anyway so that we know when to
783 		 * free the overall descriptor.  Thus we drop the frag late.
784 		 */
785 		goto drop_fragment;
786 	}
787 
788 
789  pass:
790 	/* Update maximum data size */
791 	if ((*frag)->fr_max < max)
792 		(*frag)->fr_max = max;
793 
794 	/* This is the last segment */
795 	if (!mff)
796 		(*frag)->fr_flags |= PFFRAG_SEENLAST;
797 
798 	/* Check if we are completely reassembled */
799 	if (((*frag)->fr_flags & PFFRAG_SEENLAST) &&
800 	    LIST_FIRST(&(*frag)->fr_cache)->fr_off == 0 &&
801 	    LIST_FIRST(&(*frag)->fr_cache)->fr_end == (*frag)->fr_max) {
802 		/* Remove from fragment queue */
803 		DPFPRINTF(("fragcache[%d]: done 0-%d\n", h->ip_id,
804 		    (*frag)->fr_max));
805 		pf_free_fragment(*frag);
806 		*frag = NULL;
807 	}
808 
809 	return (m);
810 
811  no_mem:
812 	*nomem = 1;
813 
814 	/* Still need to pay attention to !IP_MF */
815 	if (!mff && *frag != NULL)
816 		(*frag)->fr_flags |= PFFRAG_SEENLAST;
817 
818 	m_freem(m);
819 	return (NULL);
820 
821  drop_fragment:
822 
823 	/* Still need to pay attention to !IP_MF */
824 	if (!mff && *frag != NULL)
825 		(*frag)->fr_flags |= PFFRAG_SEENLAST;
826 
827 	if (drop) {
828 		/* This fragment has been deemed bad.  Don't reass */
829 		if (((*frag)->fr_flags & PFFRAG_DROP) == 0)
830 			DPFPRINTF(("fragcache[%d]: dropping overall fragment\n",
831 			    h->ip_id));
832 		(*frag)->fr_flags |= PFFRAG_DROP;
833 	}
834 
835 	m_freem(m);
836 	return (NULL);
837 }
838 
839 int
840 pf_normalize_ip(struct mbuf **m0, int dir, struct pfi_kif *kif, u_short *reason)
841 {
842 	struct mbuf		*m = *m0;
843 	struct pf_rule		*r;
844 	struct pf_frent		*frent;
845 	struct pf_fragment	*frag = NULL;
846 	struct ip		*h = mtod(m, struct ip *);
847 	int			 mff = (h->ip_off & IP_MF);
848 	int			 hlen = h->ip_hl << 2;
849 	u_int16_t		 fragoff = (h->ip_off & IP_OFFMASK) << 3;
850 	u_int16_t		 max;
851 	int			 ip_len;
852 	int			 ip_off;
853 
854 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
855 	while (r != NULL) {
856 		r->evaluations++;
857 		if (r->kif != NULL &&
858 		    (r->kif != kif && r->kif != kif->pfik_parent) == !r->ifnot)
859 			r = r->skip[PF_SKIP_IFP].ptr;
860 		else if (r->direction && r->direction != dir)
861 			r = r->skip[PF_SKIP_DIR].ptr;
862 		else if (r->af && r->af != AF_INET)
863 			r = r->skip[PF_SKIP_AF].ptr;
864 		else if (r->proto && r->proto != h->ip_p)
865 			r = r->skip[PF_SKIP_PROTO].ptr;
866 		else if (PF_MISMATCHAW(&r->src.addr,
867 		    (struct pf_addr *)&h->ip_src.s_addr, AF_INET, r->src.not))
868 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
869 		else if (PF_MISMATCHAW(&r->dst.addr,
870 		    (struct pf_addr *)&h->ip_dst.s_addr, AF_INET, r->dst.not))
871 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
872 		else
873 			break;
874 	}
875 
876 	if (r == NULL)
877 		return (PF_PASS);
878 	else
879 		r->packets++;
880 
881 	/* Check for illegal packets */
882 	if (hlen < (int)sizeof(struct ip))
883 		goto drop;
884 
885 	if (hlen > h->ip_len)
886 		goto drop;
887 
888 	/* Clear IP_DF if the rule uses the no-df option */
889 	if (r->rule_flag & PFRULE_NODF)
890 		h->ip_off &= ~IP_DF;
891 
892 	/* We will need other tests here */
893 	if (!fragoff && !mff)
894 		goto no_fragment;
895 
896 	/* We're dealing with a fragment now. Don't allow fragments
897 	 * with IP_DF to enter the cache. If the flag was cleared by
898 	 * no-df above, fine. Otherwise drop it.
