xref: /freebsd/sys/netpfil/pf/pf_norm.c (revision 4bb3b365776458bd8f710e40f97e2c68994e3306)
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/if_var.h>
46 #include <net/if_private.h>
47 #include <net/vnet.h>
48 #include <net/pfvar.h>
49 #include <net/if_pflog.h>
50 
51 #include <netinet/in.h>
52 #include <netinet/ip.h>
53 #include <netinet/ip_var.h>
54 #include <netinet6/in6_var.h>
55 #include <netinet6/nd6.h>
56 #include <netinet6/ip6_var.h>
57 #include <netinet6/scope6_var.h>
58 #include <netinet/tcp.h>
59 #include <netinet/tcp_fsm.h>
60 #include <netinet/tcp_seq.h>
61 #include <netinet/sctp_constants.h>
62 #include <netinet/sctp_header.h>
63 
64 #ifdef INET6
65 #include <netinet/ip6.h>
66 #endif /* INET6 */
67 
68 struct pf_frent {
69 	TAILQ_ENTRY(pf_frent)	fr_next;
70 	struct mbuf	*fe_m;
71 	uint16_t	fe_hdrlen;	/* ipv4 header length with ip options
72 					   ipv6, extension, fragment header */
73 	uint16_t	fe_extoff;	/* last extension header offset or 0 */
74 	uint16_t	fe_len;		/* fragment length */
75 	uint16_t	fe_off;		/* fragment offset */
76 	uint16_t	fe_mff;		/* more fragment flag */
77 };
78 
79 RB_HEAD(pf_frag_tree, pf_fragment);
80 struct pf_frnode {
81 	struct pf_addr		fn_src;		/* ip source address */
82 	struct pf_addr		fn_dst;		/* ip destination address */
83 	sa_family_t		fn_af;		/* address family */
84 	u_int8_t		fn_proto;	/* protocol for fragments in fn_tree */
85 	u_int32_t		fn_fragments;	/* number of entries in fn_tree */
86 
87 	RB_ENTRY(pf_frnode)	fn_entry;
88 	struct pf_frag_tree	fn_tree;	/* matching fragments, lookup by id */
89 };
90 
91 struct pf_fragment {
92 	uint32_t	fr_id;	/* fragment id for reassemble */
93 
94 	/* pointers to queue element */
95 	struct pf_frent	*fr_firstoff[PF_FRAG_ENTRY_POINTS];
96 	/* count entries between pointers */
97 	uint8_t	fr_entries[PF_FRAG_ENTRY_POINTS];
98 	RB_ENTRY(pf_fragment) fr_entry;
99 	TAILQ_ENTRY(pf_fragment) frag_next;
100 	uint32_t	fr_timeout;
101 	TAILQ_HEAD(pf_fragq, pf_frent) fr_queue;
102 	uint16_t	fr_maxlen;	/* maximum length of single fragment */
103 	u_int16_t	fr_holes;	/* number of holes in the queue */
104 	struct pf_frnode *fr_node;	/* ip src/dst/proto/af for fragments */
105 };
106 
107 VNET_DEFINE_STATIC(struct mtx, pf_frag_mtx);
108 #define V_pf_frag_mtx		VNET(pf_frag_mtx)
109 #define PF_FRAG_LOCK()		mtx_lock(&V_pf_frag_mtx)
110 #define PF_FRAG_UNLOCK()	mtx_unlock(&V_pf_frag_mtx)
111 #define PF_FRAG_ASSERT()	mtx_assert(&V_pf_frag_mtx, MA_OWNED)
112 
113 VNET_DEFINE(uma_zone_t, pf_state_scrub_z);	/* XXX: shared with pfsync */
114 
115 VNET_DEFINE_STATIC(uma_zone_t, pf_frent_z);
116 #define	V_pf_frent_z	VNET(pf_frent_z)
117 VNET_DEFINE_STATIC(uma_zone_t, pf_frnode_z);
118 #define	V_pf_frnode_z	VNET(pf_frnode_z)
119 VNET_DEFINE_STATIC(uma_zone_t, pf_frag_z);
120 #define	V_pf_frag_z	VNET(pf_frag_z)
121 
122 TAILQ_HEAD(pf_fragqueue, pf_fragment);
123 TAILQ_HEAD(pf_cachequeue, pf_fragment);
124 RB_HEAD(pf_frnode_tree, pf_frnode);
125 VNET_DEFINE_STATIC(struct pf_fragqueue,	pf_fragqueue);
126 #define	V_pf_fragqueue			VNET(pf_fragqueue)
127 static __inline int	pf_frnode_compare(struct pf_frnode *,
128 			    struct pf_frnode *);
129 VNET_DEFINE_STATIC(struct pf_frnode_tree, pf_frnode_tree);
130 #define	V_pf_frnode_tree		VNET(pf_frnode_tree)
131 RB_PROTOTYPE(pf_frnode_tree, pf_frnode, fn_entry, pf_frnode_compare);
132 RB_GENERATE(pf_frnode_tree, pf_frnode, fn_entry, pf_frnode_compare);
133 
134 static int		 pf_frag_compare(struct pf_fragment *,
135 			    struct pf_fragment *);
136 static RB_PROTOTYPE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
137 static RB_GENERATE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
138 
139 static void	pf_flush_fragments(void);
140 static void	pf_free_fragment(struct pf_fragment *);
141 
142 static struct pf_frent *pf_create_fragment(u_short *);
143 static int	pf_frent_holes(struct pf_frent *frent);
144 static struct pf_fragment	*pf_find_fragment(struct pf_frnode *, u_int32_t);
145 static inline int	pf_frent_index(struct pf_frent *);
146 static int	pf_frent_insert(struct pf_fragment *,
147 			    struct pf_frent *, struct pf_frent *);
148 void			pf_frent_remove(struct pf_fragment *,
149 			    struct pf_frent *);
150 struct pf_frent		*pf_frent_previous(struct pf_fragment *,
151 			    struct pf_frent *);
152 static struct pf_fragment *pf_fillup_fragment(struct pf_frnode *, u_int32_t,
153 		    struct pf_frent *, u_short *);
154 static struct mbuf *pf_join_fragment(struct pf_fragment *);
155 #ifdef INET
156 static int	pf_reassemble(struct mbuf **, u_short *);
157 #endif	/* INET */
158 #ifdef INET6
159 static int	pf_reassemble6(struct mbuf **,
160 		    struct ip6_frag *, uint16_t, uint16_t, u_short *);
161 #endif	/* INET6 */
162 
163 #ifdef INET
164 static void
pf_ip2key(struct ip * ip,struct pf_frnode * key)165 pf_ip2key(struct ip *ip, struct pf_frnode *key)
166 {
167 
168 	key->fn_src.v4 = ip->ip_src;
169 	key->fn_dst.v4 = ip->ip_dst;
170 	key->fn_af = AF_INET;
171 	key->fn_proto = ip->ip_p;
172 }
173 #endif	/* INET */
174 
175 void
pf_normalize_init(void)176 pf_normalize_init(void)
177 {
178 
179 	V_pf_frag_z = uma_zcreate("pf frags", sizeof(struct pf_fragment),
180 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
181 	V_pf_frnode_z = uma_zcreate("pf fragment node",
182 	    sizeof(struct pf_frnode), NULL, NULL, NULL, NULL,
183 	    UMA_ALIGN_PTR, 0);
184 	V_pf_frent_z = uma_zcreate("pf frag entries", sizeof(struct pf_frent),
185 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
186 	V_pf_state_scrub_z = uma_zcreate("pf state scrubs",
187 	    sizeof(struct pf_state_scrub),  NULL, NULL, NULL, NULL,
188 	    UMA_ALIGN_PTR, 0);
189 
190 	mtx_init(&V_pf_frag_mtx, "pf fragments", NULL, MTX_DEF);
191 
192 	V_pf_limits[PF_LIMIT_FRAGS].zone = V_pf_frent_z;
193 	V_pf_limits[PF_LIMIT_FRAGS].limit = PFFRAG_FRENT_HIWAT;
194 	uma_zone_set_max(V_pf_frent_z, PFFRAG_FRENT_HIWAT);
195 	uma_zone_set_warning(V_pf_frent_z, "PF frag entries limit reached");
196 
197 	TAILQ_INIT(&V_pf_fragqueue);
198 }
199 
200 void
pf_normalize_cleanup(void)201 pf_normalize_cleanup(void)
202 {
203 
204 	uma_zdestroy(V_pf_state_scrub_z);
205 	uma_zdestroy(V_pf_frent_z);
206 	uma_zdestroy(V_pf_frnode_z);
207 	uma_zdestroy(V_pf_frag_z);
208 
209 	mtx_destroy(&V_pf_frag_mtx);
210 }
211 
212 static int
pf_frnode_compare(struct pf_frnode * a,struct pf_frnode * b)213 pf_frnode_compare(struct pf_frnode *a, struct pf_frnode *b)
214 {
215 	int	diff;
216 
217 	if ((diff = a->fn_proto - b->fn_proto) != 0)
218 		return (diff);
219 	if ((diff = a->fn_af - b->fn_af) != 0)
220 		return (diff);
221 	if ((diff = pf_addr_cmp(&a->fn_src, &b->fn_src, a->fn_af)) != 0)
222 		return (diff);
223 	if ((diff = pf_addr_cmp(&a->fn_dst, &b->fn_dst, a->fn_af)) != 0)
224 		return (diff);
225 	return (0);
226 }
227 
228 static __inline int
pf_frag_compare(struct pf_fragment * a,struct pf_fragment * b)229 pf_frag_compare(struct pf_fragment *a, struct pf_fragment *b)
230 {
231 	int	diff;
232 
233 	if ((diff = a->fr_id - b->fr_id) != 0)
234 		return (diff);
235 
236 	return (0);
237 }
238 
239 void
pf_purge_expired_fragments(void)240 pf_purge_expired_fragments(void)
241 {
242 	u_int32_t	expire = time_uptime -
243 			    V_pf_default_rule.timeout[PFTM_FRAG];
244 
245 	pf_purge_fragments(expire);
246 }
247 
248 void
pf_purge_fragments(uint32_t expire)249 pf_purge_fragments(uint32_t expire)
250 {
251 	struct pf_fragment	*frag;
252 
253 	PF_FRAG_LOCK();
254 	while ((frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue)) != NULL) {
255 		if (frag->fr_timeout > expire)
256 			break;
257 
258 		DPFPRINTF(PF_DEBUG_MISC, "expiring %d(%p)",
259 		    frag->fr_id, frag);
260 		pf_free_fragment(frag);
261 	}
262 
263 	PF_FRAG_UNLOCK();
264 }
265 
266 /*
267  * Try to flush old fragments to make space for new ones
268  */
269 static void
pf_flush_fragments(void)270 pf_flush_fragments(void)
271 {
272 	struct pf_fragment	*frag;
273 	int			 goal;
274 
275 	PF_FRAG_ASSERT();
276 
277 	goal = uma_zone_get_cur(V_pf_frent_z) * 9 / 10;
278 	DPFPRINTF(PF_DEBUG_MISC, "trying to free %d frag entriess", goal);
279 	while (goal < uma_zone_get_cur(V_pf_frent_z)) {
280 		frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue);
281 		if (frag)
282 			pf_free_fragment(frag);
283 		else
284 			break;
285 	}
286 }
287 
288 /*
289  * Remove a fragment from the fragment queue, free its fragment entries,
290  * and free the fragment itself.
