xref: /netbsd-src/sys/dist/pf/net/pf_norm.c (revision 7788a0781fe6ff2cce37368b4578a7ade0850cb1)
1 /*	$NetBSD: pf_norm.c,v 1.26 2011/11/28 08:05:05 tls 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 1.26 2011/11/28 08:05:05 tls 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 #include <sys/cprng.h>
51 #else
52 #include <dev/rndvar.h>
53 #endif /* !__NetBSD__ */
54 #include <net/if.h>
55 #include <net/if_types.h>
56 #include <net/bpf.h>
57 #include <net/route.h>
58 #include <net/if_pflog.h>
59 
60 #include <netinet/in.h>
61 #include <netinet/in_var.h>
62 #include <netinet/in_systm.h>
63 #include <netinet/ip.h>
64 #include <netinet/ip_var.h>
65 #include <netinet/tcp.h>
66 #include <netinet/tcp_seq.h>
67 #include <netinet/udp.h>
68 #include <netinet/ip_icmp.h>
69 
70 #ifdef INET6
71 #include <netinet/ip6.h>
72 #endif /* INET6 */
73 
74 #include <net/pfvar.h>
75 
76 struct pf_frent {
77 	LIST_ENTRY(pf_frent) fr_next;
78 	struct ip *fr_ip;
79 	struct mbuf *fr_m;
80 };
81 
82 struct pf_frcache {
83 	LIST_ENTRY(pf_frcache) fr_next;
84 	uint16_t	fr_off;
85 	uint16_t	fr_end;
86 };
87 
88 #define PFFRAG_SEENLAST	0x0001		/* Seen the last fragment for this */
89 #define PFFRAG_NOBUFFER	0x0002		/* Non-buffering fragment cache */
90 #define PFFRAG_DROP	0x0004		/* Drop all fragments */
91 #define BUFFER_FRAGMENTS(fr)	(!((fr)->fr_flags & PFFRAG_NOBUFFER))
92 
93 struct pf_fragment {
94 	RB_ENTRY(pf_fragment) fr_entry;
95 	TAILQ_ENTRY(pf_fragment) frag_next;
96 	struct in_addr	fr_src;
97 	struct in_addr	fr_dst;
98 	u_int8_t	fr_p;		/* protocol of this fragment */
99 	u_int8_t	fr_flags;	/* status flags */
100 	u_int16_t	fr_id;		/* fragment id for reassemble */
101 	u_int16_t	fr_max;		/* fragment data max */
102 	u_int32_t	fr_timeout;
103 #define fr_queue	fr_u.fru_queue
104 #define fr_cache	fr_u.fru_cache
105 	union {
106 		LIST_HEAD(pf_fragq, pf_frent) fru_queue;	/* buffering */
107 		LIST_HEAD(pf_cacheq, pf_frcache) fru_cache;	/* non-buf */
108 	} fr_u;
109 };
110 
111 TAILQ_HEAD(pf_fragqueue, pf_fragment)	pf_fragqueue;
112 TAILQ_HEAD(pf_cachequeue, pf_fragment)	pf_cachequeue;
113 
114 static __inline int	 pf_frag_compare(struct pf_fragment *,
115 			    struct pf_fragment *);
116 RB_HEAD(pf_frag_tree, pf_fragment)	pf_frag_tree, pf_cache_tree;
117 RB_PROTOTYPE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
118 RB_GENERATE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
119 
120 /* Private prototypes */
121 void			 pf_ip2key(struct pf_fragment *, struct ip *);
122 void			 pf_remove_fragment(struct pf_fragment *);
123 void			 pf_flush_fragments(void);
124 void			 pf_free_fragment(struct pf_fragment *);
125 struct pf_fragment	*pf_find_fragment(struct ip *, struct pf_frag_tree *);
126 struct mbuf		*pf_reassemble(struct mbuf **, struct pf_fragment **,
127 			    struct pf_frent *, int);
128 struct mbuf		*pf_fragcache(struct mbuf **, struct ip*,
129 			    struct pf_fragment **, int, int, int *);
130 int			 pf_normalize_tcpopt(struct pf_rule *, struct mbuf *,
131 			    struct tcphdr *, int);
132 
133 #define	DPFPRINTF(x) do {				\
134 	if (pf_status.debug >= PF_DEBUG_MISC) {		\
135 		printf("%s: ", __func__);		\
136 		printf x ;				\
137 	}						\
138 } while(0)
139 
140 /* Globals */
141 struct pool		 pf_frent_pl, pf_frag_pl, pf_cache_pl, pf_cent_pl;
142 struct pool		 pf_state_scrub_pl;
143 int			 pf_nfrents, pf_ncache;
144 
145 void
146 pf_normalize_init(void)
147 {
148 #ifdef __NetBSD__
149 	pool_init(&pf_frent_pl, sizeof(struct pf_frent), 0, 0, 0, "pffrent",
150 	    NULL, IPL_SOFTNET);
151 	pool_init(&pf_frag_pl, sizeof(struct pf_fragment), 0, 0, 0, "pffrag",
152 	    NULL, IPL_SOFTNET);
153 	pool_init(&pf_cache_pl, sizeof(struct pf_fragment), 0, 0, 0,
154 	    "pffrcache", NULL, IPL_SOFTNET);
155 	pool_init(&pf_cent_pl, sizeof(struct pf_frcache), 0, 0, 0, "pffrcent",
156 	    NULL, IPL_SOFTNET);
157 	pool_init(&pf_state_scrub_pl, sizeof(struct pf_state_scrub), 0, 0, 0,
158 	    "pfstscr", NULL, IPL_SOFTNET);
159 #else
160 	pool_init(&pf_frent_pl, sizeof(struct pf_frent), 0, 0, 0, "pffrent",
161 	    NULL);
162 	pool_init(&pf_frag_pl, sizeof(struct pf_fragment), 0, 0, 0, "pffrag",
163 	    NULL);
164 	pool_init(&pf_cache_pl, sizeof(struct pf_fragment), 0, 0, 0,
165 	    "pffrcache", NULL);
166 	pool_init(&pf_cent_pl, sizeof(struct pf_frcache), 0, 0, 0, "pffrcent",
167 	    NULL);
168 	pool_init(&pf_state_scrub_pl, sizeof(struct pf_state_scrub), 0, 0, 0,
169 	    "pfstscr", NULL);
170 #endif /* !__NetBSD__ */
171 
172 	pool_sethiwat(&pf_frag_pl, PFFRAG_FRAG_HIWAT);
173 	pool_sethardlimit(&pf_frent_pl, PFFRAG_FRENT_HIWAT, NULL, 0);
174 	pool_sethardlimit(&pf_cache_pl, PFFRAG_FRCACHE_HIWAT, NULL, 0);
175 	pool_sethardlimit(&pf_cent_pl, PFFRAG_FRCENT_HIWAT, NULL, 0);
176 
177 	TAILQ_INIT(&pf_fragqueue);
178 	TAILQ_INIT(&pf_cachequeue);
179 }
180 
181 #ifdef _MODULE
182 void
183 pf_normalize_destroy(void)
184 {
185 	pool_destroy(&pf_state_scrub_pl);
186 	pool_destroy(&pf_cent_pl);
187 	pool_destroy(&pf_cache_pl);
188 	pool_destroy(&pf_frag_pl);
189 	pool_destroy(&pf_frent_pl);
190 }
191 #endif /* _MODULE */
192 
193 static __inline int
194 pf_frag_compare(struct pf_fragment *a, struct pf_fragment *b)
195 {
196 	int	diff;
197 
198 	if ((diff = a->fr_id - b->fr_id))
199 		return (diff);
200 	else if ((diff = a->fr_p - b->fr_p))
201 		return (diff);
202 	else if (a->fr_src.s_addr < b->fr_src.s_addr)
203 		return (-1);
204 	else if (a->fr_src.s_addr > b->fr_src.s_addr)
205 		return (1);
206 	else if (a->fr_dst.s_addr < b->fr_dst.s_addr)
207 		return (-1);
208 	else if (a->fr_dst.s_addr > b->fr_dst.s_addr)
209 		return (1);
210 	return (0);
211 }
212 
213 void
214 pf_purge_expired_fragments(void)
215 {
216 	struct pf_fragment	*frag;
217 	u_int32_t		 expire = time_second -
218 				    pf_default_rule.timeout[PFTM_FRAG];
219 
220 	while ((frag = TAILQ_LAST(&pf_fragqueue, pf_fragqueue)) != NULL) {
221 		KASSERT(BUFFER_FRAGMENTS(frag));
222 		if (frag->fr_timeout > expire)
223 			break;
224 
225 		DPFPRINTF(("expiring %d(%p)\n", frag->fr_id, frag));
226 		pf_free_fragment(frag);
227 	}
228 
229 	while ((frag = TAILQ_LAST(&pf_cachequeue, pf_cachequeue)) != NULL) {
230 		KASSERT(!BUFFER_FRAGMENTS(frag));
231 		if (frag->fr_timeout > expire)
232 			break;
233 
234 		DPFPRINTF(("expiring %d(%p)\n", frag->fr_id, frag));
235 		pf_free_fragment(frag);
236 		KASSERT(TAILQ_EMPTY(&pf_cachequeue) ||
237 		    TAILQ_LAST(&pf_cachequeue, pf_cachequeue) != frag);
238 	}
239 }
240 
241 /*
242  * Try to flush old fragments to make space for new ones
243  */
244 
245 void
246 pf_flush_fragments(void)
247 {
248 	struct pf_fragment	*frag;
249 	int			 goal;
250 
251 	goal = pf_nfrents * 9 / 10;
252 	DPFPRINTF(("trying to free > %d frents\n",
253 	    pf_nfrents - goal));
254 	while (goal < pf_nfrents) {
255 		frag = TAILQ_LAST(&pf_fragqueue, pf_fragqueue);
256 		if (frag == NULL)
257 			break;
258 		pf_free_fragment(frag);
259 	}
260 
261 
262 	goal = pf_ncache * 9 / 10;
263 	DPFPRINTF(("trying to free > %d cache entries\n",
264 	    pf_ncache - goal));
265 	while (goal < pf_ncache) {
266 		frag = TAILQ_LAST(&pf_cachequeue, pf_cachequeue);
267 		if (frag == NULL)
268 			break;
269 		pf_free_fragment(frag);
270 	}
271 }
272 
273 /* Frees the fragments and all associated entries */
274 
275 void
276 pf_free_fragment(struct pf_fragment *frag)
277 {
278 	struct pf_frent		*frent;
279 	struct pf_frcache	*frcache;
280 
281 	/* Free all fragments */
282 	if (BUFFER_FRAGMENTS(frag)) {
283 		for (frent = LIST_FIRST(&frag->fr_queue); frent;
284 		    frent = LIST_FIRST(&frag->fr_queue)) {
285 			LIST_REMOVE(frent, fr_next);
