xref: /openbsd-src/sys/net/pf_norm.c (revision d13be5d47e4149db2549a9828e244d59dbc43f15)
1 /*	$OpenBSD: pf_norm.c,v 1.140 2011/07/18 21:03:10 mikeb Exp $ */
2 
3 /*
4  * Copyright 2001 Niels Provos <provos@citi.umich.edu>
5  * Copyright 2009 Henning Brauer <henning@openbsd.org>
6  * Copyright 2011 Alexander Bluhm <bluhm@openbsd.org>
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
19  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
20  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
21  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
22  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
23  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
24  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
25  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
26  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
27  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
28  */
29 
30 #include "pflog.h"
31 
32 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/mbuf.h>
35 #include <sys/filio.h>
36 #include <sys/fcntl.h>
37 #include <sys/socket.h>
38 #include <sys/kernel.h>
39 #include <sys/time.h>
40 #include <sys/pool.h>
41 #include <sys/syslog.h>
42 
43 #include <dev/rndvar.h>
44 #include <net/if.h>
45 #include <net/if_types.h>
46 #include <net/bpf.h>
47 #include <net/route.h>
48 #include <net/if_pflog.h>
49 
50 #include <netinet/in.h>
51 #include <netinet/in_var.h>
52 #include <netinet/in_systm.h>
53 #include <netinet/ip.h>
54 #include <netinet/ip_var.h>
55 #include <netinet/tcp.h>
56 #include <netinet/tcp_seq.h>
57 #include <netinet/udp.h>
58 #include <netinet/ip_icmp.h>
59 
60 #ifdef INET6
61 #include <netinet/ip6.h>
62 #include <netinet6/ip6_var.h>
63 #endif /* INET6 */
64 
65 #include <net/pfvar.h>
66 
67 struct pf_frent {
68 	TAILQ_ENTRY(pf_frent) fr_next;
69 	struct mbuf	*fe_m;
70 	u_int16_t	 fe_hdrlen;	/* ipv4 header lenght with ip options
71 					   ipv6, extension, fragment header */
72 	u_int16_t	 fe_extoff;	/* last extension header offset or 0 */
73 	u_int16_t	 fe_len;	/* fragment length */
74 	u_int16_t	 fe_off;	/* fragment offset */
75 	u_int16_t	 fe_mff;	/* more fragment flag */
76 };
77 
78 /* keep synced with struct pf_fragment, used in RB_FIND */
79 struct pf_fragment_cmp {
80 	struct pf_addr	fr_src;
81 	struct pf_addr	fr_dst;
82 	u_int32_t	fr_id;
83 	sa_family_t	fr_af;
84 	u_int8_t	fr_proto;
85 	u_int8_t	fr_direction;
86 };
87 
88 struct pf_fragment {
89 	struct pf_addr	fr_src;		/* ip source address */
90 	struct pf_addr	fr_dst;		/* ip destination address */
91 	u_int32_t	fr_id;		/* fragment id for reassemble */
92 	sa_family_t	fr_af;		/* address family */
93 	u_int8_t	fr_proto;	/* protocol of this fragment */
94 	u_int8_t	fr_direction;	/* pf packet direction */
95 
96 	RB_ENTRY(pf_fragment) fr_entry;
97 	TAILQ_ENTRY(pf_fragment) frag_next;
98 	u_int32_t	fr_timeout;
99 	u_int16_t	fr_maxlen;	/* maximum length of single fragment */
100 	TAILQ_HEAD(pf_fragq, pf_frent) fr_queue;
101 };
102 
103 struct pf_fragment_tag {
104 	u_int16_t	 ft_hdrlen;	/* header lenght of reassembled pkt */
105 	u_int16_t	 ft_extoff;	/* last extension header offset or 0 */
106 	u_int16_t	 ft_maxlen;	/* maximum fragment payload length */
107 };
108 
109 TAILQ_HEAD(pf_fragqueue, pf_fragment)	pf_fragqueue;
110 
111 static __inline int	 pf_frag_compare(struct pf_fragment *,
112 			    struct pf_fragment *);
113 RB_HEAD(pf_frag_tree, pf_fragment)	pf_frag_tree, pf_cache_tree;
114 RB_PROTOTYPE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
115 RB_GENERATE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
116 
117 /* Private prototypes */
118 void			 pf_remove_fragment(struct pf_fragment *);
119 void			 pf_flush_fragments(void);
120 void			 pf_free_fragment(struct pf_fragment *);
121 struct pf_fragment	*pf_find_fragment(struct pf_fragment_cmp *,
122 			    struct pf_frag_tree *);
123 struct pf_frent		*pf_create_fragment(u_short *);
124 struct pf_fragment	*pf_fillup_fragment(struct pf_fragment_cmp *,
125 			    struct pf_frent *, u_short *);
126 int			 pf_isfull_fragment(struct pf_fragment *);
127 struct mbuf		*pf_join_fragment(struct pf_fragment *);
128 int			 pf_reassemble(struct mbuf **, int, u_short *);
129 #ifdef INET6
130 int			 pf_reassemble6(struct mbuf **, struct ip6_frag *,
131 			    u_int16_t, u_int16_t, int, u_short *);
132 #endif
133 
134 /* Globals */
135 struct pool		 pf_frent_pl, pf_frag_pl;
136 struct pool		 pf_state_scrub_pl;
137 int			 pf_nfrents;
138 
139 void
140 pf_normalize_init(void)
141 {
142 	pool_init(&pf_frent_pl, sizeof(struct pf_frent), 0, 0, 0, "pffrent",
143 	    NULL);
144 	pool_init(&pf_frag_pl, sizeof(struct pf_fragment), 0, 0, 0, "pffrag",
145 	    NULL);
146 	pool_init(&pf_state_scrub_pl, sizeof(struct pf_state_scrub), 0, 0, 0,
147 	    "pfstscr", NULL);
148 
149 	pool_sethiwat(&pf_frag_pl, PFFRAG_FRAG_HIWAT);
150 	pool_sethardlimit(&pf_frent_pl, PFFRAG_FRENT_HIWAT, NULL, 0);
151 
152 	TAILQ_INIT(&pf_fragqueue);
153 }
154 
155 static __inline int
156 pf_frag_compare(struct pf_fragment *a, struct pf_fragment *b)
157 {
158 	int	diff;
159 
160 	if ((diff = a->fr_id - b->fr_id) != 0)
161 		return (diff);
162 	if ((diff = a->fr_proto - b->fr_proto) != 0)
163 		return (diff);
164 	if ((diff = a->fr_af - b->fr_af) != 0)
165 		return (diff);
166 	if ((diff = pf_addr_compare(&a->fr_src, &b->fr_src, a->fr_af)) != 0)
167 		return (diff);
168 	if ((diff = pf_addr_compare(&a->fr_dst, &b->fr_dst, a->fr_af)) != 0)
169 		return (diff);
170 	return (0);
171 }
172 
173 void
174 pf_purge_expired_fragments(void)
175 {
176 	struct pf_fragment	*frag;
177 	u_int32_t		 expire = time_second -
178 				    pf_default_rule.timeout[PFTM_FRAG];
179 
180 	while ((frag = TAILQ_LAST(&pf_fragqueue, pf_fragqueue)) != NULL) {