291  */
292 static void
pf_free_fragment(struct pf_fragment * frag)293 pf_free_fragment(struct pf_fragment *frag)
294 {
295 	struct pf_frent		*frent;
296 	struct pf_frnode	*frnode;
297 
298 	PF_FRAG_ASSERT();
299 
300 	frnode = frag->fr_node;
301 	RB_REMOVE(pf_frag_tree, &frnode->fn_tree, frag);
302 	MPASS(frnode->fn_fragments >= 1);
303 	frnode->fn_fragments--;
304 	if (frnode->fn_fragments == 0) {
305 		MPASS(RB_EMPTY(&frnode->fn_tree));
306 		RB_REMOVE(pf_frnode_tree, &V_pf_frnode_tree, frnode);
307 		uma_zfree(V_pf_frnode_z, frnode);
308 	}
309 
310 	TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next);
311 
312 	/* Free all fragment entries */
313 	while ((frent = TAILQ_FIRST(&frag->fr_queue)) != NULL) {
314 		TAILQ_REMOVE(&frag->fr_queue, frent, fr_next);
315 
316 		m_freem(frent->fe_m);
317 		uma_zfree(V_pf_frent_z, frent);
318 	}
319 
320 	uma_zfree(V_pf_frag_z, frag);
321 }
322 
323 static struct pf_fragment *
pf_find_fragment(struct pf_frnode * key,uint32_t id)324 pf_find_fragment(struct pf_frnode *key, uint32_t id)
325 {
326 	struct pf_fragment	*frag, idkey;
327 	struct pf_frnode	*frnode;
328 
329 	PF_FRAG_ASSERT();
330 
331 	frnode = RB_FIND(pf_frnode_tree, &V_pf_frnode_tree, key);
332 	if (frnode == NULL)
333 		return (NULL);
334 	MPASS(frnode->fn_fragments >= 1);
335 	idkey.fr_id = id;
336 	frag = RB_FIND(pf_frag_tree, &frnode->fn_tree, &idkey);
337 	if (frag == NULL)
338 		return (NULL);
339 	TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next);
340 	TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next);
341 
342 	return (frag);
343 }
344 
345 static struct pf_frent *
pf_create_fragment(u_short * reason)346 pf_create_fragment(u_short *reason)
347 {
348 	struct pf_frent *frent;
349 
350 	PF_FRAG_ASSERT();
351 
352 	frent = uma_zalloc(V_pf_frent_z, M_NOWAIT);
353 	if (frent == NULL) {
354 		pf_flush_fragments();
355 		frent = uma_zalloc(V_pf_frent_z, M_NOWAIT);
356 		if (frent == NULL) {
357 			REASON_SET(reason, PFRES_MEMORY);
358 			return (NULL);
359 		}
360 	}
361 
362 	return (frent);
363 }
364 
365 /*
366  * Calculate the additional holes that were created in the fragment
367  * queue by inserting this fragment.  A fragment in the middle
368  * creates one more hole by splitting.  For each connected side,
369  * it loses one hole.
370  * Fragment entry must be in the queue when calling this function.
371  */
372 static int
pf_frent_holes(struct pf_frent * frent)373 pf_frent_holes(struct pf_frent *frent)
374 {
375 	struct pf_frent *prev = TAILQ_PREV(frent, pf_fragq, fr_next);
376 	struct pf_frent *next = TAILQ_NEXT(frent, fr_next);
377 	int holes = 1;
378 
379 	if (prev == NULL) {
380 		if (frent->fe_off == 0)
381 			holes--;
382 	} else {
383 		KASSERT(frent->fe_off != 0, ("frent->fe_off != 0"));
384 		if (frent->fe_off == prev->fe_off + prev->fe_len)
385 			holes--;
386 	}
387 	if (next == NULL) {
388 		if (!frent->fe_mff)
389 			holes--;
390 	} else {
391 		KASSERT(frent->fe_mff, ("frent->fe_mff"));
392 		if (next->fe_off == frent->fe_off + frent->fe_len)
393 			holes--;
394 	}
395 	return holes;
396 }
397 
398 static inline int
pf_frent_index(struct pf_frent * frent)399 pf_frent_index(struct pf_frent *frent)
400 {
401 	/*
402 	 * We have an array of 16 entry points to the queue.  A full size
403 	 * 65535 octet IP packet can have 8192 fragments.  So the queue
404 	 * traversal length is at most 512 and at most 16 entry points are
405 	 * checked.  We need 128 additional bytes on a 64 bit architecture.
406 	 */
407 	CTASSERT(((u_int16_t)0xffff &~ 7) / (0x10000 / PF_FRAG_ENTRY_POINTS) ==
408 	    16 - 1);
409 	CTASSERT(((u_int16_t)0xffff >> 3) / PF_FRAG_ENTRY_POINTS == 512 - 1);
410 
411 	return frent->fe_off / (0x10000 / PF_FRAG_ENTRY_POINTS);
412 }
413 
414 static int
pf_frent_insert(struct pf_fragment * frag,struct pf_frent * frent,struct pf_frent * prev)415 pf_frent_insert(struct pf_fragment *frag, struct pf_frent *frent,
416     struct pf_frent *prev)
417 {
418 	int index;
419 
420 	CTASSERT(PF_FRAG_ENTRY_LIMIT <= 0xff);
421 
422 	/*
423 	 * A packet has at most 65536 octets.  With 16 entry points, each one
424 	 * spawns 4096 octets.  We limit these to 64 fragments each, which
425 	 * means on average every fragment must have at least 64 octets.
426 	 */
427 	index = pf_frent_index(frent);
428 	if (frag->fr_entries[index] >= PF_FRAG_ENTRY_LIMIT)
429 		return ENOBUFS;
430 	frag->fr_entries[index]++;
431 
432 	if (prev == NULL) {
433 		TAILQ_INSERT_HEAD(&frag->fr_queue, frent, fr_next);
434 	} else {
435 		KASSERT(prev->fe_off + prev->fe_len <= frent->fe_off,
436 		    ("overlapping fragment"));
437 		TAILQ_INSERT_AFTER(&frag->fr_queue, prev, frent, fr_next);
438 	}
439 
440 	if (frag->fr_firstoff[index] == NULL) {
441 		KASSERT(prev == NULL || pf_frent_index(prev) < index,
442 		    ("prev == NULL || pf_frent_index(pref) < index"));
443 		frag->fr_firstoff[index] = frent;
444 	} else {
445 		if (frent->fe_off < frag->fr_firstoff[index]->fe_off) {
446 			KASSERT(prev == NULL || pf_frent_index(prev) < index,
447 			    ("prev == NULL || pf_frent_index(pref) < index"));
448 			frag->fr_firstoff[index] = frent;
449 		} else {
450 			KASSERT(prev != NULL, ("prev != NULL"));
451 			KASSERT(pf_frent_index(prev) == index,
452 			    ("pf_frent_index(prev) == index"));
453 		}
454 	}
455 
456 	frag->fr_holes += pf_frent_holes(frent);
457 
458 	return 0;
459 }
460 
461 void
pf_frent_remove(struct pf_fragment * frag,struct pf_frent * frent)462 pf_frent_remove(struct pf_fragment *frag, struct pf_frent *frent)
463 {
464 #ifdef INVARIANTS
465 	struct pf_frent *prev = TAILQ_PREV(frent, pf_fragq, fr_next);
466 #endif /* INVARIANTS */
467 	struct pf_frent *next = TAILQ_NEXT(frent, fr_next);
468 	int index;
469 
470 	frag->fr_holes -= pf_frent_holes(frent);
471 
472 	index = pf_frent_index(frent);
473 	KASSERT(frag->fr_firstoff[index] != NULL, ("frent not found"));
474 	if (frag->fr_firstoff[index]->fe_off == frent->fe_off) {
475 		if (next == NULL) {
476 			frag->fr_firstoff[index] = NULL;
477 		} else {
478 			KASSERT(frent->fe_off + frent->fe_len <= next->fe_off,
479 			    ("overlapping fragment"));
480 			if (pf_frent_index(next) == index) {
481 				frag->fr_firstoff[index] = next;
482 			} else {
483 				frag->fr_firstoff[index] = NULL;
484 			}
485 		}
486 	} else {
487 		KASSERT(frag->fr_firstoff[index]->fe_off < frent->fe_off,
488 		    ("frag->fr_firstoff[index]->fe_off < frent->fe_off"));
489 		KASSERT(prev != NULL, ("prev != NULL"));
490 		KASSERT(prev->fe_off + prev->fe_len <= frent->fe_off,
491 		    ("overlapping fragment"));
492 		KASSERT(pf_frent_index(prev) == index,
493 		    ("pf_frent_index(prev) == index"));
494 	}
495 
496 	TAILQ_REMOVE(&frag->fr_queue, frent, fr_next);
497 
498 	KASSERT(frag->fr_entries[index] > 0, ("No fragments remaining"));
499 	frag->fr_entries[index]--;
500 }
501 
502 struct pf_frent *
pf_frent_previous(struct pf_fragment * frag,struct pf_frent * frent)503 pf_frent_previous(struct pf_fragment *frag, struct pf_frent *frent)
504 {
505 	struct pf_frent *prev, *next;
506 	int index;
507 
508 	/*
509 	 * If there are no fragments after frag, take the final one.  Assume
510 	 * that the global queue is not empty.
511 	 */
512 	prev = TAILQ_LAST(&frag->fr_queue, pf_fragq);
513 	KASSERT(prev != NULL, ("prev != NULL"));
514 	if (prev->fe_off <= frent->fe_off)
515 		return prev;
516 	/*
517 	 * We want to find a fragment entry that is before frag, but still
518 	 * close to it.  Find the first fragment entry that is in the same
519 	 * entry point or in the first entry point after that.  As we have
520 	 * already checked that there are entries behind frag, this will
521 	 * succeed.
522 	 */
523 	for (index = pf_frent_index(frent); index < PF_FRAG_ENTRY_POINTS;
524 	    index++) {
525 		prev = frag->fr_firstoff[index];
526 		if (prev != NULL)
527 			break;
528 	}
529 	KASSERT(prev != NULL, ("prev != NULL"));
530 	/*
531 	 * In prev we may have a fragment from the same entry point that is
532 	 * before frent, or one that is just one position behind frent.
533 	 * In the latter case, we go back one step and have the predecessor.
534 	 * There may be none if the new fragment will be the first one.
535 	 */
536 	if (prev->fe_off > frent->fe_off) {
537 		prev = TAILQ_PREV(prev, pf_fragq, fr_next);
538 		if (prev == NULL)
539 			return NULL;
540 		KASSERT(prev->fe_off <= frent->fe_off,
541 		    ("prev->fe_off <= frent->fe_off"));
542 		return prev;
543 	}
544 	/*
545 	 * In prev is the first fragment of the entry point.  The offset
546 	 * of frag is behind it.  Find the closest previous fragment.