286 
287 			m_freem(frent->fr_m);
288 			pool_put(&pf_frent_pl, frent);
289 			pf_nfrents--;
290 		}
291 	} else {
292 		for (frcache = LIST_FIRST(&frag->fr_cache); frcache;
293 		    frcache = LIST_FIRST(&frag->fr_cache)) {
294 			LIST_REMOVE(frcache, fr_next);
295 
296 			KASSERT(LIST_EMPTY(&frag->fr_cache) ||
297 			    LIST_FIRST(&frag->fr_cache)->fr_off >
298 			    frcache->fr_end);
299 
300 			pool_put(&pf_cent_pl, frcache);
301 			pf_ncache--;
302 		}
303 	}
304 
305 	pf_remove_fragment(frag);
306 }
307 
308 void
309 pf_ip2key(struct pf_fragment *key, struct ip *ip)
310 {
311 	key->fr_p = ip->ip_p;
312 	key->fr_id = ip->ip_id;
313 	key->fr_src.s_addr = ip->ip_src.s_addr;
314 	key->fr_dst.s_addr = ip->ip_dst.s_addr;
315 }
316 
317 struct pf_fragment *
318 pf_find_fragment(struct ip *ip, struct pf_frag_tree *tree)
319 {
320 	struct pf_fragment	 key;
321 	struct pf_fragment	*frag;
322 
323 	pf_ip2key(&key, ip);
324 
325 	frag = RB_FIND(pf_frag_tree, tree, &key);
326 	if (frag != NULL) {
327 		/* XXX Are we sure we want to update the timeout? */
328 		frag->fr_timeout = time_second;
329 		if (BUFFER_FRAGMENTS(frag)) {
330 			TAILQ_REMOVE(&pf_fragqueue, frag, frag_next);
331 			TAILQ_INSERT_HEAD(&pf_fragqueue, frag, frag_next);
332 		} else {
333 			TAILQ_REMOVE(&pf_cachequeue, frag, frag_next);
334 			TAILQ_INSERT_HEAD(&pf_cachequeue, frag, frag_next);
335 		}
336 	}
337 
338 	return (frag);
339 }
340 
341 /* Removes a fragment from the fragment queue and frees the fragment */
342 
343 void
344 pf_remove_fragment(struct pf_fragment *frag)
345 {
346 	if (BUFFER_FRAGMENTS(frag)) {
347 		RB_REMOVE(pf_frag_tree, &pf_frag_tree, frag);
348 		TAILQ_REMOVE(&pf_fragqueue, frag, frag_next);
349 		pool_put(&pf_frag_pl, frag);
350 	} else {
351 		RB_REMOVE(pf_frag_tree, &pf_cache_tree, frag);
352 		TAILQ_REMOVE(&pf_cachequeue, frag, frag_next);
353 		pool_put(&pf_cache_pl, frag);
354 	}
355 }
356 
357 #define FR_IP_OFF(fr)	((ntohs((fr)->fr_ip->ip_off) & IP_OFFMASK) << 3)
358 struct mbuf *
359 pf_reassemble(struct mbuf **m0, struct pf_fragment **frag,
360     struct pf_frent *frent, int mff)
361 {
362 	struct mbuf	*m = *m0, *m2;
363 	struct pf_frent	*frea, *next;
364 	struct pf_frent	*frep = NULL;
365 	struct ip	*ip = frent->fr_ip;
366 	int		 hlen = ip->ip_hl << 2;
367 	u_int16_t	 off = (ntohs(ip->ip_off) & IP_OFFMASK) << 3;
368 	u_int16_t	 ip_len = ntohs(ip->ip_len) - ip->ip_hl * 4;
369 	u_int16_t	 frmax = ip_len + off;
370 
371 	KASSERT(*frag == NULL || BUFFER_FRAGMENTS(*frag));
372 
373 	/* Strip off ip header */
374 	m->m_data += hlen;
375 	m->m_len -= hlen;
376 
377 	/* Create a new reassembly queue for this packet */
378 	if (*frag == NULL) {
379 		*frag = pool_get(&pf_frag_pl, PR_NOWAIT);
380 		if (*frag == NULL) {
381 			pf_flush_fragments();
382 			*frag = pool_get(&pf_frag_pl, PR_NOWAIT);
383 			if (*frag == NULL)
384 				goto drop_fragment;
385 		}
386 
387 		(*frag)->fr_flags = 0;
388 		(*frag)->fr_max = 0;
389 		(*frag)->fr_src = frent->fr_ip->ip_src;
390 		(*frag)->fr_dst = frent->fr_ip->ip_dst;
391 		(*frag)->fr_p = frent->fr_ip->ip_p;
392 		(*frag)->fr_id = frent->fr_ip->ip_id;
393 		(*frag)->fr_timeout = time_second;
394 		LIST_INIT(&(*frag)->fr_queue);
395 
396 		RB_INSERT(pf_frag_tree, &pf_frag_tree, *frag);
397 		TAILQ_INSERT_HEAD(&pf_fragqueue, *frag, frag_next);
398 
399 		/* We do not have a previous fragment */
400 		frep = NULL;
401 		goto insert;
402 	}
403 
404 	/*
405 	 * Find a fragment after the current one:
406 	 *  - off contains the real shifted offset.
407 	 */
408 	LIST_FOREACH(frea, &(*frag)->fr_queue, fr_next) {
409 		if (FR_IP_OFF(frea) > off)
410 			break;
411 		frep = frea;
412 	}
413 
414 	KASSERT(frep != NULL || frea != NULL);
415 
416 	if (frep != NULL &&
417 	    FR_IP_OFF(frep) + ntohs(frep->fr_ip->ip_len) - frep->fr_ip->ip_hl *
418 	    4 > off)
419 	{
420 		u_int16_t	precut;
421 
422 		precut = FR_IP_OFF(frep) + ntohs(frep->fr_ip->ip_len) -
423 		    frep->fr_ip->ip_hl * 4 - off;
424 		if (precut >= ip_len)
425 			goto drop_fragment;
426 		m_adj(frent->fr_m, precut);
427 		DPFPRINTF(("overlap -%d\n", precut));
428 		/* Enforce 8 byte boundaries */
429 		ip->ip_off = htons(ntohs(ip->ip_off) + (precut >> 3));
430 		off = (ntohs(ip->ip_off) & IP_OFFMASK) << 3;
431 		ip_len -= precut;
432 		ip->ip_len = htons(ip_len);
433 	}
434 
435 	for (; frea != NULL && ip_len + off > FR_IP_OFF(frea);
436 	    frea = next)
437 	{
438 		u_int16_t	aftercut;
439 
440 		aftercut = ip_len + off - FR_IP_OFF(frea);
441 		DPFPRINTF(("adjust overlap %d\n", aftercut));
442 		if (aftercut < ntohs(frea->fr_ip->ip_len) - frea->fr_ip->ip_hl
443 		    * 4)
444 		{
445 			frea->fr_ip->ip_len =
446 			    htons(ntohs(frea->fr_ip->ip_len) - aftercut);
447 			frea->fr_ip->ip_off = htons(ntohs(frea->fr_ip->ip_off) +
448 			    (aftercut >> 3));
449 			m_adj(frea->fr_m, aftercut);
450 			break;
451 		}
452 
453 		/* This fragment is completely overlapped, lose it */
454 		next = LIST_NEXT(frea, fr_next);
455 		m_freem(frea->fr_m);
456 		LIST_REMOVE(frea, fr_next);
457 		pool_put(&pf_frent_pl, frea);
458 		pf_nfrents--;
459 	}
460 
461  insert:
462 	/* Update maximum data size */
463 	if ((*frag)->fr_max < frmax)
464 		(*frag)->fr_max = frmax;
465 	/* This is the last segment */
466 	if (!mff)
467 		(*frag)->fr_flags |= PFFRAG_SEENLAST;
468 
469 	if (frep == NULL)
470 		LIST_INSERT_HEAD(&(*frag)->fr_queue, frent, fr_next);
471 	else
472 		LIST_INSERT_AFTER(frep, frent, fr_next);
473 
474 	/* Check if we are completely reassembled */
475 	if (!((*frag)->fr_flags & PFFRAG_SEENLAST))
476 		return (NULL);
477 
478 	/* Check if we have all the data */
479 	off = 0;
480 	for (frep = LIST_FIRST(&(*frag)->fr_queue); frep; frep = next) {
481 		next = LIST_NEXT(frep, fr_next);
482 
483 		off += ntohs(frep->fr_ip->ip_len) - frep->fr_ip->ip_hl * 4;
484 		if (off < (*frag)->fr_max &&
485 		    (next == NULL || FR_IP_OFF(next) != off))
486 		{
487 			DPFPRINTF(("missing fragment at %d, next %d, max %d\n",
488 			    off, next == NULL ? -1 : FR_IP_OFF(next),
489 			    (*frag)->fr_max));
490 			return (NULL);
491 		}
492 	}
493 	DPFPRINTF(("%d < %d?\n", off, (*frag)->fr_max));
494 	if (off < (*frag)->fr_max)
495 		return (NULL);
496 
497 	/* We have all the data */
498 	frent = LIST_FIRST(&(*frag)->fr_queue);
499 	KASSERT(frent != NULL);
500 	if ((frent->fr_ip->ip_hl << 2) + off > IP_MAXPACKET) {
501 		DPFPRINTF(("drop: too big: %d\n", off));
502 		pf_free_fragment(*frag);
503 		*frag = NULL;
504 		return (NULL);
505 	}
506 	next = LIST_NEXT(frent, fr_next);
507 
508 	/* Magic from ip_input */
509 	ip = frent->fr_ip;
510 	m = frent->fr_m;
511 	m2 = m->m_next;
512 	m->m_next = NULL;
513 	m_cat(m, m2);
514 	pool_put(&pf_frent_pl, frent);
515 	pf_nfrents--;
516 	for (frent = next; frent != NULL; frent = next) {
517 		next = LIST_NEXT(frent, fr_next);
518 
519 		m2 = frent->fr_m;
520 		pool_put(&pf_frent_pl, frent);
521 		pf_nfrents--;
522 		m_cat(m, m2);
523 	}
524 
525 	ip->ip_src = (*frag)->fr_src;
526 	ip->ip_dst = (*frag)->fr_dst;
527 
528 	/* Remove from fragment queue */
529 	pf_remove_fragment(*frag);
530 	*frag = NULL;
531 
532 	hlen = ip->ip_hl << 2;
533 	ip->ip_len = htons(off + hlen);
534 	m->m_len += hlen;
535 	m->m_data -= hlen;
536 
537 	/* some debugging cruft by sklower, below, will go away soon */
538 	/* XXX this should be done elsewhere */
539 	if (m->m_flags & M_PKTHDR) {
540 		int plen = 0;
541 		for (m2 = m; m2; m2 = m2->m_next)
542 			plen += m2->m_len;
543 		m->m_pkthdr.len = plen;
544 #ifdef __NetBSD__
545 		m->m_pkthdr.csum_flags = 0;
