xref: /freebsd/sys/netpfil/pf/pf_norm.c (revision 029532e77b92d0c74976e3e3fc79a0ca5e0e3dc0)
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 VNET_DEFINE(uma_zone_t, pf_anchor_z);
122 VNET_DEFINE(uma_zone_t, pf_eth_anchor_z);
123 
124 TAILQ_HEAD(pf_fragqueue, pf_fragment);
125 TAILQ_HEAD(pf_cachequeue, pf_fragment);
126 RB_HEAD(pf_frnode_tree, pf_frnode);
127 VNET_DEFINE_STATIC(struct pf_fragqueue,	pf_fragqueue);
128 #define	V_pf_fragqueue			VNET(pf_fragqueue)
129 static __inline int	pf_frnode_compare(struct pf_frnode *,
130 			    struct pf_frnode *);
131 VNET_DEFINE_STATIC(struct pf_frnode_tree, pf_frnode_tree);
132 #define	V_pf_frnode_tree		VNET(pf_frnode_tree)
133 RB_PROTOTYPE(pf_frnode_tree, pf_frnode, fn_entry, pf_frnode_compare);
134 RB_GENERATE(pf_frnode_tree, pf_frnode, fn_entry, pf_frnode_compare);
135 
136 static int		 pf_frag_compare(struct pf_fragment *,
137 			    struct pf_fragment *);
138 static RB_PROTOTYPE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
139 static RB_GENERATE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
140 
141 static void	pf_flush_fragments(void);
142 static void	pf_free_fragment(struct pf_fragment *);
143 
144 static struct pf_frent *pf_create_fragment(u_short *);
145 static int	pf_frent_holes(struct pf_frent *frent);
146 static struct pf_fragment	*pf_find_fragment(struct pf_frnode *, u_int32_t);
147 static inline int	pf_frent_index(struct pf_frent *);
148 static int	pf_frent_insert(struct pf_fragment *,
149 			    struct pf_frent *, struct pf_frent *);
150 void			pf_frent_remove(struct pf_fragment *,
151 			    struct pf_frent *);
152 struct pf_frent		*pf_frent_previous(struct pf_fragment *,
153 			    struct pf_frent *);
154 static struct pf_fragment *pf_fillup_fragment(struct pf_frnode *, u_int32_t,
155 		    struct pf_frent *, u_short *);
156 static struct mbuf *pf_join_fragment(struct pf_fragment *);
157 #ifdef INET
158 static int	pf_reassemble(struct mbuf **, u_short *);
159 #endif	/* INET */
160 #ifdef INET6
161 static int	pf_reassemble6(struct mbuf **,
162 		    struct ip6_frag *, uint16_t, uint16_t, u_short *);
163 #endif	/* INET6 */
164 
165 #ifdef INET
166 static void
pf_ip2key(struct ip * ip,struct pf_frnode * key)167 pf_ip2key(struct ip *ip, struct pf_frnode *key)
168 {
169 
170 	key->fn_src.v4 = ip->ip_src;
171 	key->fn_dst.v4 = ip->ip_dst;
172 	key->fn_af = AF_INET;
173 	key->fn_proto = ip->ip_p;
174 }
175 #endif	/* INET */
176 
177 void
pf_normalize_init(void)178 pf_normalize_init(void)
179 {
180 
181 	V_pf_frag_z = uma_zcreate("pf frags", sizeof(struct pf_fragment),
182 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
183 	V_pf_frnode_z = uma_zcreate("pf fragment node",
184 	    sizeof(struct pf_frnode), NULL, NULL, NULL, NULL,
185 	    UMA_ALIGN_PTR, 0);
186 	V_pf_frent_z = uma_zcreate("pf frag entries", sizeof(struct pf_frent),
187 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