547 	 */
548 	for (next = TAILQ_NEXT(prev, fr_next); next != NULL;
549 	    next = TAILQ_NEXT(next, fr_next)) {
550 		if (next->fe_off > frent->fe_off)
551 			break;
552 		prev = next;
553 	}
554 	return prev;
555 }
556 
557 static struct pf_fragment *
pf_fillup_fragment(struct pf_frnode * key,uint32_t id,struct pf_frent * frent,u_short * reason)558 pf_fillup_fragment(struct pf_frnode *key, uint32_t id,
559     struct pf_frent *frent, u_short *reason)
560 {
561 	struct pf_frent		*after, *next, *prev;
562 	struct pf_fragment	*frag;
563 	struct pf_frnode	*frnode;
564 	uint16_t		total;
565 
566 	PF_FRAG_ASSERT();
567 
568 	/* No empty fragments. */
569 	if (frent->fe_len == 0) {
570 		DPFPRINTF(PF_DEBUG_MISC, "bad fragment: len 0");
571 		goto bad_fragment;
572 	}
573 
574 	/* All fragments are 8 byte aligned. */
575 	if (frent->fe_mff && (frent->fe_len & 0x7)) {
576 		DPFPRINTF(PF_DEBUG_MISC, "bad fragment: mff and len %d",
577 		    frent->fe_len);
578 		goto bad_fragment;
579 	}
580 
581 	/* Respect maximum length, IP_MAXPACKET == IPV6_MAXPACKET. */
582 	if (frent->fe_off + frent->fe_len > IP_MAXPACKET) {
583 		DPFPRINTF(PF_DEBUG_MISC, "bad fragment: max packet %d",
584 		    frent->fe_off + frent->fe_len);
585 		goto bad_fragment;
586 	}
587 
588 	if (key->fn_af == AF_INET)
589 		DPFPRINTF(PF_DEBUG_MISC, "reass frag %d @ %d-%d\n",
590 		    id, frent->fe_off, frent->fe_off + frent->fe_len);
591 	else
592 		DPFPRINTF(PF_DEBUG_MISC, "reass frag %#08x @ %d-%d",
593 		    id, frent->fe_off, frent->fe_off + frent->fe_len);
594 
595 	/* Fully buffer all of the fragments in this fragment queue. */
596 	frag = pf_find_fragment(key, id);
597 
598 	/* Create a new reassembly queue for this packet. */
599 	if (frag == NULL) {
600 		frag = uma_zalloc(V_pf_frag_z, M_NOWAIT);
601 		if (frag == NULL) {
602 			pf_flush_fragments();
603 			frag = uma_zalloc(V_pf_frag_z, M_NOWAIT);
604 			if (frag == NULL) {
605 				REASON_SET(reason, PFRES_MEMORY);
606 				goto drop_fragment;
607 			}
608 		}
609 
610 		frnode = RB_FIND(pf_frnode_tree, &V_pf_frnode_tree, key);
611 		if (frnode == NULL) {
612 			frnode = uma_zalloc(V_pf_frnode_z, M_NOWAIT);
613 			if (frnode == NULL) {
614 				pf_flush_fragments();
615 				frnode = uma_zalloc(V_pf_frnode_z, M_NOWAIT);
616 				if (frnode == NULL) {
617 					REASON_SET(reason, PFRES_MEMORY);
618 					uma_zfree(V_pf_frag_z, frag);
619 					goto drop_fragment;
620 				}
621 			}
622 			*frnode = *key;
623 			RB_INIT(&frnode->fn_tree);
624 			frnode->fn_fragments = 0;
625 		}
626 		memset(frag->fr_firstoff, 0, sizeof(frag->fr_firstoff));
627 		memset(frag->fr_entries, 0, sizeof(frag->fr_entries));
628 		frag->fr_timeout = time_uptime;
629 		TAILQ_INIT(&frag->fr_queue);
630 		frag->fr_maxlen = frent->fe_len;
631 		frag->fr_holes = 1;
632 
633 		frag->fr_id = id;
634 		frag->fr_node = frnode;
635 		/* RB_INSERT cannot fail as pf_find_fragment() found nothing */
636 		RB_INSERT(pf_frag_tree, &frnode->fn_tree, frag);
637 		frnode->fn_fragments++;
638 		if (frnode->fn_fragments == 1)
639 			RB_INSERT(pf_frnode_tree, &V_pf_frnode_tree, frnode);
640 
641 		TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next);
642 
643 		/* We do not have a previous fragment, cannot fail. */
644 		pf_frent_insert(frag, frent, NULL);
645 
646 		return (frag);
647 	}
648 
649 	KASSERT(!TAILQ_EMPTY(&frag->fr_queue), ("!TAILQ_EMPTY()->fr_queue"));
650 	MPASS(frag->fr_node);
651 
652 	/* Remember maximum fragment len for refragmentation. */
653 	if (frent->fe_len > frag->fr_maxlen)
654 		frag->fr_maxlen = frent->fe_len;
655 
656 	/* Maximum data we have seen already. */
657 	total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
658 		TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
659 
660 	/* Non terminal fragments must have more fragments flag. */
661 	if (frent->fe_off + frent->fe_len < total && !frent->fe_mff)
662 		goto free_ipv6_fragment;
663 
664 	/* Check if we saw the last fragment already. */
665 	if (!TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_mff) {
666 		if (frent->fe_off + frent->fe_len > total ||
667 		    (frent->fe_off + frent->fe_len == total && frent->fe_mff))
668 			goto free_ipv6_fragment;
669 	} else {
670 		if (frent->fe_off + frent->fe_len == total && !frent->fe_mff)
671 			goto free_ipv6_fragment;
672 	}
673 
674 	/* Find neighbors for newly inserted fragment */
675 	prev = pf_frent_previous(frag, frent);
676 	if (prev == NULL) {
677 		after = TAILQ_FIRST(&frag->fr_queue);
678 		KASSERT(after != NULL, ("after != NULL"));
679 	} else {
680 		after = TAILQ_NEXT(prev, fr_next);
681 	}
682 
683 	if (prev != NULL && prev->fe_off + prev->fe_len > frent->fe_off) {
684 		uint16_t precut;
685 
686 		if (frag->fr_node->fn_af == AF_INET6)
687 			goto free_fragment;
688 
689 		precut = prev->fe_off + prev->fe_len - frent->fe_off;
690 		if (precut >= frent->fe_len) {
691 			DPFPRINTF(PF_DEBUG_MISC, "new frag overlapped");
692 			goto drop_fragment;
693 		}
694 		DPFPRINTF(PF_DEBUG_MISC, "frag head overlap %d", precut);
695 		m_adj(frent->fe_m, precut);
696 		frent->fe_off += precut;
697 		frent->fe_len -= precut;
698 	}
699 
700 	for (; after != NULL && frent->fe_off + frent->fe_len > after->fe_off;
701 	    after = next) {
702 		uint16_t aftercut;
703 
704 		aftercut = frent->fe_off + frent->fe_len - after->fe_off;
705 		if (aftercut < after->fe_len) {
706 			DPFPRINTF(PF_DEBUG_MISC, "frag tail overlap %d",
707 			    aftercut);
708 			m_adj(after->fe_m, aftercut);
709 			/* Fragment may switch queue as fe_off changes */
710 			pf_frent_remove(frag, after);
711 			after->fe_off += aftercut;
712 			after->fe_len -= aftercut;
713 			/* Insert into correct queue */
714 			if (pf_frent_insert(frag, after, prev)) {
715 				DPFPRINTF(PF_DEBUG_MISC,
716 				    "fragment requeue limit exceeded");
717 				m_freem(after->fe_m);
718 				uma_zfree(V_pf_frent_z, after);
719 				/* There is not way to recover */
720 				goto free_fragment;
721 			}
722 			break;
723 		}
724 
725 		/* This fragment is completely overlapped, lose it. */
726 		DPFPRINTF(PF_DEBUG_MISC, "old frag overlapped");
727 		next = TAILQ_NEXT(after, fr_next);
728 		pf_frent_remove(frag, after);
729 		m_freem(after->fe_m);
730 		uma_zfree(V_pf_frent_z, after);
731 	}
732 
733 	/* If part of the queue gets too long, there is not way to recover. */
734 	if (pf_frent_insert(frag, frent, prev)) {
735 		DPFPRINTF(PF_DEBUG_MISC, "fragment queue limit exceeded");
736 		goto bad_fragment;
737 	}
738 
739 	return (frag);
740 
741 free_ipv6_fragment:
742 	if (frag->fr_node->fn_af == AF_INET)
743 		goto bad_fragment;
744 free_fragment:
745 	/*
746 	 * RFC 5722, Errata 3089:  When reassembling an IPv6 datagram, if one
747 	 * or more its constituent fragments is determined to be an overlapping
748 	 * fragment, the entire datagram (and any constituent fragments) MUST
749 	 * be silently discarded.
750 	 */
751 	DPFPRINTF(PF_DEBUG_MISC, "flush overlapping fragments");
752 	pf_free_fragment(frag);
753 
754 bad_fragment:
755 	REASON_SET(reason, PFRES_FRAG);
756 drop_fragment:
757 	uma_zfree(V_pf_frent_z, frent);
758 	return (NULL);
759 }
760 
761 static struct mbuf *
pf_join_fragment(struct pf_fragment * frag)762 pf_join_fragment(struct pf_fragment *frag)
763 {
764 	struct mbuf *m, *m2;
765 	struct pf_frent	*frent;
766 
767 	frent = TAILQ_FIRST(&frag->fr_queue);
768 	TAILQ_REMOVE(&frag->fr_queue, frent, fr_next);
769 
770 	m = frent->fe_m;
771 	if ((frent->fe_hdrlen + frent->fe_len) < m->m_pkthdr.len)
772 		m_adj(m, (frent->fe_hdrlen + frent->fe_len) - m->m_pkthdr.len);
773 	uma_zfree(V_pf_frent_z, frent);
774 	while ((frent = TAILQ_FIRST(&frag->fr_queue)) != NULL) {
775 		TAILQ_REMOVE(&frag->fr_queue, frent, fr_next);
776 
777 		m2 = frent->fe_m;
778 		/* Strip off ip header. */
779 		m_adj(m2, frent->fe_hdrlen);
780 		/* Strip off any trailing bytes. */
781 		if (frent->fe_len < m2->m_pkthdr.len)
782 			m_adj(m2, frent->fe_len - m2->m_pkthdr.len);
783 
784 		uma_zfree(V_pf_frent_z, frent);
785 		m_cat(m, m2);
786 	}
787 
788 	/* Remove from fragment queue. */
789 	pf_free_fragment(frag);
790 
791 	return (m);
792 }
793 
794 #ifdef INET
795 static int
pf_reassemble(struct mbuf ** m0,u_short * reason)796 pf_reassemble(struct mbuf **m0, u_short *reason)
797 {
798 	struct mbuf		*m = *m0;
799 	struct ip		*ip = mtod(m, struct ip *);
800 	struct pf_frent		*frent;
801 	struct pf_fragment	*frag;
802 	struct m_tag		*mtag;
803 	struct pf_fragment_tag	*ftag;
804 	struct pf_frnode	 key;
805 	uint16_t		 total, hdrlen;
806 	uint32_t		 frag_id;
807 	uint16_t		 maxlen;
808 
809 	/* Get an entry for the fragment queue */
810 	if ((frent = pf_create_fragment(reason)) == NULL)
811 		return (PF_DROP);
812 
813 	frent->fe_m = m;
814 	frent->fe_hdrlen = ip->ip_hl << 2;
815 	frent->fe_extoff = 0;
816 	frent->fe_len = ntohs(ip->ip_len) - (ip->ip_hl << 2);
817 	frent->fe_off = (ntohs(ip->ip_off) & IP_OFFMASK) << 3;
818 	frent->fe_mff = ntohs(ip->ip_off) & IP_MF;
819 
820 	pf_ip2key(ip, &key);
821 
822 	if ((frag = pf_fillup_fragment(&key, ip->ip_id, frent, reason)) == NULL)
823 		return (PF_DROP);
824 
825 	/* The mbuf is part of the fragment entry, no direct free or access */
826 	m = *m0 = NULL;
827 
828 	if (frag->fr_holes) {
829 		DPFPRINTF(PF_DEBUG_MISC, "frag %d, holes %d",
830 		    frag->fr_id, frag->fr_holes);
831 		return (PF_PASS);  /* drop because *m0 is NULL, no error */
832 	}
833 
834 	/* We have all the data */
835 	frent = TAILQ_FIRST(&frag->fr_queue);
836 	KASSERT(frent != NULL, ("frent != NULL"));
837 	total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
838 		TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
839 	hdrlen = frent->fe_hdrlen;
840 
841 	maxlen = frag->fr_maxlen;
842 	frag_id = frag->fr_id;
843 	m = *m0 = pf_join_fragment(frag);
844 	frag = NULL;
845 
846 	if (m->m_flags & M_PKTHDR) {
847 		int plen = 0;
848 		for (m = *m0; m; m = m->m_next)
849 			plen += m->m_len;
850 		m = *m0;
851 		m->m_pkthdr.len = plen;
852 	}
853 
854 	if ((mtag = m_tag_get(PACKET_TAG_PF_REASSEMBLED,
855 	    sizeof(struct pf_fragment_tag), M_NOWAIT)) == NULL) {
856 		REASON_SET(reason, PFRES_SHORT);
857 		/* PF_DROP requires a valid mbuf *m0 in pf_test() */
858 		return (PF_DROP);
859 	}
860 	ftag = (struct pf_fragment_tag *)(mtag + 1);
861 	ftag->ft_hdrlen = hdrlen;
862 	ftag->ft_extoff = 0;
863 	ftag->ft_maxlen = maxlen;
864 	ftag->ft_id = frag_id;
865 	m_tag_prepend(m, mtag);
866 
867 	ip = mtod(m, struct ip *);
868 	ip->ip_sum = pf_cksum_fixup(ip->ip_sum, ip->ip_len,
869 	    htons(hdrlen + total), 0);
870 	ip->ip_len = htons(hdrlen + total);
871 	ip->ip_sum = pf_cksum_fixup(ip->ip_sum, ip->ip_off,
872 	    ip->ip_off & ~(IP_MF|IP_OFFMASK), 0);
873 	ip->ip_off &= ~(IP_MF|IP_OFFMASK);
874 
875 	if (hdrlen + total > IP_MAXPACKET) {
876 		DPFPRINTF(PF_DEBUG_MISC, "drop: too big: %d", total);
877 		ip->ip_len = 0;
878 		REASON_SET(reason, PFRES_SHORT);
879 		/* PF_DROP requires a valid mbuf *m0 in pf_test() */
880 		return (PF_DROP);
881 	}
882 
883 	DPFPRINTF(PF_DEBUG_MISC, "complete: %p(%d)", m, ntohs(ip->ip_len));
884 	return (PF_PASS);
885 }
886 #endif	/* INET */
887 
888 #ifdef INET6
889 static int
pf_reassemble6(struct mbuf ** m0,struct ip6_frag * fraghdr,uint16_t hdrlen,uint16_t extoff,u_short * reason)890 pf_reassemble6(struct mbuf **m0, struct ip6_frag *fraghdr,
