xref: /netbsd-src/sys/dist/pf/net/pf_norm.c (revision 10ad5ffa714ce1a679dcc9dd8159648df2d67b5a)
1 /*	$NetBSD: pf_norm.c,v 1.21 2009/07/28 18:15:26 minskim Exp $	*/
2 /*	$OpenBSD: pf_norm.c,v 1.109 2007/05/28 17:16:39 henning Exp $ */
3 
4 /*
5  * Copyright 2001 Niels Provos <provos@citi.umich.edu>
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __KERNEL_RCSID(0, "$NetBSD: pf_norm.c,v 1.21 2009/07/28 18:15:26 minskim Exp $");
31 
32 #ifdef _KERNEL_OPT
33 #include "opt_inet.h"
34 #endif
35 
36 #include "pflog.h"
37 
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/mbuf.h>
41 #include <sys/filio.h>
42 #include <sys/fcntl.h>
43 #include <sys/socket.h>
44 #include <sys/kernel.h>
45 #include <sys/time.h>
46 #include <sys/pool.h>
47 
48 #ifdef __NetBSD__
49 #include <sys/rnd.h>
50 #else
51 #include <dev/rndvar.h>
52 #endif /* !__NetBSD__ */
53 #include <net/if.h>
54 #include <net/if_types.h>
55 #include <net/bpf.h>
56 #include <net/route.h>
57 #include <net/if_pflog.h>
58 
59 #include <netinet/in.h>
60 #include <netinet/in_var.h>
61 #include <netinet/in_systm.h>
62 #include <netinet/ip.h>
63 #include <netinet/ip_var.h>
64 #include <netinet/tcp.h>
65 #include <netinet/tcp_seq.h>
66 #include <netinet/udp.h>
67 #include <netinet/ip_icmp.h>
68 
69 #ifdef INET6
70 #include <netinet/ip6.h>
71 #endif /* INET6 */
72 
73 #include <net/pfvar.h>
74 
75 struct pf_frent {
76 	LIST_ENTRY(pf_frent) fr_next;
77 	struct ip *fr_ip;
78 	struct mbuf *fr_m;
79 };
80 
81 struct pf_frcache {
82 	LIST_ENTRY(pf_frcache) fr_next;
83 	uint16_t	fr_off;
84 	uint16_t	fr_end;
85 };
86 
87 #define PFFRAG_SEENLAST	0x0001		/* Seen the last fragment for this */
88 #define PFFRAG_NOBUFFER	0x0002		/* Non-buffering fragment cache */
89 #define PFFRAG_DROP	0x0004		/* Drop all fragments */
90 #define BUFFER_FRAGMENTS(fr)	(!((fr)->fr_flags & PFFRAG_NOBUFFER))
91 
92 struct pf_fragment {
93 	RB_ENTRY(pf_fragment) fr_entry;
94 	TAILQ_ENTRY(pf_fragment) frag_next;
95 	struct in_addr	fr_src;
96 	struct in_addr	fr_dst;
97 	u_int8_t	fr_p;		/* protocol of this fragment */
98 	u_int8_t	fr_flags;	/* status flags */
99 	u_int16_t	fr_id;		/* fragment id for reassemble */
100 	u_int16_t	fr_max;		/* fragment data max */
101 	u_int32_t	fr_timeout;
102 #define fr_queue	fr_u.fru_queue
103 #define fr_cache	fr_u.fru_cache
104 	union {
105 		LIST_HEAD(pf_fragq, pf_frent) fru_queue;	/* buffering */
106 		LIST_HEAD(pf_cacheq, pf_frcache) fru_cache;	/* non-buf */
107 	} fr_u;
108 };
109 
110 TAILQ_HEAD(pf_fragqueue, pf_fragment)	pf_fragqueue;
111 TAILQ_HEAD(pf_cachequeue, pf_fragment)	pf_cachequeue;
112 
113 static __inline int	 pf_frag_compare(struct pf_fragment *,
114 			    struct pf_fragment *);
115 RB_HEAD(pf_frag_tree, pf_fragment)	pf_frag_tree, pf_cache_tree;
116 RB_PROTOTYPE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
117 RB_GENERATE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
118 
119 /* Private prototypes */
120 void			 pf_ip2key(struct pf_fragment *, struct ip *);
121 void			 pf_remove_fragment(struct pf_fragment *);
122 void			 pf_flush_fragments(void);
123 void			 pf_free_fragment(struct pf_fragment *);
124 struct pf_fragment	*pf_find_fragment(struct ip *, struct pf_frag_tree *);
125 struct mbuf		*pf_reassemble(struct mbuf **, struct pf_fragment **,
126 			    struct pf_frent *, int);
127 struct mbuf		*pf_fragcache(struct mbuf **, struct ip*,
128 			    struct pf_fragment **, int, int, int *);
129 int			 pf_normalize_tcpopt(struct pf_rule *, struct mbuf *,
130 			    struct tcphdr *, int);
131 
132 #define	DPFPRINTF(x) do {				\
133 	if (pf_status.debug >= PF_DEBUG_MISC) {		\
134 		printf("%s: ", __func__);		\
135 		printf x ;				\
136 	}						\
137 } while(0)
138 
139 /* Globals */
140 struct pool		 pf_frent_pl, pf_frag_pl, pf_cache_pl, pf_cent_pl;
141 struct pool		 pf_state_scrub_pl;
142 int			 pf_nfrents, pf_ncache;
143 
144 void
145 pf_normalize_init(void)
146 {
147 #ifdef __NetBSD__
148 	pool_init(&pf_frent_pl, sizeof(struct pf_frent), 0, 0, 0, "pffrent",
149 	    NULL, IPL_SOFTNET);
150 	pool_init(&pf_frag_pl, sizeof(struct pf_fragment), 0, 0, 0, "pffrag",
151 	    NULL, IPL_SOFTNET);
152 	pool_init(&pf_cache_pl, sizeof(struct pf_fragment), 0, 0, 0,
153 	    "pffrcache", NULL, IPL_SOFTNET);
154 	pool_init(&pf_cent_pl, sizeof(struct pf_frcache), 0, 0, 0, "pffrcent",
155 	    NULL, IPL_SOFTNET);
156 	pool_init(&pf_state_scrub_pl, sizeof(struct pf_state_scrub), 0, 0, 0,
157 	    "pfstscr", NULL, IPL_SOFTNET);
158 #else
159 	pool_init(&pf_frent_pl, sizeof(struct pf_frent), 0, 0, 0, "pffrent",
160 	    NULL);
161 	pool_init(&pf_frag_pl, sizeof(struct pf_fragment), 0, 0, 0, "pffrag",
162 	    NULL);
163 	pool_init(&pf_cache_pl, sizeof(struct pf_fragment), 0, 0, 0,
164 	    "pffrcache", NULL);
165 	pool_init(&pf_cent_pl, sizeof(struct pf_frcache), 0, 0, 0, "pffrcent",
166 	    NULL);
167 	pool_init(&pf_state_scrub_pl, sizeof(struct pf_state_scrub), 0, 0, 0,
168 	    "pfstscr", NULL);
169 #endif /* !__NetBSD__ */
170 
171 	pool_sethiwat(&pf_frag_pl, PFFRAG_FRAG_HIWAT);
172 	pool_sethardlimit(&pf_frent_pl, PFFRAG_FRENT_HIWAT, NULL, 0);
173 	pool_sethardlimit(&pf_cache_pl, PFFRAG_FRCACHE_HIWAT, NULL, 0);
174 	pool_sethardlimit(&pf_cent_pl, PFFRAG_FRCENT_HIWAT, NULL, 0);
175 
176 	TAILQ_INIT(&pf_fragqueue);
177 	TAILQ_INIT(&pf_cachequeue);
178 }
179 
180 static __inline int
181 pf_frag_compare(struct pf_fragment *a, struct pf_fragment *b)
182 {
183 	int	diff;
184 
185 	if ((diff = a->fr_id - b->fr_id))
186 		return (diff);
187 	else if ((diff = a->fr_p - b->fr_p))
188 		return (diff);
189 	else if (a->fr_src.s_addr < b->fr_src.s_addr)
190 		return (-1);
191 	else if (a->fr_src.s_addr > b->fr_src.s_addr)
192 		return (1);
193 	else if (a->fr_dst.s_addr < b->fr_dst.s_addr)
194 		return (-1);
195 	else if (a->fr_dst.s_addr > b->fr_dst.s_addr)
196 		return (1);
197 	return (0);
198 }
199 
200 void
201 pf_purge_expired_fragments(void)
202 {
203 	struct pf_fragment	*frag;
204 	u_int32_t		 expire = time_second -
205 				    pf_default_rule.timeout[PFTM_FRAG];
206 
207 	while ((frag = TAILQ_LAST(&pf_fragqueue, pf_fragqueue)) != NULL) {
208 		KASSERT(BUFFER_FRAGMENTS(frag));
209 		if (frag->fr_timeout > expire)
210 			break;
211 
212 		DPFPRINTF(("expiring %d(%p)\n", frag->fr_id, frag));
213 		pf_free_fragment(frag);
214 	}
215 
216 	while ((frag = TAILQ_LAST(&pf_cachequeue, pf_cachequeue)) != NULL) {
217 		KASSERT(!BUFFER_FRAGMENTS(frag));
218 		if (frag->fr_timeout > expire)
219 			break;
220 
221 		DPFPRINTF(("expiring %d(%p)\n", frag->fr_id, frag));
222 		pf_free_fragment(frag);
223 		KASSERT(TAILQ_EMPTY(&pf_cachequeue) ||
224 		    TAILQ_LAST(&pf_cachequeue, pf_cachequeue) != frag);
225 	}
226 }
227 
228 /*
229  * Try to flush old fragments to make space for new ones
230  */
231 
232 void
233 pf_flush_fragments(void)
234 {
235 	struct pf_fragment	*frag;
236 	int			 goal;
237 
238 	goal = pf_nfrents * 9 / 10;
239 	DPFPRINTF(("trying to free > %d frents\n",
240 	    pf_nfrents - goal));
241 	while (goal < pf_nfrents) {
242 		frag = TAILQ_LAST(&pf_fragqueue, pf_fragqueue);
243 		if (frag == NULL)
244 			break;
245 		pf_free_fragment(frag);
246 	}
247 
248 
249 	goal = pf_ncache * 9 / 10;
250 	DPFPRINTF(("trying to free > %d cache entries\n",
251 	    pf_ncache - goal));
252 	while (goal < pf_ncache) {
253 		frag = TAILQ_LAST(&pf_cachequeue, pf_cachequeue);
254 		if (frag == NULL)
255 			break;
256 		pf_free_fragment(frag);
257 	}
258 }
259 
260 /* Frees the fragments and all associated entries */
261 
262 void