181 		if (frag->fr_timeout > expire)
182 			break;
183 
184 		DPFPRINTF(LOG_NOTICE, "expiring %d(%p)", frag->fr_id, frag);
185 		pf_free_fragment(frag);
186 	}
187 }
188 
189 /*
190  * Try to flush old fragments to make space for new ones
191  */
192 
193 void
194 pf_flush_fragments(void)
195 {
196 	struct pf_fragment	*frag;
197 	int			 goal;
198 
199 	goal = pf_nfrents * 9 / 10;
200 	DPFPRINTF(LOG_NOTICE, "trying to free > %d frents",
201 	    pf_nfrents - goal);
202 	while (goal < pf_nfrents) {
203 		frag = TAILQ_LAST(&pf_fragqueue, pf_fragqueue);
204 		if (frag == NULL)
205 			break;
206 		pf_free_fragment(frag);
207 	}
208 }
209 
210 /* Frees the fragments and all associated entries */
211 
212 void
213 pf_free_fragment(struct pf_fragment *frag)
214 {
215 	struct pf_frent		*frent;
216 
217 	/* Free all fragments */
218 	for (frent = TAILQ_FIRST(&frag->fr_queue); frent;
219 	    frent = TAILQ_FIRST(&frag->fr_queue)) {
220 		TAILQ_REMOVE(&frag->fr_queue, frent, fr_next);
221 
222 		m_freem(frent->fe_m);
223 		pool_put(&pf_frent_pl, frent);
224 		pf_nfrents--;
225 	}
226 
227 	pf_remove_fragment(frag);
228 }
229 
230 struct pf_fragment *
231 pf_find_fragment(struct pf_fragment_cmp *key, struct pf_frag_tree *tree)
232 {
233 	struct pf_fragment	*frag;
234 
235 	frag = RB_FIND(pf_frag_tree, tree, (struct pf_fragment *)key);
236 	if (frag != NULL) {
237 		/* XXX Are we sure we want to update the timeout? */
238 		frag->fr_timeout = time_second;
239 		TAILQ_REMOVE(&pf_fragqueue, frag, frag_next);
240 		TAILQ_INSERT_HEAD(&pf_fragqueue, frag, frag_next);
241 	}
242 
243 	return (frag);
244 }
245 
246 /* Removes a fragment from the fragment queue and frees the fragment */
247 
248 void
249 pf_remove_fragment(struct pf_fragment *frag)
250 {
251 	RB_REMOVE(pf_frag_tree, &pf_frag_tree, frag);
252 	TAILQ_REMOVE(&pf_fragqueue, frag, frag_next);
253 	pool_put(&pf_frag_pl, frag);
254 }
255 
256 struct pf_frent *
257 pf_create_fragment(u_short *reason)
258 {
259 	struct pf_frent	*frent;
260 
261 	frent = pool_get(&pf_frent_pl, PR_NOWAIT);
262 	if (frent == NULL) {
263 		pf_flush_fragments();
264 		frent = pool_get(&pf_frent_pl, PR_NOWAIT);
265 		if (frent == NULL) {
266 			REASON_SET(reason, PFRES_MEMORY);
267 			return (NULL);
268 		}
269 	}
270 	pf_nfrents++;
271 
272 	return (frent);
273 }
274 
275 struct pf_fragment *
276 pf_fillup_fragment(struct pf_fragment_cmp *key, struct pf_frent *frent,
277     u_short *reason)
278 {
279 	struct pf_frent		*after, *next, *prev;
280 	struct pf_fragment	*frag;
281 	u_int16_t		 total;
282 
283 	/* No empty fragments */
284 	if (frent->fe_len == 0) {
285 		DPFPRINTF(LOG_NOTICE, "bad fragment: len 0");
286 		goto bad_fragment;
287 	}
288 
289 	/* All fragments are 8 byte aligned */
290 	if (frent->fe_mff && (frent->fe_len & 0x7)) {
291 		DPFPRINTF(LOG_NOTICE, "bad fragment: mff and len %d",
292 		    frent->fe_len);
293 		goto bad_fragment;
294 	}
295 
296 	/* Respect maximum length, IP_MAXPACKET == IPV6_MAXPACKET */
297 	if (frent->fe_off + frent->fe_len > IP_MAXPACKET) {
298 		DPFPRINTF(LOG_NOTICE, "bad fragment: max packet %d",
299 		    frent->fe_off + frent->fe_len);
300 		goto bad_fragment;
301 	}
302 
303 	DPFPRINTF(LOG_NOTICE, key->fr_af == AF_INET ?
304 	    "reass frag %d @ %d-%d" : "reass frag %#08x @ %d-%d",
305 	    key->fr_id, frent->fe_off, frent->fe_off + frent->fe_len);
306 
307 	/* Fully buffer all of the fragments in this fragment queue */
308 	frag = pf_find_fragment(key, &pf_frag_tree);
309 
310 	/* Create a new reassembly queue for this packet */
311 	if (frag == NULL) {
312 		frag = pool_get(&pf_frag_pl, PR_NOWAIT);
313 		if (frag == NULL) {
314 			pf_flush_fragments();
315 			frag = pool_get(&pf_frag_pl, PR_NOWAIT);
316 			if (frag == NULL) {
317 				REASON_SET(reason, PFRES_MEMORY);
318 				goto drop_fragment;
319 			}
320 		}
321 
322 		*(struct pf_fragment_cmp *)frag = *key;
323 		frag->fr_timeout = time_second;
324 		frag->fr_maxlen = frent->fe_len;
325 		TAILQ_INIT(&frag->fr_queue);
326 
327 		RB_INSERT(pf_frag_tree, &pf_frag_tree, frag);
328 		TAILQ_INSERT_HEAD(&pf_fragqueue, frag, frag_next);
329 
330 		/* We do not have a previous fragment */
331 		TAILQ_INSERT_HEAD(&frag->fr_queue, frent, fr_next);
332 
333 		return (frag);
334 	}
335 
336 	KASSERT(!TAILQ_EMPTY(&frag->fr_queue));
337 
338 	/* Remember maximum fragment len for refragmentation */
339 	if (frent->fe_len > frag->fr_maxlen)
340 		frag->fr_maxlen = frent->fe_len;
341 
342 	/* Maximum data we have seen already */
343 	total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
344 	    TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
345 
346 	/* Non terminal fragments must have more fragments flag */
347 	if (frent->fe_off + frent->fe_len < total && !frent->fe_mff)
348 		goto bad_fragment;
349 
350 	/* Check if we saw the last fragment already */
351 	if (!TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_mff) {
352 		if (frent->fe_off + frent->fe_len > total ||
353 		    (frent->fe_off + frent->fe_len == total && frent->fe_mff))
354 			goto bad_fragment;
355 	} else {
356 		if (frent->fe_off + frent->fe_len == total && !frent->fe_mff)
357 			goto bad_fragment;
358 	}
359 
360 	/* Find a fragment after the current one */
361 	prev = NULL;
362 	TAILQ_FOREACH(after, &frag->fr_queue, fr_next) {
363 		if (after->fe_off > frent->fe_off)
364 			break;
365 		prev = after;
366 	}
367 
368 	KASSERT(prev != NULL || after != NULL);
369 
370 	if (prev != NULL && prev->fe_off + prev->fe_len > frent->fe_off) {
371 		u_int16_t	precut;
372 
373 		precut = prev->fe_off + prev->fe_len - frent->fe_off;
374 		if (precut >= frent->fe_len)
375 			goto bad_fragment;
376 		DPFPRINTF(LOG_NOTICE, "overlap -%d", precut);
377 		m_adj(frent->fe_m, precut);