188 	V_pf_state_scrub_z = uma_zcreate("pf state scrubs",
189 	    sizeof(struct pf_state_scrub),  NULL, NULL, NULL, NULL,
190 	    UMA_ALIGN_PTR, 0);
191 
192 	mtx_init(&V_pf_frag_mtx, "pf fragments", NULL, MTX_DEF);
193 
194 	V_pf_limits[PF_LIMIT_FRAGS].zone = V_pf_frent_z;
195 	V_pf_limits[PF_LIMIT_FRAGS].limit = PFFRAG_FRENT_HIWAT;
196 	uma_zone_set_max(V_pf_frent_z, PFFRAG_FRENT_HIWAT);
197 	uma_zone_set_warning(V_pf_frent_z, "PF frag entries limit reached");
198 
199 	TAILQ_INIT(&V_pf_fragqueue);
200 }
201 
202 void
pf_normalize_cleanup(void)203 pf_normalize_cleanup(void)
204 {
205 
206 	uma_zdestroy(V_pf_state_scrub_z);
207 	uma_zdestroy(V_pf_frent_z);
208 	uma_zdestroy(V_pf_frnode_z);
209 	uma_zdestroy(V_pf_frag_z);
210 
211 	mtx_destroy(&V_pf_frag_mtx);
212 }
213 
214 static int
pf_frnode_compare(struct pf_frnode * a,struct pf_frnode * b)215 pf_frnode_compare(struct pf_frnode *a, struct pf_frnode *b)
216 {
217 	int	diff;
218 
219 	if ((diff = a->fn_proto - b->fn_proto) != 0)
220 		return (diff);
221 	if ((diff = a->fn_af - b->fn_af) != 0)
222 		return (diff);
223 	if ((diff = pf_addr_cmp(&a->fn_src, &b->fn_src, a->fn_af)) != 0)
224 		return (diff);
225 	if ((diff = pf_addr_cmp(&a->fn_dst, &b->fn_dst, a->fn_af)) != 0)
226 		return (diff);
227 	return (0);
228 }
229 
230 static __inline int
pf_frag_compare(struct pf_fragment * a,struct pf_fragment * b)231 pf_frag_compare(struct pf_fragment *a, struct pf_fragment *b)
232 {
233 	int	diff;
234 
235 	if ((diff = a->fr_id - b->fr_id) != 0)
236 		return (diff);
237 
238 	return (0);
239 }
240 
241 void
pf_purge_expired_fragments(void)242 pf_purge_expired_fragments(void)
243 {
244 	u_int32_t	expire = time_uptime -
245 			    V_pf_default_rule.timeout[PFTM_FRAG];
246 
247 	pf_purge_fragments(expire);
248 }
249 
250 void
pf_purge_fragments(uint32_t expire)251 pf_purge_fragments(uint32_t expire)
252 {
253 	struct pf_fragment	*frag;
254 
255 	PF_FRAG_LOCK();
256 	while ((frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue)) != NULL) {
257 		if (frag->fr_timeout > expire)
258 			break;
259 
260 		DPFPRINTF(PF_DEBUG_MISC, "expiring %d(%p)",
261 		    frag->fr_id, frag);
262 		pf_free_fragment(frag);
263 	}
264 
265 	PF_FRAG_UNLOCK();
266 }
267 
268 /*
269  * Try to flush old fragments to make space for new ones
270  */
271 static void
pf_flush_fragments(void)272 pf_flush_fragments(void)
273 {
274 	struct pf_fragment	*frag;
275 	int			 goal;
276 
277 	PF_FRAG_ASSERT();
278 
279 	goal = uma_zone_get_cur(V_pf_frent_z) * 9 / 10;
280 	DPFPRINTF(PF_DEBUG_MISC, "trying to free %d frag entriess", goal);
281 	while (goal < uma_zone_get_cur(V_pf_frent_z)) {
282 		frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue);
283 		if (frag)
284 			pf_free_fragment(frag);
285 		else
286 			break;
287 	}
288 }
289 
290 /*
291  * Remove a fragment from the fragment queue, free its fragment entries,
292  * and free the fragment itself.
