xref: /freebsd-14-stable/sys/netpfil/pf/pf_norm.c (revision 2e0f053ad52b38bd8bca72f817d7347df87dbe98)
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 	TAILQ_HEAD(pf_fragq, pf_frent) fr_queue;
99 	uint16_t	fr_maxlen;	/* maximum length of single fragment */
100 	u_int16_t	fr_holes;	/* number of holes in the 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
pf_ip2key(struct ip * ip,int dir,struct pf_fragment_cmp * key)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
pf_normalize_init(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
pf_normalize_cleanup(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
pf_frag_compare(struct pf_fragment * a,struct pf_fragment * b)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
pf_purge_expired_fragments(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
pf_purge_fragments(uint32_t expire)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
pf_flush_fragments(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
pf_free_fragment(struct pf_fragment * frag)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 *
pf_find_fragment(struct pf_fragment_cmp * key,struct pf_frag_tree * tree)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
pf_remove_fragment(struct pf_fragment * frag)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 *
pf_create_fragment(u_short * reason)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
pf_frent_holes(struct pf_frent * frent)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
pf_frent_index(struct pf_frent * frent)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
pf_frent_insert(struct pf_fragment * frag,struct pf_frent * frent,struct pf_frent * prev)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
pf_frent_remove(struct pf_fragment * frag,struct pf_frent * frent)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 *
pf_frent_previous(struct pf_fragment * frag,struct pf_frent * 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 *
pf_fillup_fragment(struct pf_fragment_cmp * key,struct pf_frent * frent,u_short * reason)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 
551 	PF_FRAG_ASSERT();
552 
553 	/* No empty fragments. */
554 	if (frent->fe_len == 0) {
555 		DPFPRINTF(("bad fragment: len 0\n"));
556 		goto bad_fragment;
557 	}
558 
559 	/* All fragments are 8 byte aligned. */
560 	if (frent->fe_mff && (frent->fe_len & 0x7)) {
561 		DPFPRINTF(("bad fragment: mff and len %d\n", frent->fe_len));
562 		goto bad_fragment;
563 	}
564 
565 	/* Respect maximum length, IP_MAXPACKET == IPV6_MAXPACKET. */
566 	if (frent->fe_off + frent->fe_len > IP_MAXPACKET) {
567 		DPFPRINTF(("bad fragment: max packet %d\n",
568 		    frent->fe_off + frent->fe_len));
569 		goto bad_fragment;
570 	}
571 
572 	DPFPRINTF((key->frc_af == AF_INET ?
573 	    "reass frag %d @ %d-%d\n" : "reass frag %#08x @ %d-%d\n",
574 	    key->frc_id, frent->fe_off, frent->fe_off + frent->fe_len));
575 
576 	/* Fully buffer all of the fragments in this fragment queue. */
577 	frag = pf_find_fragment(key, &V_pf_frag_tree);
578 
579 	/* Create a new reassembly queue for this packet. */
580 	if (frag == NULL) {
581 		frag = uma_zalloc(V_pf_frag_z, M_NOWAIT);
582 		if (frag == NULL) {
583 			pf_flush_fragments();
584 			frag = uma_zalloc(V_pf_frag_z, M_NOWAIT);
585 			if (frag == NULL) {
586 				REASON_SET(reason, PFRES_MEMORY);
587 				goto drop_fragment;
588 			}
589 		}
590 
591 		*(struct pf_fragment_cmp *)frag = *key;
592 		memset(frag->fr_firstoff, 0, sizeof(frag->fr_firstoff));
593 		memset(frag->fr_entries, 0, sizeof(frag->fr_entries));
594 		frag->fr_timeout = time_uptime;
595 		TAILQ_INIT(&frag->fr_queue);
596 		frag->fr_maxlen = frent->fe_len;
597 		frag->fr_holes = 1;
598 
599 		RB_INSERT(pf_frag_tree, &V_pf_frag_tree, frag);
600 		TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next);
601 
602 		/* We do not have a previous fragment, cannot fail. */
603 		pf_frent_insert(frag, frent, NULL);
604 
605 		return (frag);
606 	}
607 
608 	KASSERT(!TAILQ_EMPTY(&frag->fr_queue), ("!TAILQ_EMPTY()->fr_queue"));
609 
610 	/* Remember maximum fragment len for refragmentation. */
611 	if (frent->fe_len > frag->fr_maxlen)
612 		frag->fr_maxlen = frent->fe_len;
613 
614 	/* Maximum data we have seen already. */
615 	total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
616 		TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
617 
618 	/* Non terminal fragments must have more fragments flag. */
619 	if (frent->fe_off + frent->fe_len < total && !frent->fe_mff)
620 		goto bad_fragment;
621 
622 	/* Check if we saw the last fragment already. */
623 	if (!TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_mff) {
624 		if (frent->fe_off + frent->fe_len > total ||
625 		    (frent->fe_off + frent->fe_len == total && frent->fe_mff))
626 			goto bad_fragment;
627 	} else {
628 		if (frent->fe_off + frent->fe_len == total && !frent->fe_mff)
629 			goto bad_fragment;
630 	}
631 
632 	/* Find neighbors for newly inserted fragment */
633 	prev = pf_frent_previous(frag, frent);
634 	if (prev == NULL) {
635 		after = TAILQ_FIRST(&frag->fr_queue);
636 		KASSERT(after != NULL, ("after != NULL"));
637 	} else {
638 		after = TAILQ_NEXT(prev, fr_next);
639 	}
640 
641 	if (prev != NULL && prev->fe_off + prev->fe_len > frent->fe_off) {
642 		uint16_t precut;
643 
644 		if (frag->fr_af == AF_INET6)
645 			goto free_fragment;
646 
647 		precut = prev->fe_off + prev->fe_len - frent->fe_off;
648 		if (precut >= frent->fe_len) {
649 			DPFPRINTF(("new frag overlapped\n"));
650 			goto drop_fragment;
651 		}
652 		DPFPRINTF(("frag head overlap %d\n", precut));
653 		m_adj(frent->fe_m, precut);
654 		frent->fe_off += precut;
655 		frent->fe_len -= precut;
656 	}
657 
658 	for (; after != NULL && frent->fe_off + frent->fe_len > after->fe_off;
659 	    after = next) {
660 		uint16_t aftercut;
661 
662 		aftercut = frent->fe_off + frent->fe_len - after->fe_off;
663 		if (aftercut < after->fe_len) {
664 			DPFPRINTF(("frag tail overlap %d", aftercut));
665 			m_adj(after->fe_m, aftercut);
666 			/* Fragment may switch queue as fe_off changes */
667 			pf_frent_remove(frag, after);
668 			after->fe_off += aftercut;
669 			after->fe_len -= aftercut;
670 			/* Insert into correct queue */
671 			if (pf_frent_insert(frag, after, prev)) {
672 				DPFPRINTF(("fragment requeue limit exceeded"));
673 				m_freem(after->fe_m);
674 				uma_zfree(V_pf_frent_z, after);
675 				/* There is not way to recover */
676 				goto free_fragment;
677 			}
678 			break;
679 		}
680 
681 		/* This fragment is completely overlapped, lose it. */
682 		DPFPRINTF(("old frag overlapped\n"));
683 		next = TAILQ_NEXT(after, fr_next);
684 		pf_frent_remove(frag, after);
685 		m_freem(after->fe_m);
686 		uma_zfree(V_pf_frent_z, after);
687 	}
688 
689 	/* If part of the queue gets too long, there is not way to recover. */
690 	if (pf_frent_insert(frag, frent, prev)) {
691 		DPFPRINTF(("fragment queue limit exceeded\n"));
692 		goto bad_fragment;
693 	}
694 
695 	return (frag);
696 
697 free_fragment:
698 	/*
699 	 * RFC 5722, Errata 3089:  When reassembling an IPv6 datagram, if one
700 	 * or more its constituent fragments is determined to be an overlapping
701 	 * fragment, the entire datagram (and any constituent fragments) MUST
702 	 * be silently discarded.