546 #endif /* __NetBSD__ */
547 	}
548 
549 	DPFPRINTF(("complete: %p(%d)\n", m, ntohs(ip->ip_len)));
550 	return (m);
551 
552  drop_fragment:
553 	/* Oops - fail safe - drop packet */
554 	pool_put(&pf_frent_pl, frent);
555 	pf_nfrents--;
556 	m_freem(m);
557 	return (NULL);
558 }
559 
560 struct mbuf *
561 pf_fragcache(struct mbuf **m0, struct ip *h, struct pf_fragment **frag, int mff,
562     int drop, int *nomem)
563 {
564 	struct mbuf		*m = *m0;
565 	struct pf_frcache	*frp, *fra, *cur = NULL;
566 	int			 ip_len = ntohs(h->ip_len) - (h->ip_hl << 2);
567 	u_int16_t		 off = ntohs(h->ip_off) << 3;
568 	u_int16_t		 frmax = ip_len + off;
569 	int			 hosed = 0;
570 
571 	KASSERT(*frag == NULL || !BUFFER_FRAGMENTS(*frag));
572 
573 	/* Create a new range queue for this packet */
574 	if (*frag == NULL) {
575 		*frag = pool_get(&pf_cache_pl, PR_NOWAIT);
576 		if (*frag == NULL) {
577 			pf_flush_fragments();
578 			*frag = pool_get(&pf_cache_pl, PR_NOWAIT);
579 			if (*frag == NULL)
580 				goto no_mem;
581 		}
582 
583 		/* Get an entry for the queue */
584 		cur = pool_get(&pf_cent_pl, PR_NOWAIT);
585 		if (cur == NULL) {
586 			pool_put(&pf_cache_pl, *frag);
587 			*frag = NULL;
588 			goto no_mem;
589 		}
590 		pf_ncache++;
591 
592 		(*frag)->fr_flags = PFFRAG_NOBUFFER;
593 		(*frag)->fr_max = 0;
594 		(*frag)->fr_src = h->ip_src;
595 		(*frag)->fr_dst = h->ip_dst;
596 		(*frag)->fr_p = h->ip_p;
597 		(*frag)->fr_id = h->ip_id;
598 		(*frag)->fr_timeout = time_second;
599 
600 		cur->fr_off = off;
601 		cur->fr_end = frmax;
602 		LIST_INIT(&(*frag)->fr_cache);
603 		LIST_INSERT_HEAD(&(*frag)->fr_cache, cur, fr_next);
604 
605 		RB_INSERT(pf_frag_tree, &pf_cache_tree, *frag);
606 		TAILQ_INSERT_HEAD(&pf_cachequeue, *frag, frag_next);
607 
608 		DPFPRINTF(("fragcache[%d]: new %d-%d\n", h->ip_id, off, frmax));
609 
610 		goto pass;
611 	}
612 
613 	/*
614 	 * Find a fragment after the current one:
615 	 *  - off contains the real shifted offset.
616 	 */
617 	frp = NULL;
618 	LIST_FOREACH(fra, &(*frag)->fr_cache, fr_next) {
619 		if (fra->fr_off > off)
620 			break;
621 		frp = fra;
622 	}
623 
624 	KASSERT(frp != NULL || fra != NULL);
625 
626 	if (frp != NULL) {
627 		int	precut;
628 
629 		precut = frp->fr_end - off;
630 		if (precut >= ip_len) {
631 			/* Fragment is entirely a duplicate */
632 			DPFPRINTF(("fragcache[%d]: dead (%d-%d) %d-%d\n",
633 			    h->ip_id, frp->fr_off, frp->fr_end, off, frmax));
634 			goto drop_fragment;
635 		}
636 		if (precut == 0) {
637 			/* They are adjacent.  Fixup cache entry */
638 			DPFPRINTF(("fragcache[%d]: adjacent (%d-%d) %d-%d\n",
639 			    h->ip_id, frp->fr_off, frp->fr_end, off, frmax));
640 			frp->fr_end = frmax;
641 		} else if (precut > 0) {
642 			/* The first part of this payload overlaps with a
643 			 * fragment that has already been passed.
644 			 * Need to trim off the first part of the payload.
645 			 * But to do so easily, we need to create another
646 			 * mbuf to throw the original header into.
647 			 */
648 
649 			DPFPRINTF(("fragcache[%d]: chop %d (%d-%d) %d-%d\n",
650 			    h->ip_id, precut, frp->fr_off, frp->fr_end, off,
651 			    frmax));
652 
653 			off += precut;
654 			frmax -= precut;
655 			/* Update the previous frag to encompass this one */
656 			frp->fr_end = frmax;
657 
658 			if (!drop) {
659 				/* XXX Optimization opportunity
660 				 * This is a very heavy way to trim the payload.
661 				 * we could do it much faster by diddling mbuf
662 				 * internals but that would be even less legible
663 				 * than this mbuf magic.  For my next trick,
664 				 * I'll pull a rabbit out of my laptop.
665 				 */
666 				*m0 = m_dup(m, 0, h->ip_hl << 2, M_NOWAIT);
667 				if (*m0 == NULL)
668 					goto no_mem;
669 				KASSERT((*m0)->m_next == NULL);
670 				m_adj(m, precut + (h->ip_hl << 2));
671 				m_cat(*m0, m);
672 				m = *m0;
673 				if (m->m_flags & M_PKTHDR) {
674 					int plen = 0;
675 					struct mbuf *t;
676 					for (t = m; t; t = t->m_next)
677 						plen += t->m_len;
678 					m->m_pkthdr.len = plen;
679 				}
680 
681 
682 				h = mtod(m, struct ip *);
683 
684 
685 				KASSERT((int)m->m_len ==
686 				    ntohs(h->ip_len) - precut);
687 				h->ip_off = htons(ntohs(h->ip_off) +
688 				    (precut >> 3));
689 				h->ip_len = htons(ntohs(h->ip_len) - precut);
690 			} else {
691 				hosed++;
692 			}
693 		} else {
694 			/* There is a gap between fragments */
695 
696 			DPFPRINTF(("fragcache[%d]: gap %d (%d-%d) %d-%d\n",
697 			    h->ip_id, -precut, frp->fr_off, frp->fr_end, off,
698 			    frmax));
699 
700 			cur = pool_get(&pf_cent_pl, PR_NOWAIT);
701 			if (cur == NULL)
702 				goto no_mem;
703 			pf_ncache++;
704 
705 			cur->fr_off = off;
706 			cur->fr_end = frmax;
707 			LIST_INSERT_AFTER(frp, cur, fr_next);
708 		}
709 	}
710 
711 	if (fra != NULL) {
712 		int	aftercut;
713 		int	merge = 0;
714 
715 		aftercut = frmax - fra->fr_off;
716 		if (aftercut == 0) {
717 			/* Adjacent fragments */
718 			DPFPRINTF(("fragcache[%d]: adjacent %d-%d (%d-%d)\n",
719 			    h->ip_id, off, frmax, fra->fr_off, fra->fr_end));
720 			fra->fr_off = off;
721 			merge = 1;
722 		} else if (aftercut > 0) {
723 			/* Need to chop off the tail of this fragment */
724 			DPFPRINTF(("fragcache[%d]: chop %d %d-%d (%d-%d)\n",
725 			    h->ip_id, aftercut, off, frmax, fra->fr_off,
726 			    fra->fr_end));
727 			fra->fr_off = off;
728 			frmax -= aftercut;
729 
730 			merge = 1;
731 
732 			if (!drop) {
733 				m_adj(m, -aftercut);
734 				if (m->m_flags & M_PKTHDR) {
735 					int plen = 0;
736 					struct mbuf *t;
737 					for (t = m; t; t = t->m_next)
738 						plen += t->m_len;
739 					m->m_pkthdr.len = plen;
740 				}
741 				h = mtod(m, struct ip *);
742 				KASSERT((int)m->m_len ==
743 				    ntohs(h->ip_len) - aftercut);
744 				h->ip_len = htons(ntohs(h->ip_len) - aftercut);
745 			} else {
746 				hosed++;
747 			}
748 		} else if (frp == NULL) {
749 			/* There is a gap between fragments */
750 			DPFPRINTF(("fragcache[%d]: gap %d %d-%d (%d-%d)\n",
751 			    h->ip_id, -aftercut, off, frmax, fra->fr_off,
752 			    fra->fr_end));
753 
754 			cur = pool_get(&pf_cent_pl, PR_NOWAIT);
755 			if (cur == NULL)
756 				goto no_mem;
757 			pf_ncache++;
758 
759 			cur->fr_off = off;
760 			cur->fr_end = frmax;
761 			LIST_INSERT_BEFORE(fra, cur, fr_next);
762 		}
763 
764 
765 		/* Need to glue together two separate fragment descriptors */
766 		if (merge) {
767 			if (cur && fra->fr_off <= cur->fr_end) {
768 				/* Need to merge in a previous 'cur' */
769 				DPFPRINTF(("fragcache[%d]: adjacent(merge "
770 				    "%d-%d) %d-%d (%d-%d)\n",
771 				    h->ip_id, cur->fr_off, cur->fr_end, off,
772 				    frmax, fra->fr_off, fra->fr_end));
773 				fra->fr_off = cur->fr_off;
774 				LIST_REMOVE(cur, fr_next);
775 				pool_put(&pf_cent_pl, cur);
776 				pf_ncache--;
777 				cur = NULL;
778 
779 			} else if (frp && fra->fr_off <= frp->fr_end) {
780 				/* Need to merge in a modified 'frp' */
781 				KASSERT(cur == NULL);
782 				DPFPRINTF(("fragcache[%d]: adjacent(merge "
783 				    "%d-%d) %d-%d (%d-%d)\n",
784 				    h->ip_id, frp->fr_off, frp->fr_end, off,
785 				    frmax, fra->fr_off, fra->fr_end));
786 				fra->fr_off = frp->fr_off;
787 				LIST_REMOVE(frp, fr_next);
788 				pool_put(&pf_cent_pl, frp);
789 				pf_ncache--;
790 				frp = NULL;
791 
792 			}
793 		}
794 	}
795 
796 	if (hosed) {
797 		/*
798 		 * We must keep tracking the overall fragment even when
799 		 * we're going to drop it anyway so that we know when to
800 		 * free the overall descriptor.  Thus we drop the frag late.