899 	 */
900 	if (h->ip_off & IP_DF) {
901 		DPFPRINTF(("IP_DF\n"));
902 		goto bad;
903 	}
904 
905 	ip_len = h->ip_len - hlen;
906 	ip_off = (h->ip_off & IP_OFFMASK) << 3;
907 
908 	/* All fragments are 8 byte aligned */
909 	if (mff && (ip_len & 0x7)) {
910 		DPFPRINTF(("mff and %d\n", ip_len));
911 		goto bad;
912 	}
913 
914 	/* Respect maximum length */
915 	if (fragoff + ip_len > IP_MAXPACKET) {
916 		DPFPRINTF(("max packet %d\n", fragoff + ip_len));
917 		goto bad;
918 	}
919 	max = fragoff + ip_len;
920 
921 	if ((r->rule_flag & (PFRULE_FRAGCROP|PFRULE_FRAGDROP)) == 0) {
922 		/* Fully buffer all of the fragments */
923 
924 		frag = pf_find_fragment(h, &pf_frag_tree);
925 
926 		/* Check if we saw the last fragment already */
927 		if (frag != NULL && (frag->fr_flags & PFFRAG_SEENLAST) &&
928 		    max > frag->fr_max)
929 			goto bad;
930 
931 		/* Get an entry for the fragment queue */
932 		frent = pool_get(&pf_frent_pl, PR_NOWAIT);
933 		if (frent == NULL) {
934 			REASON_SET(reason, PFRES_MEMORY);
935 			return (PF_DROP);
936 		}
937 		pf_nfrents++;
938 		frent->fr_ip = h;
939 		frent->fr_m = m;
940 
941 		/* Might return a completely reassembled mbuf, or NULL */
942 		DPFPRINTF(("reass frag %d @ %d-%d\n", h->ip_id, fragoff, max));
943 		*m0 = m = pf_reassemble(m0, &frag, frent, mff);
944 
945 		if (m == NULL)
946 			return (PF_DROP);
947 
948 		if (frag != NULL && (frag->fr_flags & PFFRAG_DROP))
949 			goto drop;
950 
951 		h = mtod(m, struct ip *);
952 	} else {
953 		/* non-buffering fragment cache (drops or masks overlaps) */
954 		int	nomem = 0;
955 
956 		if (dir == PF_OUT) {
957 			if (m->m_pkthdr.fw_flags & PF_MBUF_FRAGCACHE) {
958 				/* Already passed the fragment cache in the
959 				 * input direction.  If we continued, it would
960 				 * appear to be a dup and would be dropped.
