xref: /freebsd/sys/netpfil/pf/pf_norm.c (revision fcb560670601b2a4d87bb31d7531c8dcc37ee71b)
1 /*-
2  * Copyright 2001 Niels Provos <provos@citi.umich.edu>
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
15  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
16  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
17  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
18  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
19  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
20  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
21  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
22  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
23  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
24  *
25  *	$OpenBSD: pf_norm.c,v 1.114 2009/01/29 14:11:45 henning Exp $
26  */
27 
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
30 
31 #include "opt_inet.h"
32 #include "opt_inet6.h"
33 #include "opt_pf.h"
34 
35 #include <sys/param.h>
36 #include <sys/lock.h>
37 #include <sys/mbuf.h>
38 #include <sys/mutex.h>
39 #include <sys/refcount.h>
40 #include <sys/rwlock.h>
41 #include <sys/socket.h>
42 
43 #include <net/if.h>
44 #include <net/vnet.h>
45 #include <net/pfvar.h>
46 #include <net/if_pflog.h>
47 
48 #include <netinet/in.h>
49 #include <netinet/ip.h>
50 #include <netinet/ip_var.h>
51 #include <netinet/tcp.h>
52 #include <netinet/tcp_fsm.h>
53 #include <netinet/tcp_seq.h>
54 
55 #ifdef INET6
56 #include <netinet/ip6.h>
57 #endif /* INET6 */
58 
59 struct pf_frent {
60 	LIST_ENTRY(pf_frent) fr_next;
61 	union {
62 		struct {
63 			struct ip *_fr_ip;
64 			struct mbuf *_fr_m;
65 		} _frag;
66 		struct {
67 			uint16_t _fr_off;
68 			uint16_t _fr_end;
69 		} _cache;
70 	} _u;
71 };
72 #define	fr_ip	_u._frag._fr_ip
73 #define	fr_m	_u._frag._fr_m
74 #define	fr_off	_u._cache._fr_off
75 #define	fr_end	_u._cache._fr_end
76 
77 struct pf_fragment {
78 	RB_ENTRY(pf_fragment) fr_entry;
79 	TAILQ_ENTRY(pf_fragment) frag_next;
80 	struct in_addr	fr_src;
81 	struct in_addr	fr_dst;
82 	u_int8_t	fr_p;		/* protocol of this fragment */
83 	u_int8_t	fr_flags;	/* status flags */
84 #define PFFRAG_SEENLAST	0x0001		/* Seen the last fragment for this */
85 #define PFFRAG_NOBUFFER	0x0002		/* Non-buffering fragment cache */
86 #define PFFRAG_DROP	0x0004		/* Drop all fragments */
87 #define BUFFER_FRAGMENTS(fr)	(!((fr)->fr_flags & PFFRAG_NOBUFFER))
88 	u_int16_t	fr_id;		/* fragment id for reassemble */
89 	u_int16_t	fr_max;		/* fragment data max */
90 	u_int32_t	fr_timeout;
91 	LIST_HEAD(, pf_frent) fr_queue;
92 };
93 
94 static struct mtx pf_frag_mtx;
95 #define PF_FRAG_LOCK()		mtx_lock(&pf_frag_mtx)
96 #define PF_FRAG_UNLOCK()	mtx_unlock(&pf_frag_mtx)
97 #define PF_FRAG_ASSERT()	mtx_assert(&pf_frag_mtx, MA_OWNED)
98 
99 VNET_DEFINE(uma_zone_t, pf_state_scrub_z);	/* XXX: shared with pfsync */
100 
101 static VNET_DEFINE(uma_zone_t, pf_frent_z);
102 #define	V_pf_frent_z	VNET(pf_frent_z)
103 static VNET_DEFINE(uma_zone_t, pf_frag_z);
104 #define	V_pf_frag_z	VNET(pf_frag_z)
105 
106 TAILQ_HEAD(pf_fragqueue, pf_fragment);
107 TAILQ_HEAD(pf_cachequeue, pf_fragment);
108 static VNET_DEFINE(struct pf_fragqueue,	pf_fragqueue);
109 #define	V_pf_fragqueue			VNET(pf_fragqueue)
110 static VNET_DEFINE(struct pf_cachequeue,	pf_cachequeue);
111 #define	V_pf_cachequeue			VNET(pf_cachequeue)
112 RB_HEAD(pf_frag_tree, pf_fragment);
113 static VNET_DEFINE(struct pf_frag_tree,	pf_frag_tree);
114 #define	V_pf_frag_tree			VNET(pf_frag_tree)
115 static VNET_DEFINE(struct pf_frag_tree,	pf_cache_tree);
116 #define	V_pf_cache_tree			VNET(pf_cache_tree)
117 static int		 pf_frag_compare(struct pf_fragment *,
118 			    struct pf_fragment *);
119 static RB_PROTOTYPE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
120 static RB_GENERATE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
121 
122 /* Private prototypes */
123 static void		 pf_free_fragment(struct pf_fragment *);
124 static void		 pf_remove_fragment(struct pf_fragment *);
125 static int		 pf_normalize_tcpopt(struct pf_rule *, struct mbuf *,
126 			    struct tcphdr *, int, sa_family_t);
127 #ifdef INET
128 static void		 pf_ip2key(struct pf_fragment *, struct ip *);
129 static void		 pf_scrub_ip(struct mbuf **, u_int32_t, u_int8_t,
130 			    u_int8_t);
131 static void		 pf_flush_fragments(void);
132 static struct pf_fragment *pf_find_fragment(struct ip *, struct pf_frag_tree *);
133 static struct mbuf	*pf_reassemble(struct mbuf **, struct pf_fragment **,
134 			    struct pf_frent *, int);
135 static struct mbuf	*pf_fragcache(struct mbuf **, struct ip*,
136 			    struct pf_fragment **, int, int, int *);
137 #endif /* INET */
138 #ifdef INET6
139 static void		 pf_scrub_ip6(struct mbuf **, u_int8_t);
140 #endif
141 #define	DPFPRINTF(x) do {				\
142 	if (V_pf_status.debug >= PF_DEBUG_MISC) {	\
143 		printf("%s: ", __func__);		\
144 		printf x ;				\
145 	}						\
146 } while(0)
147 
148 void
149 pf_normalize_init(void)
150 {
151 
152 	V_pf_frag_z = uma_zcreate("pf frags", sizeof(struct pf_fragment),
153 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
154 	V_pf_frent_z = uma_zcreate("pf frag entries", sizeof(struct pf_frent),
155 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
156 	V_pf_state_scrub_z = uma_zcreate("pf state scrubs",
157 	    sizeof(struct pf_state_scrub),  NULL, NULL, NULL, NULL,
158 	    UMA_ALIGN_PTR, 0);
159 
160 	V_pf_limits[PF_LIMIT_FRAGS].zone = V_pf_frent_z;
161 	V_pf_limits[PF_LIMIT_FRAGS].limit = PFFRAG_FRENT_HIWAT;
162 	uma_zone_set_max(V_pf_frent_z, PFFRAG_FRENT_HIWAT);
163 	uma_zone_set_warning(V_pf_frent_z, "PF frag entries limit reached");
164 
165 	mtx_init(&pf_frag_mtx, "pf fragments", NULL, MTX_DEF);
166 
167 	TAILQ_INIT(&V_pf_fragqueue);
168 	TAILQ_INIT(&V_pf_cachequeue);
169 }
170 
171 void
172 pf_normalize_cleanup(void)
173 {
174 
175 	uma_zdestroy(V_pf_state_scrub_z);
176 	uma_zdestroy(V_pf_frent_z);
177 	uma_zdestroy(V_pf_frag_z);
178 
179 	mtx_destroy(&pf_frag_mtx);
180 }
181 
182 static int
183 pf_frag_compare(struct pf_fragment *a, struct pf_fragment *b)
184 {
185 	int	diff;
186 
187 	if ((diff = a->fr_id - b->fr_id))
188 		return (diff);
189 	else if ((diff = a->fr_p - b->fr_p))
190 		return (diff);
191 	else if (a->fr_src.s_addr < b->fr_src.s_addr)
192 		return (-1);
193 	else if (a->fr_src.s_addr > b->fr_src.s_addr)
194 		return (1);
195 	else if (a->fr_dst.s_addr < b->fr_dst.s_addr)
196 		return (-1);
197 	else if (a->fr_dst.s_addr > b->fr_dst.s_addr)
198 		return (1);
199 	return (0);
200 }
201 
202 void
203 pf_purge_expired_fragments(void)
204 {
205 	struct pf_fragment	*frag;
206 	u_int32_t		 expire = time_uptime -
207 				    V_pf_default_rule.timeout[PFTM_FRAG];
208 
209 	PF_FRAG_LOCK();
210 	while ((frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue)) != NULL) {
211 		KASSERT((BUFFER_FRAGMENTS(frag)),
212 		    ("BUFFER_FRAGMENTS(frag) == 0: %s", __FUNCTION__));
213 		if (frag->fr_timeout > expire)
214 			break;
215 
216 		DPFPRINTF(("expiring %d(%p)\n", frag->fr_id, frag));
217 		pf_free_fragment(frag);
218 	}
219 
220 	while ((frag = TAILQ_LAST(&V_pf_cachequeue, pf_cachequeue)) != NULL) {
221 		KASSERT((!BUFFER_FRAGMENTS(frag)),
222 		    ("BUFFER_FRAGMENTS(frag) != 0: %s", __FUNCTION__));
223 		if (frag->fr_timeout > expire)
224 			break;
225 
226 		DPFPRINTF(("expiring %d(%p)\n", frag->fr_id, frag));
227 		pf_free_fragment(frag);
228 		KASSERT((TAILQ_EMPTY(&V_pf_cachequeue) ||