801 		 */
802 		goto drop_fragment;
803 	}
804 
805 
806  pass:
807 	/* Update maximum data size */
808 	if ((*frag)->fr_max < frmax)
809 		(*frag)->fr_max = frmax;
810 
811 	/* This is the last segment */
812 	if (!mff)
813 		(*frag)->fr_flags |= PFFRAG_SEENLAST;
814 
815 	/* Check if we are completely reassembled */
816 	if (((*frag)->fr_flags & PFFRAG_SEENLAST) &&
817 	    LIST_FIRST(&(*frag)->fr_cache)->fr_off == 0 &&
818 	    LIST_FIRST(&(*frag)->fr_cache)->fr_end == (*frag)->fr_max) {
819 		/* Remove from fragment queue */
820 		DPFPRINTF(("fragcache[%d]: done 0-%d\n", h->ip_id,
821 		    (*frag)->fr_max));
822 		pf_free_fragment(*frag);
823 		*frag = NULL;
824 	}
825 
826 	return (m);
827 
828  no_mem:
829 	*nomem = 1;
830 
831 	/* Still need to pay attention to !IP_MF */
832 	if (!mff && *frag != NULL)
833 		(*frag)->fr_flags |= PFFRAG_SEENLAST;
834 
835 	m_freem(m);
836 	return (NULL);
837 
838  drop_fragment:
839 
840 	/* Still need to pay attention to !IP_MF */
841 	if (!mff && *frag != NULL)
842 		(*frag)->fr_flags |= PFFRAG_SEENLAST;
843 
844 	if (drop) {
845 		/* This fragment has been deemed bad.  Don't reass */
846 		if (((*frag)->fr_flags & PFFRAG_DROP) == 0)
847 			DPFPRINTF(("fragcache[%d]: dropping overall fragment\n",
848 			    h->ip_id));
849 		(*frag)->fr_flags |= PFFRAG_DROP;
850 	}
851 
852 	m_freem(m);
853 	return (NULL);
854 }
855 
856 int
857 pf_normalize_ip(struct mbuf **m0, int dir, struct pfi_kif *kif, u_short *reason,
858     struct pf_pdesc *pd)
859 {
860 	struct mbuf		*m = *m0;
861 	struct pf_rule		*r;
862 	struct pf_frent		*frent;
863 	struct pf_fragment	*frag = NULL;
864 	struct ip		*h = mtod(m, struct ip *);
865 	int			 mff = (ntohs(h->ip_off) & IP_MF);
866 	int			 hlen = h->ip_hl << 2;
867 	u_int16_t		 fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
868 	u_int16_t		 frmax;
869 	int			 ip_len;
870 	int			 ip_off;
871 
872 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
873 	while (r != NULL) {
874 		r->evaluations++;
875 		if (pfi_kif_match(r->kif, kif) == r->ifnot)
876 			r = r->skip[PF_SKIP_IFP].ptr;
877 		else if (r->direction && r->direction != dir)
878 			r = r->skip[PF_SKIP_DIR].ptr;
879 		else if (r->af && r->af != AF_INET)
880 			r = r->skip[PF_SKIP_AF].ptr;
881 		else if (r->proto && r->proto != h->ip_p)
882 			r = r->skip[PF_SKIP_PROTO].ptr;
883 		else if (PF_MISMATCHAW(&r->src.addr,
884 		    (struct pf_addr *)&h->ip_src.s_addr, AF_INET,
885 		    r->src.neg, kif))
886 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
887 		else if (PF_MISMATCHAW(&r->dst.addr,
888 		    (struct pf_addr *)&h->ip_dst.s_addr, AF_INET,
889 		    r->dst.neg, NULL))
890 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
891 		else
892 			break;
893 	}
894 
895 	if (r == NULL || r->action == PF_NOSCRUB)
896 		return (PF_PASS);
897 	else {
898 		r->packets[dir == PF_OUT]++;
899 		r->bytes[dir == PF_OUT] += pd->tot_len;
900 	}
901 
902 	/* Check for illegal packets */
903 	if (hlen < (int)sizeof(struct ip))
904 		goto drop;
905 
906 	if (hlen > ntohs(h->ip_len))
907 		goto drop;
908 
909 	/* Clear IP_DF if the rule uses the no-df option */
910 	if (r->rule_flag & PFRULE_NODF && h->ip_off & htons(IP_DF)) {
911 		u_int16_t off = h->ip_off;
912 
913 		h->ip_off &= htons(~IP_DF);
914 		h->ip_sum = pf_cksum_fixup(h->ip_sum, off, h->ip_off, 0);
915 	}
916 
917 	/* We will need other tests here */
918 	if (!fragoff && !mff)
919 		goto no_fragment;
920 
921 	/* We're dealing with a fragment now. Don't allow fragments
922 	 * with IP_DF to enter the cache. If the flag was cleared by
923 	 * no-df above, fine. Otherwise drop it.
924 	 */
925 	if (h->ip_off & htons(IP_DF)) {
926 		DPFPRINTF(("IP_DF\n"));
927 		goto bad;
928 	}
929 
930 	ip_len = ntohs(h->ip_len) - hlen;
931 	ip_off = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
932 
933 	/* All fragments are 8 byte aligned */
934 	if (mff && (ip_len & 0x7)) {
935 		DPFPRINTF(("mff and %d\n", ip_len));
936 		goto bad;
937 	}
938 
939 	/* Respect maximum length */
940 	if (fragoff + ip_len > IP_MAXPACKET) {
941 		DPFPRINTF(("max packet %d\n", fragoff + ip_len));
942 		goto bad;
943 	}
944 	frmax = fragoff + ip_len;
945 
946 	if ((r->rule_flag & (PFRULE_FRAGCROP|PFRULE_FRAGDROP)) == 0) {
947 		/* Fully buffer all of the fragments */
948 
949 		frag = pf_find_fragment(h, &pf_frag_tree);
950 
951 		/* Check if we saw the last fragment already */
952 		if (frag != NULL && (frag->fr_flags & PFFRAG_SEENLAST) &&
953 		    frmax > frag->fr_max)
954 			goto bad;
955 
956 		/* Get an entry for the fragment queue */
957 		frent = pool_get(&pf_frent_pl, PR_NOWAIT);
958 		if (frent == NULL) {
959 			REASON_SET(reason, PFRES_MEMORY);
960 			return (PF_DROP);
961 		}
962 		pf_nfrents++;
963 		frent->fr_ip = h;
964 		frent->fr_m = m;
965 
966 		/* Might return a completely reassembled mbuf, or NULL */
967 		DPFPRINTF(("reass frag %d @ %d-%d\n", h->ip_id, fragoff, frmax));
968 		*m0 = m = pf_reassemble(m0, &frag, frent, mff);
969 
970 		if (m == NULL)
971 			return (PF_DROP);
972 
973 		if (frag != NULL && (frag->fr_flags & PFFRAG_DROP))
974 			goto drop;
975 
976 		h = mtod(m, struct ip *);
977 	} else {
978 		/* non-buffering fragment cache (drops or masks overlaps) */
979 		int	nomem = 0;
980 
981 #ifdef __NetBSD__
982 		struct pf_mtag *pf_mtag = pf_find_mtag(m);
983 		KASSERT(pf_mtag != NULL);
984 
985 		if (dir == PF_OUT && pf_mtag->flags & PF_TAG_FRAGCACHE) {
986 #else
987 		if (dir == PF_OUT && m->m_pkthdr.pf.flags & PF_TAG_FRAGCACHE) {
988 #endif /* !__NetBSD__ */
989 			/*
990 			 * Already passed the fragment cache in the
991 			 * input direction.  If we continued, it would
992 			 * appear to be a dup and would be dropped.