961 				 */
962 				goto fragment_pass;
963 			}
964 		}
965 
966 		frag = pf_find_fragment(h, &pf_cache_tree);
967 
968 		/* Check if we saw the last fragment already */
969 		if (frag != NULL && (frag->fr_flags & PFFRAG_SEENLAST) &&
970 		    max > frag->fr_max) {
971 			if (r->rule_flag & PFRULE_FRAGDROP)
972 				frag->fr_flags |= PFFRAG_DROP;
973 			goto bad;
974 		}
975 
976 		*m0 = m = pf_fragcache(m0, h, &frag, mff,
977 		    (r->rule_flag & PFRULE_FRAGDROP) ? 1 : 0, &nomem);
978 		if (m == NULL) {
979 			if (nomem)
980 				goto no_mem;
981 			goto drop;
982 		}
983 
984 		if (dir == PF_IN)
985 			m->m_pkthdr.fw_flags |= PF_MBUF_FRAGCACHE;
986 
987 		if (frag != NULL && (frag->fr_flags & PFFRAG_DROP))
988 			goto drop;
989 		goto fragment_pass;
990 	}
991 
992  no_fragment:
993 	/* At this point, only IP_DF is allowed in ip_off */
994 	h->ip_off &= IP_DF;
995 
996 	/* Enforce a minimum ttl, may cause endless packet loops */
997 	if (r->min_ttl && h->ip_ttl < r->min_ttl)
998 		h->ip_ttl = r->min_ttl;
999 
1000 	if (r->rule_flag & PFRULE_RANDOMID) {
1001 #ifdef RANDOM_IP_ID
1002 		h->ip_id = ip_randomid();
1003 #else
1004 		h->ip_id = htons(ip_id++);
1005 #endif
1006 	}
1007 
1008 	return (PF_PASS);
1009 
1010  fragment_pass:
1011 	/* Enforce a minimum ttl, may cause endless packet loops */
1012 	if (r->min_ttl && h->ip_ttl < r->min_ttl)
1013 		h->ip_ttl = r->min_ttl;
1014 
1015 	return (PF_PASS);
1016 
1017  no_mem:
1018 	REASON_SET(reason, PFRES_MEMORY);
1019 	if (r != NULL && r->log)
1020 		PFLOG_PACKET(kif, h, m, AF_INET, dir, *reason, r, NULL, NULL);
1021 	return (PF_DROP);
1022 
1023  drop:
1024 	REASON_SET(reason, PFRES_NORM);
1025 	if (r != NULL && r->log)
1026 		PFLOG_PACKET(kif, h, m, AF_INET, dir, *reason, r, NULL, NULL);
1027 	return (PF_DROP);
1028 
1029  bad:
1030 	DPFPRINTF(("dropping bad fragment\n"));
1031 
1032 	/* Free associated fragments */
1033 	if (frag != NULL)
1034 		pf_free_fragment(frag);
1035 
1036 	REASON_SET(reason, PFRES_FRAG);
1037 	if (r != NULL && r->log)
1038 		PFLOG_PACKET(kif, h, m, AF_INET, dir, *reason, r, NULL, NULL);
1039 
1040 	return (PF_DROP);
1041 }
1042 
1043 #ifdef INET6
1044 int
1045 pf_normalize_ip6(struct mbuf **m0, int dir, struct pfi_kif *kif,
1046     u_short *reason)
1047 {
1048 	struct mbuf		*m = *m0;
1049 	struct pf_rule		*r;
1050 	struct ip6_hdr		*h = mtod(m, struct ip6_hdr *);
1051 	int			 off;
1052 	struct ip6_ext		 ext;
1053 	struct ip6_opt		 opt;
1054 	struct ip6_opt_jumbo	 jumbo;
1055 	struct ip6_frag		 frag;
1056 	u_int32_t		 jumbolen = 0, plen;
1057 	u_int16_t		 fragoff = 0;
1058 	int			 optend;
1059 	int			 ooff;
1060 	u_int8_t		 proto;
1061 	int			 terminal;
1062 
1063 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1064 	while (r != NULL) {
1065 		r->evaluations++;
1066 		if (r->kif != NULL &&
1067 		    (r->kif != kif && r->kif != kif->pfik_parent) == !r->ifnot)
1068 			r = r->skip[PF_SKIP_IFP].ptr;
1069 		else if (r->direction && r->direction != dir)
1070 			r = r->skip[PF_SKIP_DIR].ptr;
1071 		else if (r->af && r->af != AF_INET6)
