xref: /freebsd/sys/netpfil/pf/pf_norm.c (revision 6137b5f7b8c183ee8806d79b3f1d8e5e3ddb3df3)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright 2001 Niels Provos <provos@citi.umich.edu>
5  * Copyright 2011-2018 Alexander Bluhm <bluhm@openbsd.org>
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27  *
28  *	$OpenBSD: pf_norm.c,v 1.114 2009/01/29 14:11:45 henning Exp $
29  */
30 
31 #include <sys/cdefs.h>
32 #include "opt_inet.h"
33 #include "opt_inet6.h"
34 #include "opt_pf.h"
35 
36 #include <sys/param.h>
37 #include <sys/kernel.h>
38 #include <sys/lock.h>
39 #include <sys/mbuf.h>
40 #include <sys/mutex.h>
41 #include <sys/refcount.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 <netinet6/scope6_var.h>
54 #include <netinet/tcp.h>
55 #include <netinet/tcp_fsm.h>
56 #include <netinet/tcp_seq.h>
57 #include <netinet/sctp_constants.h>
58 #include <netinet/sctp_header.h>
59 
60 #ifdef INET6
61 #include <netinet/ip6.h>
62 #endif /* INET6 */
63 
64 struct pf_frent {
65 	TAILQ_ENTRY(pf_frent)	fr_next;
66 	struct mbuf	*fe_m;
67 	uint16_t	fe_hdrlen;	/* ipv4 header length with ip options
68 					   ipv6, extension, fragment header */
69 	uint16_t	fe_extoff;	/* last extension header offset or 0 */
70 	uint16_t	fe_len;		/* fragment length */
71 	uint16_t	fe_off;		/* fragment offset */
72 	uint16_t	fe_mff;		/* more fragment flag */
73 };
74 
75 struct pf_fragment_cmp {
76 	struct pf_addr	frc_src;
77 	struct pf_addr	frc_dst;
78 	uint32_t	frc_id;
79 	sa_family_t	frc_af;
80 	uint8_t		frc_proto;
81 };
82 
83 struct pf_fragment {
84 	struct pf_fragment_cmp	fr_key;
85 #define fr_src	fr_key.frc_src
86 #define fr_dst	fr_key.frc_dst
87 #define fr_id	fr_key.frc_id
88 #define fr_af	fr_key.frc_af
89 #define fr_proto	fr_key.frc_proto
90 
91 	/* pointers to queue element */
92 	struct pf_frent	*fr_firstoff[PF_FRAG_ENTRY_POINTS];
93 	/* count entries between pointers */
94 	uint8_t	fr_entries[PF_FRAG_ENTRY_POINTS];
95 	RB_ENTRY(pf_fragment) fr_entry;
96 	TAILQ_ENTRY(pf_fragment) frag_next;
97 	uint32_t	fr_timeout;
98 	uint16_t	fr_maxlen;	/* maximum length of single fragment */
99 	u_int16_t	fr_holes;	/* number of holes in the queue */
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 	uint32_t	ft_id;		/* fragment id */
108 };
109 
110 VNET_DEFINE_STATIC(struct mtx, pf_frag_mtx);
111 #define V_pf_frag_mtx		VNET(pf_frag_mtx)
112 #define PF_FRAG_LOCK()		mtx_lock(&V_pf_frag_mtx)
113 #define PF_FRAG_UNLOCK()	mtx_unlock(&V_pf_frag_mtx)
114 #define PF_FRAG_ASSERT()	mtx_assert(&V_pf_frag_mtx, MA_OWNED)
115 
116 VNET_DEFINE(uma_zone_t, pf_state_scrub_z);	/* XXX: shared with pfsync */
117 
118 VNET_DEFINE_STATIC(uma_zone_t, pf_frent_z);
119 #define	V_pf_frent_z	VNET(pf_frent_z)
120 VNET_DEFINE_STATIC(uma_zone_t, pf_frag_z);
121 #define	V_pf_frag_z	VNET(pf_frag_z)
122 
123 TAILQ_HEAD(pf_fragqueue, pf_fragment);
124 TAILQ_HEAD(pf_cachequeue, pf_fragment);
125 VNET_DEFINE_STATIC(struct pf_fragqueue,	pf_fragqueue);
126 #define	V_pf_fragqueue			VNET(pf_fragqueue)
127 RB_HEAD(pf_frag_tree, pf_fragment);
128 VNET_DEFINE_STATIC(struct pf_frag_tree,	pf_frag_tree);
129 #define	V_pf_frag_tree			VNET(pf_frag_tree)
130 static int		 pf_frag_compare(struct pf_fragment *,
131 			    struct pf_fragment *);
132 static RB_PROTOTYPE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
133 static RB_GENERATE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
134 
135 static void	pf_flush_fragments(void);
136 static void	pf_free_fragment(struct pf_fragment *);
137 static void	pf_remove_fragment(struct pf_fragment *);
138 
139 static struct pf_frent *pf_create_fragment(u_short *);
140 static int	pf_frent_holes(struct pf_frent *frent);
141 static struct pf_fragment *pf_find_fragment(struct pf_fragment_cmp *key,
142 		    struct pf_frag_tree *tree);
143 static inline int	pf_frent_index(struct pf_frent *);
144 static int	pf_frent_insert(struct pf_fragment *,
145 			    struct pf_frent *, struct pf_frent *);
146 void			pf_frent_remove(struct pf_fragment *,
147 			    struct pf_frent *);
148 struct pf_frent		*pf_frent_previous(struct pf_fragment *,
149 			    struct pf_frent *);
150 static struct pf_fragment *pf_fillup_fragment(struct pf_fragment_cmp *,
151 		    struct pf_frent *, u_short *);
152 static struct mbuf *pf_join_fragment(struct pf_fragment *);
153 #ifdef INET
154 static int	pf_reassemble(struct mbuf **, struct ip *, int, u_short *);
155 #endif	/* INET */
156 #ifdef INET6
157 static int	pf_reassemble6(struct mbuf **, struct ip6_hdr *,
158 		    struct ip6_frag *, uint16_t, uint16_t, u_short *);
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 }
179 #endif	/* INET */
180 
181 void
182 pf_normalize_init(void)
183 {
184 
185 	V_pf_frag_z = uma_zcreate("pf frags", sizeof(struct pf_fragment),
186 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
187 	V_pf_frent_z = uma_zcreate("pf frag entries", sizeof(struct pf_frent),
188 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
189 	V_pf_state_scrub_z = uma_zcreate("pf state scrubs",
190 	    sizeof(struct pf_state_scrub),  NULL, NULL, NULL, NULL,
191 	    UMA_ALIGN_PTR, 0);
192 
193 	mtx_init(&V_pf_frag_mtx, "pf fragments", NULL, MTX_DEF);
194 
195 	V_pf_limits[PF_LIMIT_FRAGS].zone = V_pf_frent_z;
196 	V_pf_limits[PF_LIMIT_FRAGS].limit = PFFRAG_FRENT_HIWAT;
197 	uma_zone_set_max(V_pf_frent_z, PFFRAG_FRENT_HIWAT);
198 	uma_zone_set_warning(V_pf_frent_z, "PF frag entries limit reached");
199 
200 	TAILQ_INIT(&V_pf_fragqueue);
201 }
202 
203 void
204 pf_normalize_cleanup(void)
205 {
206 
207 	uma_zdestroy(V_pf_state_scrub_z);
208 	uma_zdestroy(V_pf_frent_z);
209 	uma_zdestroy(V_pf_frag_z);
210 
211 	mtx_destroy(&V_pf_frag_mtx);
212 }
213 
214 static int
215 pf_frag_compare(struct pf_fragment *a, struct pf_fragment *b)
216 {
217 	int	diff;
218 
219 	if ((diff = a->fr_id - b->fr_id) != 0)
220 		return (diff);
221 	if ((diff = a->fr_proto - b->fr_proto) != 0)
222 		return (diff);
223 	if ((diff = a->fr_af - b->fr_af) != 0)
224 		return (diff);
225 	if ((diff = pf_addr_cmp(&a->fr_src, &b->fr_src, a->fr_af)) != 0)
226 		return (diff);
227 	if ((diff = pf_addr_cmp(&a->fr_dst, &b->fr_dst, a->fr_af)) != 0)
228 		return (diff);
229 	return (0);
230 }
231 
232 void
233 pf_purge_expired_fragments(void)
234 {
235 	u_int32_t	expire = time_uptime -
236 			    V_pf_default_rule.timeout[PFTM_FRAG];
237 
238 	pf_purge_fragments(expire);
239 }
240 
241 void
242 pf_purge_fragments(uint32_t expire)
243 {
244 	struct pf_fragment	*frag;
245 
246 	PF_FRAG_LOCK();
247 	while ((frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue)) != NULL) {
248 		if (frag->fr_timeout > expire)
249 			break;
250 
251 		DPFPRINTF(("expiring %d(%p)\n", frag->fr_id, frag));
252 		pf_free_fragment(frag);
253 	}
254 
255 	PF_FRAG_UNLOCK();
256 }
257 
258 /*
259  * Try to flush old fragments to make space for new ones
260  */
261 static void
262 pf_flush_fragments(void)
263 {
264 	struct pf_fragment	*frag;
265 	int			 goal;
266 
267 	PF_FRAG_ASSERT();
268 
269 	goal = uma_zone_get_cur(V_pf_frent_z) * 9 / 10;
270 	DPFPRINTF(("trying to free %d frag entriess\n", goal));
271 	while (goal < uma_zone_get_cur(V_pf_frent_z)) {
272 		frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue);
273 		if (frag)
274 			pf_free_fragment(frag);
275 		else
276 			break;
277 	}
278 }
279 
280 /* Frees the fragments and all associated entries */
281 static void
282 pf_free_fragment(struct pf_fragment *frag)
283 {
284 	struct pf_frent		*frent;
285 
286 	PF_FRAG_ASSERT();
287 
288 	/* Free all fragments */
289 	for (frent = TAILQ_FIRST(&frag->fr_queue); frent;
290 	    frent = TAILQ_FIRST(&frag->fr_queue)) {
291 		TAILQ_REMOVE(&frag->fr_queue, frent, fr_next);
292 
293 		m_freem(frent->fe_m);
294 		uma_zfree(V_pf_frent_z, frent);
295 	}
296 
297 	pf_remove_fragment(frag);
298 }
299 
300 static struct pf_fragment *
301 pf_find_fragment(struct pf_fragment_cmp *key, struct pf_frag_tree *tree)
302 {
303 	struct pf_fragment	*frag;
304 
305 	PF_FRAG_ASSERT();
306 
307 	frag = RB_FIND(pf_frag_tree, tree, (struct pf_fragment *)key);
308 	if (frag != NULL) {
309 		/* XXX Are we sure we want to update the timeout? */
310 		frag->fr_timeout = time_uptime;
311 		TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next);
312 		TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next);
313 	}
314 
315 	return (frag);
316 }
317 
318 /* Removes a fragment from the fragment queue and frees the fragment */
319 static void
320 pf_remove_fragment(struct pf_fragment *frag)
321 {
322 
323 	PF_FRAG_ASSERT();
324 	KASSERT(frag, ("frag != NULL"));
325 
326 	RB_REMOVE(pf_frag_tree, &V_pf_frag_tree, frag);
327 	TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next);
328 	uma_zfree(V_pf_frag_z, frag);
329 }
330 
331 static struct pf_frent *
332 pf_create_fragment(u_short *reason)
333 {
334 	struct pf_frent *frent;
335 
336 	PF_FRAG_ASSERT();
337 
338 	frent = uma_zalloc(V_pf_frent_z, M_NOWAIT);
339 	if (frent == NULL) {
340 		pf_flush_fragments();
341 		frent = uma_zalloc(V_pf_frent_z, M_NOWAIT);
342 		if (frent == NULL) {
343 			REASON_SET(reason, PFRES_MEMORY);
344 			return (NULL);
345 		}
346 	}
347 
348 	return (frent);
349 }
350 
351 /*
352  * Calculate the additional holes that were created in the fragment
353  * queue by inserting this fragment.  A fragment in the middle
354  * creates one more hole by splitting.  For each connected side,
355  * it loses one hole.
356  * Fragment entry must be in the queue when calling this function.