378 		frent->fe_off += precut;
379 		frent->fe_len -= precut;
380 	}
381 
382 	for (; after != NULL && frent->fe_off + frent->fe_len > after->fe_off;
383 	    after = next)
384 	{
385 		u_int16_t	aftercut;
386 
387 		aftercut = frent->fe_off + frent->fe_len - after->fe_off;
388 		DPFPRINTF(LOG_NOTICE, "adjust overlap %d", aftercut);
389 		if (aftercut < after->fe_len) {
390 			m_adj(after->fe_m, aftercut);
391 			after->fe_off += aftercut;
392 			after->fe_len -= aftercut;
393 			break;
394 		}
395 
396 		/* This fragment is completely overlapped, lose it */
397 		next = TAILQ_NEXT(after, fr_next);
398 		m_freem(after->fe_m);
399 		TAILQ_REMOVE(&frag->fr_queue, after, fr_next);
400 		pool_put(&pf_frent_pl, after);
401 		pf_nfrents--;
402 	}
403 
404 	if (prev == NULL)
405 		TAILQ_INSERT_HEAD(&frag->fr_queue, frent, fr_next);
406 	else
407 		TAILQ_INSERT_AFTER(&frag->fr_queue, prev, frent, fr_next);
408 
409 	return (frag);
410 
411  bad_fragment:
412 	REASON_SET(reason, PFRES_FRAG);
413  drop_fragment:
414 	pool_put(&pf_frent_pl, frent);
415 	pf_nfrents--;
416 	return (NULL);
417 }
418 
419 int
420 pf_isfull_fragment(struct pf_fragment *frag)
421 {
422 	struct pf_frent		*frent, *next;
423 	u_int16_t		 off, total;
424 
425 	/* Check if we are completely reassembled */
426 	if (TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_mff)
427 		return (0);
428 
429 	/* Maximum data we have seen already */
430 	total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
431 	    TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
432 
433 	/* Check if we have all the data */
434 	off = 0;
435 	for (frent = TAILQ_FIRST(&frag->fr_queue); frent; frent = next) {
436 		next = TAILQ_NEXT(frent, fr_next);
437 
438 		off += frent->fe_len;
439 		if (off < total && (next == NULL || next->fe_off != off)) {
440 			DPFPRINTF(LOG_NOTICE,
441 			    "missing fragment at %d, next %d, total %d",
442 			    off, next == NULL ? -1 : next->fe_off, total);
443 			return (0);
444 		}
445 	}
446 	DPFPRINTF(LOG_NOTICE, "%d < %d?", off, total);
447 	if (off < total)
448 		return (0);
449 	KASSERT(off == total);
450 
451 	return (1);
452 }
453 
454 struct mbuf *
455 pf_join_fragment(struct pf_fragment *frag)
456 {
457 	struct mbuf		*m, *m2;
458 	struct pf_frent		*frent, *next;
459 
460 	frent = TAILQ_FIRST(&frag->fr_queue);
461 	next = TAILQ_NEXT(frent, fr_next);
462 
463 	/* Magic from ip_input */
464 	m = frent->fe_m;
465 	m2 = m->m_next;
466 	m->m_next = NULL;
467 	m_cat(m, m2);
468 	pool_put(&pf_frent_pl, frent);
469 	pf_nfrents--;
470 	for (frent = next; frent != NULL; frent = next) {
471 		next = TAILQ_NEXT(frent, fr_next);
472 
473 		m2 = frent->fe_m;
474 		/* Strip off ip header */
475 		m_adj(m2, frent->fe_hdrlen);
476 		pool_put(&pf_frent_pl, frent);
477 		pf_nfrents--;
478 		m_cat(m, m2);
479 	}
480 
481 	/* Remove from fragment queue */
482 	pf_remove_fragment(frag);
483 
484 	return (m);
485 }
486 
487 int
488 pf_reassemble(struct mbuf **m0, int dir, u_short *reason)
489 {
490 	struct mbuf		*m = *m0;
491 	struct ip		*ip = mtod(m, struct ip *);
492 	struct pf_frent		*frent;
493 	struct pf_fragment	*frag;
494 	struct pf_fragment_cmp	 key;
495 	u_int16_t		 total, hdrlen;
496 
497 	/* Get an entry for the fragment queue */
498 	if ((frent = pf_create_fragment(reason)) == NULL)
499 		return (PF_DROP);
500 
501 	frent->fe_m = m;
502 	frent->fe_hdrlen = ip->ip_hl << 2;
503 	frent->fe_extoff = 0;
504 	frent->fe_len = ntohs(ip->ip_len) - (ip->ip_hl << 2);
505 	frent->fe_off = (ntohs(ip->ip_off) & IP_OFFMASK) << 3;
506 	frent->fe_mff = ntohs(ip->ip_off) & IP_MF;
507 
508 	key.fr_src.v4 = ip->ip_src;
509 	key.fr_dst.v4 = ip->ip_dst;
510 	key.fr_af = AF_INET;
511 	key.fr_proto = ip->ip_p;
512 	key.fr_id = ip->ip_id;
513 	key.fr_direction = dir;
514 
515 	if ((frag = pf_fillup_fragment(&key, frent, reason)) == NULL)
516 		return (PF_DROP);
517 
518 	/* The mbuf is part of the fragment entry, no direct free or access */
519 	m = *m0 = NULL;
520 
521 	if (!pf_isfull_fragment(frag))
522 		return (PF_PASS);  /* drop because *m0 is NULL, no error */
523 
524 	/* We have all the data */
525 	frent = TAILQ_FIRST(&frag->fr_queue);
526 	KASSERT(frent != NULL);
527 	total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
528 	    TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
529 	hdrlen = frent->fe_hdrlen;
530 
531 	m = *m0 = pf_join_fragment(frag);
532 	frag = NULL;
533 
534 	if (m->m_flags & M_PKTHDR) {
535 		int plen = 0;
536 		for (m = *m0; m; m = m->m_next)
537 			plen += m->m_len;
538 		m = *m0;
539 		m->m_pkthdr.len = plen;
540 	}
541 
542 	ip = mtod(m, struct ip *);
543 	ip->ip_len = htons(hdrlen + total);
544 	ip->ip_off &= ~(IP_MF|IP_OFFMASK);
545 
546 	if (hdrlen + total > IP_MAXPACKET) {
547 		DPFPRINTF(LOG_NOTICE, "drop: too big: %d", total);
548 		ip->ip_len = 0;
549 		REASON_SET(reason, PFRES_SHORT);
550 		/* PF_DROP requires a valid mbuf *m0 in pf_test() */
551 		return (PF_DROP);
552 	}
553 
554 	DPFPRINTF(LOG_NOTICE, "complete: %p(%d)", m, ntohs(ip->ip_len));
555 	return (PF_PASS);
556 }
557 
558 #ifdef INET6
559 int
560 pf_reassemble6(struct mbuf **m0, struct ip6_frag *fraghdr,
561     u_int16_t hdrlen, u_int16_t extoff, int dir, u_short *reason)
562 {
563 	struct mbuf		*m = *m0;
564 	struct ip6_hdr		*ip6 = mtod(m, struct ip6_hdr *);
565 	struct m_tag		*mtag;
566 	struct pf_fragment_tag	*ftag;
567 	struct pf_frent		*frent;
568 	struct pf_fragment	*frag;
569 	struct pf_fragment_cmp	 key;
570 	int			 off;
571 	u_int16_t		 total, maxlen;
572 	u_int8_t		 proto;
573 
574 	/* Get an entry for the fragment queue */
575 	if ((frent = pf_create_fragment(reason)) == NULL)
576 		return (PF_DROP);
577 
578 	frent->fe_m = m;
579 	frent->fe_hdrlen = hdrlen;
580 	frent->fe_extoff = extoff;