1072 			r = r->skip[PF_SKIP_AF].ptr;
1073 #if 0 /* header chain! */
1074 		else if (r->proto && r->proto != h->ip6_nxt)
1075 			r = r->skip[PF_SKIP_PROTO].ptr;
1076 #endif
1077 		else if (PF_MISMATCHAW(&r->src.addr,
1078 		    (struct pf_addr *)&h->ip6_src, AF_INET6, r->src.not))
1079 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
1080 		else if (PF_MISMATCHAW(&r->dst.addr,
1081 		    (struct pf_addr *)&h->ip6_dst, AF_INET6, r->dst.not))
1082 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
1083 		else
1084 			break;
1085 	}
1086 
1087 	if (r == NULL)
1088 		return (PF_PASS);
1089 	else
1090 		r->packets++;
1091 
1092 	/* Check for illegal packets */
1093 	if (sizeof(struct ip6_hdr) + IPV6_MAXPACKET < m->m_pkthdr.len)
1094 		goto drop;
1095 
1096 	off = sizeof(struct ip6_hdr);
1097 	proto = h->ip6_nxt;
1098 	terminal = 0;
1099 	do {
1100 		switch (proto) {
1101 		case IPPROTO_FRAGMENT:
1102 			goto fragment;
1103 			break;
1104 		case IPPROTO_AH:
1105 		case IPPROTO_ROUTING:
1106 		case IPPROTO_DSTOPTS:
1107 			if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL,
1108 			    NULL, AF_INET6))
1109 				goto shortpkt;
1110 			if (proto == IPPROTO_AH)
1111 				off += (ext.ip6e_len + 2) * 4;
1112 			else
1113 				off += (ext.ip6e_len + 1) * 8;
1114 			proto = ext.ip6e_nxt;
1115 			break;
1116 		case IPPROTO_HOPOPTS:
1117 			if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL,
1118 			    NULL, AF_INET6))
1119 				goto shortpkt;
1120 			optend = off + (ext.ip6e_len + 1) * 8;
1121 			ooff = off + sizeof(ext);
1122 			do {
1123 				if (!pf_pull_hdr(m, ooff, &opt.ip6o_type,
1124 				    sizeof(opt.ip6o_type), NULL, NULL,
1125 				    AF_INET6))
1126 					goto shortpkt;
1127 				if (opt.ip6o_type == IP6OPT_PAD1) {
1128 					ooff++;
1129 					continue;
1130 				}
1131 				if (!pf_pull_hdr(m, ooff, &opt, sizeof(opt),
1132 				    NULL, NULL, AF_INET6))
1133 					goto shortpkt;
1134 				if (ooff + sizeof(opt) + opt.ip6o_len > optend)
1135 					goto drop;
1136 				switch (opt.ip6o_type) {
1137 				case IP6OPT_JUMBO:
1138 					if (h->ip6_plen != 0)
1139 						goto drop;
1140 					if (!pf_pull_hdr(m, ooff, &jumbo,
1141 					    sizeof(jumbo), NULL, NULL,
1142 					    AF_INET6))
1143 						goto shortpkt;
1144 					memcpy(&jumbolen, jumbo.ip6oj_jumbo_len,
1145 					    sizeof(jumbolen));
1146 					jumbolen = ntohl(jumbolen);
1147 					if (jumbolen <= IPV6_MAXPACKET)
1148 						goto drop;
1149 					if (sizeof(struct ip6_hdr) + jumbolen !=
1150 					    m->m_pkthdr.len)
1151 						goto drop;
1152 					break;
1153 				default:
1154 					break;
1155 				}
1156 				ooff += sizeof(opt) + opt.ip6o_len;
1157 			} while (ooff < optend);
1158 
1159 			off = optend;
1160 			proto = ext.ip6e_nxt;
1161 			break;
1162 		default:
1163 			terminal = 1;
1164 			break;
1165 		}
1166 	} while (!terminal);
1167 
1168 	/* jumbo payload option must be present, or plen > 0 */
1169 	if (ntohs(h->ip6_plen) == 0)
1170 		plen = jumbolen;
1171 	else
1172 		plen = ntohs(h->ip6_plen);
1173 	if (plen == 0)
1174 		goto drop;