993 			 */
994 			goto fragment_pass;
995 		}
996 
997 		frag = pf_find_fragment(h, &pf_cache_tree);
998 
999 		/* Check if we saw the last fragment already */
1000 		if (frag != NULL && (frag->fr_flags & PFFRAG_SEENLAST) &&
1001 		    frmax > frag->fr_max) {
1002 			if (r->rule_flag & PFRULE_FRAGDROP)
1003 				frag->fr_flags |= PFFRAG_DROP;
1004 			goto bad;
1005 		}
1006 
1007 		*m0 = m = pf_fragcache(m0, h, &frag, mff,
1008 		    (r->rule_flag & PFRULE_FRAGDROP) ? 1 : 0, &nomem);
1009 		if (m == NULL) {
1010 			if (nomem)
1011 				goto no_mem;
1012 			goto drop;
1013 		}
1014 
1015 		if (dir == PF_IN)
1016 #ifdef __NetBSD__
1017 			pf_mtag = pf_find_mtag(m);
1018 			KASSERT(pf_mtag != NULL);
1019 
1020 			pf_mtag->flags |= PF_TAG_FRAGCACHE;
1021 #else
1022 			m->m_pkthdr.pf.flags |= PF_TAG_FRAGCACHE;
1023 #endif /* !__NetBSD__ */
1024 
1025 		if (frag != NULL && (frag->fr_flags & PFFRAG_DROP))
1026 			goto drop;
1027 		goto fragment_pass;
1028 	}
1029 
1030  no_fragment:
1031 	/* At this point, only IP_DF is allowed in ip_off */
1032 	if (h->ip_off & ~htons(IP_DF)) {
1033 		u_int16_t off = h->ip_off;
1034 
1035 		h->ip_off &= htons(IP_DF);
1036 		h->ip_sum = pf_cksum_fixup(h->ip_sum, off, h->ip_off, 0);
1037 	}
1038 
1039 	/* Enforce a minimum ttl, may cause endless packet loops */
1040 	if (r->min_ttl && h->ip_ttl < r->min_ttl) {
1041 		u_int16_t ip_ttl = h->ip_ttl;
1042 
1043 		h->ip_ttl = r->min_ttl;
1044 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_ttl, h->ip_ttl, 0);
1045 	}
1046 
1047 	if (r->rule_flag & PFRULE_RANDOMID) {
1048 		u_int16_t id = h->ip_id;
1049 
1050 		h->ip_id = ip_randomid(ip_ids, 0);
1051 		h->ip_sum = pf_cksum_fixup(h->ip_sum, id, h->ip_id, 0);
1052 	}
1053 	if ((r->rule_flag & (PFRULE_FRAGCROP|PFRULE_FRAGDROP)) == 0)
1054 		pd->flags |= PFDESC_IP_REAS;
1055 
1056 	return (PF_PASS);
1057 
1058  fragment_pass:
1059 	/* Enforce a minimum ttl, may cause endless packet loops */
1060 	if (r->min_ttl && h->ip_ttl < r->min_ttl) {
1061 		u_int16_t ip_ttl = h->ip_ttl;
1062 
1063 		h->ip_ttl = r->min_ttl;
1064 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_ttl, h->ip_ttl, 0);
1065 	}
1066 	if ((r->rule_flag & (PFRULE_FRAGCROP|PFRULE_FRAGDROP)) == 0)
1067 		pd->flags |= PFDESC_IP_REAS;
1068 	return (PF_PASS);
1069 
1070  no_mem:
1071 	REASON_SET(reason, PFRES_MEMORY);
1072 	if (r != NULL && r->log)
1073 		PFLOG_PACKET(kif, h, m, AF_INET, dir, *reason, r, NULL, NULL, pd);
1074 	return (PF_DROP);
1075 
1076  drop:
1077 	REASON_SET(reason, PFRES_NORM);
1078 	if (r != NULL && r->log)
1079 		PFLOG_PACKET(kif, h, m, AF_INET, dir, *reason, r, NULL, NULL, pd);
1080 	return (PF_DROP);
1081 
1082  bad:
1083 	DPFPRINTF(("dropping bad fragment\n"));
1084 
1085 	/* Free associated fragments */
1086 	if (frag != NULL)
1087 		pf_free_fragment(frag);
1088 
1089 	REASON_SET(reason, PFRES_FRAG);
1090 	if (r != NULL && r->log)
1091 		PFLOG_PACKET(kif, h, m, AF_INET, dir, *reason, r, NULL, NULL, pd);
1092 
1093 	return (PF_DROP);
1094 }
1095 
1096 #ifdef INET6
1097 int
1098 pf_normalize_ip6(struct mbuf **m0, int dir, struct pfi_kif *kif,
1099     u_short *reason, struct pf_pdesc *pd)
1100 {
1101 	struct mbuf		*m = *m0;
1102 	struct pf_rule		*r;
1103 	struct ip6_hdr		*h = mtod(m, struct ip6_hdr *);
1104 	int			 off;
1105 	struct ip6_ext		 ext;
1106 	struct ip6_opt		 opt;
1107 	struct ip6_opt_jumbo	 jumbo;
1108 	struct ip6_frag		 frag;
1109 	u_int32_t		 jumbolen = 0, plen;
1110 	u_int16_t		 fragoff = 0;
1111 	int			 optend;
1112 	int			 ooff;
1113 	u_int8_t		 proto;
1114 	int			 terminal;
1115 
1116 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1117 	while (r != NULL) {
1118 		r->evaluations++;
1119 		if (pfi_kif_match(r->kif, kif) == r->ifnot)
1120 			r = r->skip[PF_SKIP_IFP].ptr;
1121 		else if (r->direction && r->direction != dir)
1122 			r = r->skip[PF_SKIP_DIR].ptr;
1123 		else if (r->af && r->af != AF_INET6)
1124 			r = r->skip[PF_SKIP_AF].ptr;
1125 #if 0 /* header chain! */
1126 		else if (r->proto && r->proto != h->ip6_nxt)
1127 			r = r->skip[PF_SKIP_PROTO].ptr;
1128 #endif
1129 		else if (PF_MISMATCHAW(&r->src.addr,
1130 		    (struct pf_addr *)&h->ip6_src, AF_INET6,
1131 		    r->src.neg, kif))
1132 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
1133 		else if (PF_MISMATCHAW(&r->dst.addr,
1134 		    (struct pf_addr *)&h->ip6_dst, AF_INET6,
1135 		    r->dst.neg, NULL))
1136 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
1137 		else
1138 			break;
1139 	}
1140 
1141 	if (r == NULL || r->action == PF_NOSCRUB)
1142 		return (PF_PASS);
1143 	else {
1144 		r->packets[dir == PF_OUT]++;
1145 		r->bytes[dir == PF_OUT] += pd->tot_len;
1146 	}
1147 
1148 	/* Check for illegal packets */
1149 	if (sizeof(struct ip6_hdr) + IPV6_MAXPACKET < m->m_pkthdr.len)
1150 		goto drop;
1151 
1152 	off = sizeof(struct ip6_hdr);
1153 	proto = h->ip6_nxt;
1154 	terminal = 0;
1155 	do {
1156 		switch (proto) {
1157 		case IPPROTO_FRAGMENT:
1158 			goto fragment;
1159 			break;
1160 		case IPPROTO_AH:
1161 		case IPPROTO_ROUTING:
1162 		case IPPROTO_DSTOPTS:
1163 			if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL,
1164 			    NULL, AF_INET6))
1165 				goto shortpkt;
1166 			if (proto == IPPROTO_AH)
1167 				off += (ext.ip6e_len + 2) * 4;
1168 			else
1169 				off += (ext.ip6e_len + 1) * 8;
1170 			proto = ext.ip6e_nxt;
1171 			break;
1172 		case IPPROTO_HOPOPTS:
1173 			if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL,
1174 			    NULL, AF_INET6))
1175 				goto shortpkt;
1176 			optend = off + (ext.ip6e_len + 1) * 8;
1177 			ooff = off + sizeof(ext);
1178 			do {
1179 				if (!pf_pull_hdr(m, ooff, &opt.ip6o_type,
1180 				    sizeof(opt.ip6o_type), NULL, NULL,
1181 				    AF_INET6))
1182 					goto shortpkt;
1183 				if (opt.ip6o_type == IP6OPT_PAD1) {
1184 					ooff++;
1185 					continue;
1186 				}
1187 				if (!pf_pull_hdr(m, ooff, &opt, sizeof(opt),
1188 				    NULL, NULL, AF_INET6))
1189 					goto shortpkt;
1190 				if (ooff + sizeof(opt) + opt.ip6o_len > optend)
1191 					goto drop;
1192 				switch (opt.ip6o_type) {
1193 				case IP6OPT_JUMBO:
1194 					if (h->ip6_plen != 0)
1195 						goto drop;
1196 					if (!pf_pull_hdr(m, ooff, &jumbo,
1197 					    sizeof(jumbo), NULL, NULL,
1198 					    AF_INET6))
1199 						goto shortpkt;
1200 					memcpy(&jumbolen, jumbo.ip6oj_jumbo_len,
1201 					    sizeof(jumbolen));
1202 					jumbolen = ntohl(jumbolen);
1203 					if (jumbolen <= IPV6_MAXPACKET)
1204 						goto drop;
1205 					if (sizeof(struct ip6_hdr) + jumbolen !=
1206 					    m->m_pkthdr.len)
1207 						goto drop;
1208 					break;
1209 				default:
1210 					break;
1211 				}
1212 				ooff += sizeof(opt) + opt.ip6o_len;
1213 			} while (ooff < optend);
1214 
1215 			off = optend;
1216 			proto = ext.ip6e_nxt;
1217 			break;
1218 		default:
1219 			terminal = 1;
1220 			break;
1221 		}
1222 	} while (!terminal);
1223 
1224 	/* jumbo payload option must be present, or plen > 0 */
1225 	if (ntohs(h->ip6_plen) == 0)
1226 		plen = jumbolen;
1227 	else
1228 		plen = ntohs(h->ip6_plen);
1229 	if (plen == 0)
1230 		goto drop;
1231 	if (sizeof(struct ip6_hdr) + plen > m->m_pkthdr.len)
1232 		goto shortpkt;
1233 
1234 	/* Enforce a minimum ttl, may cause endless packet loops */
1235 	if (r->min_ttl && h->ip6_hlim < r->min_ttl)
1236 		h->ip6_hlim = r->min_ttl;
1237 
1238 	return (PF_PASS);
1239 
1240  fragment:
1241 	if (ntohs(h->ip6_plen) == 0 || jumbolen)
1242 		goto drop;
1243 	plen = ntohs(h->ip6_plen);
1244 
1245 	if (!pf_pull_hdr(m, off, &frag, sizeof(frag), NULL, NULL, AF_INET6))
1246 		goto shortpkt;
1247 	fragoff = ntohs(frag.ip6f_offlg & IP6F_OFF_MASK);
1248 	if (fragoff + (plen - off - sizeof(frag)) > IPV6_MAXPACKET)
1249 		goto badfrag;
1250 