293  */
294 static void
pf_free_fragment(struct pf_fragment * frag)295 pf_free_fragment(struct pf_fragment *frag)
296 {
297 	struct pf_frent		*frent;
298 	struct pf_frnode	*frnode;
299 
300 	PF_FRAG_ASSERT();
301 
302 	frnode = frag->fr_node;
303 	RB_REMOVE(pf_frag_tree, &frnode->fn_tree, frag);
304 	MPASS(frnode->fn_fragments >= 1);
305 	frnode->fn_fragments--;
306 	if (frnode->fn_fragments == 0) {
307 		MPASS(RB_EMPTY(&frnode->fn_tree));
308 		RB_REMOVE(pf_frnode_tree, &V_pf_frnode_tree, frnode);
309 		uma_zfree(V_pf_frnode_z, frnode);
310 	}
311 
312 	TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next);
313 
314 	/* Free all fragment entries */
315 	while ((frent = TAILQ_FIRST(&frag->fr_queue)) != NULL) {
316 		TAILQ_REMOVE(&frag->fr_queue, frent, fr_next);
317 
318 		m_freem(frent->fe_m);
319 		uma_zfree(V_pf_frent_z, frent);
320 	}
321 
322 	uma_zfree(V_pf_frag_z, frag);
323 }
324 
325 static struct pf_fragment *
pf_find_fragment(struct pf_frnode * key,uint32_t id)326 pf_find_fragment(struct pf_frnode *key, uint32_t id)
327 {
328 	struct pf_fragment	*frag, idkey;
329 	struct pf_frnode	*frnode;
330 
331 	PF_FRAG_ASSERT();
332 
333 	frnode = RB_FIND(pf_frnode_tree, &V_pf_frnode_tree, key);
334 	if (frnode == NULL)
335 		return (NULL);
336 	MPASS(frnode->fn_fragments >= 1);
337 	idkey.fr_id = id;
338 	frag = RB_FIND(pf_frag_tree, &frnode->fn_tree, &idkey);
339 	if (frag == NULL)
340 		return (NULL);
341 	TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next);
342 	TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next);
343 
344 	return (frag);
345 }
346 
347 static struct pf_frent *
pf_create_fragment(u_short * reason)348 pf_create_fragment(u_short *reason)
349 {
350 	struct pf_frent *frent;
351 
352 	PF_FRAG_ASSERT();
353 
354 	frent = uma_zalloc(V_pf_frent_z, M_NOWAIT);
355 	if (frent == NULL) {
356 		pf_flush_fragments();
357 		frent = uma_zalloc(V_pf_frent_z, M_NOWAIT);
358 		if (frent == NULL) {
359 			REASON_SET(reason, PFRES_MEMORY);
360 			return (NULL);
361 		}
362 	}
363 
364 	return (frent);
365 }
366 
367 /*
368  * Calculate the additional holes that were created in the fragment
369  * queue by inserting this fragment.  A fragment in the middle
370  * creates one more hole by splitting.  For each connected side,
371  * it loses one hole.
372  * Fragment entry must be in the queue when calling this function.
373  */
374 static int
pf_frent_holes(struct pf_frent * frent)375 pf_frent_holes(struct pf_frent *frent)
376 {
377 	struct pf_frent *prev = TAILQ_PREV(frent, pf_fragq, fr_next);
378 	struct pf_frent *next = TAILQ_NEXT(frent, fr_next);
379 	int holes = 1;
380 
381 	if (prev == NULL) {
382 		if (frent->fe_off == 0)
383 			holes--;
384 	} else {
385 		KASSERT(frent->fe_off != 0, ("frent->fe_off != 0"));
386 		if (frent->fe_off == prev->fe_off + prev->fe_len)
387 			holes--;
388 	}
389 	if (next == NULL) {
390 		if (!frent->fe_mff)
391 			holes--;
392 	} else {
393 		KASSERT(frent->fe_mff, ("frent->fe_mff"));
394 		if (next->fe_off == frent->fe_off + frent->fe_len)
395 			holes--;
396 	}
397 	return holes;
398 }
399 
400 static inline int
pf_frent_index(struct pf_frent * frent)401 pf_frent_index(struct pf_frent *frent)
402 {
403 	/*
404 	 * We have an array of 16 entry points to the queue.  A full size
405 	 * 65535 octet IP packet can have 8192 fragments.  So the queue
406 	 * traversal length is at most 512 and at most 16 entry points are
407 	 * checked.  We need 128 additional bytes on a 64 bit architecture.