703 	 */
704 	DPFPRINTF(("flush overlapping fragments\n"));
705 	pf_free_fragment(frag);
706 
707 bad_fragment:
708 	REASON_SET(reason, PFRES_FRAG);
709 drop_fragment:
710 	uma_zfree(V_pf_frent_z, frent);
711 	return (NULL);
712 }
713 
714 static struct mbuf *
pf_join_fragment(struct pf_fragment * frag)715 pf_join_fragment(struct pf_fragment *frag)
716 {
717 	struct mbuf *m, *m2;
718 	struct pf_frent	*frent, *next;
719 
720 	frent = TAILQ_FIRST(&frag->fr_queue);
721 	next = TAILQ_NEXT(frent, fr_next);
722 
723 	m = frent->fe_m;
724 	m_adj(m, (frent->fe_hdrlen + frent->fe_len) - m->m_pkthdr.len);
725 	uma_zfree(V_pf_frent_z, frent);
726 	for (frent = next; frent != NULL; frent = next) {
727 		next = TAILQ_NEXT(frent, fr_next);
728 
729 		m2 = frent->fe_m;
730 		/* Strip off ip header. */
731 		m_adj(m2, frent->fe_hdrlen);
732 		/* Strip off any trailing bytes. */
733 		m_adj(m2, frent->fe_len - m2->m_pkthdr.len);
734 
735 		uma_zfree(V_pf_frent_z, frent);
736 		m_cat(m, m2);
737 	}
738 
739 	/* Remove from fragment queue. */
740 	pf_remove_fragment(frag);
741 
742 	return (m);
743 }
744 
745 #ifdef INET
746 static int
pf_reassemble(struct mbuf ** m0,struct ip * ip,int dir,u_short * reason)747 pf_reassemble(struct mbuf **m0, struct ip *ip, int dir, u_short *reason)
748 {
749 	struct mbuf		*m = *m0;
750 	struct pf_frent		*frent;
751 	struct pf_fragment	*frag;
752 	struct pf_fragment_cmp	key;
753 	uint16_t		total, hdrlen;
754 
755 	/* Get an entry for the fragment queue */
756 	if ((frent = pf_create_fragment(reason)) == NULL)
757 		return (PF_DROP);
758 
759 	frent->fe_m = m;
760 	frent->fe_hdrlen = ip->ip_hl << 2;
761 	frent->fe_extoff = 0;
762 	frent->fe_len = ntohs(ip->ip_len) - (ip->ip_hl << 2);
763 	frent->fe_off = (ntohs(ip->ip_off) & IP_OFFMASK) << 3;
764 	frent->fe_mff = ntohs(ip->ip_off) & IP_MF;
765 
766 	pf_ip2key(ip, dir, &key);
767 
768 	if ((frag = pf_fillup_fragment(&key, frent, reason)) == NULL)
769 		return (PF_DROP);
770 
771 	/* The mbuf is part of the fragment entry, no direct free or access */
772 	m = *m0 = NULL;
773 
774 	if (frag->fr_holes) {
775 		DPFPRINTF(("frag %d, holes %d\n", frag->fr_id, frag->fr_holes));
776 		return (PF_PASS);  /* drop because *m0 is NULL, no error */
777 	}
778 
779 	/* We have all the data */
780 	frent = TAILQ_FIRST(&frag->fr_queue);
781 	KASSERT(frent != NULL, ("frent != NULL"));
782 	total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
783 		TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
784 	hdrlen = frent->fe_hdrlen;
785 
786 	m = *m0 = pf_join_fragment(frag);
787 	frag = NULL;
788 
789 	if (m->m_flags & M_PKTHDR) {
790 		int plen = 0;
791 		for (m = *m0; m; m = m->m_next)
792 			plen += m->m_len;
793 		m = *m0;
794 		m->m_pkthdr.len = plen;
795 	}
796 
797 	ip = mtod(m, struct ip *);
798 	ip->ip_sum = pf_cksum_fixup(ip->ip_sum, ip->ip_len,
799 	    htons(hdrlen + total), 0);
800 	ip->ip_len = htons(hdrlen + total);
801 	ip->ip_sum = pf_cksum_fixup(ip->ip_sum, ip->ip_off,
802 	    ip->ip_off & ~(IP_MF|IP_OFFMASK), 0);
803 	ip->ip_off &= ~(IP_MF|IP_OFFMASK);
804 
805 	if (hdrlen + total > IP_MAXPACKET) {
806 		DPFPRINTF(("drop: too big: %d\n", total));
807 		ip->ip_len = 0;
808 		REASON_SET(reason, PFRES_SHORT);
809 		/* PF_DROP requires a valid mbuf *m0 in pf_test() */
810 		return (PF_DROP);
811 	}
812 
813 	DPFPRINTF(("complete: %p(%d)\n", m, ntohs(ip->ip_len)));
814 	return (PF_PASS);
815 }
816 #endif	/* INET */
817 
818 #ifdef INET6
819 static int
pf_reassemble6(struct mbuf ** m0,struct ip6_hdr * ip6,struct ip6_frag * fraghdr,uint16_t hdrlen,uint16_t extoff,u_short * reason)820 pf_reassemble6(struct mbuf **m0, struct ip6_hdr *ip6, struct ip6_frag *fraghdr,
821     uint16_t hdrlen, uint16_t extoff, u_short *reason)
822 {
823 	struct mbuf		*m = *m0;
824 	struct pf_frent		*frent;
825 	struct pf_fragment	*frag;
826 	struct pf_fragment_cmp	 key;
827 	struct m_tag		*mtag;
828 	struct pf_fragment_tag	*ftag;
829 	int			 off;
830 	uint32_t		 frag_id;
831 	uint16_t		 total, maxlen;
832 	uint8_t			 proto;
833 
834 	PF_FRAG_LOCK();
835 
836 	/* Get an entry for the fragment queue. */
837 	if ((frent = pf_create_fragment(reason)) == NULL) {
838 		PF_FRAG_UNLOCK();
839 		return (PF_DROP);
840 	}
841 
842 	frent->fe_m = m;
843 	frent->fe_hdrlen = hdrlen;
844 	frent->fe_extoff = extoff;
845 	frent->fe_len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - hdrlen;
846 	frent->fe_off = ntohs(fraghdr->ip6f_offlg & IP6F_OFF_MASK);
847 	frent->fe_mff = fraghdr->ip6f_offlg & IP6F_MORE_FRAG;
848 
849 	key.frc_src.v6 = ip6->ip6_src;
850 	key.frc_dst.v6 = ip6->ip6_dst;
851 	key.frc_af = AF_INET6;
852 	/* Only the first fragment's protocol is relevant. */
853 	key.frc_proto = 0;
854 	key.frc_id = fraghdr->ip6f_ident;
855 
856 	if ((frag = pf_fillup_fragment(&key, frent, reason)) == NULL) {
857 		PF_FRAG_UNLOCK();
858 		return (PF_DROP);
859 	}
860 
861 	/* The mbuf is part of the fragment entry, no direct free or access. */
862 	m = *m0 = NULL;
863 
864 	if (frag->fr_holes) {
865 		DPFPRINTF(("frag %d, holes %d\n", frag->fr_id,
866 		    frag->fr_holes));
867 		PF_FRAG_UNLOCK();
868 		return (PF_PASS);  /* Drop because *m0 is NULL, no error. */
869 	}
870 
871 	/* We have all the data. */
872 	frent = TAILQ_FIRST(&frag->fr_queue);
873 	KASSERT(frent != NULL, ("frent != NULL"));
874 	extoff = frent->fe_extoff;
875 	maxlen = frag->fr_maxlen;
876 	frag_id = frag->fr_id;
877 	total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
878 		TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
879 	hdrlen = frent->fe_hdrlen - sizeof(struct ip6_frag);
880 
881 	m = *m0 = pf_join_fragment(frag);
882 	frag = NULL;
883 
884 	PF_FRAG_UNLOCK();
885 
886 	/* Take protocol from first fragment header. */
887 	m = m_getptr(m, hdrlen + offsetof(struct ip6_frag, ip6f_nxt), &off);
888 	KASSERT(m, ("%s: short mbuf chain", __func__));
889 	proto = *(mtod(m, uint8_t *) + off);
890 	m = *m0;
891 
892 	/* Delete frag6 header */
893 	if (ip6_deletefraghdr(m, hdrlen, M_NOWAIT) != 0)
894 		goto fail;
895 
896 	if (m->m_flags & M_PKTHDR) {
897 		int plen = 0;
898 		for (m = *m0; m; m = m->m_next)
899 			plen += m->m_len;
900 		m = *m0;
901 		m->m_pkthdr.len = plen;
902 	}
903 
904 	if ((mtag = m_tag_get(PACKET_TAG_PF_REASSEMBLED,
905 	    sizeof(struct pf_fragment_tag), M_NOWAIT)) == NULL)
906 		goto fail;
907 	ftag = (struct pf_fragment_tag *)(mtag + 1);
908 	ftag->ft_hdrlen = hdrlen;
909 	ftag->ft_extoff = extoff;
910 	ftag->ft_maxlen = maxlen;
911 	ftag->ft_id = frag_id;
912 	m_tag_prepend(m, mtag);
913 
914 	ip6 = mtod(m, struct ip6_hdr *);
915 	ip6->ip6_plen = htons(hdrlen - sizeof(struct ip6_hdr) + total);
916 	if (extoff) {
917 		/* Write protocol into next field of last extension header. */
918 		m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt),
919 		    &off);
920 		KASSERT(m, ("%s: short mbuf chain", __func__));
921 		*(mtod(m, char *) + off) = proto;
922 		m = *m0;
923 	} else
924 		ip6->ip6_nxt = proto;
925 
926 	if (hdrlen - sizeof(struct ip6_hdr) + total > IPV6_MAXPACKET) {
927 		DPFPRINTF(("drop: too big: %d\n", total));
928 		ip6->ip6_plen = 0;
929 		REASON_SET(reason, PFRES_SHORT);
930 		/* PF_DROP requires a valid mbuf *m0 in pf_test6(). */
931 		return (PF_DROP);
932 	}
933 
934 	DPFPRINTF(("complete: %p(%d)\n", m, ntohs(ip6->ip6_plen)));
935 	return (PF_PASS);
936 
937 fail:
938 	REASON_SET(reason, PFRES_MEMORY);
939 	/* PF_DROP requires a valid mbuf *m0 in pf_test6(), will free later. */
940 	return (PF_DROP);
941 }
942 #endif	/* INET6 */
943 
944 #ifdef INET6
945 int
pf_refragment6(struct ifnet * ifp,struct mbuf ** m0,struct m_tag * mtag,bool forward)946 pf_refragment6(struct ifnet *ifp, struct mbuf **m0, struct m_tag *mtag,
947     bool forward)
948 {
949 	struct mbuf		*m = *m0, *t;
950 	struct ip6_hdr		*hdr;
951 	struct pf_fragment_tag	*ftag = (struct pf_fragment_tag *)(mtag + 1);
952 	struct pf_pdesc		 pd;
953 	uint32_t		 frag_id;
954 	uint16_t		 hdrlen, extoff, maxlen;
955 	uint8_t			 proto;
956 	int			 error, action;
957 
958 	hdrlen = ftag->ft_hdrlen;
959 	extoff = ftag->ft_extoff;
960 	maxlen = ftag->ft_maxlen;
961 	frag_id = ftag->ft_id;
962 	m_tag_delete(m, mtag);
963 	mtag = NULL;
964 	ftag = NULL;
965 
966 	if (extoff) {
967 		int off;
968 
969 		/* Use protocol from next field of last extension header */
970 		m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt),
971 		    &off);
972 		KASSERT((m != NULL), ("pf_refragment6: short mbuf chain"));
973 		proto = *(mtod(m, uint8_t *) + off);
974 		*(mtod(m, char *) + off) = IPPROTO_FRAGMENT;
975 		m = *m0;
976 	} else {
977 		hdr = mtod(m, struct ip6_hdr *);
978 		proto = hdr->ip6_nxt;
979 		hdr->ip6_nxt = IPPROTO_FRAGMENT;
980 	}
981 
982 	/* In case of link-local traffic we'll need a scope set. */
983 	hdr = mtod(m, struct ip6_hdr *);
984 
985 	in6_setscope(&hdr->ip6_src, ifp, NULL);
986 	in6_setscope(&hdr->ip6_dst, ifp, NULL);
987 
988 	/* The MTU must be a multiple of 8 bytes, or we risk doing the
989 	 * fragmentation wrong. */
990 	maxlen = maxlen & ~7;
991 
992 	/*
993 	 * Maxlen may be less than 8 if there was only a single
994 	 * fragment.  As it was fragmented before, add a fragment
995 	 * header also for a single fragment.  If total or maxlen
996 	 * is less than 8, ip6_fragment() will return EMSGSIZE and
997 	 * we drop the packet.