891     uint16_t hdrlen, uint16_t extoff, u_short *reason)
892 {
893 	struct mbuf		*m = *m0;
894 	struct ip6_hdr		*ip6 = mtod(m, struct ip6_hdr *);
895 	struct pf_frent		*frent;
896 	struct pf_fragment	*frag;
897 	struct pf_frnode	 key;
898 	struct m_tag		*mtag;
899 	struct pf_fragment_tag	*ftag;
900 	int			 off;
901 	uint32_t		 frag_id;
902 	uint16_t		 total, maxlen;
903 	uint8_t			 proto;
904 
905 	PF_FRAG_LOCK();
906 
907 	/* Get an entry for the fragment queue. */
908 	if ((frent = pf_create_fragment(reason)) == NULL) {
909 		PF_FRAG_UNLOCK();
910 		return (PF_DROP);
911 	}
912 
913 	frent->fe_m = m;
914 	frent->fe_hdrlen = hdrlen;
915 	frent->fe_extoff = extoff;
916 	frent->fe_len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - hdrlen;
917 	frent->fe_off = ntohs(fraghdr->ip6f_offlg & IP6F_OFF_MASK);
918 	frent->fe_mff = fraghdr->ip6f_offlg & IP6F_MORE_FRAG;
919 
920 	key.fn_src.v6 = ip6->ip6_src;
921 	key.fn_dst.v6 = ip6->ip6_dst;
922 	key.fn_af = AF_INET6;
923 	/* Only the first fragment's protocol is relevant. */
924 	key.fn_proto = 0;
925 
926 	if ((frag = pf_fillup_fragment(&key, fraghdr->ip6f_ident, frent, reason)) == NULL) {
927 		PF_FRAG_UNLOCK();
928 		return (PF_DROP);
929 	}
930 
931 	/* The mbuf is part of the fragment entry, no direct free or access. */
932 	m = *m0 = NULL;
933 
934 	if (frag->fr_holes) {
935 		DPFPRINTF(PF_DEBUG_MISC, "frag %d, holes %d", frag->fr_id,
936 		    frag->fr_holes);
937 		PF_FRAG_UNLOCK();
938 		return (PF_PASS);  /* Drop because *m0 is NULL, no error. */
939 	}
940 
941 	/* We have all the data. */
942 	frent = TAILQ_FIRST(&frag->fr_queue);
943 	KASSERT(frent != NULL, ("frent != NULL"));
944 	extoff = frent->fe_extoff;
945 	maxlen = frag->fr_maxlen;
946 	frag_id = frag->fr_id;
947 	total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
948 		TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
949 	hdrlen = frent->fe_hdrlen - sizeof(struct ip6_frag);
950 
951 	m = *m0 = pf_join_fragment(frag);
952 	frag = NULL;
953 
954 	PF_FRAG_UNLOCK();
955 
956 	/* Take protocol from first fragment header. */
957 	m = m_getptr(m, hdrlen + offsetof(struct ip6_frag, ip6f_nxt), &off);
958 	KASSERT(m, ("%s: short mbuf chain", __func__));
959 	proto = *(mtod(m, uint8_t *) + off);
960 	m = *m0;
961 
962 	/* Delete frag6 header */
963 	if (ip6_deletefraghdr(m, hdrlen, M_NOWAIT) != 0)
964 		goto fail;
965 
966 	if (m->m_flags & M_PKTHDR) {
967 		int plen = 0;
968 		for (m = *m0; m; m = m->m_next)
969 			plen += m->m_len;
970 		m = *m0;
971 		m->m_pkthdr.len = plen;
972 	}
973 
974 	if ((mtag = m_tag_get(PACKET_TAG_PF_REASSEMBLED,
975 	    sizeof(struct pf_fragment_tag), M_NOWAIT)) == NULL)
976 		goto fail;
977 	ftag = (struct pf_fragment_tag *)(mtag + 1);
978 	ftag->ft_hdrlen = hdrlen;
979 	ftag->ft_extoff = extoff;
980 	ftag->ft_maxlen = maxlen;
981 	ftag->ft_id = frag_id;
982 	m_tag_prepend(m, mtag);
983 
984 	ip6 = mtod(m, struct ip6_hdr *);
985 	ip6->ip6_plen = htons(hdrlen - sizeof(struct ip6_hdr) + total);
986 	if (extoff) {
987 		/* Write protocol into next field of last extension header. */
988 		m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt),
989 		    &off);
990 		KASSERT(m, ("%s: short mbuf chain", __func__));
991 		*(mtod(m, char *) + off) = proto;
992 		m = *m0;
993 	} else
994 		ip6->ip6_nxt = proto;
995 
996 	if (hdrlen - sizeof(struct ip6_hdr) + total > IPV6_MAXPACKET) {
997 		DPFPRINTF(PF_DEBUG_MISC, "drop: too big: %d", total);
998 		ip6->ip6_plen = 0;
999 		REASON_SET(reason, PFRES_SHORT);
1000 		/* PF_DROP requires a valid mbuf *m0 in pf_test6(). */
1001 		return (PF_DROP);
1002 	}
1003 
1004 	DPFPRINTF(PF_DEBUG_MISC, "complete: %p(%d)", m,
1005 	    ntohs(ip6->ip6_plen));
1006 	return (PF_PASS);
1007 
1008 fail:
1009 	REASON_SET(reason, PFRES_MEMORY);
1010 	/* PF_DROP requires a valid mbuf *m0 in pf_test6(), will free later. */
1011 	return (PF_DROP);
1012 }
1013 #endif	/* INET6 */
1014 
1015 #ifdef INET6
1016 int
pf_max_frag_size(struct mbuf * m)1017 pf_max_frag_size(struct mbuf *m)
1018 {
1019 	struct m_tag *tag;
1020 	struct pf_fragment_tag *ftag;
1021 
1022 	tag = m_tag_find(m, PACKET_TAG_PF_REASSEMBLED, NULL);
1023 	if (tag == NULL)
1024 		return (m->m_pkthdr.len);
1025 
1026 	ftag = (struct pf_fragment_tag *)(tag + 1);
1027 
1028 	return (ftag->ft_maxlen);
1029 }
1030 
1031 int
pf_refragment6(struct ifnet * ifp,struct mbuf ** m0,struct m_tag * mtag,struct ifnet * rt,bool forward)1032 pf_refragment6(struct ifnet *ifp, struct mbuf **m0, struct m_tag *mtag,
1033     struct ifnet *rt, bool forward)
1034 {
1035 	struct mbuf		*m = *m0, *t;
1036 	struct ip6_hdr		*hdr;
1037 	struct pf_fragment_tag	*ftag = (struct pf_fragment_tag *)(mtag + 1);
1038 	struct pf_pdesc		 pd;
1039 	uint32_t		 frag_id;
1040 	uint16_t		 hdrlen, extoff, maxlen;
1041 	uint8_t			 proto;
1042 	int			 error, action;
1043 
1044 	hdrlen = ftag->ft_hdrlen;
1045 	extoff = ftag->ft_extoff;
1046 	maxlen = ftag->ft_maxlen;
1047 	frag_id = ftag->ft_id;
1048 	m_tag_delete(m, mtag);
1049 	mtag = NULL;
1050 	ftag = NULL;
1051 
1052 	if (extoff) {
1053 		int off;
1054 
1055 		/* Use protocol from next field of last extension header */
1056 		m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt),
1057 		    &off);
1058 		KASSERT((m != NULL), ("pf_refragment6: short mbuf chain"));
1059 		proto = *(mtod(m, uint8_t *) + off);
1060 		*(mtod(m, char *) + off) = IPPROTO_FRAGMENT;
1061 		m = *m0;
1062 	} else {
1063 		hdr = mtod(m, struct ip6_hdr *);
1064 		proto = hdr->ip6_nxt;
1065 		hdr->ip6_nxt = IPPROTO_FRAGMENT;
1066 	}
1067 
1068 	/* In case of link-local traffic we'll need a scope set. */
1069 	hdr = mtod(m, struct ip6_hdr *);
1070 
1071 	in6_setscope(&hdr->ip6_src, ifp, NULL);
1072 	in6_setscope(&hdr->ip6_dst, ifp, NULL);
1073 
1074 	/* The MTU must be a multiple of 8 bytes, or we risk doing the
1075 	 * fragmentation wrong. */
1076 	maxlen = maxlen & ~7;
1077 
1078 	/*
1079 	 * Maxlen may be less than 8 if there was only a single
1080 	 * fragment.  As it was fragmented before, add a fragment
1081 	 * header also for a single fragment.  If total or maxlen
1082 	 * is less than 8, ip6_fragment() will return EMSGSIZE and
1083 	 * we drop the packet.
1084 	 */
1085 	error = ip6_fragment(ifp, m, hdrlen, proto, maxlen, frag_id);
1086 	m = (*m0)->m_nextpkt;
1087 	(*m0)->m_nextpkt = NULL;
1088 	if (error == 0) {
1089 		/* The first mbuf contains the unfragmented packet. */
1090 		m_freem(*m0);
1091 		*m0 = NULL;
1092 		action = PF_PASS;
1093 	} else {
1094 		/* Drop expects an mbuf to free. */
1095 		DPFPRINTF(PF_DEBUG_MISC, "refragment error %d", error);
1096 		action = PF_DROP;
1097 	}
1098 	for (; m; m = t) {
1099 		t = m->m_nextpkt;
1100 		m->m_nextpkt = NULL;
1101 		m->m_flags |= M_SKIP_FIREWALL;
1102 		memset(&pd, 0, sizeof(pd));
1103 		pd.pf_mtag = pf_find_mtag(m);
1104 		if (error != 0) {
1105 			m_freem(m);
1106 			continue;
1107 		}
1108 		if (rt != NULL) {
1109 			struct sockaddr_in6	dst;
1110 			hdr = mtod(m, struct ip6_hdr *);
1111 
1112 			bzero(&dst, sizeof(dst));
1113 			dst.sin6_family = AF_INET6;
1114 			dst.sin6_len = sizeof(dst);
1115 			dst.sin6_addr = hdr->ip6_dst;
1116 
1117 			if (m->m_pkthdr.len <= if_getmtu(ifp)) {
1118 				nd6_output_ifp(rt, rt, m, &dst, NULL);
1119 			} else {
1120 				in6_ifstat_inc(ifp, ifs6_in_toobig);
1121 				icmp6_error(m, ICMP6_PACKET_TOO_BIG, 0,
1122 				    if_getmtu(ifp));
1123 			}
1124 		} else if (forward) {
1125 			MPASS(m->m_pkthdr.rcvif != NULL);
1126 			ip6_forward(m, 0);
1127 		} else {
1128 			(void)ip6_output(m, NULL, NULL, 0, NULL, NULL,
1129 			    NULL);
1130 		}
1131 	}
1132 
1133 	return (action);
1134 }
1135 #endif /* INET6 */
1136 
1137 #ifdef INET
1138 int
pf_normalize_ip(u_short * reason,struct pf_pdesc * pd)1139 pf_normalize_ip(u_short *reason, struct pf_pdesc *pd)
1140 {
1141 	struct pf_krule		*r;
1142 	struct ip		*h = mtod(pd->m, struct ip *);
1143 	int			 mff = (ntohs(h->ip_off) & IP_MF);
1144 	int			 hlen = h->ip_hl << 2;
1145 	u_int16_t		 fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
1146 	u_int16_t		 max;
1147 	int			 ip_len;
1148 	int			 tag = -1;
1149 	int			 verdict;
1150 	bool			 scrub_compat;
1151 
1152 	PF_RULES_RASSERT();
1153 
1154 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1155 	/*
1156 	 * Check if there are any scrub rules, matching or not.
1157 	 * Lack of scrub rules means:
1158 	 *  - enforced packet normalization operation just like in OpenBSD
1159 	 *  - fragment reassembly depends on V_pf_status.reass
1160 	 * With scrub rules:
1161 	 *  - packet normalization is performed if there is a matching scrub rule
1162 	 *  - fragment reassembly is performed if the matching rule has no
1163 	 *    PFRULE_FRAGMENT_NOREASS flag
1164 	 */
1165 	scrub_compat = (r != NULL);
1166 	while (r != NULL) {
1167 		pf_counter_u64_add(&r->evaluations, 1);
1168 		if (pfi_kkif_match(r->kif, pd->kif) == r->ifnot)
1169 			r = r->skip[PF_SKIP_IFP];
1170 		else if (r->direction && r->direction != pd->dir)
1171 			r = r->skip[PF_SKIP_DIR];
1172 		else if (r->af && r->af != AF_INET)
1173 			r = r->skip[PF_SKIP_AF];
1174 		else if (r->proto && r->proto != h->ip_p)
1175 			r = r->skip[PF_SKIP_PROTO];
1176 		else if (PF_MISMATCHAW(&r->src.addr,
1177 		    (struct pf_addr *)&h->ip_src.s_addr, AF_INET,
1178 		    r->src.neg, pd->kif, M_GETFIB(pd->m)))
1179 			r = r->skip[PF_SKIP_SRC_ADDR];
1180 		else if (PF_MISMATCHAW(&r->dst.addr,
1181 		    (struct pf_addr *)&h->ip_dst.s_addr, AF_INET,
1182 		    r->dst.neg, NULL, M_GETFIB(pd->m)))
1183 			r = r->skip[PF_SKIP_DST_ADDR];
1184 		else if (r->match_tag && !pf_match_tag(pd->m, r, &tag,
1185 		    pd->pf_mtag ? pd->pf_mtag->tag : 0))
1186 			r = TAILQ_NEXT(r, entries);
1187 		else
1188 			break;
1189 	}
1190 
1191 	if (scrub_compat) {
1192 		/* With scrub rules present IPv4 normalization happens only
1193 		 * if one of rules has matched and it's not a "no scrub" rule */
1194 		if (r == NULL || r->action == PF_NOSCRUB)
1195 			return (PF_PASS);
1196 
1197 		pf_counter_u64_critical_enter();
1198 		pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1);
1199 		pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len);
1200 		pf_counter_u64_critical_exit();
1201 		pf_rule_to_actions(r, &pd->act);
1202 	}
1203 
1204 	/* Check for illegal packets */
1205 	if (hlen < (int)sizeof(struct ip)) {
1206 		REASON_SET(reason, PFRES_NORM);
1207 		goto drop;
1208 	}
1209 
1210 	if (hlen > ntohs(h->ip_len)) {
1211 		REASON_SET(reason, PFRES_NORM);
1212 		goto drop;
1213 	}
1214 
1215 	/* Clear IP_DF if the rule uses the no-df option or we're in no-df mode */
1216 	if (((!scrub_compat && V_pf_status.reass & PF_REASS_NODF) ||
1217 	    (r != NULL && r->rule_flag & PFRULE_NODF)) &&
1218 	    (h->ip_off & htons(IP_DF))
1219 	) {
1220 		u_int16_t ip_off = h->ip_off;
1221 
1222 		h->ip_off &= htons(~IP_DF);
1223 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
1224 	}
1225 
1226 	/* We will need other tests here */
1227 	if (!fragoff && !mff)
1228 		goto no_fragment;
1229 
1230 	/* We're dealing with a fragment now. Don't allow fragments
1231 	 * with IP_DF to enter the cache. If the flag was cleared by
1232 	 * no-df above, fine. Otherwise drop it.