357  */
358 static int
359 pf_frent_holes(struct pf_frent *frent)
360 {
361 	struct pf_frent *prev = TAILQ_PREV(frent, pf_fragq, fr_next);
362 	struct pf_frent *next = TAILQ_NEXT(frent, fr_next);
363 	int holes = 1;
364 
365 	if (prev == NULL) {
366 		if (frent->fe_off == 0)
367 			holes--;
368 	} else {
369 		KASSERT(frent->fe_off != 0, ("frent->fe_off != 0"));
370 		if (frent->fe_off == prev->fe_off + prev->fe_len)
371 			holes--;
372 	}
373 	if (next == NULL) {
374 		if (!frent->fe_mff)
375 			holes--;
376 	} else {
377 		KASSERT(frent->fe_mff, ("frent->fe_mff"));
378 		if (next->fe_off == frent->fe_off + frent->fe_len)
379 			holes--;
380 	}
381 	return holes;
382 }
383 
384 static inline int
385 pf_frent_index(struct pf_frent *frent)
386 {
387 	/*
388 	 * We have an array of 16 entry points to the queue.  A full size
389 	 * 65535 octet IP packet can have 8192 fragments.  So the queue
390 	 * traversal length is at most 512 and at most 16 entry points are
391 	 * checked.  We need 128 additional bytes on a 64 bit architecture.
392 	 */
393 	CTASSERT(((u_int16_t)0xffff &~ 7) / (0x10000 / PF_FRAG_ENTRY_POINTS) ==
394 	    16 - 1);
395 	CTASSERT(((u_int16_t)0xffff >> 3) / PF_FRAG_ENTRY_POINTS == 512 - 1);
396 
397 	return frent->fe_off / (0x10000 / PF_FRAG_ENTRY_POINTS);
398 }
399 
400 static int
401 pf_frent_insert(struct pf_fragment *frag, struct pf_frent *frent,
402     struct pf_frent *prev)
403 {
404 	int index;
405 
406 	CTASSERT(PF_FRAG_ENTRY_LIMIT <= 0xff);
407 
408 	/*
409 	 * A packet has at most 65536 octets.  With 16 entry points, each one
410 	 * spawns 4096 octets.  We limit these to 64 fragments each, which
411 	 * means on average every fragment must have at least 64 octets.
412 	 */
413 	index = pf_frent_index(frent);
414 	if (frag->fr_entries[index] >= PF_FRAG_ENTRY_LIMIT)
415 		return ENOBUFS;
416 	frag->fr_entries[index]++;
417 
418 	if (prev == NULL) {
419 		TAILQ_INSERT_HEAD(&frag->fr_queue, frent, fr_next);
420 	} else {
421 		KASSERT(prev->fe_off + prev->fe_len <= frent->fe_off,
422 		    ("overlapping fragment"));
423 		TAILQ_INSERT_AFTER(&frag->fr_queue, prev, frent, fr_next);
424 	}
425 
426 	if (frag->fr_firstoff[index] == NULL) {
427 		KASSERT(prev == NULL || pf_frent_index(prev) < index,
428 		    ("prev == NULL || pf_frent_index(pref) < index"));
429 		frag->fr_firstoff[index] = frent;
430 	} else {
431 		if (frent->fe_off < frag->fr_firstoff[index]->fe_off) {
432 			KASSERT(prev == NULL || pf_frent_index(prev) < index,
433 			    ("prev == NULL || pf_frent_index(pref) < index"));
434 			frag->fr_firstoff[index] = frent;
435 		} else {
436 			KASSERT(prev != NULL, ("prev != NULL"));
437 			KASSERT(pf_frent_index(prev) == index,
438 			    ("pf_frent_index(prev) == index"));
439 		}
440 	}
441 
442 	frag->fr_holes += pf_frent_holes(frent);
443 
444 	return 0;
445 }
446 
447 void
448 pf_frent_remove(struct pf_fragment *frag, struct pf_frent *frent)
449 {
450 #ifdef INVARIANTS
451 	struct pf_frent *prev = TAILQ_PREV(frent, pf_fragq, fr_next);
452 #endif
453 	struct pf_frent *next = TAILQ_NEXT(frent, fr_next);
454 	int index;
455 
456 	frag->fr_holes -= pf_frent_holes(frent);
457 
458 	index = pf_frent_index(frent);
459 	KASSERT(frag->fr_firstoff[index] != NULL, ("frent not found"));
460 	if (frag->fr_firstoff[index]->fe_off == frent->fe_off) {
461 		if (next == NULL) {
462 			frag->fr_firstoff[index] = NULL;
463 		} else {
464 			KASSERT(frent->fe_off + frent->fe_len <= next->fe_off,
465 			    ("overlapping fragment"));
466 			if (pf_frent_index(next) == index) {
467 				frag->fr_firstoff[index] = next;
468 			} else {
469 				frag->fr_firstoff[index] = NULL;
470 			}
471 		}
472 	} else {
473 		KASSERT(frag->fr_firstoff[index]->fe_off < frent->fe_off,
474 		    ("frag->fr_firstoff[index]->fe_off < frent->fe_off"));
475 		KASSERT(prev != NULL, ("prev != NULL"));
476 		KASSERT(prev->fe_off + prev->fe_len <= frent->fe_off,
477 		    ("overlapping fragment"));
478 		KASSERT(pf_frent_index(prev) == index,
479 		    ("pf_frent_index(prev) == index"));
480 	}
481 
482 	TAILQ_REMOVE(&frag->fr_queue, frent, fr_next);
483 
484 	KASSERT(frag->fr_entries[index] > 0, ("No fragments remaining"));
485 	frag->fr_entries[index]--;
486 }
487 
488 struct pf_frent *
489 pf_frent_previous(struct pf_fragment *frag, struct pf_frent *frent)
490 {
491 	struct pf_frent *prev, *next;
492 	int index;
493 
494 	/*
495 	 * If there are no fragments after frag, take the final one.  Assume
496 	 * that the global queue is not empty.
497 	 */
498 	prev = TAILQ_LAST(&frag->fr_queue, pf_fragq);
499 	KASSERT(prev != NULL, ("prev != NULL"));
500 	if (prev->fe_off <= frent->fe_off)
501 		return prev;
502 	/*
503 	 * We want to find a fragment entry that is before frag, but still
504 	 * close to it.  Find the first fragment entry that is in the same
505 	 * entry point or in the first entry point after that.  As we have
506 	 * already checked that there are entries behind frag, this will
507 	 * succeed.
508 	 */
509 	for (index = pf_frent_index(frent); index < PF_FRAG_ENTRY_POINTS;
510 	    index++) {
511 		prev = frag->fr_firstoff[index];
512 		if (prev != NULL)
513 			break;
514 	}
515 	KASSERT(prev != NULL, ("prev != NULL"));
516 	/*
517 	 * In prev we may have a fragment from the same entry point that is
518 	 * before frent, or one that is just one position behind frent.
519 	 * In the latter case, we go back one step and have the predecessor.
520 	 * There may be none if the new fragment will be the first one.
521 	 */
522 	if (prev->fe_off > frent->fe_off) {
523 		prev = TAILQ_PREV(prev, pf_fragq, fr_next);
524 		if (prev == NULL)
525 			return NULL;
526 		KASSERT(prev->fe_off <= frent->fe_off,
527 		    ("prev->fe_off <= frent->fe_off"));
528 		return prev;
529 	}
530 	/*
531 	 * In prev is the first fragment of the entry point.  The offset
532 	 * of frag is behind it.  Find the closest previous fragment.
533 	 */
534 	for (next = TAILQ_NEXT(prev, fr_next); next != NULL;
535 	    next = TAILQ_NEXT(next, fr_next)) {
536 		if (next->fe_off > frent->fe_off)
537 			break;
538 		prev = next;
539 	}
540 	return prev;
541 }
542 
543 static struct pf_fragment *
544 pf_fillup_fragment(struct pf_fragment_cmp *key, struct pf_frent *frent,
545     u_short *reason)
546 {
547 	struct pf_frent		*after, *next, *prev;
548 	struct pf_fragment	*frag;
549 	uint16_t		total;
550 	int			old_index, new_index;
551 
552 	PF_FRAG_ASSERT();
553 
554 	/* No empty fragments. */
555 	if (frent->fe_len == 0) {
556 		DPFPRINTF(("bad fragment: len 0\n"));
557 		goto bad_fragment;
558 	}
559 
560 	/* All fragments are 8 byte aligned. */
561 	if (frent->fe_mff && (frent->fe_len & 0x7)) {
562 		DPFPRINTF(("bad fragment: mff and len %d\n", frent->fe_len));
563 		goto bad_fragment;
564 	}
565 
566 	/* Respect maximum length, IP_MAXPACKET == IPV6_MAXPACKET. */
567 	if (frent->fe_off + frent->fe_len > IP_MAXPACKET) {
568 		DPFPRINTF(("bad fragment: max packet %d\n",
569 		    frent->fe_off + frent->fe_len));
570 		goto bad_fragment;
