xref: /freebsd/sys/netpfil/pf/pf_norm.c (revision ec0e626bafb335b30c499d06066997f54b10c092)
1 /*-
2  * Copyright 2001 Niels Provos <provos@citi.umich.edu>
3  * Copyright 2011 Alexander Bluhm <bluhm@openbsd.org>
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
18  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
19  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
20  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
21  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
22  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
23  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
24  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25  *
26  *	$OpenBSD: pf_norm.c,v 1.114 2009/01/29 14:11:45 henning Exp $
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include "opt_inet.h"
33 #include "opt_inet6.h"
34 #include "opt_pf.h"
35 
36 #include <sys/param.h>
37 #include <sys/lock.h>
38 #include <sys/mbuf.h>
39 #include <sys/mutex.h>
40 #include <sys/refcount.h>
41 #include <sys/rwlock.h>
42 #include <sys/socket.h>
43 
44 #include <net/if.h>
45 #include <net/vnet.h>
46 #include <net/pfvar.h>
47 #include <net/if_pflog.h>
48 
49 #include <netinet/in.h>
50 #include <netinet/ip.h>
51 #include <netinet/ip_var.h>
52 #include <netinet6/ip6_var.h>
53 #include <netinet/tcp.h>
54 #include <netinet/tcp_fsm.h>
55 #include <netinet/tcp_seq.h>
56 
57 #ifdef INET6
58 #include <netinet/ip6.h>
59 #endif /* INET6 */
60 
61 struct pf_frent {
62 	TAILQ_ENTRY(pf_frent)	fr_next;
63 	struct mbuf	*fe_m;
64 	uint16_t	fe_hdrlen;	/* ipv4 header lenght with ip options
65 					   ipv6, extension, fragment header */
66 	uint16_t	fe_extoff;	/* last extension header offset or 0 */
67 	uint16_t	fe_len;		/* fragment length */
68 	uint16_t	fe_off;		/* fragment offset */
69 	uint16_t	fe_mff;		/* more fragment flag */
70 };
71 
72 struct pf_fragment_cmp {
73 	struct pf_addr	frc_src;
74 	struct pf_addr	frc_dst;
75 	uint32_t	frc_id;
76 	sa_family_t	frc_af;
77 	uint8_t		frc_proto;
78 	uint8_t		frc_direction;
79 };
80 
81 struct pf_fragment {
82 	struct pf_fragment_cmp	fr_key;
83 #define fr_src	fr_key.frc_src
84 #define fr_dst	fr_key.frc_dst
85 #define fr_id	fr_key.frc_id
86 #define fr_af	fr_key.frc_af
87 #define fr_proto	fr_key.frc_proto
88 #define fr_direction	fr_key.frc_direction
89 
90 	RB_ENTRY(pf_fragment) fr_entry;
91 	TAILQ_ENTRY(pf_fragment) frag_next;
92 	uint8_t		fr_flags;	/* status flags */
93 #define PFFRAG_SEENLAST		0x0001	/* Seen the last fragment for this */
94 #define PFFRAG_NOBUFFER		0x0002	/* Non-buffering fragment cache */
95 #define PFFRAG_DROP		0x0004	/* Drop all fragments */
96 #define BUFFER_FRAGMENTS(fr)	(!((fr)->fr_flags & PFFRAG_NOBUFFER))
97 	uint16_t	fr_max;		/* fragment data max */
98 	uint32_t	fr_timeout;
99 	uint16_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 	uint16_t	ft_hdrlen;	/* header length of reassembled pkt */
105 	uint16_t	ft_extoff;	/* last extension header offset or 0 */
106 	uint16_t	ft_maxlen;	/* maximum fragment payload length */
107 };
108 
109 static struct mtx pf_frag_mtx;
110 #define PF_FRAG_LOCK()		mtx_lock(&pf_frag_mtx)
111 #define PF_FRAG_UNLOCK()	mtx_unlock(&pf_frag_mtx)
112 #define PF_FRAG_ASSERT()	mtx_assert(&pf_frag_mtx, MA_OWNED)
113 
114 VNET_DEFINE(uma_zone_t, pf_state_scrub_z);	/* XXX: shared with pfsync */
115 
116 static VNET_DEFINE(uma_zone_t, pf_frent_z);
117 #define	V_pf_frent_z	VNET(pf_frent_z)
118 static VNET_DEFINE(uma_zone_t, pf_frag_z);
119 #define	V_pf_frag_z	VNET(pf_frag_z)
120 
121 TAILQ_HEAD(pf_fragqueue, pf_fragment);
122 TAILQ_HEAD(pf_cachequeue, pf_fragment);
123 static VNET_DEFINE(struct pf_fragqueue,	pf_fragqueue);
124 #define	V_pf_fragqueue			VNET(pf_fragqueue)
125 static VNET_DEFINE(struct pf_cachequeue,	pf_cachequeue);
126 #define	V_pf_cachequeue			VNET(pf_cachequeue)
127 RB_HEAD(pf_frag_tree, pf_fragment);
128 static VNET_DEFINE(struct pf_frag_tree,	pf_frag_tree);
129 #define	V_pf_frag_tree			VNET(pf_frag_tree)
130 static VNET_DEFINE(struct pf_frag_tree,	pf_cache_tree);
131 #define	V_pf_cache_tree			VNET(pf_cache_tree)
132 static int		 pf_frag_compare(struct pf_fragment *,
133 			    struct pf_fragment *);
134 static RB_PROTOTYPE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
135 static RB_GENERATE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
136 
137 static void	pf_flush_fragments(void);
138 static void	pf_free_fragment(struct pf_fragment *);
139 static void	pf_remove_fragment(struct pf_fragment *);
140 static int	pf_normalize_tcpopt(struct pf_rule *, struct mbuf *,
141 		    struct tcphdr *, int, sa_family_t);
142 static struct pf_frent *pf_create_fragment(u_short *);
143 static struct pf_fragment *pf_find_fragment(struct pf_fragment_cmp *key,
144 		    struct pf_frag_tree *tree);
145 static struct pf_fragment *pf_fillup_fragment(struct pf_fragment_cmp *,
146 		    struct pf_frent *, u_short *);
147 static int	pf_isfull_fragment(struct pf_fragment *);
148 static struct mbuf *pf_join_fragment(struct pf_fragment *);
149 #ifdef INET
150 static void	pf_scrub_ip(struct mbuf **, uint32_t, uint8_t, uint8_t);
151 static int	pf_reassemble(struct mbuf **, struct ip *, int, u_short *);
152 static struct mbuf *pf_fragcache(struct mbuf **, struct ip*,
153 		    struct pf_fragment **, int, int, int *);
154 #endif	/* INET */
155 #ifdef INET6
156 static int	pf_reassemble6(struct mbuf **, struct ip6_hdr *,
157 		    struct ip6_frag *, uint16_t, uint16_t, int, u_short *);
158 static void	pf_scrub_ip6(struct mbuf **, uint8_t);
159 #endif	/* INET6 */
160 
161 #define	DPFPRINTF(x) do {				\
162 	if (V_pf_status.debug >= PF_DEBUG_MISC) {	\
163 		printf("%s: ", __func__);		\
164 		printf x ;				\
165 	}						\
166 } while(0)
167 
168 #ifdef INET
169 static void
170 pf_ip2key(struct ip *ip, int dir, struct pf_fragment_cmp *key)
171 {
172 
173 	key->frc_src.v4 = ip->ip_src;
174 	key->frc_dst.v4 = ip->ip_dst;
175 	key->frc_af = AF_INET;
176 	key->frc_proto = ip->ip_p;
177 	key->frc_id = ip->ip_id;
178 	key->frc_direction = dir;
179 }
180 #endif	/* INET */
181 
182 void
183 pf_normalize_init(void)
184 {
185 
186 	V_pf_frag_z = uma_zcreate("pf frags", sizeof(struct pf_fragment),
187 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
188 	V_pf_frent_z = uma_zcreate("pf frag entries", sizeof(struct pf_frent),
189 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
190 	V_pf_state_scrub_z = uma_zcreate("pf state scrubs",
191 	    sizeof(struct pf_state_scrub),  NULL, NULL, NULL, NULL,
192 	    UMA_ALIGN_PTR, 0);
193 
194 	V_pf_limits[PF_LIMIT_FRAGS].zone = V_pf_frent_z;
195 	V_pf_limits[PF_LIMIT_FRAGS].limit = PFFRAG_FRENT_HIWAT;
196 	uma_zone_set_max(V_pf_frent_z, PFFRAG_FRENT_HIWAT);
197 	uma_zone_set_warning(V_pf_frent_z, "PF frag entries limit reached");
198 
199 	mtx_init(&pf_frag_mtx, "pf fragments", NULL, MTX_DEF);
200 
201 	TAILQ_INIT(&V_pf_fragqueue);
202 	TAILQ_INIT(&V_pf_cachequeue);
203 }
204 
205 void
206 pf_normalize_cleanup(void)
207 {
208 
209 	uma_zdestroy(V_pf_state_scrub_z);
210 	uma_zdestroy(V_pf_frent_z);
211 	uma_zdestroy(V_pf_frag_z);
212 
213 	mtx_destroy(&pf_frag_mtx);
214 }
215 
216 static int
217 pf_frag_compare(struct pf_fragment *a, struct pf_fragment *b)
218 {
219 	int	diff;
220 
221 	if ((diff = a->fr_id - b->fr_id) != 0)
222 		return (diff);
223 	if ((diff = a->fr_proto - b->fr_proto) != 0)
224 		return (diff);
225 	if ((diff = a->fr_af - b->fr_af) != 0)
226 		return (diff);
227 	if ((diff = pf_addr_cmp(&a->fr_src, &b->fr_src, a->fr_af)) != 0)
228 		return (diff);
229 	if ((diff = pf_addr_cmp(&a->fr_dst, &b->fr_dst, a->fr_af)) != 0)
230 		return (diff);
231 	return (0);
232 }
233 
234 void
235 pf_purge_expired_fragments(void)
236 {
237 	struct pf_fragment	*frag;
238 	u_int32_t		 expire = time_uptime -
239 				    V_pf_default_rule.timeout[PFTM_FRAG];
240 
241 	PF_FRAG_LOCK();
242 	while ((frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue)) != NULL) {
243 		KASSERT((BUFFER_FRAGMENTS(frag)),
244 		    ("BUFFER_FRAGMENTS(frag) == 0: %s", __FUNCTION__));
245 		if (frag->fr_timeout > expire)
246 			break;
247 
248 		DPFPRINTF(("expiring %d(%p)\n", frag->fr_id, frag));
249 		pf_free_fragment(frag);
250 	}
251 
252 	while ((frag = TAILQ_LAST(&V_pf_cachequeue, pf_cachequeue)) != NULL) {
253 		KASSERT((!BUFFER_FRAGMENTS(frag)),
254 		    ("BUFFER_FRAGMENTS(frag) != 0: %s", __FUNCTION__));
255 		if (frag->fr_timeout > expire)
256 			break;
257 
258 		DPFPRINTF(("expiring %d(%p)\n", frag->fr_id, frag));
259 		pf_free_fragment(frag);
260 		KASSERT((TAILQ_EMPTY(&V_pf_cachequeue) ||
261 		    TAILQ_LAST(&V_pf_cachequeue, pf_cachequeue) != frag),
262 		    ("!(TAILQ_EMPTY() || TAILQ_LAST() == farg): %s",
263 		    __FUNCTION__));
264 	}
265 	PF_FRAG_UNLOCK();
266 }
267 
268 /*
269  * Try to flush old fragments to make space for new ones
270  */
271 static void
272 pf_flush_fragments(void)
273 {
274 	struct pf_fragment	*frag, *cache;
275 	int			 goal;
276 
277 	PF_FRAG_ASSERT();
278 
279 	goal = uma_zone_get_cur(V_pf_frent_z) * 9 / 10;
280 	DPFPRINTF(("trying to free %d frag entriess\n", goal));
281 	while (goal < uma_zone_get_cur(V_pf_frent_z)) {
282 		frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue);
283 		if (frag)