998 	 */
999 	error = ip6_fragment(ifp, m, hdrlen, proto, maxlen, frag_id);
1000 	m = (*m0)->m_nextpkt;
1001 	(*m0)->m_nextpkt = NULL;
1002 	if (error == 0) {
1003 		/* The first mbuf contains the unfragmented packet. */
1004 		m_freem(*m0);
1005 		*m0 = NULL;
1006 		action = PF_PASS;
1007 	} else {
1008 		/* Drop expects an mbuf to free. */
1009 		DPFPRINTF(("refragment error %d\n", error));
1010 		action = PF_DROP;
1011 	}
1012 	for (; m; m = t) {
1013 		t = m->m_nextpkt;
1014 		m->m_nextpkt = NULL;
1015 		m->m_flags |= M_SKIP_FIREWALL;
1016 		memset(&pd, 0, sizeof(pd));
1017 		pd.pf_mtag = pf_find_mtag(m);
1018 		if (error == 0)
1019 			if (forward) {
1020 				MPASS(m->m_pkthdr.rcvif != NULL);
1021 				ip6_forward(m, 0);
1022 			} else {
1023 				(void)ip6_output(m, NULL, NULL, 0, NULL, NULL,
1024 				    NULL);
1025 			}
1026 		else
1027 			m_freem(m);
1028 	}
1029 
1030 	return (action);
1031 }
1032 #endif /* INET6 */
1033 
1034 #ifdef INET
1035 int
pf_normalize_ip(struct mbuf ** m0,struct pfi_kkif * kif,u_short * reason,struct pf_pdesc * pd)1036 pf_normalize_ip(struct mbuf **m0, struct pfi_kkif *kif, u_short *reason,
1037     struct pf_pdesc *pd)
1038 {
1039 	struct mbuf		*m = *m0;
1040 	struct pf_krule		*r;
1041 	struct ip		*h = mtod(m, struct ip *);
1042 	int			 mff = (ntohs(h->ip_off) & IP_MF);
1043 	int			 hlen = h->ip_hl << 2;
1044 	u_int16_t		 fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
1045 	u_int16_t		 max;
1046 	int			 ip_len;
1047 	int			 tag = -1;
1048 	int			 verdict;
1049 	bool			 scrub_compat;
1050 
1051 	PF_RULES_RASSERT();
1052 
1053 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1054 	/*
1055 	 * Check if there are any scrub rules, matching or not.
1056 	 * Lack of scrub rules means:
1057 	 *  - enforced packet normalization operation just like in OpenBSD
1058 	 *  - fragment reassembly depends on V_pf_status.reass
1059 	 * With scrub rules:
1060 	 *  - packet normalization is performed if there is a matching scrub rule
1061 	 *  - fragment reassembly is performed if the matching rule has no
1062 	 *    PFRULE_FRAGMENT_NOREASS flag
1063 	 */
1064 	scrub_compat = (r != NULL);
1065 	while (r != NULL) {
1066 		pf_counter_u64_add(&r->evaluations, 1);
1067 		if (pfi_kkif_match(r->kif, kif) == r->ifnot)
1068 			r = r->skip[PF_SKIP_IFP].ptr;
1069 		else if (r->direction && r->direction != pd->dir)
1070 			r = r->skip[PF_SKIP_DIR].ptr;
1071 		else if (r->af && r->af != AF_INET)
1072 			r = r->skip[PF_SKIP_AF].ptr;
1073 		else if (r->proto && r->proto != h->ip_p)
1074 			r = r->skip[PF_SKIP_PROTO].ptr;
1075 		else if (PF_MISMATCHAW(&r->src.addr,
1076 		    (struct pf_addr *)&h->ip_src.s_addr, AF_INET,
1077 		    r->src.neg, kif, M_GETFIB(m)))
1078 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
1079 		else if (PF_MISMATCHAW(&r->dst.addr,
1080 		    (struct pf_addr *)&h->ip_dst.s_addr, AF_INET,
1081 		    r->dst.neg, NULL, M_GETFIB(m)))
1082 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
1083 		else if (r->match_tag && !pf_match_tag(m, r, &tag,
1084 		    pd->pf_mtag ? pd->pf_mtag->tag : 0))
1085 			r = TAILQ_NEXT(r, entries);
1086 		else
1087 			break;
1088 	}
1089 
1090 	if (scrub_compat) {
1091 		/* With scrub rules present IPv4 normalization happens only
1092 		 * if one of rules has matched and it's not a "no scrub" rule */
1093 		if (r == NULL || r->action == PF_NOSCRUB)
1094 			return (PF_PASS);
1095 
1096 		pf_counter_u64_critical_enter();
1097 		pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1);
1098 		pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len);
1099 		pf_counter_u64_critical_exit();
1100 		pf_rule_to_actions(r, &pd->act);
1101 	}
1102 
1103 	/* Check for illegal packets */
1104 	if (hlen < (int)sizeof(struct ip)) {
1105 		REASON_SET(reason, PFRES_NORM);
1106 		goto drop;
1107 	}
1108 
1109 	if (hlen > ntohs(h->ip_len)) {
1110 		REASON_SET(reason, PFRES_NORM);
1111 		goto drop;
1112 	}
1113 
1114 	/* Clear IP_DF if the rule uses the no-df option or we're in no-df mode */
1115 	if (((!scrub_compat && V_pf_status.reass & PF_REASS_NODF) ||
1116 	    (r != NULL && r->rule_flag & PFRULE_NODF)) &&
1117 	    (h->ip_off & htons(IP_DF))
1118 	) {
1119 		u_int16_t ip_off = h->ip_off;
1120 
1121 		h->ip_off &= htons(~IP_DF);
1122 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
1123 	}
1124 
1125 	/* We will need other tests here */
1126 	if (!fragoff && !mff)
1127 		goto no_fragment;
1128 
1129 	/* We're dealing with a fragment now. Don't allow fragments
1130 	 * with IP_DF to enter the cache. If the flag was cleared by
1131 	 * no-df above, fine. Otherwise drop it.
1132 	 */
1133 	if (h->ip_off & htons(IP_DF)) {
1134 		DPFPRINTF(("IP_DF\n"));
1135 		goto bad;
1136 	}
1137 
1138 	ip_len = ntohs(h->ip_len) - hlen;
1139 
1140 	/* All fragments are 8 byte aligned */
1141 	if (mff && (ip_len & 0x7)) {
1142 		DPFPRINTF(("mff and %d\n", ip_len));
1143 		goto bad;
1144 	}
1145 
1146 	/* Respect maximum length */
1147 	if (fragoff + ip_len > IP_MAXPACKET) {
1148 		DPFPRINTF(("max packet %d\n", fragoff + ip_len));
1149 		goto bad;
1150 	}
1151 
1152 	if ((!scrub_compat && V_pf_status.reass) ||
1153 	    (r != NULL && !(r->rule_flag & PFRULE_FRAGMENT_NOREASS))
1154 	) {
1155 		max = fragoff + ip_len;
1156 
1157 		/* Fully buffer all of the fragments
1158 		 * Might return a completely reassembled mbuf, or NULL */
1159 		PF_FRAG_LOCK();
1160 		DPFPRINTF(("reass frag %d @ %d-%d\n", h->ip_id, fragoff, max));
1161 		verdict = pf_reassemble(m0, h, pd->dir, reason);
1162 		PF_FRAG_UNLOCK();
1163 
1164 		if (verdict != PF_PASS)
1165 			return (PF_DROP);
1166 
1167 		m = *m0;
1168 		if (m == NULL)
1169 			return (PF_DROP);
1170 
1171 		h = mtod(m, struct ip *);
1172 
1173  no_fragment:
1174 		/* At this point, only IP_DF is allowed in ip_off */
1175 		if (h->ip_off & ~htons(IP_DF)) {
1176 			u_int16_t ip_off = h->ip_off;
1177 
1178 			h->ip_off &= htons(IP_DF);
1179 			h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
1180 		}
1181 	}
1182 
1183 	return (PF_PASS);
1184 
1185  bad:
1186 	DPFPRINTF(("dropping bad fragment\n"));
1187 	REASON_SET(reason, PFRES_FRAG);
1188  drop:
1189 	if (r != NULL && r->log)
1190 		PFLOG_PACKET(kif, m, AF_INET, *reason, r, NULL, NULL, pd, 1);
1191 
1192 	return (PF_DROP);
1193 }
1194 #endif
1195 
1196 #ifdef INET6
1197 int
pf_normalize_ip6(struct mbuf ** m0,struct pfi_kkif * kif,u_short * reason,struct pf_pdesc * pd)1198 pf_normalize_ip6(struct mbuf **m0, struct pfi_kkif *kif,
1199     u_short *reason, struct pf_pdesc *pd)
1200 {
1201 	struct mbuf		*m = *m0;
1202 	struct pf_krule		*r;
1203 	struct ip6_hdr		*h = mtod(m, struct ip6_hdr *);
1204 	int			 extoff;
1205 	int			 off;
1206 	struct ip6_ext		 ext;
1207 	struct ip6_opt		 opt;
1208 	struct ip6_frag		 frag;
1209 	u_int32_t		 plen;
1210 	int			 optend;
1211 	int			 ooff;
1212 	u_int8_t		 proto;
1213 	int			 terminal;
1214 	bool			 scrub_compat;
1215 
1216 	PF_RULES_RASSERT();
1217 
1218 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1219 	/*
1220 	 * Check if there are any scrub rules, matching or not.
1221 	 * Lack of scrub rules means:
1222 	 *  - enforced packet normalization operation just like in OpenBSD
1223 	 * With scrub rules:
1224 	 *  - packet normalization is performed if there is a matching scrub rule
1225 	 * XXX: Fragment reassembly always performed for IPv6!