229 		    TAILQ_LAST(&V_pf_cachequeue, pf_cachequeue) != frag),
230 		    ("!(TAILQ_EMPTY() || TAILQ_LAST() == farg): %s",
231 		    __FUNCTION__));
232 	}
233 	PF_FRAG_UNLOCK();
234 }
235 
236 #ifdef INET
237 /*
238  * Try to flush old fragments to make space for new ones
239  */
240 static void
241 pf_flush_fragments(void)
242 {
243 	struct pf_fragment	*frag, *cache;
244 	int			 goal;
245 
246 	PF_FRAG_ASSERT();
247 
248 	goal = uma_zone_get_cur(V_pf_frent_z) * 9 / 10;
249 	DPFPRINTF(("trying to free %d frag entriess\n", goal));
250 	while (goal < uma_zone_get_cur(V_pf_frent_z)) {
251 		frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue);
252 		if (frag)
253 			pf_free_fragment(frag);
254 		cache = TAILQ_LAST(&V_pf_cachequeue, pf_cachequeue);
255 		if (cache)
256 			pf_free_fragment(cache);
257 		if (frag == NULL && cache == NULL)
258 			break;
259 	}
260 }
261 #endif /* INET */
262 
263 /* Frees the fragments and all associated entries */
264 static void
265 pf_free_fragment(struct pf_fragment *frag)
266 {
267 	struct pf_frent		*frent;
268 
269 	PF_FRAG_ASSERT();
270 
271 	/* Free all fragments */
272 	if (BUFFER_FRAGMENTS(frag)) {
273 		for (frent = LIST_FIRST(&frag->fr_queue); frent;
274 		    frent = LIST_FIRST(&frag->fr_queue)) {
275 			LIST_REMOVE(frent, fr_next);
276 
277 			m_freem(frent->fr_m);
278 			uma_zfree(V_pf_frent_z, frent);
279 		}
280 	} else {
281 		for (frent = LIST_FIRST(&frag->fr_queue); frent;
282 		    frent = LIST_FIRST(&frag->fr_queue)) {
283 			LIST_REMOVE(frent, fr_next);
284 
285 			KASSERT((LIST_EMPTY(&frag->fr_queue) ||
286 			    LIST_FIRST(&frag->fr_queue)->fr_off >
287 			    frent->fr_end),
288 			    ("! (LIST_EMPTY() || LIST_FIRST()->fr_off >"
289 			    " frent->fr_end): %s", __func__));
290 
291 			uma_zfree(V_pf_frent_z, frent);
292 		}
293 	}
294 
295 	pf_remove_fragment(frag);
296 }
297 
298 #ifdef INET
299 static void
300 pf_ip2key(struct pf_fragment *key, struct ip *ip)
301 {
302 	key->fr_p = ip->ip_p;
303 	key->fr_id = ip->ip_id;
304 	key->fr_src.s_addr = ip->ip_src.s_addr;
305 	key->fr_dst.s_addr = ip->ip_dst.s_addr;
306 }
307 
308 static struct pf_fragment *
309 pf_find_fragment(struct ip *ip, struct pf_frag_tree *tree)
310 {
311 	struct pf_fragment	 key;
312 	struct pf_fragment	*frag;
313 
314 	PF_FRAG_ASSERT();
315 
316 	pf_ip2key(&key, ip);
317 
318 	frag = RB_FIND(pf_frag_tree, tree, &key);
319 	if (frag != NULL) {
320 		/* XXX Are we sure we want to update the timeout? */
321 		frag->fr_timeout = time_uptime;
322 		if (BUFFER_FRAGMENTS(frag)) {
323 			TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next);
324 			TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next);
325 		} else {
326 			TAILQ_REMOVE(&V_pf_cachequeue, frag, frag_next);
327 			TAILQ_INSERT_HEAD(&V_pf_cachequeue, frag, frag_next);
328 		}
329 	}
330 
331 	return (frag);
332 }
333 #endif /* INET */
334 
335 /* Removes a fragment from the fragment queue and frees the fragment */
336 
337 static void
338 pf_remove_fragment(struct pf_fragment *frag)
339 {
340 
341 	PF_FRAG_ASSERT();
342 
343 	if (BUFFER_FRAGMENTS(frag)) {
344 		RB_REMOVE(pf_frag_tree, &V_pf_frag_tree, frag);
345 		TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next);
346 		uma_zfree(V_pf_frag_z, frag);
347 	} else {
348 		RB_REMOVE(pf_frag_tree, &V_pf_cache_tree, frag);
349 		TAILQ_REMOVE(&V_pf_cachequeue, frag, frag_next);
350 		uma_zfree(V_pf_frag_z, frag);
351 	}
352 }
353 
354 #ifdef INET
355 #define FR_IP_OFF(fr)	((ntohs((fr)->fr_ip->ip_off) & IP_OFFMASK) << 3)
356 static struct mbuf *
357 pf_reassemble(struct mbuf **m0, struct pf_fragment **frag,
358     struct pf_frent *frent, int mff)
359 {
360 	struct mbuf	*m = *m0, *m2;
361 	struct pf_frent	*frea, *next;
362 	struct pf_frent	*frep = NULL;
363 	struct ip	*ip = frent->fr_ip;
364 	int		 hlen = ip->ip_hl << 2;
365 	u_int16_t	 off = (ntohs(ip->ip_off) & IP_OFFMASK) << 3;
366 	u_int16_t	 ip_len = ntohs(ip->ip_len) - ip->ip_hl * 4;
367 	u_int16_t	 max = ip_len + off;
368 
369 	PF_FRAG_ASSERT();
370 	KASSERT((*frag == NULL || BUFFER_FRAGMENTS(*frag)),
371 	    ("! (*frag == NULL || BUFFER_FRAGMENTS(*frag)): %s", __FUNCTION__));
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 = uma_zalloc(V_pf_frag_z, M_NOWAIT);
380 		if (*frag == NULL) {
381 			pf_flush_fragments();
382 			*frag = uma_zalloc(V_pf_frag_z, M_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_uptime;
394 		LIST_INIT(&(*frag)->fr_queue);
395 
396 		RB_INSERT(pf_frag_tree, &V_pf_frag_tree, *frag);
397 		TAILQ_INSERT_HEAD(&V_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 	    ("!(frep != NULL || frea != NULL): %s", __FUNCTION__));;
416 
417 	if (frep != NULL &&
418 	    FR_IP_OFF(frep) + ntohs(frep->fr_ip->ip_len) - frep->fr_ip->ip_hl *
419 	    4 > off)
420 	{
421 		u_int16_t	precut;
422 
423 		precut = FR_IP_OFF(frep) + ntohs(frep->fr_ip->ip_len) -
424 		    frep->fr_ip->ip_hl * 4 - off;
425 		if (precut >= ip_len)
426 			goto drop_fragment;
427 		m_adj(frent->fr_m, precut);
428 		DPFPRINTF(("overlap -%d\n", precut));
429 		/* Enforce 8 byte boundaries */
430 		ip->ip_off = htons(ntohs(ip->ip_off) + (precut >> 3));
431 		off = (ntohs(ip->ip_off) & IP_OFFMASK) << 3;
432 		ip_len -= precut;
433 		ip->ip_len = htons(ip_len);
434 	}
435 
436 	for (; frea != NULL && ip_len + off > FR_IP_OFF(frea);
437 	    frea = next)
438 	{
439 		u_int16_t	aftercut;
440 
441 		aftercut = ip_len + off - FR_IP_OFF(frea);
442 		DPFPRINTF(("adjust overlap %d\n", aftercut));
443 		if (aftercut < ntohs(frea->fr_ip->ip_len) - frea->fr_ip->ip_hl
444 		    * 4)
445 		{
446 			frea->fr_ip->ip_len =
447 			    htons(ntohs(frea->fr_ip->ip_len) - aftercut);
448 			frea->fr_ip->ip_off = htons(ntohs(frea->fr_ip->ip_off) +
449 			    (aftercut >> 3));
450 			m_adj(frea->fr_m, aftercut);
451 			break;
452 		}
453 
454 		/* This fragment is completely overlapped, lose it */
455 		next = LIST_NEXT(frea, fr_next);
456 		m_freem(frea->fr_m);
457 		LIST_REMOVE(frea, fr_next);
458 		uma_zfree(V_pf_frent_z, frea);
459 	}
460 
461  insert:
462 	/* Update maximum data size */
463 	if ((*frag)->fr_max < max)
464 		(*frag)->fr_max = max;
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), ("frent == NULL: %s", __FUNCTION__));
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 	uma_zfree(V_pf_frent_z, frent);
515 	for (frent = next; frent != NULL; frent = next) {
516 		next = LIST_NEXT(frent, fr_next);
517 
518 		m2 = frent->fr_m;
519 		uma_zfree(V_pf_frent_z, frent);
520 		m->m_pkthdr.csum_flags &= m2->m_pkthdr.csum_flags;
521 		m->m_pkthdr.csum_data += m2->m_pkthdr.csum_data;
522 		m_cat(m, m2);
523 	}
524 
525 	while (m->m_pkthdr.csum_data & 0xffff0000)
526 		m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) +
527 		    (m->m_pkthdr.csum_data >> 16);
528 	ip->ip_src = (*frag)->fr_src;
529 	ip->ip_dst = (*frag)->fr_dst;
530 
531 	/* Remove from fragment queue */
532 	pf_remove_fragment(*frag);
533 	*frag = NULL;
534 
535 	hlen = ip->ip_hl << 2;
536 	ip->ip_len = htons(off + hlen);
537 	m->m_len += hlen;
538 	m->m_data -= hlen;
539 
540 	/* some debugging cruft by sklower, below, will go away soon */
541 	/* XXX this should be done elsewhere */
542 	if (m->m_flags & M_PKTHDR) {
543 		int plen = 0;
544 		for (m2 = m; m2; m2 = m2->m_next)
545 			plen += m2->m_len;
546 		m->m_pkthdr.len = plen;
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 	uma_zfree(V_pf_frent_z, frent);
555 	m_freem(m);
556 	return (NULL);