263 pf_free_fragment(struct pf_fragment *frag)
264 {
265 	struct pf_frent		*frent;
266 	struct pf_frcache	*frcache;
267 
268 	/* Free all fragments */
269 	if (BUFFER_FRAGMENTS(frag)) {
270 		for (frent = LIST_FIRST(&frag->fr_queue); frent;
271 		    frent = LIST_FIRST(&frag->fr_queue)) {
272 			LIST_REMOVE(frent, fr_next);
273 
274 			m_freem(frent->fr_m);
275 			pool_put(&pf_frent_pl, frent);
276 			pf_nfrents--;
277 		}
278 	} else {
279 		for (frcache = LIST_FIRST(&frag->fr_cache); frcache;
280 		    frcache = LIST_FIRST(&frag->fr_cache)) {
281 			LIST_REMOVE(frcache, fr_next);
282 
283 			KASSERT(LIST_EMPTY(&frag->fr_cache) ||
284 			    LIST_FIRST(&frag->fr_cache)->fr_off >
285 			    frcache->fr_end);
286 
287 			pool_put(&pf_cent_pl, frcache);
288 			pf_ncache--;
289 		}
290 	}
291 
292 	pf_remove_fragment(frag);
293 }
294 
295 void
296 pf_ip2key(struct pf_fragment *key, struct ip *ip)
297 {
298 	key->fr_p = ip->ip_p;
299 	key->fr_id = ip->ip_id;
300 	key->fr_src.s_addr = ip->ip_src.s_addr;
301 	key->fr_dst.s_addr = ip->ip_dst.s_addr;
302 }
303 
304 struct pf_fragment *
305 pf_find_fragment(struct ip *ip, struct pf_frag_tree *tree)
306 {
307 	struct pf_fragment	 key;
308 	struct pf_fragment	*frag;
309 
310 	pf_ip2key(&key, ip);
311 
312 	frag = RB_FIND(pf_frag_tree, tree, &key);
313 	if (frag != NULL) {
314 		/* XXX Are we sure we want to update the timeout? */
315 		frag->fr_timeout = time_second;
316 		if (BUFFER_FRAGMENTS(frag)) {
317 			TAILQ_REMOVE(&pf_fragqueue, frag, frag_next);
318 			TAILQ_INSERT_HEAD(&pf_fragqueue, frag, frag_next);
319 		} else {
320 			TAILQ_REMOVE(&pf_cachequeue, frag, frag_next);
321 			TAILQ_INSERT_HEAD(&pf_cachequeue, frag, frag_next);
322 		}
323 	}
324 
325 	return (frag);
326 }
327 
328 /* Removes a fragment from the fragment queue and frees the fragment */
329 
330 void
331 pf_remove_fragment(struct pf_fragment *frag)
332 {
333 	if (BUFFER_FRAGMENTS(frag)) {
334 		RB_REMOVE(pf_frag_tree, &pf_frag_tree, frag);
335 		TAILQ_REMOVE(&pf_fragqueue, frag, frag_next);
336 		pool_put(&pf_frag_pl, frag);
337 	} else {
338 		RB_REMOVE(pf_frag_tree, &pf_cache_tree, frag);
339 		TAILQ_REMOVE(&pf_cachequeue, frag, frag_next);
340 		pool_put(&pf_cache_pl, frag);
341 	}
342 }
343 
344 #define FR_IP_OFF(fr)	((ntohs((fr)->fr_ip->ip_off) & IP_OFFMASK) << 3)
345 struct mbuf *
346 pf_reassemble(struct mbuf **m0, struct pf_fragment **frag,
347     struct pf_frent *frent, int mff)
348 {
349 	struct mbuf	*m = *m0, *m2;
350 	struct pf_frent	*frea, *next;
351 	struct pf_frent	*frep = NULL;
352 	struct ip	*ip = frent->fr_ip;
353 	int		 hlen = ip->ip_hl << 2;
354 	u_int16_t	 off = (ntohs(ip->ip_off) & IP_OFFMASK) << 3;
355 	u_int16_t	 ip_len = ntohs(ip->ip_len) - ip->ip_hl * 4;
356 	u_int16_t	 max = ip_len + off;
357 
358 	KASSERT(*frag == NULL || BUFFER_FRAGMENTS(*frag));
359 
360 	/* Strip off ip header */
361 	m->m_data += hlen;
362 	m->m_len -= hlen;
363 
364 	/* Create a new reassembly queue for this packet */
365 	if (*frag == NULL) {
366 		*frag = pool_get(&pf_frag_pl, PR_NOWAIT);
367 		if (*frag == NULL) {
368 			pf_flush_fragments();
369 			*frag = pool_get(&pf_frag_pl, PR_NOWAIT);
370 			if (*frag == NULL)
371 				goto drop_fragment;
372 		}
373 
374 		(*frag)->fr_flags = 0;
375 		(*frag)->fr_max = 0;
376 		(*frag)->fr_src = frent->fr_ip->ip_src;
377 		(*frag)->fr_dst = frent->fr_ip->ip_dst;
378 		(*frag)->fr_p = frent->fr_ip->ip_p;
379 		(*frag)->fr_id = frent->fr_ip->ip_id;
380 		(*frag)->fr_timeout = time_second;
381 		LIST_INIT(&(*frag)->fr_queue);
382 
383 		RB_INSERT(pf_frag_tree, &pf_frag_tree, *frag);
384 		TAILQ_INSERT_HEAD(&pf_fragqueue, *frag, frag_next);
385 
386 		/* We do not have a previous fragment */
387 		frep = NULL;
388 		goto insert;
389 	}
390 
391 	/*
392 	 * Find a fragment after the current one:
393 	 *  - off contains the real shifted offset.
394 	 */
395 	LIST_FOREACH(frea, &(*frag)->fr_queue, fr_next) {
396 		if (FR_IP_OFF(frea) > off)
397 			break;
398 		frep = frea;
399 	}
400 
401 	KASSERT(frep != NULL || frea != NULL);
402 
403 	if (frep != NULL &&
404 	    FR_IP_OFF(frep) + ntohs(frep->fr_ip->ip_len) - frep->fr_ip->ip_hl *
405 	    4 > off)
406 	{
407 		u_int16_t	precut;
408 
409 		precut = FR_IP_OFF(frep) + ntohs(frep->fr_ip->ip_len) -
410 		    frep->fr_ip->ip_hl * 4 - off;
411 		if (precut >= ip_len)
412 			goto drop_fragment;
413 		m_adj(frent->fr_m, precut);
414 		DPFPRINTF(("overlap -%d\n", precut));
415 		/* Enforce 8 byte boundaries */
416 		ip->ip_off = htons(ntohs(ip->ip_off) + (precut >> 3));
417 		off = (ntohs(ip->ip_off) & IP_OFFMASK) << 3;
418 		ip_len -= precut;
419 		ip->ip_len = htons(ip_len);
420 	}
421 
422 	for (; frea != NULL && ip_len + off > FR_IP_OFF(frea);
423 	    frea = next)
424 	{
425 		u_int16_t	aftercut;
426 
427 		aftercut = ip_len + off - FR_IP_OFF(frea);
428 		DPFPRINTF(("adjust overlap %d\n", aftercut));
429 		if (aftercut < ntohs(frea->fr_ip->ip_len) - frea->fr_ip->ip_hl
430 		    * 4)
431 		{
432 			frea->fr_ip->ip_len =
433 			    htons(ntohs(frea->fr_ip->ip_len) - aftercut);
434 			frea->fr_ip->ip_off = htons(ntohs(frea->fr_ip->ip_off) +
435 			    (aftercut >> 3));
436 			m_adj(frea->fr_m, aftercut);
437 			break;
438 		}
439 
440 		/* This fragment is completely overlapped, lose it */
441 		next = LIST_NEXT(frea, fr_next);
442 		m_freem(frea->fr_m);
443 		LIST_REMOVE(frea, fr_next);
444 		pool_put(&pf_frent_pl, frea);
445 		pf_nfrents--;
446 	}
447 
448  insert:
449 	/* Update maximum data size */
450 	if ((*frag)->fr_max < max)
451 		(*frag)->fr_max = max;
452 	/* This is the last segment */
453 	if (!mff)
454 		(*frag)->fr_flags |= PFFRAG_SEENLAST;
455 
456 	if (frep == NULL)
457 		LIST_INSERT_HEAD(&(*frag)->fr_queue, frent, fr_next);
458 	else
459 		LIST_INSERT_AFTER(frep, frent, fr_next);
460 
461 	/* Check if we are completely reassembled */
462 	if (!((*frag)->fr_flags & PFFRAG_SEENLAST))
463 		return (NULL);
464 
465 	/* Check if we have all the data */
466 	off = 0;
467 	for (frep = LIST_FIRST(&(*frag)->fr_queue); frep; frep = next) {
468 		next = LIST_NEXT(frep, fr_next);
469 
470 		off += ntohs(frep->fr_ip->ip_len) - frep->fr_ip->ip_hl * 4;
471 		if (off < (*frag)->fr_max &&
472 		    (next == NULL || FR_IP_OFF(next) != off))
473 		{
474 			DPFPRINTF(("missing fragment at %d, next %d, max %d\n",
475 			    off, next == NULL ? -1 : FR_IP_OFF(next),
476 			    (*frag)->fr_max));
477 			return (NULL);
478 		}
479 	}
480 	DPFPRINTF(("%d < %d?\n", off, (*frag)->fr_max));
481 	if (off < (*frag)->fr_max)
482 		return (NULL);
483 
484 	/* We have all the data */
485 	frent = LIST_FIRST(&(*frag)->fr_queue);
486 	KASSERT(frent != NULL);
487 	if ((frent->fr_ip->ip_hl << 2) + off > IP_MAXPACKET) {
488 		DPFPRINTF(("drop: too big: %d\n", off));
489 		pf_free_fragment(*frag);
490 		*frag = NULL;
491 		return (NULL);
492 	}
493 	next = LIST_NEXT(frent, fr_next);
494 
495 	/* Magic from ip_input */
496 	ip = frent->fr_ip;
497 	m = frent->fr_m;
498 	m2 = m->m_next;
499 	m->m_next = NULL;
500 	m_cat(m, m2);
501 	pool_put(&pf_frent_pl, frent);
502 	pf_nfrents--;
503 	for (frent = next; frent != NULL; frent = next) {
504 		next = LIST_NEXT(frent, fr_next);
505 
506 		m2 = frent->fr_m;
507 		pool_put(&pf_frent_pl, frent);
508 		pf_nfrents--;
509 		m_cat(m, m2);
510 	}
511 
512 	ip->ip_src = (*frag)->fr_src;
513 	ip->ip_dst = (*frag)->fr_dst;
514 
515 	/* Remove from fragment queue */
516 	pf_remove_fragment(*frag);
517 	*frag = NULL;
518 
519 	hlen = ip->ip_hl << 2;