284 			pf_free_fragment(frag);
285 		cache = TAILQ_LAST(&V_pf_cachequeue, pf_cachequeue);
286 		if (cache)
287 			pf_free_fragment(cache);
288 		if (frag == NULL && cache == NULL)
289 			break;
290 	}
291 }
292 
293 /* Frees the fragments and all associated entries */
294 static void
295 pf_free_fragment(struct pf_fragment *frag)
296 {
297 	struct pf_frent		*frent;
298 
299 	PF_FRAG_ASSERT();
300 
301 	/* Free all fragments */
302 	if (BUFFER_FRAGMENTS(frag)) {
303 		for (frent = TAILQ_FIRST(&frag->fr_queue); frent;
304 		    frent = TAILQ_FIRST(&frag->fr_queue)) {
305 			TAILQ_REMOVE(&frag->fr_queue, frent, fr_next);
306 
307 			m_freem(frent->fe_m);
308 			uma_zfree(V_pf_frent_z, frent);
309 		}
310 	} else {
311 		for (frent = TAILQ_FIRST(&frag->fr_queue); frent;
312 		    frent = TAILQ_FIRST(&frag->fr_queue)) {
313 			TAILQ_REMOVE(&frag->fr_queue, frent, fr_next);
314 
315 			KASSERT((TAILQ_EMPTY(&frag->fr_queue) ||
316 			    TAILQ_FIRST(&frag->fr_queue)->fe_off >
317 			    frent->fe_len),
318 			    ("! (TAILQ_EMPTY() || TAILQ_FIRST()->fe_off >"
319 			    " frent->fe_len): %s", __func__));
320 
321 			uma_zfree(V_pf_frent_z, frent);
322 		}
323 	}
324 
325 	pf_remove_fragment(frag);
326 }
327 
328 static struct pf_fragment *
329 pf_find_fragment(struct pf_fragment_cmp *key, struct pf_frag_tree *tree)
330 {
331 	struct pf_fragment	*frag;
332 
333 	PF_FRAG_ASSERT();
334 
335 	frag = RB_FIND(pf_frag_tree, tree, (struct pf_fragment *)key);
336 	if (frag != NULL) {
337 		/* XXX Are we sure we want to update the timeout? */
338 		frag->fr_timeout = time_uptime;
339 		if (BUFFER_FRAGMENTS(frag)) {
340 			TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next);
341 			TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next);
342 		} else {
343 			TAILQ_REMOVE(&V_pf_cachequeue, frag, frag_next);
344 			TAILQ_INSERT_HEAD(&V_pf_cachequeue, frag, frag_next);
345 		}
346 	}
347 
348 	return (frag);
349 }
350 
351 /* Removes a fragment from the fragment queue and frees the fragment */
352 static void
353 pf_remove_fragment(struct pf_fragment *frag)
354 {
355 
356 	PF_FRAG_ASSERT();
357 
358 	if (BUFFER_FRAGMENTS(frag)) {
359 		RB_REMOVE(pf_frag_tree, &V_pf_frag_tree, frag);
360 		TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next);
361 		uma_zfree(V_pf_frag_z, frag);
362 	} else {
363 		RB_REMOVE(pf_frag_tree, &V_pf_cache_tree, frag);
364 		TAILQ_REMOVE(&V_pf_cachequeue, frag, frag_next);
365 		uma_zfree(V_pf_frag_z, frag);
366 	}
367 }
368 
369 static struct pf_frent *
370 pf_create_fragment(u_short *reason)
371 {
372 	struct pf_frent *frent;
373 
374 	PF_FRAG_ASSERT();
375 
376 	frent = uma_zalloc(V_pf_frent_z, M_NOWAIT);
377 	if (frent == NULL) {
378 		pf_flush_fragments();
379 		frent = uma_zalloc(V_pf_frent_z, M_NOWAIT);
380 		if (frent == NULL) {
381 			REASON_SET(reason, PFRES_MEMORY);
382 			return (NULL);
383 		}
384 	}
385 
386 	return (frent);
387 }
388 
389 struct pf_fragment *
390 pf_fillup_fragment(struct pf_fragment_cmp *key, struct pf_frent *frent,
391 		u_short *reason)
392 {
393 	struct pf_frent		*after, *next, *prev;
394 	struct pf_fragment	*frag;
395 	uint16_t		total;
396 
397 	PF_FRAG_ASSERT();
398 
399 	/* No empty fragments. */
400 	if (frent->fe_len == 0) {
401 		DPFPRINTF(("bad fragment: len 0"));
402 		goto bad_fragment;
403 	}
404 
405 	/* All fragments are 8 byte aligned. */
406 	if (frent->fe_mff && (frent->fe_len & 0x7)) {
407 		DPFPRINTF(("bad fragment: mff and len %d", frent->fe_len));
408 		goto bad_fragment;
409 	}
410 
411 	/* Respect maximum length, IP_MAXPACKET == IPV6_MAXPACKET. */
412 	if (frent->fe_off + frent->fe_len > IP_MAXPACKET) {
413 		DPFPRINTF(("bad fragment: max packet %d",
414 		    frent->fe_off + frent->fe_len));
415 		goto bad_fragment;
416 	}
417 
418 	DPFPRINTF((key->frc_af == AF_INET ?
419 	    "reass frag %d @ %d-%d" : "reass frag %#08x @ %d-%d",
420 	    key->frc_id, frent->fe_off, frent->fe_off + frent->fe_len));
421 
422 	/* Fully buffer all of the fragments in this fragment queue. */
423 	frag = pf_find_fragment(key, &V_pf_frag_tree);
424 
425 	/* Create a new reassembly queue for this packet. */
426 	if (frag == NULL) {
427 		frag = uma_zalloc(V_pf_frag_z, M_NOWAIT);
428 		if (frag == NULL) {
429 			pf_flush_fragments();
430 			frag = uma_zalloc(V_pf_frag_z, M_NOWAIT);
431 			if (frag == NULL) {
432 				REASON_SET(reason, PFRES_MEMORY);
433 				goto drop_fragment;
434 			}
435 		}
436 
437 		*(struct pf_fragment_cmp *)frag = *key;
438 		frag->fr_timeout = time_second;
439 		frag->fr_maxlen = frent->fe_len;
440 		TAILQ_INIT(&frag->fr_queue);
441 
442 		RB_INSERT(pf_frag_tree, &V_pf_frag_tree, frag);
443 		TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next);
444 
445 		/* We do not have a previous fragment. */
446 		TAILQ_INSERT_HEAD(&frag->fr_queue, frent, fr_next);
447 
448 		return (frag);
449 	}
450 
451 	KASSERT(!TAILQ_EMPTY(&frag->fr_queue), ("!TAILQ_EMPTY()->fr_queue"));
452 
453 	/* Remember maximum fragment len for refragmentation. */
454 	if (frent->fe_len > frag->fr_maxlen)
455 		frag->fr_maxlen = frent->fe_len;
456 
457 	/* Maximum data we have seen already. */
458 	total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
459 		TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
460 
461 	/* Non terminal fragments must have more fragments flag. */
462 	if (frent->fe_off + frent->fe_len < total && !frent->fe_mff)
463 		goto bad_fragment;
464 
465 	/* Check if we saw the last fragment already. */
466 	if (!TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_mff) {
467 		if (frent->fe_off + frent->fe_len > total ||
468 		    (frent->fe_off + frent->fe_len == total && frent->fe_mff))
469 			goto bad_fragment;
470 	} else {
471 		if (frent->fe_off + frent->fe_len == total && !frent->fe_mff)
472 			goto bad_fragment;
473 	}
474 
475 	/* Find a fragment after the current one. */
476 	prev = NULL;
477 	TAILQ_FOREACH(after, &frag->fr_queue, fr_next) {
478 		if (after->fe_off > frent->fe_off)
479 			break;
480 		prev = after;
481 	}
482 
483 	KASSERT(prev != NULL || after != NULL,
484 	    ("prev != NULL || after != NULL"));
485 
486 	if (prev != NULL && prev->fe_off + prev->fe_len > frent->fe_off) {
487 		uint16_t precut;
488 
489 		precut = prev->fe_off + prev->fe_len - frent->fe_off;
490 		if (precut >= frent->fe_len)
491 			goto bad_fragment;
492 		DPFPRINTF(("overlap -%d", precut));
493 		m_adj(frent->fe_m, precut);
494 		frent->fe_off += precut;
495 		frent->fe_len -= precut;
496 	}
497 
498 	for (; after != NULL && frent->fe_off + frent->fe_len > after->fe_off;
499 	    after = next) {
500 		uint16_t aftercut;
501 
502 		aftercut = frent->fe_off + frent->fe_len - after->fe_off;
503 		DPFPRINTF(("adjust overlap %d", aftercut));
504 		if (aftercut < after->fe_len) {
505 			m_adj(after->fe_m, aftercut);
506 			after->fe_off += aftercut;
507 			after->fe_len -= aftercut;
508 			break;
509 		}
510 
511 		/* This fragment is completely overlapped, lose it. */
512 		next = TAILQ_NEXT(after, fr_next);
513 		m_freem(after->fe_m);
514 		TAILQ_REMOVE(&frag->fr_queue, after, fr_next);
515 		uma_zfree(V_pf_frent_z, after);
516 	}
517 
518 	if (prev == NULL)
519 		TAILQ_INSERT_HEAD(&frag->fr_queue, frent, fr_next);
520 	else
521 		TAILQ_INSERT_AFTER(&frag->fr_queue, prev, frent, fr_next);
522 
523 	return (frag);
524 
525 bad_fragment:
526 	REASON_SET(reason, PFRES_FRAG);
527 drop_fragment:
528 	uma_zfree(V_pf_frent_z, frent);
529 	return (NULL);
530 }
531 
532 static int
533 pf_isfull_fragment(struct pf_fragment *frag)
534 {
535 	struct pf_frent	*frent, *next;
536 	uint16_t off, total;
537 
538 	/* Check if we are completely reassembled */
539 	if (TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_mff)
540 		return (0);
541 
542 	/* Maximum data we have seen already */
543 	total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
544 		TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
545 
546 	/* Check if we have all the data */
547 	off = 0;
548 	for (frent = TAILQ_FIRST(&frag->fr_queue); frent; frent = next) {
549 		next = TAILQ_NEXT(frent, fr_next);
550 
551 		off += frent->fe_len;
552 		if (off < total && (next == NULL || next->fe_off != off)) {
553 			DPFPRINTF(("missing fragment at %d, next %d, total %d",
554 			    off, next == NULL ? -1 : next->fe_off, total));
555 			return (0);
556 		}
557 	}
558 	DPFPRINTF(("%d < %d?", off, total));
559 	if (off < total)
560 		return (0);
561 	KASSERT(off == total, ("off == total"));
562 
563 	return (1);
564 }
565 
566 static struct mbuf *
567 pf_join_fragment(struct pf_fragment *frag)
568 {
569 	struct mbuf *m, *m2;
570 	struct pf_frent	*frent, *next;
571 
572 	frent = TAILQ_FIRST(&frag->fr_queue);
573 	next = TAILQ_NEXT(frent, fr_next);
574 
575 	/* Magic from ip_input. */
576 	m = frent->fe_m;
577 	m2 = m->m_next;
578 	m->m_next = NULL;
579 	m_cat(m, m2);
580 	uma_zfree(V_pf_frent_z, frent);
581 	for (frent = next; frent != NULL; frent = next) {
582 		next = TAILQ_NEXT(frent, fr_next);
583 
584 		m2 = frent->fe_m;
585 		/* Strip off ip header. */
586 		m_adj(m2, frent->fe_hdrlen);
587 		uma_zfree(V_pf_frent_z, frent);
588 		m_cat(m, m2);
589 	}
590 
591 	/* Remove from fragment queue. */