557 }
558 
559 static struct mbuf *
560 pf_fragcache(struct mbuf **m0, struct ip *h, struct pf_fragment **frag, int mff,
561     int drop, int *nomem)
562 {
563 	struct mbuf		*m = *m0;
564 	struct pf_frent		*frp, *fra, *cur = NULL;
565 	int			 ip_len = ntohs(h->ip_len) - (h->ip_hl << 2);
566 	u_int16_t		 off = ntohs(h->ip_off) << 3;
567 	u_int16_t		 max = ip_len + off;
568 	int			 hosed = 0;
569 
570 	PF_FRAG_ASSERT();
571 	KASSERT((*frag == NULL || !BUFFER_FRAGMENTS(*frag)),
572 	    ("!(*frag == NULL || !BUFFER_FRAGMENTS(*frag)): %s", __FUNCTION__));
573 
574 	/* Create a new range queue for this packet */
575 	if (*frag == NULL) {
576 		*frag = uma_zalloc(V_pf_frag_z, M_NOWAIT);
577 		if (*frag == NULL) {
578 			pf_flush_fragments();
579 			*frag = uma_zalloc(V_pf_frag_z, M_NOWAIT);
580 			if (*frag == NULL)
581 				goto no_mem;
582 		}
583 
584 		/* Get an entry for the queue */
585 		cur = uma_zalloc(V_pf_frent_z, M_NOWAIT);
586 		if (cur == NULL) {
587 			uma_zfree(V_pf_frag_z, *frag);
588 			*frag = NULL;
589 			goto no_mem;
590 		}
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_uptime;
599 
600 		cur->fr_off = off;
601 		cur->fr_end = max;
602 		LIST_INIT(&(*frag)->fr_queue);
603 		LIST_INSERT_HEAD(&(*frag)->fr_queue, cur, fr_next);
604 
605 		RB_INSERT(pf_frag_tree, &V_pf_cache_tree, *frag);
606 		TAILQ_INSERT_HEAD(&V_pf_cachequeue, *frag, frag_next);
607 
608 		DPFPRINTF(("fragcache[%d]: new %d-%d\n", h->ip_id, off, max));
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_queue, fr_next) {
619 		if (fra->fr_off > off)
620 			break;
621 		frp = fra;
622 	}
623 
624 	KASSERT((frp != NULL || fra != NULL),
625 	    ("!(frp != NULL || fra != NULL): %s", __FUNCTION__));
626 
627 	if (frp != NULL) {
628 		int	precut;
629 
630 		precut = frp->fr_end - off;
631 		if (precut >= ip_len) {
632 			/* Fragment is entirely a duplicate */
633 			DPFPRINTF(("fragcache[%d]: dead (%d-%d) %d-%d\n",
634 			    h->ip_id, frp->fr_off, frp->fr_end, off, max));
635 			goto drop_fragment;
636 		}
637 		if (precut == 0) {
638 			/* They are adjacent.  Fixup cache entry */
639 			DPFPRINTF(("fragcache[%d]: adjacent (%d-%d) %d-%d\n",
640 			    h->ip_id, frp->fr_off, frp->fr_end, off, max));
641 			frp->fr_end = max;
642 		} else if (precut > 0) {
643 			/* The first part of this payload overlaps with a
644 			 * fragment that has already been passed.
645 			 * Need to trim off the first part of the payload.
646 			 * But to do so easily, we need to create another
647 			 * mbuf to throw the original header into.
648 			 */
649 
650 			DPFPRINTF(("fragcache[%d]: chop %d (%d-%d) %d-%d\n",
651 			    h->ip_id, precut, frp->fr_off, frp->fr_end, off,
652 			    max));
653 
654 			off += precut;
655 			max -= precut;
656 			/* Update the previous frag to encompass this one */
657 			frp->fr_end = max;
658 
659 			if (!drop) {
660 				/* XXX Optimization opportunity
661 				 * This is a very heavy way to trim the payload.
662 				 * we could do it much faster by diddling mbuf
663 				 * internals but that would be even less legible
664 				 * than this mbuf magic.  For my next trick,
665 				 * I'll pull a rabbit out of my laptop.
666 				 */
667 				*m0 = m_dup(m, M_NOWAIT);
668 				if (*m0 == NULL)
669 					goto no_mem;
670 				/* From KAME Project : We have missed this! */
671 				m_adj(*m0, (h->ip_hl << 2) -
672 				    (*m0)->m_pkthdr.len);
673 
674 				KASSERT(((*m0)->m_next == NULL),
675 				    ("(*m0)->m_next != NULL: %s",
676 				    __FUNCTION__));
677 				m_adj(m, precut + (h->ip_hl << 2));
678 				m_cat(*m0, m);
679 				m = *m0;
680 				if (m->m_flags & M_PKTHDR) {
681 					int plen = 0;
682 					struct mbuf *t;
683 					for (t = m; t; t = t->m_next)
684 						plen += t->m_len;
685 					m->m_pkthdr.len = plen;
686 				}
687 
688 
689 				h = mtod(m, struct ip *);
690 
691 				KASSERT(((int)m->m_len ==
692 				    ntohs(h->ip_len) - precut),
693 				    ("m->m_len != ntohs(h->ip_len) - precut: %s",
694 				    __FUNCTION__));
695 				h->ip_off = htons(ntohs(h->ip_off) +
696 				    (precut >> 3));
697 				h->ip_len = htons(ntohs(h->ip_len) - precut);
698 			} else {
699 				hosed++;
700 			}
701 		} else {
702 			/* There is a gap between fragments */
703 
704 			DPFPRINTF(("fragcache[%d]: gap %d (%d-%d) %d-%d\n",
705 			    h->ip_id, -precut, frp->fr_off, frp->fr_end, off,
706 			    max));
707 
708 			cur = uma_zalloc(V_pf_frent_z, M_NOWAIT);
709 			if (cur == NULL)
710 				goto no_mem;
711 
712 			cur->fr_off = off;
713 			cur->fr_end = max;
714 			LIST_INSERT_AFTER(frp, cur, fr_next);
715 		}
716 	}
717 
718 	if (fra != NULL) {
719 		int	aftercut;
720 		int	merge = 0;
721 
722 		aftercut = max - fra->fr_off;
723 		if (aftercut == 0) {
724 			/* Adjacent fragments */
725 			DPFPRINTF(("fragcache[%d]: adjacent %d-%d (%d-%d)\n",
726 			    h->ip_id, off, max, fra->fr_off, fra->fr_end));
727 			fra->fr_off = off;
728 			merge = 1;
729 		} else if (aftercut > 0) {
730 			/* Need to chop off the tail of this fragment */
731 			DPFPRINTF(("fragcache[%d]: chop %d %d-%d (%d-%d)\n",
732 			    h->ip_id, aftercut, off, max, fra->fr_off,
733 			    fra->fr_end));
734 			fra->fr_off = off;
735 			max -= aftercut;
736 
737 			merge = 1;
738 
739 			if (!drop) {
740 				m_adj(m, -aftercut);
741 				if (m->m_flags & M_PKTHDR) {
742 					int plen = 0;
743 					struct mbuf *t;
744 					for (t = m; t; t = t->m_next)
745 						plen += t->m_len;
746 					m->m_pkthdr.len = plen;
747 				}
748 				h = mtod(m, struct ip *);
749 				KASSERT(((int)m->m_len == ntohs(h->ip_len) - aftercut),
750 				    ("m->m_len != ntohs(h->ip_len) - aftercut: %s",
751 				    __FUNCTION__));
752 				h->ip_len = htons(ntohs(h->ip_len) - aftercut);
753 			} else {
754 				hosed++;
755 			}
756 		} else if (frp == NULL) {
757 			/* There is a gap between fragments */
758 			DPFPRINTF(("fragcache[%d]: gap %d %d-%d (%d-%d)\n",
759 			    h->ip_id, -aftercut, off, max, fra->fr_off,
760 			    fra->fr_end));
761 
762 			cur = uma_zalloc(V_pf_frent_z, M_NOWAIT);
763 			if (cur == NULL)
764 				goto no_mem;
765 
766 			cur->fr_off = off;
767 			cur->fr_end = max;
768 			LIST_INSERT_BEFORE(fra, cur, fr_next);
769 		}
770 
771 
772 		/* Need to glue together two separate fragment descriptors */
773 		if (merge) {
774 			if (cur && fra->fr_off <= cur->fr_end) {
775 				/* Need to merge in a previous 'cur' */
776 				DPFPRINTF(("fragcache[%d]: adjacent(merge "
777 				    "%d-%d) %d-%d (%d-%d)\n",
778 				    h->ip_id, cur->fr_off, cur->fr_end, off,
779 				    max, fra->fr_off, fra->fr_end));
780 				fra->fr_off = cur->fr_off;
781 				LIST_REMOVE(cur, fr_next);
782 				uma_zfree(V_pf_frent_z, cur);
783 				cur = NULL;
784 
785 			} else if (frp && fra->fr_off <= frp->fr_end) {
786 				/* Need to merge in a modified 'frp' */
787 				KASSERT((cur == NULL), ("cur != NULL: %s",
788 				    __FUNCTION__));
789 				DPFPRINTF(("fragcache[%d]: adjacent(merge "
790 				    "%d-%d) %d-%d (%d-%d)\n",
791 				    h->ip_id, frp->fr_off, frp->fr_end, off,
792 				    max, fra->fr_off, fra->fr_end));
793 				fra->fr_off = frp->fr_off;
794 				LIST_REMOVE(frp, fr_next);
795 				uma_zfree(V_pf_frent_z, frp);
796 				frp = NULL;
797 
798 			}
799 		}
800 	}
801 
802 	if (hosed) {
803 		/*
804 		 * We must keep tracking the overall fragment even when
805 		 * we're going to drop it anyway so that we know when to
806 		 * free the overall descriptor.  Thus we drop the frag late.