1226 	 */
1227 	scrub_compat = (r != NULL);
1228 	while (r != NULL) {
1229 		pf_counter_u64_add(&r->evaluations, 1);
1230 		if (pfi_kkif_match(r->kif, kif) == r->ifnot)
1231 			r = r->skip[PF_SKIP_IFP].ptr;
1232 		else if (r->direction && r->direction != pd->dir)
1233 			r = r->skip[PF_SKIP_DIR].ptr;
1234 		else if (r->af && r->af != AF_INET6)
1235 			r = r->skip[PF_SKIP_AF].ptr;
1236 #if 0 /* header chain! */
1237 		else if (r->proto && r->proto != h->ip6_nxt)
1238 			r = r->skip[PF_SKIP_PROTO].ptr;
1239 #endif
1240 		else if (PF_MISMATCHAW(&r->src.addr,
1241 		    (struct pf_addr *)&h->ip6_src, AF_INET6,
1242 		    r->src.neg, kif, M_GETFIB(m)))
1243 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
1244 		else if (PF_MISMATCHAW(&r->dst.addr,
1245 		    (struct pf_addr *)&h->ip6_dst, AF_INET6,
1246 		    r->dst.neg, NULL, M_GETFIB(m)))
1247 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
1248 		else
1249 			break;
1250 	}
1251 
1252 	if (scrub_compat) {
1253 		/* With scrub rules present IPv6 normalization happens only
1254 		 * if one of rules has matched and it's not a "no scrub" rule */
1255 		if (r == NULL || r->action == PF_NOSCRUB)
1256 			return (PF_PASS);
1257 
1258 		pf_counter_u64_critical_enter();
1259 		pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1);
1260 		pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len);
1261 		pf_counter_u64_critical_exit();
1262 		pf_rule_to_actions(r, &pd->act);
1263 	}
1264 
1265 	/* Check for illegal packets */
1266 	if (sizeof(struct ip6_hdr) + IPV6_MAXPACKET < m->m_pkthdr.len)
1267 		goto drop;
1268 
1269 again:
1270 	h = mtod(m, struct ip6_hdr *);
1271 	plen = ntohs(h->ip6_plen);
1272 	/* jumbo payload option not supported */
1273 	if (plen == 0)
1274 		goto drop;
1275 
1276 	extoff = 0;
1277 	off = sizeof(struct ip6_hdr);
1278 	proto = h->ip6_nxt;
1279 	terminal = 0;
1280 	do {
1281 		switch (proto) {
1282 		case IPPROTO_FRAGMENT:
1283 			goto fragment;
1284 			break;
1285 		case IPPROTO_AH:
1286 		case IPPROTO_ROUTING:
1287 		case IPPROTO_DSTOPTS:
1288 			if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL,
1289 			    NULL, AF_INET6))
1290 				goto shortpkt;
1291 			extoff = off;
1292 			if (proto == IPPROTO_AH)
1293 				off += (ext.ip6e_len + 2) * 4;
1294 			else
1295 				off += (ext.ip6e_len + 1) * 8;
1296 			proto = ext.ip6e_nxt;
1297 			break;
1298 		case IPPROTO_HOPOPTS:
1299 			if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL,
1300 			    NULL, AF_INET6))
1301 				goto shortpkt;
1302 			extoff = off;
1303 			optend = off + (ext.ip6e_len + 1) * 8;
1304 			ooff = off + sizeof(ext);
1305 			do {
1306 				if (!pf_pull_hdr(m, ooff, &opt.ip6o_type,
1307 				    sizeof(opt.ip6o_type), NULL, NULL,
1308 				    AF_INET6))
1309 					goto shortpkt;
1310 				if (opt.ip6o_type == IP6OPT_PAD1) {
1311 					ooff++;
1312 					continue;
1313 				}
1314 				if (!pf_pull_hdr(m, ooff, &opt, sizeof(opt),
1315 				    NULL, NULL, AF_INET6))
1316 					goto shortpkt;
1317 				if (ooff + sizeof(opt) + opt.ip6o_len > optend)
1318 					goto drop;
1319 				if (opt.ip6o_type == IP6OPT_JUMBO)
1320 					goto drop;
1321 				ooff += sizeof(opt) + opt.ip6o_len;
1322 			} while (ooff < optend);
1323 
1324 			off = optend;
1325 			proto = ext.ip6e_nxt;
1326 			break;
1327 		default:
1328 			terminal = 1;
1329 			break;
1330 		}
1331 	} while (!terminal);
1332 
1333 	if (sizeof(struct ip6_hdr) + plen > m->m_pkthdr.len)
1334 		goto shortpkt;
1335 
1336 	return (PF_PASS);
1337 
1338  fragment:
1339 	if (pd->flags & PFDESC_IP_REAS)
1340 		return (PF_DROP);
1341 	if (sizeof(struct ip6_hdr) + plen > m->m_pkthdr.len)
1342 		goto shortpkt;
1343 
1344 	if (!pf_pull_hdr(m, off, &frag, sizeof(frag), NULL, NULL, AF_INET6))
1345 		goto shortpkt;
1346 
1347 	/* Offset now points to data portion. */
1348 	off += sizeof(frag);
1349 
1350 	/* Returns PF_DROP or *m0 is NULL or completely reassembled mbuf. */
1351 	if (pf_reassemble6(m0, h, &frag, off, extoff, reason) != PF_PASS)
1352 		return (PF_DROP);
1353 	m = *m0;
1354 	if (m == NULL)
1355 		return (PF_DROP);
1356 
1357 	pd->flags |= PFDESC_IP_REAS;
1358 	goto again;
1359 
1360  shortpkt:
1361 	REASON_SET(reason, PFRES_SHORT);
1362 	if (r != NULL && r->log)
1363 		PFLOG_PACKET(kif, m, AF_INET6, *reason, r, NULL, NULL, pd, 1);
1364 	return (PF_DROP);
1365 
1366  drop:
1367 	REASON_SET(reason, PFRES_NORM);
1368 	if (r != NULL && r->log)
1369 		PFLOG_PACKET(kif, m, AF_INET6, *reason, r, NULL, NULL, pd, 1);
1370 	return (PF_DROP);
1371 }
1372 #endif /* INET6 */
1373 
1374 int
pf_normalize_tcp(struct pfi_kkif * kif,struct mbuf * m,int ipoff,int off,void * h,struct pf_pdesc * pd)1375 pf_normalize_tcp(struct pfi_kkif *kif, struct mbuf *m, int ipoff,
1376     int off, void *h, struct pf_pdesc *pd)
1377 {
1378 	struct pf_krule	*r, *rm = NULL;
1379 	struct tcphdr	*th = &pd->hdr.tcp;
1380 	int		 rewrite = 0;
1381 	u_short		 reason;
1382 	u_int8_t	 flags;
1383 	sa_family_t	 af = pd->af;
1384 	int		 srs;
1385 
1386 	PF_RULES_RASSERT();
1387 
1388 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1389 	/* Check if there any scrub rules. Lack of scrub rules means enforced
1390 	 * packet normalization operation just like in OpenBSD. */
1391 	srs = (r != NULL);
1392 	while (r != NULL) {
1393 		pf_counter_u64_add(&r->evaluations, 1);
1394 		if (pfi_kkif_match(r->kif, kif) == r->ifnot)
1395 			r = r->skip[PF_SKIP_IFP].ptr;
1396 		else if (r->direction && r->direction != pd->dir)
1397 			r = r->skip[PF_SKIP_DIR].ptr;
1398 		else if (r->af && r->af != af)
1399 			r = r->skip[PF_SKIP_AF].ptr;
1400 		else if (r->proto && r->proto != pd->proto)
1401 			r = r->skip[PF_SKIP_PROTO].ptr;
1402 		else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
1403 		    r->src.neg, kif, M_GETFIB(m)))
1404 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
1405 		else if (r->src.port_op && !pf_match_port(r->src.port_op,
1406 			    r->src.port[0], r->src.port[1], th->th_sport))
1407 			r = r->skip[PF_SKIP_SRC_PORT].ptr;
1408 		else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
1409 		    r->dst.neg, NULL, M_GETFIB(m)))
1410 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
1411 		else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
1412 			    r->dst.port[0], r->dst.port[1], th->th_dport))
1413 			r = r->skip[PF_SKIP_DST_PORT].ptr;
1414 		else if (r->os_fingerprint != PF_OSFP_ANY && !pf_osfp_match(
1415 			    pf_osfp_fingerprint(pd, m, off, th),
1416 			    r->os_fingerprint))
1417 			r = TAILQ_NEXT(r, entries);
1418 		else {
1419 			rm = r;
1420 			break;
1421 		}
1422 	}
1423 
1424 	if (srs) {
1425 		/* With scrub rules present TCP normalization happens only
1426 		 * if one of rules has matched and it's not a "no scrub" rule */
1427 		if (rm == NULL || rm->action == PF_NOSCRUB)
1428 			return (PF_PASS);
1429 
1430 		pf_counter_u64_critical_enter();
1431 		pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1);
1432 		pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len);
1433 		pf_counter_u64_critical_exit();
1434 		pf_rule_to_actions(rm, &pd->act);
1435 	}
1436 
1437 	if (rm && rm->rule_flag & PFRULE_REASSEMBLE_TCP)
1438 		pd->flags |= PFDESC_TCP_NORM;
1439 
1440 	flags = th->th_flags;
1441 	if (flags & TH_SYN) {
1442 		/* Illegal packet */
1443 		if (flags & TH_RST)
1444 			goto tcp_drop;
1445 
1446 		if (flags & TH_FIN)
1447 			goto tcp_drop;