592 	pf_remove_fragment(frag);
593 
594 	return (m);
595 }
596 
597 #ifdef INET
598 static int
599 pf_reassemble(struct mbuf **m0, struct ip *ip, int dir, u_short *reason)
600 {
601 	struct mbuf		*m = *m0;
602 	struct pf_frent		*frent;
603 	struct pf_fragment	*frag;
604 	struct pf_fragment_cmp	key;
605 	uint16_t		total, hdrlen;
606 
607 	/* Get an entry for the fragment queue */
608 	if ((frent = pf_create_fragment(reason)) == NULL)
609 		return (PF_DROP);
610 
611 	frent->fe_m = m;
612 	frent->fe_hdrlen = ip->ip_hl << 2;
613 	frent->fe_extoff = 0;
614 	frent->fe_len = ntohs(ip->ip_len) - (ip->ip_hl << 2);
615 	frent->fe_off = (ntohs(ip->ip_off) & IP_OFFMASK) << 3;
616 	frent->fe_mff = ntohs(ip->ip_off) & IP_MF;
617 
618 	pf_ip2key(ip, dir, &key);
619 
620 	if ((frag = pf_fillup_fragment(&key, frent, reason)) == NULL)
621 		return (PF_DROP);
622 
623 	/* The mbuf is part of the fragment entry, no direct free or access */
624 	m = *m0 = NULL;
625 
626 	if (!pf_isfull_fragment(frag))
627 		return (PF_PASS);  /* drop because *m0 is NULL, no error */
628 
629 	/* We have all the data */
630 	frent = TAILQ_FIRST(&frag->fr_queue);
631 	KASSERT(frent != NULL, ("frent != NULL"));
632 	total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
633 		TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
634 	hdrlen = frent->fe_hdrlen;
635 
636 	m = *m0 = pf_join_fragment(frag);
637 	frag = NULL;
638 
639 	if (m->m_flags & M_PKTHDR) {
640 		int plen = 0;
641 		for (m = *m0; m; m = m->m_next)
642 			plen += m->m_len;
643 		m = *m0;
644 		m->m_pkthdr.len = plen;
645 	}
646 
647 	ip = mtod(m, struct ip *);
648 	ip->ip_len = htons(hdrlen + total);
649 	ip->ip_off &= ~(IP_MF|IP_OFFMASK);
650 
651 	if (hdrlen + total > IP_MAXPACKET) {
652 		DPFPRINTF(("drop: too big: %d", total));
653 		ip->ip_len = 0;
654 		REASON_SET(reason, PFRES_SHORT);
655 		/* PF_DROP requires a valid mbuf *m0 in pf_test() */
656 		return (PF_DROP);
657 	}
658 
659 	DPFPRINTF(("complete: %p(%d)\n", m, ntohs(ip->ip_len)));
660 	return (PF_PASS);
661 }
662 #endif	/* INET */
663 
664 #ifdef INET6
665 static int
666 pf_reassemble6(struct mbuf **m0, struct ip6_hdr *ip6, struct ip6_frag *fraghdr,
667     uint16_t hdrlen, uint16_t extoff, int dir, u_short *reason)
668 {
669 	struct mbuf		*m = *m0;
670 	struct pf_frent		*frent;
671 	struct pf_fragment	*frag;
672 	struct pf_fragment_cmp	 key;
673 	struct m_tag		*mtag;
674 	struct pf_fragment_tag	*ftag;
675 	int			 off;
676 	uint16_t		 total, maxlen;
677 	uint8_t			 proto;
678 
679 	PF_FRAG_LOCK();
680 
681 	/* Get an entry for the fragment queue. */
682 	if ((frent = pf_create_fragment(reason)) == NULL) {
683 		PF_FRAG_UNLOCK();
684 		return (PF_DROP);
685 	}
686 
687 	frent->fe_m = m;
688 	frent->fe_hdrlen = hdrlen;
689 	frent->fe_extoff = extoff;
690 	frent->fe_len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - hdrlen;
691 	frent->fe_off = ntohs(fraghdr->ip6f_offlg & IP6F_OFF_MASK);
692 	frent->fe_mff = fraghdr->ip6f_offlg & IP6F_MORE_FRAG;
693 
694 	key.frc_src.v6 = ip6->ip6_src;
695 	key.frc_dst.v6 = ip6->ip6_dst;
696 	key.frc_af = AF_INET6;
697 	/* Only the first fragment's protocol is relevant. */
698 	key.frc_proto = 0;
699 	key.frc_id = fraghdr->ip6f_ident;
700 	key.frc_direction = dir;
701 
702 	if ((frag = pf_fillup_fragment(&key, frent, reason)) == NULL) {
703 		PF_FRAG_UNLOCK();
704 		return (PF_DROP);
705 	}
706 
707 	/* The mbuf is part of the fragment entry, no direct free or access. */
708 	m = *m0 = NULL;
709 
710 	if (!pf_isfull_fragment(frag)) {
711 		PF_FRAG_UNLOCK();
712 		return (PF_PASS);  /* Drop because *m0 is NULL, no error. */
713 	}
714 
715 	/* We have all the data. */
716 	extoff = frent->fe_extoff;
717 	maxlen = frag->fr_maxlen;
718 	frent = TAILQ_FIRST(&frag->fr_queue);
719 	KASSERT(frent != NULL, ("frent != NULL"));
720 	total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
721 		TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
722 	hdrlen = frent->fe_hdrlen - sizeof(struct ip6_frag);
723 
724 	m = *m0 = pf_join_fragment(frag);
725 	frag = NULL;
726 
727 	PF_FRAG_UNLOCK();
728 
729 	/* Take protocol from first fragment header. */
730 	m = m_getptr(m, hdrlen + offsetof(struct ip6_frag, ip6f_nxt), &off);
731 	KASSERT(m, ("%s: short mbuf chain", __func__));
732 	proto = *(mtod(m, caddr_t) + off);
733 	m = *m0;
734 
735 	/* Delete frag6 header */
736 	if (ip6_deletefraghdr(m, hdrlen, M_NOWAIT) != 0)
737 		goto fail;
738 
739 	if (m->m_flags & M_PKTHDR) {
740 		int plen = 0;
741 		for (m = *m0; m; m = m->m_next)
742 			plen += m->m_len;
743 		m = *m0;
744 		m->m_pkthdr.len = plen;
745 	}
746 
747 	if ((mtag = m_tag_get(PF_REASSEMBLED, sizeof(struct pf_fragment_tag),
748 	    M_NOWAIT)) == NULL)
749 		goto fail;
750 	ftag = (struct pf_fragment_tag *)(mtag + 1);
751 	ftag->ft_hdrlen = hdrlen;
752 	ftag->ft_extoff = extoff;
753 	ftag->ft_maxlen = maxlen;
754 	m_tag_prepend(m, mtag);
755 
756 	ip6 = mtod(m, struct ip6_hdr *);
757 	ip6->ip6_plen = htons(hdrlen - sizeof(struct ip6_hdr) + total);
758 	if (extoff) {
759 		/* Write protocol into next field of last extension header. */
760 		m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt),
761 		    &off);
762 		KASSERT(m, ("%s: short mbuf chain", __func__));
763 		*(mtod(m, char *) + off) = proto;
764 		m = *m0;
765 	} else
766 		ip6->ip6_nxt = proto;
767 
768 	if (hdrlen - sizeof(struct ip6_hdr) + total > IPV6_MAXPACKET) {
769 		DPFPRINTF(("drop: too big: %d", total));
770 		ip6->ip6_plen = 0;
771 		REASON_SET(reason, PFRES_SHORT);
772 		/* PF_DROP requires a valid mbuf *m0 in pf_test6(). */
773 		return (PF_DROP);
774 	}
775 
776 	DPFPRINTF(("complete: %p(%d)", m, ntohs(ip6->ip6_plen)));
777 	return (PF_PASS);
778 
779 fail:
780 	REASON_SET(reason, PFRES_MEMORY);
781 	/* PF_DROP requires a valid mbuf *m0 in pf_test6(), will free later. */
782 	return (PF_DROP);
783 }
784 #endif	/* INET6 */
785 
786 #ifdef INET
787 static struct mbuf *
788 pf_fragcache(struct mbuf **m0, struct ip *h, struct pf_fragment **frag, int mff,
789     int drop, int *nomem)
790 {
791 	struct mbuf		*m = *m0;
792 	struct pf_frent		*frp, *fra, *cur = NULL;
793 	int			 ip_len = ntohs(h->ip_len) - (h->ip_hl << 2);
794 	u_int16_t		 off = ntohs(h->ip_off) << 3;
795 	u_int16_t		 max = ip_len + off;
796 	int			 hosed = 0;
797 
798 	PF_FRAG_ASSERT();
799 	KASSERT((*frag == NULL || !BUFFER_FRAGMENTS(*frag)),
800 	    ("!(*frag == NULL || !BUFFER_FRAGMENTS(*frag)): %s", __FUNCTION__));
801 
802 	/* Create a new range queue for this packet */
803 	if (*frag == NULL) {
804 		*frag = uma_zalloc(V_pf_frag_z, M_NOWAIT);
805 		if (*frag == NULL) {
806 			pf_flush_fragments();
807 			*frag = uma_zalloc(V_pf_frag_z, M_NOWAIT);
808 			if (*frag == NULL)
809 				goto no_mem;
810 		}
811 
812 		/* Get an entry for the queue */
813 		cur = uma_zalloc(V_pf_frent_z, M_NOWAIT);
814 		if (cur == NULL) {
815 			uma_zfree(V_pf_frag_z, *frag);
816 			*frag = NULL;
817 			goto no_mem;
818 		}
819 
820 		(*frag)->fr_flags = PFFRAG_NOBUFFER;
821 		(*frag)->fr_max = 0;
822 		(*frag)->fr_src.v4 = h->ip_src;
823 		(*frag)->fr_dst.v4 = h->ip_dst;
824 		(*frag)->fr_id = h->ip_id;
825 		(*frag)->fr_timeout = time_uptime;
826 
827 		cur->fe_off = off;
828 		cur->fe_len = max; /* TODO: fe_len = max - off ? */
829 		TAILQ_INIT(&(*frag)->fr_queue);
830 		TAILQ_INSERT_HEAD(&(*frag)->fr_queue, cur, fr_next);
831 
832 		RB_INSERT(pf_frag_tree, &V_pf_cache_tree, *frag);
833 		TAILQ_INSERT_HEAD(&V_pf_cachequeue, *frag, frag_next);
834 
835 		DPFPRINTF(("fragcache[%d]: new %d-%d\n", h->ip_id, off, max));
836 
837 		goto pass;
838 	}
839 
840 	/*
841 	 * Find a fragment after the current one:
842 	 *  - off contains the real shifted offset.
843 	 */
844 	frp = NULL;
845 	TAILQ_FOREACH(fra, &(*frag)->fr_queue, fr_next) {
846 		if (fra->fe_off > off)
847 			break;
848 		frp = fra;
849 	}
850 
851 	KASSERT((frp != NULL || fra != NULL),
852 	    ("!(frp != NULL || fra != NULL): %s", __FUNCTION__));
853 
854 	if (frp != NULL) {
855 		int	precut;
856 
857 		precut = frp->fe_len - off;
858 		if (precut >= ip_len) {
859 			/* Fragment is entirely a duplicate */
860 			DPFPRINTF(("fragcache[%d]: dead (%d-%d) %d-%d\n",
861 			    h->ip_id, frp->fe_off, frp->fe_len, off, max));
862 			goto drop_fragment;
863 		}
864 		if (precut == 0) {
865 			/* They are adjacent.  Fixup cache entry */
866 			DPFPRINTF(("fragcache[%d]: adjacent (%d-%d) %d-%d\n",
867 			    h->ip_id, frp->fe_off, frp->fe_len, off, max));
868 			frp->fe_len = max;
869 		} else if (precut > 0) {
870 			/* The first part of this payload overlaps with a
871 			 * fragment that has already been passed.
872 			 * Need to trim off the first part of the payload.
873 			 * But to do so easily, we need to create another
874 			 * mbuf to throw the original header into.
875 			 */
876 
877 			DPFPRINTF(("fragcache[%d]: chop %d (%d-%d) %d-%d\n",
878 			    h->ip_id, precut, frp->fe_off, frp->fe_len, off,
879 			    max));
880 
881 			off += precut;
882 			max -= precut;
883 			/* Update the previous frag to encompass this one */
884 			frp->fe_len = max;
885 
886 			if (!drop) {
887 				/* XXX Optimization opportunity
888 				 * This is a very heavy way to trim the payload.