571 	}
572 
573 	DPFPRINTF((key->frc_af == AF_INET ?
574 	    "reass frag %d @ %d-%d\n" : "reass frag %#08x @ %d-%d\n",
575 	    key->frc_id, frent->fe_off, frent->fe_off + frent->fe_len));
576 
577 	/* Fully buffer all of the fragments in this fragment queue. */
578 	frag = pf_find_fragment(key, &V_pf_frag_tree);
579 
580 	/* Create a new reassembly queue for this packet. */
581 	if (frag == NULL) {
582 		frag = uma_zalloc(V_pf_frag_z, M_NOWAIT);
583 		if (frag == NULL) {
584 			pf_flush_fragments();
585 			frag = uma_zalloc(V_pf_frag_z, M_NOWAIT);
586 			if (frag == NULL) {
587 				REASON_SET(reason, PFRES_MEMORY);
588 				goto drop_fragment;
589 			}
590 		}
591 
592 		*(struct pf_fragment_cmp *)frag = *key;
593 		memset(frag->fr_firstoff, 0, sizeof(frag->fr_firstoff));
594 		memset(frag->fr_entries, 0, sizeof(frag->fr_entries));
595 		frag->fr_timeout = time_uptime;
596 		frag->fr_maxlen = frent->fe_len;
597 		frag->fr_holes = 1;
598 		TAILQ_INIT(&frag->fr_queue);
599 
600 		RB_INSERT(pf_frag_tree, &V_pf_frag_tree, frag);
601 		TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next);
602 
603 		/* We do not have a previous fragment, cannot fail. */
604 		pf_frent_insert(frag, frent, NULL);
605 
606 		return (frag);
607 	}
608 
609 	KASSERT(!TAILQ_EMPTY(&frag->fr_queue), ("!TAILQ_EMPTY()->fr_queue"));
610 
611 	/* Remember maximum fragment len for refragmentation. */
612 	if (frent->fe_len > frag->fr_maxlen)
613 		frag->fr_maxlen = frent->fe_len;
614 
615 	/* Maximum data we have seen already. */
616 	total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
617 		TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
618 
619 	/* Non terminal fragments must have more fragments flag. */
620 	if (frent->fe_off + frent->fe_len < total && !frent->fe_mff)
621 		goto bad_fragment;
622 
623 	/* Check if we saw the last fragment already. */
624 	if (!TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_mff) {
625 		if (frent->fe_off + frent->fe_len > total ||
626 		    (frent->fe_off + frent->fe_len == total && frent->fe_mff))
627 			goto bad_fragment;
628 	} else {
629 		if (frent->fe_off + frent->fe_len == total && !frent->fe_mff)
630 			goto bad_fragment;
631 	}
632 
633 	/* Find neighbors for newly inserted fragment */
634 	prev = pf_frent_previous(frag, frent);
635 	if (prev == NULL) {
636 		after = TAILQ_FIRST(&frag->fr_queue);
637 		KASSERT(after != NULL, ("after != NULL"));
638 	} else {
639 		after = TAILQ_NEXT(prev, fr_next);
640 	}
641 
642 	if (prev != NULL && prev->fe_off + prev->fe_len > frent->fe_off) {
643 		uint16_t precut;
644 
645 		precut = prev->fe_off + prev->fe_len - frent->fe_off;
646 		if (precut >= frent->fe_len)
647 			goto bad_fragment;
648 		DPFPRINTF(("overlap -%d\n", precut));
649 		m_adj(frent->fe_m, precut);
650 		frent->fe_off += precut;
651 		frent->fe_len -= precut;
652 	}
653 
654 	for (; after != NULL && frent->fe_off + frent->fe_len > after->fe_off;
655 	    after = next) {
656 		uint16_t aftercut;
657 
658 		aftercut = frent->fe_off + frent->fe_len - after->fe_off;
659 		DPFPRINTF(("adjust overlap %d\n", aftercut));
660 		if (aftercut < after->fe_len) {
661 			m_adj(after->fe_m, aftercut);
662 			old_index = pf_frent_index(after);
663 			after->fe_off += aftercut;
664 			after->fe_len -= aftercut;
665 			new_index = pf_frent_index(after);
666 			if (old_index != new_index) {
667 				DPFPRINTF(("frag index %d, new %d",
668 				    old_index, new_index));
669 				/* Fragment switched queue as fe_off changed */
670 				after->fe_off -= aftercut;
671 				after->fe_len += aftercut;
672 				/* Remove restored fragment from old queue */
673 				pf_frent_remove(frag, after);
674 				after->fe_off += aftercut;
675 				after->fe_len -= aftercut;
676 				/* Insert into correct queue */
677 				if (pf_frent_insert(frag, after, prev)) {
678 					DPFPRINTF(
679 					    ("fragment requeue limit exceeded"));
680 					m_freem(after->fe_m);
681 					uma_zfree(V_pf_frent_z, after);
682 					/* There is not way to recover */
683 					goto bad_fragment;
684 				}
685 			}
686 			break;
687 		}
688 
689 		/* This fragment is completely overlapped, lose it. */
690 		next = TAILQ_NEXT(after, fr_next);
691 		pf_frent_remove(frag, after);
692 		m_freem(after->fe_m);
693 		uma_zfree(V_pf_frent_z, after);
694 	}
695 
696 	/* If part of the queue gets too long, there is not way to recover. */
697 	if (pf_frent_insert(frag, frent, prev)) {
698 		DPFPRINTF(("fragment queue limit exceeded\n"));
699 		goto bad_fragment;
700 	}
701 
702 	return (frag);
703 
704 bad_fragment:
705 	REASON_SET(reason, PFRES_FRAG);
706 drop_fragment:
707 	uma_zfree(V_pf_frent_z, frent);
708 	return (NULL);
709 }
710 
711 static struct mbuf *
712 pf_join_fragment(struct pf_fragment *frag)
713 {
714 	struct mbuf *m, *m2;
715 	struct pf_frent	*frent, *next;
716 
717 	frent = TAILQ_FIRST(&frag->fr_queue);
718 	next = TAILQ_NEXT(frent, fr_next);
719 
720 	m = frent->fe_m;
721 	m_adj(m, (frent->fe_hdrlen + frent->fe_len) - m->m_pkthdr.len);
722 	uma_zfree(V_pf_frent_z, frent);
723 	for (frent = next; frent != NULL; frent = next) {
724 		next = TAILQ_NEXT(frent, fr_next);
725 
726 		m2 = frent->fe_m;
727 		/* Strip off ip header. */
728 		m_adj(m2, frent->fe_hdrlen);
729 		/* Strip off any trailing bytes. */
730 		m_adj(m2, frent->fe_len - m2->m_pkthdr.len);
731 
732 		uma_zfree(V_pf_frent_z, frent);
733 		m_cat(m, m2);
734 	}
735 
736 	/* Remove from fragment queue. */
737 	pf_remove_fragment(frag);
738 
739 	return (m);
740 }
741 
742 #ifdef INET
743 static int
744 pf_reassemble(struct mbuf **m0, struct ip *ip, int dir, u_short *reason)
745 {
746 	struct mbuf		*m = *m0;
747 	struct pf_frent		*frent;
748 	struct pf_fragment	*frag;
749 	struct pf_fragment_cmp	key;
750 	uint16_t		total, hdrlen;
751 
752 	/* Get an entry for the fragment queue */
753 	if ((frent = pf_create_fragment(reason)) == NULL)
754 		return (PF_DROP);
755 
756 	frent->fe_m = m;
757 	frent->fe_hdrlen = ip->ip_hl << 2;
758 	frent->fe_extoff = 0;
759 	frent->fe_len = ntohs(ip->ip_len) - (ip->ip_hl << 2);
760 	frent->fe_off = (ntohs(ip->ip_off) & IP_OFFMASK) << 3;
761 	frent->fe_mff = ntohs(ip->ip_off) & IP_MF;
762 
763 	pf_ip2key(ip, dir, &key);
764 
765 	if ((frag = pf_fillup_fragment(&key, frent, reason)) == NULL)
766 		return (PF_DROP);
767 
768 	/* The mbuf is part of the fragment entry, no direct free or access */
769 	m = *m0 = NULL;
770 
771 	if (frag->fr_holes) {
772 		DPFPRINTF(("frag %d, holes %d\n", frag->fr_id, frag->fr_holes));
773 		return (PF_PASS);  /* drop because *m0 is NULL, no error */
774 	}
775 
776 	/* We have all the data */
777 	frent = TAILQ_FIRST(&frag->fr_queue);
778 	KASSERT(frent != NULL, ("frent != NULL"));
779 	total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
780 		TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
781 	hdrlen = frent->fe_hdrlen;
782 
783 	m = *m0 = pf_join_fragment(frag);
784 	frag = NULL;
785 
786 	if (m->m_flags & M_PKTHDR) {
787 		int plen = 0;
788 		for (m = *m0; m; m = m->m_next)
789 			plen += m->m_len;
790 		m = *m0;
791 		m->m_pkthdr.len = plen;
792 	}
793 
794 	ip = mtod(m, struct ip *);
795 	ip->ip_sum = pf_cksum_fixup(ip->ip_sum, ip->ip_len,
796 	    htons(hdrlen + total), 0);
797 	ip->ip_len = htons(hdrlen + total);
798 	ip->ip_sum = pf_cksum_fixup(ip->ip_sum, ip->ip_off,
799 	    ip->ip_off & ~(IP_MF|IP_OFFMASK), 0);
800 	ip->ip_off &= ~(IP_MF|IP_OFFMASK);
801 
802 	if (hdrlen + total > IP_MAXPACKET) {
803 		DPFPRINTF(("drop: too big: %d\n", total));
804 		ip->ip_len = 0;
805 		REASON_SET(reason, PFRES_SHORT);
806 		/* PF_DROP requires a valid mbuf *m0 in pf_test() */
807 		return (PF_DROP);
808 	}
809 
810 	DPFPRINTF(("complete: %p(%d)\n", m, ntohs(ip->ip_len)));
811 	return (PF_PASS);
812 }
813 #endif	/* INET */
814 
815 #ifdef INET6
816 static int
817 pf_reassemble6(struct mbuf **m0, struct ip6_hdr *ip6, struct ip6_frag *fraghdr,
818     uint16_t hdrlen, uint16_t extoff, u_short *reason)
819 {
820 	struct mbuf		*m = *m0;
821 	struct pf_frent		*frent;
822 	struct pf_fragment	*frag;
823 	struct pf_fragment_cmp	 key;
824 	struct m_tag		*mtag;
825 	struct pf_fragment_tag	*ftag;
826 	int			 off;
827 	uint32_t		 frag_id;
828 	uint16_t		 total, maxlen;
829 	uint8_t			 proto;
830 
831 	PF_FRAG_LOCK();
832 
833 	/* Get an entry for the fragment queue. */
834 	if ((frent = pf_create_fragment(reason)) == NULL) {
835 		PF_FRAG_UNLOCK();
836 		return (PF_DROP);
837 	}
838 
839 	frent->fe_m = m;
840 	frent->fe_hdrlen = hdrlen;
841 	frent->fe_extoff = extoff;
842 	frent->fe_len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - hdrlen;
843 	frent->fe_off = ntohs(fraghdr->ip6f_offlg & IP6F_OFF_MASK);
844 	frent->fe_mff = fraghdr->ip6f_offlg & IP6F_MORE_FRAG;
845 
846 	key.frc_src.v6 = ip6->ip6_src;
847 	key.frc_dst.v6 = ip6->ip6_dst;
848 	key.frc_af = AF_INET6;
849 	/* Only the first fragment's protocol is relevant. */
850 	key.frc_proto = 0;
851 	key.frc_id = fraghdr->ip6f_ident;
852 
853 	if ((frag = pf_fillup_fragment(&key, frent, reason)) == NULL) {
854 		PF_FRAG_UNLOCK();
855 		return (PF_DROP);
856 	}
857 
858 	/* The mbuf is part of the fragment entry, no direct free or access. */
859 	m = *m0 = NULL;
860 
861 	if (frag->fr_holes) {
862 		DPFPRINTF(("frag %d, holes %d\n", frag->fr_id,
863 		    frag->fr_holes));
864 		PF_FRAG_UNLOCK();
865 		return (PF_PASS);  /* Drop because *m0 is NULL, no error. */
866 	}
867 
868 	/* We have all the data. */
869 	frent = TAILQ_FIRST(&frag->fr_queue);
870 	KASSERT(frent != NULL, ("frent != NULL"));
871 	extoff = frent->fe_extoff;
872 	maxlen = frag->fr_maxlen;
873 	frag_id = frag->fr_id;
874 	total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
875 		TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
876 	hdrlen = frent->fe_hdrlen - sizeof(struct ip6_frag);
877 
878 	m = *m0 = pf_join_fragment(frag);
879 	frag = NULL;
880 
881 	PF_FRAG_UNLOCK();
882 
883 	/* Take protocol from first fragment header. */
884 	m = m_getptr(m, hdrlen + offsetof(struct ip6_frag, ip6f_nxt), &off);
885 	KASSERT(m, ("%s: short mbuf chain", __func__));
886 	proto = *(mtod(m, uint8_t *) + off);
887 	m = *m0;
888 
889 	/* Delete frag6 header */
890 	if (ip6_deletefraghdr(m, hdrlen, M_NOWAIT) != 0)
891 		goto fail;
892 
893 	if (m->m_flags & M_PKTHDR) {
894 		int plen = 0;
895 		for (m = *m0; m; m = m->m_next)
896 			plen += m->m_len;
897 		m = *m0;
898 		m->m_pkthdr.len = plen;
899 	}
900 
901 	if ((mtag = m_tag_get(PACKET_TAG_PF_REASSEMBLED,
902 	    sizeof(struct pf_fragment_tag), M_NOWAIT)) == NULL)
903 		goto fail;
904 	ftag = (struct pf_fragment_tag *)(mtag + 1);
905 	ftag->ft_hdrlen = hdrlen;
906 	ftag->ft_extoff = extoff;
907 	ftag->ft_maxlen = maxlen;
908 	ftag->ft_id = frag_id;
909 	m_tag_prepend(m, mtag);
910 
911 	ip6 = mtod(m, struct ip6_hdr *);
912 	ip6->ip6_plen = htons(hdrlen - sizeof(struct ip6_hdr) + total);
913 	if (extoff) {
914 		/* Write protocol into next field of last extension header. */
915 		m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt),
916 		    &off);
917 		KASSERT(m, ("%s: short mbuf chain", __func__));
918 		*(mtod(m, char *) + off) = proto;
919 		m = *m0;
920 	} else
921 		ip6->ip6_nxt = proto;
922 
923 	if (hdrlen - sizeof(struct ip6_hdr) + total > IPV6_MAXPACKET) {
924 		DPFPRINTF(("drop: too big: %d\n", total));
925 		ip6->ip6_plen = 0;
926 		REASON_SET(reason, PFRES_SHORT);
927 		/* PF_DROP requires a valid mbuf *m0 in pf_test6(). */
928 		return (PF_DROP);
929 	}
930 
931 	DPFPRINTF(("complete: %p(%d)\n", m, ntohs(ip6->ip6_plen)));
932 	return (PF_PASS);
933 
934 fail:
935 	REASON_SET(reason, PFRES_MEMORY);
936 	/* PF_DROP requires a valid mbuf *m0 in pf_test6(), will free later. */
937 	return (PF_DROP);
938 }
939 #endif	/* INET6 */
940 
941 #ifdef INET6
942 int
943 pf_refragment6(struct ifnet *ifp, struct mbuf **m0, struct m_tag *mtag,
944     bool forward)
945 {
946 	struct mbuf		*m = *m0, *t;
947 	struct ip6_hdr		*hdr;
948 	struct pf_fragment_tag	*ftag = (struct pf_fragment_tag *)(mtag + 1);
949 	struct pf_pdesc		 pd;
950 	uint32_t		 frag_id;
951 	uint16_t		 hdrlen, extoff, maxlen;
952 	uint8_t			 proto;
953 	int			 error, action;
954 
955 	hdrlen = ftag->ft_hdrlen;
956 	extoff = ftag->ft_extoff;
957 	maxlen = ftag->ft_maxlen;
958 	frag_id = ftag->ft_id;
959 	m_tag_delete(m, mtag);
960 	mtag = NULL;
961 	ftag = NULL;
962 
963 	if (extoff) {
964 		int off;
965 
966 		/* Use protocol from next field of last extension header */
967 		m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt),
968 		    &off);
969 		KASSERT((m != NULL), ("pf_refragment6: short mbuf chain"));
970 		proto = *(mtod(m, uint8_t *) + off);
971 		*(mtod(m, char *) + off) = IPPROTO_FRAGMENT;
972 		m = *m0;
973 	} else {
974 		hdr = mtod(m, struct ip6_hdr *);
975 		proto = hdr->ip6_nxt;
976 		hdr->ip6_nxt = IPPROTO_FRAGMENT;
977 	}
978 
979 	/* In case of link-local traffic we'll need a scope set. */
980 	hdr = mtod(m, struct ip6_hdr *);
981 
982 	in6_setscope(&hdr->ip6_src, ifp, NULL);
983 	in6_setscope(&hdr->ip6_dst, ifp, NULL);
984 
985 	/* The MTU must be a multiple of 8 bytes, or we risk doing the
986 	 * fragmentation wrong. */
987 	maxlen = maxlen & ~7;
988 
989 	/*
990 	 * Maxlen may be less than 8 if there was only a single
991 	 * fragment.  As it was fragmented before, add a fragment
992 	 * header also for a single fragment.  If total or maxlen
993 	 * is less than 8, ip6_fragment() will return EMSGSIZE and
994 	 * we drop the packet.