807 		 */
808 		goto drop_fragment;
809 	}
810 
811 
812  pass:
813 	/* Update maximum data size */
814 	if ((*frag)->fr_max < max)
815 		(*frag)->fr_max = max;
816 
817 	/* This is the last segment */
818 	if (!mff)
819 		(*frag)->fr_flags |= PFFRAG_SEENLAST;
820 
821 	/* Check if we are completely reassembled */
822 	if (((*frag)->fr_flags & PFFRAG_SEENLAST) &&
823 	    LIST_FIRST(&(*frag)->fr_queue)->fr_off == 0 &&
824 	    LIST_FIRST(&(*frag)->fr_queue)->fr_end == (*frag)->fr_max) {
825 		/* Remove from fragment queue */
826 		DPFPRINTF(("fragcache[%d]: done 0-%d\n", h->ip_id,
827 		    (*frag)->fr_max));
828 		pf_free_fragment(*frag);
829 		*frag = NULL;
830 	}
831 
832 	return (m);
833 
834  no_mem:
835 	*nomem = 1;
836 
837 	/* Still need to pay attention to !IP_MF */
838 	if (!mff && *frag != NULL)
839 		(*frag)->fr_flags |= PFFRAG_SEENLAST;
840 
841 	m_freem(m);
842 	return (NULL);
843 
844  drop_fragment:
845 
846 	/* Still need to pay attention to !IP_MF */
847 	if (!mff && *frag != NULL)
848 		(*frag)->fr_flags |= PFFRAG_SEENLAST;
849 
850 	if (drop) {
851 		/* This fragment has been deemed bad.  Don't reass */
852 		if (((*frag)->fr_flags & PFFRAG_DROP) == 0)
853 			DPFPRINTF(("fragcache[%d]: dropping overall fragment\n",
854 			    h->ip_id));
855 		(*frag)->fr_flags |= PFFRAG_DROP;
856 	}
857 
858 	m_freem(m);
859 	return (NULL);
860 }
861 
862 int
863 pf_normalize_ip(struct mbuf **m0, int dir, struct pfi_kif *kif, u_short *reason,
864     struct pf_pdesc *pd)
865 {
866 	struct mbuf		*m = *m0;
867 	struct pf_rule		*r;
868 	struct pf_frent		*frent;
869 	struct pf_fragment	*frag = NULL;
870 	struct ip		*h = mtod(m, struct ip *);
871 	int			 mff = (ntohs(h->ip_off) & IP_MF);
872 	int			 hlen = h->ip_hl << 2;
873 	u_int16_t		 fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
874 	u_int16_t		 max;
875 	int			 ip_len;
876 	int			 ip_off;
877 	int			 tag = -1;
878 
879 	PF_RULES_RASSERT();
880 
881 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
882 	while (r != NULL) {
883 		r->evaluations++;
884 		if (pfi_kif_match(r->kif, kif) == r->ifnot)
885 			r = r->skip[PF_SKIP_IFP].ptr;
886 		else if (r->direction && r->direction != dir)
887 			r = r->skip[PF_SKIP_DIR].ptr;
888 		else if (r->af && r->af != AF_INET)
889 			r = r->skip[PF_SKIP_AF].ptr;
890 		else if (r->proto && r->proto != h->ip_p)
891 			r = r->skip[PF_SKIP_PROTO].ptr;
892 		else if (PF_MISMATCHAW(&r->src.addr,
893 		    (struct pf_addr *)&h->ip_src.s_addr, AF_INET,
894 		    r->src.neg, kif, M_GETFIB(m)))
895 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
896 		else if (PF_MISMATCHAW(&r->dst.addr,
897 		    (struct pf_addr *)&h->ip_dst.s_addr, AF_INET,
898 		    r->dst.neg, NULL, M_GETFIB(m)))
899 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
900 		else if (r->match_tag && !pf_match_tag(m, r, &tag,
901 		    pd->pf_mtag ? pd->pf_mtag->tag : 0))
902 			r = TAILQ_NEXT(r, entries);
903 		else
904 			break;
905 	}
906 
907 	if (r == NULL || r->action == PF_NOSCRUB)
908 		return (PF_PASS);
909 	else {
910 		r->packets[dir == PF_OUT]++;
911 		r->bytes[dir == PF_OUT] += pd->tot_len;
912 	}
913 
914 	/* Check for illegal packets */
915 	if (hlen < (int)sizeof(struct ip))
916 		goto drop;
917 
918 	if (hlen > ntohs(h->ip_len))
919 		goto drop;
920 
921 	/* Clear IP_DF if the rule uses the no-df option */
922 	if (r->rule_flag & PFRULE_NODF && h->ip_off & htons(IP_DF)) {
923 		u_int16_t ip_off = h->ip_off;
924 
925 		h->ip_off &= htons(~IP_DF);
926 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
927 	}
928 
929 	/* We will need other tests here */
930 	if (!fragoff && !mff)
931 		goto no_fragment;
932 
933 	/* We're dealing with a fragment now. Don't allow fragments
934 	 * with IP_DF to enter the cache. If the flag was cleared by
935 	 * no-df above, fine. Otherwise drop it.