408 	 */
409 	CTASSERT(((u_int16_t)0xffff &~ 7) / (0x10000 / PF_FRAG_ENTRY_POINTS) ==
410 	    16 - 1);
411 	CTASSERT(((u_int16_t)0xffff >> 3) / PF_FRAG_ENTRY_POINTS == 512 - 1);
412 
413 	return frent->fe_off / (0x10000 / PF_FRAG_ENTRY_POINTS);
414 }
415 
416 static int
pf_frent_insert(struct pf_fragment * frag,struct pf_frent * frent,struct pf_frent * prev)417 pf_frent_insert(struct pf_fragment *frag, struct pf_frent *frent,
418     struct pf_frent *prev)
419 {
420 	int index;
421 
422 	CTASSERT(PF_FRAG_ENTRY_LIMIT <= 0xff);
423 
424 	/*
425 	 * A packet has at most 65536 octets.  With 16 entry points, each one
426 	 * spawns 4096 octets.  We limit these to 64 fragments each, which
427 	 * means on average every fragment must have at least 64 octets.
428 	 */
429 	index = pf_frent_index(frent);
430 	if (frag->fr_entries[index] >= PF_FRAG_ENTRY_LIMIT)
431 		return ENOBUFS;
432 	frag->fr_entries[index]++;
433 
434 	if (prev == NULL) {
435 		TAILQ_INSERT_HEAD(&frag->fr_queue, frent, fr_next);
436 	} else {
437 		KASSERT(prev->fe_off + prev->fe_len <= frent->fe_off,
438 		    ("overlapping fragment"));
439 		TAILQ_INSERT_AFTER(&frag->fr_queue, prev, frent, fr_next);
440 	}
441 
442 	if (frag->fr_firstoff[index] == NULL) {
443 		KASSERT(prev == NULL || pf_frent_index(prev) < index,
444 		    ("prev == NULL || pf_frent_index(pref) < index"));
445 		frag->fr_firstoff[index] = frent;
446 	} else {
447 		if (frent->fe_off < frag->fr_firstoff[index]->fe_off) {
448 			KASSERT(prev == NULL || pf_frent_index(prev) < index,
449 			    ("prev == NULL || pf_frent_index(pref) < index"));
450 			frag->fr_firstoff[index] = frent;
451 		} else {
452 			KASSERT(prev != NULL, ("prev != NULL"));
453 			KASSERT(pf_frent_index(prev) == index,
454 			    ("pf_frent_index(prev) == index"));
455 		}
456 	}
457 
458 	frag->fr_holes += pf_frent_holes(frent);
459 
460 	return 0;
461 }
462 
463 void
pf_frent_remove(struct pf_fragment * frag,struct pf_frent * frent)464 pf_frent_remove(struct pf_fragment *frag, struct pf_frent *frent)
465 {
466 #ifdef INVARIANTS
467 	struct pf_frent *prev = TAILQ_PREV(frent, pf_fragq, fr_next);
468 #endif /* INVARIANTS */
469 	struct pf_frent *next = TAILQ_NEXT(frent, fr_next);
470 	int index;
471 
472 	frag->fr_holes -= pf_frent_holes(frent);
473 
474 	index = pf_frent_index(frent);
475 	KASSERT(frag->fr_firstoff[index] != NULL, ("frent not found"));
476 	if (frag->fr_firstoff[index]->fe_off == frent->fe_off) {
477 		if (next == NULL) {
478 			frag->fr_firstoff[index] = NULL;
479 		} else {
480 			KASSERT(frent->fe_off + frent->fe_len <= next->fe_off,
481 			    ("overlapping fragment"));
482 			if (pf_frent_index(next) == index) {
483 				frag->fr_firstoff[index] = next;
484 			} else {
485 				frag->fr_firstoff[index] = NULL;
486 			}
487 		}
488 	} else {
489 		KASSERT(frag->fr_firstoff[index]->fe_off < frent->fe_off,
490 		    ("frag->fr_firstoff[index]->fe_off < frent->fe_off"));
491 		KASSERT(prev != NULL, ("prev != NULL"));
492 		KASSERT(prev->fe_off + prev->fe_len <= frent->fe_off,
493 		    ("overlapping fragment"));
494 		KASSERT(pf_frent_index(prev) == index,
495 		    ("pf_frent_index(prev) == index"));
496 	}
497 
498 	TAILQ_REMOVE(&frag->fr_queue, frent, fr_next);
499 
500 	KASSERT(frag->fr_entries[index] > 0, ("No fragments remaining"));
501 	frag->fr_entries[index]--;
502 }
503 
504 struct pf_frent *
pf_frent_previous(struct pf_fragment * frag,struct pf_frent * frent)505 pf_frent_previous(struct pf_fragment *frag, struct pf_frent *frent)
506 {
507 	struct pf_frent *prev, *next;
508 	int index;
509 
510 	/*
511 	 * If there are no fragments after frag, take the final one.  Assume
512 	 * that the global queue is not empty.