1251 	/* do something about it */
1252 	/* remember to set pd->flags |= PFDESC_IP_REAS */
1253 	return (PF_PASS);
1254 
1255  shortpkt:
1256 	REASON_SET(reason, PFRES_SHORT);
1257 	if (r != NULL && r->log)
1258 		PFLOG_PACKET(kif, h, m, AF_INET6, dir, *reason, r, NULL, NULL, pd);
1259 	return (PF_DROP);
1260 
1261  drop:
1262 	REASON_SET(reason, PFRES_NORM);
1263 	if (r != NULL && r->log)
1264 		PFLOG_PACKET(kif, h, m, AF_INET6, dir, *reason, r, NULL, NULL, pd);
1265 	return (PF_DROP);
1266 
1267  badfrag:
1268 	REASON_SET(reason, PFRES_FRAG);
1269 	if (r != NULL && r->log)
1270 		PFLOG_PACKET(kif, h, m, AF_INET6, dir, *reason, r, NULL, NULL, pd);
1271 	return (PF_DROP);
1272 }
1273 #endif /* INET6 */
1274 
1275 int
1276 pf_normalize_tcp(int dir, struct pfi_kif *kif, struct mbuf *m,
1277     int ipoff, int off, void *h, struct pf_pdesc *pd)
1278 {
1279 	struct pf_rule	*r, *rm = NULL;
1280 	struct tcphdr	*th = pd->hdr.tcp;
1281 	int		 rewrite = 0;
1282 	u_short		 reason;
1283 	u_int8_t	 flags;
1284 	sa_family_t	 af = pd->af;
1285 
1286 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1287 	while (r != NULL) {
1288 		r->evaluations++;
1289 		if (pfi_kif_match(r->kif, kif) == r->ifnot)
1290 			r = r->skip[PF_SKIP_IFP].ptr;
1291 		else if (r->direction && r->direction != dir)
1292 			r = r->skip[PF_SKIP_DIR].ptr;
1293 		else if (r->af && r->af != af)
1294 			r = r->skip[PF_SKIP_AF].ptr;
1295 		else if (r->proto && r->proto != pd->proto)
1296 			r = r->skip[PF_SKIP_PROTO].ptr;
1297 		else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
1298 		    r->src.neg, kif))
1299 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
1300 		else if (r->src.port_op && !pf_match_port(r->src.port_op,
1301 			    r->src.port[0], r->src.port[1], th->th_sport))
1302 			r = r->skip[PF_SKIP_SRC_PORT].ptr;
1303 		else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
1304 		    r->dst.neg, NULL))
1305 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
1306 		else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
1307 			    r->dst.port[0], r->dst.port[1], th->th_dport))
1308 			r = r->skip[PF_SKIP_DST_PORT].ptr;
1309 		else if (r->os_fingerprint != PF_OSFP_ANY && !pf_osfp_match(
1310 			    pf_osfp_fingerprint(pd, m, off, th),
1311 			    r->os_fingerprint))
1312 			r = TAILQ_NEXT(r, entries);
1313 		else {
1314 			rm = r;
1315 			break;
1316 		}
1317 	}
1318 
1319 	if (rm == NULL || rm->action == PF_NOSCRUB)
1320 		return (PF_PASS);
1321 	else {
1322 		r->packets[dir == PF_OUT]++;
1323 		r->bytes[dir == PF_OUT] += pd->tot_len;
1324 	}
1325 
1326 	if (rm->rule_flag & PFRULE_REASSEMBLE_TCP)
1327 		pd->flags |= PFDESC_TCP_NORM;
1328 
1329 	flags = th->th_flags;
1330 	if (flags & TH_SYN) {
1331 		/* Illegal packet */
1332 		if (flags & TH_RST)
1333 			goto tcp_drop;
1334 
1335 		if (flags & TH_FIN)
1336 			flags &= ~TH_FIN;
1337 	} else {
1338 		/* Illegal packet */
1339 		if (!(flags & (TH_ACK|TH_RST)))
1340 			goto tcp_drop;
1341 	}
1342 
1343 	if (!(flags & TH_ACK)) {
1344 		/* These flags are only valid if ACK is set */
1345 		if ((flags & TH_FIN) || (flags & TH_PUSH) || (flags & TH_URG))
1346 			goto tcp_drop;
1347 	}
1348 
1349 	/* Check for illegal header length */
1350 	if (th->th_off < (sizeof(struct tcphdr) >> 2))
1351 		goto tcp_drop;
1352 
1353 	/* If flags changed, or reserved data set, then adjust */
1354 	if (flags != th->th_flags || th->th_x2 != 0) {
1355 		u_int16_t	ov, nv;
1356 
1357 		ov = *(u_int16_t *)(&th->th_ack + 1);
1358 		th->th_flags = flags;
1359 		th->th_x2 = 0;
1360 		nv = *(u_int16_t *)(&th->th_ack + 1);
1361 
1362 		th->th_sum = pf_cksum_fixup(th->th_sum, ov, nv, 0);
1363 		rewrite = 1;
1364 	}
1365 
1366 	/* Remove urgent pointer, if TH_URG is not set */
1367 	if (!(flags & TH_URG) && th->th_urp) {
1368 		th->th_sum = pf_cksum_fixup(th->th_sum, th->th_urp, 0, 0);
1369 		th->th_urp = 0;
1370 		rewrite = 1;
1371 	}
1372 
1373 	/* Process options */
1374 	if (r->max_mss && pf_normalize_tcpopt(r, m, th, off))
1375 		rewrite = 1;
1376 
1377 	/* copy back packet headers if we sanitized */
1378 	if (rewrite)
1379 		m_copyback(m, off, sizeof(*th), th);
1380 
1381 	return (PF_PASS);
1382 
1383  tcp_drop:
1384 	REASON_SET(&reason, PFRES_NORM);
1385 	if (rm != NULL && r->log)
1386 		PFLOG_PACKET(kif, h, m, AF_INET, dir, reason, r, NULL, NULL, pd);
1387 	return (PF_DROP);
1388 }
1389 
1390 int
1391 pf_normalize_tcp_init(struct mbuf *m, int off, struct pf_pdesc *pd,
1392     struct tcphdr *th, struct pf_state_peer *src,
1393     struct pf_state_peer *dst)
1394 {
1395 	u_int32_t tsval, tsecr;
1396 	u_int8_t hdr[60];
1397 	u_int8_t *opt;
1398 
1399 	KASSERT(src->scrub == NULL);
1400 
1401 	src->scrub = pool_get(&pf_state_scrub_pl, PR_NOWAIT);
1402 	if (src->scrub == NULL)
1403 		return (1);
1404 	bzero(src->scrub, sizeof(*src->scrub));
1405 
1406 	switch (pd->af) {
1407 #ifdef INET
1408 	case AF_INET: {
1409 		struct ip *h = mtod(m, struct ip *);
1410 		src->scrub->pfss_ttl = h->ip_ttl;
1411 		break;
1412 	}
1413 #endif /* INET */
1414 #ifdef INET6
1415 	case AF_INET6: {
1416 		struct ip6_hdr *h = mtod(m, struct ip6_hdr *);
1417 		src->scrub->pfss_ttl = h->ip6_hlim;
1418 		break;
1419 	}
1420 #endif /* INET6 */
1421 	}
1422 
1423 
1424 	/*
1425 	 * All normalizations below are only begun if we see the start of
1426 	 * the connections.  They must all set an enabled bit in pfss_flags
1427 	 */
1428 	if ((th->th_flags & TH_SYN) == 0)
1429 		return (0);
1430 
1431 
1432 	if (th->th_off > (sizeof(struct tcphdr) >> 2) && src->scrub &&
1433 	    pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) {
1434 		/* Diddle with TCP options */
1435 		int hlen;
1436 		opt = hdr + sizeof(struct tcphdr);
1437 		hlen = (th->th_off << 2) - sizeof(struct tcphdr);
1438 		while (hlen >= TCPOLEN_TIMESTAMP) {
1439 			switch (*opt) {
1440 			case TCPOPT_EOL:	/* FALLTHROUGH */
1441 			case TCPOPT_NOP:
1442 				opt++;
1443 				hlen--;
1444 				break;
1445 			case TCPOPT_TIMESTAMP:
1446 				if (opt[1] >= TCPOLEN_TIMESTAMP) {
1447 					src->scrub->pfss_flags |=
1448 					    PFSS_TIMESTAMP;
1449 					src->scrub->pfss_ts_mod =
1450 					    htonl(cprng_fast32());
1451 
1452 					/* note PFSS_PAWS not set yet */
1453 					memcpy(&tsval, &opt[2],
1454 					    sizeof(u_int32_t));
1455 					memcpy(&tsecr, &opt[6],
1456 					    sizeof(u_int32_t));
1457 					src->scrub->pfss_tsval0 = ntohl(tsval);
1458 					src->scrub->pfss_tsval = ntohl(tsval);
1459 					src->scrub->pfss_tsecr = ntohl(tsecr);
1460 					getmicrouptime(&src->scrub->pfss_last);
1461 				}
1462 				/* FALLTHROUGH */
1463 			default:
1464 				hlen -= MAX(opt[1], 2);
1465 				opt += MAX(opt[1], 2);
1466 				break;
1467 			}
1468 		}
1469 	}
1470 
1471 	return (0);
1472 }
1473 
1474 void
1475 pf_normalize_tcp_cleanup(struct pf_state *state)
1476 {
1477 	if (state->src.scrub)
1478 		pool_put(&pf_state_scrub_pl, state->src.scrub);
1479 	if (state->dst.scrub)
1480 		pool_put(&pf_state_scrub_pl, state->dst.scrub);
1481 
1482 	/* Someday... flush the TCP segment reassembly descriptors. */
1483 }
1484 
1485 int
1486 pf_normalize_tcp_stateful(struct mbuf *m, int off, struct pf_pdesc *pd,
1487     u_short *reason, struct tcphdr *th, struct pf_state *state,
1488     struct pf_state_peer *src, struct pf_state_peer *dst, int *writeback)
1489 {
1490 	struct timeval uptime;
1491 	u_int32_t tsval = 0, tsecr = 0;
1492 	u_int tsval_from_last;
1493 	u_int8_t hdr[60];
1494 	u_int8_t *opt;
1495 	int copyback = 0;
1496 	int got_ts = 0;
1497 
1498 	KASSERT(src->scrub || dst->scrub);
1499 
1500 	/*
1501 	 * Enforce the minimum TTL seen for this connection.  Negate a common
1502 	 * technique to evade an intrusion detection system and confuse
1503 	 * firewall state code.