936 	 */
937 	if (h->ip_off & htons(IP_DF)) {
938 		DPFPRINTF(("IP_DF\n"));
939 		goto bad;
940 	}
941 
942 	ip_len = ntohs(h->ip_len) - hlen;
943 	ip_off = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
944 
945 	/* All fragments are 8 byte aligned */
946 	if (mff && (ip_len & 0x7)) {
947 		DPFPRINTF(("mff and %d\n", ip_len));
948 		goto bad;
949 	}
950 
951 	/* Respect maximum length */
952 	if (fragoff + ip_len > IP_MAXPACKET) {
953 		DPFPRINTF(("max packet %d\n", fragoff + ip_len));
954 		goto bad;
955 	}
956 	max = fragoff + ip_len;
957 
958 	if ((r->rule_flag & (PFRULE_FRAGCROP|PFRULE_FRAGDROP)) == 0) {
959 
960 		/* Fully buffer all of the fragments */
961 		PF_FRAG_LOCK();
962 		frag = pf_find_fragment(h, &V_pf_frag_tree);
963 
964 		/* Check if we saw the last fragment already */
965 		if (frag != NULL && (frag->fr_flags & PFFRAG_SEENLAST) &&
966 		    max > frag->fr_max)
967 			goto bad;
968 
969 		/* Get an entry for the fragment queue */
970 		frent = uma_zalloc(V_pf_frent_z, M_NOWAIT);
971 		if (frent == NULL) {
972 			PF_FRAG_UNLOCK();
973 			REASON_SET(reason, PFRES_MEMORY);
974 			return (PF_DROP);
975 		}
976 		frent->fr_ip = h;
977 		frent->fr_m = m;
978 
979 		/* Might return a completely reassembled mbuf, or NULL */
980 		DPFPRINTF(("reass frag %d @ %d-%d\n", h->ip_id, fragoff, max));
981 		*m0 = m = pf_reassemble(m0, &frag, frent, mff);
982 		PF_FRAG_UNLOCK();
983 
984 		if (m == NULL)
985 			return (PF_DROP);
986 
987 		if (frag != NULL && (frag->fr_flags & PFFRAG_DROP))
988 			goto drop;
989 
990 		h = mtod(m, struct ip *);
991 	} else {
992 		/* non-buffering fragment cache (drops or masks overlaps) */
993 		int	nomem = 0;
994 
995 		if (dir == PF_OUT && pd->pf_mtag &&
996 		    pd->pf_mtag->flags & PF_TAG_FRAGCACHE) {
997 			/*
998 			 * Already passed the fragment cache in the
999 			 * input direction.  If we continued, it would
1000 			 * appear to be a dup and would be dropped.
1001 			 */
1002 			goto fragment_pass;
1003 		}
1004 
1005 		PF_FRAG_LOCK();
1006 		frag = pf_find_fragment(h, &V_pf_cache_tree);
1007 
1008 		/* Check if we saw the last fragment already */
1009 		if (frag != NULL && (frag->fr_flags & PFFRAG_SEENLAST) &&
1010 		    max > frag->fr_max) {
1011 			if (r->rule_flag & PFRULE_FRAGDROP)
1012 				frag->fr_flags |= PFFRAG_DROP;
1013 			goto bad;
1014 		}
1015 
1016 		*m0 = m = pf_fragcache(m0, h, &frag, mff,
1017 		    (r->rule_flag & PFRULE_FRAGDROP) ? 1 : 0, &nomem);
1018 		PF_FRAG_UNLOCK();
1019 		if (m == NULL) {
1020 			if (nomem)
1021 				goto no_mem;
1022 			goto drop;
1023 		}
1024 
1025 		if (dir == PF_IN) {
1026 			/* Use mtag from copied and trimmed mbuf chain. */
1027 			pd->pf_mtag = pf_get_mtag(m);
1028 			if (pd->pf_mtag == NULL) {
1029 				m_freem(m);
1030 				*m0 = NULL;
1031 				goto no_mem;
1032 			}
1033 			pd->pf_mtag->flags |= PF_TAG_FRAGCACHE;
1034 		}
1035 
1036 		if (frag != NULL && (frag->fr_flags & PFFRAG_DROP))
1037 			goto drop;
1038 		goto fragment_pass;
1039 	}
1040 
1041  no_fragment:
1042 	/* At this point, only IP_DF is allowed in ip_off */
1043 	if (h->ip_off & ~htons(IP_DF)) {
1044 		u_int16_t ip_off = h->ip_off;
1045 
1046 		h->ip_off &= htons(IP_DF);
1047 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
1048 	}
1049 
1050 	/* not missing a return here */
1051 
1052  fragment_pass:
1053 	pf_scrub_ip(&m, r->rule_flag, r->min_ttl, r->set_tos);
1054 
1055 	if ((r->rule_flag & (PFRULE_FRAGCROP|PFRULE_FRAGDROP)) == 0)
1056 		pd->flags |= PFDESC_IP_REAS;
1057 	return (PF_PASS);
1058 
1059  no_mem:
1060 	REASON_SET(reason, PFRES_MEMORY);
1061 	if (r != NULL && r->log)
1062 		PFLOG_PACKET(kif, m, AF_INET, dir, *reason, r, NULL, NULL, pd,
1063 		    1);
1064 	return (PF_DROP);
1065 
1066  drop:
1067 	REASON_SET(reason, PFRES_NORM);
1068 	if (r != NULL && r->log)
1069 		PFLOG_PACKET(kif, m, AF_INET, dir, *reason, r, NULL, NULL, pd,
1070 		    1);
1071 	return (PF_DROP);
1072 
1073  bad:
1074 	DPFPRINTF(("dropping bad fragment\n"));
1075 
1076 	/* Free associated fragments */
1077 	if (frag != NULL) {
1078 		pf_free_fragment(frag);
1079 		PF_FRAG_UNLOCK();
1080 	}
1081 
1082 	REASON_SET(reason, PFRES_FRAG);
1083 	if (r != NULL && r->log)
1084 		PFLOG_PACKET(kif, m, AF_INET, dir, *reason, r, NULL, NULL, pd,
1085 		    1);
1086 
1087 	return (PF_DROP);
1088 }
1089 #endif
1090 
1091 #ifdef INET6
1092 int
1093 pf_normalize_ip6(struct mbuf **m0, int dir, struct pfi_kif *kif,
1094     u_short *reason, struct pf_pdesc *pd)
1095 {
1096 	struct mbuf		*m = *m0;
1097 	struct pf_rule		*r;
1098 	struct ip6_hdr		*h = mtod(m, struct ip6_hdr *);
1099 	int			 off;
1100 	struct ip6_ext		 ext;
1101 	struct ip6_opt		 opt;
1102 	struct ip6_opt_jumbo	 jumbo;
1103 	struct ip6_frag		 frag;
1104 	u_int32_t		 jumbolen = 0, plen;
1105 	u_int16_t		 fragoff = 0;
1106 	int			 optend;
1107 	int			 ooff;
1108 	u_int8_t		 proto;
1109 	int			 terminal;
1110 
1111 	PF_RULES_RASSERT();
1112 
1113 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1114 	while (r != NULL) {
1115 		r->evaluations++;
1116 		if (pfi_kif_match(r->kif, kif) == r->ifnot)
1117 			r = r->skip[PF_SKIP_IFP].ptr;
1118 		else if (r->direction && r->direction != dir)
1119 			r = r->skip[PF_SKIP_DIR].ptr;
1120 		else if (r->af && r->af != AF_INET6)
1121 			r = r->skip[PF_SKIP_AF].ptr;
1122 #if 0 /* header chain! */
1123 		else if (r->proto && r->proto != h->ip6_nxt)
1124 			r = r->skip[PF_SKIP_PROTO].ptr;
1125 #endif
1126 		else if (PF_MISMATCHAW(&r->src.addr,
1127 		    (struct pf_addr *)&h->ip6_src, AF_INET6,
1128 		    r->src.neg, kif, M_GETFIB(m)))
1129 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
1130 		else if (PF_MISMATCHAW(&r->dst.addr,
1131 		    (struct pf_addr *)&h->ip6_dst, AF_INET6,
1132 		    r->dst.neg, NULL, M_GETFIB(m)))
1133 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
1134 		else
1135 			break;
1136 	}
1137 
1138 	if (r == NULL || r->action == PF_NOSCRUB)
1139 		return (PF_PASS);
1140 	else {
1141 		r->packets[dir == PF_OUT]++;
1142 		r->bytes[dir == PF_OUT] += pd->tot_len;
1143 	}
1144 
1145 	/* Check for illegal packets */
1146 	if (sizeof(struct ip6_hdr) + IPV6_MAXPACKET < m->m_pkthdr.len)
1147 		goto drop;
1148 
1149 	off = sizeof(struct ip6_hdr);
1150 	proto = h->ip6_nxt;
1151 	terminal = 0;
1152 	do {
1153 		switch (proto) {
1154 		case IPPROTO_FRAGMENT:
1155 			goto fragment;
1156 			break;
1157 		case IPPROTO_AH:
1158 		case IPPROTO_ROUTING:
1159 		case IPPROTO_DSTOPTS:
1160 			if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL,
1161 			    NULL, AF_INET6))
1162 				goto shortpkt;
1163 			if (proto == IPPROTO_AH)
1164 				off += (ext.ip6e_len + 2) * 4;
1165 			else
1166 				off += (ext.ip6e_len + 1) * 8;
1167 			proto = ext.ip6e_nxt;
1168 			break;
1169 		case IPPROTO_HOPOPTS:
1170 			if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL,
1171 			    NULL, AF_INET6))
1172 				goto shortpkt;
1173 			optend = off + (ext.ip6e_len + 1) * 8;
1174 			ooff = off + sizeof(ext);
1175 			do {
1176 				if (!pf_pull_hdr(m, ooff, &opt.ip6o_type,
1177 				    sizeof(opt.ip6o_type), NULL, NULL,
1178 				    AF_INET6))
1179 					goto shortpkt;
1180 				if (opt.ip6o_type == IP6OPT_PAD1) {
1181 					ooff++;
1182 					continue;
1183 				}
1184 				if (!pf_pull_hdr(m, ooff, &opt, sizeof(opt),
1185 				    NULL, NULL, AF_INET6))
1186 					goto shortpkt;
1187 				if (ooff + sizeof(opt) + opt.ip6o_len > optend)