513 	 */
514 	prev = TAILQ_LAST(&frag->fr_queue, pf_fragq);
515 	KASSERT(prev != NULL, ("prev != NULL"));
516 	if (prev->fe_off <= frent->fe_off)
517 		return prev;
518 	/*
519 	 * We want to find a fragment entry that is before frag, but still
520 	 * close to it.  Find the first fragment entry that is in the same
521 	 * entry point or in the first entry point after that.  As we have
522 	 * already checked that there are entries behind frag, this will
523 	 * succeed.
524 	 */
525 	for (index = pf_frent_index(frent); index < PF_FRAG_ENTRY_POINTS;
526 	    index++) {
527 		prev = frag->fr_firstoff[index];
528 		if (prev != NULL)
529 			break;
530 	}
531 	KASSERT(prev != NULL, ("prev != NULL"));
532 	/*
533 	 * In prev we may have a fragment from the same entry point that is
534 	 * before frent, or one that is just one position behind frent.
535 	 * In the latter case, we go back one step and have the predecessor.
536 	 * There may be none if the new fragment will be the first one.
537 	 */
538 	if (prev->fe_off > frent->fe_off) {
539 		prev = TAILQ_PREV(prev, pf_fragq, fr_next);
540 		if (prev == NULL)
541 			return NULL;
542 		KASSERT(prev->fe_off <= frent->fe_off,
543 		    ("prev->fe_off <= frent->fe_off"));
544 		return prev;
545 	}
546 	/*
547 	 * In prev is the first fragment of the entry point.  The offset
548 	 * of frag is behind it.  Find the closest previous fragment.
549 	 */
550 	for (next = TAILQ_NEXT(prev, fr_next); next != NULL;
551 	    next = TAILQ_NEXT(next, fr_next)) {
552 		if (next->fe_off > frent->fe_off)
553 			break;
554 		prev = next;
555 	}
556 	return prev;
557 }
558 
559 static struct pf_fragment *
pf_fillup_fragment(struct pf_frnode * key,uint32_t id,struct pf_frent * frent,u_short * reason)560 pf_fillup_fragment(struct pf_frnode *key, uint32_t id,
561     struct pf_frent *frent, u_short *reason)
562 {
563 	struct pf_frent		*after, *next, *prev;
564 	struct pf_fragment	*frag;
565 	struct pf_frnode	*frnode;
566 	uint16_t		total;
567 
568 	PF_FRAG_ASSERT();
569 
570 	/* No empty fragments. */
571 	if (frent->fe_len == 0) {
572 		DPFPRINTF(PF_DEBUG_MISC, "bad fragment: len 0");
573 		goto bad_fragment;
574 	}
575 
576 	/* All fragments are 8 byte aligned. */
577 	if (frent->fe_mff && (frent->fe_len & 0x7)) {
578 		DPFPRINTF(PF_DEBUG_MISC, "bad fragment: mff and len %d",
579 		    frent->fe_len);
580 		goto bad_fragment;
581 	}
582 
583 	/* Respect maximum length, IP_MAXPACKET == IPV6_MAXPACKET. */
584 	if (frent->fe_off + frent->fe_len > IP_MAXPACKET) {
585 		DPFPRINTF(PF_DEBUG_MISC, "bad fragment: max packet %d",
586 		    frent->fe_off + frent->fe_len);
587 		goto bad_fragment;
588 	}
589 
590 	if (key->fn_af == AF_INET)
591 		DPFPRINTF(PF_DEBUG_MISC, "reass frag %d @ %d-%d\n",
592 		    id, frent->fe_off, frent->fe_off + frent->fe_len);
593 	else
594 		DPFPRINTF(PF_DEBUG_MISC, "reass frag %#08x @ %d-%d",
595 		    id, frent->fe_off, frent->fe_off + frent->fe_len);
596 
597 	/* Fully buffer all of the fragments in this fragment queue. */