1175 	if (sizeof(struct ip6_hdr) + plen > m->m_pkthdr.len)
1176 		goto shortpkt;
1177 
1178 	/* Enforce a minimum ttl, may cause endless packet loops */
1179 	if (r->min_ttl && h->ip6_hlim < r->min_ttl)
1180 		h->ip6_hlim = r->min_ttl;
1181 
1182 	return (PF_PASS);
1183 
1184  fragment:
1185 	if (ntohs(h->ip6_plen) == 0 || jumbolen)
1186 		goto drop;
1187 	plen = ntohs(h->ip6_plen);
1188 
1189 	if (!pf_pull_hdr(m, off, &frag, sizeof(frag), NULL, NULL, AF_INET6))
1190 		goto shortpkt;
1191 	fragoff = ntohs(frag.ip6f_offlg & IP6F_OFF_MASK);
1192 	if (fragoff + (plen - off - sizeof(frag)) > IPV6_MAXPACKET)
1193 		goto badfrag;
1194 
1195 	/* do something about it */
1196 	return (PF_PASS);
1197 
1198  shortpkt:
1199 	REASON_SET(reason, PFRES_SHORT);
1200 	if (r != NULL && r->log)
1201 		PFLOG_PACKET(kif, h, m, AF_INET6, dir, *reason, r, NULL, NULL);
1202 	return (PF_DROP);
1203 
1204  drop:
1205 	REASON_SET(reason, PFRES_NORM);
1206 	if (r != NULL && r->log)
1207 		PFLOG_PACKET(kif, h, m, AF_INET6, dir, *reason, r, NULL, NULL);
1208 	return (PF_DROP);
1209 
1210  badfrag:
1211 	REASON_SET(reason, PFRES_FRAG);
1212 	if (r != NULL && r->log)
1213 		PFLOG_PACKET(kif, h, m, AF_INET6, dir, *reason, r, NULL, NULL);
1214 	return (PF_DROP);
1215 }
1216 #endif
1217 
1218 int
1219 pf_normalize_tcp(int dir, struct pfi_kif *kif, struct mbuf *m, int ipoff,
1220     int off, void *h, struct pf_pdesc *pd)
1221 {
1222 	struct pf_rule	*r, *rm = NULL;
1223 	struct tcphdr	*th = pd->hdr.tcp;
1224 	int		 rewrite = 0;
1225 	u_short		 reason;
1226 	u_int8_t	 flags;
1227 	sa_family_t	 af = pd->af;
1228 
1229 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1230 	while (r != NULL) {
1231 		r->evaluations++;
1232 		if (r->kif != NULL &&
1233 		    (r->kif != kif && r->kif != kif->pfik_parent) == !r->ifnot)
1234 			r = r->skip[PF_SKIP_IFP].ptr;
1235 		else if (r->direction && r->direction != dir)
1236 			r = r->skip[PF_SKIP_DIR].ptr;
1237 		else if (r->af && r->af != af)
1238 			r = r->skip[PF_SKIP_AF].ptr;
1239 		else if (r->proto && r->proto != pd->proto)
1240 			r = r->skip[PF_SKIP_PROTO].ptr;
1241 		else if (PF_MISMATCHAW(&r->src.addr, pd->src, af, r->src.not))
1242 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
1243 		else if (r->src.port_op && !pf_match_port(r->src.port_op,
1244 			    r->src.port[0], r->src.port[1], th->th_sport))
1245 			r = r->skip[PF_SKIP_SRC_PORT].ptr;
1246 		else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af, r->dst.not))
1247 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
1248 		else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
1249 			    r->dst.port[0], r->dst.port[1], th->th_dport))
1250 			r = r->skip[PF_SKIP_DST_PORT].ptr;
1251 		else if (r->os_fingerprint != PF_OSFP_ANY && !pf_osfp_match(
1252 			    pf_osfp_fingerprint(pd, m, off, th),
1253 			    r->os_fingerprint))
1254 			r = TAILQ_NEXT(r, entries);
1255 		else {
1256 			rm = r;
1257 			break;
1258 		}
1259 	}
1260 
1261 	if (rm == NULL)
1262 		return (PF_PASS);
1263 	else
1264 		r->packets++;
1265 