581 	frent->fe_len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - hdrlen;
582 	frent->fe_off = ntohs(fraghdr->ip6f_offlg & IP6F_OFF_MASK);
583 	frent->fe_mff = fraghdr->ip6f_offlg & IP6F_MORE_FRAG;
584 
585 	key.fr_src.v6 = ip6->ip6_src;
586 	key.fr_dst.v6 = ip6->ip6_dst;
587 	key.fr_af = AF_INET6;
588 	/* Only the first fragment's protocol is relevant */
589 	key.fr_proto = 0;
590 	key.fr_id = fraghdr->ip6f_ident;
591 	key.fr_direction = dir;
592 
593 	if ((frag = pf_fillup_fragment(&key, frent, reason)) == NULL)
594 		return (PF_DROP);
595 
596 	/* The mbuf is part of the fragment entry, no direct free or access */
597 	m = *m0 = NULL;
598 
599 	if (!pf_isfull_fragment(frag))
600 		return (PF_PASS);  /* drop because *m0 is NULL, no error */
601 
602 	/* We have all the data */
603 	extoff = frent->fe_extoff;
604 	maxlen = frag->fr_maxlen;
605 	frent = TAILQ_FIRST(&frag->fr_queue);
606 	KASSERT(frent != NULL);
607 	total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
608 	    TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
609 	hdrlen = frent->fe_hdrlen - sizeof(struct ip6_frag);
610 
611 	m = *m0 = pf_join_fragment(frag);
612 	frag = NULL;
613 
614 	/* Take protocol from first fragment header */
615 	if ((m = m_getptr(m, hdrlen + offsetof(struct ip6_frag, ip6f_nxt),
616 	    &off)) == NULL)
617 		panic("pf_reassemble6: short mbuf chain");
618 	proto = *(mtod(m, caddr_t) + off);
619 	m = *m0;
620 
621 	/* Delete frag6 header */
622 	if (frag6_deletefraghdr(m, hdrlen) != 0)
623 		goto fail;
624 
625 	if (m->m_flags & M_PKTHDR) {
626 		int plen = 0;
627 		for (m = *m0; m; m = m->m_next)
628 			plen += m->m_len;
629 		m = *m0;
630 		m->m_pkthdr.len = plen;
631 	}
632 
633 	if ((mtag = m_tag_get(PACKET_TAG_PF_REASSEMBLED, sizeof(struct
634 	    pf_fragment_tag), M_NOWAIT)) == NULL)
635 		goto fail;
636 	ftag = (struct pf_fragment_tag *)(mtag + 1);
637 	ftag->ft_hdrlen = hdrlen;
638 	ftag->ft_extoff = extoff;
639 	ftag->ft_maxlen = maxlen;
640 	m_tag_prepend(m, mtag);
641 
642 	ip6 = mtod(m, struct ip6_hdr *);
643 	ip6->ip6_plen = htons(hdrlen - sizeof(struct ip6_hdr) + total);
644 	if (extoff) {
645 		/* Write protocol into next field of last extension header */
646 		if ((m = m_getptr(m, extoff + offsetof(struct ip6_ext,
647 		    ip6e_nxt), &off)) == NULL)
648 			panic("pf_reassemble6: short mbuf chain");
649 		*(mtod(m, caddr_t) + off) = proto;
650 		m = *m0;
651 	} else
652 		ip6->ip6_nxt = proto;
653 
654 	if (hdrlen - sizeof(struct ip6_hdr) + total > IPV6_MAXPACKET) {
655 		DPFPRINTF(LOG_NOTICE, "drop: too big: %d", total);
656 		ip6->ip6_plen = 0;
657 		REASON_SET(reason, PFRES_SHORT);
658 		/* PF_DROP requires a valid mbuf *m0 in pf_test6() */
659 		return (PF_DROP);
660 	}
661 
662 	DPFPRINTF(LOG_NOTICE, "complete: %p(%d)", m, ntohs(ip6->ip6_plen));
663 	return (PF_PASS);
664 
665  fail:
666 	REASON_SET(reason, PFRES_MEMORY);
667 	/* PF_DROP requires a valid mbuf *m0 in pf_test6(), will free later */
668 	return (PF_DROP);
669 }
670 
671 int
672 pf_refragment6(struct mbuf **m0, struct m_tag *mtag, int dir)
673 {
674 	struct mbuf		*m = *m0, *t;
675 	struct pf_fragment_tag	*ftag = (struct pf_fragment_tag *)(mtag + 1);
676 	u_int32_t		 mtu;
677 	u_int16_t		 hdrlen, extoff, maxlen;
678 	u_int8_t		 proto;
679 	int			 error, action;
680 
681 	hdrlen = ftag->ft_hdrlen;
682 	extoff = ftag->ft_extoff;
683 	maxlen = ftag->ft_maxlen;
684 	m_tag_delete(m, mtag);
685 	mtag = NULL;
686 	ftag = NULL;
687 
688 	if (extoff) {
689 		int off;
690 
691 		/* Use protocol from next field of last extension header */
692 		if ((m = m_getptr(m, extoff + offsetof(struct ip6_ext,
693 		    ip6e_nxt), &off)) == NULL)
694 			panic("pf_refragment6: short mbuf chain");
695 		proto = *(mtod(m, caddr_t) + off);
696 		*(mtod(m, caddr_t) + off) = IPPROTO_FRAGMENT;
697 		m = *m0;
698 	} else {
699 		struct ip6_hdr *hdr;
700 
701 		hdr = mtod(m, struct ip6_hdr *);
702 		proto = hdr->ip6_nxt;
703 		hdr->ip6_nxt = IPPROTO_FRAGMENT;
704 	}
705 
706 	/*
707 	 * Maxlen may be less than 8 iff there was only a single
708 	 * fragment.  As it was fragmented before, add a fragment
709 	 * header also for a single fragment.  If total or maxlen
710 	 * is less than 8, ip6_fragment() will return EMSGSIZE and
711 	 * we drop the packet.
712 	 */
713 
714 	mtu = hdrlen + sizeof(struct ip6_frag) + maxlen;
715 	error = ip6_fragment(m, hdrlen, proto, mtu);
716 
717 	m = (*m0)->m_nextpkt;
718 	(*m0)->m_nextpkt = NULL;
719 	if (error == 0) {
720 		/* The first mbuf contains the unfragmented packet */
721 		m_freem(*m0);
722 		*m0 = NULL;
723 		action = PF_PASS;
724 	} else {
725 		/* Drop expects an mbuf to free */
726 		DPFPRINTF(LOG_NOTICE, "refragment error %d", error);
727 		action = PF_DROP;
728 	}
729 	for (t = m; m; m = t) {
730 		t = m->m_nextpkt;
731 		m->m_nextpkt = NULL;
732 		m->m_pkthdr.pf.flags |= PF_TAG_REFRAGMENTED;
733 		if (error == 0)
734 			ip6_forward(m, 0);
735 		else
736 			m_freem(m);
737 	}
738 
739 	return (action);
740 }
741 #endif /* INET6 */
742 
743 int
744 pf_normalize_ip(struct mbuf **m0, int dir, u_short *reason)
745 {
746 	struct mbuf		*m = *m0;
747 	struct ip		*h = mtod(m, struct ip *);
748 	u_int16_t		 fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
749 	u_int16_t		 mff = (ntohs(h->ip_off) & IP_MF);
750 
751 	/* We will need other tests here */
752 	if (!fragoff && !mff)
753 		goto no_fragment;
754 
755 	/* Clear IP_DF if we're in no-df mode */
756 	if (pf_status.reass & PF_REASS_NODF && h->ip_off & htons(IP_DF))
757 		h->ip_off &= htons(~IP_DF);
758 
759 	/* We're dealing with a fragment now. Don't allow fragments
760 	 * with IP_DF to enter the cache. If the flag was cleared by
761 	 * no-df above, fine. Otherwise drop it.