889 				 * we could do it much faster by diddling mbuf
890 				 * internals but that would be even less legible
891 				 * than this mbuf magic.  For my next trick,
892 				 * I'll pull a rabbit out of my laptop.
893 				 */
894 				*m0 = m_dup(m, M_NOWAIT);
895 				if (*m0 == NULL)
896 					goto no_mem;
897 				/* From KAME Project : We have missed this! */
898 				m_adj(*m0, (h->ip_hl << 2) -
899 				    (*m0)->m_pkthdr.len);
900 
901 				KASSERT(((*m0)->m_next == NULL),
902 				    ("(*m0)->m_next != NULL: %s",
903 				    __FUNCTION__));
904 				m_adj(m, precut + (h->ip_hl << 2));
905 				m_cat(*m0, m);
906 				m = *m0;
907 				if (m->m_flags & M_PKTHDR) {
908 					int plen = 0;
909 					struct mbuf *t;
910 					for (t = m; t; t = t->m_next)
911 						plen += t->m_len;
912 					m->m_pkthdr.len = plen;
913 				}
914 
915 
916 				h = mtod(m, struct ip *);
917 
918 				KASSERT(((int)m->m_len ==
919 				    ntohs(h->ip_len) - precut),
920 				    ("m->m_len != ntohs(h->ip_len) - precut: %s",
921 				    __FUNCTION__));
922 				h->ip_off = htons(ntohs(h->ip_off) +
923 				    (precut >> 3));
924 				h->ip_len = htons(ntohs(h->ip_len) - precut);
925 			} else {
926 				hosed++;
927 			}
928 		} else {
929 			/* There is a gap between fragments */
930 
931 			DPFPRINTF(("fragcache[%d]: gap %d (%d-%d) %d-%d\n",
932 			    h->ip_id, -precut, frp->fe_off, frp->fe_len, off,
933 			    max));
934 
935 			cur = uma_zalloc(V_pf_frent_z, M_NOWAIT);
936 			if (cur == NULL)
937 				goto no_mem;
938 
939 			cur->fe_off = off;
940 			cur->fe_len = max;
941 			TAILQ_INSERT_AFTER(&(*frag)->fr_queue, frp, cur, fr_next);
942 		}
943 	}
944 
945 	if (fra != NULL) {
946 		int	aftercut;
947 		int	merge = 0;
948 
949 		aftercut = max - fra->fe_off;
950 		if (aftercut == 0) {
951 			/* Adjacent fragments */
952 			DPFPRINTF(("fragcache[%d]: adjacent %d-%d (%d-%d)\n",
953 			    h->ip_id, off, max, fra->fe_off, fra->fe_len));
954 			fra->fe_off = off;
955 			merge = 1;
956 		} else if (aftercut > 0) {
957 			/* Need to chop off the tail of this fragment */
958 			DPFPRINTF(("fragcache[%d]: chop %d %d-%d (%d-%d)\n",
959 			    h->ip_id, aftercut, off, max, fra->fe_off,
960 			    fra->fe_len));
961 			fra->fe_off = off;
962 			max -= aftercut;
963 
964 			merge = 1;
965 
966 			if (!drop) {
967 				m_adj(m, -aftercut);
968 				if (m->m_flags & M_PKTHDR) {
969 					int plen = 0;
970 					struct mbuf *t;
971 					for (t = m; t; t = t->m_next)
972 						plen += t->m_len;
973 					m->m_pkthdr.len = plen;
974 				}
975 				h = mtod(m, struct ip *);
976 				KASSERT(((int)m->m_len == ntohs(h->ip_len) - aftercut),
977 				    ("m->m_len != ntohs(h->ip_len) - aftercut: %s",
978 				    __FUNCTION__));
979 				h->ip_len = htons(ntohs(h->ip_len) - aftercut);
980 			} else {
981 				hosed++;
982 			}
983 		} else if (frp == NULL) {
984 			/* There is a gap between fragments */
985 			DPFPRINTF(("fragcache[%d]: gap %d %d-%d (%d-%d)\n",
986 			    h->ip_id, -aftercut, off, max, fra->fe_off,
987 			    fra->fe_len));
988 
989 			cur = uma_zalloc(V_pf_frent_z, M_NOWAIT);
990 			if (cur == NULL)
991 				goto no_mem;
992 
993 			cur->fe_off = off;
994 			cur->fe_len = max;
995 			TAILQ_INSERT_HEAD(&(*frag)->fr_queue, cur, fr_next);
996 		}
997 
998 
999 		/* Need to glue together two separate fragment descriptors */
1000 		if (merge) {
1001 			if (cur && fra->fe_off <= cur->fe_len) {
1002 				/* Need to merge in a previous 'cur' */
1003 				DPFPRINTF(("fragcache[%d]: adjacent(merge "
1004 				    "%d-%d) %d-%d (%d-%d)\n",
1005 				    h->ip_id, cur->fe_off, cur->fe_len, off,
1006 				    max, fra->fe_off, fra->fe_len));
1007 				fra->fe_off = cur->fe_off;
1008 				TAILQ_REMOVE(&(*frag)->fr_queue, cur, fr_next);
1009 				uma_zfree(V_pf_frent_z, cur);
1010 				cur = NULL;
1011 
1012 			} else if (frp && fra->fe_off <= frp->fe_len) {
1013 				/* Need to merge in a modified 'frp' */
1014 				KASSERT((cur == NULL), ("cur != NULL: %s",
1015 				    __FUNCTION__));
1016 				DPFPRINTF(("fragcache[%d]: adjacent(merge "
1017 				    "%d-%d) %d-%d (%d-%d)\n",
1018 				    h->ip_id, frp->fe_off, frp->fe_len, off,
1019 				    max, fra->fe_off, fra->fe_len));
1020 				fra->fe_off = frp->fe_off;
1021 				TAILQ_REMOVE(&(*frag)->fr_queue, frp, fr_next);
1022 				uma_zfree(V_pf_frent_z, frp);
1023 				frp = NULL;
1024 
1025 			}
1026 		}
1027 	}
1028 
1029 	if (hosed) {
1030 		/*
1031 		 * We must keep tracking the overall fragment even when
1032 		 * we're going to drop it anyway so that we know when to
1033 		 * free the overall descriptor.  Thus we drop the frag late.
1034 		 */
1035 		goto drop_fragment;
1036 	}
1037 
1038 
1039  pass:
1040 	/* Update maximum data size */
1041 	if ((*frag)->fr_max < max)
1042 		(*frag)->fr_max = max;
1043 
1044 	/* This is the last segment */
1045 	if (!mff)
1046 		(*frag)->fr_flags |= PFFRAG_SEENLAST;
1047 
1048 	/* Check if we are completely reassembled */
1049 	if (((*frag)->fr_flags & PFFRAG_SEENLAST) &&
1050 	    TAILQ_FIRST(&(*frag)->fr_queue)->fe_off == 0 &&
1051 	    TAILQ_FIRST(&(*frag)->fr_queue)->fe_len == (*frag)->fr_max) {
1052 		/* Remove from fragment queue */
1053 		DPFPRINTF(("fragcache[%d]: done 0-%d\n", h->ip_id,
1054 		    (*frag)->fr_max));
1055 		pf_free_fragment(*frag);
1056 		*frag = NULL;
1057 	}
1058 
1059 	return (m);
1060 
1061  no_mem:
1062 	*nomem = 1;
1063 
1064 	/* Still need to pay attention to !IP_MF */
1065 	if (!mff && *frag != NULL)
1066 		(*frag)->fr_flags |= PFFRAG_SEENLAST;
1067 
1068 	m_freem(m);
1069 	return (NULL);
1070 
1071  drop_fragment:
1072 
1073 	/* Still need to pay attention to !IP_MF */
1074 	if (!mff && *frag != NULL)
1075 		(*frag)->fr_flags |= PFFRAG_SEENLAST;
1076 
1077 	if (drop) {
1078 		/* This fragment has been deemed bad.  Don't reass */
1079 		if (((*frag)->fr_flags & PFFRAG_DROP) == 0)
1080 			DPFPRINTF(("fragcache[%d]: dropping overall fragment\n",
1081 			    h->ip_id));
1082 		(*frag)->fr_flags |= PFFRAG_DROP;
1083 	}
1084 
1085 	m_freem(m);
1086 	return (NULL);
1087 }
1088 #endif	/* INET */
1089 
1090 #ifdef INET6
1091 int
1092 pf_refragment6(struct ifnet *ifp, struct mbuf **m0, struct m_tag *mtag)
1093 {
1094 	struct mbuf		*m = *m0, *t;
1095 	struct pf_fragment_tag	*ftag = (struct pf_fragment_tag *)(mtag + 1);
1096 	struct pf_pdesc		 pd;
1097 	uint16_t		 hdrlen, extoff, maxlen;
1098 	uint8_t			 proto;
1099 	int			 error, action;
1100 
1101 	hdrlen = ftag->ft_hdrlen;
1102 	extoff = ftag->ft_extoff;
1103 	maxlen = ftag->ft_maxlen;
1104 	m_tag_delete(m, mtag);
1105 	mtag = NULL;
1106 	ftag = NULL;
1107 
1108 	if (extoff) {
1109 		int off;
1110 
1111 		/* Use protocol from next field of last extension header */
1112 		m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt),
1113 		    &off);
1114 		KASSERT((m != NULL), ("pf_refragment6: short mbuf chain"));
1115 		proto = *(mtod(m, caddr_t) + off);
1116 		*(mtod(m, char *) + off) = IPPROTO_FRAGMENT;
1117 		m = *m0;
1118 	} else {
1119 		struct ip6_hdr *hdr;
1120 
1121 		hdr = mtod(m, struct ip6_hdr *);
1122 		proto = hdr->ip6_nxt;
1123 		hdr->ip6_nxt = IPPROTO_FRAGMENT;
1124 	}
1125 
1126 	/*
1127 	 * Maxlen may be less than 8 if there was only a single
1128 	 * fragment.  As it was fragmented before, add a fragment
1129 	 * header also for a single fragment.  If total or maxlen
1130 	 * is less than 8, ip6_fragment() will return EMSGSIZE and
1131 	 * we drop the packet.