520 	ip->ip_len = htons(off + hlen);
521 	m->m_len += hlen;
522 	m->m_data -= hlen;
523 
524 	/* some debugging cruft by sklower, below, will go away soon */
525 	/* XXX this should be done elsewhere */
526 	if (m->m_flags & M_PKTHDR) {
527 		int plen = 0;
528 		for (m2 = m; m2; m2 = m2->m_next)
529 			plen += m2->m_len;
530 		m->m_pkthdr.len = plen;
531 #ifdef __NetBSD__
532 		m->m_pkthdr.csum_flags = 0;
533 #endif /* __NetBSD__ */
534 	}
535 
536 	DPFPRINTF(("complete: %p(%d)\n", m, ntohs(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 = ntohs(h->ip_len) - (h->ip_hl << 2);
554 	u_int16_t		 off = ntohs(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 
560 	/* Create a new range queue for this packet */
561 	if (*frag == NULL) {
562 		*frag = pool_get(&pf_cache_pl, PR_NOWAIT);
563 		if (*frag == NULL) {
564 			pf_flush_fragments();
565 			*frag = pool_get(&pf_cache_pl, PR_NOWAIT);
566 			if (*frag == NULL)
567 				goto no_mem;
568 		}
569 
570 		/* Get an entry for the queue */
571 		cur = pool_get(&pf_cent_pl, PR_NOWAIT);
572 		if (cur == NULL) {
573 			pool_put(&pf_cache_pl, *frag);
574 			*frag = NULL;
575 			goto no_mem;
576 		}
577 		pf_ncache++;
578 
579 		(*frag)->fr_flags = PFFRAG_NOBUFFER;
580 		(*frag)->fr_max = 0;
581 		(*frag)->fr_src = h->ip_src;
582 		(*frag)->fr_dst = h->ip_dst;
583 		(*frag)->fr_p = h->ip_p;
584 		(*frag)->fr_id = h->ip_id;
585 		(*frag)->fr_timeout = time_second;
586 
587 		cur->fr_off = off;
588 		cur->fr_end = max;
589 		LIST_INIT(&(*frag)->fr_cache);
590 		LIST_INSERT_HEAD(&(*frag)->fr_cache, cur, fr_next);
591 
592 		RB_INSERT(pf_frag_tree, &pf_cache_tree, *frag);
593 		TAILQ_INSERT_HEAD(&pf_cachequeue, *frag, frag_next);
594 
595 		DPFPRINTF(("fragcache[%d]: new %d-%d\n", h->ip_id, off, max));
596 
597 		goto pass;
598 	}
599 
600 	/*
601 	 * Find a fragment after the current one:
602 	 *  - off contains the real shifted offset.
603 	 */
604 	frp = NULL;
605 	LIST_FOREACH(fra, &(*frag)->fr_cache, fr_next) {
606 		if (fra->fr_off > off)
607 			break;
608 		frp = fra;
609 	}
610 
611 	KASSERT(frp != NULL || fra != NULL);
612 
613 	if (frp != NULL) {
614 		int	precut;
615 
616 		precut = frp->fr_end - off;
617 		if (precut >= ip_len) {
618 			/* Fragment is entirely a duplicate */
619 			DPFPRINTF(("fragcache[%d]: dead (%d-%d) %d-%d\n",
620 			    h->ip_id, frp->fr_off, frp->fr_end, off, max));
621 			goto drop_fragment;
622 		}
623 		if (precut == 0) {
624 			/* They are adjacent.  Fixup cache entry */
625 			DPFPRINTF(("fragcache[%d]: adjacent (%d-%d) %d-%d\n",
626 			    h->ip_id, frp->fr_off, frp->fr_end, off, max));
627 			frp->fr_end = max;
628 		} else if (precut > 0) {
629 			/* The first part of this payload overlaps with a
630 			 * fragment that has already been passed.
631 			 * Need to trim off the first part of the payload.
632 			 * But to do so easily, we need to create another
633 			 * mbuf to throw the original header into.
634 			 */
635 
636 			DPFPRINTF(("fragcache[%d]: chop %d (%d-%d) %d-%d\n",
637 			    h->ip_id, precut, frp->fr_off, frp->fr_end, off,
638 			    max));
639 
640 			off += precut;
641 			max -= precut;
642 			/* Update the previous frag to encompass this one */
643 			frp->fr_end = max;
644 
645 			if (!drop) {
646 				/* XXX Optimization opportunity
647 				 * This is a very heavy way to trim the payload.
648 				 * we could do it much faster by diddling mbuf
649 				 * internals but that would be even less legible
650 				 * than this mbuf magic.  For my next trick,
651 				 * I'll pull a rabbit out of my laptop.
652 				 */
653 				*m0 = m_dup(m, 0, h->ip_hl << 2, M_NOWAIT);
654 				if (*m0 == NULL)
655 					goto no_mem;
656 				KASSERT((*m0)->m_next == NULL);
657 				m_adj(m, precut + (h->ip_hl << 2));
658 				m_cat(*m0, m);
659 				m = *m0;
660 				if (m->m_flags & M_PKTHDR) {
661 					int plen = 0;
662 					struct mbuf *t;
663 					for (t = m; t; t = t->m_next)
664 						plen += t->m_len;
665 					m->m_pkthdr.len = plen;
666 				}
667 
668 
669 				h = mtod(m, struct ip *);
670 
671 
672 				KASSERT((int)m->m_len ==
673 				    ntohs(h->ip_len) - precut);
674 				h->ip_off = htons(ntohs(h->ip_off) +
675 				    (precut >> 3));
676 				h->ip_len = htons(ntohs(h->ip_len) - precut);
677 			} else {
678 				hosed++;
679 			}
680 		} else {
681 			/* There is a gap between fragments */
682 
683 			DPFPRINTF(("fragcache[%d]: gap %d (%d-%d) %d-%d\n",
684 			    h->ip_id, -precut, frp->fr_off, frp->fr_end, off,
685 			    max));
686 
687 			cur = pool_get(&pf_cent_pl, PR_NOWAIT);
688 			if (cur == NULL)
689 				goto no_mem;
690 			pf_ncache++;
691 
692 			cur->fr_off = off;
693 			cur->fr_end = max;
694 			LIST_INSERT_AFTER(frp, cur, fr_next);
695 		}
696 	}
697 
698 	if (fra != NULL) {
699 		int	aftercut;
700 		int	merge = 0;
701 
702 		aftercut = max - fra->fr_off;
703 		if (aftercut == 0) {
704 			/* Adjacent fragments */
705 			DPFPRINTF(("fragcache[%d]: adjacent %d-%d (%d-%d)\n",
706 			    h->ip_id, off, max, fra->fr_off, fra->fr_end));
707 			fra->fr_off = off;
708 			merge = 1;
709 		} else if (aftercut > 0) {
710 			/* Need to chop off the tail of this fragment */
711 			DPFPRINTF(("fragcache[%d]: chop %d %d-%d (%d-%d)\n",
712 			    h->ip_id, aftercut, off, max, fra->fr_off,
713 			    fra->fr_end));
714 			fra->fr_off = off;
715 			max -= aftercut;
716 
717 			merge = 1;
718 
719 			if (!drop) {
720 				m_adj(m, -aftercut);
721 				if (m->m_flags & M_PKTHDR) {
722 					int plen = 0;
723 					struct mbuf *t;
724 					for (t = m; t; t = t->m_next)
725 						plen += t->m_len;
726 					m->m_pkthdr.len = plen;
727 				}
728 				h = mtod(m, struct ip *);
729 				KASSERT((int)m->m_len ==
730 				    ntohs(h->ip_len) - aftercut);
731 				h->ip_len = htons(ntohs(h->ip_len) - aftercut);
732 			} else {
733 				hosed++;
734 			}
735 		} else if (frp == NULL) {
736 			/* There is a gap between fragments */
737 			DPFPRINTF(("fragcache[%d]: gap %d %d-%d (%d-%d)\n",
738 			    h->ip_id, -aftercut, off, max, fra->fr_off,
739 			    fra->fr_end));
740 
741 			cur = pool_get(&pf_cent_pl, PR_NOWAIT);
742 			if (cur == NULL)
743 				goto no_mem;
744 			pf_ncache++;
745 
746 			cur->fr_off = off;
747 			cur->fr_end = max;
748 			LIST_INSERT_BEFORE(fra, cur, fr_next);
749 		}
750 
751 
752 		/* Need to glue together two separate fragment descriptors */
753 		if (merge) {
754 			if (cur && fra->fr_off <= cur->fr_end) {
755 				/* Need to merge in a previous 'cur' */
756 				DPFPRINTF(("fragcache[%d]: adjacent(merge "
757 				    "%d-%d) %d-%d (%d-%d)\n",
758 				    h->ip_id, cur->fr_off, cur->fr_end, off,
759 				    max, fra->fr_off, fra->fr_end));
760 				fra->fr_off = cur->fr_off;
761 				LIST_REMOVE(cur, fr_next);
762 				pool_put(&pf_cent_pl, cur);
763 				pf_ncache--;
764 				cur = NULL;
765 
766 			} else if (frp && fra->fr_off <= frp->fr_end) {
767 				/* Need to merge in a modified 'frp' */
768 				KASSERT(cur == NULL);
769 				DPFPRINTF(("fragcache[%d]: adjacent(merge "
770 				    "%d-%d) %d-%d (%d-%d)\n",
771 				    h->ip_id, frp->fr_off, frp->fr_end, off,
772 				    max, fra->fr_off, fra->fr_end));
773 				fra->fr_off = frp->fr_off;
774 				LIST_REMOVE(frp, fr_next);
775 				pool_put(&pf_cent_pl, frp);
776 				pf_ncache--;
777 				frp = NULL;
778 
779 			}
780 		}
781 	}
782 
783 	if (hosed) {
784 		/*
785 		 * We must keep tracking the overall fragment even when
786 		 * we're going to drop it anyway so that we know when to
787 		 * free the overall descriptor.  Thus we drop the frag late.