1266 	if (rm->rule_flag & PFRULE_REASSEMBLE_TCP)
1267 		pd->flags |= PFDESC_TCP_NORM;
1268 
1269 	flags = th->th_flags;
1270 	if (flags & TH_SYN) {
1271 		/* Illegal packet */
1272 		if (flags & TH_RST)
1273 			goto tcp_drop;
1274 
1275 		if (flags & TH_FIN)
1276 			flags &= ~TH_FIN;
1277 	} else {
1278 		/* Illegal packet */
1279 		if (!(flags & (TH_ACK|TH_RST)))
1280 			goto tcp_drop;
1281 	}
1282 
1283 	if (!(flags & TH_ACK)) {
1284 		/* These flags are only valid if ACK is set */
1285 		if ((flags & TH_FIN) || (flags & TH_PUSH) || (flags & TH_URG))
1286 			goto tcp_drop;
1287 	}
1288 
1289 	/* Check for illegal header length */
1290 	if (th->th_off < (sizeof(struct tcphdr) >> 2))
1291 		goto tcp_drop;
1292 
1293 	/* If flags changed, or reserved data set, then adjust */
1294 	if (flags != th->th_flags || th->th_x2 != 0) {
1295 		u_int16_t	ov, nv;
1296 
1297 		ov = *(u_int16_t *)(&th->th_ack + 1);
1298 		th->th_flags = flags;
1299 		th->th_x2 = 0;
1300 		nv = *(u_int16_t *)(&th->th_ack + 1);
1301 
1302 		th->th_sum = pf_cksum_fixup(th->th_sum, ov, nv);
1303 		rewrite = 1;
1304 	}
1305 
1306 	/* Remove urgent pointer, if TH_URG is not set */
1307 	if (!(flags & TH_URG) && th->th_urp) {
1308 		th->th_sum = pf_cksum_fixup(th->th_sum, th->th_urp, 0);
1309 		th->th_urp = 0;
1310 		rewrite = 1;
1311 	}
1312 
1313 	/* Process options */
1314 	if (r->max_mss && pf_normalize_tcpopt(r, m, th, off))
1315 		rewrite = 1;
1316 
1317 	/* copy back packet headers if we sanitized */
1318 	if (rewrite)
1319 		m_copyback(m, off, sizeof(*th), (caddr_t)th);
1320 
1321 	return (PF_PASS);
1322 
1323  tcp_drop:
1324 	REASON_SET(&reason, PFRES_NORM);
1325 	if (rm != NULL && r->log)
1326 		PFLOG_PACKET(kif, h, m, AF_INET, dir, reason, r, NULL, NULL);
1327 	return (PF_DROP);
1328 }
1329 
1330 int
1331 pf_normalize_tcp_init(struct mbuf *m, int off, struct pf_pdesc *pd,
1332     struct tcphdr *th, struct pf_state_peer *src, struct pf_state_peer *dst)
1333 {
1334 	u_int8_t hdr[60];
1335 	u_int8_t *opt;
1336 
1337 	KASSERT((src->scrub == NULL),
1338 	    ("pf_normalize_tcp_init: src->scrub != NULL"));
1339 
1340 	src->scrub = pool_get(&pf_state_scrub_pl, PR_NOWAIT);
1341 	if (src->scrub == NULL)
1342 		return (1);
1343 	bzero(src->scrub, sizeof(*src->scrub));
1344 
1345 	switch (pd->af) {
1346 #ifdef INET
1347 	case AF_INET: {
1348 		struct ip *h = mtod(m, struct ip *);
1349 		src->scrub->pfss_ttl = h->ip_ttl;
1350 		break;
1351 	}
1352 #endif /* INET */
1353 #ifdef INET6
1354 	case AF_INET6: {
1355 		struct ip6_hdr *h = mtod(m, struct ip6_hdr *);
1356 		src->scrub->pfss_ttl = h->ip6_hlim;
1357 		break;
1358 	}
1359 #endif /* INET6 */
1360 	}
1361 
1362 
1363 	/*
1364 	 * All normalizations below are only begun if we see the start of
1365 	 * the connections.  They must all set an enabled bit in pfss_flags
1366 	 */
1367 	if ((th->th_flags & TH_SYN) == 0)
1368 		return (0);
1369 
1370 
1371 	if (th->th_off > (sizeof(struct tcphdr) >> 2) && src->scrub &&
1372 	    pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) {