995 	 */
996 	error = ip6_fragment(ifp, m, hdrlen, proto, maxlen, frag_id);
997 	m = (*m0)->m_nextpkt;
998 	(*m0)->m_nextpkt = NULL;
999 	if (error == 0) {
1000 		/* The first mbuf contains the unfragmented packet. */
1001 		m_freem(*m0);
1002 		*m0 = NULL;
1003 		action = PF_PASS;
1004 	} else {
1005 		/* Drop expects an mbuf to free. */
1006 		DPFPRINTF(("refragment error %d\n", error));
1007 		action = PF_DROP;
1008 	}
1009 	for (; m; m = t) {
1010 		t = m->m_nextpkt;
1011 		m->m_nextpkt = NULL;
1012 		m->m_flags |= M_SKIP_FIREWALL;
1013 		memset(&pd, 0, sizeof(pd));
1014 		pd.pf_mtag = pf_find_mtag(m);
1015 		if (error == 0)
1016 			if (forward) {
1017 				MPASS(m->m_pkthdr.rcvif != NULL);
1018 				ip6_forward(m, 0);
1019 			} else {
1020 				(void)ip6_output(m, NULL, NULL, 0, NULL, NULL,
1021 				    NULL);
1022 			}
1023 		else
1024 			m_freem(m);
1025 	}
1026 
1027 	return (action);
1028 }
1029 #endif /* INET6 */
1030 
1031 #ifdef INET
1032 int
1033 pf_normalize_ip(struct mbuf **m0, struct pfi_kkif *kif, u_short *reason,
1034     struct pf_pdesc *pd)
1035 {
1036 	struct mbuf		*m = *m0;
1037 	struct pf_krule		*r;
1038 	struct ip		*h = mtod(m, struct ip *);
1039 	int			 mff = (ntohs(h->ip_off) & IP_MF);
1040 	int			 hlen = h->ip_hl << 2;
1041 	u_int16_t		 fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
1042 	u_int16_t		 max;
1043 	int			 ip_len;
1044 	int			 tag = -1;
1045 	int			 verdict;
1046 	bool			 scrub_compat;
1047 
1048 	PF_RULES_RASSERT();
1049 
1050 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1051 	/*
1052 	 * Check if there are any scrub rules, matching or not.
1053 	 * Lack of scrub rules means:
1054 	 *  - enforced packet normalization operation just like in OpenBSD
1055 	 *  - fragment reassembly depends on V_pf_status.reass
1056 	 * With scrub rules:
1057 	 *  - packet normalization is performed if there is a matching scrub rule
1058 	 *  - fragment reassembly is performed if the matching rule has no
1059 	 *    PFRULE_FRAGMENT_NOREASS flag
1060 	 */
1061 	scrub_compat = (r != NULL);
1062 	while (r != NULL) {
1063 		pf_counter_u64_add(&r->evaluations, 1);
1064 		if (pfi_kkif_match(r->kif, kif) == r->ifnot)
1065 			r = r->skip[PF_SKIP_IFP].ptr;
1066 		else if (r->direction && r->direction != pd->dir)
1067 			r = r->skip[PF_SKIP_DIR].ptr;
1068 		else if (r->af && r->af != AF_INET)
1069 			r = r->skip[PF_SKIP_AF].ptr;
1070 		else if (r->proto && r->proto != h->ip_p)
1071 			r = r->skip[PF_SKIP_PROTO].ptr;
1072 		else if (PF_MISMATCHAW(&r->src.addr,
1073 		    (struct pf_addr *)&h->ip_src.s_addr, AF_INET,
1074 		    r->src.neg, kif, M_GETFIB(m)))
1075 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
1076 		else if (PF_MISMATCHAW(&r->dst.addr,
1077 		    (struct pf_addr *)&h->ip_dst.s_addr, AF_INET,
1078 		    r->dst.neg, NULL, M_GETFIB(m)))
1079 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
1080 		else if (r->match_tag && !pf_match_tag(m, r, &tag,
1081 		    pd->pf_mtag ? pd->pf_mtag->tag : 0))
1082 			r = TAILQ_NEXT(r, entries);
1083 		else
1084 			break;
1085 	}
1086 
1087 	if (scrub_compat) {
1088 		/* With scrub rules present IPv4 normalization happens only
1089 		 * if one of rules has matched and it's not a "no scrub" rule */
1090 		if (r == NULL || r->action == PF_NOSCRUB)
1091 			return (PF_PASS);
1092 
1093 		pf_counter_u64_critical_enter();
1094 		pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1);
1095 		pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len);
1096 		pf_counter_u64_critical_exit();
1097 		pf_rule_to_actions(r, &pd->act);
1098 	}
1099 
1100 	/* Check for illegal packets */
1101 	if (hlen < (int)sizeof(struct ip)) {
1102 		REASON_SET(reason, PFRES_NORM);
1103 		goto drop;
1104 	}
1105 
1106 	if (hlen > ntohs(h->ip_len)) {
1107 		REASON_SET(reason, PFRES_NORM);
1108 		goto drop;
1109 	}
1110 
1111 	/* Clear IP_DF if the rule uses the no-df option or we're in no-df mode */
1112 	if (((!scrub_compat && V_pf_status.reass & PF_REASS_NODF) ||
1113 	    (r != NULL && r->rule_flag & PFRULE_NODF)) &&
1114 	    (h->ip_off & htons(IP_DF))
1115 	) {
1116 		u_int16_t ip_off = h->ip_off;
1117 
1118 		h->ip_off &= htons(~IP_DF);
1119 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
1120 	}
1121 
1122 	/* We will need other tests here */
1123 	if (!fragoff && !mff)
1124 		goto no_fragment;
1125 
1126 	/* We're dealing with a fragment now. Don't allow fragments
1127 	 * with IP_DF to enter the cache. If the flag was cleared by
1128 	 * no-df above, fine. Otherwise drop it.
1129 	 */
1130 	if (h->ip_off & htons(IP_DF)) {
1131 		DPFPRINTF(("IP_DF\n"));
1132 		goto bad;
1133 	}
1134 
1135 	ip_len = ntohs(h->ip_len) - hlen;
1136 
1137 	/* All fragments are 8 byte aligned */
1138 	if (mff && (ip_len & 0x7)) {
1139 		DPFPRINTF(("mff and %d\n", ip_len));
1140 		goto bad;
1141 	}
1142 
1143 	/* Respect maximum length */
1144 	if (fragoff + ip_len > IP_MAXPACKET) {
1145 		DPFPRINTF(("max packet %d\n", fragoff + ip_len));
1146 		goto bad;
1147 	}
1148 
1149 	if ((!scrub_compat && V_pf_status.reass) ||
1150 	    (r != NULL && !(r->rule_flag & PFRULE_FRAGMENT_NOREASS))
1151 	) {
1152 		max = fragoff + ip_len;
1153 
1154 		/* Fully buffer all of the fragments
1155 		 * Might return a completely reassembled mbuf, or NULL */
1156 		PF_FRAG_LOCK();
1157 		DPFPRINTF(("reass frag %d @ %d-%d\n", h->ip_id, fragoff, max));
1158 		verdict = pf_reassemble(m0, h, pd->dir, reason);
1159 		PF_FRAG_UNLOCK();
1160 
1161 		if (verdict != PF_PASS)
1162 			return (PF_DROP);
1163 
1164 		m = *m0;
1165 		if (m == NULL)
1166 			return (PF_DROP);
1167 
1168 		h = mtod(m, struct ip *);
1169 
1170  no_fragment:
1171 		/* At this point, only IP_DF is allowed in ip_off */
1172 		if (h->ip_off & ~htons(IP_DF)) {
1173 			u_int16_t ip_off = h->ip_off;
1174 
1175 			h->ip_off &= htons(IP_DF);
1176 			h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
1177 		}
1178 	}
1179 
1180 	return (PF_PASS);
1181 
1182  bad:
1183 	DPFPRINTF(("dropping bad fragment\n"));
1184 	REASON_SET(reason, PFRES_FRAG);
1185  drop:
1186 	if (r != NULL && r->log)
1187 		PFLOG_PACKET(kif, m, AF_INET, PF_DROP, *reason, r, NULL, NULL, pd, 1);
1188 
1189 	return (PF_DROP);
1190 }
1191 #endif
1192 
1193 #ifdef INET6
1194 int
1195 pf_normalize_ip6(struct mbuf **m0, struct pfi_kkif *kif,
1196     u_short *reason, struct pf_pdesc *pd)
1197 {
1198 	struct mbuf		*m = *m0;
1199 	struct pf_krule		*r;
1200 	struct ip6_hdr		*h = mtod(m, struct ip6_hdr *);
1201 	int			 extoff;
1202 	int			 off;
1203 	struct ip6_ext		 ext;
1204 	struct ip6_opt		 opt;
1205 	struct ip6_frag		 frag;
1206 	u_int32_t		 plen;
1207 	int			 optend;
1208 	int			 ooff;
1209 	u_int8_t		 proto;
1210 	int			 terminal;
1211 	bool			 scrub_compat;
1212 
1213 	PF_RULES_RASSERT();
1214 
1215 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1216 	/*
1217 	 * Check if there are any scrub rules, matching or not.
1218 	 * Lack of scrub rules means:
1219 	 *  - enforced packet normalization operation just like in OpenBSD
1220 	 * With scrub rules:
1221 	 *  - packet normalization is performed if there is a matching scrub rule
1222 	 * XXX: Fragment reassembly always performed for IPv6!