598 	frag = pf_find_fragment(key, id);
599 
600 	/* Create a new reassembly queue for this packet. */
601 	if (frag == NULL) {
602 		frag = uma_zalloc(V_pf_frag_z, M_NOWAIT);
603 		if (frag == NULL) {
604 			pf_flush_fragments();
605 			frag = uma_zalloc(V_pf_frag_z, M_NOWAIT);
606 			if (frag == NULL) {
607 				REASON_SET(reason, PFRES_MEMORY);
608 				goto drop_fragment;
609 			}
610 		}
611 
612 		frnode = RB_FIND(pf_frnode_tree, &V_pf_frnode_tree, key);
613 		if (frnode == NULL) {
614 			frnode = uma_zalloc(V_pf_frnode_z, M_NOWAIT);
615 			if (frnode == NULL) {
616 				pf_flush_fragments();
617 				frnode = uma_zalloc(V_pf_frnode_z, M_NOWAIT);
618 				if (frnode == NULL) {
619 					REASON_SET(reason, PFRES_MEMORY);
620 					uma_zfree(V_pf_frag_z, frag);
621 					goto drop_fragment;
622 				}
623 			}
624 			*frnode = *key;
625 			RB_INIT(&frnode->fn_tree);
626 			frnode->fn_fragments = 0;
627 		}
628 		memset(frag->fr_firstoff, 0, sizeof(frag->fr_firstoff));
629 		memset(frag->fr_entries, 0, sizeof(frag->fr_entries));
630 		frag->fr_timeout = time_uptime;
631 		TAILQ_INIT(&frag->fr_queue);
632 		frag->fr_maxlen = frent->fe_len;
633 		frag->fr_holes = 1;
634 
635 		frag->fr_id = id;
636 		frag->fr_node = frnode;
637 		/* RB_INSERT cannot fail as pf_find_fragment() found nothing */
638 		RB_INSERT(pf_frag_tree, &frnode->fn_tree, frag);
639 		frnode->fn_fragments++;
640 		if (frnode->fn_fragments == 1)
641 			RB_INSERT(pf_frnode_tree, &V_pf_frnode_tree, frnode);
642 
643 		TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next);
644 
645 		/* We do not have a previous fragment, cannot fail. */
646 		pf_frent_insert(frag, frent, NULL);
647 
648 		return (frag);
649 	}
650 
651 	KASSERT(!TAILQ_EMPTY(&frag->fr_queue), ("!TAILQ_EMPTY()->fr_queue"));
652 	MPASS(frag->fr_node);
653 
654 	/* Remember maximum fragment len for refragmentation. */
655 	if (frent->fe_len > frag->fr_maxlen)
656 		frag->fr_maxlen = frent->fe_len;
657 
658 	/* Maximum data we have seen already. */
659 	total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
660 		TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
661 
662 	/* Non terminal fragments must have more fragments flag. */
663 	if (frent->fe_off + frent->fe_len < total && !frent->fe_mff)
664 		goto free_ipv6_fragment;
665 
666 	/* Check if we saw the last fragment already. */
667 	if (!TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_mff) {
668 		if (frent->fe_off + frent->fe_len > total ||
669 		    (frent->fe_off + frent->fe_len == total && frent->fe_mff))
670 			goto free_ipv6_fragment;
671 	} else {
672 		if (frent->fe_off + frent->fe_len == total && !frent->fe_mff)
673 			goto free_ipv6_fragment;
674 	}
675 
676 	/* Find neighbors for newly inserted fragment */
677 	prev = pf_frent_previous(frag, frent);
678 	if (prev == NULL) {
679 		after = TAILQ_FIRST(&frag->fr_queue);
680 		KASSERT(after != NULL, ("after != NULL"));
681 	} else {
682 		after = TAILQ_NEXT(prev, fr_next);
683 	}
684 