1373 		/* Diddle with TCP options */
1374 		int hlen;
1375 		opt = hdr + sizeof(struct tcphdr);
1376 		hlen = (th->th_off << 2) - sizeof(struct tcphdr);
1377 		while (hlen >= TCPOLEN_TIMESTAMP) {
1378 			switch (*opt) {
1379 			case TCPOPT_EOL:	/* FALLTHROUGH */
1380 			case TCPOPT_NOP:
1381 				opt++;
1382 				hlen--;
1383 				break;
1384 			case TCPOPT_TIMESTAMP:
1385 				if (opt[1] >= TCPOLEN_TIMESTAMP) {
1386 					src->scrub->pfss_flags |=
1387 					    PFSS_TIMESTAMP;
1388 					src->scrub->pfss_ts_mod = arc4random();
1389 				}
1390 				/* FALLTHROUGH */
1391 			default:
1392 				hlen -= MAX(opt[1], 2);
1393 				opt += MAX(opt[1], 2);
1394 				break;
1395 			}
1396 		}
1397 	}
1398 
1399 	return (0);
1400 }
1401 
1402 void
1403 pf_normalize_tcp_cleanup(struct pf_state *state)
1404 {
1405 	if (state->src.scrub)
1406 		pool_put(&pf_state_scrub_pl, state->src.scrub);
1407 	if (state->dst.scrub)
1408 		pool_put(&pf_state_scrub_pl, state->dst.scrub);
1409 
1410 	/* Someday... flush the TCP segment reassembly descriptors. */
1411 }
1412 
1413 int
1414 pf_normalize_tcp_stateful(struct mbuf *m, int off, struct pf_pdesc *pd,
1415     u_short *reason, struct tcphdr *th, struct pf_state_peer *src,
1416     struct pf_state_peer *dst, int *writeback)
1417 {
1418 	u_int8_t hdr[60];
1419 	u_int8_t *opt;
1420 	int copyback = 0;
1421 
1422 	KASSERT((src->scrub || dst->scrub),
1423 	    ("pf_normalize_tcp_statefull: src->scrub && dst->scrub!"));
1424 
1425 	/*
1426 	 * Enforce the minimum TTL seen for this connection.  Negate a common
1427 	 * technique to evade an intrusion detection system and confuse
1428 	 * firewall state code.
1429 	 */
1430 	switch (pd->af) {
1431 #ifdef INET
1432 	case AF_INET: {
1433 		if (src->scrub) {
1434 			struct ip *h = mtod(m, struct ip *);
1435 			if (h->ip_ttl > src->scrub->pfss_ttl)
1436 				src->scrub->pfss_ttl = h->ip_ttl;
1437 			h->ip_ttl = src->scrub->pfss_ttl;
1438 		}
1439 		break;
1440 	}
1441 #endif /* INET */
1442 #ifdef INET6
1443 	case AF_INET6: {
1444 		if (src->scrub) {
1445 			struct ip6_hdr *h = mtod(m, struct ip6_hdr *);
1446 			if (h->ip6_hlim > src->scrub->pfss_ttl)
1447 				src->scrub->pfss_ttl = h->ip6_hlim;
1448 			h->ip6_hlim = src->scrub->pfss_ttl;
1449 		}
1450 		break;
1451 	}
1452 #endif /* INET6 */
1453 	}
1454 
1455 	if (th->th_off > (sizeof(struct tcphdr) >> 2) &&
1456 	    ((src->scrub && (src->scrub->pfss_flags & PFSS_TIMESTAMP)) ||
1457 	    (dst->scrub && (dst->scrub->pfss_flags & PFSS_TIMESTAMP))) &&
1458 	    pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) {
1459 		/* Diddle with TCP options */
1460 		int hlen;
1461 		opt = hdr + sizeof(struct tcphdr);
1462 		hlen = (th->th_off << 2) - sizeof(struct tcphdr);
1463 		while (hlen >= TCPOLEN_TIMESTAMP) {
1464 			switch (*opt) {
1465 			case TCPOPT_EOL:	/* FALLTHROUGH */
1466 			case TCPOPT_NOP:
1467 				opt++;
1468 				hlen--;
1469 				break;
1470 			case TCPOPT_TIMESTAMP:
1471 				/* Modulate the timestamps.  Can be used for
1472 				 * NAT detection, OS uptime determination or
1473 				 * reboot detection.