1448 	} else {
1449 		/* Illegal packet */
1450 		if (!(flags & (TH_ACK|TH_RST)))
1451 			goto tcp_drop;
1452 	}
1453 
1454 	if (!(flags & TH_ACK)) {
1455 		/* These flags are only valid if ACK is set */
1456 		if ((flags & TH_FIN) || (flags & TH_PUSH) || (flags & TH_URG))
1457 			goto tcp_drop;
1458 	}
1459 
1460 	/* Check for illegal header length */
1461 	if (th->th_off < (sizeof(struct tcphdr) >> 2))
1462 		goto tcp_drop;
1463 
1464 	/* If flags changed, or reserved data set, then adjust */
1465 	if (flags != th->th_flags || th->th_x2 != 0) {
1466 		u_int16_t	ov, nv;
1467 
1468 		ov = *(u_int16_t *)(&th->th_ack + 1);
1469 		th->th_flags = flags;
1470 		th->th_x2 = 0;
1471 		nv = *(u_int16_t *)(&th->th_ack + 1);
1472 
1473 		th->th_sum = pf_proto_cksum_fixup(m, th->th_sum, ov, nv, 0);
1474 		rewrite = 1;
1475 	}
1476 
1477 	/* Remove urgent pointer, if TH_URG is not set */
1478 	if (!(flags & TH_URG) && th->th_urp) {
1479 		th->th_sum = pf_proto_cksum_fixup(m, th->th_sum, th->th_urp,
1480 		    0, 0);
1481 		th->th_urp = 0;
1482 		rewrite = 1;
1483 	}
1484 
1485 	/* copy back packet headers if we sanitized */
1486 	if (rewrite)
1487 		m_copyback(m, off, sizeof(*th), (caddr_t)th);
1488 
1489 	return (PF_PASS);
1490 
1491  tcp_drop:
1492 	REASON_SET(&reason, PFRES_NORM);
1493 	if (rm != NULL && r->log)
1494 		PFLOG_PACKET(kif, m, AF_INET, reason, r, NULL, NULL, pd, 1);
1495 	return (PF_DROP);
1496 }
1497 
1498 int
pf_normalize_tcp_init(struct mbuf * m,int off,struct pf_pdesc * pd,struct tcphdr * th,struct pf_state_peer * src,struct pf_state_peer * dst)1499 pf_normalize_tcp_init(struct mbuf *m, int off, struct pf_pdesc *pd,
1500     struct tcphdr *th, struct pf_state_peer *src, struct pf_state_peer *dst)
1501 {
1502 	u_int32_t tsval, tsecr;
1503 	u_int8_t hdr[60];
1504 	u_int8_t *opt;
1505 
1506 	KASSERT((src->scrub == NULL),
1507 	    ("pf_normalize_tcp_init: src->scrub != NULL"));
1508 
1509 	src->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT);
1510 	if (src->scrub == NULL)
1511 		return (1);
1512 
1513 	switch (pd->af) {
1514 #ifdef INET
1515 	case AF_INET: {
1516 		struct ip *h = mtod(m, struct ip *);
1517 		src->scrub->pfss_ttl = h->ip_ttl;
1518 		break;
1519 	}
1520 #endif /* INET */
1521 #ifdef INET6
1522 	case AF_INET6: {
1523 		struct ip6_hdr *h = mtod(m, struct ip6_hdr *);
1524 		src->scrub->pfss_ttl = h->ip6_hlim;
1525 		break;
1526 	}
1527 #endif /* INET6 */
1528 	}
1529 
1530 	/*
1531 	 * All normalizations below are only begun if we see the start of
1532 	 * the connections.  They must all set an enabled bit in pfss_flags
1533 	 */
1534 	if ((th->th_flags & TH_SYN) == 0)
1535 		return (0);
1536 
1537 	if (th->th_off > (sizeof(struct tcphdr) >> 2) && src->scrub &&
1538 	    pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) {
1539 		/* Diddle with TCP options */
1540 		int hlen;
1541 		opt = hdr + sizeof(struct tcphdr);
1542 		hlen = (th->th_off << 2) - sizeof(struct tcphdr);
1543 		while (hlen >= TCPOLEN_TIMESTAMP) {
1544 			switch (*opt) {
1545 			case TCPOPT_EOL:	/* FALLTHROUGH */
1546 			case TCPOPT_NOP:
1547 				opt++;
1548 				hlen--;
1549 				break;
1550 			case TCPOPT_TIMESTAMP:
1551 				if (opt[1] >= TCPOLEN_TIMESTAMP) {
1552 					src->scrub->pfss_flags |=
1553 					    PFSS_TIMESTAMP;
1554 					src->scrub->pfss_ts_mod =
1555 					    htonl(arc4random());
1556 
1557 					/* note PFSS_PAWS not set yet */
1558 					memcpy(&tsval, &opt[2],
1559 					    sizeof(u_int32_t));
1560 					memcpy(&tsecr, &opt[6],
1561 					    sizeof(u_int32_t));
1562 					src->scrub->pfss_tsval0 = ntohl(tsval);
1563 					src->scrub->pfss_tsval = ntohl(tsval);
1564 					src->scrub->pfss_tsecr = ntohl(tsecr);
1565 					getmicrouptime(&src->scrub->pfss_last);
1566 				}
1567 				/* FALLTHROUGH */
1568 			default:
1569 				hlen -= MAX(opt[1], 2);
1570 				opt += MAX(opt[1], 2);
1571 				break;
1572 			}
1573 		}
1574 	}
1575 
1576 	return (0);
1577 }
1578 
1579 void
pf_normalize_tcp_cleanup(struct pf_kstate * state)1580 pf_normalize_tcp_cleanup(struct pf_kstate *state)
1581 {
1582 	/* XXX Note: this also cleans up SCTP. */
1583 	uma_zfree(V_pf_state_scrub_z, state->src.scrub);
1584 	uma_zfree(V_pf_state_scrub_z, state->dst.scrub);
1585 
1586 	/* Someday... flush the TCP segment reassembly descriptors. */
1587 }
1588 int
pf_normalize_sctp_init(struct mbuf * m,int off,struct pf_pdesc * pd,struct pf_state_peer * src,struct pf_state_peer * dst)1589 pf_normalize_sctp_init(struct mbuf *m, int off, struct pf_pdesc *pd,
1590 	    struct pf_state_peer *src, struct pf_state_peer *dst)
1591 {
1592 	src->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT);
1593 	if (src->scrub == NULL)
1594 		return (1);
1595 
1596 	dst->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT);
1597 	if (dst->scrub == NULL) {
1598 		uma_zfree(V_pf_state_scrub_z, src);
1599 		return (1);
1600 	}
1601 
1602 	dst->scrub->pfss_v_tag = pd->sctp_initiate_tag;
1603 
1604 	return (0);
1605 }
1606 
1607 int
pf_normalize_tcp_stateful(struct mbuf * m,int off,struct pf_pdesc * pd,u_short * reason,struct tcphdr * th,struct pf_kstate * state,struct pf_state_peer * src,struct pf_state_peer * dst,int * writeback)1608 pf_normalize_tcp_stateful(struct mbuf *m, int off, struct pf_pdesc *pd,
1609     u_short *reason, struct tcphdr *th, struct pf_kstate *state,
1610     struct pf_state_peer *src, struct pf_state_peer *dst, int *writeback)
1611 {
1612 	struct timeval uptime;
1613 	u_int32_t tsval, tsecr;
1614 	u_int tsval_from_last;
1615 	u_int8_t hdr[60];
1616 	u_int8_t *opt;
1617 	int copyback = 0;
1618 	int got_ts = 0;
1619 	size_t startoff;
1620 
1621 	KASSERT((src->scrub || dst->scrub),
1622 	    ("%s: src->scrub && dst->scrub!", __func__));
1623 
1624 	/*
1625 	 * Enforce the minimum TTL seen for this connection.  Negate a common
1626 	 * technique to evade an intrusion detection system and confuse
1627 	 * firewall state code.
1628 	 */
1629 	switch (pd->af) {
1630 #ifdef INET
1631 	case AF_INET: {
1632 		if (src->scrub) {
1633 			struct ip *h = mtod(m, struct ip *);
1634 			if (h->ip_ttl > src->scrub->pfss_ttl)
1635 				src->scrub->pfss_ttl = h->ip_ttl;
1636 			h->ip_ttl = src->scrub->pfss_ttl;
1637 		}
1638 		break;
1639 	}
1640 #endif /* INET */
1641 #ifdef INET6
1642 	case AF_INET6: {
1643 		if (src->scrub) {
1644 			struct ip6_hdr *h = mtod(m, struct ip6_hdr *);
1645 			if (h->ip6_hlim > src->scrub->pfss_ttl)
1646 				src->scrub->pfss_ttl = h->ip6_hlim;
1647 			h->ip6_hlim = src->scrub->pfss_ttl;
1648 		}
1649 		break;
1650 	}
1651 #endif /* INET6 */
1652 	}
1653 
1654 	if (th->th_off > (sizeof(struct tcphdr) >> 2) &&
1655 	    ((src->scrub && (src->scrub->pfss_flags & PFSS_TIMESTAMP)) ||
1656 	    (dst->scrub && (dst->scrub->pfss_flags & PFSS_TIMESTAMP))) &&
1657 	    pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) {
1658 		/* Diddle with TCP options */
1659 		int hlen;
1660 		opt = hdr + sizeof(struct tcphdr);
1661 		hlen = (th->th_off << 2) - sizeof(struct tcphdr);
1662 		while (hlen >= TCPOLEN_TIMESTAMP) {
1663 			startoff = opt - (hdr + sizeof(struct tcphdr));
1664 			switch (*opt) {
1665 			case TCPOPT_EOL:	/* FALLTHROUGH */
1666 			case TCPOPT_NOP:
1667 				opt++;
1668 				hlen--;
1669 				break;
1670 			case TCPOPT_TIMESTAMP:
1671 				/* Modulate the timestamps.  Can be used for
1672 				 * NAT detection, OS uptime determination or
1673 				 * reboot detection.