1188 					goto drop;
1189 				switch (opt.ip6o_type) {
1190 				case IP6OPT_JUMBO:
1191 					if (h->ip6_plen != 0)
1192 						goto drop;
1193 					if (!pf_pull_hdr(m, ooff, &jumbo,
1194 					    sizeof(jumbo), NULL, NULL,
1195 					    AF_INET6))
1196 						goto shortpkt;
1197 					memcpy(&jumbolen, jumbo.ip6oj_jumbo_len,
1198 					    sizeof(jumbolen));
1199 					jumbolen = ntohl(jumbolen);
1200 					if (jumbolen <= IPV6_MAXPACKET)
1201 						goto drop;
1202 					if (sizeof(struct ip6_hdr) + jumbolen !=
1203 					    m->m_pkthdr.len)
1204 						goto drop;
1205 					break;
1206 				default:
1207 					break;
1208 				}
1209 				ooff += sizeof(opt) + opt.ip6o_len;
1210 			} while (ooff < optend);
1211 
1212 			off = optend;
1213 			proto = ext.ip6e_nxt;
1214 			break;
1215 		default:
1216 			terminal = 1;
1217 			break;
1218 		}
1219 	} while (!terminal);
1220 
1221 	/* jumbo payload option must be present, or plen > 0 */
1222 	if (ntohs(h->ip6_plen) == 0)
1223 		plen = jumbolen;
1224 	else
1225 		plen = ntohs(h->ip6_plen);
1226 	if (plen == 0)
1227 		goto drop;
1228 	if (sizeof(struct ip6_hdr) + plen > m->m_pkthdr.len)
1229 		goto shortpkt;
1230 
1231 	pf_scrub_ip6(&m, r->min_ttl);
1232 
1233 	return (PF_PASS);
1234 
1235  fragment:
1236 	if (ntohs(h->ip6_plen) == 0 || jumbolen)
1237 		goto drop;
1238 	plen = ntohs(h->ip6_plen);
1239 
1240 	if (!pf_pull_hdr(m, off, &frag, sizeof(frag), NULL, NULL, AF_INET6))
1241 		goto shortpkt;
1242 	fragoff = ntohs(frag.ip6f_offlg & IP6F_OFF_MASK);
1243 	if (fragoff + (plen - off - sizeof(frag)) > IPV6_MAXPACKET)
1244 		goto badfrag;
1245 
1246 	/* do something about it */
1247 	/* remember to set pd->flags |= PFDESC_IP_REAS */
1248 	return (PF_PASS);
1249 
1250  shortpkt:
1251 	REASON_SET(reason, PFRES_SHORT);
1252 	if (r != NULL && r->log)
1253 		PFLOG_PACKET(kif, m, AF_INET6, dir, *reason, r, NULL, NULL, pd,
1254 		    1);
1255 	return (PF_DROP);
1256 
1257  drop:
1258 	REASON_SET(reason, PFRES_NORM);
1259 	if (r != NULL && r->log)
1260 		PFLOG_PACKET(kif, m, AF_INET6, dir, *reason, r, NULL, NULL, pd,
1261 		    1);
1262 	return (PF_DROP);
1263 
1264  badfrag:
1265 	REASON_SET(reason, PFRES_FRAG);
1266 	if (r != NULL && r->log)
1267 		PFLOG_PACKET(kif, m, AF_INET6, dir, *reason, r, NULL, NULL, pd,
1268 		    1);
1269 	return (PF_DROP);
1270 }
1271 #endif /* INET6 */
1272 
1273 int
1274 pf_normalize_tcp(int dir, struct pfi_kif *kif, struct mbuf *m, int ipoff,
1275     int off, void *h, struct pf_pdesc *pd)
1276 {
1277 	struct pf_rule	*r, *rm = NULL;
1278 	struct tcphdr	*th = pd->hdr.tcp;
1279 	int		 rewrite = 0;
1280 	u_short		 reason;
1281 	u_int8_t	 flags;
1282 	sa_family_t	 af = pd->af;
1283 
1284 	PF_RULES_RASSERT();
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, M_GETFIB(m)))
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, M_GETFIB(m)))
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, pd->af))
1375 		rewrite = 1;
1376 
1377 	/* copy back packet headers if we sanitized */
1378 	if (rewrite)
1379 		m_copyback(m, off, sizeof(*th), (caddr_t)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, m, AF_INET, dir, reason, r, NULL, NULL, pd,
1387 		    1);
1388 	return (PF_DROP);
1389 }
1390 
1391 int
1392 pf_normalize_tcp_init(struct mbuf *m, int off, struct pf_pdesc *pd,
1393     struct tcphdr *th, struct pf_state_peer *src, 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 	    ("pf_normalize_tcp_init: src->scrub != NULL"));
1401 
1402 	src->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT);
1403 	if (src->scrub == NULL)
1404 		return (1);
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(arc4random());
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 		uma_zfree(V_pf_state_scrub_z, state->src.scrub);
1479 	if (state->dst.scrub)
1480 		uma_zfree(V_pf_state_scrub_z, 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, tsecr;
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 	    ("%s: src->scrub && dst->scrub!", __func__));
1500 
1501 	/*
1502 	 * Enforce the minimum TTL seen for this connection.  Negate a common
1503 	 * technique to evade an intrusion detection system and confuse
1504 	 * firewall state code.
1505 	 */
1506 	switch (pd->af) {
1507 #ifdef INET
1508 	case AF_INET: {
1509 		if (src->scrub) {
1510 			struct ip *h = mtod(m, struct ip *);
1511 			if (h->ip_ttl > src->scrub->pfss_ttl)
1512 				src->scrub->pfss_ttl = h->ip_ttl;
1513 			h->ip_ttl = src->scrub->pfss_ttl;
1514 		}
1515 		break;
1516 	}
1517 #endif /* INET */
1518 #ifdef INET6
1519 	case AF_INET6: {
1520 		if (src->scrub) {
1521 			struct ip6_hdr *h = mtod(m, struct ip6_hdr *);
1522 			if (h->ip6_hlim > src->scrub->pfss_ttl)
1523 				src->scrub->pfss_ttl = h->ip6_hlim;
1524 			h->ip6_hlim = src->scrub->pfss_ttl;
1525 		}
1526 		break;
1527 	}
1528 #endif /* INET6 */
1529 	}
1530 
1531 	if (th->th_off > (sizeof(struct tcphdr) >> 2) &&
1532 	    ((src->scrub && (src->scrub->pfss_flags & PFSS_TIMESTAMP)) ||
1533 	    (dst->scrub && (dst->scrub->pfss_flags & PFSS_TIMESTAMP))) &&
1534 	    pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) {
1535 		/* Diddle with TCP options */
1536 		int hlen;
1537 		opt = hdr + sizeof(struct tcphdr);
1538 		hlen = (th->th_off << 2) - sizeof(struct tcphdr);
1539 		while (hlen >= TCPOLEN_TIMESTAMP) {
1540 			switch (*opt) {
1541 			case TCPOPT_EOL:	/* FALLTHROUGH */
1542 			case TCPOPT_NOP:
1543 				opt++;
1544 				hlen--;
1545 				break;
1546 			case TCPOPT_TIMESTAMP:
1547 				/* Modulate the timestamps.  Can be used for
1548 				 * NAT detection, OS uptime determination or
1549 				 * reboot detection.
1550 				 */
1551 
1552 				if (got_ts) {
1553 					/* Huh?  Multiple timestamps!? */
1554 					if (V_pf_status.debug >= PF_DEBUG_MISC) {
1555 						DPFPRINTF(("multiple TS??"));
1556 						pf_print_state(state);
1557 						printf("\n");
1558 					}
1559 					REASON_SET(reason, PFRES_TS);
1560 					return (PF_DROP);
1561 				}
1562 				if (opt[1] >= TCPOLEN_TIMESTAMP) {
1563 					memcpy(&tsval, &opt[2],
1564 					    sizeof(u_int32_t));
1565 					if (tsval && src->scrub &&
1566 					    (src->scrub->pfss_flags &
1567 					    PFSS_TIMESTAMP)) {
1568 						tsval = ntohl(tsval);
1569 						pf_change_a(&opt[2],
1570 						    &th->th_sum,
1571 						    htonl(tsval +
1572 						    src->scrub->pfss_ts_mod),
1573 						    0);
1574 						copyback = 1;
1575 					}
1576 
1577 					/* Modulate TS reply iff valid (!0) */
1578 					memcpy(&tsecr, &opt[6],
1579 					    sizeof(u_int32_t));
1580 					if (tsecr && dst->scrub &&
1581 					    (dst->scrub->pfss_flags &
1582 					    PFSS_TIMESTAMP)) {
1583 						tsecr = ntohl(tsecr)
1584 						    - dst->scrub->pfss_ts_mod;
1585 						pf_change_a(&opt[6],
1586 						    &th->th_sum, htonl(tsecr),
1587 						    0);
1588 						copyback = 1;
1589 					}
1590 					got_ts = 1;
1591 				}
1592 				/* FALLTHROUGH */
1593 			default:
1594 				hlen -= MAX(opt[1], 2);
1595 				opt += MAX(opt[1], 2);
1596 				break;
1597 			}
1598 		}
1599 		if (copyback) {
1600 			/* Copyback the options, caller copys back header */
1601 			*writeback = 1;
1602 			m_copyback(m, off + sizeof(struct tcphdr),
1603 			    (th->th_off << 2) - sizeof(struct tcphdr), hdr +
1604 			    sizeof(struct tcphdr));
1605 		}
1606 	}
1607 
1608 
1609 	/*
1610 	 * Must invalidate PAWS checks on connections idle for too long.