1233 	 */
1234 	if (h->ip_off & htons(IP_DF)) {
1235 		DPFPRINTF(PF_DEBUG_MISC, "IP_DF");
1236 		goto bad;
1237 	}
1238 
1239 	ip_len = ntohs(h->ip_len) - hlen;
1240 
1241 	/* All fragments are 8 byte aligned */
1242 	if (mff && (ip_len & 0x7)) {
1243 		DPFPRINTF(PF_DEBUG_MISC, "mff and %d", ip_len);
1244 		goto bad;
1245 	}
1246 
1247 	/* Respect maximum length */
1248 	if (fragoff + ip_len > IP_MAXPACKET) {
1249 		DPFPRINTF(PF_DEBUG_MISC, "max packet %d", fragoff + ip_len);
1250 		goto bad;
1251 	}
1252 
1253 	if ((!scrub_compat && V_pf_status.reass) ||
1254 	    (r != NULL && !(r->rule_flag & PFRULE_FRAGMENT_NOREASS))
1255 	) {
1256 		max = fragoff + ip_len;
1257 
1258 		/* Fully buffer all of the fragments
1259 		 * Might return a completely reassembled mbuf, or NULL */
1260 		PF_FRAG_LOCK();
1261 		DPFPRINTF(PF_DEBUG_MISC, "reass frag %d @ %d-%d",
1262 		    h->ip_id, fragoff, max);
1263 		verdict = pf_reassemble(&pd->m, reason);
1264 		PF_FRAG_UNLOCK();
1265 
1266 		if (verdict != PF_PASS)
1267 			return (PF_DROP);
1268 
1269 		if (pd->m == NULL)
1270 			return (PF_DROP);
1271 
1272 		h = mtod(pd->m, struct ip *);
1273 		pd->tot_len = htons(h->ip_len);
1274 
1275  no_fragment:
1276 		/* At this point, only IP_DF is allowed in ip_off */
1277 		if (h->ip_off & ~htons(IP_DF)) {
1278 			u_int16_t ip_off = h->ip_off;
1279 
1280 			h->ip_off &= htons(IP_DF);
1281 			h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
1282 		}
1283 	}
1284 
1285 	return (PF_PASS);
1286 
1287  bad:
1288 	DPFPRINTF(PF_DEBUG_MISC, "dropping bad fragment");
1289 	REASON_SET(reason, PFRES_FRAG);
1290  drop:
1291 	if (r != NULL && r->log)
1292 		PFLOG_PACKET(PF_DROP, *reason, r, NULL, NULL, pd, 1, NULL);
1293 
1294 	return (PF_DROP);
1295 }
1296 #endif
1297 
1298 #ifdef INET6
1299 int
pf_normalize_ip6(int off,u_short * reason,struct pf_pdesc * pd)1300 pf_normalize_ip6(int off, u_short *reason,
1301     struct pf_pdesc *pd)
1302 {
1303 	struct pf_krule		*r;
1304 	struct ip6_hdr		*h;
1305 	struct ip6_frag		 frag;
1306 	bool			 scrub_compat;
1307 
1308 	PF_RULES_RASSERT();
1309 
1310 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1311 	/*
1312 	 * Check if there are any scrub rules, matching or not.
1313 	 * Lack of scrub rules means:
1314 	 *  - enforced packet normalization operation just like in OpenBSD
1315 	 * With scrub rules:
1316 	 *  - packet normalization is performed if there is a matching scrub rule
1317 	 * XXX: Fragment reassembly always performed for IPv6!
1318 	 */
1319 	scrub_compat = (r != NULL);
1320 	while (r != NULL) {
1321 		pf_counter_u64_add(&r->evaluations, 1);
1322 		if (pfi_kkif_match(r->kif, pd->kif) == r->ifnot)
1323 			r = r->skip[PF_SKIP_IFP];
1324 		else if (r->direction && r->direction != pd->dir)
1325 			r = r->skip[PF_SKIP_DIR];
1326 		else if (r->af && r->af != AF_INET6)
1327 			r = r->skip[PF_SKIP_AF];
1328 		else if (r->proto && r->proto != pd->proto)
1329 			r = r->skip[PF_SKIP_PROTO];
1330 		else if (PF_MISMATCHAW(&r->src.addr,
1331 		    (struct pf_addr *)&pd->src, AF_INET6,
1332 		    r->src.neg, pd->kif, M_GETFIB(pd->m)))
1333 			r = r->skip[PF_SKIP_SRC_ADDR];
1334 		else if (PF_MISMATCHAW(&r->dst.addr,
1335 		    (struct pf_addr *)&pd->dst, AF_INET6,
1336 		    r->dst.neg, NULL, M_GETFIB(pd->m)))
1337 			r = r->skip[PF_SKIP_DST_ADDR];
1338 		else
1339 			break;
1340 	}
1341 
1342 	if (scrub_compat) {
1343 		/* With scrub rules present IPv6 normalization happens only
1344 		 * if one of rules has matched and it's not a "no scrub" rule */
1345 		if (r == NULL || r->action == PF_NOSCRUB)
1346 			return (PF_PASS);
1347 
1348 		pf_counter_u64_critical_enter();
1349 		pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1);
1350 		pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len);
1351 		pf_counter_u64_critical_exit();
1352 		pf_rule_to_actions(r, &pd->act);
1353 	}
1354 
1355 	if (!pf_pull_hdr(pd->m, off, &frag, sizeof(frag), NULL, reason, AF_INET6))
1356 		return (PF_DROP);
1357 
1358 	/* Offset now points to data portion. */
1359 	off += sizeof(frag);
1360 
1361 	if (pd->virtual_proto == PF_VPROTO_FRAGMENT) {
1362 		/* Returns PF_DROP or *m0 is NULL or completely reassembled
1363 		 * mbuf. */
1364 		if (pf_reassemble6(&pd->m, &frag, off, pd->extoff, reason) != PF_PASS)
1365 			return (PF_DROP);
1366 		if (pd->m == NULL)
1367 			return (PF_DROP);
1368 		h = mtod(pd->m, struct ip6_hdr *);
1369 		pd->tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr);
1370 	}
1371 
1372 	return (PF_PASS);
1373 }
1374 #endif /* INET6 */
1375 
1376 int
pf_normalize_tcp(struct pf_pdesc * pd)1377 pf_normalize_tcp(struct pf_pdesc *pd)
1378 {
1379 	struct pf_krule	*r, *rm = NULL;
1380 	struct tcphdr	*th = &pd->hdr.tcp;
1381 	int		 rewrite = 0;
1382 	u_short		 reason;
1383 	u_int16_t	 flags;
1384 	sa_family_t	 af = pd->af;
1385 	int		 srs;
1386 
1387 	PF_RULES_RASSERT();
1388 
1389 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1390 	/* Check if there any scrub rules. Lack of scrub rules means enforced
1391 	 * packet normalization operation just like in OpenBSD. */
1392 	srs = (r != NULL);
1393 	while (r != NULL) {
1394 		pf_counter_u64_add(&r->evaluations, 1);
1395 		if (pfi_kkif_match(r->kif, pd->kif) == r->ifnot)
1396 			r = r->skip[PF_SKIP_IFP];
1397 		else if (r->direction && r->direction != pd->dir)
1398 			r = r->skip[PF_SKIP_DIR];
1399 		else if (r->af && r->af != af)
1400 			r = r->skip[PF_SKIP_AF];
1401 		else if (r->proto && r->proto != pd->proto)
1402 			r = r->skip[PF_SKIP_PROTO];
1403 		else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
1404 		    r->src.neg, pd->kif, M_GETFIB(pd->m)))
1405 			r = r->skip[PF_SKIP_SRC_ADDR];
1406 		else if (r->src.port_op && !pf_match_port(r->src.port_op,
1407 			    r->src.port[0], r->src.port[1], th->th_sport))
1408 			r = r->skip[PF_SKIP_SRC_PORT];
1409 		else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
1410 		    r->dst.neg, NULL, M_GETFIB(pd->m)))
1411 			r = r->skip[PF_SKIP_DST_ADDR];
1412 		else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
1413 			    r->dst.port[0], r->dst.port[1], th->th_dport))
1414 			r = r->skip[PF_SKIP_DST_PORT];
1415 		else if (r->os_fingerprint != PF_OSFP_ANY && !pf_osfp_match(
1416 			    pf_osfp_fingerprint(pd, th),
1417 			    r->os_fingerprint))
1418 			r = TAILQ_NEXT(r, entries);
1419 		else {
1420 			rm = r;
1421 			break;
1422 		}
1423 	}
1424 
1425 	if (srs) {
1426 		/* With scrub rules present TCP normalization happens only
1427 		 * if one of rules has matched and it's not a "no scrub" rule */
1428 		if (rm == NULL || rm->action == PF_NOSCRUB)
1429 			return (PF_PASS);
1430 
1431 		pf_counter_u64_critical_enter();
1432 		pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1);
1433 		pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len);
1434 		pf_counter_u64_critical_exit();
1435 		pf_rule_to_actions(rm, &pd->act);
1436 	}
1437 
1438 	if (rm && rm->rule_flag & PFRULE_REASSEMBLE_TCP)
1439 		pd->flags |= PFDESC_TCP_NORM;
1440 
1441 	flags = tcp_get_flags(th);
1442 	if (flags & TH_SYN) {
1443 		/* Illegal packet */
1444 		if (flags & TH_RST)
1445 			goto tcp_drop;
1446 
1447 		if (flags & TH_FIN)
1448 			goto tcp_drop;
1449 	} else {
1450 		/* Illegal packet */
1451 		if (!(flags & (TH_ACK|TH_RST)))
1452 			goto tcp_drop;
1453 	}
1454 
1455 	if (!(flags & TH_ACK)) {
1456 		/* These flags are only valid if ACK is set */
1457 		if ((flags & TH_FIN) || (flags & TH_PUSH) || (flags & TH_URG))
1458 			goto tcp_drop;
1459 	}
1460 
1461 	/* Check for illegal header length */
1462 	if (th->th_off < (sizeof(struct tcphdr) >> 2))
1463 		goto tcp_drop;
1464 
1465 	/* If flags changed, or reserved data set, then adjust */
1466 	if (flags != tcp_get_flags(th) ||
1467 	    (tcp_get_flags(th) & (TH_RES1|TH_RES2|TH_RES2)) != 0) {
1468 		u_int16_t	ov, nv;
1469 
1470 		ov = *(u_int16_t *)(&th->th_ack + 1);
1471 		flags &= ~(TH_RES1 | TH_RES2 | TH_RES3);
1472 		tcp_set_flags(th, flags);
1473 		nv = *(u_int16_t *)(&th->th_ack + 1);
1474 
1475 		th->th_sum = pf_proto_cksum_fixup(pd->m, th->th_sum, ov, nv, 0);
1476 		rewrite = 1;
1477 	}
1478 
1479 	/* Remove urgent pointer, if TH_URG is not set */
1480 	if (!(flags & TH_URG) && th->th_urp) {
1481 		th->th_sum = pf_proto_cksum_fixup(pd->m, th->th_sum, th->th_urp,
1482 		    0, 0);
1483 		th->th_urp = 0;
1484 		rewrite = 1;
1485 	}
1486 
1487 	/* copy back packet headers if we sanitized */
1488 	if (rewrite)
1489 		m_copyback(pd->m, pd->off, sizeof(*th), (caddr_t)th);
1490 
1491 	return (PF_PASS);
1492 
1493  tcp_drop:
1494 	REASON_SET(&reason, PFRES_NORM);
1495 	if (rm != NULL && r->log)
1496 		PFLOG_PACKET(PF_DROP, reason, r, NULL, NULL, pd, 1, NULL);
1497 	return (PF_DROP);
1498 }
1499 
1500 int
pf_normalize_tcp_init(struct pf_pdesc * pd,struct tcphdr * th,struct pf_state_peer * src)1501 pf_normalize_tcp_init(struct pf_pdesc *pd, struct tcphdr *th,
1502     struct pf_state_peer *src)
1503 {