788 		 */
789 		goto drop_fragment;
790 	}
791 
792 
793  pass:
794 	/* Update maximum data size */
795 	if ((*frag)->fr_max < max)
796 		(*frag)->fr_max = max;
797 
798 	/* This is the last segment */
799 	if (!mff)
800 		(*frag)->fr_flags |= PFFRAG_SEENLAST;
801 
802 	/* Check if we are completely reassembled */
803 	if (((*frag)->fr_flags & PFFRAG_SEENLAST) &&
804 	    LIST_FIRST(&(*frag)->fr_cache)->fr_off == 0 &&
805 	    LIST_FIRST(&(*frag)->fr_cache)->fr_end == (*frag)->fr_max) {
806 		/* Remove from fragment queue */
807 		DPFPRINTF(("fragcache[%d]: done 0-%d\n", h->ip_id,
808 		    (*frag)->fr_max));
809 		pf_free_fragment(*frag);
810 		*frag = NULL;
811 	}
812 
813 	return (m);
814 
815  no_mem:
816 	*nomem = 1;
817 
818 	/* Still need to pay attention to !IP_MF */
819 	if (!mff && *frag != NULL)
820 		(*frag)->fr_flags |= PFFRAG_SEENLAST;
821 
822 	m_freem(m);
823 	return (NULL);
824 
825  drop_fragment:
826 
827 	/* Still need to pay attention to !IP_MF */
828 	if (!mff && *frag != NULL)
829 		(*frag)->fr_flags |= PFFRAG_SEENLAST;
830 
831 	if (drop) {
832 		/* This fragment has been deemed bad.  Don't reass */
833 		if (((*frag)->fr_flags & PFFRAG_DROP) == 0)
834 			DPFPRINTF(("fragcache[%d]: dropping overall fragment\n",
835 			    h->ip_id));
836 		(*frag)->fr_flags |= PFFRAG_DROP;
837 	}
838 
839 	m_freem(m);
840 	return (NULL);
841 }
842 
843 int
844 pf_normalize_ip(struct mbuf **m0, int dir, struct pfi_kif *kif, u_short *reason,
845     struct pf_pdesc *pd)
846 {
847 	struct mbuf		*m = *m0;
848 	struct pf_rule		*r;
849 	struct pf_frent		*frent;
850 	struct pf_fragment	*frag = NULL;
851 	struct ip		*h = mtod(m, struct ip *);
852 	int			 mff = (ntohs(h->ip_off) & IP_MF);
853 	int			 hlen = h->ip_hl << 2;
854 	u_int16_t		 fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
855 	u_int16_t		 max;
856 	int			 ip_len;
857 	int			 ip_off;
858 
859 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
860 	while (r != NULL) {
861 		r->evaluations++;
862 		if (pfi_kif_match(r->kif, kif) == r->ifnot)
863 			r = r->skip[PF_SKIP_IFP].ptr;
864 		else if (r->direction && r->direction != dir)
865 			r = r->skip[PF_SKIP_DIR].ptr;
866 		else if (r->af && r->af != AF_INET)
867 			r = r->skip[PF_SKIP_AF].ptr;
868 		else if (r->proto && r->proto != h->ip_p)
869 			r = r->skip[PF_SKIP_PROTO].ptr;
870 		else if (PF_MISMATCHAW(&r->src.addr,
871 		    (struct pf_addr *)&h->ip_src.s_addr, AF_INET,
872 		    r->src.neg, kif))
873 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
874 		else if (PF_MISMATCHAW(&r->dst.addr,
875 		    (struct pf_addr *)&h->ip_dst.s_addr, AF_INET,
876 		    r->dst.neg, NULL))
877 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
878 		else
879 			break;
880 	}
881 
882 	if (r == NULL || r->action == PF_NOSCRUB)
883 		return (PF_PASS);
884 	else {
885 		r->packets[dir == PF_OUT]++;
886 		r->bytes[dir == PF_OUT] += pd->tot_len;
887 	}
888 
889 	/* Check for illegal packets */
890 	if (hlen < (int)sizeof(struct ip))
891 		goto drop;
892 
893 	if (hlen > ntohs(h->ip_len))
894 		goto drop;
895 
896 	/* Clear IP_DF if the rule uses the no-df option */
897 	if (r->rule_flag & PFRULE_NODF && h->ip_off & htons(IP_DF)) {
898 		u_int16_t ip_off = h->ip_off;
899 
900 		h->ip_off &= htons(~IP_DF);
901 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
902 	}
903 
904 	/* We will need other tests here */
905 	if (!fragoff && !mff)
906 		goto no_fragment;
907 
908 	/* We're dealing with a fragment now. Don't allow fragments
909 	 * with IP_DF to enter the cache. If the flag was cleared by
910 	 * no-df above, fine. Otherwise drop it.
911 	 */
912 	if (h->ip_off & htons(IP_DF)) {
913 		DPFPRINTF(("IP_DF\n"));
914 		goto bad;
915 	}
916 
917 	ip_len = ntohs(h->ip_len) - hlen;
918 	ip_off = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
919 
920 	/* All fragments are 8 byte aligned */
921 	if (mff && (ip_len & 0x7)) {
922 		DPFPRINTF(("mff and %d\n", ip_len));
923 		goto bad;
924 	}
925 
926 	/* Respect maximum length */
927 	if (fragoff + ip_len > IP_MAXPACKET) {
928 		DPFPRINTF(("max packet %d\n", fragoff + ip_len));
929 		goto bad;
930 	}
931 	max = fragoff + ip_len;
932 
933 	if ((r->rule_flag & (PFRULE_FRAGCROP|PFRULE_FRAGDROP)) == 0) {
934 		/* Fully buffer all of the fragments */
935 
936 		frag = pf_find_fragment(h, &pf_frag_tree);
937 
938 		/* Check if we saw the last fragment already */
939 		if (frag != NULL && (frag->fr_flags & PFFRAG_SEENLAST) &&
940 		    max > frag->fr_max)
941 			goto bad;
942 
943 		/* Get an entry for the fragment queue */
944 		frent = pool_get(&pf_frent_pl, PR_NOWAIT);
945 		if (frent == NULL) {
946 			REASON_SET(reason, PFRES_MEMORY);
947 			return (PF_DROP);
948 		}
949 		pf_nfrents++;
950 		frent->fr_ip = h;
951 		frent->fr_m = m;
952 
953 		/* Might return a completely reassembled mbuf, or NULL */
954 		DPFPRINTF(("reass frag %d @ %d-%d\n", h->ip_id, fragoff, max));
955 		*m0 = m = pf_reassemble(m0, &frag, frent, mff);
956 
957 		if (m == NULL)
958 			return (PF_DROP);
959 
960 		if (frag != NULL && (frag->fr_flags & PFFRAG_DROP))
961 			goto drop;
962 
963 		h = mtod(m, struct ip *);
964 	} else {
965 		/* non-buffering fragment cache (drops or masks overlaps) */
966 		int	nomem = 0;
967 
968 #ifdef __NetBSD__
969 		struct pf_mtag *pf_mtag = pf_find_mtag(m);
970 		KASSERT(pf_mtag != NULL);
971 
972 		if (dir == PF_OUT && pf_mtag->flags & PF_TAG_FRAGCACHE) {
973 #else
974 		if (dir == PF_OUT && m->m_pkthdr.pf.flags & PF_TAG_FRAGCACHE) {
975 #endif /* !__NetBSD__ */
976 			/*
977 			 * Already passed the fragment cache in the
978 			 * input direction.  If we continued, it would
979 			 * appear to be a dup and would be dropped.