1132 	 */
1133 	error = ip6_fragment(ifp, m, hdrlen, proto, maxlen);
1134 	m = (*m0)->m_nextpkt;
1135 	(*m0)->m_nextpkt = NULL;
1136 	if (error == 0) {
1137 		/* The first mbuf contains the unfragmented packet. */
1138 		m_freem(*m0);
1139 		*m0 = NULL;
1140 		action = PF_PASS;
1141 	} else {
1142 		/* Drop expects an mbuf to free. */
1143 		DPFPRINTF(("refragment error %d", error));
1144 		action = PF_DROP;
1145 	}
1146 	for (t = m; m; m = t) {
1147 		t = m->m_nextpkt;
1148 		m->m_nextpkt = NULL;
1149 		memset(&pd, 0, sizeof(pd));
1150 		pd.pf_mtag = pf_find_mtag(m);
1151 		if (error == 0)
1152 			ip6_forward(m, 0);
1153 		else
1154 			m_freem(m);
1155 	}
1156 
1157 	return (action);
1158 }
1159 #endif /* INET6 */
1160 
1161 #ifdef INET
1162 int
1163 pf_normalize_ip(struct mbuf **m0, int dir, struct pfi_kif *kif, u_short *reason,
1164     struct pf_pdesc *pd)
1165 {
1166 	struct mbuf		*m = *m0;
1167 	struct pf_rule		*r;
1168 	struct pf_fragment	*frag = NULL;
1169 	struct pf_fragment_cmp	key;
1170 	struct ip		*h = mtod(m, struct ip *);
1171 	int			 mff = (ntohs(h->ip_off) & IP_MF);
1172 	int			 hlen = h->ip_hl << 2;
1173 	u_int16_t		 fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
1174 	u_int16_t		 max;
1175 	int			 ip_len;
1176 	int			 ip_off;
1177 	int			 tag = -1;
1178 	int			 verdict;
1179 
1180 	PF_RULES_RASSERT();
1181 
1182 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1183 	while (r != NULL) {
1184 		r->evaluations++;
1185 		if (pfi_kif_match(r->kif, kif) == r->ifnot)
1186 			r = r->skip[PF_SKIP_IFP].ptr;
1187 		else if (r->direction && r->direction != dir)
1188 			r = r->skip[PF_SKIP_DIR].ptr;
1189 		else if (r->af && r->af != AF_INET)
1190 			r = r->skip[PF_SKIP_AF].ptr;
1191 		else if (r->proto && r->proto != h->ip_p)
1192 			r = r->skip[PF_SKIP_PROTO].ptr;
1193 		else if (PF_MISMATCHAW(&r->src.addr,
1194 		    (struct pf_addr *)&h->ip_src.s_addr, AF_INET,
1195 		    r->src.neg, kif, M_GETFIB(m)))
1196 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
1197 		else if (PF_MISMATCHAW(&r->dst.addr,
1198 		    (struct pf_addr *)&h->ip_dst.s_addr, AF_INET,
1199 		    r->dst.neg, NULL, M_GETFIB(m)))
1200 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
1201 		else if (r->match_tag && !pf_match_tag(m, r, &tag,
1202 		    pd->pf_mtag ? pd->pf_mtag->tag : 0))
1203 			r = TAILQ_NEXT(r, entries);
1204 		else
1205 			break;
1206 	}
1207 
1208 	if (r == NULL || r->action == PF_NOSCRUB)
1209 		return (PF_PASS);
1210 	else {
1211 		r->packets[dir == PF_OUT]++;
1212 		r->bytes[dir == PF_OUT] += pd->tot_len;
1213 	}
1214 
1215 	/* Check for illegal packets */
1216 	if (hlen < (int)sizeof(struct ip))
1217 		goto drop;
1218 
1219 	if (hlen > ntohs(h->ip_len))
1220 		goto drop;
1221 
1222 	/* Clear IP_DF if the rule uses the no-df option */
1223 	if (r->rule_flag & PFRULE_NODF && h->ip_off & htons(IP_DF)) {
1224 		u_int16_t ip_off = h->ip_off;
1225 
1226 		h->ip_off &= htons(~IP_DF);
1227 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
1228 	}
1229 
1230 	/* We will need other tests here */
1231 	if (!fragoff && !mff)
1232 		goto no_fragment;
1233 
1234 	/* We're dealing with a fragment now. Don't allow fragments
1235 	 * with IP_DF to enter the cache. If the flag was cleared by
1236 	 * no-df above, fine. Otherwise drop it.
1237 	 */
1238 	if (h->ip_off & htons(IP_DF)) {
1239 		DPFPRINTF(("IP_DF\n"));
1240 		goto bad;
1241 	}
1242 
1243 	ip_len = ntohs(h->ip_len) - hlen;
1244 	ip_off = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
1245 
1246 	/* All fragments are 8 byte aligned */
1247 	if (mff && (ip_len & 0x7)) {
1248 		DPFPRINTF(("mff and %d\n", ip_len));
1249 		goto bad;
1250 	}
1251 
1252 	/* Respect maximum length */
1253 	if (fragoff + ip_len > IP_MAXPACKET) {
1254 		DPFPRINTF(("max packet %d\n", fragoff + ip_len));
1255 		goto bad;
1256 	}
1257 	max = fragoff + ip_len;
1258 
1259 	if ((r->rule_flag & (PFRULE_FRAGCROP|PFRULE_FRAGDROP)) == 0) {
1260 
1261 		/* Fully buffer all of the fragments */
1262 		PF_FRAG_LOCK();
1263 
1264 		pf_ip2key(h, dir, &key);
1265 		frag = pf_find_fragment(&key, &V_pf_frag_tree);
1266 
1267 		/* Check if we saw the last fragment already */
1268 		if (frag != NULL && (frag->fr_flags & PFFRAG_SEENLAST) &&
1269 		    max > frag->fr_max)
1270 			goto bad;
1271 
1272 		/* Might return a completely reassembled mbuf, or NULL */
1273 		DPFPRINTF(("reass frag %d @ %d-%d\n", h->ip_id, fragoff, max));
1274 		verdict = pf_reassemble(m0, h, dir, reason);
1275 		PF_FRAG_UNLOCK();
1276 
1277 		if (verdict != PF_PASS)
1278 			return (PF_DROP);
1279 
1280 		m = *m0;
1281 		if (m == NULL)
1282 			return (PF_DROP);
1283 
1284 		if (frag != NULL && (frag->fr_flags & PFFRAG_DROP))
1285 			goto drop;
1286 
1287 		h = mtod(m, struct ip *);
1288 	} else {
1289 		/* non-buffering fragment cache (drops or masks overlaps) */
1290 		int	nomem = 0;
1291 
1292 		if (dir == PF_OUT && pd->pf_mtag &&
1293 		    pd->pf_mtag->flags & PF_TAG_FRAGCACHE) {
1294 			/*
1295 			 * Already passed the fragment cache in the
1296 			 * input direction.  If we continued, it would
1297 			 * appear to be a dup and would be dropped.
1298 			 */
1299 			goto fragment_pass;
1300 		}
1301 
1302 		PF_FRAG_LOCK();
1303 		pf_ip2key(h, dir, &key);
1304 		frag = pf_find_fragment(&key, &V_pf_cache_tree);
1305 
1306 		/* Check if we saw the last fragment already */
1307 		if (frag != NULL && (frag->fr_flags & PFFRAG_SEENLAST) &&
1308 		    max > frag->fr_max) {
1309 			if (r->rule_flag & PFRULE_FRAGDROP)
1310 				frag->fr_flags |= PFFRAG_DROP;
1311 			goto bad;
1312 		}
1313 
1314 		*m0 = m = pf_fragcache(m0, h, &frag, mff,
1315 		    (r->rule_flag & PFRULE_FRAGDROP) ? 1 : 0, &nomem);
1316 		PF_FRAG_UNLOCK();
1317 		if (m == NULL) {
1318 			if (nomem)
1319 				goto no_mem;
1320 			goto drop;
1321 		}
1322 
1323 		if (dir == PF_IN) {
1324 			/* Use mtag from copied and trimmed mbuf chain. */
1325 			pd->pf_mtag = pf_get_mtag(m);
1326 			if (pd->pf_mtag == NULL) {
1327 				m_freem(m);
1328 				*m0 = NULL;
1329 				goto no_mem;
1330 			}
1331 			pd->pf_mtag->flags |= PF_TAG_FRAGCACHE;
1332 		}
1333 
1334 		if (frag != NULL && (frag->fr_flags & PFFRAG_DROP))
1335 			goto drop;
1336 		goto fragment_pass;
1337 	}
1338 
1339  no_fragment:
1340 	/* At this point, only IP_DF is allowed in ip_off */
1341 	if (h->ip_off & ~htons(IP_DF)) {
1342 		u_int16_t ip_off = h->ip_off;
1343 
1344 		h->ip_off &= htons(IP_DF);
1345 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
1346 	}
1347 
1348 	/* not missing a return here */
1349 
1350  fragment_pass:
1351 	pf_scrub_ip(&m, r->rule_flag, r->min_ttl, r->set_tos);
1352 
1353 	if ((r->rule_flag & (PFRULE_FRAGCROP|PFRULE_FRAGDROP)) == 0)
1354 		pd->flags |= PFDESC_IP_REAS;
1355 	return (PF_PASS);
1356 
1357  no_mem:
1358 	REASON_SET(reason, PFRES_MEMORY);
1359 	if (r != NULL && r->log)
1360 		PFLOG_PACKET(kif, m, AF_INET, dir, *reason, r, NULL, NULL, pd,
1361 		    1);
1362 	return (PF_DROP);
1363 
1364  drop:
1365 	REASON_SET(reason, PFRES_NORM);
1366 	if (r != NULL && r->log)
1367 		PFLOG_PACKET(kif, m, AF_INET, dir, *reason, r, NULL, NULL, pd,
1368 		    1);
1369 	return (PF_DROP);
1370 
1371  bad:
1372 	DPFPRINTF(("dropping bad fragment\n"));
1373 
1374 	/* Free associated fragments */
1375 	if (frag != NULL) {
1376 		pf_free_fragment(frag);
1377 		PF_FRAG_UNLOCK();
1378 	}
1379 
1380 	REASON_SET(reason, PFRES_FRAG);
1381 	if (r != NULL && r->log)
1382 		PFLOG_PACKET(kif, m, AF_INET, dir, *reason, r, NULL, NULL, pd,
1383 		    1);
1384 
1385 	return (PF_DROP);
1386 }
1387 #endif
1388 
1389 #ifdef INET6
1390 int
1391 pf_normalize_ip6(struct mbuf **m0, int dir, struct pfi_kif *kif,
1392     u_short *reason, struct pf_pdesc *pd)
1393 {
1394 	struct mbuf		*m = *m0;
1395 	struct pf_rule		*r;
1396 	struct ip6_hdr		*h = mtod(m, struct ip6_hdr *);
1397 	int			 extoff;
1398 	int			 off;
1399 	struct ip6_ext		 ext;
1400 	struct ip6_opt		 opt;
1401 	struct ip6_opt_jumbo	 jumbo;
1402 	struct ip6_frag		 frag;
1403 	u_int32_t		 jumbolen = 0, plen;
1404 	int			 optend;
1405 	int			 ooff;
1406 	u_int8_t		 proto;
1407 	int			 terminal;
1408 
1409 	PF_RULES_RASSERT();
1410 
1411 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1412 	while (r != NULL) {
1413 		r->evaluations++;
1414 		if (pfi_kif_match(r->kif, kif) == r->ifnot)
1415 			r = r->skip[PF_SKIP_IFP].ptr;
1416 		else if (r->direction && r->direction != dir)
1417 			r = r->skip[PF_SKIP_DIR].ptr;
1418 		else if (r->af && r->af != AF_INET6)
1419 			r = r->skip[PF_SKIP_AF].ptr;
1420 #if 0 /* header chain! */
1421 		else if (r->proto && r->proto != h->ip6_nxt)
1422 			r = r->skip[PF_SKIP_PROTO].ptr;
1423 #endif
1424 		else if (PF_MISMATCHAW(&r->src.addr,
1425 		    (struct pf_addr *)&h->ip6_src, AF_INET6,
1426 		    r->src.neg, kif, M_GETFIB(m)))
1427 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
1428 		else if (PF_MISMATCHAW(&r->dst.addr,
1429 		    (struct pf_addr *)&h->ip6_dst, AF_INET6,
1430 		    r->dst.neg, NULL, M_GETFIB(m)))
1431 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
1432 		else
1433 			break;
1434 	}
1435 
1436 	if (r == NULL || r->action == PF_NOSCRUB)
1437 		return (PF_PASS);
1438 	else {
1439 		r->packets[dir == PF_OUT]++;
1440 		r->bytes[dir == PF_OUT] += pd->tot_len;
1441 	}
1442 
1443 	/* Check for illegal packets */
1444 	if (sizeof(struct ip6_hdr) + IPV6_MAXPACKET < m->m_pkthdr.len)
1445 		goto drop;
1446 
1447 	extoff = 0;
1448 	off = sizeof(struct ip6_hdr);
1449 	proto = h->ip6_nxt;
1450 	terminal = 0;
1451 	do {
1452 		switch (proto) {
1453 		case IPPROTO_FRAGMENT:
1454 			goto fragment;
1455 			break;
1456 		case IPPROTO_AH:
1457 		case IPPROTO_ROUTING:
1458 		case IPPROTO_DSTOPTS:
1459 			if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL,
1460 			    NULL, AF_INET6))
1461 				goto shortpkt;
1462 			extoff = off;
1463 			if (proto == IPPROTO_AH)
1464 				off += (ext.ip6e_len + 2) * 4;
1465 			else
1466 				off += (ext.ip6e_len + 1) * 8;
1467 			proto = ext.ip6e_nxt;
1468 			break;
1469 		case IPPROTO_HOPOPTS:
1470 			if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL,
1471 			    NULL, AF_INET6))