1674 				 */
1675 
1676 				if (got_ts) {
1677 					/* Huh?  Multiple timestamps!? */
1678 					if (V_pf_status.debug >= PF_DEBUG_MISC) {
1679 						DPFPRINTF(("multiple TS??\n"));
1680 						pf_print_state(state);
1681 						printf("\n");
1682 					}
1683 					REASON_SET(reason, PFRES_TS);
1684 					return (PF_DROP);
1685 				}
1686 				if (opt[1] >= TCPOLEN_TIMESTAMP) {
1687 					memcpy(&tsval, &opt[2],
1688 					    sizeof(u_int32_t));
1689 					if (tsval && src->scrub &&
1690 					    (src->scrub->pfss_flags &
1691 					    PFSS_TIMESTAMP)) {
1692 						tsval = ntohl(tsval);
1693 						pf_patch_32_unaligned(m,
1694 						    &th->th_sum,
1695 						    &opt[2],
1696 						    htonl(tsval +
1697 						    src->scrub->pfss_ts_mod),
1698 						    PF_ALGNMNT(startoff),
1699 						    0);
1700 						copyback = 1;
1701 					}
1702 
1703 					/* Modulate TS reply iff valid (!0) */
1704 					memcpy(&tsecr, &opt[6],
1705 					    sizeof(u_int32_t));
1706 					if (tsecr && dst->scrub &&
1707 					    (dst->scrub->pfss_flags &
1708 					    PFSS_TIMESTAMP)) {
1709 						tsecr = ntohl(tsecr)
1710 						    - dst->scrub->pfss_ts_mod;
1711 						pf_patch_32_unaligned(m,
1712 						    &th->th_sum,
1713 						    &opt[6],
1714 						    htonl(tsecr),
1715 						    PF_ALGNMNT(startoff),
1716 						    0);
1717 						copyback = 1;
1718 					}
1719 					got_ts = 1;
1720 				}
1721 				/* FALLTHROUGH */
1722 			default:
1723 				hlen -= MAX(opt[1], 2);
1724 				opt += MAX(opt[1], 2);
1725 				break;
1726 			}
1727 		}
1728 		if (copyback) {
1729 			/* Copyback the options, caller copys back header */
1730 			*writeback = 1;
1731 			m_copyback(m, off + sizeof(struct tcphdr),
1732 			    (th->th_off << 2) - sizeof(struct tcphdr), hdr +
1733 			    sizeof(struct tcphdr));
1734 		}
1735 	}
1736 
1737 	/*
1738 	 * Must invalidate PAWS checks on connections idle for too long.
1739 	 * The fastest allowed timestamp clock is 1ms.  That turns out to
1740 	 * be about 24 days before it wraps.  XXX Right now our lowerbound
1741 	 * TS echo check only works for the first 12 days of a connection
1742 	 * when the TS has exhausted half its 32bit space
1743 	 */
1744 #define TS_MAX_IDLE	(24*24*60*60)
1745 #define TS_MAX_CONN	(12*24*60*60)	/* XXX remove when better tsecr check */
1746 
1747 	getmicrouptime(&uptime);
1748 	if (src->scrub && (src->scrub->pfss_flags & PFSS_PAWS) &&
1749 	    (uptime.tv_sec - src->scrub->pfss_last.tv_sec > TS_MAX_IDLE ||
1750 	    time_uptime - state->creation > TS_MAX_CONN))  {
1751 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
1752 			DPFPRINTF(("src idled out of PAWS\n"));
1753 			pf_print_state(state);
1754 			printf("\n");
1755 		}
1756 		src->scrub->pfss_flags = (src->scrub->pfss_flags & ~PFSS_PAWS)
1757 		    | PFSS_PAWS_IDLED;
1758 	}
1759 	if (dst->scrub && (dst->scrub->pfss_flags & PFSS_PAWS) &&
1760 	    uptime.tv_sec - dst->scrub->pfss_last.tv_sec > TS_MAX_IDLE) {
1761 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
1762 			DPFPRINTF(("dst idled out of PAWS\n"));
1763 			pf_print_state(state);
1764 			printf("\n");
1765 		}
1766 		dst->scrub->pfss_flags = (dst->scrub->pfss_flags & ~PFSS_PAWS)
1767 		    | PFSS_PAWS_IDLED;
1768 	}
1769 
1770 	if (got_ts && src->scrub && dst->scrub &&
1771 	    (src->scrub->pfss_flags & PFSS_PAWS) &&
1772 	    (dst->scrub->pfss_flags & PFSS_PAWS)) {
1773 		/* Validate that the timestamps are "in-window".
1774 		 * RFC1323 describes TCP Timestamp options that allow
1775 		 * measurement of RTT (round trip time) and PAWS
1776 		 * (protection against wrapped sequence numbers).  PAWS
1777 		 * gives us a set of rules for rejecting packets on
1778 		 * long fat pipes (packets that were somehow delayed
1779 		 * in transit longer than the time it took to send the
1780 		 * full TCP sequence space of 4Gb).  We can use these
1781 		 * rules and infer a few others that will let us treat
1782 		 * the 32bit timestamp and the 32bit echoed timestamp
1783 		 * as sequence numbers to prevent a blind attacker from
1784 		 * inserting packets into a connection.
1785 		 *
1786 		 * RFC1323 tells us:
1787 		 *  - The timestamp on this packet must be greater than
1788 		 *    or equal to the last value echoed by the other
1789 		 *    endpoint.  The RFC says those will be discarded
1790 		 *    since it is a dup that has already been acked.
1791 		 *    This gives us a lowerbound on the timestamp.
1792 		 *        timestamp >= other last echoed timestamp
1793 		 *  - The timestamp will be less than or equal to
1794 		 *    the last timestamp plus the time between the
1795 		 *    last packet and now.  The RFC defines the max
1796 		 *    clock rate as 1ms.  We will allow clocks to be
1797 		 *    up to 10% fast and will allow a total difference
1798 		 *    or 30 seconds due to a route change.  And this
1799 		 *    gives us an upperbound on the timestamp.
1800 		 *        timestamp <= last timestamp + max ticks
1801 		 *    We have to be careful here.  Windows will send an
1802 		 *    initial timestamp of zero and then initialize it
1803 		 *    to a random value after the 3whs; presumably to
1804 		 *    avoid a DoS by having to call an expensive RNG
1805 		 *    during a SYN flood.  Proof MS has at least one
1806 		 *    good security geek.
1807 		 *
1808 		 *  - The TCP timestamp option must also echo the other
1809 		 *    endpoints timestamp.  The timestamp echoed is the
1810 		 *    one carried on the earliest unacknowledged segment
1811 		 *    on the left edge of the sequence window.  The RFC
1812 		 *    states that the host will reject any echoed
1813 		 *    timestamps that were larger than any ever sent.
1814 		 *    This gives us an upperbound on the TS echo.
1815 		 *        tescr <= largest_tsval
1816 		 *  - The lowerbound on the TS echo is a little more
1817 		 *    tricky to determine.  The other endpoint's echoed
1818 		 *    values will not decrease.  But there may be
1819 		 *    network conditions that re-order packets and
1820 		 *    cause our view of them to decrease.  For now the
1821 		 *    only lowerbound we can safely determine is that
1822 		 *    the TS echo will never be less than the original
1823 		 *    TS.  XXX There is probably a better lowerbound.
1824 		 *    Remove TS_MAX_CONN with better lowerbound check.
1825 		 *        tescr >= other original TS
1826 		 *
1827 		 * It is also important to note that the fastest
1828 		 * timestamp clock of 1ms will wrap its 32bit space in
1829 		 * 24 days.  So we just disable TS checking after 24
1830 		 * days of idle time.  We actually must use a 12d
1831 		 * connection limit until we can come up with a better
1832 		 * lowerbound to the TS echo check.
1833 		 */
1834 		struct timeval delta_ts;
1835 		int ts_fudge;
1836 
1837 		/*
1838 		 * PFTM_TS_DIFF is how many seconds of leeway to allow
1839 		 * a host's timestamp.  This can happen if the previous
1840 		 * packet got delayed in transit for much longer than
1841 		 * this packet.
1842 		 */
1843 		if ((ts_fudge = state->rule.ptr->timeout[PFTM_TS_DIFF]) == 0)
1844 			ts_fudge = V_pf_default_rule.timeout[PFTM_TS_DIFF];
1845 
1846 		/* Calculate max ticks since the last timestamp */
1847 #define TS_MAXFREQ	1100		/* RFC max TS freq of 1Khz + 10% skew */
1848 #define TS_MICROSECS	1000000		/* microseconds per second */
1849 		delta_ts = uptime;
1850 		timevalsub(&delta_ts, &src->scrub->pfss_last);
1851 		tsval_from_last = (delta_ts.tv_sec + ts_fudge) * TS_MAXFREQ;
1852 		tsval_from_last += delta_ts.tv_usec / (TS_MICROSECS/TS_MAXFREQ);
1853 
1854 		if ((src->state >= TCPS_ESTABLISHED &&
1855 		    dst->state >= TCPS_ESTABLISHED) &&
1856 		    (SEQ_LT(tsval, dst->scrub->pfss_tsecr) ||
1857 		    SEQ_GT(tsval, src->scrub->pfss_tsval + tsval_from_last) ||
1858 		    (tsecr && (SEQ_GT(tsecr, dst->scrub->pfss_tsval) ||
1859 		    SEQ_LT(tsecr, dst->scrub->pfss_tsval0))))) {
1860 			/* Bad RFC1323 implementation or an insertion attack.
1861 			 *
1862 			 * - Solaris 2.6 and 2.7 are known to send another ACK
1863 			 *   after the FIN,FIN|ACK,ACK closing that carries
1864 			 *   an old timestamp.
1865 			 */
1866 
1867 			DPFPRINTF(("Timestamp failed %c%c%c%c\n",
1868 			    SEQ_LT(tsval, dst->scrub->pfss_tsecr) ? '0' : ' ',
1869 			    SEQ_GT(tsval, src->scrub->pfss_tsval +
1870 			    tsval_from_last) ? '1' : ' ',
1871 			    SEQ_GT(tsecr, dst->scrub->pfss_tsval) ? '2' : ' ',
1872 			    SEQ_LT(tsecr, dst->scrub->pfss_tsval0)? '3' : ' '));
1873 			DPFPRINTF((" tsval: %u  tsecr: %u  +ticks: %u  "
1874 			    "idle: %jus %lums\n",
1875 			    tsval, tsecr, tsval_from_last,
1876 			    (uintmax_t)delta_ts.tv_sec,
1877 			    delta_ts.tv_usec / 1000));
1878 			DPFPRINTF((" src->tsval: %u  tsecr: %u\n",
1879 			    src->scrub->pfss_tsval, src->scrub->pfss_tsecr));
1880 			DPFPRINTF((" dst->tsval: %u  tsecr: %u  tsval0: %u"
1881 			    "\n", dst->scrub->pfss_tsval,
1882 			    dst->scrub->pfss_tsecr, dst->scrub->pfss_tsval0));
1883 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
1884 				pf_print_state(state);
1885 				pf_print_flags(th->th_flags);
1886 				printf("\n");
1887 			}
1888 			REASON_SET(reason, PFRES_TS);
1889 			return (PF_DROP);
1890 		}
1891 
1892 		/* XXX I'd really like to require tsecr but it's optional */
1893 
1894 	} else if (!got_ts && (th->th_flags & TH_RST) == 0 &&
1895 	    ((src->state == TCPS_ESTABLISHED && dst->state == TCPS_ESTABLISHED)
1896 	    || pd->p_len > 0 || (th->th_flags & TH_SYN)) &&
1897 	    src->scrub && dst->scrub &&
1898 	    (src->scrub->pfss_flags & PFSS_PAWS) &&
1899 	    (dst->scrub->pfss_flags & PFSS_PAWS)) {
1900 		/* Didn't send a timestamp.  Timestamps aren't really useful
1901 		 * when:
1902 		 *  - connection opening or closing (often not even sent).