762 	 */
763 	if (h->ip_off & htons(IP_DF)) {
764 		DPFPRINTF(LOG_NOTICE, "bad fragment: IP_DF");
765 		REASON_SET(reason, PFRES_FRAG);
766 		return (PF_DROP);
767 	}
768 
769 	if (!pf_status.reass)
770 		return (PF_PASS);	/* no reassembly */
771 
772 	/* Returns PF_DROP or *m0 is NULL or completely reassembled mbuf */
773 	if (pf_reassemble(m0, dir, reason) != PF_PASS)
774 		return (PF_DROP);
775 	m = *m0;
776 	if (m == NULL)
777 		return (PF_PASS);  /* packet has been reassembled, no error */
778 
779 	h = mtod(m, struct ip *);
780 
781  no_fragment:
782 	/* At this point, only IP_DF is allowed in ip_off */
783 	if (h->ip_off & ~htons(IP_DF))
784 		h->ip_off &= htons(IP_DF);
785 
786 	return (PF_PASS);
787 }
788 
789 #ifdef INET6
790 int
791 pf_normalize_ip6(struct mbuf **m0, int dir, int off, int extoff,
792     u_short *reason)
793 {
794 	struct mbuf		*m = *m0;
795 	struct ip6_frag		 frag;
796 
797 	if (!pf_pull_hdr(m, off, &frag, sizeof(frag), NULL, reason, AF_INET6))
798 		return (PF_DROP);
799 	/* offset now points to data portion */
800 	off += sizeof(frag);
801 
802 	/* Returns PF_DROP or *m0 is NULL or completely reassembled mbuf */
803 	if (pf_reassemble6(m0, &frag, off, extoff, dir, reason) != PF_PASS)
804 		return (PF_DROP);
805 
806 	return (PF_PASS);
807 }
808 #endif /* INET6 */
809 
810 int
811 pf_normalize_tcp(int dir, struct mbuf *m, int off, struct pf_pdesc *pd)
812 {
813 	struct tcphdr	*th = pd->hdr.tcp;
814 	u_short		 reason;
815 	u_int8_t	 flags;
816 	u_int		 rewrite = 0;
817 
818 	flags = th->th_flags;
819 	if (flags & TH_SYN) {
820 		/* Illegal packet */
821 		if (flags & TH_RST)
822 			goto tcp_drop;
823 
824 		if (flags & TH_FIN)
825 			flags &= ~TH_FIN;
826 	} else {
827 		/* Illegal packet */
828 		if (!(flags & (TH_ACK|TH_RST)))
829 			goto tcp_drop;
830 	}
831 
832 	if (!(flags & TH_ACK)) {
833 		/* These flags are only valid if ACK is set */
834 		if ((flags & TH_FIN) || (flags & TH_PUSH) || (flags & TH_URG))
835 			goto tcp_drop;
836 	}
837 
838 	/* Check for illegal header length */
839 	if (th->th_off < (sizeof(struct tcphdr) >> 2))
840 		goto tcp_drop;
841 
842 	/* If flags changed, or reserved data set, then adjust */
843 	if (flags != th->th_flags || th->th_x2 != 0) {
844 		u_int16_t	ov, nv;
845 
846 		ov = *(u_int16_t *)(&th->th_ack + 1);
847 		th->th_flags = flags;
848 		th->th_x2 = 0;
849 		nv = *(u_int16_t *)(&th->th_ack + 1);
850 
851 		th->th_sum = pf_cksum_fixup(th->th_sum, ov, nv, 0);
852 		rewrite = 1;
853 	}
854 
855 	/* Remove urgent pointer, if TH_URG is not set */
856 	if (!(flags & TH_URG) && th->th_urp) {
857 		th->th_sum = pf_cksum_fixup(th->th_sum, th->th_urp, 0, 0);
858 		th->th_urp = 0;
859 		rewrite = 1;
860 	}
861 
862 	/* copy back packet headers if we sanitized */
863 	if (rewrite)
864 		m_copyback(m, off, sizeof(*th), th, M_NOWAIT);
865 
866 	return (PF_PASS);
867 
868  tcp_drop:
869 	REASON_SET(&reason, PFRES_NORM);
870 	return (PF_DROP);
871 }
872 
873 int
874 pf_normalize_tcp_init(struct mbuf *m, int off, struct pf_pdesc *pd,
875     struct tcphdr *th, struct pf_state_peer *src, struct pf_state_peer *dst)
876 {
877 	u_int32_t tsval, tsecr;
878 	u_int8_t hdr[60];
879 	u_int8_t *opt;
880 
881 	KASSERT(src->scrub == NULL);
882 
883 	src->scrub = pool_get(&pf_state_scrub_pl, PR_NOWAIT);
884 	if (src->scrub == NULL)
885 		return (1);
886 	bzero(src->scrub, sizeof(*src->scrub));
887 
888 	switch (pd->af) {
889 #ifdef INET
890 	case AF_INET: {
891 		struct ip *h = mtod(m, struct ip *);
892 		src->scrub->pfss_ttl = h->ip_ttl;
893 		break;
894 	}
895 #endif /* INET */
896 #ifdef INET6
897 	case AF_INET6: {
898 		struct ip6_hdr *h = mtod(m, struct ip6_hdr *);
899 		src->scrub->pfss_ttl = h->ip6_hlim;
900 		break;
901 	}
902 #endif /* INET6 */
903 	}
904 
905 
906 	/*
907 	 * All normalizations below are only begun if we see the start of
908 	 * the connections.  They must all set an enabled bit in pfss_flags
909 	 */
910 	if ((th->th_flags & TH_SYN) == 0)
911 		return (0);
912 
913 
914 	if (th->th_off > (sizeof(struct tcphdr) >> 2) && src->scrub &&
915 	    pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) {
916 		/* Diddle with TCP options */
917 		int hlen;
918 		opt = hdr + sizeof(struct tcphdr);
919 		hlen = (th->th_off << 2) - sizeof(struct tcphdr);
920 		while (hlen >= TCPOLEN_TIMESTAMP) {
921 			switch (*opt) {
922 			case TCPOPT_EOL:	/* FALLTHROUGH */
923 			case TCPOPT_NOP:
924 				opt++;
925 				hlen--;
926 				break;
927 			case TCPOPT_TIMESTAMP:
928 				if (opt[1] >= TCPOLEN_TIMESTAMP) {
929 					src->scrub->pfss_flags |=
930 					    PFSS_TIMESTAMP;
931 					src->scrub->pfss_ts_mod =
932 					    htonl(arc4random());
933 
934 					/* note PFSS_PAWS not set yet */
935 					memcpy(&tsval, &opt[2],
936 					    sizeof(u_int32_t));
937 					memcpy(&tsecr, &opt[6],
938 					    sizeof(u_int32_t));
939 					src->scrub->pfss_tsval0 = ntohl(tsval);
940 					src->scrub->pfss_tsval = ntohl(tsval);
941 					src->scrub->pfss_tsecr = ntohl(tsecr);
942 					getmicrouptime(&src->scrub->pfss_last);
943 				}
944 				/* FALLTHROUGH */
945 			default:
946 				hlen -= MAX(opt[1], 2);
947 				opt += MAX(opt[1], 2);
948 				break;
949 			}
950 		}
951 	}
952 
953 	return (0);
954 }
955 
956 void
957 pf_normalize_tcp_cleanup(struct pf_state *state)
958 {
959 	if (state->src.scrub)
960 		pool_put(&pf_state_scrub_pl, state->src.scrub);
961 	if (state->dst.scrub)
962 		pool_put(&pf_state_scrub_pl, state->dst.scrub);
963 
964 	/* Someday... flush the TCP segment reassembly descriptors. */
965 }
966 
967 int
968 pf_normalize_tcp_stateful(struct mbuf *m, int off, struct pf_pdesc *pd,
969     u_short *reason, struct tcphdr *th, struct pf_state *state,
970     struct pf_state_peer *src, struct pf_state_peer *dst, int *writeback)
971 {
972 	struct timeval uptime;
973 	u_int32_t tsval, tsecr;
974 	u_int tsval_from_last;
975 	u_int8_t hdr[60];
976 	u_int8_t *opt;
977 	int copyback = 0;
978 	int got_ts = 0;
979 
980 	KASSERT(src->scrub || dst->scrub);
981 
982 	/*
983 	 * Enforce the minimum TTL seen for this connection.  Negate a common
984 	 * technique to evade an intrusion detection system and confuse
985 	 * firewall state code.