1472 				goto shortpkt;
1473 			extoff = off;
1474 			optend = off + (ext.ip6e_len + 1) * 8;
1475 			ooff = off + sizeof(ext);
1476 			do {
1477 				if (!pf_pull_hdr(m, ooff, &opt.ip6o_type,
1478 				    sizeof(opt.ip6o_type), NULL, NULL,
1479 				    AF_INET6))
1480 					goto shortpkt;
1481 				if (opt.ip6o_type == IP6OPT_PAD1) {
1482 					ooff++;
1483 					continue;
1484 				}
1485 				if (!pf_pull_hdr(m, ooff, &opt, sizeof(opt),
1486 				    NULL, NULL, AF_INET6))
1487 					goto shortpkt;
1488 				if (ooff + sizeof(opt) + opt.ip6o_len > optend)
1489 					goto drop;
1490 				switch (opt.ip6o_type) {
1491 				case IP6OPT_JUMBO:
1492 					if (h->ip6_plen != 0)
1493 						goto drop;
1494 					if (!pf_pull_hdr(m, ooff, &jumbo,
1495 					    sizeof(jumbo), NULL, NULL,
1496 					    AF_INET6))
1497 						goto shortpkt;
1498 					memcpy(&jumbolen, jumbo.ip6oj_jumbo_len,
1499 					    sizeof(jumbolen));
1500 					jumbolen = ntohl(jumbolen);
1501 					if (jumbolen <= IPV6_MAXPACKET)
1502 						goto drop;
1503 					if (sizeof(struct ip6_hdr) + jumbolen !=
1504 					    m->m_pkthdr.len)
1505 						goto drop;
1506 					break;
1507 				default:
1508 					break;
1509 				}
1510 				ooff += sizeof(opt) + opt.ip6o_len;
1511 			} while (ooff < optend);
1512 
1513 			off = optend;
1514 			proto = ext.ip6e_nxt;
1515 			break;
1516 		default:
1517 			terminal = 1;
1518 			break;
1519 		}
1520 	} while (!terminal);
1521 
1522 	/* jumbo payload option must be present, or plen > 0 */
1523 	if (ntohs(h->ip6_plen) == 0)
1524 		plen = jumbolen;
1525 	else
1526 		plen = ntohs(h->ip6_plen);
1527 	if (plen == 0)
1528 		goto drop;
1529 	if (sizeof(struct ip6_hdr) + plen > m->m_pkthdr.len)
1530 		goto shortpkt;
1531 
1532 	pf_scrub_ip6(&m, r->min_ttl);
1533 
1534 	return (PF_PASS);
1535 
1536  fragment:
1537 	/* Jumbo payload packets cannot be fragmented. */
1538 	plen = ntohs(h->ip6_plen);
1539 	if (plen == 0 || jumbolen)
1540 		goto drop;
1541 	if (sizeof(struct ip6_hdr) + plen > m->m_pkthdr.len)
1542 		goto shortpkt;
1543 
1544 	if (!pf_pull_hdr(m, off, &frag, sizeof(frag), NULL, NULL, AF_INET6))
1545 		goto shortpkt;
1546 
1547 	/* Offset now points to data portion. */
1548 	off += sizeof(frag);
1549 
1550 	/* Returns PF_DROP or *m0 is NULL or completely reassembled mbuf. */
1551 	if (pf_reassemble6(m0, h, &frag, off, extoff, dir, reason) != PF_PASS)
1552 		return (PF_DROP);
1553 	m = *m0;
1554 	if (m == NULL)
1555 		return (PF_DROP);
1556 
1557 	pd->flags |= PFDESC_IP_REAS;
1558 	return (PF_PASS);
1559 
1560  shortpkt:
1561 	REASON_SET(reason, PFRES_SHORT);
1562 	if (r != NULL && r->log)
1563 		PFLOG_PACKET(kif, m, AF_INET6, dir, *reason, r, NULL, NULL, pd,
1564 		    1);
1565 	return (PF_DROP);
1566 
1567  drop:
1568 	REASON_SET(reason, PFRES_NORM);
1569 	if (r != NULL && r->log)
1570 		PFLOG_PACKET(kif, m, AF_INET6, dir, *reason, r, NULL, NULL, pd,
1571 		    1);
1572 	return (PF_DROP);
1573 }
1574 #endif /* INET6 */
1575 
1576 int
1577 pf_normalize_tcp(int dir, struct pfi_kif *kif, struct mbuf *m, int ipoff,
1578     int off, void *h, struct pf_pdesc *pd)
1579 {
1580 	struct pf_rule	*r, *rm = NULL;
1581 	struct tcphdr	*th = pd->hdr.tcp;
1582 	int		 rewrite = 0;
1583 	u_short		 reason;
1584 	u_int8_t	 flags;
1585 	sa_family_t	 af = pd->af;
1586 
1587 	PF_RULES_RASSERT();
1588 
1589 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1590 	while (r != NULL) {
1591 		r->evaluations++;
1592 		if (pfi_kif_match(r->kif, kif) == r->ifnot)
1593 			r = r->skip[PF_SKIP_IFP].ptr;
1594 		else if (r->direction && r->direction != dir)
1595 			r = r->skip[PF_SKIP_DIR].ptr;
1596 		else if (r->af && r->af != af)
1597 			r = r->skip[PF_SKIP_AF].ptr;
1598 		else if (r->proto && r->proto != pd->proto)
1599 			r = r->skip[PF_SKIP_PROTO].ptr;
1600 		else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
1601 		    r->src.neg, kif, M_GETFIB(m)))
1602 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
1603 		else if (r->src.port_op && !pf_match_port(r->src.port_op,
1604 			    r->src.port[0], r->src.port[1], th->th_sport))
1605 			r = r->skip[PF_SKIP_SRC_PORT].ptr;
1606 		else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
1607 		    r->dst.neg, NULL, M_GETFIB(m)))
1608 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
1609 		else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
1610 			    r->dst.port[0], r->dst.port[1], th->th_dport))
1611 			r = r->skip[PF_SKIP_DST_PORT].ptr;
1612 		else if (r->os_fingerprint != PF_OSFP_ANY && !pf_osfp_match(
1613 			    pf_osfp_fingerprint(pd, m, off, th),
1614 			    r->os_fingerprint))
1615 			r = TAILQ_NEXT(r, entries);
1616 		else {
1617 			rm = r;
1618 			break;
1619 		}
1620 	}
1621 
1622 	if (rm == NULL || rm->action == PF_NOSCRUB)
1623 		return (PF_PASS);
1624 	else {
1625 		r->packets[dir == PF_OUT]++;
1626 		r->bytes[dir == PF_OUT] += pd->tot_len;
1627 	}
1628 
1629 	if (rm->rule_flag & PFRULE_REASSEMBLE_TCP)
1630 		pd->flags |= PFDESC_TCP_NORM;
1631 
1632 	flags = th->th_flags;
1633 	if (flags & TH_SYN) {
1634 		/* Illegal packet */
1635 		if (flags & TH_RST)
1636 			goto tcp_drop;
1637 
1638 		if (flags & TH_FIN)
1639 			flags &= ~TH_FIN;
1640 	} else {
1641 		/* Illegal packet */
1642 		if (!(flags & (TH_ACK|TH_RST)))
1643 			goto tcp_drop;
1644 	}
1645 
1646 	if (!(flags & TH_ACK)) {
1647 		/* These flags are only valid if ACK is set */
1648 		if ((flags & TH_FIN) || (flags & TH_PUSH) || (flags & TH_URG))
1649 			goto tcp_drop;
1650 	}
1651 
1652 	/* Check for illegal header length */
1653 	if (th->th_off < (sizeof(struct tcphdr) >> 2))
1654 		goto tcp_drop;
1655 
1656 	/* If flags changed, or reserved data set, then adjust */
1657 	if (flags != th->th_flags || th->th_x2 != 0) {
1658 		u_int16_t	ov, nv;
1659 
1660 		ov = *(u_int16_t *)(&th->th_ack + 1);
1661 		th->th_flags = flags;
1662 		th->th_x2 = 0;
1663 		nv = *(u_int16_t *)(&th->th_ack + 1);
1664 
1665 		th->th_sum = pf_cksum_fixup(th->th_sum, ov, nv, 0);
1666 		rewrite = 1;
1667 	}
1668 
1669 	/* Remove urgent pointer, if TH_URG is not set */
1670 	if (!(flags & TH_URG) && th->th_urp) {
1671 		th->th_sum = pf_cksum_fixup(th->th_sum, th->th_urp, 0, 0);
1672 		th->th_urp = 0;
1673 		rewrite = 1;
1674 	}
1675 
1676 	/* Process options */
1677 	if (r->max_mss && pf_normalize_tcpopt(r, m, th, off, pd->af))
1678 		rewrite = 1;
1679 
1680 	/* copy back packet headers if we sanitized */
1681 	if (rewrite)
1682 		m_copyback(m, off, sizeof(*th), (caddr_t)th);
1683 
1684 	return (PF_PASS);
1685 
1686  tcp_drop:
1687 	REASON_SET(&reason, PFRES_NORM);
1688 	if (rm != NULL && r->log)
1689 		PFLOG_PACKET(kif, m, AF_INET, dir, reason, r, NULL, NULL, pd,
1690 		    1);
1691 	return (PF_DROP);
1692 }
1693 
1694 int
1695 pf_normalize_tcp_init(struct mbuf *m, int off, struct pf_pdesc *pd,
1696     struct tcphdr *th, struct pf_state_peer *src, struct pf_state_peer *dst)
1697 {
1698 	u_int32_t tsval, tsecr;
1699 	u_int8_t hdr[60];
1700 	u_int8_t *opt;
1701 
1702 	KASSERT((src->scrub == NULL),
1703 	    ("pf_normalize_tcp_init: src->scrub != NULL"));
1704 
1705 	src->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT);
1706 	if (src->scrub == NULL)
1707 		return (1);
1708 
1709 	switch (pd->af) {
1710 #ifdef INET
1711 	case AF_INET: {
1712 		struct ip *h = mtod(m, struct ip *);
1713 		src->scrub->pfss_ttl = h->ip_ttl;
1714 		break;
1715 	}
1716 #endif /* INET */
1717 #ifdef INET6
1718 	case AF_INET6: {
1719 		struct ip6_hdr *h = mtod(m, struct ip6_hdr *);
1720 		src->scrub->pfss_ttl = h->ip6_hlim;
1721 		break;
1722 	}
1723 #endif /* INET6 */
1724 	}
1725 
1726 
1727 	/*
1728 	 * All normalizations below are only begun if we see the start of
1729 	 * the connections.  They must all set an enabled bit in pfss_flags
1730 	 */
1731 	if ((th->th_flags & TH_SYN) == 0)
1732 		return (0);
1733 
1734 
1735 	if (th->th_off > (sizeof(struct tcphdr) >> 2) && src->scrub &&
1736 	    pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) {
1737 		/* Diddle with TCP options */
1738 		int hlen;
1739 		opt = hdr + sizeof(struct tcphdr);
1740 		hlen = (th->th_off << 2) - sizeof(struct tcphdr);
1741 		while (hlen >= TCPOLEN_TIMESTAMP) {
1742 			switch (*opt) {
1743 			case TCPOPT_EOL:	/* FALLTHROUGH */
1744 			case TCPOPT_NOP:
1745 				opt++;
1746 				hlen--;
1747 				break;
1748 			case TCPOPT_TIMESTAMP:
1749 				if (opt[1] >= TCPOLEN_TIMESTAMP) {
1750 					src->scrub->pfss_flags |=
1751 					    PFSS_TIMESTAMP;
1752 					src->scrub->pfss_ts_mod =
1753 					    htonl(arc4random());
1754 
1755 					/* note PFSS_PAWS not set yet */
1756 					memcpy(&tsval, &opt[2],
1757 					    sizeof(u_int32_t));
1758 					memcpy(&tsecr, &opt[6],
1759 					    sizeof(u_int32_t));
1760 					src->scrub->pfss_tsval0 = ntohl(tsval);
1761 					src->scrub->pfss_tsval = ntohl(tsval);
1762 					src->scrub->pfss_tsecr = ntohl(tsecr);
1763 					getmicrouptime(&src->scrub->pfss_last);
1764 				}
1765 				/* FALLTHROUGH */
1766 			default:
1767 				hlen -= MAX(opt[1], 2);
1768 				opt += MAX(opt[1], 2);
1769 				break;
1770 			}
1771 		}
1772 	}
1773 
1774 	return (0);
1775 }
1776 
1777 void
1778 pf_normalize_tcp_cleanup(struct pf_state *state)
1779 {
1780 	if (state->src.scrub)
1781 		uma_zfree(V_pf_state_scrub_z, state->src.scrub);
1782 	if (state->dst.scrub)
1783 		uma_zfree(V_pf_state_scrub_z, state->dst.scrub);
1784 
1785 	/* Someday... flush the TCP segment reassembly descriptors. */
1786 }
1787 
1788 int
1789 pf_normalize_tcp_stateful(struct mbuf *m, int off, struct pf_pdesc *pd,
1790     u_short *reason, struct tcphdr *th, struct pf_state *state,
1791     struct pf_state_peer *src, struct pf_state_peer *dst, int *writeback)
1792 {
1793 	struct timeval uptime;
1794 	u_int32_t tsval, tsecr;
1795 	u_int tsval_from_last;
1796 	u_int8_t hdr[60];
1797 	u_int8_t *opt;
1798 	int copyback = 0;
1799 	int got_ts = 0;
1800 
1801 	KASSERT((src->scrub || dst->scrub),
1802 	    ("%s: src->scrub && dst->scrub!", __func__));
1803 
1804 	/*
1805 	 * Enforce the minimum TTL seen for this connection.  Negate a common
1806 	 * technique to evade an intrusion detection system and confuse
1807 	 * firewall state code.