1504 	u_int32_t tsval, tsecr;
1505 	int		 olen;
1506 	uint8_t		 opts[MAX_TCPOPTLEN], *opt;
1507 
1508 	KASSERT((src->scrub == NULL),
1509 	    ("pf_normalize_tcp_init: src->scrub != NULL"));
1510 
1511 	src->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT);
1512 	if (src->scrub == NULL)
1513 		return (1);
1514 
1515 	switch (pd->af) {
1516 #ifdef INET
1517 	case AF_INET: {
1518 		struct ip *h = mtod(pd->m, struct ip *);
1519 		src->scrub->pfss_ttl = h->ip_ttl;
1520 		break;
1521 	}
1522 #endif /* INET */
1523 #ifdef INET6
1524 	case AF_INET6: {
1525 		struct ip6_hdr *h = mtod(pd->m, struct ip6_hdr *);
1526 		src->scrub->pfss_ttl = h->ip6_hlim;
1527 		break;
1528 	}
1529 #endif /* INET6 */
1530 	default:
1531 		unhandled_af(pd->af);
1532 	}
1533 
1534 	/*
1535 	 * All normalizations below are only begun if we see the start of
1536 	 * the connections.  They must all set an enabled bit in pfss_flags
1537 	 */
1538 	if ((tcp_get_flags(th) & TH_SYN) == 0)
1539 		return (0);
1540 
1541 	olen = (th->th_off << 2) - sizeof(*th);
1542 	if (olen < TCPOLEN_TIMESTAMP || !pf_pull_hdr(pd->m,
1543 	    pd->off + sizeof(*th), opts, olen, NULL, NULL, pd->af))
1544 		return (0);
1545 
1546 	opt = opts;
1547 	while ((opt = pf_find_tcpopt(opt, opts, olen,
1548 	    TCPOPT_TIMESTAMP, TCPOLEN_TIMESTAMP)) != NULL) {
1549 		src->scrub->pfss_flags |= PFSS_TIMESTAMP;
1550 		src->scrub->pfss_ts_mod = arc4random();
1551 		/* note PFSS_PAWS not set yet */
1552 		memcpy(&tsval, &opt[2], sizeof(u_int32_t));
1553 		memcpy(&tsecr, &opt[6], sizeof(u_int32_t));
1554 		src->scrub->pfss_tsval0 = ntohl(tsval);
1555 		src->scrub->pfss_tsval = ntohl(tsval);
1556 		src->scrub->pfss_tsecr = ntohl(tsecr);
1557 		getmicrouptime(&src->scrub->pfss_last);
1558 
1559 		opt += opt[1];
1560 	}
1561 
1562 	return (0);
1563 }
1564 
1565 void
pf_normalize_tcp_cleanup(struct pf_kstate * state)1566 pf_normalize_tcp_cleanup(struct pf_kstate *state)
1567 {
1568 	/* XXX Note: this also cleans up SCTP. */
1569 	uma_zfree(V_pf_state_scrub_z, state->src.scrub);
1570 	uma_zfree(V_pf_state_scrub_z, state->dst.scrub);
1571 
1572 	/* Someday... flush the TCP segment reassembly descriptors. */
1573 }
1574 int
pf_normalize_sctp_init(struct pf_pdesc * pd,struct pf_state_peer * src,struct pf_state_peer * dst)1575 pf_normalize_sctp_init(struct pf_pdesc *pd, struct pf_state_peer *src,
1576     struct pf_state_peer *dst)
1577 {
1578 	src->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT);
1579 	if (src->scrub == NULL)
1580 		return (1);
1581 
1582 	dst->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT);
1583 	if (dst->scrub == NULL) {
1584 		uma_zfree(V_pf_state_scrub_z, src);
1585 		return (1);
1586 	}
1587 
1588 	dst->scrub->pfss_v_tag = pd->sctp_initiate_tag;
1589 
1590 	return (0);
1591 }
1592 
1593 int
pf_normalize_tcp_stateful(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)1594 pf_normalize_tcp_stateful(struct pf_pdesc *pd,
1595     u_short *reason, struct tcphdr *th, struct pf_kstate *state,
1596     struct pf_state_peer *src, struct pf_state_peer *dst, int *writeback)
1597 {
1598 	struct timeval uptime;
1599 	u_int tsval_from_last;
1600 	uint32_t tsval, tsecr;
1601 	int copyback = 0;
1602 	int got_ts = 0;
1603 	int olen;
1604 	uint8_t opts[MAX_TCPOPTLEN], *opt;
1605 
1606 	KASSERT((src->scrub || dst->scrub),
1607 	    ("%s: src->scrub && dst->scrub!", __func__));
1608 
1609 	/*
1610 	 * Enforce the minimum TTL seen for this connection.  Negate a common
1611 	 * technique to evade an intrusion detection system and confuse
1612 	 * firewall state code.
1613 	 */
1614 	switch (pd->af) {
1615 #ifdef INET
1616 	case AF_INET: {
1617 		if (src->scrub) {
1618 			struct ip *h = mtod(pd->m, struct ip *);
1619 			if (h->ip_ttl > src->scrub->pfss_ttl)
1620 				src->scrub->pfss_ttl = h->ip_ttl;
1621 			h->ip_ttl = src->scrub->pfss_ttl;
1622 		}
1623 		break;
1624 	}
1625 #endif /* INET */
1626 #ifdef INET6
1627 	case AF_INET6: {
1628 		if (src->scrub) {
1629 			struct ip6_hdr *h = mtod(pd->m, struct ip6_hdr *);
1630 			if (h->ip6_hlim > src->scrub->pfss_ttl)
1631 				src->scrub->pfss_ttl = h->ip6_hlim;
1632 			h->ip6_hlim = src->scrub->pfss_ttl;
1633 		}
1634 		break;
1635 	}
1636 #endif /* INET6 */
1637 	default:
1638 		unhandled_af(pd->af);
1639 	}
1640 
1641 	olen = (th->th_off << 2) - sizeof(*th);
1642 
1643 	if (olen >= TCPOLEN_TIMESTAMP &&
1644 	    ((src->scrub && (src->scrub->pfss_flags & PFSS_TIMESTAMP)) ||
1645 	    (dst->scrub && (dst->scrub->pfss_flags & PFSS_TIMESTAMP))) &&
1646 	    pf_pull_hdr(pd->m, pd->off + sizeof(*th), opts, olen, NULL, NULL, pd->af)) {
1647 		/* Modulate the timestamps.  Can be used for NAT detection, OS
1648 		 * uptime determination or reboot detection.
1649 		 */
1650 		opt = opts;
1651 		while ((opt = pf_find_tcpopt(opt, opts, olen,
1652 		    TCPOPT_TIMESTAMP, TCPOLEN_TIMESTAMP)) != NULL) {
1653 			uint8_t *ts = opt + 2;
1654 			uint8_t *tsr = opt + 6;
1655 
1656 			if (got_ts) {
1657 				/* Huh?  Multiple timestamps!? */
1658 				if (V_pf_status.debug >= PF_DEBUG_MISC) {
1659 					printf("pf: %s: multiple TS??", __func__);
1660 					pf_print_state(state);
1661 					printf("\n");
1662 				}
1663 				REASON_SET(reason, PFRES_TS);
1664 				return (PF_DROP);
1665 			}
1666 
1667 			memcpy(&tsval, ts, sizeof(u_int32_t));
1668 			memcpy(&tsecr, tsr, sizeof(u_int32_t));
1669 
1670 			/* modulate TS */
1671 			if (tsval && src->scrub &&
1672 			    (src->scrub->pfss_flags & PFSS_TIMESTAMP)) {
1673 				/* tsval used further on */
1674 				tsval = ntohl(tsval);
1675 				pf_patch_32(pd,
1676 				    ts, htonl(tsval + src->scrub->pfss_ts_mod),
1677 				    PF_ALGNMNT(ts - opts));
1678 				copyback = 1;
1679 			}
1680 
1681 			/* modulate TS reply if any (!0) */
1682 			if (tsecr && dst->scrub &&
1683 			    (dst->scrub->pfss_flags & PFSS_TIMESTAMP)) {
1684 				/* tsecr used further on */
1685 				tsecr = ntohl(tsecr) - dst->scrub->pfss_ts_mod;
1686 				pf_patch_32(pd, tsr, htonl(tsecr),
1687 				    PF_ALGNMNT(tsr - opts));
1688 				copyback = 1;
1689 			}
1690 
1691 			got_ts = 1;
1692 			opt += opt[1];
1693 		}
1694 
1695 		if (copyback) {
1696 			/* Copyback the options, caller copys back header */
1697 			*writeback = 1;
1698 			m_copyback(pd->m, pd->off + sizeof(*th), olen, opts);
1699 		}
1700 	}
1701 
1702 	/*
1703 	 * Must invalidate PAWS checks on connections idle for too long.
1704 	 * The fastest allowed timestamp clock is 1ms.  That turns out to
1705 	 * be about 24 days before it wraps.  XXX Right now our lowerbound
1706 	 * TS echo check only works for the first 12 days of a connection
1707 	 * when the TS has exhausted half its 32bit space
1708 	 */
1709 #define TS_MAX_IDLE	(24*24*60*60)
1710 #define TS_MAX_CONN	(12*24*60*60)	/* XXX remove when better tsecr check */
1711 
1712 	getmicrouptime(&uptime);
1713 	if (src->scrub && (src->scrub->pfss_flags & PFSS_PAWS) &&
1714 	    (uptime.tv_sec - src->scrub->pfss_last.tv_sec > TS_MAX_IDLE ||
1715 	    time_uptime - (state->creation / 1000) > TS_MAX_CONN))  {
1716 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
1717 			DPFPRINTF(PF_DEBUG_MISC, "src idled out of PAWS");
1718 			pf_print_state(state);
1719 			printf("\n");
1720 		}
1721 		src->scrub->pfss_flags = (src->scrub->pfss_flags & ~PFSS_PAWS)
1722 		    | PFSS_PAWS_IDLED;
1723 	}
1724 	if (dst->scrub && (dst->scrub->pfss_flags & PFSS_PAWS) &&
1725 	    uptime.tv_sec - dst->scrub->pfss_last.tv_sec > TS_MAX_IDLE) {
1726 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
1727 			DPFPRINTF(PF_DEBUG_MISC, "dst idled out of PAWS");
1728 			pf_print_state(state);
1729 			printf("\n");
1730 		}
1731 		dst->scrub->pfss_flags = (dst->scrub->pfss_flags & ~PFSS_PAWS)
1732 		    | PFSS_PAWS_IDLED;
1733 	}
1734 
1735 	if (got_ts && src->scrub && dst->scrub &&
1736 	    (src->scrub->pfss_flags & PFSS_PAWS) &&
1737 	    (dst->scrub->pfss_flags & PFSS_PAWS)) {
1738 		/* Validate that the timestamps are "in-window".
1739 		 * RFC1323 describes TCP Timestamp options that allow
1740 		 * measurement of RTT (round trip time) and PAWS
1741 		 * (protection against wrapped sequence numbers).  PAWS
1742 		 * gives us a set of rules for rejecting packets on
1743 		 * long fat pipes (packets that were somehow delayed
1744 		 * in transit longer than the time it took to send the
1745 		 * full TCP sequence space of 4Gb).  We can use these
1746 		 * rules and infer a few others that will let us treat
1747 		 * the 32bit timestamp and the 32bit echoed timestamp
1748 		 * as sequence numbers to prevent a blind attacker from
1749 		 * inserting packets into a connection.
1750 		 *
1751 		 * RFC1323 tells us:
1752 		 *  - The timestamp on this packet must be greater than
1753 		 *    or equal to the last value echoed by the other
1754 		 *    endpoint.  The RFC says those will be discarded
1755 		 *    since it is a dup that has already been acked.
1756 		 *    This gives us a lowerbound on the timestamp.