1223 	 */
1224 	scrub_compat = (r != NULL);
1225 	while (r != NULL) {
1226 		pf_counter_u64_add(&r->evaluations, 1);
1227 		if (pfi_kkif_match(r->kif, kif) == r->ifnot)
1228 			r = r->skip[PF_SKIP_IFP].ptr;
1229 		else if (r->direction && r->direction != pd->dir)
1230 			r = r->skip[PF_SKIP_DIR].ptr;
1231 		else if (r->af && r->af != AF_INET6)
1232 			r = r->skip[PF_SKIP_AF].ptr;
1233 #if 0 /* header chain! */
1234 		else if (r->proto && r->proto != h->ip6_nxt)
1235 			r = r->skip[PF_SKIP_PROTO].ptr;
1236 #endif
1237 		else if (PF_MISMATCHAW(&r->src.addr,
1238 		    (struct pf_addr *)&h->ip6_src, AF_INET6,
1239 		    r->src.neg, kif, M_GETFIB(m)))
1240 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
1241 		else if (PF_MISMATCHAW(&r->dst.addr,
1242 		    (struct pf_addr *)&h->ip6_dst, AF_INET6,
1243 		    r->dst.neg, NULL, M_GETFIB(m)))
1244 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
1245 		else
1246 			break;
1247 	}
1248 
1249 	if (scrub_compat) {
1250 		/* With scrub rules present IPv6 normalization happens only
1251 		 * if one of rules has matched and it's not a "no scrub" rule */
1252 		if (r == NULL || r->action == PF_NOSCRUB)
1253 			return (PF_PASS);
1254 
1255 		pf_counter_u64_critical_enter();
1256 		pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1);
1257 		pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len);
1258 		pf_counter_u64_critical_exit();
1259 		pf_rule_to_actions(r, &pd->act);
1260 	}
1261 
1262 	/* Check for illegal packets */
1263 	if (sizeof(struct ip6_hdr) + IPV6_MAXPACKET < m->m_pkthdr.len)
1264 		goto drop;
1265 
1266 again:
1267 	h = mtod(m, struct ip6_hdr *);
1268 	plen = ntohs(h->ip6_plen);
1269 	/* jumbo payload option not supported */
1270 	if (plen == 0)
1271 		goto drop;
1272 
1273 	extoff = 0;
1274 	off = sizeof(struct ip6_hdr);
1275 	proto = h->ip6_nxt;
1276 	terminal = 0;
1277 	do {
1278 		switch (proto) {
1279 		case IPPROTO_FRAGMENT:
1280 			goto fragment;
1281 			break;
1282 		case IPPROTO_AH:
1283 		case IPPROTO_ROUTING:
1284 		case IPPROTO_DSTOPTS:
1285 			if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL,
1286 			    NULL, AF_INET6))
1287 				goto shortpkt;
1288 			extoff = off;
1289 			if (proto == IPPROTO_AH)
1290 				off += (ext.ip6e_len + 2) * 4;
1291 			else
1292 				off += (ext.ip6e_len + 1) * 8;
1293 			proto = ext.ip6e_nxt;
1294 			break;
1295 		case IPPROTO_HOPOPTS:
1296 			if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL,
1297 			    NULL, AF_INET6))
1298 				goto shortpkt;
1299 			extoff = off;
1300 			optend = off + (ext.ip6e_len + 1) * 8;
1301 			ooff = off + sizeof(ext);
1302 			do {
1303 				if (!pf_pull_hdr(m, ooff, &opt.ip6o_type,
1304 				    sizeof(opt.ip6o_type), NULL, NULL,
1305 				    AF_INET6))
1306 					goto shortpkt;
1307 				if (opt.ip6o_type == IP6OPT_PAD1) {
1308 					ooff++;
1309 					continue;
1310 				}
1311 				if (!pf_pull_hdr(m, ooff, &opt, sizeof(opt),
1312 				    NULL, NULL, AF_INET6))
1313 					goto shortpkt;
1314 				if (ooff + sizeof(opt) + opt.ip6o_len > optend)
1315 					goto drop;
1316 				if (opt.ip6o_type == IP6OPT_JUMBO)
1317 					goto drop;
1318 				ooff += sizeof(opt) + opt.ip6o_len;
1319 			} while (ooff < optend);
1320 
1321 			off = optend;
1322 			proto = ext.ip6e_nxt;
1323 			break;
1324 		default:
1325 			terminal = 1;
1326 			break;
1327 		}
1328 	} while (!terminal);
1329 
1330 	if (sizeof(struct ip6_hdr) + plen > m->m_pkthdr.len)
1331 		goto shortpkt;
1332 
1333 	return (PF_PASS);
1334 
1335  fragment:
1336 	if (pd->flags & PFDESC_IP_REAS)
1337 		return (PF_DROP);
1338 	if (sizeof(struct ip6_hdr) + plen > m->m_pkthdr.len)
1339 		goto shortpkt;
1340 
1341 	if (!pf_pull_hdr(m, off, &frag, sizeof(frag), NULL, NULL, AF_INET6))
1342 		goto shortpkt;
1343 
1344 	/* Offset now points to data portion. */
1345 	off += sizeof(frag);
1346 
1347 	/* Returns PF_DROP or *m0 is NULL or completely reassembled mbuf. */
1348 	if (pf_reassemble6(m0, h, &frag, off, extoff, reason) != PF_PASS)
1349 		return (PF_DROP);
1350 	m = *m0;
1351 	if (m == NULL)
1352 		return (PF_DROP);
1353 
1354 	pd->flags |= PFDESC_IP_REAS;
1355 	goto again;
1356 
1357  shortpkt:
1358 	REASON_SET(reason, PFRES_SHORT);
1359 	if (r != NULL && r->log)
1360 		PFLOG_PACKET(kif, m, AF_INET6, PF_DROP, *reason, r, NULL, NULL, pd, 1);
1361 	return (PF_DROP);
1362 
1363  drop:
1364 	REASON_SET(reason, PFRES_NORM);
1365 	if (r != NULL && r->log)
1366 		PFLOG_PACKET(kif, m, AF_INET6, PF_DROP, *reason, r, NULL, NULL, pd, 1);
1367 	return (PF_DROP);
1368 }
1369 #endif /* INET6 */
1370 
1371 int
1372 pf_normalize_tcp(struct pfi_kkif *kif, struct mbuf *m, int ipoff,
1373     int off, void *h, struct pf_pdesc *pd)
1374 {
1375 	struct pf_krule	*r, *rm = NULL;
1376 	struct tcphdr	*th = &pd->hdr.tcp;
1377 	int		 rewrite = 0;
1378 	u_short		 reason;
1379 	u_int16_t	 flags;
1380 	sa_family_t	 af = pd->af;
1381 	int		 srs;
1382 
1383 	PF_RULES_RASSERT();
1384 
1385 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1386 	/* Check if there any scrub rules. Lack of scrub rules means enforced
1387 	 * packet normalization operation just like in OpenBSD. */
1388 	srs = (r != NULL);
1389 	while (r != NULL) {
1390 		pf_counter_u64_add(&r->evaluations, 1);
1391 		if (pfi_kkif_match(r->kif, kif) == r->ifnot)
1392 			r = r->skip[PF_SKIP_IFP].ptr;
1393 		else if (r->direction && r->direction != pd->dir)
1394 			r = r->skip[PF_SKIP_DIR].ptr;
1395 		else if (r->af && r->af != af)
1396 			r = r->skip[PF_SKIP_AF].ptr;
1397 		else if (r->proto && r->proto != pd->proto)
1398 			r = r->skip[PF_SKIP_PROTO].ptr;
1399 		else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
1400 		    r->src.neg, kif, M_GETFIB(m)))
1401 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
1402 		else if (r->src.port_op && !pf_match_port(r->src.port_op,
1403 			    r->src.port[0], r->src.port[1], th->th_sport))
1404 			r = r->skip[PF_SKIP_SRC_PORT].ptr;
1405 		else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
1406 		    r->dst.neg, NULL, M_GETFIB(m)))
1407 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
1408 		else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
1409 			    r->dst.port[0], r->dst.port[1], th->th_dport))
1410 			r = r->skip[PF_SKIP_DST_PORT].ptr;
1411 		else if (r->os_fingerprint != PF_OSFP_ANY && !pf_osfp_match(
1412 			    pf_osfp_fingerprint(pd, m, off, th),
1413 			    r->os_fingerprint))
1414 			r = TAILQ_NEXT(r, entries);
1415 		else {
1416 			rm = r;
1417 			break;
1418 		}
1419 	}
1420 
1421 	if (srs) {
1422 		/* With scrub rules present TCP normalization happens only
1423 		 * if one of rules has matched and it's not a "no scrub" rule */
1424 		if (rm == NULL || rm->action == PF_NOSCRUB)
1425 			return (PF_PASS);
1426 
1427 		pf_counter_u64_critical_enter();
1428 		pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1);
1429 		pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len);
1430 		pf_counter_u64_critical_exit();
1431 		pf_rule_to_actions(rm, &pd->act);
1432 	}
1433 
1434 	if (rm && rm->rule_flag & PFRULE_REASSEMBLE_TCP)
1435 		pd->flags |= PFDESC_TCP_NORM;
1436 
1437 	flags = tcp_get_flags(th);
1438 	if (flags & TH_SYN) {
1439 		/* Illegal packet */
1440 		if (flags & TH_RST)
1441 			goto tcp_drop;
1442 
1443 		if (flags & TH_FIN)
1444 			goto tcp_drop;
1445 	} else {
1446 		/* Illegal packet */
1447 		if (!(flags & (TH_ACK|TH_RST)))
1448 			goto tcp_drop;
1449 	}
1450 
1451 	if (!(flags & TH_ACK)) {
1452 		/* These flags are only valid if ACK is set */
1453 		if ((flags & TH_FIN) || (flags & TH_PUSH) || (flags & TH_URG))
1454 			goto tcp_drop;
1455 	}
1456 
1457 	/* Check for illegal header length */
1458 	if (th->th_off < (sizeof(struct tcphdr) >> 2))
1459 		goto tcp_drop;
1460 
1461 	/* If flags changed, or reserved data set, then adjust */
1462 	if (flags != tcp_get_flags(th) ||
1463 	    (tcp_get_flags(th) & (TH_RES1|TH_RES2|TH_RES2)) != 0) {
1464 		u_int16_t	ov, nv;
1465 
1466 		ov = *(u_int16_t *)(&th->th_ack + 1);
1467 		flags &= ~(TH_RES1 | TH_RES2 | TH_RES3);
1468 		tcp_set_flags(th, flags);
1469 		nv = *(u_int16_t *)(&th->th_ack + 1);
1470 
1471 		th->th_sum = pf_proto_cksum_fixup(m, th->th_sum, ov, nv, 0);
1472 		rewrite = 1;
1473 	}
1474 
1475 	/* Remove urgent pointer, if TH_URG is not set */
1476 	if (!(flags & TH_URG) && th->th_urp) {
1477 		th->th_sum = pf_proto_cksum_fixup(m, th->th_sum, th->th_urp,
1478 		    0, 0);
1479 		th->th_urp = 0;
1480 		rewrite = 1;
1481 	}
1482 
1483 	/* copy back packet headers if we sanitized */
1484 	if (rewrite)
1485 		m_copyback(m, off, sizeof(*th), (caddr_t)th);
1486 
1487 	return (PF_PASS);
1488 
1489  tcp_drop:
1490 	REASON_SET(&reason, PFRES_NORM);
1491 	if (rm != NULL && r->log)
1492 		PFLOG_PACKET(kif, m, AF_INET, PF_DROP, reason, r, NULL, NULL, pd, 1);
1493 	return (PF_DROP);
1494 }
1495 
1496 int
1497 pf_normalize_tcp_init(struct mbuf *m, int off, struct pf_pdesc *pd,
1498     struct tcphdr *th, struct pf_state_peer *src, struct pf_state_peer *dst)
1499 {
1500 	u_int32_t tsval, tsecr;
1501 	u_int8_t hdr[60];
1502 	u_int8_t *opt;
1503 
1504 	KASSERT((src->scrub == NULL),
1505 	    ("pf_normalize_tcp_init: src->scrub != NULL"));
1506 
1507 	src->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT);
1508 	if (src->scrub == NULL)
1509 		return (1);
1510 
1511 	switch (pd->af) {
1512 #ifdef INET
1513 	case AF_INET: {
1514 		struct ip *h = mtod(m, struct ip *);
1515 		src->scrub->pfss_ttl = h->ip_ttl;
1516 		break;
1517 	}
1518 #endif /* INET */
1519 #ifdef INET6
1520 	case AF_INET6: {
1521 		struct ip6_hdr *h = mtod(m, struct ip6_hdr *);
1522 		src->scrub->pfss_ttl = h->ip6_hlim;
1523 		break;
1524 	}
1525 #endif /* INET6 */
1526 	}
1527 
1528 	/*
1529 	 * All normalizations below are only begun if we see the start of