986 	 */
987 	switch (pd->af) {
988 #ifdef INET
989 	case AF_INET: {
990 		if (src->scrub) {
991 			struct ip *h = mtod(m, struct ip *);
992 			if (h->ip_ttl > src->scrub->pfss_ttl)
993 				src->scrub->pfss_ttl = h->ip_ttl;
994 			h->ip_ttl = src->scrub->pfss_ttl;
995 		}
996 		break;
997 	}
998 #endif /* INET */
999 #ifdef INET6
1000 	case AF_INET6: {
1001 		if (src->scrub) {
1002 			struct ip6_hdr *h = mtod(m, struct ip6_hdr *);
1003 			if (h->ip6_hlim > src->scrub->pfss_ttl)
1004 				src->scrub->pfss_ttl = h->ip6_hlim;
1005 			h->ip6_hlim = src->scrub->pfss_ttl;
1006 		}
1007 		break;
1008 	}
1009 #endif /* INET6 */
1010 	}
1011 
1012 	if (th->th_off > (sizeof(struct tcphdr) >> 2) &&
1013 	    ((src->scrub && (src->scrub->pfss_flags & PFSS_TIMESTAMP)) ||
1014 	    (dst->scrub && (dst->scrub->pfss_flags & PFSS_TIMESTAMP))) &&
1015 	    pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) {
1016 		/* Diddle with TCP options */
1017 		int hlen;
1018 		opt = hdr + sizeof(struct tcphdr);
1019 		hlen = (th->th_off << 2) - sizeof(struct tcphdr);
1020 		while (hlen >= TCPOLEN_TIMESTAMP) {
1021 			switch (*opt) {
1022 			case TCPOPT_EOL:	/* FALLTHROUGH */
1023 			case TCPOPT_NOP:
1024 				opt++;
1025 				hlen--;
1026 				break;
1027 			case TCPOPT_TIMESTAMP:
1028 				/* Modulate the timestamps.  Can be used for
1029 				 * NAT detection, OS uptime determination or
1030 				 * reboot detection.
1031 				 */
1032 
1033 				if (got_ts) {
1034 					/* Huh?  Multiple timestamps!? */
1035 					if (pf_status.debug >= LOG_NOTICE) {
1036 						log(LOG_NOTICE,
1037 						    "pf: %s: multiple TS??",
1038 						    __func__);
1039 						pf_print_state(state);
1040 						addlog("\n");
1041 					}
1042 					REASON_SET(reason, PFRES_TS);
1043 					return (PF_DROP);
1044 				}
1045 				if (opt[1] >= TCPOLEN_TIMESTAMP) {
1046 					memcpy(&tsval, &opt[2],
1047 					    sizeof(u_int32_t));
1048 					if (tsval && src->scrub &&
1049 					    (src->scrub->pfss_flags &
1050 					    PFSS_TIMESTAMP)) {
1051 						tsval = ntohl(tsval);
1052 						pf_change_a(&opt[2],
1053 						    &th->th_sum,
1054 						    htonl(tsval +
1055 						    src->scrub->pfss_ts_mod),
1056 						    0);
1057 						copyback = 1;
1058 					}
1059 
1060 					/* Modulate TS reply iff valid (!0) */
1061 					memcpy(&tsecr, &opt[6],
1062 					    sizeof(u_int32_t));
1063 					if (tsecr && dst->scrub &&
1064 					    (dst->scrub->pfss_flags &
1065 					    PFSS_TIMESTAMP)) {
1066 						tsecr = ntohl(tsecr)
1067 						    - dst->scrub->pfss_ts_mod;
1068 						pf_change_a(&opt[6],
1069 						    &th->th_sum, htonl(tsecr),
1070 						    0);
1071 						copyback = 1;
1072 					}
1073 					got_ts = 1;
1074 				}
1075 				/* FALLTHROUGH */
1076 			default:
1077 				hlen -= MAX(opt[1], 2);
1078 				opt += MAX(opt[1], 2);
1079 				break;
1080 			}
1081 		}
1082 		if (copyback) {
1083 			/* Copyback the options, caller copys back header */
1084 			*writeback = 1;
1085 			m_copyback(m, off + sizeof(struct tcphdr),
1086 			    (th->th_off << 2) - sizeof(struct tcphdr), hdr +
1087 			    sizeof(struct tcphdr), M_NOWAIT);
1088 		}
1089 	}
1090 
1091 
1092 	/*
1093 	 * Must invalidate PAWS checks on connections idle for too long.
1094 	 * The fastest allowed timestamp clock is 1ms.  That turns out to
1095 	 * be about 24 days before it wraps.  XXX Right now our lowerbound
1096 	 * TS echo check only works for the first 12 days of a connection
1097 	 * when the TS has exhausted half its 32bit space
1098 	 */
1099 #define TS_MAX_IDLE	(24*24*60*60)
1100 #define TS_MAX_CONN	(12*24*60*60)	/* XXX remove when better tsecr check */
1101 
1102 	getmicrouptime(&uptime);
1103 	if (src->scrub && (src->scrub->pfss_flags & PFSS_PAWS) &&
1104 	    (uptime.tv_sec - src->scrub->pfss_last.tv_sec > TS_MAX_IDLE ||
1105 	    time_second - state->creation > TS_MAX_CONN))  {
1106 		if (pf_status.debug >= LOG_NOTICE) {
1107 			log(LOG_NOTICE, "pf: src idled out of PAWS ");
1108 			pf_print_state(state);
1109 			addlog("\n");
1110 		}
1111 		src->scrub->pfss_flags = (src->scrub->pfss_flags & ~PFSS_PAWS)
1112 		    | PFSS_PAWS_IDLED;
1113 	}
1114 	if (dst->scrub && (dst->scrub->pfss_flags & PFSS_PAWS) &&
1115 	    uptime.tv_sec - dst->scrub->pfss_last.tv_sec > TS_MAX_IDLE) {
1116 		if (pf_status.debug >= LOG_NOTICE) {
1117 			log(LOG_NOTICE, "pf: dst idled out of PAWS ");
1118 			pf_print_state(state);
1119 			addlog("\n");
1120 		}
1121 		dst->scrub->pfss_flags = (dst->scrub->pfss_flags & ~PFSS_PAWS)
1122 		    | PFSS_PAWS_IDLED;
1123 	}
1124 
1125 	if (got_ts && src->scrub && dst->scrub &&
1126 	    (src->scrub->pfss_flags & PFSS_PAWS) &&
1127 	    (dst->scrub->pfss_flags & PFSS_PAWS)) {
1128 		/* Validate that the timestamps are "in-window".
1129 		 * RFC1323 describes TCP Timestamp options that allow
1130 		 * measurement of RTT (round trip time) and PAWS
1131 		 * (protection against wrapped sequence numbers).  PAWS
1132 		 * gives us a set of rules for rejecting packets on
1133 		 * long fat pipes (packets that were somehow delayed
1134 		 * in transit longer than the time it took to send the
1135 		 * full TCP sequence space of 4Gb).  We can use these
1136 		 * rules and infer a few others that will let us treat
1137 		 * the 32bit timestamp and the 32bit echoed timestamp
1138 		 * as sequence numbers to prevent a blind attacker from
1139 		 * inserting packets into a connection.
1140 		 *
1141 		 * RFC1323 tells us:
1142 		 *  - The timestamp on this packet must be greater than
1143 		 *    or equal to the last value echoed by the other
1144 		 *    endpoint.  The RFC says those will be discarded
1145 		 *    since it is a dup that has already been acked.
1146 		 *    This gives us a lowerbound on the timestamp.