1504 	 */
1505 	switch (pd->af) {
1506 #ifdef INET
1507 	case AF_INET: {
1508 		if (src->scrub) {
1509 			struct ip *h = mtod(m, struct ip *);
1510 			if (h->ip_ttl > src->scrub->pfss_ttl)
1511 				src->scrub->pfss_ttl = h->ip_ttl;
1512 			h->ip_ttl = src->scrub->pfss_ttl;
1513 		}
1514 		break;
1515 	}
1516 #endif /* INET */
1517 #ifdef INET6
1518 	case AF_INET6: {
1519 		if (src->scrub) {
1520 			struct ip6_hdr *h = mtod(m, struct ip6_hdr *);
1521 			if (h->ip6_hlim > src->scrub->pfss_ttl)
1522 				src->scrub->pfss_ttl = h->ip6_hlim;
1523 			h->ip6_hlim = src->scrub->pfss_ttl;
1524 		}
1525 		break;
1526 	}
1527 #endif /* INET6 */
1528 	}
1529 
1530 	if (th->th_off > (sizeof(struct tcphdr) >> 2) &&
1531 	    ((src->scrub && (src->scrub->pfss_flags & PFSS_TIMESTAMP)) ||
1532 	    (dst->scrub && (dst->scrub->pfss_flags & PFSS_TIMESTAMP))) &&
1533 	    pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) {
1534 		/* Diddle with TCP options */
1535 		int hlen;
1536 		opt = hdr + sizeof(struct tcphdr);
1537 		hlen = (th->th_off << 2) - sizeof(struct tcphdr);
1538 		while (hlen >= TCPOLEN_TIMESTAMP) {
1539 			switch (*opt) {
1540 			case TCPOPT_EOL:	/* FALLTHROUGH */
1541 			case TCPOPT_NOP:
1542 				opt++;
1543 				hlen--;
1544 				break;
1545 			case TCPOPT_TIMESTAMP:
1546 				/* Modulate the timestamps.  Can be used for
1547 				 * NAT detection, OS uptime determination or
1548 				 * reboot detection.
1549 				 */
1550 
1551 				if (got_ts) {
1552 					/* Huh?  Multiple timestamps!? */
1553 					if (pf_status.debug >= PF_DEBUG_MISC) {
1554 						DPFPRINTF(("multiple TS??"));
1555 						pf_print_state(state);
1556 						printf("\n");
1557 					}
1558 					REASON_SET(reason, PFRES_TS);
1559 					return (PF_DROP);
1560 				}
1561 				if (opt[1] >= TCPOLEN_TIMESTAMP) {
1562 					memcpy(&tsval, &opt[2],
1563 					    sizeof(u_int32_t));
1564 					if (tsval && src->scrub &&
1565 					    (src->scrub->pfss_flags &
1566 					    PFSS_TIMESTAMP)) {
1567 						tsval = ntohl(tsval);
1568 						pf_change_a(&opt[2],
1569 						    &th->th_sum,
1570 						    htonl(tsval +
1571 						    src->scrub->pfss_ts_mod),
1572 						    0);
1573 						copyback = 1;
1574 					}
1575 
1576 					/* Modulate TS reply iff valid (!0) */
1577 					memcpy(&tsecr, &opt[6],
1578 					    sizeof(u_int32_t));
1579 					if (tsecr && dst->scrub &&
1580 					    (dst->scrub->pfss_flags &
1581 					    PFSS_TIMESTAMP)) {
1582 						tsecr = ntohl(tsecr)
1583 						    - dst->scrub->pfss_ts_mod;
1584 						pf_change_a(&opt[6],
1585 						    &th->th_sum, htonl(tsecr),
1586 						    0);
1587 						copyback = 1;
1588 					}
1589 					got_ts = 1;
1590 				}
1591 				/* FALLTHROUGH */
1592 			default:
1593 				hlen -= MAX(opt[1], 2);
1594 				opt += MAX(opt[1], 2);
1595 				break;
1596 			}
1597 		}
1598 		if (copyback) {
1599 			/* Copyback the options, caller copys back header */
1600 			*writeback = 1;
1601 			m_copyback(m, off + sizeof(struct tcphdr),
1602 			    (th->th_off << 2) - sizeof(struct tcphdr), hdr +
1603 			    sizeof(struct tcphdr));
1604 		}
1605 	}
1606 
1607 
1608 	/*
1609 	 * Must invalidate PAWS checks on connections idle for too long.
1610 	 * The fastest allowed timestamp clock is 1ms.  That turns out to
1611 	 * be about 24 days before it wraps.  XXX Right now our lowerbound
1612 	 * TS echo check only works for the first 12 days of a connection
1613 	 * when the TS has exhausted half its 32bit space
1614 	 */
1615 #define TS_MAX_IDLE	(24*24*60*60)
1616 #define TS_MAX_CONN	(12*24*60*60)	/* XXX remove when better tsecr check */
1617 
1618 	getmicrouptime(&uptime);
1619 	if (src->scrub && (src->scrub->pfss_flags & PFSS_PAWS) &&
1620 	    (uptime.tv_sec - src->scrub->pfss_last.tv_sec > TS_MAX_IDLE ||
1621 	    time_second - state->creation > TS_MAX_CONN))  {
1622 		if (pf_status.debug >= PF_DEBUG_MISC) {
1623 			DPFPRINTF(("src idled out of PAWS\n"));
1624 			pf_print_state(state);
1625 			printf("\n");
1626 		}
1627 		src->scrub->pfss_flags = (src->scrub->pfss_flags & ~PFSS_PAWS)
1628 		    | PFSS_PAWS_IDLED;
1629 	}
1630 	if (dst->scrub && (dst->scrub->pfss_flags & PFSS_PAWS) &&
1631 	    uptime.tv_sec - dst->scrub->pfss_last.tv_sec > TS_MAX_IDLE) {
1632 		if (pf_status.debug >= PF_DEBUG_MISC) {
1633 			DPFPRINTF(("dst idled out of PAWS\n"));
1634 			pf_print_state(state);
1635 			printf("\n");
1636 		}
1637 		dst->scrub->pfss_flags = (dst->scrub->pfss_flags & ~PFSS_PAWS)
1638 		    | PFSS_PAWS_IDLED;
1639 	}
1640 
1641 	if (got_ts && src->scrub && dst->scrub &&
1642 	    (src->scrub->pfss_flags & PFSS_PAWS) &&
1643 	    (dst->scrub->pfss_flags & PFSS_PAWS)) {
1644 		/* Validate that the timestamps are "in-window".
1645 		 * RFC1323 describes TCP Timestamp options that allow
1646 		 * measurement of RTT (round trip time) and PAWS
1647 		 * (protection against wrapped sequence numbers).  PAWS
1648 		 * gives us a set of rules for rejecting packets on
1649 		 * long fat pipes (packets that were somehow delayed
1650 		 * in transit longer than the time it took to send the
1651 		 * full TCP sequence space of 4Gb).  We can use these
1652 		 * rules and infer a few others that will let us treat
1653 		 * the 32bit timestamp and the 32bit echoed timestamp
1654 		 * as sequence numbers to prevent a blind attacker from
1655 		 * inserting packets into a connection.
1656 		 *
1657 		 * RFC1323 tells us:
1658 		 *  - The timestamp on this packet must be greater than
1659 		 *    or equal to the last value echoed by the other
1660 		 *    endpoint.  The RFC says those will be discarded
1661 		 *    since it is a dup that has already been acked.
1662 		 *    This gives us a lowerbound on the timestamp.
1663 		 *        timestamp >= other last echoed timestamp
1664 		 *  - The timestamp will be less than or equal to
1665 		 *    the last timestamp plus the time between the
1666 		 *    last packet and now.  The RFC defines the max
1667 		 *    clock rate as 1ms.  We will allow clocks to be
1668 		 *    up to 10% fast and will allow a total difference
1669 		 *    or 30 seconds due to a route change.  And this
1670 		 *    gives us an upperbound on the timestamp.
1671 		 *        timestamp <= last timestamp + max ticks
1672 		 *    We have to be careful here.  Windows will send an
1673 		 *    initial timestamp of zero and then initialize it
1674 		 *    to a random value after the 3whs; presumably to
1675 		 *    avoid a DoS by having to call an expensive RNG
1676 		 *    during a SYN flood.  Proof MS has at least one
1677 		 *    good security geek.
1678 		 *
1679 		 *  - The TCP timestamp option must also echo the other
1680 		 *    endpoints timestamp.  The timestamp echoed is the
1681 		 *    one carried on the earliest unacknowledged segment
1682 		 *    on the left edge of the sequence window.  The RFC
1683 		 *    states that the host will reject any echoed
1684 		 *    timestamps that were larger than any ever sent.
1685 		 *    This gives us an upperbound on the TS echo.
1686 		 *        tescr <= largest_tsval
1687 		 *  - The lowerbound on the TS echo is a little more
1688 		 *    tricky to determine.  The other endpoint's echoed
1689 		 *    values will not decrease.  But there may be
1690 		 *    network conditions that re-order packets and
1691 		 *    cause our view of them to decrease.  For now the
1692 		 *    only lowerbound we can safely determine is that
1693 		 *    the TS echo will never be less than the original
1694 		 *    TS.  XXX There is probably a better lowerbound.
1695 		 *    Remove TS_MAX_CONN with better lowerbound check.
1696 		 *        tescr >= other original TS
1697 		 *
1698 		 * It is also important to note that the fastest
1699 		 * timestamp clock of 1ms will wrap its 32bit space in
1700 		 * 24 days.  So we just disable TS checking after 24
1701 		 * days of idle time.  We actually must use a 12d
1702 		 * connection limit until we can come up with a better
1703 		 * lowerbound to the TS echo check.
1704 		 */
1705 		struct timeval delta_ts;
1706 		int ts_fudge;
1707 
1708 
1709 		/*
1710 		 * PFTM_TS_DIFF is how many seconds of leeway to allow
1711 		 * a host's timestamp.  This can happen if the previous
1712 		 * packet got delayed in transit for much longer than
1713 		 * this packet.