980 			 */
981 			goto fragment_pass;
982 		}
983 
984 		frag = pf_find_fragment(h, &pf_cache_tree);
985 
986 		/* Check if we saw the last fragment already */
987 		if (frag != NULL && (frag->fr_flags & PFFRAG_SEENLAST) &&
988 		    max > frag->fr_max) {
989 			if (r->rule_flag & PFRULE_FRAGDROP)
990 				frag->fr_flags |= PFFRAG_DROP;
991 			goto bad;
992 		}
993 
994 		*m0 = m = pf_fragcache(m0, h, &frag, mff,
995 		    (r->rule_flag & PFRULE_FRAGDROP) ? 1 : 0, &nomem);
996 		if (m == NULL) {
997 			if (nomem)
998 				goto no_mem;
999 			goto drop;
1000 		}
1001 
1002 		if (dir == PF_IN)
1003 #ifdef __NetBSD__
1004 			pf_mtag = pf_find_mtag(m);
1005 			KASSERT(pf_mtag != NULL);
1006 
1007 			pf_mtag->flags |= PF_TAG_FRAGCACHE;
1008 #else
1009 			m->m_pkthdr.pf.flags |= PF_TAG_FRAGCACHE;
1010 #endif /* !__NetBSD__ */
1011 
1012 		if (frag != NULL && (frag->fr_flags & PFFRAG_DROP))
1013 			goto drop;
1014 		goto fragment_pass;
1015 	}
1016 
1017  no_fragment:
1018 	/* At this point, only IP_DF is allowed in ip_off */
1019 	if (h->ip_off & ~htons(IP_DF)) {
1020 		u_int16_t ip_off = h->ip_off;
1021 
1022 		h->ip_off &= htons(IP_DF);
1023 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
1024 	}
1025 
1026 	/* Enforce a minimum ttl, may cause endless packet loops */
1027 	if (r->min_ttl && h->ip_ttl < r->min_ttl) {
1028 		u_int16_t ip_ttl = h->ip_ttl;
1029 
1030 		h->ip_ttl = r->min_ttl;
1031 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_ttl, h->ip_ttl, 0);
1032 	}
1033 
1034 	if (r->rule_flag & PFRULE_RANDOMID) {
1035 		u_int16_t ip_id = h->ip_id;
1036 
1037 		h->ip_id = ip_randomid(0);
1038 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_id, h->ip_id, 0);
1039 	}
1040 	if ((r->rule_flag & (PFRULE_FRAGCROP|PFRULE_FRAGDROP)) == 0)
1041 		pd->flags |= PFDESC_IP_REAS;
1042 
1043 	return (PF_PASS);
1044 
1045  fragment_pass:
1046 	/* Enforce a minimum ttl, may cause endless packet loops */
1047 	if (r->min_ttl && h->ip_ttl < r->min_ttl) {
1048 		u_int16_t ip_ttl = h->ip_ttl;
1049 
1050 		h->ip_ttl = r->min_ttl;
1051 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_ttl, h->ip_ttl, 0);
1052 	}
1053 	if ((r->rule_flag & (PFRULE_FRAGCROP|PFRULE_FRAGDROP)) == 0)
1054 		pd->flags |= PFDESC_IP_REAS;
1055 	return (PF_PASS);
1056 
1057  no_mem:
1058 	REASON_SET(reason, PFRES_MEMORY);
1059 	if (r != NULL && r->log)
1060 		PFLOG_PACKET(kif, h, m, AF_INET, dir, *reason, r, NULL, NULL, pd);
1061 	return (PF_DROP);
1062 
1063  drop:
1064 	REASON_SET(reason, PFRES_NORM);
1065 	if (r != NULL && r->log)
1066 		PFLOG_PACKET(kif, h, m, AF_INET, dir, *reason, r, NULL, NULL, pd);
1067 	return (PF_DROP);
1068 
1069  bad:
1070 	DPFPRINTF(("dropping bad fragment\n"));
1071 
1072 	/* Free associated fragments */
1073 	if (frag != NULL)
1074 		pf_free_fragment(frag);
1075 
1076 	REASON_SET(reason, PFRES_FRAG);
1077 	if (r != NULL && r->log)
1078 		PFLOG_PACKET(kif, h, m, AF_INET, dir, *reason, r, NULL, NULL, pd);
1079 
1080 	return (PF_DROP);
1081 }
1082 
1083 #ifdef INET6
1084 int
1085 pf_normalize_ip6(struct mbuf **m0, int dir, struct pfi_kif *kif,
1086     u_short *reason, struct pf_pdesc *pd)
1087 {
1088 	struct mbuf		*m = *m0;
1089 	struct pf_rule		*r;
1090 	struct ip6_hdr		*h = mtod(m, struct ip6_hdr *);
1091 	int			 off;
1092 	struct ip6_ext		 ext;
1093 	struct ip6_opt		 opt;
1094 	struct ip6_opt_jumbo	 jumbo;
1095 	struct ip6_frag		 frag;
1096 	u_int32_t		 jumbolen = 0, plen;
1097 	u_int16_t		 fragoff = 0;
1098 	int			 optend;
1099 	int			 ooff;
1100 	u_int8_t		 proto;
1101 	int			 terminal;
1102 
1103 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1104 	while (r != NULL) {
1105 		r->evaluations++;
1106 		if (pfi_kif_match(r->kif, kif) == r->ifnot)
1107 			r = r->skip[PF_SKIP_IFP].ptr;
1108 		else if (r->direction && r->direction != dir)
1109 			r = r->skip[PF_SKIP_DIR].ptr;
1110 		else if (r->af && r->af != AF_INET6)
1111 			r = r->skip[PF_SKIP_AF].ptr;
1112 #if 0 /* header chain! */
1113 		else if (r->proto && r->proto != h->ip6_nxt)
1114 			r = r->skip[PF_SKIP_PROTO].ptr;
1115 #endif
1116 		else if (PF_MISMATCHAW(&r->src.addr,
1117 		    (struct pf_addr *)&h->ip6_src, AF_INET6,
1118 		    r->src.neg, kif))
1119 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
1120 		else if (PF_MISMATCHAW(&r->dst.addr,
1121 		    (struct pf_addr *)&h->ip6_dst, AF_INET6,
1122 		    r->dst.neg, NULL))
1123 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
1124 		else
1125 			break;
1126 	}
1127 
1128 	if (r == NULL || r->action == PF_NOSCRUB)
1129 		return (PF_PASS);
1130 	else {
1131 		r->packets[dir == PF_OUT]++;
1132 		r->bytes[dir == PF_OUT] += pd->tot_len;
1133 	}
1134 
1135 	/* Check for illegal packets */
1136 	if (sizeof(struct ip6_hdr) + IPV6_MAXPACKET < m->m_pkthdr.len)
1137 		goto drop;
1138 
1139 	off = sizeof(struct ip6_hdr);
1140 	proto = h->ip6_nxt;
1141 	terminal = 0;
1142 	do {
1143 		switch (proto) {
1144 		case IPPROTO_FRAGMENT:
1145 			goto fragment;
1146 			break;
1147 		case IPPROTO_AH:
1148 		case IPPROTO_ROUTING:
1149 		case IPPROTO_DSTOPTS:
1150 			if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL,
1151 			    NULL, AF_INET6))
1152 				goto shortpkt;
1153 			if (proto == IPPROTO_AH)
1154 				off += (ext.ip6e_len + 2) * 4;
1155 			else
1156 				off += (ext.ip6e_len + 1) * 8;
1157 			proto = ext.ip6e_nxt;
1158 			break;
1159 		case IPPROTO_HOPOPTS:
1160 			if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL,
1161 			    NULL, AF_INET6))
1162 				goto shortpkt;
1163 			optend = off + (ext.ip6e_len + 1) * 8;
1164 			ooff = off + sizeof(ext);
1165 			do {
1166 				if (!pf_pull_hdr(m, ooff, &opt.ip6o_type,
1167 				    sizeof(opt.ip6o_type), NULL, NULL,
1168 				    AF_INET6))
1169 					goto shortpkt;
1170 				if (opt.ip6o_type == IP6OPT_PAD1) {
1171 					ooff++;
1172 					continue;
1173 				}
1174 				if (!pf_pull_hdr(m, ooff, &opt, sizeof(opt),
1175 				    NULL, NULL, AF_INET6))
1176 					goto shortpkt;
1177 				if (ooff + sizeof(opt) + opt.ip6o_len > optend)
1178 					goto drop;
1179 				switch (opt.ip6o_type) {
1180 				case IP6OPT_JUMBO:
1181 					if (h->ip6_plen != 0)
1182 						goto drop;
1183 					if (!pf_pull_hdr(m, ooff, &jumbo,
1184 					    sizeof(jumbo), NULL, NULL,
1185 					    AF_INET6))
1186 						goto shortpkt;
1187 					memcpy(&jumbolen, jumbo.ip6oj_jumbo_len,
1188 					    sizeof(jumbolen));
1189 					jumbolen = ntohl(jumbolen);
1190 					if (jumbolen <= IPV6_MAXPACKET)
1191 						goto drop;
1192 					if (sizeof(struct ip6_hdr) + jumbolen !=
1193 					    m->m_pkthdr.len)
1194 						goto drop;
1195 					break;
1196 				default:
1197 					break;
1198 				}
1199 				ooff += sizeof(opt) + opt.ip6o_len;
1200 			} while (ooff < optend);
1201 
1202 			off = optend;
1203 			proto = ext.ip6e_nxt;
1204 			break;
1205 		default:
1206 			terminal = 1;
1207 			break;
1208 		}
1209 	} while (!terminal);
1210 
1211 	/* jumbo payload option must be present, or plen > 0 */
1212 	if (ntohs(h->ip6_plen) == 0)
1213 		plen = jumbolen;
1214 	else
1215 		plen = ntohs(h->ip6_plen);
1216 	if (plen == 0)
1217 		goto drop;
1218 	if (sizeof(struct ip6_hdr) + plen > m->m_pkthdr.len)
1219 		goto shortpkt;
1220 
1221 	/* Enforce a minimum ttl, may cause endless packet loops */
1222 	if (r->min_ttl && h->ip6_hlim < r->min_ttl)
1223 		h->ip6_hlim = r->min_ttl;
1224 
1225 	return (PF_PASS);
1226 
1227  fragment:
1228 	if (ntohs(h->ip6_plen) == 0 || jumbolen)
1229 		goto drop;
1230 	plen = ntohs(h->ip6_plen);
1231 
1232 	if (!pf_pull_hdr(m, off, &frag, sizeof(frag), NULL, NULL, AF_INET6))
1233 		goto shortpkt;
1234 	fragoff = ntohs(frag.ip6f_offlg & IP6F_OFF_MASK);
1235 	if (fragoff + (plen - off - sizeof(frag)) > IPV6_MAXPACKET)
1236 		goto badfrag;
1237 