1147 		 *        timestamp >= other last echoed timestamp
1148 		 *  - The timestamp will be less than or equal to
1149 		 *    the last timestamp plus the time between the
1150 		 *    last packet and now.  The RFC defines the max
1151 		 *    clock rate as 1ms.  We will allow clocks to be
1152 		 *    up to 10% fast and will allow a total difference
1153 		 *    or 30 seconds due to a route change.  And this
1154 		 *    gives us an upperbound on the timestamp.
1155 		 *        timestamp <= last timestamp + max ticks
1156 		 *    We have to be careful here.  Windows will send an
1157 		 *    initial timestamp of zero and then initialize it
1158 		 *    to a random value after the 3whs; presumably to
1159 		 *    avoid a DoS by having to call an expensive RNG
1160 		 *    during a SYN flood.  Proof MS has at least one
1161 		 *    good security geek.
1162 		 *
1163 		 *  - The TCP timestamp option must also echo the other
1164 		 *    endpoints timestamp.  The timestamp echoed is the
1165 		 *    one carried on the earliest unacknowledged segment
1166 		 *    on the left edge of the sequence window.  The RFC
1167 		 *    states that the host will reject any echoed
1168 		 *    timestamps that were larger than any ever sent.
1169 		 *    This gives us an upperbound on the TS echo.
1170 		 *        tescr <= largest_tsval
1171 		 *  - The lowerbound on the TS echo is a little more
1172 		 *    tricky to determine.  The other endpoint's echoed
1173 		 *    values will not decrease.  But there may be
1174 		 *    network conditions that re-order packets and
1175 		 *    cause our view of them to decrease.  For now the
1176 		 *    only lowerbound we can safely determine is that
1177 		 *    the TS echo will never be less than the original
1178 		 *    TS.  XXX There is probably a better lowerbound.
1179 		 *    Remove TS_MAX_CONN with better lowerbound check.
1180 		 *        tescr >= other original TS
1181 		 *
1182 		 * It is also important to note that the fastest
1183 		 * timestamp clock of 1ms will wrap its 32bit space in
1184 		 * 24 days.  So we just disable TS checking after 24
1185 		 * days of idle time.  We actually must use a 12d
1186 		 * connection limit until we can come up with a better
1187 		 * lowerbound to the TS echo check.
1188 		 */
1189 		struct timeval delta_ts;
1190 		int ts_fudge;
1191 
1192 
1193 		/*
1194 		 * PFTM_TS_DIFF is how many seconds of leeway to allow
1195 		 * a host's timestamp.  This can happen if the previous
1196 		 * packet got delayed in transit for much longer than
1197 		 * this packet.
1198 		 */
1199 		if ((ts_fudge = state->rule.ptr->timeout[PFTM_TS_DIFF]) == 0)
1200 			ts_fudge = pf_default_rule.timeout[PFTM_TS_DIFF];
1201 
1202 
1203 		/* Calculate max ticks since the last timestamp */
1204 #define TS_MAXFREQ	1100		/* RFC max TS freq of 1Khz + 10% skew */
1205 #define TS_MICROSECS	1000000		/* microseconds per second */
1206 		timersub(&uptime, &src->scrub->pfss_last, &delta_ts);
1207 		tsval_from_last = (delta_ts.tv_sec + ts_fudge) * TS_MAXFREQ;
1208 		tsval_from_last += delta_ts.tv_usec / (TS_MICROSECS/TS_MAXFREQ);
1209 
1210 
1211 		if ((src->state >= TCPS_ESTABLISHED &&
1212 		    dst->state >= TCPS_ESTABLISHED) &&
1213 		    (SEQ_LT(tsval, dst->scrub->pfss_tsecr) ||
1214 		    SEQ_GT(tsval, src->scrub->pfss_tsval + tsval_from_last) ||
1215 		    (tsecr && (SEQ_GT(tsecr, dst->scrub->pfss_tsval) ||
1216 		    SEQ_LT(tsecr, dst->scrub->pfss_tsval0))))) {
1217 			/* Bad RFC1323 implementation or an insertion attack.
1218 			 *
1219 			 * - Solaris 2.6 and 2.7 are known to send another ACK
1220 			 *   after the FIN,FIN|ACK,ACK closing that carries
1221 			 *   an old timestamp.
1222 			 */
1223 
1224 			DPFPRINTF(LOG_NOTICE, "Timestamp failed %c%c%c%c",
1225 			    SEQ_LT(tsval, dst->scrub->pfss_tsecr) ? '0' : ' ',
1226 			    SEQ_GT(tsval, src->scrub->pfss_tsval +
1227 			    tsval_from_last) ? '1' : ' ',
1228 			    SEQ_GT(tsecr, dst->scrub->pfss_tsval) ? '2' : ' ',
1229 			    SEQ_LT(tsecr, dst->scrub->pfss_tsval0)? '3' : ' ');
1230 			DPFPRINTF(LOG_NOTICE,
1231 			    " tsval: %lu  tsecr: %lu  +ticks: %lu  "
1232 			    "idle: %lus %lums",
1233 			    tsval, tsecr, tsval_from_last, delta_ts.tv_sec,
1234 			    delta_ts.tv_usec / 1000);
1235 			DPFPRINTF(LOG_NOTICE,
1236 			    " src->tsval: %lu  tsecr: %lu",
1237 			    src->scrub->pfss_tsval, src->scrub->pfss_tsecr);
1238 			DPFPRINTF(LOG_NOTICE,
1239 			    " dst->tsval: %lu  tsecr: %lu  tsval0: %lu",
1240 			    dst->scrub->pfss_tsval, dst->scrub->pfss_tsecr,
1241 			    dst->scrub->pfss_tsval0);
1242 			if (pf_status.debug >= LOG_NOTICE) {
1243 				log(LOG_NOTICE, "pf: ");
1244 				pf_print_state(state);
1245 				pf_print_flags(th->th_flags);
1246 				addlog("\n");
1247 			}
1248 			REASON_SET(reason, PFRES_TS);
1249 			return (PF_DROP);
1250 		}
1251 
1252 		/* XXX I'd really like to require tsecr but it's optional */
1253 
1254 	} else if (!got_ts && (th->th_flags & TH_RST) == 0 &&
1255 	    ((src->state == TCPS_ESTABLISHED && dst->state == TCPS_ESTABLISHED)
1256 	    || pd->p_len > 0 || (th->th_flags & TH_SYN)) &&
1257 	    src->scrub && dst->scrub &&
1258 	    (src->scrub->pfss_flags & PFSS_PAWS) &&
1259 	    (dst->scrub->pfss_flags & PFSS_PAWS)) {
1260 		/* Didn't send a timestamp.  Timestamps aren't really useful
1261 		 * when:
1262 		 *  - connection opening or closing (often not even sent).
1263 		 *    but we must not let an attacker to put a FIN on a
1264 		 *    data packet to sneak it through our ESTABLISHED check.
1265 		 *  - on a TCP reset.  RFC suggests not even looking at TS.
1266 		 *  - on an empty ACK.  The TS will not be echoed so it will
1267 		 *    probably not help keep the RTT calculation in sync and
1268 		 *    there isn't as much danger when the sequence numbers
1269 		 *    got wrapped.  So some stacks don't include TS on empty
1270 		 *    ACKs :-(
1271 		 *
1272 		 * To minimize the disruption to mostly RFC1323 conformant
1273 		 * stacks, we will only require timestamps on data packets.
1274 		 *
1275 		 * And what do ya know, we cannot require timestamps on data
1276 		 * packets.  There appear to be devices that do legitimate
1277 		 * TCP connection hijacking.  There are HTTP devices that allow
1278 		 * a 3whs (with timestamps) and then buffer the HTTP request.
1279 		 * If the intermediate device has the HTTP response cache, it
1280 		 * will spoof the response but not bother timestamping its
1281 		 * packets.  So we can look for the presence of a timestamp in
1282 		 * the first data packet and if there, require it in all future
1283 		 * packets.