1530 	 * the connections.  They must all set an enabled bit in pfss_flags
1531 	 */
1532 	if ((th->th_flags & TH_SYN) == 0)
1533 		return (0);
1534 
1535 	if (th->th_off > (sizeof(struct tcphdr) >> 2) && src->scrub &&
1536 	    pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) {
1537 		/* Diddle with TCP options */
1538 		int hlen;
1539 		opt = hdr + sizeof(struct tcphdr);
1540 		hlen = (th->th_off << 2) - sizeof(struct tcphdr);
1541 		while (hlen >= TCPOLEN_TIMESTAMP) {
1542 			switch (*opt) {
1543 			case TCPOPT_EOL:	/* FALLTHROUGH */
1544 			case TCPOPT_NOP:
1545 				opt++;
1546 				hlen--;
1547 				break;
1548 			case TCPOPT_TIMESTAMP:
1549 				if (opt[1] >= TCPOLEN_TIMESTAMP) {
1550 					src->scrub->pfss_flags |=
1551 					    PFSS_TIMESTAMP;
1552 					src->scrub->pfss_ts_mod =
1553 					    htonl(arc4random());
1554 
1555 					/* note PFSS_PAWS not set yet */
1556 					memcpy(&tsval, &opt[2],
1557 					    sizeof(u_int32_t));
1558 					memcpy(&tsecr, &opt[6],
1559 					    sizeof(u_int32_t));
1560 					src->scrub->pfss_tsval0 = ntohl(tsval);
1561 					src->scrub->pfss_tsval = ntohl(tsval);
1562 					src->scrub->pfss_tsecr = ntohl(tsecr);
1563 					getmicrouptime(&src->scrub->pfss_last);
1564 				}
1565 				/* FALLTHROUGH */
1566 			default:
1567 				hlen -= MAX(opt[1], 2);
1568 				opt += MAX(opt[1], 2);
1569 				break;
1570 			}
1571 		}
1572 	}
1573 
1574 	return (0);
1575 }
1576 
1577 void
1578 pf_normalize_tcp_cleanup(struct pf_kstate *state)
1579 {
1580 	/* XXX Note: this also cleans up SCTP. */
1581 	uma_zfree(V_pf_state_scrub_z, state->src.scrub);
1582 	uma_zfree(V_pf_state_scrub_z, state->dst.scrub);
1583 
1584 	/* Someday... flush the TCP segment reassembly descriptors. */
1585 }
1586 int
1587 pf_normalize_sctp_init(struct mbuf *m, int off, struct pf_pdesc *pd,
1588 	    struct pf_state_peer *src, struct pf_state_peer *dst)
1589 {
1590 	src->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT);
1591 	if (src->scrub == NULL)
1592 		return (1);
1593 
1594 	dst->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT);
1595 	if (dst->scrub == NULL) {
1596 		uma_zfree(V_pf_state_scrub_z, src);
1597 		return (1);
1598 	}
1599 
1600 	dst->scrub->pfss_v_tag = pd->sctp_initiate_tag;
1601 
1602 	return (0);
1603 }
1604 
1605 int
1606 pf_normalize_tcp_stateful(struct mbuf *m, int off, struct pf_pdesc *pd,
1607     u_short *reason, struct tcphdr *th, struct pf_kstate *state,
1608     struct pf_state_peer *src, struct pf_state_peer *dst, int *writeback)
1609 {
1610 	struct timeval uptime;
1611 	u_int32_t tsval, tsecr;
1612 	u_int tsval_from_last;
1613 	u_int8_t hdr[60];
1614 	u_int8_t *opt;
1615 	int copyback = 0;
1616 	int got_ts = 0;
1617 	size_t startoff;
1618 
1619 	KASSERT((src->scrub || dst->scrub),
1620 	    ("%s: src->scrub && dst->scrub!", __func__));
1621 
1622 	/*
1623 	 * Enforce the minimum TTL seen for this connection.  Negate a common
1624 	 * technique to evade an intrusion detection system and confuse
1625 	 * firewall state code.
1626 	 */
1627 	switch (pd->af) {
1628 #ifdef INET
1629 	case AF_INET: {
1630 		if (src->scrub) {
1631 			struct ip *h = mtod(m, struct ip *);
1632 			if (h->ip_ttl > src->scrub->pfss_ttl)
1633 				src->scrub->pfss_ttl = h->ip_ttl;
1634 			h->ip_ttl = src->scrub->pfss_ttl;
1635 		}
1636 		break;
1637 	}
1638 #endif /* INET */
1639 #ifdef INET6
1640 	case AF_INET6: {
1641 		if (src->scrub) {
1642 			struct ip6_hdr *h = mtod(m, struct ip6_hdr *);
1643 			if (h->ip6_hlim > src->scrub->pfss_ttl)
1644 				src->scrub->pfss_ttl = h->ip6_hlim;
1645 			h->ip6_hlim = src->scrub->pfss_ttl;
1646 		}
1647 		break;
1648 	}
1649 #endif /* INET6 */
1650 	}
1651 
1652 	if (th->th_off > (sizeof(struct tcphdr) >> 2) &&
1653 	    ((src->scrub && (src->scrub->pfss_flags & PFSS_TIMESTAMP)) ||
1654 	    (dst->scrub && (dst->scrub->pfss_flags & PFSS_TIMESTAMP))) &&
1655 	    pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) {
1656 		/* Diddle with TCP options */
1657 		int hlen;
1658 		opt = hdr + sizeof(struct tcphdr);
1659 		hlen = (th->th_off << 2) - sizeof(struct tcphdr);
1660 		while (hlen >= TCPOLEN_TIMESTAMP) {
1661 			startoff = opt - (hdr + sizeof(struct tcphdr));
1662 			switch (*opt) {
1663 			case TCPOPT_EOL:	/* FALLTHROUGH */
1664 			case TCPOPT_NOP:
1665 				opt++;
1666 				hlen--;
1667 				break;
1668 			case TCPOPT_TIMESTAMP:
1669 				/* Modulate the timestamps.  Can be used for
1670 				 * NAT detection, OS uptime determination or
1671 				 * reboot detection.
1672 				 */
1673 
1674 				if (got_ts) {
1675 					/* Huh?  Multiple timestamps!? */
1676 					if (V_pf_status.debug >= PF_DEBUG_MISC) {
1677 						DPFPRINTF(("multiple TS??\n"));
1678 						pf_print_state(state);
1679 						printf("\n");
1680 					}
1681 					REASON_SET(reason, PFRES_TS);
1682 					return (PF_DROP);
1683 				}
1684 				if (opt[1] >= TCPOLEN_TIMESTAMP) {
1685 					memcpy(&tsval, &opt[2],
1686 					    sizeof(u_int32_t));
1687 					if (tsval && src->scrub &&
1688 					    (src->scrub->pfss_flags &
1689 					    PFSS_TIMESTAMP)) {
1690 						tsval = ntohl(tsval);
1691 						pf_patch_32_unaligned(m,
1692 						    &th->th_sum,
1693 						    &opt[2],
1694 						    htonl(tsval +
1695 						    src->scrub->pfss_ts_mod),
1696 						    PF_ALGNMNT(startoff),
1697 						    0);
1698 						copyback = 1;
1699 					}
1700 
1701 					/* Modulate TS reply iff valid (!0) */
1702 					memcpy(&tsecr, &opt[6],
1703 					    sizeof(u_int32_t));
1704 					if (tsecr && dst->scrub &&
1705 					    (dst->scrub->pfss_flags &
1706 					    PFSS_TIMESTAMP)) {
1707 						tsecr = ntohl(tsecr)
1708 						    - dst->scrub->pfss_ts_mod;
1709 						pf_patch_32_unaligned(m,
1710 						    &th->th_sum,
1711 						    &opt[6],
1712 						    htonl(tsecr),
1713 						    PF_ALGNMNT(startoff),
1714 						    0);
1715 						copyback = 1;
1716 					}
1717 					got_ts = 1;
1718 				}
1719 				/* FALLTHROUGH */
1720 			default:
1721 				hlen -= MAX(opt[1], 2);
1722 				opt += MAX(opt[1], 2);
1723 				break;
1724 			}
1725 		}
1726 		if (copyback) {
1727 			/* Copyback the options, caller copys back header */
1728 			*writeback = 1;
1729 			m_copyback(m, off + sizeof(struct tcphdr),
1730 			    (th->th_off << 2) - sizeof(struct tcphdr), hdr +
1731 			    sizeof(struct tcphdr));
1732 		}
1733 	}
1734 
1735 	/*
1736 	 * Must invalidate PAWS checks on connections idle for too long.
1737 	 * The fastest allowed timestamp clock is 1ms.  That turns out to
1738 	 * be about 24 days before it wraps.  XXX Right now our lowerbound
1739 	 * TS echo check only works for the first 12 days of a connection
1740 	 * when the TS has exhausted half its 32bit space
1741 	 */
1742 #define TS_MAX_IDLE	(24*24*60*60)
1743 #define TS_MAX_CONN	(12*24*60*60)	/* XXX remove when better tsecr check */
1744 
1745 	getmicrouptime(&uptime);
1746 	if (src->scrub && (src->scrub->pfss_flags & PFSS_PAWS) &&
1747 	    (uptime.tv_sec - src->scrub->pfss_last.tv_sec > TS_MAX_IDLE ||
1748 	    time_uptime - state->creation > TS_MAX_CONN))  {
1749 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
1750 			DPFPRINTF(("src idled out of PAWS\n"));
1751 			pf_print_state(state);
1752 			printf("\n");
1753 		}
1754 		src->scrub->pfss_flags = (src->scrub->pfss_flags & ~PFSS_PAWS)
1755 		    | PFSS_PAWS_IDLED;
1756 	}
1757 	if (dst->scrub && (dst->scrub->pfss_flags & PFSS_PAWS) &&
1758 	    uptime.tv_sec - dst->scrub->pfss_last.tv_sec > TS_MAX_IDLE) {
1759 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
1760 			DPFPRINTF(("dst idled out of PAWS\n"));
1761 			pf_print_state(state);
1762 			printf("\n");
1763 		}
1764 		dst->scrub->pfss_flags = (dst->scrub->pfss_flags & ~PFSS_PAWS)
1765 		    | PFSS_PAWS_IDLED;
1766 	}
1767 
1768 	if (got_ts && src->scrub && dst->scrub &&
1769 	    (src->scrub->pfss_flags & PFSS_PAWS) &&
1770 	    (dst->scrub->pfss_flags & PFSS_PAWS)) {
1771 		/* Validate that the timestamps are "in-window".
1772 		 * RFC1323 describes TCP Timestamp options that allow
1773 		 * measurement of RTT (round trip time) and PAWS
1774 		 * (protection against wrapped sequence numbers).  PAWS
1775 		 * gives us a set of rules for rejecting packets on
1776 		 * long fat pipes (packets that were somehow delayed
1777 		 * in transit longer than the time it took to send the
1778 		 * full TCP sequence space of 4Gb).  We can use these
1779 		 * rules and infer a few others that will let us treat
1780 		 * the 32bit timestamp and the 32bit echoed timestamp
1781 		 * as sequence numbers to prevent a blind attacker from
1782 		 * inserting packets into a connection.
1783 		 *
1784 		 * RFC1323 tells us:
1785 		 *  - The timestamp on this packet must be greater than
1786 		 *    or equal to the last value echoed by the other
1787 		 *    endpoint.  The RFC says those will be discarded
1788 		 *    since it is a dup that has already been acked.
1789 		 *    This gives us a lowerbound on the timestamp.
1790 		 *        timestamp >= other last echoed timestamp
1791 		 *  - The timestamp will be less than or equal to
1792 		 *    the last timestamp plus the time between the
1793 		 *    last packet and now.  The RFC defines the max
1794 		 *    clock rate as 1ms.  We will allow clocks to be
1795 		 *    up to 10% fast and will allow a total difference
1796 		 *    or 30 seconds due to a route change.  And this
1797 		 *    gives us an upperbound on the timestamp.
1798 		 *        timestamp <= last timestamp + max ticks
1799 		 *    We have to be careful here.  Windows will send an
1800 		 *    initial timestamp of zero and then initialize it
1801 		 *    to a random value after the 3whs; presumably to
1802 		 *    avoid a DoS by having to call an expensive RNG
1803 		 *    during a SYN flood.  Proof MS has at least one
1804 		 *    good security geek.