1903 		 *    but we must not let an attacker to put a FIN on a
1904 		 *    data packet to sneak it through our ESTABLISHED check.
1905 		 *  - on a TCP reset.  RFC suggests not even looking at TS.
1906 		 *  - on an empty ACK.  The TS will not be echoed so it will
1907 		 *    probably not help keep the RTT calculation in sync and
1908 		 *    there isn't as much danger when the sequence numbers
1909 		 *    got wrapped.  So some stacks don't include TS on empty
1910 		 *    ACKs :-(
1911 		 *
1912 		 * To minimize the disruption to mostly RFC1323 conformant
1913 		 * stacks, we will only require timestamps on data packets.
1914 		 *
1915 		 * And what do ya know, we cannot require timestamps on data
1916 		 * packets.  There appear to be devices that do legitimate
1917 		 * TCP connection hijacking.  There are HTTP devices that allow
1918 		 * a 3whs (with timestamps) and then buffer the HTTP request.
1919 		 * If the intermediate device has the HTTP response cache, it
1920 		 * will spoof the response but not bother timestamping its
1921 		 * packets.  So we can look for the presence of a timestamp in
1922 		 * the first data packet and if there, require it in all future
1923 		 * packets.
1924 		 */
1925 
1926 		if (pd->p_len > 0 && (src->scrub->pfss_flags & PFSS_DATA_TS)) {
1927 			/*
1928 			 * Hey!  Someone tried to sneak a packet in.  Or the
1929 			 * stack changed its RFC1323 behavior?!?!
1930 			 */
1931 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
1932 				DPFPRINTF(("Did not receive expected RFC1323 "
1933 				    "timestamp\n"));
1934 				pf_print_state(state);
1935 				pf_print_flags(th->th_flags);
1936 				printf("\n");
1937 			}
1938 			REASON_SET(reason, PFRES_TS);
1939 			return (PF_DROP);
1940 		}
1941 	}
1942 
1943 	/*
1944 	 * We will note if a host sends his data packets with or without
1945 	 * timestamps.  And require all data packets to contain a timestamp
1946 	 * if the first does.  PAWS implicitly requires that all data packets be
1947 	 * timestamped.  But I think there are middle-man devices that hijack
1948 	 * TCP streams immediately after the 3whs and don't timestamp their
1949 	 * packets (seen in a WWW accelerator or cache).
1950 	 */
1951 	if (pd->p_len > 0 && src->scrub && (src->scrub->pfss_flags &
1952 	    (PFSS_TIMESTAMP|PFSS_DATA_TS|PFSS_DATA_NOTS)) == PFSS_TIMESTAMP) {
1953 		if (got_ts)
1954 			src->scrub->pfss_flags |= PFSS_DATA_TS;
1955 		else {
1956 			src->scrub->pfss_flags |= PFSS_DATA_NOTS;
1957 			if (V_pf_status.debug >= PF_DEBUG_MISC && dst->scrub &&
1958 			    (dst->scrub->pfss_flags & PFSS_TIMESTAMP)) {
1959 				/* Don't warn if other host rejected RFC1323 */
1960 				DPFPRINTF(("Broken RFC1323 stack did not "
1961 				    "timestamp data packet. Disabled PAWS "
1962 				    "security.\n"));
1963 				pf_print_state(state);
1964 				pf_print_flags(th->th_flags);
1965 				printf("\n");
1966 			}
1967 		}
1968 	}
1969 
1970 	/*
1971 	 * Update PAWS values
1972 	 */
1973 	if (got_ts && src->scrub && PFSS_TIMESTAMP == (src->scrub->pfss_flags &
1974 	    (PFSS_PAWS_IDLED|PFSS_TIMESTAMP))) {
1975 		getmicrouptime(&src->scrub->pfss_last);
1976 		if (SEQ_GEQ(tsval, src->scrub->pfss_tsval) ||
1977 		    (src->scrub->pfss_flags & PFSS_PAWS) == 0)
1978 			src->scrub->pfss_tsval = tsval;
1979 
1980 		if (tsecr) {
1981 			if (SEQ_GEQ(tsecr, src->scrub->pfss_tsecr) ||
1982 			    (src->scrub->pfss_flags & PFSS_PAWS) == 0)
1983 				src->scrub->pfss_tsecr = tsecr;
1984 
1985 			if ((src->scrub->pfss_flags & PFSS_PAWS) == 0 &&
1986 			    (SEQ_LT(tsval, src->scrub->pfss_tsval0) ||
1987 			    src->scrub->pfss_tsval0 == 0)) {
1988 				/* tsval0 MUST be the lowest timestamp */
1989 				src->scrub->pfss_tsval0 = tsval;
1990 			}
1991 
1992 			/* Only fully initialized after a TS gets echoed */
1993 			if ((src->scrub->pfss_flags & PFSS_PAWS) == 0)
1994 				src->scrub->pfss_flags |= PFSS_PAWS;
1995 		}
1996 	}
1997 
1998 	/* I have a dream....  TCP segment reassembly.... */
1999 	return (0);
2000 }
2001 
2002 int
pf_normalize_mss(struct mbuf * m,int off,struct pf_pdesc * pd)2003 pf_normalize_mss(struct mbuf *m, int off, struct pf_pdesc *pd)
2004 {
2005 	struct tcphdr	*th = &pd->hdr.tcp;
2006 	u_int16_t	*mss;
2007 	int		 thoff;
2008 	int		 opt, cnt, optlen = 0;
2009 	u_char		 opts[TCP_MAXOLEN];
2010 	u_char		*optp = opts;
2011 	size_t		 startoff;
2012 
2013 	thoff = th->th_off << 2;
2014 	cnt = thoff - sizeof(struct tcphdr);
2015 
2016 	if (cnt > 0 && !pf_pull_hdr(m, off + sizeof(*th), opts, cnt,
2017 	    NULL, NULL, pd->af))
2018 		return (0);
2019 
2020 	for (; cnt > 0; cnt -= optlen, optp += optlen) {
2021 		startoff = optp - opts;
2022 		opt = optp[0];
2023 		if (opt == TCPOPT_EOL)
2024 			break;
2025 		if (opt == TCPOPT_NOP)
2026 			optlen = 1;
2027 		else {
2028 			if (cnt < 2)
2029 				break;
2030 			optlen = optp[1];
2031 			if (optlen < 2 || optlen > cnt)
2032 				break;
2033 		}
2034 		switch (opt) {
2035 		case TCPOPT_MAXSEG:
2036 			mss = (u_int16_t *)(optp + 2);
2037 			if ((ntohs(*mss)) > pd->act.max_mss) {
2038 				pf_patch_16_unaligned(m,
2039 				    &th->th_sum,
2040 				    mss, htons(pd->act.max_mss),
2041 				    PF_ALGNMNT(startoff),
2042 				    0);
2043 				m_copyback(m, off + sizeof(*th),
2044 				    thoff - sizeof(*th), opts);
2045 				m_copyback(m, off, sizeof(*th), (caddr_t)th);
2046 			}
2047 			break;
2048 		default:
2049 			break;
2050 		}
2051 	}
2052 
2053 	return (0);
2054 }
2055 
2056 static int
pf_scan_sctp(struct mbuf * m,int ipoff,int off,struct pf_pdesc * pd,struct pfi_kkif * kif)2057 pf_scan_sctp(struct mbuf *m, int ipoff, int off, struct pf_pdesc *pd,
2058     struct pfi_kkif *kif)
2059 {
2060 	struct sctp_chunkhdr ch = { };
2061 	int chunk_off = sizeof(struct sctphdr);
2062 	int chunk_start;
2063 	int ret;
2064 
2065 	while (off + chunk_off < pd->tot_len) {
2066 		if (!pf_pull_hdr(m, off + chunk_off, &ch, sizeof(ch), NULL,
2067 		    NULL, pd->af))
2068 			return (PF_DROP);
2069 
2070 		/* Length includes the header, this must be at least 4. */
2071 		if (ntohs(ch.chunk_length) < 4)
2072 			return (PF_DROP);
2073 
2074 		chunk_start = chunk_off;
2075 		chunk_off += roundup(ntohs(ch.chunk_length), 4);
2076 
2077 		switch (ch.chunk_type) {
2078 		case SCTP_INITIATION:
2079 		case SCTP_INITIATION_ACK: {
2080 			struct sctp_init_chunk init;
2081 
2082 			if (!pf_pull_hdr(m, off + chunk_start, &init,
2083 			    sizeof(init), NULL, NULL, pd->af))
2084 				return (PF_DROP);
2085 
2086 			/*
2087 			 * RFC 9620, Section 3.3.2, "The Initiate Tag is allowed to have
2088 			 * any value except 0."
2089 			 */
2090 			if (init.init.initiate_tag == 0)
2091 				return (PF_DROP);
2092 			if (init.init.num_inbound_streams == 0)
2093 				return (PF_DROP);
2094 			if (init.init.num_outbound_streams == 0)
2095 				return (PF_DROP);
2096 			if (ntohl(init.init.a_rwnd) < SCTP_MIN_RWND)
2097 				return (PF_DROP);
2098 
2099 			/*
2100 			 * RFC 9260, Section 3.1, INIT chunks MUST have zero
2101 			 * verification tag.