1611 	 * The fastest allowed timestamp clock is 1ms.  That turns out to
1612 	 * be about 24 days before it wraps.  XXX Right now our lowerbound
1613 	 * TS echo check only works for the first 12 days of a connection
1614 	 * when the TS has exhausted half its 32bit space
1615 	 */
1616 #define TS_MAX_IDLE	(24*24*60*60)
1617 #define TS_MAX_CONN	(12*24*60*60)	/* XXX remove when better tsecr check */
1618 
1619 	getmicrouptime(&uptime);
1620 	if (src->scrub && (src->scrub->pfss_flags & PFSS_PAWS) &&
1621 	    (uptime.tv_sec - src->scrub->pfss_last.tv_sec > TS_MAX_IDLE ||
1622 	    time_uptime - state->creation > TS_MAX_CONN))  {
1623 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
1624 			DPFPRINTF(("src idled out of PAWS\n"));
1625 			pf_print_state(state);
1626 			printf("\n");
1627 		}
1628 		src->scrub->pfss_flags = (src->scrub->pfss_flags & ~PFSS_PAWS)
1629 		    | PFSS_PAWS_IDLED;
1630 	}
1631 	if (dst->scrub && (dst->scrub->pfss_flags & PFSS_PAWS) &&
1632 	    uptime.tv_sec - dst->scrub->pfss_last.tv_sec > TS_MAX_IDLE) {
1633 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
1634 			DPFPRINTF(("dst idled out of PAWS\n"));
1635 			pf_print_state(state);
1636 			printf("\n");
1637 		}
1638 		dst->scrub->pfss_flags = (dst->scrub->pfss_flags & ~PFSS_PAWS)
1639 		    | PFSS_PAWS_IDLED;
1640 	}
1641 
1642 	if (got_ts && src->scrub && dst->scrub &&
1643 	    (src->scrub->pfss_flags & PFSS_PAWS) &&
1644 	    (dst->scrub->pfss_flags & PFSS_PAWS)) {
1645 		/* Validate that the timestamps are "in-window".
1646 		 * RFC1323 describes TCP Timestamp options that allow
1647 		 * measurement of RTT (round trip time) and PAWS
1648 		 * (protection against wrapped sequence numbers).  PAWS
1649 		 * gives us a set of rules for rejecting packets on
1650 		 * long fat pipes (packets that were somehow delayed
1651 		 * in transit longer than the time it took to send the
1652 		 * full TCP sequence space of 4Gb).  We can use these
1653 		 * rules and infer a few others that will let us treat
1654 		 * the 32bit timestamp and the 32bit echoed timestamp
1655 		 * as sequence numbers to prevent a blind attacker from
1656 		 * inserting packets into a connection.
1657 		 *
1658 		 * RFC1323 tells us:
1659 		 *  - The timestamp on this packet must be greater than
1660 		 *    or equal to the last value echoed by the other
1661 		 *    endpoint.  The RFC says those will be discarded
1662 		 *    since it is a dup that has already been acked.
1663 		 *    This gives us a lowerbound on the timestamp.
1664 		 *        timestamp >= other last echoed timestamp
1665 		 *  - The timestamp will be less than or equal to
1666 		 *    the last timestamp plus the time between the
1667 		 *    last packet and now.  The RFC defines the max
1668 		 *    clock rate as 1ms.  We will allow clocks to be
1669 		 *    up to 10% fast and will allow a total difference
1670 		 *    or 30 seconds due to a route change.  And this
1671 		 *    gives us an upperbound on the timestamp.
1672 		 *        timestamp <= last timestamp + max ticks
1673 		 *    We have to be careful here.  Windows will send an
1674 		 *    initial timestamp of zero and then initialize it
1675 		 *    to a random value after the 3whs; presumably to
1676 		 *    avoid a DoS by having to call an expensive RNG
1677 		 *    during a SYN flood.  Proof MS has at least one
1678 		 *    good security geek.
1679 		 *
1680 		 *  - The TCP timestamp option must also echo the other
1681 		 *    endpoints timestamp.  The timestamp echoed is the
1682 		 *    one carried on the earliest unacknowledged segment
1683 		 *    on the left edge of the sequence window.  The RFC
1684 		 *    states that the host will reject any echoed
1685 		 *    timestamps that were larger than any ever sent.
1686 		 *    This gives us an upperbound on the TS echo.
1687 		 *        tescr <= largest_tsval
1688 		 *  - The lowerbound on the TS echo is a little more
1689 		 *    tricky to determine.  The other endpoint's echoed
1690 		 *    values will not decrease.  But there may be
1691 		 *    network conditions that re-order packets and
1692 		 *    cause our view of them to decrease.  For now the
1693 		 *    only lowerbound we can safely determine is that
1694 		 *    the TS echo will never be less than the original
1695 		 *    TS.  XXX There is probably a better lowerbound.
1696 		 *    Remove TS_MAX_CONN with better lowerbound check.
1697 		 *        tescr >= other original TS
1698 		 *
1699 		 * It is also important to note that the fastest
1700 		 * timestamp clock of 1ms will wrap its 32bit space in
1701 		 * 24 days.  So we just disable TS checking after 24
1702 		 * days of idle time.  We actually must use a 12d
1703 		 * connection limit until we can come up with a better
1704 		 * lowerbound to the TS echo check.
1705 		 */
1706 		struct timeval delta_ts;
1707 		int ts_fudge;
1708 
1709 
1710 		/*
1711 		 * PFTM_TS_DIFF is how many seconds of leeway to allow
1712 		 * a host's timestamp.  This can happen if the previous
1713 		 * packet got delayed in transit for much longer than
1714 		 * this packet.
1715 		 */
1716 		if ((ts_fudge = state->rule.ptr->timeout[PFTM_TS_DIFF]) == 0)
1717 			ts_fudge = V_pf_default_rule.timeout[PFTM_TS_DIFF];
1718 
1719 		/* Calculate max ticks since the last timestamp */
1720 #define TS_MAXFREQ	1100		/* RFC max TS freq of 1Khz + 10% skew */
1721 #define TS_MICROSECS	1000000		/* microseconds per second */
1722 		delta_ts = uptime;
1723 		timevalsub(&delta_ts, &src->scrub->pfss_last);
1724 		tsval_from_last = (delta_ts.tv_sec + ts_fudge) * TS_MAXFREQ;
1725 		tsval_from_last += delta_ts.tv_usec / (TS_MICROSECS/TS_MAXFREQ);
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: %u  tsecr: %u  +ticks: %u  "
1747 			    "idle: %jus %lums\n",
1748 			    tsval, tsecr, tsval_from_last,
1749 			    (uintmax_t)delta_ts.tv_sec,
1750 			    delta_ts.tv_usec / 1000));
1751 			DPFPRINTF((" src->tsval: %u  tsecr: %u\n",
1752 			    src->scrub->pfss_tsval, src->scrub->pfss_tsecr));
1753 			DPFPRINTF((" dst->tsval: %u  tsecr: %u  tsval0: %u"
1754 			    "\n", dst->scrub->pfss_tsval,
1755 			    dst->scrub->pfss_tsecr, dst->scrub->pfss_tsval0));
1756 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
1757 				pf_print_state(state);
1758 				pf_print_flags(th->th_flags);
1759 				printf("\n");
1760 			}
1761 			REASON_SET(reason, PFRES_TS);
1762 			return (PF_DROP);
1763 		}
1764 
1765 		/* XXX I'd really like to require tsecr but it's optional */
1766 
1767 	} else if (!got_ts && (th->th_flags & TH_RST) == 0 &&
1768 	    ((src->state == TCPS_ESTABLISHED && dst->state == TCPS_ESTABLISHED)
1769 	    || pd->p_len > 0 || (th->th_flags & TH_SYN)) &&
1770 	    src->scrub && dst->scrub &&
1771 	    (src->scrub->pfss_flags & PFSS_PAWS) &&
1772 	    (dst->scrub->pfss_flags & PFSS_PAWS)) {
1773 		/* Didn't send a timestamp.  Timestamps aren't really useful
1774 		 * when:
1775 		 *  - connection opening or closing (often not even sent).
1776 		 *    but we must not let an attacker to put a FIN on a
1777 		 *    data packet to sneak it through our ESTABLISHED check.