1714 		 */
1715 		if ((ts_fudge = state->rule.ptr->timeout[PFTM_TS_DIFF]) == 0)
1716 			ts_fudge = pf_default_rule.timeout[PFTM_TS_DIFF];
1717 
1718 
1719 		/* Calculate max ticks since the last timestamp */
1720 #define TS_MAXFREQ	1100		/* RFC max TS freq of 1 kHz + 10% skew */
1721 #define TS_MICROSECS	1000000		/* microseconds per second */
1722 		timersub(&uptime, &src->scrub->pfss_last, &delta_ts);
1723 		tsval_from_last = (delta_ts.tv_sec + ts_fudge) * TS_MAXFREQ;
1724 		tsval_from_last += delta_ts.tv_usec / (TS_MICROSECS/TS_MAXFREQ);
1725 
1726 
1727 		if ((src->state >= TCPS_ESTABLISHED &&
1728 		    dst->state >= TCPS_ESTABLISHED) &&
1729 		    (SEQ_LT(tsval, dst->scrub->pfss_tsecr) ||
1730 		    SEQ_GT(tsval, src->scrub->pfss_tsval + tsval_from_last) ||
1731 		    (tsecr && (SEQ_GT(tsecr, dst->scrub->pfss_tsval) ||
1732 		    SEQ_LT(tsecr, dst->scrub->pfss_tsval0))))) {
1733 			/* Bad RFC1323 implementation or an insertion attack.
1734 			 *
1735 			 * - Solaris 2.6 and 2.7 are known to send another ACK
1736 			 *   after the FIN,FIN|ACK,ACK closing that carries
1737 			 *   an old timestamp.
1738 			 */
1739 
1740 			DPFPRINTF(("Timestamp failed %c%c%c%c\n",
1741 			    SEQ_LT(tsval, dst->scrub->pfss_tsecr) ? '0' : ' ',
1742 			    SEQ_GT(tsval, src->scrub->pfss_tsval +
1743 			    tsval_from_last) ? '1' : ' ',
1744 			    SEQ_GT(tsecr, dst->scrub->pfss_tsval) ? '2' : ' ',
1745 			    SEQ_LT(tsecr, dst->scrub->pfss_tsval0)? '3' : ' '));
1746 			DPFPRINTF((" tsval: %" PRIu32 "  tsecr: %" PRIu32
1747 			    "  +ticks: %" PRIu32 "  idle: %"PRIx64"s %ums\n",
1748 			    tsval, tsecr, tsval_from_last, delta_ts.tv_sec,
1749 			    delta_ts.tv_usec / 1000U));
1750 			DPFPRINTF((" src->tsval: %" PRIu32 "  tsecr: %" PRIu32
1751 			    "\n",
1752 			    src->scrub->pfss_tsval, src->scrub->pfss_tsecr));
1753 			DPFPRINTF((" dst->tsval: %" PRIu32 "  tsecr: %" PRIu32
1754 			    "  tsval0: %" PRIu32 "\n",
1755 			    dst->scrub->pfss_tsval,
1756 			    dst->scrub->pfss_tsecr, dst->scrub->pfss_tsval0));
1757 			if (pf_status.debug >= PF_DEBUG_MISC) {
1758 				pf_print_state(state);
1759 				pf_print_flags(th->th_flags);
1760 				printf("\n");
1761 			}
1762 			REASON_SET(reason, PFRES_TS);
1763 			return (PF_DROP);
1764 		}
1765 
1766 		/* XXX I'd really like to require tsecr but it's optional */
1767 
1768 	} else if (!got_ts && (th->th_flags & TH_RST) == 0 &&
1769 	    ((src->state == TCPS_ESTABLISHED && dst->state == TCPS_ESTABLISHED)
1770 	    || pd->p_len > 0 || (th->th_flags & TH_SYN)) &&
1771 	    src->scrub && dst->scrub &&
1772 	    (src->scrub->pfss_flags & PFSS_PAWS) &&
1773 	    (dst->scrub->pfss_flags & PFSS_PAWS)) {
1774 		/* Didn't send a timestamp.  Timestamps aren't really useful
1775 		 * when:
1776 		 *  - connection opening or closing (often not even sent).
1777 		 *    but we must not let an attacker to put a FIN on a
1778 		 *    data packet to sneak it through our ESTABLISHED check.
1779 		 *  - on a TCP reset.  RFC suggests not even looking at TS.
1780 		 *  - on an empty ACK.  The TS will not be echoed so it will
1781 		 *    probably not help keep the RTT calculation in sync and
1782 		 *    there isn't as much danger when the sequence numbers
1783 		 *    got wrapped.  So some stacks don't include TS on empty
1784 		 *    ACKs :-(
1785 		 *
1786 		 * To minimize the disruption to mostly RFC1323 conformant
1787 		 * stacks, we will only require timestamps on data packets.
1788 		 *
1789 		 * And what do ya know, we cannot require timestamps on data
1790 		 * packets.  There appear to be devices that do legitimate
1791 		 * TCP connection hijacking.  There are HTTP devices that allow
1792 		 * a 3whs (with timestamps) and then buffer the HTTP request.
1793 		 * If the intermediate device has the HTTP response cache, it
1794 		 * will spoof the response but not bother timestamping its
1795 		 * packets.  So we can look for the presence of a timestamp in
1796 		 * the first data packet and if there, require it in all future
1797 		 * packets.
1798 		 */
1799 
1800 		if (pd->p_len > 0 && (src->scrub->pfss_flags & PFSS_DATA_TS)) {
1801 			/*
1802 			 * Hey!  Someone tried to sneak a packet in.  Or the
1803 			 * stack changed its RFC1323 behavior?!?!
1804 			 */
1805 			if (pf_status.debug >= PF_DEBUG_MISC) {
1806 				DPFPRINTF(("Did not receive expected RFC1323 "
1807 				    "timestamp\n"));
1808 				pf_print_state(state);
1809 				pf_print_flags(th->th_flags);
1810 				printf("\n");
1811 			}
1812 			REASON_SET(reason, PFRES_TS);
1813 			return (PF_DROP);
1814 		}
1815 	}
1816 
1817 
1818 	/*
1819 	 * We will note if a host sends his data packets with or without
1820 	 * timestamps.  And require all data packets to contain a timestamp
1821 	 * if the first does.  PAWS implicitly requires that all data packets be
1822 	 * timestamped.  But I think there are middle-man devices that hijack
1823 	 * TCP streams immediately after the 3whs and don't timestamp their
1824 	 * packets (seen in a WWW accelerator or cache).
1825 	 */
1826 	if (pd->p_len > 0 && src->scrub && (src->scrub->pfss_flags &
1827 	    (PFSS_TIMESTAMP|PFSS_DATA_TS|PFSS_DATA_NOTS)) == PFSS_TIMESTAMP) {
1828 		if (got_ts)
1829 			src->scrub->pfss_flags |= PFSS_DATA_TS;
1830 		else {
1831 			src->scrub->pfss_flags |= PFSS_DATA_NOTS;
1832 			if (pf_status.debug >= PF_DEBUG_MISC && dst->scrub &&
1833 			    (dst->scrub->pfss_flags & PFSS_TIMESTAMP)) {
1834 				/* Don't warn if other host rejected RFC1323 */
1835 				DPFPRINTF(("Broken RFC1323 stack did not "
1836 				    "timestamp data packet. Disabled PAWS "
1837 				    "security.\n"));
1838 				pf_print_state(state);
1839 				pf_print_flags(th->th_flags);
1840 				printf("\n");
1841 			}
1842 		}
1843 	}
1844 
1845 
1846 	/*
1847 	 * Update PAWS values
1848 	 */
1849 	if (got_ts && src->scrub && PFSS_TIMESTAMP == (src->scrub->pfss_flags &
1850 	    (PFSS_PAWS_IDLED|PFSS_TIMESTAMP))) {
1851 		getmicrouptime(&src->scrub->pfss_last);
1852 		if (SEQ_GEQ(tsval, src->scrub->pfss_tsval) ||
1853 		    (src->scrub->pfss_flags & PFSS_PAWS) == 0)
1854 			src->scrub->pfss_tsval = tsval;
1855 
1856 		if (tsecr) {
1857 			if (SEQ_GEQ(tsecr, src->scrub->pfss_tsecr) ||
1858 			    (src->scrub->pfss_flags & PFSS_PAWS) == 0)
1859 				src->scrub->pfss_tsecr = tsecr;
1860 
1861 			if ((src->scrub->pfss_flags & PFSS_PAWS) == 0 &&
1862 			    (SEQ_LT(tsval, src->scrub->pfss_tsval0) ||
1863 			    src->scrub->pfss_tsval0 == 0)) {
1864 				/* tsval0 MUST be the lowest timestamp */
1865 				src->scrub->pfss_tsval0 = tsval;
1866 			}
1867 
1868 			/* Only fully initialized after a TS gets echoed */
1869 			if ((src->scrub->pfss_flags & PFSS_PAWS) == 0)
1870 				src->scrub->pfss_flags |= PFSS_PAWS;
1871 		}
1872 	}
1873 
1874 	/* I have a dream....  TCP segment reassembly.... */
1875 	return (0);
1876 }
1877 
1878 int
1879 pf_normalize_tcpopt(struct pf_rule *r, struct mbuf *m, struct tcphdr *th,
1880     int off)
1881 {
1882 	u_int16_t	*mss;
1883 	int		 thoff;
1884 	int		 opt, cnt, optlen = 0;
1885 	int		 rewrite = 0;
1886 	u_char		*optp;
1887 
1888 	thoff = th->th_off << 2;
1889 	cnt = thoff - sizeof(struct tcphdr);
1890 	optp = mtod(m, u_char *) + off + sizeof(struct tcphdr);
1891 
1892 	for (; cnt > 0; cnt -= optlen, optp += optlen) {
1893 		opt = optp[0];
1894 		if (opt == TCPOPT_EOL)
1895 			break;
1896 		if (opt == TCPOPT_NOP)
1897 			optlen = 1;
1898 		else {
1899 			if (cnt < 2)
1900 				break;
1901 			optlen = optp[1];
1902 			if (optlen < 2 || optlen > cnt)
1903 				break;
1904 		}
1905 		switch (opt) {
1906 		case TCPOPT_MAXSEG:
1907 			mss = (u_int16_t *)(optp + 2);
1908 			if ((ntohs(*mss)) > r->max_mss) {
1909 				th->th_sum = pf_cksum_fixup(th->th_sum,
1910 				    *mss, htons(r->max_mss), 0);
1911 				*mss = htons(r->max_mss);
1912 				rewrite = 1;
1913 			}
1914 			break;
1915 		default:
1916 			break;
1917 		}
1918 	}
1919 
1920 	return (rewrite);
1921 }
1922