2102 			 */
2103 			if (ch.chunk_type == SCTP_INITIATION &&
2104 			    pd->hdr.sctp.v_tag != 0)
2105 				return (PF_DROP);
2106 
2107 			pd->sctp_initiate_tag = init.init.initiate_tag;
2108 
2109 			if (ch.chunk_type == SCTP_INITIATION)
2110 				pd->sctp_flags |= PFDESC_SCTP_INIT;
2111 			else
2112 				pd->sctp_flags |= PFDESC_SCTP_INIT_ACK;
2113 
2114 			ret = pf_multihome_scan_init(m, off + chunk_start,
2115 			    ntohs(init.ch.chunk_length), pd, kif);
2116 			if (ret != PF_PASS)
2117 				return (ret);
2118 
2119 			break;
2120 		}
2121 		case SCTP_ABORT_ASSOCIATION:
2122 			pd->sctp_flags |= PFDESC_SCTP_ABORT;
2123 			break;
2124 		case SCTP_SHUTDOWN:
2125 		case SCTP_SHUTDOWN_ACK:
2126 			pd->sctp_flags |= PFDESC_SCTP_SHUTDOWN;
2127 			break;
2128 		case SCTP_SHUTDOWN_COMPLETE:
2129 			pd->sctp_flags |= PFDESC_SCTP_SHUTDOWN_COMPLETE;
2130 			break;
2131 		case SCTP_COOKIE_ECHO:
2132 			pd->sctp_flags |= PFDESC_SCTP_COOKIE;
2133 			break;
2134 		case SCTP_COOKIE_ACK:
2135 			pd->sctp_flags |= PFDESC_SCTP_COOKIE_ACK;
2136 			break;
2137 		case SCTP_DATA:
2138 			pd->sctp_flags |= PFDESC_SCTP_DATA;
2139 			break;
2140 		case SCTP_HEARTBEAT_REQUEST:
2141 			pd->sctp_flags |= PFDESC_SCTP_HEARTBEAT;
2142 			break;
2143 		case SCTP_HEARTBEAT_ACK:
2144 			pd->sctp_flags |= PFDESC_SCTP_HEARTBEAT_ACK;
2145 			break;
2146 		case SCTP_ASCONF:
2147 			pd->sctp_flags |= PFDESC_SCTP_ASCONF;
2148 
2149 			ret = pf_multihome_scan_asconf(m, off + chunk_start,
2150 			    ntohs(ch.chunk_length), pd, kif);
2151 			if (ret != PF_PASS)
2152 				return (ret);
2153 			break;
2154 		default:
2155 			pd->sctp_flags |= PFDESC_SCTP_OTHER;
2156 			break;
2157 		}
2158 	}
2159 
2160 	/* Validate chunk lengths vs. packet length. */
2161 	if (off + chunk_off != pd->tot_len)
2162 		return (PF_DROP);
2163 
2164 	/*
2165 	 * INIT, INIT_ACK or SHUTDOWN_COMPLETE chunks must always be the only
2166 	 * one in a packet.
2167 	 */
2168 	if ((pd->sctp_flags & PFDESC_SCTP_INIT) &&
2169 	    (pd->sctp_flags & ~PFDESC_SCTP_INIT))
2170 		return (PF_DROP);
2171 	if ((pd->sctp_flags & PFDESC_SCTP_INIT_ACK) &&
2172 	    (pd->sctp_flags & ~PFDESC_SCTP_INIT_ACK))
2173 		return (PF_DROP);
2174 	if ((pd->sctp_flags & PFDESC_SCTP_SHUTDOWN_COMPLETE) &&
2175 	    (pd->sctp_flags & ~PFDESC_SCTP_SHUTDOWN_COMPLETE))
2176 		return (PF_DROP);
2177 	if ((pd->sctp_flags & PFDESC_SCTP_ABORT) &&
2178 	    (pd->sctp_flags & PFDESC_SCTP_DATA)) {
2179 		/*
2180 		 * RFC4960 3.3.7: DATA chunks MUST NOT be
2181 		 * bundled with ABORT.
2182 		 */
2183 		return (PF_DROP);
2184 	}
2185 
2186 	return (PF_PASS);
2187 }
2188 
2189 int
pf_normalize_sctp(int dir,struct pfi_kkif * kif,struct mbuf * m,int ipoff,int off,void * h,struct pf_pdesc * pd)2190 pf_normalize_sctp(int dir, struct pfi_kkif *kif, struct mbuf *m, int ipoff,
2191     int off, void *h, struct pf_pdesc *pd)
2192 {
2193 	struct pf_krule	*r, *rm = NULL;
2194 	struct sctphdr	*sh = &pd->hdr.sctp;
2195 	u_short		 reason;
2196 	sa_family_t	 af = pd->af;
2197 	int		 srs;
2198 
2199 	PF_RULES_RASSERT();
2200 
2201 	/* Unconditionally scan the SCTP packet, because we need to look for
2202 	 * things like shutdown and asconf chunks. */
2203 	if (pf_scan_sctp(m, ipoff, off, pd, kif) != PF_PASS)
2204 		goto sctp_drop;
2205 
2206 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
2207 	/* Check if there any scrub rules. Lack of scrub rules means enforced
2208 	 * packet normalization operation just like in OpenBSD. */
2209 	srs = (r != NULL);
2210 	while (r != NULL) {
2211 		pf_counter_u64_add(&r->evaluations, 1);
2212 		if (pfi_kkif_match(r->kif, kif) == r->ifnot)
2213 			r = r->skip[PF_SKIP_IFP].ptr;
2214 		else if (r->direction && r->direction != dir)
2215 			r = r->skip[PF_SKIP_DIR].ptr;
2216 		else if (r->af && r->af != af)
2217 			r = r->skip[PF_SKIP_AF].ptr;
2218 		else if (r->proto && r->proto != pd->proto)
2219 			r = r->skip[PF_SKIP_PROTO].ptr;
2220 		else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
2221 		    r->src.neg, kif, M_GETFIB(m)))
2222 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
2223 		else if (r->src.port_op && !pf_match_port(r->src.port_op,
2224 			    r->src.port[0], r->src.port[1], sh->src_port))
2225 			r = r->skip[PF_SKIP_SRC_PORT].ptr;
2226 		else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
2227 		    r->dst.neg, NULL, M_GETFIB(m)))
2228 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
2229 		else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
2230 			    r->dst.port[0], r->dst.port[1], sh->dest_port))
2231 			r = r->skip[PF_SKIP_DST_PORT].ptr;
2232 		else {
2233 			rm = r;
2234 			break;
2235 		}
2236 	}
2237 
2238 	if (srs) {
2239 		/* With scrub rules present SCTP normalization happens only
2240 		 * if one of rules has matched and it's not a "no scrub" rule */
2241 		if (rm == NULL || rm->action == PF_NOSCRUB)
2242 			return (PF_PASS);
2243 
2244 		pf_counter_u64_critical_enter();
2245 		pf_counter_u64_add_protected(&r->packets[dir == PF_OUT], 1);
2246 		pf_counter_u64_add_protected(&r->bytes[dir == PF_OUT], pd->tot_len);
2247 		pf_counter_u64_critical_exit();
2248 	}
2249 
2250 	/* Verify we're a multiple of 4 bytes long */
2251 	if ((pd->tot_len - off - sizeof(struct sctphdr)) % 4)
2252 		goto sctp_drop;
2253 
2254 	/* INIT chunk needs to be the only chunk */
2255 	if (pd->sctp_flags & PFDESC_SCTP_INIT)
2256 		if (pd->sctp_flags & ~PFDESC_SCTP_INIT)
2257 			goto sctp_drop;
2258 
2259 	return (PF_PASS);
2260 
2261 sctp_drop:
2262 	REASON_SET(&reason, PFRES_NORM);
2263 	if (rm != NULL && r->log)
2264 		PFLOG_PACKET(kif, m, AF_INET, reason, r, NULL, NULL, pd,
2265 		    1);
2266 
2267 	return (PF_DROP);
2268 }
2269 
2270 #ifdef INET
2271 void
pf_scrub_ip(struct mbuf ** m0,struct pf_pdesc * pd)2272 pf_scrub_ip(struct mbuf **m0, struct pf_pdesc *pd)
2273 {
2274 	struct mbuf		*m = *m0;
2275 	struct ip		*h = mtod(m, struct ip *);
2276 
2277 	/* Clear IP_DF if no-df was requested */
2278 	if (pd->act.flags & PFSTATE_NODF && h->ip_off & htons(IP_DF)) {
2279 		u_int16_t ip_off = h->ip_off;
2280 
2281 		h->ip_off &= htons(~IP_DF);
2282 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
2283 	}
2284 
2285 	/* Enforce a minimum ttl, may cause endless packet loops */
2286 	if (pd->act.min_ttl && h->ip_ttl < pd->act.min_ttl) {
2287 		u_int16_t ip_ttl = h->ip_ttl;
2288 
2289 		h->ip_ttl = pd->act.min_ttl;
2290 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_ttl, h->ip_ttl, 0);
2291 	}
2292 
2293 	/* Enforce tos */
2294 	if (pd->act.flags & PFSTATE_SETTOS) {
2295 		u_int16_t	ov, nv;
2296 
2297 		ov = *(u_int16_t *)h;
2298 		h->ip_tos = pd->act.set_tos | (h->ip_tos & IPTOS_ECN_MASK);
2299 		nv = *(u_int16_t *)h;
2300 
2301 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ov, nv, 0);
2302 	}
2303 
2304 	/* random-id, but not for fragments */
2305 	if (pd->act.flags & PFSTATE_RANDOMID && !(h->ip_off & ~htons(IP_DF))) {
2306 		uint16_t ip_id = h->ip_id;
2307 
2308 		ip_fillid(h);
2309 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_id, h->ip_id, 0);
2310 	}
2311 }
2312 #endif /* INET */
2313 
2314 #ifdef INET6
2315 void
pf_scrub_ip6(struct mbuf ** m0,struct pf_pdesc * pd)2316 pf_scrub_ip6(struct mbuf **m0, struct pf_pdesc *pd)
2317 {
2318 	struct mbuf		*m = *m0;
2319 	struct ip6_hdr		*h = mtod(m, struct ip6_hdr *);
2320 
2321 	/* Enforce a minimum ttl, may cause endless packet loops */
2322 	if (pd->act.min_ttl && h->ip6_hlim < pd->act.min_ttl)
2323 		h->ip6_hlim = pd->act.min_ttl;
2324 
2325 	/* Enforce tos. Set traffic class bits */
2326 	if (pd->act.flags & PFSTATE_SETTOS) {
2327 		h->ip6_flow &= IPV6_FLOWLABEL_MASK | IPV6_VERSION_MASK;
2328 		h->ip6_flow |= htonl((pd->act.set_tos | IPV6_ECN(h)) << 20);
2329 	}
2330 }
2331 #endif
2332