1284 		 */
1285 
1286 		if (pd->p_len > 0 && (src->scrub->pfss_flags & PFSS_DATA_TS)) {
1287 			/*
1288 			 * Hey!  Someone tried to sneak a packet in.  Or the
1289 			 * stack changed its RFC1323 behavior?!?!
1290 			 */
1291 			if (pf_status.debug >= LOG_NOTICE) {
1292 				log(LOG_NOTICE,
1293 				    "pf: did not receive expected RFC1323 "
1294 				    "timestamp");
1295 				pf_print_state(state);
1296 				pf_print_flags(th->th_flags);
1297 				addlog("\n");
1298 			}
1299 			REASON_SET(reason, PFRES_TS);
1300 			return (PF_DROP);
1301 		}
1302 	}
1303 
1304 
1305 	/*
1306 	 * We will note if a host sends his data packets with or without
1307 	 * timestamps.  And require all data packets to contain a timestamp
1308 	 * if the first does.  PAWS implicitly requires that all data packets be
1309 	 * timestamped.  But I think there are middle-man devices that hijack
1310 	 * TCP streams immediately after the 3whs and don't timestamp their
1311 	 * packets (seen in a WWW accelerator or cache).
1312 	 */
1313 	if (pd->p_len > 0 && src->scrub && (src->scrub->pfss_flags &
1314 	    (PFSS_TIMESTAMP|PFSS_DATA_TS|PFSS_DATA_NOTS)) == PFSS_TIMESTAMP) {
1315 		if (got_ts)
1316 			src->scrub->pfss_flags |= PFSS_DATA_TS;
1317 		else {
1318 			src->scrub->pfss_flags |= PFSS_DATA_NOTS;
1319 			if (pf_status.debug >= LOG_NOTICE && dst->scrub &&
1320 			    (dst->scrub->pfss_flags & PFSS_TIMESTAMP)) {
1321 				/* Don't warn if other host rejected RFC1323 */
1322 				log(LOG_NOTICE,
1323 				    "pf: broken RFC1323 stack did not "
1324 				    "timestamp data packet. Disabled PAWS "
1325 				    "security.");
1326 				pf_print_state(state);
1327 				pf_print_flags(th->th_flags);
1328 				addlog("\n");
1329 			}
1330 		}
1331 	}
1332 
1333 
1334 	/*
1335 	 * Update PAWS values
1336 	 */
1337 	if (got_ts && src->scrub && PFSS_TIMESTAMP == (src->scrub->pfss_flags &
1338 	    (PFSS_PAWS_IDLED|PFSS_TIMESTAMP))) {
1339 		getmicrouptime(&src->scrub->pfss_last);
1340 		if (SEQ_GEQ(tsval, src->scrub->pfss_tsval) ||
1341 		    (src->scrub->pfss_flags & PFSS_PAWS) == 0)
1342 			src->scrub->pfss_tsval = tsval;
1343 
1344 		if (tsecr) {
1345 			if (SEQ_GEQ(tsecr, src->scrub->pfss_tsecr) ||
1346 			    (src->scrub->pfss_flags & PFSS_PAWS) == 0)
1347 				src->scrub->pfss_tsecr = tsecr;
1348 
1349 			if ((src->scrub->pfss_flags & PFSS_PAWS) == 0 &&
1350 			    (SEQ_LT(tsval, src->scrub->pfss_tsval0) ||
1351 			    src->scrub->pfss_tsval0 == 0)) {
1352 				/* tsval0 MUST be the lowest timestamp */
1353 				src->scrub->pfss_tsval0 = tsval;
1354 			}
1355 
1356 			/* Only fully initialized after a TS gets echoed */
1357 			if ((src->scrub->pfss_flags & PFSS_PAWS) == 0)
1358 				src->scrub->pfss_flags |= PFSS_PAWS;
1359 		}
1360 	}
1361 
1362 	/* I have a dream....  TCP segment reassembly.... */
1363 	return (0);
1364 }
1365 
1366 int
1367 pf_normalize_mss(struct mbuf *m, int off, struct pf_pdesc *pd, u_int16_t maxmss)
1368 {
1369 	struct tcphdr	*th = pd->hdr.tcp;
1370 	u_int16_t	 mss;
1371 	int		 thoff;
1372 	int		 opt, cnt, optlen = 0;
1373 	u_char		 opts[MAX_TCPOPTLEN];
1374 	u_char		*optp = opts;
1375 
1376 	thoff = th->th_off << 2;
1377 	cnt = thoff - sizeof(struct tcphdr);
1378 
1379 	if (cnt > 0 && !pf_pull_hdr(m, off + sizeof(*th), opts, cnt,
1380 	    NULL, NULL, pd->af))
1381 		return (0);
1382 
1383 	for (; cnt > 0; cnt -= optlen, optp += optlen) {
1384 		opt = optp[0];
1385 		if (opt == TCPOPT_EOL)
1386 			break;
1387 		if (opt == TCPOPT_NOP)
1388 			optlen = 1;
1389 		else {
1390 			if (cnt < 2)
1391 				break;
1392 			optlen = optp[1];
1393 			if (optlen < 2 || optlen > cnt)
1394 				break;
1395 		}
1396 		switch (opt) {
1397 		case TCPOPT_MAXSEG:
1398 			bcopy((caddr_t)(optp + 2), (caddr_t)&mss, 2);
1399 			if (ntohs(mss) > maxmss) {
1400 				th->th_sum = pf_cksum_fixup(th->th_sum,
1401 				    mss, htons(maxmss), 0);
1402 				mss = htons(maxmss);
1403 				m_copyback(m,
1404 				    off + sizeof(*th) + optp + 2 - opts,
1405 				    2, &mss, M_NOWAIT);
1406 				m_copyback(m, off, sizeof(*th), th, M_NOWAIT);
1407 			}
1408 			break;
1409 		default:
1410 			break;
1411 		}
1412 	}
1413 
1414 
1415 
1416 	return (0);
1417 }
1418 
1419 void
1420 pf_scrub(struct mbuf *m, u_int16_t flags, sa_family_t af, u_int8_t min_ttl,
1421     u_int8_t tos)
1422 {
1423 	struct ip		*h = mtod(m, struct ip *);
1424 #ifdef INET6
1425 	struct ip6_hdr		*h6 = mtod(m, struct ip6_hdr *);
1426 #endif
1427 
1428 	/* Clear IP_DF if no-df was requested */
1429 	if (flags & PFSTATE_NODF && af == AF_INET && h->ip_off & htons(IP_DF))
1430 		h->ip_off &= htons(~IP_DF);
1431 
1432 	/* Enforce a minimum ttl, may cause endless packet loops */
1433 	if (min_ttl && af == AF_INET && h->ip_ttl < min_ttl)
1434 		h->ip_ttl = min_ttl;
1435 #ifdef INET6
1436 	if (min_ttl && af == AF_INET6 && h6->ip6_hlim < min_ttl)
1437 		h6->ip6_hlim = min_ttl;
1438 #endif
1439 
1440 	/* Enforce tos */
1441 	if (flags & PFSTATE_SETTOS) {
1442 		if (af == AF_INET)
1443 			h->ip_tos = tos;
1444 #ifdef INET6
1445 		if (af == AF_INET6) {
1446 			/* drugs are unable to explain such idiocy */
1447 			h6->ip6_flow &= ~htonl(0x0ff00000);
1448 			h6->ip6_flow |= htonl(((u_int32_t)tos) << 20);
1449 		}
1450 #endif
1451 	}
1452 
1453 	/* random-id, but not for fragments */
1454 	if (flags & PFSTATE_RANDOMID && af == AF_INET &&
1455 	    !(h->ip_off & ~htons(IP_DF)))
1456 		h->ip_id = htons(ip_randomid());
1457 }
1458