1757 		 *        timestamp >= other last echoed timestamp
1758 		 *  - The timestamp will be less than or equal to
1759 		 *    the last timestamp plus the time between the
1760 		 *    last packet and now.  The RFC defines the max
1761 		 *    clock rate as 1ms.  We will allow clocks to be
1762 		 *    up to 10% fast and will allow a total difference
1763 		 *    or 30 seconds due to a route change.  And this
1764 		 *    gives us an upperbound on the timestamp.
1765 		 *        timestamp <= last timestamp + max ticks
1766 		 *    We have to be careful here.  Windows will send an
1767 		 *    initial timestamp of zero and then initialize it
1768 		 *    to a random value after the 3whs; presumably to
1769 		 *    avoid a DoS by having to call an expensive RNG
1770 		 *    during a SYN flood.  Proof MS has at least one
1771 		 *    good security geek.
1772 		 *
1773 		 *  - The TCP timestamp option must also echo the other
1774 		 *    endpoints timestamp.  The timestamp echoed is the
1775 		 *    one carried on the earliest unacknowledged segment
1776 		 *    on the left edge of the sequence window.  The RFC
1777 		 *    states that the host will reject any echoed
1778 		 *    timestamps that were larger than any ever sent.
1779 		 *    This gives us an upperbound on the TS echo.
1780 		 *        tescr <= largest_tsval
1781 		 *  - The lowerbound on the TS echo is a little more
1782 		 *    tricky to determine.  The other endpoint's echoed
1783 		 *    values will not decrease.  But there may be
1784 		 *    network conditions that re-order packets and
1785 		 *    cause our view of them to decrease.  For now the
1786 		 *    only lowerbound we can safely determine is that
1787 		 *    the TS echo will never be less than the original
1788 		 *    TS.  XXX There is probably a better lowerbound.
1789 		 *    Remove TS_MAX_CONN with better lowerbound check.
1790 		 *        tescr >= other original TS
1791 		 *
1792 		 * It is also important to note that the fastest
1793 		 * timestamp clock of 1ms will wrap its 32bit space in
1794 		 * 24 days.  So we just disable TS checking after 24
1795 		 * days of idle time.  We actually must use a 12d
1796 		 * connection limit until we can come up with a better
1797 		 * lowerbound to the TS echo check.
1798 		 */
1799 		struct timeval delta_ts;
1800 		int ts_fudge;
1801 
1802 		/*
1803 		 * PFTM_TS_DIFF is how many seconds of leeway to allow
1804 		 * a host's timestamp.  This can happen if the previous
1805 		 * packet got delayed in transit for much longer than
1806 		 * this packet.
1807 		 */
1808 		if ((ts_fudge = state->rule->timeout[PFTM_TS_DIFF]) == 0)
1809 			ts_fudge = V_pf_default_rule.timeout[PFTM_TS_DIFF];
1810 
1811 		/* Calculate max ticks since the last timestamp */
1812 #define TS_MAXFREQ	1100		/* RFC max TS freq of 1Khz + 10% skew */
1813 #define TS_MICROSECS	1000000		/* microseconds per second */
1814 		delta_ts = uptime;
1815 		timevalsub(&delta_ts, &src->scrub->pfss_last);
1816 		tsval_from_last = (delta_ts.tv_sec + ts_fudge) * TS_MAXFREQ;
1817 		tsval_from_last += delta_ts.tv_usec / (TS_MICROSECS/TS_MAXFREQ);
1818 
1819 		if ((src->state >= TCPS_ESTABLISHED &&
1820 		    dst->state >= TCPS_ESTABLISHED) &&
1821 		    (SEQ_LT(tsval, dst->scrub->pfss_tsecr) ||
1822 		    SEQ_GT(tsval, src->scrub->pfss_tsval + tsval_from_last) ||
1823 		    (tsecr && (SEQ_GT(tsecr, dst->scrub->pfss_tsval) ||
1824 		    SEQ_LT(tsecr, dst->scrub->pfss_tsval0))))) {
1825 			/* Bad RFC1323 implementation or an insertion attack.
1826 			 *
1827 			 * - Solaris 2.6 and 2.7 are known to send another ACK
1828 			 *   after the FIN,FIN|ACK,ACK closing that carries
1829 			 *   an old timestamp.
1830 			 */
1831 
1832 			DPFPRINTF(PF_DEBUG_MISC, "Timestamp failed %c%c%c%c",
1833 			    SEQ_LT(tsval, dst->scrub->pfss_tsecr) ? '0' : ' ',
1834 			    SEQ_GT(tsval, src->scrub->pfss_tsval +
1835 			    tsval_from_last) ? '1' : ' ',
1836 			    SEQ_GT(tsecr, dst->scrub->pfss_tsval) ? '2' : ' ',
1837 			    SEQ_LT(tsecr, dst->scrub->pfss_tsval0)? '3' : ' ');
1838 			DPFPRINTF(PF_DEBUG_MISC, " tsval: %u  tsecr: %u  +ticks: "
1839 			    "%u  idle: %jus %lums",
1840 			    tsval, tsecr, tsval_from_last,
1841 			    (uintmax_t)delta_ts.tv_sec,
1842 			    delta_ts.tv_usec / 1000);
1843 			DPFPRINTF(PF_DEBUG_MISC, " src->tsval: %u  tsecr: %u",
1844 			    src->scrub->pfss_tsval, src->scrub->pfss_tsecr);
1845 			DPFPRINTF(PF_DEBUG_MISC, " dst->tsval: %u  tsecr: %u  "
1846 			    "tsval0: %u", dst->scrub->pfss_tsval,
1847 			    dst->scrub->pfss_tsecr, dst->scrub->pfss_tsval0);
1848 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
1849 				pf_print_state(state);
1850 				pf_print_flags(tcp_get_flags(th));
1851 				printf("\n");
1852 			}
1853 			REASON_SET(reason, PFRES_TS);
1854 			return (PF_DROP);
1855 		}
1856 
1857 		/* XXX I'd really like to require tsecr but it's optional */
1858 
1859 	} else if (!got_ts && (tcp_get_flags(th) & TH_RST) == 0 &&
1860 	    ((src->state == TCPS_ESTABLISHED && dst->state == TCPS_ESTABLISHED)
1861 	    || pd->p_len > 0 || (tcp_get_flags(th) & TH_SYN)) &&
1862 	    src->scrub && dst->scrub &&
1863 	    (src->scrub->pfss_flags & PFSS_PAWS) &&
1864 	    (dst->scrub->pfss_flags & PFSS_PAWS)) {
1865 		/* Didn't send a timestamp.  Timestamps aren't really useful
1866 		 * when:
1867 		 *  - connection opening or closing (often not even sent).
1868 		 *    but we must not let an attacker to put a FIN on a
1869 		 *    data packet to sneak it through our ESTABLISHED check.
1870 		 *  - on a TCP reset.  RFC suggests not even looking at TS.
1871 		 *  - on an empty ACK.  The TS will not be echoed so it will
1872 		 *    probably not help keep the RTT calculation in sync and
1873 		 *    there isn't as much danger when the sequence numbers
1874 		 *    got wrapped.  So some stacks don't include TS on empty
1875 		 *    ACKs :-(
1876 		 *
1877 		 * To minimize the disruption to mostly RFC1323 conformant
1878 		 * stacks, we will only require timestamps on data packets.
1879 		 *
1880 		 * And what do ya know, we cannot require timestamps on data
1881 		 * packets.  There appear to be devices that do legitimate
1882 		 * TCP connection hijacking.  There are HTTP devices that allow
1883 		 * a 3whs (with timestamps) and then buffer the HTTP request.
1884 		 * If the intermediate device has the HTTP response cache, it
1885 		 * will spoof the response but not bother timestamping its
1886 		 * packets.  So we can look for the presence of a timestamp in
1887 		 * the first data packet and if there, require it in all future
1888 		 * packets.
1889 		 */
1890 
1891 		if (pd->p_len > 0 && (src->scrub->pfss_flags & PFSS_DATA_TS)) {
1892 			/*
1893 			 * Hey!  Someone tried to sneak a packet in.  Or the
1894 			 * stack changed its RFC1323 behavior?!?!
1895 			 */
1896 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
1897 				DPFPRINTF(PF_DEBUG_MISC, "Did not receive expected "
1898 				    "RFC1323 timestamp");
1899 				pf_print_state(state);
1900 				pf_print_flags(tcp_get_flags(th));
1901 				printf("\n");
1902 			}
1903 			REASON_SET(reason, PFRES_TS);
1904 			return (PF_DROP);
1905 		}
1906 	}
1907 
1908 	/*
1909 	 * We will note if a host sends his data packets with or without
1910 	 * timestamps.  And require all data packets to contain a timestamp
1911 	 * if the first does.  PAWS implicitly requires that all data packets be
1912 	 * timestamped.  But I think there are middle-man devices that hijack
1913 	 * TCP streams immediately after the 3whs and don't timestamp their
1914 	 * packets (seen in a WWW accelerator or cache).
1915 	 */
1916 	if (pd->p_len > 0 && src->scrub && (src->scrub->pfss_flags &
1917 	    (PFSS_TIMESTAMP|PFSS_DATA_TS|PFSS_DATA_NOTS)) == PFSS_TIMESTAMP) {
1918 		if (got_ts)
1919 			src->scrub->pfss_flags |= PFSS_DATA_TS;
1920 		else {
1921 			src->scrub->pfss_flags |= PFSS_DATA_NOTS;
1922 			if (V_pf_status.debug >= PF_DEBUG_MISC && dst->scrub &&
1923 			    (dst->scrub->pfss_flags & PFSS_TIMESTAMP)) {
1924 				/* Don't warn if other host rejected RFC1323 */
1925 				DPFPRINTF(PF_DEBUG_MISC, "Broken RFC1323 stack did "
1926 				    "not timestamp data packet. Disabled PAWS "
1927 				    "security.");
1928 				pf_print_state(state);
1929 				pf_print_flags(tcp_get_flags(th));
1930 				printf("\n");
1931 			}
1932 		}
1933 	}
1934 
1935 	/*
1936 	 * Update PAWS values
1937 	 */
1938 	if (got_ts && src->scrub && PFSS_TIMESTAMP == (src->scrub->pfss_flags &
1939 	    (PFSS_PAWS_IDLED|PFSS_TIMESTAMP))) {
1940 		getmicrouptime(&src->scrub->pfss_last);
1941 		if (SEQ_GEQ(tsval, src->scrub->pfss_tsval) ||
1942 		    (src->scrub->pfss_flags & PFSS_PAWS) == 0)
1943 			src->scrub->pfss_tsval = tsval;
1944 
1945 		if (tsecr) {
1946 			if (SEQ_GEQ(tsecr, src->scrub->pfss_tsecr) ||
1947 			    (src->scrub->pfss_flags & PFSS_PAWS) == 0)
1948 				src->scrub->pfss_tsecr = tsecr;
1949 
1950 			if ((src->scrub->pfss_flags & PFSS_PAWS) == 0 &&
1951 			    (SEQ_LT(tsval, src->scrub->pfss_tsval0) ||
1952 			    src->scrub->pfss_tsval0 == 0)) {
1953 				/* tsval0 MUST be the lowest timestamp */
1954 				src->scrub->pfss_tsval0 = tsval;
1955 			}
1956 
1957 			/* Only fully initialized after a TS gets echoed */
1958 			if ((src->scrub->pfss_flags & PFSS_PAWS) == 0)
1959 				src->scrub->pfss_flags |= PFSS_PAWS;
1960 		}
1961 	}
1962 
1963 	/* I have a dream....  TCP segment reassembly.... */
1964 	return (0);
1965 }
1966 
1967 int
pf_normalize_mss(struct pf_pdesc * pd)1968 pf_normalize_mss(struct pf_pdesc *pd)
1969 {
1970 	int		 olen, optsoff;
1971 	uint8_t		 opts[MAX_TCPOPTLEN], *opt;
1972 
1973 	olen = (pd->hdr.tcp.th_off << 2) - sizeof(struct tcphdr);
1974 	optsoff = pd->off + sizeof(struct tcphdr);
1975 	if (olen < TCPOLEN_MAXSEG ||
1976 	    !pf_pull_hdr(pd->m, optsoff, opts, olen, NULL, NULL, pd->af))
1977 		return (0);
1978 
1979 	opt = opts;
1980 	while ((opt = pf_find_tcpopt(opt, opts, olen,
1981 	    TCPOPT_MAXSEG, TCPOLEN_MAXSEG)) != NULL) {
1982 		uint16_t	 mss;
1983 		uint8_t		*mssp = opt + 2;
1984 		memcpy(&mss, mssp, sizeof(mss));
1985 		if (ntohs(mss) > pd->act.max_mss) {
1986 			size_t mssoffopts = mssp - opts;
1987 			pf_patch_16(pd, &mss,
1988 			    htons(pd->act.max_mss), PF_ALGNMNT(mssoffopts));
1989 			m_copyback(pd->m, optsoff + mssoffopts,
1990 			    sizeof(mss), (caddr_t)&mss);
1991 			m_copyback(pd->m, pd->off,
1992 			    sizeof(struct tcphdr), (caddr_t)&pd->hdr.tcp);
1993 		}
1994 
1995 		opt += opt[1];
1996 	}
1997 
1998 	return (0);
1999 }
2000 
2001 int
pf_scan_sctp(struct pf_pdesc * pd)2002 pf_scan_sctp(struct pf_pdesc *pd)
2003 {
2004 	struct sctp_chunkhdr ch = { };
2005 	int chunk_off = sizeof(struct sctphdr);
2006 	int chunk_start;
2007 	int ret;
2008 
2009 	while (pd->off + chunk_off < pd->tot_len) {
2010 		if (!pf_pull_hdr(pd->m, pd->off + chunk_off, &ch, sizeof(ch), NULL,
2011 		    NULL, pd->af))
2012 			return (PF_DROP);
2013 
2014 		/* Length includes the header, this must be at least 4. */
2015 		if (ntohs(ch.chunk_length) < 4)
2016 			return (PF_DROP);
2017 
2018 		chunk_start = chunk_off;
2019 		chunk_off += roundup(ntohs(ch.chunk_length), 4);
2020 
2021 		switch (ch.chunk_type) {
2022 		case SCTP_INITIATION:
2023 		case SCTP_INITIATION_ACK: {
2024 			struct sctp_init_chunk init;
2025 
2026 			if (!pf_pull_hdr(pd->m, pd->off + chunk_start, &init,
2027 			    sizeof(init), NULL, NULL, pd->af))
2028 				return (PF_DROP);
2029 
2030 			/*
2031 			 * RFC 9620, Section 3.3.2, "The Initiate Tag is allowed to have
2032 			 * any value except 0."