1238 	/* do something about it */
1239 	/* remember to set pd->flags |= PFDESC_IP_REAS */
1240 	return (PF_PASS);
1241 
1242  shortpkt:
1243 	REASON_SET(reason, PFRES_SHORT);
1244 	if (r != NULL && r->log)
1245 		PFLOG_PACKET(kif, h, m, AF_INET6, dir, *reason, r, NULL, NULL, pd);
1246 	return (PF_DROP);
1247 
1248  drop:
1249 	REASON_SET(reason, PFRES_NORM);
1250 	if (r != NULL && r->log)
1251 		PFLOG_PACKET(kif, h, m, AF_INET6, dir, *reason, r, NULL, NULL, pd);
1252 	return (PF_DROP);
1253 
1254  badfrag:
1255 	REASON_SET(reason, PFRES_FRAG);
1256 	if (r != NULL && r->log)
1257 		PFLOG_PACKET(kif, h, m, AF_INET6, dir, *reason, r, NULL, NULL, pd);
1258 	return (PF_DROP);
1259 }
1260 #endif /* INET6 */
1261 
1262 int
1263 pf_normalize_tcp(int dir, struct pfi_kif *kif, struct mbuf *m,
1264     int ipoff, int off, void *h, struct pf_pdesc *pd)
1265 {
1266 	struct pf_rule	*r, *rm = NULL;
1267 	struct tcphdr	*th = pd->hdr.tcp;
1268 	int		 rewrite = 0;
1269 	u_short		 reason;
1270 	u_int8_t	 flags;
1271 	sa_family_t	 af = pd->af;
1272 
1273 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1274 	while (r != NULL) {
1275 		r->evaluations++;
1276 		if (pfi_kif_match(r->kif, kif) == r->ifnot)
1277 			r = r->skip[PF_SKIP_IFP].ptr;
1278 		else if (r->direction && r->direction != dir)
1279 			r = r->skip[PF_SKIP_DIR].ptr;
1280 		else if (r->af && r->af != af)
1281 			r = r->skip[PF_SKIP_AF].ptr;
1282 		else if (r->proto && r->proto != pd->proto)
1283 			r = r->skip[PF_SKIP_PROTO].ptr;
1284 		else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
1285 		    r->src.neg, kif))
1286 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
1287 		else if (r->src.port_op && !pf_match_port(r->src.port_op,
1288 			    r->src.port[0], r->src.port[1], th->th_sport))
1289 			r = r->skip[PF_SKIP_SRC_PORT].ptr;
1290 		else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
1291 		    r->dst.neg, NULL))
1292 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
1293 		else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
1294 			    r->dst.port[0], r->dst.port[1], th->th_dport))
1295 			r = r->skip[PF_SKIP_DST_PORT].ptr;
1296 		else if (r->os_fingerprint != PF_OSFP_ANY && !pf_osfp_match(
1297 			    pf_osfp_fingerprint(pd, m, off, th),
1298 			    r->os_fingerprint))
1299 			r = TAILQ_NEXT(r, entries);
1300 		else {
1301 			rm = r;
1302 			break;
1303 		}
1304 	}
1305 
1306 	if (rm == NULL || rm->action == PF_NOSCRUB)
1307 		return (PF_PASS);
1308 	else {
1309 		r->packets[dir == PF_OUT]++;
1310 		r->bytes[dir == PF_OUT] += pd->tot_len;
1311 	}
1312 
1313 	if (rm->rule_flag & PFRULE_REASSEMBLE_TCP)
1314 		pd->flags |= PFDESC_TCP_NORM;
1315 
1316 	flags = th->th_flags;
1317 	if (flags & TH_SYN) {
1318 		/* Illegal packet */
1319 		if (flags & TH_RST)
1320 			goto tcp_drop;
1321 
1322 		if (flags & TH_FIN)
1323 			flags &= ~TH_FIN;
1324 	} else {
1325 		/* Illegal packet */
1326 		if (!(flags & (TH_ACK|TH_RST)))
1327 			goto tcp_drop;
1328 	}
1329 
1330 	if (!(flags & TH_ACK)) {
1331 		/* These flags are only valid if ACK is set */
1332 		if ((flags & TH_FIN) || (flags & TH_PUSH) || (flags & TH_URG))
1333 			goto tcp_drop;
1334 	}
1335 
1336 	/* Check for illegal header length */
1337 	if (th->th_off < (sizeof(struct tcphdr) >> 2))
1338 		goto tcp_drop;
1339 
1340 	/* If flags changed, or reserved data set, then adjust */
1341 	if (flags != th->th_flags || th->th_x2 != 0) {
1342 		u_int16_t	ov, nv;
1343 
1344 		ov = *(u_int16_t *)(&th->th_ack + 1);
1345 		th->th_flags = flags;
1346 		th->th_x2 = 0;
1347 		nv = *(u_int16_t *)(&th->th_ack + 1);
1348 
1349 		th->th_sum = pf_cksum_fixup(th->th_sum, ov, nv, 0);
1350 		rewrite = 1;
1351 	}
1352 
1353 	/* Remove urgent pointer, if TH_URG is not set */
1354 	if (!(flags & TH_URG) && th->th_urp) {
1355 		th->th_sum = pf_cksum_fixup(th->th_sum, th->th_urp, 0, 0);
1356 		th->th_urp = 0;
1357 		rewrite = 1;
1358 	}
1359 
1360 	/* Process options */
1361 	if (r->max_mss && pf_normalize_tcpopt(r, m, th, off))
1362 		rewrite = 1;
1363 
1364 	/* copy back packet headers if we sanitized */
1365 	if (rewrite)
1366 		m_copyback(m, off, sizeof(*th), th);
1367 
1368 	return (PF_PASS);
1369 
1370  tcp_drop:
1371 	REASON_SET(&reason, PFRES_NORM);
1372 	if (rm != NULL && r->log)
1373 		PFLOG_PACKET(kif, h, m, AF_INET, dir, reason, r, NULL, NULL, pd);
1374 	return (PF_DROP);
1375 }
1376 
1377 int
1378 pf_normalize_tcp_init(struct mbuf *m, int off, struct pf_pdesc *pd,
1379     struct tcphdr *th, struct pf_state_peer *src,
1380     struct pf_state_peer *dst)
1381 {
1382 	u_int32_t tsval, tsecr;
1383 	u_int8_t hdr[60];
1384 	u_int8_t *opt;
1385 
1386 	KASSERT(src->scrub == NULL);
1387 
1388 	src->scrub = pool_get(&pf_state_scrub_pl, PR_NOWAIT);
1389 	if (src->scrub == NULL)
1390 		return (1);
1391 	bzero(src->scrub, sizeof(*src->scrub));
1392 
1393 	switch (pd->af) {
1394 #ifdef INET
1395 	case AF_INET: {
1396 		struct ip *h = mtod(m, struct ip *);
1397 		src->scrub->pfss_ttl = h->ip_ttl;
1398 		break;
1399 	}
1400 #endif /* INET */
1401 #ifdef INET6
1402 	case AF_INET6: {
1403 		struct ip6_hdr *h = mtod(m, struct ip6_hdr *);
1404 		src->scrub->pfss_ttl = h->ip6_hlim;
1405 		break;
1406 	}
1407 #endif /* INET6 */
1408 	}
1409 
1410 
1411 	/*
1412 	 * All normalizations below are only begun if we see the start of
1413 	 * the connections.  They must all set an enabled bit in pfss_flags
1414 	 */
1415 	if ((th->th_flags & TH_SYN) == 0)
1416 		return (0);
1417 
1418 
1419 	if (th->th_off > (sizeof(struct tcphdr) >> 2) && src->scrub &&
1420 	    pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) {
1421 		/* Diddle with TCP options */
1422 		int hlen;
1423 		opt = hdr + sizeof(struct tcphdr);
1424 		hlen = (th->th_off << 2) - sizeof(struct tcphdr);
1425 		while (hlen >= TCPOLEN_TIMESTAMP) {
1426 			switch (*opt) {
1427 			case TCPOPT_EOL:	/* FALLTHROUGH */
1428 			case TCPOPT_NOP:
1429 				opt++;
1430 				hlen--;
1431 				break;
1432 			case TCPOPT_TIMESTAMP:
1433 				if (opt[1] >= TCPOLEN_TIMESTAMP) {
1434 					src->scrub->pfss_flags |=
1435 					    PFSS_TIMESTAMP;
1436 					src->scrub->pfss_ts_mod =
1437 					    htonl(arc4random());
1438 
1439 					/* note PFSS_PAWS not set yet */
1440 					memcpy(&tsval, &opt[2],
1441 					    sizeof(u_int32_t));
1442 					memcpy(&tsecr, &opt[6],
1443 					    sizeof(u_int32_t));
1444 					src->scrub->pfss_tsval0 = ntohl(tsval);
1445 					src->scrub->pfss_tsval = ntohl(tsval);
1446 					src->scrub->pfss_tsecr = ntohl(tsecr);
1447 					getmicrouptime(&src->scrub->pfss_last);
1448 				}
1449 				/* FALLTHROUGH */
1450 			default:
1451 				hlen -= MAX(opt[1], 2);
1452 				opt += MAX(opt[1], 2);
1453 				break;
1454 			}
1455 		}
1456 	}
1457 
1458 	return (0);
1459 }
1460 
1461 void
1462 pf_normalize_tcp_cleanup(struct pf_state *state)
1463 {
1464 	if (state->src.scrub)
1465 		pool_put(&pf_state_scrub_pl, state->src.scrub);
1466 	if (state->dst.scrub)
1467 		pool_put(&pf_state_scrub_pl, state->dst.scrub);
1468 
1469 	/* Someday... flush the TCP segment reassembly descriptors. */
1470 }
1471 
1472 int
1473 pf_normalize_tcp_stateful(struct mbuf *m, int off, struct pf_pdesc *pd,
1474     u_short *reason, struct tcphdr *th, struct pf_state *state,
1475     struct pf_state_peer *src, struct pf_state_peer *dst, int *writeback)
1476 {
1477 	struct timeval uptime;
1478 	u_int32_t tsval, tsecr;
1479 	u_int tsval_from_last;
1480 	u_int8_t hdr[60];
1481 	u_int8_t *opt;
1482 	int copyback = 0;
1483 	int got_ts = 0;
1484 
1485 	KASSERT(src->scrub || dst->scrub);
1486 
1487 	/*
1488 	 * Enforce the minimum TTL seen for this connection.  Negate a common
1489 	 * technique to evade an intrusion detection system and confuse
1490 	 * firewall state code.