1805 		 *
1806 		 *  - The TCP timestamp option must also echo the other
1807 		 *    endpoints timestamp.  The timestamp echoed is the
1808 		 *    one carried on the earliest unacknowledged segment
1809 		 *    on the left edge of the sequence window.  The RFC
1810 		 *    states that the host will reject any echoed
1811 		 *    timestamps that were larger than any ever sent.
1812 		 *    This gives us an upperbound on the TS echo.
1813 		 *        tescr <= largest_tsval
1814 		 *  - The lowerbound on the TS echo is a little more
1815 		 *    tricky to determine.  The other endpoint's echoed
1816 		 *    values will not decrease.  But there may be
1817 		 *    network conditions that re-order packets and
1818 		 *    cause our view of them to decrease.  For now the
1819 		 *    only lowerbound we can safely determine is that
1820 		 *    the TS echo will never be less than the original
1821 		 *    TS.  XXX There is probably a better lowerbound.
1822 		 *    Remove TS_MAX_CONN with better lowerbound check.
1823 		 *        tescr >= other original TS
1824 		 *
1825 		 * It is also important to note that the fastest
1826 		 * timestamp clock of 1ms will wrap its 32bit space in
1827 		 * 24 days.  So we just disable TS checking after 24
1828 		 * days of idle time.  We actually must use a 12d
1829 		 * connection limit until we can come up with a better
1830 		 * lowerbound to the TS echo check.
1831 		 */
1832 		struct timeval delta_ts;
1833 		int ts_fudge;
1834 
1835 		/*
1836 		 * PFTM_TS_DIFF is how many seconds of leeway to allow
1837 		 * a host's timestamp.  This can happen if the previous
1838 		 * packet got delayed in transit for much longer than
1839 		 * this packet.
1840 		 */
1841 		if ((ts_fudge = state->rule.ptr->timeout[PFTM_TS_DIFF]) == 0)
1842 			ts_fudge = V_pf_default_rule.timeout[PFTM_TS_DIFF];
1843 
1844 		/* Calculate max ticks since the last timestamp */
1845 #define TS_MAXFREQ	1100		/* RFC max TS freq of 1Khz + 10% skew */
1846 #define TS_MICROSECS	1000000		/* microseconds per second */
1847 		delta_ts = uptime;
1848 		timevalsub(&delta_ts, &src->scrub->pfss_last);
1849 		tsval_from_last = (delta_ts.tv_sec + ts_fudge) * TS_MAXFREQ;
1850 		tsval_from_last += delta_ts.tv_usec / (TS_MICROSECS/TS_MAXFREQ);
1851 
1852 		if ((src->state >= TCPS_ESTABLISHED &&
1853 		    dst->state >= TCPS_ESTABLISHED) &&
1854 		    (SEQ_LT(tsval, dst->scrub->pfss_tsecr) ||
1855 		    SEQ_GT(tsval, src->scrub->pfss_tsval + tsval_from_last) ||
1856 		    (tsecr && (SEQ_GT(tsecr, dst->scrub->pfss_tsval) ||
1857 		    SEQ_LT(tsecr, dst->scrub->pfss_tsval0))))) {
1858 			/* Bad RFC1323 implementation or an insertion attack.
1859 			 *
1860 			 * - Solaris 2.6 and 2.7 are known to send another ACK
1861 			 *   after the FIN,FIN|ACK,ACK closing that carries
1862 			 *   an old timestamp.
1863 			 */
1864 
1865 			DPFPRINTF(("Timestamp failed %c%c%c%c\n",
1866 			    SEQ_LT(tsval, dst->scrub->pfss_tsecr) ? '0' : ' ',
1867 			    SEQ_GT(tsval, src->scrub->pfss_tsval +
1868 			    tsval_from_last) ? '1' : ' ',
1869 			    SEQ_GT(tsecr, dst->scrub->pfss_tsval) ? '2' : ' ',
1870 			    SEQ_LT(tsecr, dst->scrub->pfss_tsval0)? '3' : ' '));
1871 			DPFPRINTF((" tsval: %u  tsecr: %u  +ticks: %u  "
1872 			    "idle: %jus %lums\n",
1873 			    tsval, tsecr, tsval_from_last,
1874 			    (uintmax_t)delta_ts.tv_sec,
1875 			    delta_ts.tv_usec / 1000));
1876 			DPFPRINTF((" src->tsval: %u  tsecr: %u\n",
1877 			    src->scrub->pfss_tsval, src->scrub->pfss_tsecr));
1878 			DPFPRINTF((" dst->tsval: %u  tsecr: %u  tsval0: %u"
1879 			    "\n", dst->scrub->pfss_tsval,
1880 			    dst->scrub->pfss_tsecr, dst->scrub->pfss_tsval0));
1881 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
1882 				pf_print_state(state);
1883 				pf_print_flags(th->th_flags);
1884 				printf("\n");
1885 			}
1886 			REASON_SET(reason, PFRES_TS);
1887 			return (PF_DROP);
1888 		}
1889 
1890 		/* XXX I'd really like to require tsecr but it's optional */
1891 
1892 	} else if (!got_ts && (th->th_flags & TH_RST) == 0 &&
1893 	    ((src->state == TCPS_ESTABLISHED && dst->state == TCPS_ESTABLISHED)
1894 	    || pd->p_len > 0 || (th->th_flags & TH_SYN)) &&
1895 	    src->scrub && dst->scrub &&
1896 	    (src->scrub->pfss_flags & PFSS_PAWS) &&
1897 	    (dst->scrub->pfss_flags & PFSS_PAWS)) {
1898 		/* Didn't send a timestamp.  Timestamps aren't really useful
1899 		 * when:
1900 		 *  - connection opening or closing (often not even sent).
1901 		 *    but we must not let an attacker to put a FIN on a
1902 		 *    data packet to sneak it through our ESTABLISHED check.
1903 		 *  - on a TCP reset.  RFC suggests not even looking at TS.
1904 		 *  - on an empty ACK.  The TS will not be echoed so it will
1905 		 *    probably not help keep the RTT calculation in sync and
1906 		 *    there isn't as much danger when the sequence numbers
1907 		 *    got wrapped.  So some stacks don't include TS on empty
1908 		 *    ACKs :-(
1909 		 *
1910 		 * To minimize the disruption to mostly RFC1323 conformant
1911 		 * stacks, we will only require timestamps on data packets.
1912 		 *
1913 		 * And what do ya know, we cannot require timestamps on data
1914 		 * packets.  There appear to be devices that do legitimate
1915 		 * TCP connection hijacking.  There are HTTP devices that allow
1916 		 * a 3whs (with timestamps) and then buffer the HTTP request.
1917 		 * If the intermediate device has the HTTP response cache, it
1918 		 * will spoof the response but not bother timestamping its
1919 		 * packets.  So we can look for the presence of a timestamp in
1920 		 * the first data packet and if there, require it in all future
1921 		 * packets.
1922 		 */
1923 
1924 		if (pd->p_len > 0 && (src->scrub->pfss_flags & PFSS_DATA_TS)) {
1925 			/*
1926 			 * Hey!  Someone tried to sneak a packet in.  Or the
1927 			 * stack changed its RFC1323 behavior?!?!
1928 			 */
1929 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
1930 				DPFPRINTF(("Did not receive expected RFC1323 "
1931 				    "timestamp\n"));
1932 				pf_print_state(state);
1933 				pf_print_flags(th->th_flags);
1934 				printf("\n");
1935 			}
1936 			REASON_SET(reason, PFRES_TS);
1937 			return (PF_DROP);
1938 		}
1939 	}
1940 
1941 	/*
1942 	 * We will note if a host sends his data packets with or without
1943 	 * timestamps.  And require all data packets to contain a timestamp
1944 	 * if the first does.  PAWS implicitly requires that all data packets be
1945 	 * timestamped.  But I think there are middle-man devices that hijack
1946 	 * TCP streams immediately after the 3whs and don't timestamp their
1947 	 * packets (seen in a WWW accelerator or cache).
1948 	 */
1949 	if (pd->p_len > 0 && src->scrub && (src->scrub->pfss_flags &
1950 	    (PFSS_TIMESTAMP|PFSS_DATA_TS|PFSS_DATA_NOTS)) == PFSS_TIMESTAMP) {
1951 		if (got_ts)
1952 			src->scrub->pfss_flags |= PFSS_DATA_TS;
1953 		else {
1954 			src->scrub->pfss_flags |= PFSS_DATA_NOTS;
1955 			if (V_pf_status.debug >= PF_DEBUG_MISC && dst->scrub &&
1956 			    (dst->scrub->pfss_flags & PFSS_TIMESTAMP)) {
1957 				/* Don't warn if other host rejected RFC1323 */
1958 				DPFPRINTF(("Broken RFC1323 stack did not "
1959 				    "timestamp data packet. Disabled PAWS "
1960 				    "security.\n"));
1961 				pf_print_state(state);
1962 				pf_print_flags(th->th_flags);
1963 				printf("\n");
1964 			}
1965 		}
1966 	}
1967 
1968 	/*
1969 	 * Update PAWS values
1970 	 */
1971 	if (got_ts && src->scrub && PFSS_TIMESTAMP == (src->scrub->pfss_flags &
1972 	    (PFSS_PAWS_IDLED|PFSS_TIMESTAMP))) {
1973 		getmicrouptime(&src->scrub->pfss_last);
1974 		if (SEQ_GEQ(tsval, src->scrub->pfss_tsval) ||
1975 		    (src->scrub->pfss_flags & PFSS_PAWS) == 0)
1976 			src->scrub->pfss_tsval = tsval;
1977 
1978 		if (tsecr) {
1979 			if (SEQ_GEQ(tsecr, src->scrub->pfss_tsecr) ||
1980 			    (src->scrub->pfss_flags & PFSS_PAWS) == 0)
1981 				src->scrub->pfss_tsecr = tsecr;
1982 
1983 			if ((src->scrub->pfss_flags & PFSS_PAWS) == 0 &&
1984 			    (SEQ_LT(tsval, src->scrub->pfss_tsval0) ||
1985 			    src->scrub->pfss_tsval0 == 0)) {
1986 				/* tsval0 MUST be the lowest timestamp */
1987 				src->scrub->pfss_tsval0 = tsval;
1988 			}
1989 
1990 			/* Only fully initialized after a TS gets echoed */
1991 			if ((src->scrub->pfss_flags & PFSS_PAWS) == 0)
1992 				src->scrub->pfss_flags |= PFSS_PAWS;
1993 		}
1994 	}
1995 
1996 	/* I have a dream....  TCP segment reassembly.... */
1997 	return (0);
1998 }
1999 
2000 int
2001 pf_normalize_mss(struct mbuf *m, int off, struct pf_pdesc *pd)
2002 {
2003 	struct tcphdr	*th = &pd->hdr.tcp;
2004 	u_int16_t	*mss;
2005 	int		 thoff;
2006 	int		 opt, cnt, optlen = 0;
2007 	u_char		 opts[TCP_MAXOLEN];
2008 	u_char		*optp = opts;
2009 	size_t		 startoff;
2010 
2011 	thoff = th->th_off << 2;
2012 	cnt = thoff - sizeof(struct tcphdr);
2013 
2014 	if (cnt > 0 && !pf_pull_hdr(m, off + sizeof(*th), opts, cnt,
2015 	    NULL, NULL, pd->af))
2016 		return (0);
2017 
2018 	for (; cnt > 0; cnt -= optlen, optp += optlen) {
2019 		startoff = optp - opts;
2020 		opt = optp[0];
2021 		if (opt == TCPOPT_EOL)
2022 			break;
2023 		if (opt == TCPOPT_NOP)
2024 			optlen = 1;
2025 		else {
2026 			if (cnt < 2)
2027 				break;
2028 			optlen = optp[1];
2029 			if (optlen < 2 || optlen > cnt)
2030 				break;
2031 		}
2032 		switch (opt) {
2033 		case TCPOPT_MAXSEG:
2034 			mss = (u_int16_t *)(optp + 2);
2035 			if ((ntohs(*mss)) > pd->act.max_mss) {
2036 				pf_patch_16_unaligned(m,
2037 				    &th->th_sum,
2038 				    mss, htons(pd->act.max_mss),
2039 				    PF_ALGNMNT(startoff),
2040 				    0);
2041 				m_copyback(m, off + sizeof(*th),
2042 				    thoff - sizeof(*th), opts);
2043 				m_copyback(m, off, sizeof(*th), (caddr_t)th);
2044 			}
2045 			break;
2046 		default:
2047 			break;
2048 		}
2049 	}
2050 
2051 	return (0);
2052 }
2053 
2054 static int
2055 pf_scan_sctp(struct mbuf *m, int ipoff, int off, struct pf_pdesc *pd,
2056     struct pfi_kkif *kif)
2057 {
2058 	struct sctp_chunkhdr ch = { };
2059 	int chunk_off = sizeof(struct sctphdr);
2060 	int chunk_start;
2061 	int ret;
2062 
2063 	while (off + chunk_off < pd->tot_len) {
2064 		if (!pf_pull_hdr(m, off + chunk_off, &ch, sizeof(ch), NULL,
2065 		    NULL, pd->af))
2066 			return (PF_DROP);
2067 
2068 		/* Length includes the header, this must be at least 4. */
2069 		if (ntohs(ch.chunk_length) < 4)
2070 			return (PF_DROP);
2071 
2072 		chunk_start = chunk_off;
2073 		chunk_off += roundup(ntohs(ch.chunk_length), 4);
2074 
2075 		switch (ch.chunk_type) {
2076 		case SCTP_INITIATION:
2077 		case SCTP_INITIATION_ACK: {
2078 			struct sctp_init_chunk init;
2079 
2080 			if (!pf_pull_hdr(m, off + chunk_start, &init,
2081 			    sizeof(init), NULL, NULL, pd->af))
2082 				return (PF_DROP);
2083 
2084 			/*
2085 			 * RFC 9620, Section 3.3.2, "The Initiate Tag is allowed to have
2086 			 * any value except 0."