1474 				 */
1475 				if (opt[1] >= TCPOLEN_TIMESTAMP) {
1476 					u_int32_t ts_value;
1477 					if (src->scrub &&
1478 					    (src->scrub->pfss_flags &
1479 					    PFSS_TIMESTAMP)) {
1480 						memcpy(&ts_value, &opt[2],
1481 						    sizeof(u_int32_t));
1482 						ts_value = htonl(ntohl(ts_value)
1483 						    + src->scrub->pfss_ts_mod);
1484 						pf_change_a(&opt[2],
1485 						    &th->th_sum, ts_value, 0);
1486 						copyback = 1;
1487 					}
1488 
1489 					/* Modulate TS reply iff valid (!0) */
1490 					memcpy(&ts_value, &opt[6],
1491 					    sizeof(u_int32_t));
1492 					if (ts_value && dst->scrub &&
1493 					    (dst->scrub->pfss_flags &
1494 					    PFSS_TIMESTAMP)) {
1495 						ts_value = htonl(ntohl(ts_value)
1496 						    - dst->scrub->pfss_ts_mod);
1497 						pf_change_a(&opt[6],
1498 						    &th->th_sum, ts_value, 0);
1499 						copyback = 1;
1500 					}
1501 				}
1502 				/* FALLTHROUGH */
1503 			default:
1504 				hlen -= MAX(opt[1], 2);
1505 				opt += MAX(opt[1], 2);
1506 				break;
1507 			}
1508 		}
1509 		if (copyback) {
1510 			/* Copyback the options, caller copys back header */
1511 			*writeback = 1;
1512 			m_copyback(m, off + sizeof(struct tcphdr),
1513 			    (th->th_off << 2) - sizeof(struct tcphdr), hdr +
1514 			    sizeof(struct tcphdr));
1515 		}
1516 	}
1517 
1518 
1519 	/* I have a dream....  TCP segment reassembly.... */
1520 	return (0);
1521 }
1522 
1523 int
1524 pf_normalize_tcpopt(struct pf_rule *r, struct mbuf *m, struct tcphdr *th,
1525     int off)
1526 {
1527 	u_int16_t	*mss;
1528 	int		 thoff;
1529 	int		 opt, cnt, optlen = 0;
1530 	int		 rewrite = 0;
1531 	u_char		*optp;
1532 
1533 	thoff = th->th_off << 2;
1534 	cnt = thoff - sizeof(struct tcphdr);
1535 	optp = mtod(m, caddr_t) + off + sizeof(struct tcphdr);
1536 
1537 	for (; cnt > 0; cnt -= optlen, optp += optlen) {
1538 		opt = optp[0];
1539 		if (opt == TCPOPT_EOL)
1540 			break;
1541 		if (opt == TCPOPT_NOP)
1542 			optlen = 1;
1543 		else {
1544 			if (cnt < 2)
1545 				break;
1546 			optlen = optp[1];
1547 			if (optlen < 2 || optlen > cnt)
1548 				break;
1549 		}
1550 		switch (opt) {
1551 		case TCPOPT_MAXSEG:
1552 			mss = (u_int16_t *)(optp + 2);
1553 			if ((ntohs(*mss)) > r->max_mss) {
1554 				th->th_sum = pf_cksum_fixup(th->th_sum,
1555 				    *mss, htons(r->max_mss));
1556 				*mss = htons(r->max_mss);
1557 				rewrite = 1;
1558 			}
1559 			break;
1560 		default:
1561 			break;
1562 		}
1563 	}
1564 
1565 	return (rewrite);
1566 }
1567