1491 	 */
1492 	switch (pd->af) {
1493 #ifdef INET
1494 	case AF_INET: {
1495 		if (src->scrub) {
1496 			struct ip *h = mtod(m, struct ip *);
1497 			if (h->ip_ttl > src->scrub->pfss_ttl)
1498 				src->scrub->pfss_ttl = h->ip_ttl;
1499 			h->ip_ttl = src->scrub->pfss_ttl;
1500 		}
1501 		break;
1502 	}
1503 #endif /* INET */
1504 #ifdef INET6
1505 	case AF_INET6: {
1506 		if (src->scrub) {
1507 			struct ip6_hdr *h = mtod(m, struct ip6_hdr *);
1508 			if (h->ip6_hlim > src->scrub->pfss_ttl)
1509 				src->scrub->pfss_ttl = h->ip6_hlim;
1510 			h->ip6_hlim = src->scrub->pfss_ttl;
1511 		}
1512 		break;
1513 	}
1514 #endif /* INET6 */
1515 	}
1516 
1517 	if (th->th_off > (sizeof(struct tcphdr) >> 2) &&
1518 	    ((src->scrub && (src->scrub->pfss_flags & PFSS_TIMESTAMP)) ||
1519 	    (dst->scrub && (dst->scrub->pfss_flags & PFSS_TIMESTAMP))) &&
1520 	    pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) {
1521 		/* Diddle with TCP options */
1522 		int hlen;
1523 		opt = hdr + sizeof(struct tcphdr);
1524 		hlen = (th->th_off << 2) - sizeof(struct tcphdr);
1525 		while (hlen >= TCPOLEN_TIMESTAMP) {
1526 			switch (*opt) {
1527 			case TCPOPT_EOL:	/* FALLTHROUGH */
1528 			case TCPOPT_NOP:
1529 				opt++;
1530 				hlen--;
1531 				break;
1532 			case TCPOPT_TIMESTAMP:
1533 				/* Modulate the timestamps.  Can be used for
1534 				 * NAT detection, OS uptime determination or
1535 				 * reboot detection.
1536 				 */
1537 
1538 				if (got_ts) {
1539 					/* Huh?  Multiple timestamps!? */
1540 					if (pf_status.debug >= PF_DEBUG_MISC) {
1541 						DPFPRINTF(("multiple TS??"));
1542 						pf_print_state(state);
1543 						printf("\n");
1544 					}
1545 					REASON_SET(reason, PFRES_TS);
1546 					return (PF_DROP);
1547 				}
1548 				if (opt[1] >= TCPOLEN_TIMESTAMP) {
1549 					memcpy(&tsval, &opt[2],
1550 					    sizeof(u_int32_t));
1551 					if (tsval && src->scrub &&
1552 					    (src->scrub->pfss_flags &
1553 					    PFSS_TIMESTAMP)) {
1554 						tsval = ntohl(tsval);
1555 						pf_change_a(&opt[2],
1556 						    &th->th_sum,
1557 						    htonl(tsval +
1558 						    src->scrub->pfss_ts_mod),
1559 						    0);
1560 						copyback = 1;
1561 					}
1562 
1563 					/* Modulate TS reply iff valid (!0) */
1564 					memcpy(&tsecr, &opt[6],
1565 					    sizeof(u_int32_t));
1566 					if (tsecr && dst->scrub &&
1567 					    (dst->scrub->pfss_flags &
1568 					    PFSS_TIMESTAMP)) {
1569 						tsecr = ntohl(tsecr)
1570 						    - dst->scrub->pfss_ts_mod;
1571 						pf_change_a(&opt[6],
1572 						    &th->th_sum, htonl(tsecr),
1573 						    0);
1574 						copyback = 1;
1575 					}
1576 					got_ts = 1;
1577 				}
1578 				/* FALLTHROUGH */
1579 			default:
1580 				hlen -= MAX(opt[1], 2);
1581 				opt += MAX(opt[1], 2);
1582 				break;
1583 			}
1584 		}
1585 		if (copyback) {
1586 			/* Copyback the options, caller copys back header */
1587 			*writeback = 1;
1588 			m_copyback(m, off + sizeof(struct tcphdr),
1589 			    (th->th_off << 2) - sizeof(struct tcphdr), hdr +
1590 			    sizeof(struct tcphdr));
1591 		}
1592 	}
1593 
1594 
1595 	/*
1596 	 * Must invalidate PAWS checks on connections idle for too long.
1597 	 * The fastest allowed timestamp clock is 1ms.  That turns out to
1598 	 * be about 24 days before it wraps.  XXX Right now our lowerbound
1599 	 * TS echo check only works for the first 12 days of a connection
1600 	 * when the TS has exhausted half its 32bit space
1601 	 */
1602 #define TS_MAX_IDLE	(24*24*60*60)
1603 #define TS_MAX_CONN	(12*24*60*60)	/* XXX remove when better tsecr check */
1604 
1605 	getmicrouptime(&uptime);
1606 	if (src->scrub && (src->scrub->pfss_flags & PFSS_PAWS) &&
1607 	    (uptime.tv_sec - src->scrub->pfss_last.tv_sec > TS_MAX_IDLE ||
1608 	    time_second - state->creation > TS_MAX_CONN))  {
1609 		if (pf_status.debug >= PF_DEBUG_MISC) {
1610 			DPFPRINTF(("src idled out of PAWS\n"));
1611 			pf_print_state(state);
1612 			printf("\n");
1613 		}
1614 		src->scrub->pfss_flags = (src->scrub->pfss_flags & ~PFSS_PAWS)
1615 		    | PFSS_PAWS_IDLED;
1616 	}
1617 	if (dst->scrub && (dst->scrub->pfss_flags & PFSS_PAWS) &&
1618 	    uptime.tv_sec - dst->scrub->pfss_last.tv_sec > TS_MAX_IDLE) {
1619 		if (pf_status.debug >= PF_DEBUG_MISC) {
1620 			DPFPRINTF(("dst idled out of PAWS\n"));
1621 			pf_print_state(state);
1622 			printf("\n");
1623 		}
1624 		dst->scrub->pfss_flags = (dst->scrub->pfss_flags & ~PFSS_PAWS)
1625 		    | PFSS_PAWS_IDLED;
1626 	}
1627 
1628 	if (got_ts && src->scrub && dst->scrub &&
1629 	    (src->scrub->pfss_flags & PFSS_PAWS) &&
1630 	    (dst->scrub->pfss_flags & PFSS_PAWS)) {
1631 		/* Validate that the timestamps are "in-window".
1632 		 * RFC1323 describes TCP Timestamp options that allow
1633 		 * measurement of RTT (round trip time) and PAWS
1634 		 * (protection against wrapped sequence numbers).  PAWS
1635 		 * gives us a set of rules for rejecting packets on
1636 		 * long fat pipes (packets that were somehow delayed
1637 		 * in transit longer than the time it took to send the
1638 		 * full TCP sequence space of 4Gb).  We can use these
1639 		 * rules and infer a few others that will let us treat
1640 		 * the 32bit timestamp and the 32bit echoed timestamp
1641 		 * as sequence numbers to prevent a blind attacker from
1642 		 * inserting packets into a connection.
1643 		 *
1644 		 * RFC1323 tells us:
1645 		 *  - The timestamp on this packet must be greater than
1646 		 *    or equal to the last value echoed by the other
1647 		 *    endpoint.  The RFC says those will be discarded
1648 		 *    since it is a dup that has already been acked.
1649 		 *    This gives us a lowerbound on the timestamp.
1650 		 *        timestamp >= other last echoed timestamp
1651 		 *  - The timestamp will be less than or equal to
1652 		 *    the last timestamp plus the time between the
1653 		 *    last packet and now.  The RFC defines the max
1654 		 *    clock rate as 1ms.  We will allow clocks to be
1655 		 *    up to 10% fast and will allow a total difference
1656 		 *    or 30 seconds due to a route change.  And this
1657 		 *    gives us an upperbound on the timestamp.
1658 		 *        timestamp <= last timestamp + max ticks
1659 		 *    We have to be careful here.  Windows will send an
1660 		 *    initial timestamp of zero and then initialize it
1661 		 *    to a random value after the 3whs; presumably to
1662 		 *    avoid a DoS by having to call an expensive RNG
1663 		 *    during a SYN flood.  Proof MS has at least one
1664 		 *    good security geek.
1665 		 *
1666 		 *  - The TCP timestamp option must also echo the other
1667 		 *    endpoints timestamp.  The timestamp echoed is the
1668 		 *    one carried on the earliest unacknowledged segment
1669 		 *    on the left edge of the sequence window.  The RFC
1670 		 *    states that the host will reject any echoed
1671 		 *    timestamps that were larger than any ever sent.
1672 		 *    This gives us an upperbound on the TS echo.
1673 		 *        tescr <= largest_tsval
1674 		 *  - The lowerbound on the TS echo is a little more
1675 		 *    tricky to determine.  The other endpoint's echoed
1676 		 *    values will not decrease.  But there may be
1677 		 *    network conditions that re-order packets and
1678 		 *    cause our view of them to decrease.  For now the
1679 		 *    only lowerbound we can safely determine is that
1680 		 *    the TS echo will never be less than the original
1681 		 *    TS.  XXX There is probably a better lowerbound.
1682 		 *    Remove TS_MAX_CONN with better lowerbound check.
1683 		 *        tescr >= other original TS
1684 		 *
1685 		 * It is also important to note that the fastest
1686 		 * timestamp clock of 1ms will wrap its 32bit space in
1687 		 * 24 days.  So we just disable TS checking after 24
1688 		 * days of idle time.  We actually must use a 12d
1689 		 * connection limit until we can come up with a better
1690 		 * lowerbound to the TS echo check.
1691 		 */
1692 		struct timeval delta_ts;
1693 		int ts_fudge;
1694 
1695 
1696 		/*
1697 		 * PFTM_TS_DIFF is how many seconds of leeway to allow
1698 		 * a host's timestamp.  This can happen if the previous
1699 		 * packet got delayed in transit for much longer than
1700 		 * this packet.