1808 	 */
1809 	switch (pd->af) {
1810 #ifdef INET
1811 	case AF_INET: {
1812 		if (src->scrub) {
1813 			struct ip *h = mtod(m, struct ip *);
1814 			if (h->ip_ttl > src->scrub->pfss_ttl)
1815 				src->scrub->pfss_ttl = h->ip_ttl;
1816 			h->ip_ttl = src->scrub->pfss_ttl;
1817 		}
1818 		break;
1819 	}
1820 #endif /* INET */
1821 #ifdef INET6
1822 	case AF_INET6: {
1823 		if (src->scrub) {
1824 			struct ip6_hdr *h = mtod(m, struct ip6_hdr *);
1825 			if (h->ip6_hlim > src->scrub->pfss_ttl)
1826 				src->scrub->pfss_ttl = h->ip6_hlim;
1827 			h->ip6_hlim = src->scrub->pfss_ttl;
1828 		}
1829 		break;
1830 	}
1831 #endif /* INET6 */
1832 	}
1833 
1834 	if (th->th_off > (sizeof(struct tcphdr) >> 2) &&
1835 	    ((src->scrub && (src->scrub->pfss_flags & PFSS_TIMESTAMP)) ||
1836 	    (dst->scrub && (dst->scrub->pfss_flags & PFSS_TIMESTAMP))) &&
1837 	    pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) {
1838 		/* Diddle with TCP options */
1839 		int hlen;
1840 		opt = hdr + sizeof(struct tcphdr);
1841 		hlen = (th->th_off << 2) - sizeof(struct tcphdr);
1842 		while (hlen >= TCPOLEN_TIMESTAMP) {
1843 			switch (*opt) {
1844 			case TCPOPT_EOL:	/* FALLTHROUGH */
1845 			case TCPOPT_NOP:
1846 				opt++;
1847 				hlen--;
1848 				break;
1849 			case TCPOPT_TIMESTAMP:
1850 				/* Modulate the timestamps.  Can be used for
1851 				 * NAT detection, OS uptime determination or
1852 				 * reboot detection.
1853 				 */
1854 
1855 				if (got_ts) {
1856 					/* Huh?  Multiple timestamps!? */
1857 					if (V_pf_status.debug >= PF_DEBUG_MISC) {
1858 						DPFPRINTF(("multiple TS??"));
1859 						pf_print_state(state);
1860 						printf("\n");
1861 					}
1862 					REASON_SET(reason, PFRES_TS);
1863 					return (PF_DROP);
1864 				}
1865 				if (opt[1] >= TCPOLEN_TIMESTAMP) {
1866 					memcpy(&tsval, &opt[2],
1867 					    sizeof(u_int32_t));
1868 					if (tsval && src->scrub &&
1869 					    (src->scrub->pfss_flags &
1870 					    PFSS_TIMESTAMP)) {
1871 						tsval = ntohl(tsval);
1872 						pf_change_a(&opt[2],
1873 						    &th->th_sum,
1874 						    htonl(tsval +
1875 						    src->scrub->pfss_ts_mod),
1876 						    0);
1877 						copyback = 1;
1878 					}
1879 
1880 					/* Modulate TS reply iff valid (!0) */
1881 					memcpy(&tsecr, &opt[6],
1882 					    sizeof(u_int32_t));
1883 					if (tsecr && dst->scrub &&
1884 					    (dst->scrub->pfss_flags &
1885 					    PFSS_TIMESTAMP)) {
1886 						tsecr = ntohl(tsecr)
1887 						    - dst->scrub->pfss_ts_mod;
1888 						pf_change_a(&opt[6],
1889 						    &th->th_sum, htonl(tsecr),
1890 						    0);
1891 						copyback = 1;
1892 					}
1893 					got_ts = 1;
1894 				}
1895 				/* FALLTHROUGH */
1896 			default:
1897 				hlen -= MAX(opt[1], 2);
1898 				opt += MAX(opt[1], 2);
1899 				break;
1900 			}
1901 		}
1902 		if (copyback) {
1903 			/* Copyback the options, caller copys back header */
1904 			*writeback = 1;
1905 			m_copyback(m, off + sizeof(struct tcphdr),
1906 			    (th->th_off << 2) - sizeof(struct tcphdr), hdr +
1907 			    sizeof(struct tcphdr));
1908 		}
1909 	}
1910 
1911 
1912 	/*
1913 	 * Must invalidate PAWS checks on connections idle for too long.
1914 	 * The fastest allowed timestamp clock is 1ms.  That turns out to
1915 	 * be about 24 days before it wraps.  XXX Right now our lowerbound
1916 	 * TS echo check only works for the first 12 days of a connection
1917 	 * when the TS has exhausted half its 32bit space
1918 	 */
1919 #define TS_MAX_IDLE	(24*24*60*60)
1920 #define TS_MAX_CONN	(12*24*60*60)	/* XXX remove when better tsecr check */
1921 
1922 	getmicrouptime(&uptime);
1923 	if (src->scrub && (src->scrub->pfss_flags & PFSS_PAWS) &&
1924 	    (uptime.tv_sec - src->scrub->pfss_last.tv_sec > TS_MAX_IDLE ||
1925 	    time_uptime - state->creation > TS_MAX_CONN))  {
1926 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
1927 			DPFPRINTF(("src idled out of PAWS\n"));
1928 			pf_print_state(state);
1929 			printf("\n");
1930 		}
1931 		src->scrub->pfss_flags = (src->scrub->pfss_flags & ~PFSS_PAWS)
1932 		    | PFSS_PAWS_IDLED;
1933 	}
1934 	if (dst->scrub && (dst->scrub->pfss_flags & PFSS_PAWS) &&
1935 	    uptime.tv_sec - dst->scrub->pfss_last.tv_sec > TS_MAX_IDLE) {
1936 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
1937 			DPFPRINTF(("dst idled out of PAWS\n"));
1938 			pf_print_state(state);
1939 			printf("\n");
1940 		}
1941 		dst->scrub->pfss_flags = (dst->scrub->pfss_flags & ~PFSS_PAWS)
1942 		    | PFSS_PAWS_IDLED;
1943 	}
1944 
1945 	if (got_ts && src->scrub && dst->scrub &&
1946 	    (src->scrub->pfss_flags & PFSS_PAWS) &&
1947 	    (dst->scrub->pfss_flags & PFSS_PAWS)) {
1948 		/* Validate that the timestamps are "in-window".
1949 		 * RFC1323 describes TCP Timestamp options that allow
1950 		 * measurement of RTT (round trip time) and PAWS
1951 		 * (protection against wrapped sequence numbers).  PAWS
1952 		 * gives us a set of rules for rejecting packets on
1953 		 * long fat pipes (packets that were somehow delayed
1954 		 * in transit longer than the time it took to send the
1955 		 * full TCP sequence space of 4Gb).  We can use these
1956 		 * rules and infer a few others that will let us treat
1957 		 * the 32bit timestamp and the 32bit echoed timestamp
1958 		 * as sequence numbers to prevent a blind attacker from
1959 		 * inserting packets into a connection.
1960 		 *
1961 		 * RFC1323 tells us:
1962 		 *  - The timestamp on this packet must be greater than
1963 		 *    or equal to the last value echoed by the other
1964 		 *    endpoint.  The RFC says those will be discarded
1965 		 *    since it is a dup that has already been acked.
1966 		 *    This gives us a lowerbound on the timestamp.
1967 		 *        timestamp >= other last echoed timestamp
1968 		 *  - The timestamp will be less than or equal to
1969 		 *    the last timestamp plus the time between the
1970 		 *    last packet and now.  The RFC defines the max
1971 		 *    clock rate as 1ms.  We will allow clocks to be
1972 		 *    up to 10% fast and will allow a total difference
1973 		 *    or 30 seconds due to a route change.  And this
1974 		 *    gives us an upperbound on the timestamp.
1975 		 *        timestamp <= last timestamp + max ticks
1976 		 *    We have to be careful here.  Windows will send an
1977 		 *    initial timestamp of zero and then initialize it
1978 		 *    to a random value after the 3whs; presumably to
1979 		 *    avoid a DoS by having to call an expensive RNG
1980 		 *    during a SYN flood.  Proof MS has at least one
1981 		 *    good security geek.
1982 		 *
1983 		 *  - The TCP timestamp option must also echo the other
1984 		 *    endpoints timestamp.  The timestamp echoed is the
1985 		 *    one carried on the earliest unacknowledged segment
1986 		 *    on the left edge of the sequence window.  The RFC
1987 		 *    states that the host will reject any echoed
1988 		 *    timestamps that were larger than any ever sent.
1989 		 *    This gives us an upperbound on the TS echo.
1990 		 *        tescr <= largest_tsval
1991 		 *  - The lowerbound on the TS echo is a little more
1992 		 *    tricky to determine.  The other endpoint's echoed
1993 		 *    values will not decrease.  But there may be
1994 		 *    network conditions that re-order packets and
1995 		 *    cause our view of them to decrease.  For now the
1996 		 *    only lowerbound we can safely determine is that
1997 		 *    the TS echo will never be less than the original
1998 		 *    TS.  XXX There is probably a better lowerbound.
1999 		 *    Remove TS_MAX_CONN with better lowerbound check.
2000 		 *        tescr >= other original TS
2001 		 *
2002 		 * It is also important to note that the fastest
2003 		 * timestamp clock of 1ms will wrap its 32bit space in
2004 		 * 24 days.  So we just disable TS checking after 24
2005 		 * days of idle time.  We actually must use a 12d
2006 		 * connection limit until we can come up with a better
2007 		 * lowerbound to the TS echo check.
2008 		 */
2009 		struct timeval delta_ts;
2010 		int ts_fudge;
2011 
2012 
2013 		/*
2014 		 * PFTM_TS_DIFF is how many seconds of leeway to allow
2015 		 * a host's timestamp.  This can happen if the previous
2016 		 * packet got delayed in transit for much longer than
2017 		 * this packet.
2018 		 */
2019 		if ((ts_fudge = state->rule.ptr->timeout[PFTM_TS_DIFF]) == 0)
2020 			ts_fudge = V_pf_default_rule.timeout[PFTM_TS_DIFF];
2021 
2022 		/* Calculate max ticks since the last timestamp */
2023 #define TS_MAXFREQ	1100		/* RFC max TS freq of 1Khz + 10% skew */
2024 #define TS_MICROSECS	1000000		/* microseconds per second */
2025 		delta_ts = uptime;
2026 		timevalsub(&delta_ts, &src->scrub->pfss_last);
2027 		tsval_from_last = (delta_ts.tv_sec + ts_fudge) * TS_MAXFREQ;
2028 		tsval_from_last += delta_ts.tv_usec / (TS_MICROSECS/TS_MAXFREQ);
2029 
2030 		if ((src->state >= TCPS_ESTABLISHED &&
2031 		    dst->state >= TCPS_ESTABLISHED) &&
2032 		    (SEQ_LT(tsval, dst->scrub->pfss_tsecr) ||
2033 		    SEQ_GT(tsval, src->scrub->pfss_tsval + tsval_from_last) ||
2034 		    (tsecr && (SEQ_GT(tsecr, dst->scrub->pfss_tsval) ||
2035 		    SEQ_LT(tsecr, dst->scrub->pfss_tsval0))))) {
2036 			/* Bad RFC1323 implementation or an insertion attack.