2033 			 */
2034 			if (init.init.initiate_tag == 0)
2035 				return (PF_DROP);
2036 			if (init.init.num_inbound_streams == 0)
2037 				return (PF_DROP);
2038 			if (init.init.num_outbound_streams == 0)
2039 				return (PF_DROP);
2040 			if (ntohl(init.init.a_rwnd) < SCTP_MIN_RWND)
2041 				return (PF_DROP);
2042 
2043 			/*
2044 			 * RFC 9260, Section 3.1, INIT chunks MUST have zero
2045 			 * verification tag.
2046 			 */
2047 			if (ch.chunk_type == SCTP_INITIATION &&
2048 			    pd->hdr.sctp.v_tag != 0)
2049 				return (PF_DROP);
2050 
2051 			pd->sctp_initiate_tag = init.init.initiate_tag;
2052 
2053 			if (ch.chunk_type == SCTP_INITIATION)
2054 				pd->sctp_flags |= PFDESC_SCTP_INIT;
2055 			else
2056 				pd->sctp_flags |= PFDESC_SCTP_INIT_ACK;
2057 
2058 			ret = pf_multihome_scan_init(pd->off + chunk_start,
2059 			    ntohs(init.ch.chunk_length), pd);
2060 			if (ret != PF_PASS)
2061 				return (ret);
2062 
2063 			break;
2064 		}
2065 		case SCTP_ABORT_ASSOCIATION:
2066 			pd->sctp_flags |= PFDESC_SCTP_ABORT;
2067 			break;
2068 		case SCTP_SHUTDOWN:
2069 		case SCTP_SHUTDOWN_ACK:
2070 			pd->sctp_flags |= PFDESC_SCTP_SHUTDOWN;
2071 			break;
2072 		case SCTP_SHUTDOWN_COMPLETE:
2073 			pd->sctp_flags |= PFDESC_SCTP_SHUTDOWN_COMPLETE;
2074 			break;
2075 		case SCTP_COOKIE_ECHO:
2076 			pd->sctp_flags |= PFDESC_SCTP_COOKIE;
2077 			break;
2078 		case SCTP_COOKIE_ACK:
2079 			pd->sctp_flags |= PFDESC_SCTP_COOKIE_ACK;
2080 			break;
2081 		case SCTP_DATA:
2082 			pd->sctp_flags |= PFDESC_SCTP_DATA;
2083 			break;
2084 		case SCTP_HEARTBEAT_REQUEST:
2085 			pd->sctp_flags |= PFDESC_SCTP_HEARTBEAT;
2086 			break;
2087 		case SCTP_HEARTBEAT_ACK:
2088 			pd->sctp_flags |= PFDESC_SCTP_HEARTBEAT_ACK;
2089 			break;
2090 		case SCTP_ASCONF:
2091 			pd->sctp_flags |= PFDESC_SCTP_ASCONF;
2092 
2093 			ret = pf_multihome_scan_asconf(pd->off + chunk_start,
2094 			    ntohs(ch.chunk_length), pd);
2095 			if (ret != PF_PASS)
2096 				return (ret);
2097 			break;
2098 		default:
2099 			pd->sctp_flags |= PFDESC_SCTP_OTHER;
2100 			break;
2101 		}
2102 	}
2103 
2104 	/* Validate chunk lengths vs. packet length. */
2105 	if (pd->off + chunk_off != pd->tot_len)
2106 		return (PF_DROP);
2107 
2108 	/*
2109 	 * INIT, INIT_ACK or SHUTDOWN_COMPLETE chunks must always be the only
2110 	 * one in a packet.
2111 	 */
2112 	if ((pd->sctp_flags & PFDESC_SCTP_INIT) &&
2113 	    (pd->sctp_flags & ~PFDESC_SCTP_INIT))
2114 		return (PF_DROP);
2115 	if ((pd->sctp_flags & PFDESC_SCTP_INIT_ACK) &&
2116 	    (pd->sctp_flags & ~PFDESC_SCTP_INIT_ACK))
2117 		return (PF_DROP);
2118 	if ((pd->sctp_flags & PFDESC_SCTP_SHUTDOWN_COMPLETE) &&
2119 	    (pd->sctp_flags & ~PFDESC_SCTP_SHUTDOWN_COMPLETE))
2120 		return (PF_DROP);
2121 	if ((pd->sctp_flags & PFDESC_SCTP_ABORT) &&
2122 	    (pd->sctp_flags & PFDESC_SCTP_DATA)) {
2123 		/*
2124 		 * RFC4960 3.3.7: DATA chunks MUST NOT be
2125 		 * bundled with ABORT.
2126 		 */
2127 		return (PF_DROP);
2128 	}
2129 
2130 	return (PF_PASS);
2131 }
2132 
2133 int
pf_normalize_sctp(struct pf_pdesc * pd)2134 pf_normalize_sctp(struct pf_pdesc *pd)
2135 {
2136 	struct pf_krule	*r, *rm = NULL;
2137 	struct sctphdr	*sh = &pd->hdr.sctp;
2138 	u_short		 reason;
2139 	sa_family_t	 af = pd->af;
2140 	int		 srs;
2141 
2142 	PF_RULES_RASSERT();
2143 
2144 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
2145 	/* Check if there any scrub rules. Lack of scrub rules means enforced
2146 	 * packet normalization operation just like in OpenBSD. */
2147 	srs = (r != NULL);
2148 	while (r != NULL) {
2149 		pf_counter_u64_add(&r->evaluations, 1);
2150 		if (pfi_kkif_match(r->kif, pd->kif) == r->ifnot)
2151 			r = r->skip[PF_SKIP_IFP];
2152 		else if (r->direction && r->direction != pd->dir)
2153 			r = r->skip[PF_SKIP_DIR];
2154 		else if (r->af && r->af != af)
2155 			r = r->skip[PF_SKIP_AF];
2156 		else if (r->proto && r->proto != pd->proto)
2157 			r = r->skip[PF_SKIP_PROTO];
2158 		else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
2159 		    r->src.neg, pd->kif, M_GETFIB(pd->m)))
2160 			r = r->skip[PF_SKIP_SRC_ADDR];
2161 		else if (r->src.port_op && !pf_match_port(r->src.port_op,
2162 			    r->src.port[0], r->src.port[1], sh->src_port))
2163 			r = r->skip[PF_SKIP_SRC_PORT];
2164 		else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
2165 		    r->dst.neg, NULL, M_GETFIB(pd->m)))
2166 			r = r->skip[PF_SKIP_DST_ADDR];
2167 		else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
2168 			    r->dst.port[0], r->dst.port[1], sh->dest_port))
2169 			r = r->skip[PF_SKIP_DST_PORT];
2170 		else {
2171 			rm = r;
2172 			break;
2173 		}
2174 	}
2175 
2176 	if (srs) {
2177 		/* With scrub rules present SCTP normalization happens only
2178 		 * if one of rules has matched and it's not a "no scrub" rule */
2179 		if (rm == NULL || rm->action == PF_NOSCRUB)
2180 			return (PF_PASS);
2181 
2182 		pf_counter_u64_critical_enter();
2183 		pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1);
2184 		pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len);
2185 		pf_counter_u64_critical_exit();
2186 	}
2187 
2188 	/* Verify we're a multiple of 4 bytes long */
2189 	if ((pd->tot_len - pd->off - sizeof(struct sctphdr)) % 4)
2190 		goto sctp_drop;
2191 
2192 	/* INIT chunk needs to be the only chunk */
2193 	if (pd->sctp_flags & PFDESC_SCTP_INIT)
2194 		if (pd->sctp_flags & ~PFDESC_SCTP_INIT)
2195 			goto sctp_drop;
2196 
2197 	return (PF_PASS);
2198 
2199 sctp_drop:
2200 	REASON_SET(&reason, PFRES_NORM);
2201 	if (rm != NULL && r->log)
2202 		PFLOG_PACKET(PF_DROP, reason, r, NULL, NULL, pd,
2203 		    1, NULL);
2204 
2205 	return (PF_DROP);
2206 }
2207 
2208 #if defined(INET) || defined(INET6)
2209 void
pf_scrub(struct pf_pdesc * pd)2210 pf_scrub(struct pf_pdesc *pd)
2211 {
2212 
2213 	struct ip		*h = mtod(pd->m, struct ip *);
2214 #ifdef INET6
2215 	struct ip6_hdr		*h6 = mtod(pd->m, struct ip6_hdr *);
2216 #endif /* INET6 */
2217 
2218 	/* Clear IP_DF if no-df was requested */
2219 	if (pd->af == AF_INET && pd->act.flags & PFSTATE_NODF &&
2220 	    h->ip_off & htons(IP_DF))
2221 	{
2222 		u_int16_t ip_off = h->ip_off;
2223 
2224 		h->ip_off &= htons(~IP_DF);
2225 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
2226 	}
2227 
2228 	/* Enforce a minimum ttl, may cause endless packet loops */
2229 	if (pd->af == AF_INET && pd->act.min_ttl &&
2230 	    h->ip_ttl < pd->act.min_ttl) {
2231 		u_int16_t ip_ttl = h->ip_ttl;
2232 
2233 		h->ip_ttl = pd->act.min_ttl;
2234 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_ttl, h->ip_ttl, 0);
2235 	}
2236 #ifdef INET6
2237 	/* Enforce a minimum ttl, may cause endless packet loops */
2238 	if (pd->af == AF_INET6 && pd->act.min_ttl &&
2239 	    h6->ip6_hlim < pd->act.min_ttl)
2240 		h6->ip6_hlim = pd->act.min_ttl;
2241 #endif /* INET6 */
2242 	/* Enforce tos */
2243 	if (pd->act.flags & PFSTATE_SETTOS) {
2244 		switch (pd->af) {
2245 		case AF_INET: {
2246 			u_int16_t	ov, nv;
2247 
2248 			ov = *(u_int16_t *)h;
2249 			h->ip_tos = pd->act.set_tos | (h->ip_tos & IPTOS_ECN_MASK);
2250 			nv = *(u_int16_t *)h;
2251 
2252 			h->ip_sum = pf_cksum_fixup(h->ip_sum, ov, nv, 0);
2253 			break;
2254 		}
2255 #ifdef INET6
2256 		case AF_INET6:
2257 			h6->ip6_flow &= IPV6_FLOWLABEL_MASK | IPV6_VERSION_MASK;
2258 			h6->ip6_flow |= htonl((pd->act.set_tos | IPV6_ECN(h6)) << 20);
2259 			break;
2260 #endif /* INET6 */
2261 		}
2262 	}
2263 
2264 	/* random-id, but not for fragments */
2265 #ifdef INET
2266 	if (pd->af == AF_INET &&
2267 	    pd->act.flags & PFSTATE_RANDOMID && !(h->ip_off & ~htons(IP_DF))) {
2268 		uint16_t ip_id = h->ip_id;
2269 
2270 		ip_fillid(h, V_ip_random_id);
2271 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_id, h->ip_id, 0);
2272 	}
2273 #endif /* INET */
2274 }
2275 #endif /* INET || INET6 */
2276