1701 		 */
1702 		if ((ts_fudge = state->rule.ptr->timeout[PFTM_TS_DIFF]) == 0)
1703 			ts_fudge = pf_default_rule.timeout[PFTM_TS_DIFF];
1704 
1705 
1706 		/* Calculate max ticks since the last timestamp */
1707 #define TS_MAXFREQ	1100		/* RFC max TS freq of 1 kHz + 10% skew */
1708 #define TS_MICROSECS	1000000		/* microseconds per second */
1709 		timersub(&uptime, &src->scrub->pfss_last, &delta_ts);
1710 		tsval_from_last = (delta_ts.tv_sec + ts_fudge) * TS_MAXFREQ;
1711 		tsval_from_last += delta_ts.tv_usec / (TS_MICROSECS/TS_MAXFREQ);
1712 
1713 
1714 		if ((src->state >= TCPS_ESTABLISHED &&
1715 		    dst->state >= TCPS_ESTABLISHED) &&
1716 		    (SEQ_LT(tsval, dst->scrub->pfss_tsecr) ||
1717 		    SEQ_GT(tsval, src->scrub->pfss_tsval + tsval_from_last) ||
1718 		    (tsecr && (SEQ_GT(tsecr, dst->scrub->pfss_tsval) ||
1719 		    SEQ_LT(tsecr, dst->scrub->pfss_tsval0))))) {
1720 			/* Bad RFC1323 implementation or an insertion attack.
1721 			 *
1722 			 * - Solaris 2.6 and 2.7 are known to send another ACK
1723 			 *   after the FIN,FIN|ACK,ACK closing that carries
1724 			 *   an old timestamp.
1725 			 */
1726 
1727 			DPFPRINTF(("Timestamp failed %c%c%c%c\n",
1728 			    SEQ_LT(tsval, dst->scrub->pfss_tsecr) ? '0' : ' ',
1729 			    SEQ_GT(tsval, src->scrub->pfss_tsval +
1730 			    tsval_from_last) ? '1' : ' ',
1731 			    SEQ_GT(tsecr, dst->scrub->pfss_tsval) ? '2' : ' ',
1732 			    SEQ_LT(tsecr, dst->scrub->pfss_tsval0)? '3' : ' '));
1733 			DPFPRINTF((" tsval: %" PRIu32 "  tsecr: %" PRIu32
1734 			    "  +ticks: %" PRIu32 "  idle: %"PRIx64"s %ums\n",
1735 			    tsval, tsecr, tsval_from_last, delta_ts.tv_sec,
1736 			    delta_ts.tv_usec / 1000U));
1737 			DPFPRINTF((" src->tsval: %" PRIu32 "  tsecr: %" PRIu32
1738 			    "\n",
1739 			    src->scrub->pfss_tsval, src->scrub->pfss_tsecr));
1740 			DPFPRINTF((" dst->tsval: %" PRIu32 "  tsecr: %" PRIu32
1741 			    "  tsval0: %" PRIu32 "\n",
1742 			    dst->scrub->pfss_tsval,
1743 			    dst->scrub->pfss_tsecr, dst->scrub->pfss_tsval0));
1744 			if (pf_status.debug >= PF_DEBUG_MISC) {
1745 				pf_print_state(state);
1746 				pf_print_flags(th->th_flags);
1747 				printf("\n");
1748 			}
1749 			REASON_SET(reason, PFRES_TS);
1750 			return (PF_DROP);
1751 		}
1752 
1753 		/* XXX I'd really like to require tsecr but it's optional */
1754 
1755 	} else if (!got_ts && (th->th_flags & TH_RST) == 0 &&
1756 	    ((src->state == TCPS_ESTABLISHED && dst->state == TCPS_ESTABLISHED)
1757 	    || pd->p_len > 0 || (th->th_flags & TH_SYN)) &&
1758 	    src->scrub && dst->scrub &&
1759 	    (src->scrub->pfss_flags & PFSS_PAWS) &&
1760 	    (dst->scrub->pfss_flags & PFSS_PAWS)) {
1761 		/* Didn't send a timestamp.  Timestamps aren't really useful
1762 		 * when:
1763 		 *  - connection opening or closing (often not even sent).
1764 		 *    but we must not let an attacker to put a FIN on a
1765 		 *    data packet to sneak it through our ESTABLISHED check.
1766 		 *  - on a TCP reset.  RFC suggests not even looking at TS.
1767 		 *  - on an empty ACK.  The TS will not be echoed so it will
1768 		 *    probably not help keep the RTT calculation in sync and
1769 		 *    there isn't as much danger when the sequence numbers
1770 		 *    got wrapped.  So some stacks don't include TS on empty
1771 		 *    ACKs :-(
1772 		 *
1773 		 * To minimize the disruption to mostly RFC1323 conformant
1774 		 * stacks, we will only require timestamps on data packets.
1775 		 *
1776 		 * And what do ya know, we cannot require timestamps on data
1777 		 * packets.  There appear to be devices that do legitimate
1778 		 * TCP connection hijacking.  There are HTTP devices that allow
1779 		 * a 3whs (with timestamps) and then buffer the HTTP request.
1780 		 * If the intermediate device has the HTTP response cache, it
1781 		 * will spoof the response but not bother timestamping its
1782 		 * packets.  So we can look for the presence of a timestamp in
1783 		 * the first data packet and if there, require it in all future
1784 		 * packets.
1785 		 */
1786 
1787 		if (pd->p_len > 0 && (src->scrub->pfss_flags & PFSS_DATA_TS)) {
1788 			/*
1789 			 * Hey!  Someone tried to sneak a packet in.  Or the
1790 			 * stack changed its RFC1323 behavior?!?!
1791 			 */
1792 			if (pf_status.debug >= PF_DEBUG_MISC) {
1793 				DPFPRINTF(("Did not receive expected RFC1323 "
1794 				    "timestamp\n"));
1795 				pf_print_state(state);
1796 				pf_print_flags(th->th_flags);
1797 				printf("\n");
1798 			}
1799 			REASON_SET(reason, PFRES_TS);
1800 			return (PF_DROP);
1801 		}
1802 	}
1803 
1804 
1805 	/*
1806 	 * We will note if a host sends his data packets with or without
1807 	 * timestamps.  And require all data packets to contain a timestamp
1808 	 * if the first does.  PAWS implicitly requires that all data packets be
1809 	 * timestamped.  But I think there are middle-man devices that hijack
1810 	 * TCP streams immediately after the 3whs and don't timestamp their
1811 	 * packets (seen in a WWW accelerator or cache).
1812 	 */
1813 	if (pd->p_len > 0 && src->scrub && (src->scrub->pfss_flags &
1814 	    (PFSS_TIMESTAMP|PFSS_DATA_TS|PFSS_DATA_NOTS)) == PFSS_TIMESTAMP) {
1815 		if (got_ts)
1816 			src->scrub->pfss_flags |= PFSS_DATA_TS;
1817 		else {
1818 			src->scrub->pfss_flags |= PFSS_DATA_NOTS;
1819 			if (pf_status.debug >= PF_DEBUG_MISC && dst->scrub &&
1820 			    (dst->scrub->pfss_flags & PFSS_TIMESTAMP)) {
1821 				/* Don't warn if other host rejected RFC1323 */
1822 				DPFPRINTF(("Broken RFC1323 stack did not "
1823 				    "timestamp data packet. Disabled PAWS "
1824 				    "security.\n"));
1825 				pf_print_state(state);
1826 				pf_print_flags(th->th_flags);
1827 				printf("\n");
1828 			}
1829 		}
1830 	}
1831 
1832 
1833 	/*
1834 	 * Update PAWS values
1835 	 */
1836 	if (got_ts && src->scrub && PFSS_TIMESTAMP == (src->scrub->pfss_flags &
1837 	    (PFSS_PAWS_IDLED|PFSS_TIMESTAMP))) {
1838 		getmicrouptime(&src->scrub->pfss_last);
1839 		if (SEQ_GEQ(tsval, src->scrub->pfss_tsval) ||
1840 		    (src->scrub->pfss_flags & PFSS_PAWS) == 0)
1841 			src->scrub->pfss_tsval = tsval;
1842 
1843 		if (tsecr) {
1844 			if (SEQ_GEQ(tsecr, src->scrub->pfss_tsecr) ||
1845 			    (src->scrub->pfss_flags & PFSS_PAWS) == 0)
1846 				src->scrub->pfss_tsecr = tsecr;
1847 
1848 			if ((src->scrub->pfss_flags & PFSS_PAWS) == 0 &&
1849 			    (SEQ_LT(tsval, src->scrub->pfss_tsval0) ||
1850 			    src->scrub->pfss_tsval0 == 0)) {
1851 				/* tsval0 MUST be the lowest timestamp */
1852 				src->scrub->pfss_tsval0 = tsval;
1853 			}
1854 
1855 			/* Only fully initialized after a TS gets echoed */
1856 			if ((src->scrub->pfss_flags & PFSS_PAWS) == 0)
1857 				src->scrub->pfss_flags |= PFSS_PAWS;
1858 		}
1859 	}
1860 
1861 	/* I have a dream....  TCP segment reassembly.... */
1862 	return (0);
1863 }
1864 
1865 int
1866 pf_normalize_tcpopt(struct pf_rule *r, struct mbuf *m, struct tcphdr *th,
1867     int off)
1868 {
1869 	u_int16_t	*mss;
1870 	int		 thoff;
1871 	int		 opt, cnt, optlen = 0;
1872 	int		 rewrite = 0;
1873 	u_char		*optp;
1874 
1875 	thoff = th->th_off << 2;
1876 	cnt = thoff - sizeof(struct tcphdr);
1877 	optp = mtod(m, u_char *) + off + sizeof(struct tcphdr);
1878 
1879 	for (; cnt > 0; cnt -= optlen, optp += optlen) {
1880 		opt = optp[0];
1881 		if (opt == TCPOPT_EOL)
1882 			break;
1883 		if (opt == TCPOPT_NOP)
1884 			optlen = 1;
1885 		else {
1886 			if (cnt < 2)
1887 				break;
1888 			optlen = optp[1];
1889 			if (optlen < 2 || optlen > cnt)
1890 				break;
1891 		}
1892 		switch (opt) {
1893 		case TCPOPT_MAXSEG:
1894 			mss = (u_int16_t *)(optp + 2);
1895 			if ((ntohs(*mss)) > r->max_mss) {
1896 				th->th_sum = pf_cksum_fixup(th->th_sum,
1897 				    *mss, htons(r->max_mss), 0);
1898 				*mss = htons(r->max_mss);
1899 				rewrite = 1;
1900 			}
1901 			break;
1902 		default:
1903 			break;
1904 		}
1905 	}
1906 
1907 	return (rewrite);
1908 }
1909