2037 			 *
2038 			 * - Solaris 2.6 and 2.7 are known to send another ACK
2039 			 *   after the FIN,FIN|ACK,ACK closing that carries
2040 			 *   an old timestamp.
2041 			 */
2042 
2043 			DPFPRINTF(("Timestamp failed %c%c%c%c\n",
2044 			    SEQ_LT(tsval, dst->scrub->pfss_tsecr) ? '0' : ' ',
2045 			    SEQ_GT(tsval, src->scrub->pfss_tsval +
2046 			    tsval_from_last) ? '1' : ' ',
2047 			    SEQ_GT(tsecr, dst->scrub->pfss_tsval) ? '2' : ' ',
2048 			    SEQ_LT(tsecr, dst->scrub->pfss_tsval0)? '3' : ' '));
2049 			DPFPRINTF((" tsval: %u  tsecr: %u  +ticks: %u  "
2050 			    "idle: %jus %lums\n",
2051 			    tsval, tsecr, tsval_from_last,
2052 			    (uintmax_t)delta_ts.tv_sec,
2053 			    delta_ts.tv_usec / 1000));
2054 			DPFPRINTF((" src->tsval: %u  tsecr: %u\n",
2055 			    src->scrub->pfss_tsval, src->scrub->pfss_tsecr));
2056 			DPFPRINTF((" dst->tsval: %u  tsecr: %u  tsval0: %u"
2057 			    "\n", dst->scrub->pfss_tsval,
2058 			    dst->scrub->pfss_tsecr, dst->scrub->pfss_tsval0));
2059 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
2060 				pf_print_state(state);
2061 				pf_print_flags(th->th_flags);
2062 				printf("\n");
2063 			}
2064 			REASON_SET(reason, PFRES_TS);
2065 			return (PF_DROP);
2066 		}
2067 
2068 		/* XXX I'd really like to require tsecr but it's optional */
2069 
2070 	} else if (!got_ts && (th->th_flags & TH_RST) == 0 &&
2071 	    ((src->state == TCPS_ESTABLISHED && dst->state == TCPS_ESTABLISHED)
2072 	    || pd->p_len > 0 || (th->th_flags & TH_SYN)) &&
2073 	    src->scrub && dst->scrub &&
2074 	    (src->scrub->pfss_flags & PFSS_PAWS) &&
2075 	    (dst->scrub->pfss_flags & PFSS_PAWS)) {
2076 		/* Didn't send a timestamp.  Timestamps aren't really useful
2077 		 * when:
2078 		 *  - connection opening or closing (often not even sent).
2079 		 *    but we must not let an attacker to put a FIN on a
2080 		 *    data packet to sneak it through our ESTABLISHED check.
2081 		 *  - on a TCP reset.  RFC suggests not even looking at TS.
2082 		 *  - on an empty ACK.  The TS will not be echoed so it will
2083 		 *    probably not help keep the RTT calculation in sync and
2084 		 *    there isn't as much danger when the sequence numbers
2085 		 *    got wrapped.  So some stacks don't include TS on empty
2086 		 *    ACKs :-(
2087 		 *
2088 		 * To minimize the disruption to mostly RFC1323 conformant
2089 		 * stacks, we will only require timestamps on data packets.
2090 		 *
2091 		 * And what do ya know, we cannot require timestamps on data
2092 		 * packets.  There appear to be devices that do legitimate
2093 		 * TCP connection hijacking.  There are HTTP devices that allow
2094 		 * a 3whs (with timestamps) and then buffer the HTTP request.
2095 		 * If the intermediate device has the HTTP response cache, it
2096 		 * will spoof the response but not bother timestamping its
2097 		 * packets.  So we can look for the presence of a timestamp in
2098 		 * the first data packet and if there, require it in all future
2099 		 * packets.
2100 		 */
2101 
2102 		if (pd->p_len > 0 && (src->scrub->pfss_flags & PFSS_DATA_TS)) {
2103 			/*
2104 			 * Hey!  Someone tried to sneak a packet in.  Or the
2105 			 * stack changed its RFC1323 behavior?!?!
2106 			 */
2107 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
2108 				DPFPRINTF(("Did not receive expected RFC1323 "
2109 				    "timestamp\n"));
2110 				pf_print_state(state);
2111 				pf_print_flags(th->th_flags);
2112 				printf("\n");
2113 			}
2114 			REASON_SET(reason, PFRES_TS);
2115 			return (PF_DROP);
2116 		}
2117 	}
2118 
2119 
2120 	/*
2121 	 * We will note if a host sends his data packets with or without
2122 	 * timestamps.  And require all data packets to contain a timestamp
2123 	 * if the first does.  PAWS implicitly requires that all data packets be
2124 	 * timestamped.  But I think there are middle-man devices that hijack
2125 	 * TCP streams immediately after the 3whs and don't timestamp their
2126 	 * packets (seen in a WWW accelerator or cache).
2127 	 */
2128 	if (pd->p_len > 0 && src->scrub && (src->scrub->pfss_flags &
2129 	    (PFSS_TIMESTAMP|PFSS_DATA_TS|PFSS_DATA_NOTS)) == PFSS_TIMESTAMP) {
2130 		if (got_ts)
2131 			src->scrub->pfss_flags |= PFSS_DATA_TS;
2132 		else {
2133 			src->scrub->pfss_flags |= PFSS_DATA_NOTS;
2134 			if (V_pf_status.debug >= PF_DEBUG_MISC && dst->scrub &&
2135 			    (dst->scrub->pfss_flags & PFSS_TIMESTAMP)) {
2136 				/* Don't warn if other host rejected RFC1323 */
2137 				DPFPRINTF(("Broken RFC1323 stack did not "
2138 				    "timestamp data packet. Disabled PAWS "
2139 				    "security.\n"));
2140 				pf_print_state(state);
2141 				pf_print_flags(th->th_flags);
2142 				printf("\n");
2143 			}
2144 		}
2145 	}
2146 
2147 
2148 	/*
2149 	 * Update PAWS values
2150 	 */
2151 	if (got_ts && src->scrub && PFSS_TIMESTAMP == (src->scrub->pfss_flags &
2152 	    (PFSS_PAWS_IDLED|PFSS_TIMESTAMP))) {
2153 		getmicrouptime(&src->scrub->pfss_last);
2154 		if (SEQ_GEQ(tsval, src->scrub->pfss_tsval) ||
2155 		    (src->scrub->pfss_flags & PFSS_PAWS) == 0)
2156 			src->scrub->pfss_tsval = tsval;
2157 
2158 		if (tsecr) {
2159 			if (SEQ_GEQ(tsecr, src->scrub->pfss_tsecr) ||
2160 			    (src->scrub->pfss_flags & PFSS_PAWS) == 0)
2161 				src->scrub->pfss_tsecr = tsecr;
2162 
2163 			if ((src->scrub->pfss_flags & PFSS_PAWS) == 0 &&
2164 			    (SEQ_LT(tsval, src->scrub->pfss_tsval0) ||
2165 			    src->scrub->pfss_tsval0 == 0)) {
2166 				/* tsval0 MUST be the lowest timestamp */
2167 				src->scrub->pfss_tsval0 = tsval;
2168 			}
2169 
2170 			/* Only fully initialized after a TS gets echoed */
2171 			if ((src->scrub->pfss_flags & PFSS_PAWS) == 0)
2172 				src->scrub->pfss_flags |= PFSS_PAWS;
2173 		}
2174 	}
2175 
2176 	/* I have a dream....  TCP segment reassembly.... */
2177 	return (0);
2178 }
2179 
2180 static int
2181 pf_normalize_tcpopt(struct pf_rule *r, struct mbuf *m, struct tcphdr *th,
2182     int off, sa_family_t af)
2183 {
2184 	u_int16_t	*mss;
2185 	int		 thoff;
2186 	int		 opt, cnt, optlen = 0;
2187 	int		 rewrite = 0;
2188 	u_char		 opts[TCP_MAXOLEN];
2189 	u_char		*optp = opts;
2190 
2191 	thoff = th->th_off << 2;
2192 	cnt = thoff - sizeof(struct tcphdr);
2193 
2194 	if (cnt > 0 && !pf_pull_hdr(m, off + sizeof(*th), opts, cnt,
2195 	    NULL, NULL, af))
2196 		return (rewrite);
2197 
2198 	for (; cnt > 0; cnt -= optlen, optp += optlen) {
2199 		opt = optp[0];
2200 		if (opt == TCPOPT_EOL)
2201 			break;
2202 		if (opt == TCPOPT_NOP)
2203 			optlen = 1;
2204 		else {
2205 			if (cnt < 2)
2206 				break;
2207 			optlen = optp[1];
2208 			if (optlen < 2 || optlen > cnt)
2209 				break;
2210 		}
2211 		switch (opt) {
2212 		case TCPOPT_MAXSEG:
2213 			mss = (u_int16_t *)(optp + 2);
2214 			if ((ntohs(*mss)) > r->max_mss) {
2215 				th->th_sum = pf_cksum_fixup(th->th_sum,
2216 				    *mss, htons(r->max_mss), 0);
2217 				*mss = htons(r->max_mss);
2218 				rewrite = 1;
2219 			}
2220 			break;
2221 		default:
2222 			break;
2223 		}
2224 	}
2225 
2226 	if (rewrite)
2227 		m_copyback(m, off + sizeof(*th), thoff - sizeof(*th), opts);
2228 
2229 	return (rewrite);
2230 }
2231 
2232 #ifdef INET
2233 static void
2234 pf_scrub_ip(struct mbuf **m0, u_int32_t flags, u_int8_t min_ttl, u_int8_t tos)
2235 {
2236 	struct mbuf		*m = *m0;
2237 	struct ip		*h = mtod(m, struct ip *);
2238 
2239 	/* Clear IP_DF if no-df was requested */
2240 	if (flags & PFRULE_NODF && h->ip_off & htons(IP_DF)) {
2241 		u_int16_t ip_off = h->ip_off;
2242 
2243 		h->ip_off &= htons(~IP_DF);
2244 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
2245 	}
2246 
2247 	/* Enforce a minimum ttl, may cause endless packet loops */
2248 	if (min_ttl && h->ip_ttl < min_ttl) {
2249 		u_int16_t ip_ttl = h->ip_ttl;
2250 
2251 		h->ip_ttl = min_ttl;
2252 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_ttl, h->ip_ttl, 0);
2253 	}
2254 
2255 	/* Enforce tos */
2256 	if (flags & PFRULE_SET_TOS) {
2257 		u_int16_t	ov, nv;
2258 
2259 		ov = *(u_int16_t *)h;
2260 		h->ip_tos = tos;
2261 		nv = *(u_int16_t *)h;
2262 
2263 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ov, nv, 0);
2264 	}
2265 
2266 	/* random-id, but not for fragments */
2267 	if (flags & PFRULE_RANDOMID && !(h->ip_off & ~htons(IP_DF))) {
2268 		u_int16_t ip_id = h->ip_id;
2269 
2270 		h->ip_id = ip_randomid();
2271 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_id, h->ip_id, 0);
2272 	}
2273 }
2274 #endif /* INET */
2275 
2276 #ifdef INET6
2277 static void
2278 pf_scrub_ip6(struct mbuf **m0, u_int8_t min_ttl)
2279 {
2280 	struct mbuf		*m = *m0;
2281 	struct ip6_hdr		*h = mtod(m, struct ip6_hdr *);
2282 
2283 	/* Enforce a minimum ttl, may cause endless packet loops */
2284 	if (min_ttl && h->ip6_hlim < min_ttl)
2285 		h->ip6_hlim = min_ttl;
2286 }
2287 #endif
2288