685 	if (prev != NULL && prev->fe_off + prev->fe_len > frent->fe_off) {
686 		uint16_t precut;
687 
688 		if (frag->fr_node->fn_af == AF_INET6)
689 			goto free_fragment;
690 
691 		precut = prev->fe_off + prev->fe_len - frent->fe_off;
692 		if (precut >= frent->fe_len) {
693 			DPFPRINTF(PF_DEBUG_MISC, "new frag overlapped");
694 			goto drop_fragment;
695 		}
696 		DPFPRINTF(PF_DEBUG_MISC, "frag head overlap %d", precut);
697 		m_adj(frent->fe_m, precut);
698 		frent->fe_off += precut;
699 		frent->fe_len -= precut;
700 	}
701 
702 	for (; after != NULL && frent->fe_off + frent->fe_len > after->fe_off;
703 	    after = next) {
704 		uint16_t aftercut;
705 
706 		aftercut = frent->fe_off + frent->fe_len - after->fe_off;
707 		if (aftercut < after->fe_len) {
708 			DPFPRINTF(PF_DEBUG_MISC, "frag tail overlap %d",
709 			    aftercut);
710 			m_adj(after->fe_m, aftercut);
711 			/* Fragment may switch queue as fe_off changes */
712 			pf_frent_remove(frag, after);
713 			after->fe_off += aftercut;
714 			after->fe_len -= aftercut;
715 			/* Insert into correct queue */
716 			if (pf_frent_insert(frag, after, prev)) {
717 				DPFPRINTF(PF_DEBUG_MISC,
718 				    "fragment requeue limit exceeded");
719 				m_freem(after->fe_m);
720 				uma_zfree(V_pf_frent_z, after);
721 				/* There is not way to recover */
722 				goto free_fragment;
723 			}
724 			break;
725 		}
726 
727 		/* This fragment is completely overlapped, lose it. */
728 		DPFPRINTF(PF_DEBUG_MISC, "old frag overlapped");
729 		next = TAILQ_NEXT(after, fr_next);
730 		pf_frent_remove(frag, after);
731 		m_freem(after->fe_m);
732 		uma_zfree(V_pf_frent_z, after);
733 	}
734 
735 	/* If part of the queue gets too long, there is not way to recover. */
736 	if (pf_frent_insert(frag, frent, prev)) {
737 		DPFPRINTF(PF_DEBUG_MISC, "fragment queue limit exceeded");
738 		goto bad_fragment;
739 	}
740 
741 	return (frag);
742 
743 free_ipv6_fragment:
744 	if (frag->fr_node->fn_af == AF_INET)
745 		goto bad_fragment;
746 free_fragment:
747 	/*
748 	 * RFC 5722, Errata 3089:  When reassembling an IPv6 datagram, if one
749 	 * or more its constituent fragments is determined to be an overlapping
750 	 * fragment, the entire datagram (and any constituent fragments) MUST
751 	 * be silently discarded.
752 	 */
753 	DPFPRINTF(PF_DEBUG_MISC, "flush overlapping fragments");
754 	pf_free_fragment(frag);
755 
756 bad_fragment:
757 	REASON_SET(reason, PFRES_FRAG);
758 drop_fragment:
759 	uma_zfree(V_pf_frent_z, frent);
760 	return (NULL);
761 }
762 
763 static struct mbuf *
pf_join_fragment(struct pf_fragment * frag)764 pf_join_fragment(struct pf_fragment *frag)
765 {
766 	struct mbuf *m, *m2;
767 	struct pf_frent	*frent;
768 
769 	frent = TAILQ_FIRST(&frag->fr_queue);
770 	TAILQ_REMOVE(&frag->fr_queue, frent, fr_next);
771 
772 	m = frent->fe_m;
773 	if ((frent->fe_hdrlen + frent->fe_len) < m->m_pkthdr.len)
774 		m_adj(m, (frent->fe_hdrlen + frent->fe_len) - m->m_pkthdr.len);
775 	uma_zfree(V_pf_frent_z, frent);