1778 		 *  - on a TCP reset.  RFC suggests not even looking at TS.
1779 		 *  - on an empty ACK.  The TS will not be echoed so it will
1780 		 *    probably not help keep the RTT calculation in sync and
1781 		 *    there isn't as much danger when the sequence numbers
1782 		 *    got wrapped.  So some stacks don't include TS on empty
1783 		 *    ACKs :-(
1784 		 *
1785 		 * To minimize the disruption to mostly RFC1323 conformant
1786 		 * stacks, we will only require timestamps on data packets.
1787 		 *
1788 		 * And what do ya know, we cannot require timestamps on data
1789 		 * packets.  There appear to be devices that do legitimate
1790 		 * TCP connection hijacking.  There are HTTP devices that allow
1791 		 * a 3whs (with timestamps) and then buffer the HTTP request.
1792 		 * If the intermediate device has the HTTP response cache, it
1793 		 * will spoof the response but not bother timestamping its
1794 		 * packets.  So we can look for the presence of a timestamp in
1795 		 * the first data packet and if there, require it in all future
1796 		 * packets.
1797 		 */
1798 
1799 		if (pd->p_len > 0 && (src->scrub->pfss_flags & PFSS_DATA_TS)) {
1800 			/*
1801 			 * Hey!  Someone tried to sneak a packet in.  Or the
1802 			 * stack changed its RFC1323 behavior?!?!
1803 			 */
1804 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
1805 				DPFPRINTF(("Did not receive expected RFC1323 "
1806 				    "timestamp\n"));
1807 				pf_print_state(state);
1808 				pf_print_flags(th->th_flags);
1809 				printf("\n");
1810 			}
1811 			REASON_SET(reason, PFRES_TS);
1812 			return (PF_DROP);
1813 		}
1814 	}
1815 
1816 
1817 	/*
1818 	 * We will note if a host sends his data packets with or without
1819 	 * timestamps.  And require all data packets to contain a timestamp
1820 	 * if the first does.  PAWS implicitly requires that all data packets be
1821 	 * timestamped.  But I think there are middle-man devices that hijack
1822 	 * TCP streams immediately after the 3whs and don't timestamp their
1823 	 * packets (seen in a WWW accelerator or cache).
1824 	 */
1825 	if (pd->p_len > 0 && src->scrub && (src->scrub->pfss_flags &
1826 	    (PFSS_TIMESTAMP|PFSS_DATA_TS|PFSS_DATA_NOTS)) == PFSS_TIMESTAMP) {
1827 		if (got_ts)
1828 			src->scrub->pfss_flags |= PFSS_DATA_TS;
1829 		else {
1830 			src->scrub->pfss_flags |= PFSS_DATA_NOTS;
1831 			if (V_pf_status.debug >= PF_DEBUG_MISC && dst->scrub &&
1832 			    (dst->scrub->pfss_flags & PFSS_TIMESTAMP)) {
1833 				/* Don't warn if other host rejected RFC1323 */
1834 				DPFPRINTF(("Broken RFC1323 stack did not "
1835 				    "timestamp data packet. Disabled PAWS "
1836 				    "security.\n"));
1837 				pf_print_state(state);
1838 				pf_print_flags(th->th_flags);
1839 				printf("\n");
1840 			}
1841 		}
1842 	}
1843 
1844 
1845 	/*
1846 	 * Update PAWS values
1847 	 */
1848 	if (got_ts && src->scrub && PFSS_TIMESTAMP == (src->scrub->pfss_flags &
1849 	    (PFSS_PAWS_IDLED|PFSS_TIMESTAMP))) {
1850 		getmicrouptime(&src->scrub->pfss_last);
1851 		if (SEQ_GEQ(tsval, src->scrub->pfss_tsval) ||
1852 		    (src->scrub->pfss_flags & PFSS_PAWS) == 0)
1853 			src->scrub->pfss_tsval = tsval;
1854 
1855 		if (tsecr) {
1856 			if (SEQ_GEQ(tsecr, src->scrub->pfss_tsecr) ||
1857 			    (src->scrub->pfss_flags & PFSS_PAWS) == 0)
1858 				src->scrub->pfss_tsecr = tsecr;
1859 
1860 			if ((src->scrub->pfss_flags & PFSS_PAWS) == 0 &&
1861 			    (SEQ_LT(tsval, src->scrub->pfss_tsval0) ||
1862 			    src->scrub->pfss_tsval0 == 0)) {
1863 				/* tsval0 MUST be the lowest timestamp */
1864 				src->scrub->pfss_tsval0 = tsval;
1865 			}
1866 
1867 			/* Only fully initialized after a TS gets echoed */
1868 			if ((src->scrub->pfss_flags & PFSS_PAWS) == 0)
1869 				src->scrub->pfss_flags |= PFSS_PAWS;
1870 		}
1871 	}
1872 
1873 	/* I have a dream....  TCP segment reassembly.... */
1874 	return (0);
1875 }
1876 
1877 static int
1878 pf_normalize_tcpopt(struct pf_rule *r, struct mbuf *m, struct tcphdr *th,
1879     int off, sa_family_t af)
1880 {
1881 	u_int16_t	*mss;
1882 	int		 thoff;
1883 	int		 opt, cnt, optlen = 0;
1884 	int		 rewrite = 0;
1885 	u_char		 opts[TCP_MAXOLEN];
1886 	u_char		*optp = opts;
1887 
1888 	thoff = th->th_off << 2;
1889 	cnt = thoff - sizeof(struct tcphdr);
1890 
1891 	if (cnt > 0 && !pf_pull_hdr(m, off + sizeof(*th), opts, cnt,
1892 	    NULL, NULL, af))
1893 		return (rewrite);
1894 
1895 	for (; cnt > 0; cnt -= optlen, optp += optlen) {
1896 		opt = optp[0];
1897 		if (opt == TCPOPT_EOL)
1898 			break;
1899 		if (opt == TCPOPT_NOP)
1900 			optlen = 1;
1901 		else {
1902 			if (cnt < 2)
1903 				break;
1904 			optlen = optp[1];
1905 			if (optlen < 2 || optlen > cnt)
1906 				break;
1907 		}
1908 		switch (opt) {
1909 		case TCPOPT_MAXSEG:
1910 			mss = (u_int16_t *)(optp + 2);
1911 			if ((ntohs(*mss)) > r->max_mss) {
1912 				th->th_sum = pf_cksum_fixup(th->th_sum,
1913 				    *mss, htons(r->max_mss), 0);
1914 				*mss = htons(r->max_mss);
1915 				rewrite = 1;
1916 			}
1917 			break;
1918 		default:
1919 			break;
1920 		}
1921 	}
1922 
1923 	if (rewrite)
1924 		m_copyback(m, off + sizeof(*th), thoff - sizeof(*th), opts);
1925 
1926 	return (rewrite);
1927 }
1928 
1929 #ifdef INET
1930 static void
1931 pf_scrub_ip(struct mbuf **m0, u_int32_t flags, u_int8_t min_ttl, u_int8_t tos)
1932 {
1933 	struct mbuf		*m = *m0;
1934 	struct ip		*h = mtod(m, struct ip *);
1935 
1936 	/* Clear IP_DF if no-df was requested */
1937 	if (flags & PFRULE_NODF && h->ip_off & htons(IP_DF)) {
1938 		u_int16_t ip_off = h->ip_off;
1939 
1940 		h->ip_off &= htons(~IP_DF);
1941 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
1942 	}
1943 
1944 	/* Enforce a minimum ttl, may cause endless packet loops */
1945 	if (min_ttl && h->ip_ttl < min_ttl) {
1946 		u_int16_t ip_ttl = h->ip_ttl;
1947 
1948 		h->ip_ttl = min_ttl;
1949 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_ttl, h->ip_ttl, 0);
1950 	}
1951 
1952 	/* Enforce tos */
1953 	if (flags & PFRULE_SET_TOS) {
1954 		u_int16_t	ov, nv;
1955 
1956 		ov = *(u_int16_t *)h;
1957 		h->ip_tos = tos;
1958 		nv = *(u_int16_t *)h;
1959 
1960 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ov, nv, 0);
1961 	}
1962 
1963 	/* random-id, but not for fragments */
1964 	if (flags & PFRULE_RANDOMID && !(h->ip_off & ~htons(IP_DF))) {
1965 		u_int16_t ip_id = h->ip_id;
1966 
1967 		h->ip_id = ip_randomid();
1968 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_id, h->ip_id, 0);
1969 	}
1970 }
1971 #endif /* INET */
1972 
1973 #ifdef INET6
1974 static void
1975 pf_scrub_ip6(struct mbuf **m0, u_int8_t min_ttl)
1976 {
1977 	struct mbuf		*m = *m0;
1978 	struct ip6_hdr		*h = mtod(m, struct ip6_hdr *);
1979 
1980 	/* Enforce a minimum ttl, may cause endless packet loops */
1981 	if (min_ttl && h->ip6_hlim < min_ttl)
1982 		h->ip6_hlim = min_ttl;
1983 }
1984 #endif
1985