2087 			 */
2088 			if (init.init.initiate_tag == 0)
2089 				return (PF_DROP);
2090 			if (init.init.num_inbound_streams == 0)
2091 				return (PF_DROP);
2092 			if (init.init.num_outbound_streams == 0)
2093 				return (PF_DROP);
2094 			if (ntohl(init.init.a_rwnd) < SCTP_MIN_RWND)
2095 				return (PF_DROP);
2096 
2097 			/*
2098 			 * RFC 9260, Section 3.1, INIT chunks MUST have zero
2099 			 * verification tag.
2100 			 */
2101 			if (ch.chunk_type == SCTP_INITIATION &&
2102 			    pd->hdr.sctp.v_tag != 0)
2103 				return (PF_DROP);
2104 
2105 			pd->sctp_initiate_tag = init.init.initiate_tag;
2106 
2107 			if (ch.chunk_type == SCTP_INITIATION)
2108 				pd->sctp_flags |= PFDESC_SCTP_INIT;
2109 			else
2110 				pd->sctp_flags |= PFDESC_SCTP_INIT_ACK;
2111 
2112 			ret = pf_multihome_scan_init(m, off + chunk_start,
2113 			    ntohs(init.ch.chunk_length), pd, kif);
2114 			if (ret != PF_PASS)
2115 				return (ret);
2116 
2117 			break;
2118 		}
2119 		case SCTP_ABORT_ASSOCIATION:
2120 			pd->sctp_flags |= PFDESC_SCTP_ABORT;
2121 			break;
2122 		case SCTP_SHUTDOWN:
2123 		case SCTP_SHUTDOWN_ACK:
2124 			pd->sctp_flags |= PFDESC_SCTP_SHUTDOWN;
2125 			break;
2126 		case SCTP_SHUTDOWN_COMPLETE:
2127 			pd->sctp_flags |= PFDESC_SCTP_SHUTDOWN_COMPLETE;
2128 			break;
2129 		case SCTP_COOKIE_ECHO:
2130 			pd->sctp_flags |= PFDESC_SCTP_COOKIE;
2131 			break;
2132 		case SCTP_COOKIE_ACK:
2133 			pd->sctp_flags |= PFDESC_SCTP_COOKIE_ACK;
2134 			break;
2135 		case SCTP_DATA:
2136 			pd->sctp_flags |= PFDESC_SCTP_DATA;
2137 			break;
2138 		case SCTP_HEARTBEAT_REQUEST:
2139 			pd->sctp_flags |= PFDESC_SCTP_HEARTBEAT;
2140 			break;
2141 		case SCTP_HEARTBEAT_ACK:
2142 			pd->sctp_flags |= PFDESC_SCTP_HEARTBEAT_ACK;
2143 			break;
2144 		case SCTP_ASCONF:
2145 			pd->sctp_flags |= PFDESC_SCTP_ASCONF;
2146 
2147 			ret = pf_multihome_scan_asconf(m, off + chunk_start,
2148 			    ntohs(ch.chunk_length), pd, kif);
2149 			if (ret != PF_PASS)
2150 				return (ret);
2151 			break;
2152 		default:
2153 			pd->sctp_flags |= PFDESC_SCTP_OTHER;
2154 			break;
2155 		}
2156 	}
2157 
2158 	/* Validate chunk lengths vs. packet length. */
2159 	if (off + chunk_off != pd->tot_len)
2160 		return (PF_DROP);
2161 
2162 	/*
2163 	 * INIT, INIT_ACK or SHUTDOWN_COMPLETE chunks must always be the only
2164 	 * one in a packet.
2165 	 */
2166 	if ((pd->sctp_flags & PFDESC_SCTP_INIT) &&
2167 	    (pd->sctp_flags & ~PFDESC_SCTP_INIT))
2168 		return (PF_DROP);
2169 	if ((pd->sctp_flags & PFDESC_SCTP_INIT_ACK) &&
2170 	    (pd->sctp_flags & ~PFDESC_SCTP_INIT_ACK))
2171 		return (PF_DROP);
2172 	if ((pd->sctp_flags & PFDESC_SCTP_SHUTDOWN_COMPLETE) &&
2173 	    (pd->sctp_flags & ~PFDESC_SCTP_SHUTDOWN_COMPLETE))
2174 		return (PF_DROP);
2175 
2176 	return (PF_PASS);
2177 }
2178 
2179 int
2180 pf_normalize_sctp(int dir, struct pfi_kkif *kif, struct mbuf *m, int ipoff,
2181     int off, void *h, struct pf_pdesc *pd)
2182 {
2183 	struct pf_krule	*r, *rm = NULL;
2184 	struct sctphdr	*sh = &pd->hdr.sctp;
2185 	u_short		 reason;
2186 	sa_family_t	 af = pd->af;
2187 	int		 srs;
2188 
2189 	PF_RULES_RASSERT();
2190 
2191 	/* Unconditionally scan the SCTP packet, because we need to look for
2192 	 * things like shutdown and asconf chunks. */
2193 	if (pf_scan_sctp(m, ipoff, off, pd, kif) != PF_PASS)
2194 		goto sctp_drop;
2195 
2196 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
2197 	/* Check if there any scrub rules. Lack of scrub rules means enforced
2198 	 * packet normalization operation just like in OpenBSD. */
2199 	srs = (r != NULL);
2200 	while (r != NULL) {
2201 		pf_counter_u64_add(&r->evaluations, 1);
2202 		if (pfi_kkif_match(r->kif, kif) == r->ifnot)
2203 			r = r->skip[PF_SKIP_IFP].ptr;
2204 		else if (r->direction && r->direction != dir)
2205 			r = r->skip[PF_SKIP_DIR].ptr;
2206 		else if (r->af && r->af != af)
2207 			r = r->skip[PF_SKIP_AF].ptr;
2208 		else if (r->proto && r->proto != pd->proto)
2209 			r = r->skip[PF_SKIP_PROTO].ptr;
2210 		else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
2211 		    r->src.neg, kif, M_GETFIB(m)))
2212 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
2213 		else if (r->src.port_op && !pf_match_port(r->src.port_op,
2214 			    r->src.port[0], r->src.port[1], sh->src_port))
2215 			r = r->skip[PF_SKIP_SRC_PORT].ptr;
2216 		else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
2217 		    r->dst.neg, NULL, M_GETFIB(m)))
2218 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
2219 		else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
2220 			    r->dst.port[0], r->dst.port[1], sh->dest_port))
2221 			r = r->skip[PF_SKIP_DST_PORT].ptr;
2222 		else {
2223 			rm = r;
2224 			break;
2225 		}
2226 	}
2227 
2228 	if (srs) {
2229 		/* With scrub rules present SCTP normalization happens only
2230 		 * if one of rules has matched and it's not a "no scrub" rule */
2231 		if (rm == NULL || rm->action == PF_NOSCRUB)
2232 			return (PF_PASS);
2233 
2234 		pf_counter_u64_critical_enter();
2235 		pf_counter_u64_add_protected(&r->packets[dir == PF_OUT], 1);
2236 		pf_counter_u64_add_protected(&r->bytes[dir == PF_OUT], pd->tot_len);
2237 		pf_counter_u64_critical_exit();
2238 	}
2239 
2240 	/* Verify we're a multiple of 4 bytes long */
2241 	if ((pd->tot_len - off - sizeof(struct sctphdr)) % 4)
2242 		goto sctp_drop;
2243 
2244 	/* INIT chunk needs to be the only chunk */
2245 	if (pd->sctp_flags & PFDESC_SCTP_INIT)
2246 		if (pd->sctp_flags & ~PFDESC_SCTP_INIT)
2247 			goto sctp_drop;
2248 
2249 	return (PF_PASS);
2250 
2251 sctp_drop:
2252 	REASON_SET(&reason, PFRES_NORM);
2253 	if (rm != NULL && r->log)
2254 		PFLOG_PACKET(kif, m, AF_INET, PF_DROP, reason, r, NULL, NULL, pd,
2255 		    1);
2256 
2257 	return (PF_DROP);
2258 }
2259 
2260 #ifdef INET
2261 void
2262 pf_scrub_ip(struct mbuf **m0, struct pf_pdesc *pd)
2263 {
2264 	struct mbuf		*m = *m0;
2265 	struct ip		*h = mtod(m, struct ip *);
2266 
2267 	/* Clear IP_DF if no-df was requested */
2268 	if (pd->act.flags & PFSTATE_NODF && h->ip_off & htons(IP_DF)) {
2269 		u_int16_t ip_off = h->ip_off;
2270 
2271 		h->ip_off &= htons(~IP_DF);
2272 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
2273 	}
2274 
2275 	/* Enforce a minimum ttl, may cause endless packet loops */
2276 	if (pd->act.min_ttl && h->ip_ttl < pd->act.min_ttl) {
2277 		u_int16_t ip_ttl = h->ip_ttl;
2278 
2279 		h->ip_ttl = pd->act.min_ttl;
2280 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_ttl, h->ip_ttl, 0);
2281 	}
2282 
2283 	/* Enforce tos */
2284 	if (pd->act.flags & PFSTATE_SETTOS) {
2285 		u_int16_t	ov, nv;
2286 
2287 		ov = *(u_int16_t *)h;
2288 		h->ip_tos = pd->act.set_tos | (h->ip_tos & IPTOS_ECN_MASK);
2289 		nv = *(u_int16_t *)h;
2290 
2291 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ov, nv, 0);
2292 	}
2293 
2294 	/* random-id, but not for fragments */
2295 	if (pd->act.flags & PFSTATE_RANDOMID && !(h->ip_off & ~htons(IP_DF))) {
2296 		uint16_t ip_id = h->ip_id;
2297 
2298 		ip_fillid(h);
2299 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_id, h->ip_id, 0);
2300 	}
2301 }
2302 #endif /* INET */
2303 
2304 #ifdef INET6
2305 void
2306 pf_scrub_ip6(struct mbuf **m0, struct pf_pdesc *pd)
2307 {
2308 	struct mbuf		*m = *m0;
2309 	struct ip6_hdr		*h = mtod(m, struct ip6_hdr *);
2310 
2311 	/* Enforce a minimum ttl, may cause endless packet loops */
2312 	if (pd->act.min_ttl && h->ip6_hlim < pd->act.min_ttl)
2313 		h->ip6_hlim = pd->act.min_ttl;
2314 
2315 	/* Enforce tos. Set traffic class bits */
2316 	if (pd->act.flags & PFSTATE_SETTOS) {
2317 		h->ip6_flow &= IPV6_FLOWLABEL_MASK | IPV6_VERSION_MASK;
2318 		h->ip6_flow |= htonl((pd->act.set_tos | IPV6_ECN(h)) << 20);
2319 	}
2320 }
2321 #endif
2322