776 	while ((frent = TAILQ_FIRST(&frag->fr_queue)) != NULL) {
777 		TAILQ_REMOVE(&frag->fr_queue, frent, fr_next);
778 
779 		m2 = frent->fe_m;
780 		/* Strip off ip header. */
781 		m_adj(m2, frent->fe_hdrlen);
782 		/* Strip off any trailing bytes. */
783 		if (frent->fe_len < m2->m_pkthdr.len)
784 			m_adj(m2, frent->fe_len - m2->m_pkthdr.len);
785 
786 		uma_zfree(V_pf_frent_z, frent);
787 		m_cat(m, m2);
788 	}
789 
790 	/* Remove from fragment queue. */
791 	pf_free_fragment(frag);
792 
793 	return (m);
794 }
795 
796 #ifdef INET
797 static int
pf_reassemble(struct mbuf ** m0,u_short * reason)798 pf_reassemble(struct mbuf **m0, u_short *reason)
799 {
800 	struct mbuf		*m = *m0;
801 	struct ip		*ip = mtod(m, struct ip *);
802 	struct pf_frent		*frent;
803 	struct pf_fragment	*frag;
804 	struct m_tag		*mtag;
805 	struct pf_fragment_tag	*ftag;
806 	struct pf_frnode	 key;
807 	uint16_t		 total, hdrlen;
808 	uint32_t		 frag_id;
809 	uint16_t		 maxlen;
810 
811 	/* Get an entry for the fragment queue */
812 	if ((frent = pf_create_fragment(reason)) == NULL)
813 		return (PF_DROP);
814 
815 	frent->fe_m = m;
816 	frent->fe_hdrlen = ip->ip_hl << 2;
817 	frent->fe_extoff = 0;
818 	frent->fe_len = ntohs(ip->ip_len) - (ip->ip_hl << 2);
819 	frent->fe_off = (ntohs(ip->ip_off) & IP_OFFMASK) << 3;
820 	frent->fe_mff = ntohs(ip->ip_off) & IP_MF;
821 
822 	pf_ip2key(ip, &key);
823 
824 	if ((frag = pf_fillup_fragment(&key, ip->ip_id, frent, reason)) == NULL)
825 		return (PF_DROP);
826 
827 	/* The mbuf is part of the fragment entry, no direct free or access */
828 	m = *m0 = NULL;
829 
830 	if (frag->fr_holes) {
831 		DPFPRINTF(PF_DEBUG_MISC, "frag %d, holes %d",
832 		    frag->fr_id, frag->fr_holes);
833 		return (PF_PASS);  /* drop because *m0 is NULL, no error */
834 	}
835 
836 	/* We have all the data */
837 	frent = TAILQ_FIRST(&frag->fr_queue);
838 	KASSERT(frent != NULL, ("frent != NULL"));
839 	total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
840 		TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
841 	hdrlen = frent->fe_hdrlen;
842 
843 	maxlen = frag->fr_maxlen;
844 	frag_id = frag->fr_id;
845 	m = *m0 = pf_join_fragment(frag);
846 	frag = NULL;
847 
848 	if (m->m_flags & M_PKTHDR) {
849 		int plen = 0;
850 		for (m = *m0; m; m = m->m_next)
851 			plen += m->m_len;
852 		m = *m0;
853 		m->m_pkthdr.len = plen;
854 	}
855 
856 	if ((mtag = m_tag_get(PACKET_TAG_PF_REASSEMBLED,
857 	    sizeof(struct pf_fragment_tag), M_NOWAIT)) == NULL) {
858 		REASON_SET(reason, PFRES_SHORT);
859 		/* PF_DROP requires a valid mbuf *m0 in pf_test() */
860 		return (PF_DROP);
861 	}
862 	ftag = (struct pf_fragment_tag *)(mtag + 1);
863 	ftag->ft_hdrlen = hdrlen;
864 	ftag->ft_extoff = 0;
865 	ftag->ft_maxlen = maxlen;
866 	ftag->ft_id = frag_id;
867 	m_tag_prepend(m, mtag);
868 
869 	ip = mtod(m, struct ip *);
870 	ip->ip_sum = pf_cksum_fixup(ip->ip_sum, ip->ip_len,
871 	    htons(hdrlen + total), 0);
872 	ip->ip_len = htons(hdrlen + total);
873 	ip->ip_sum = pf_cksum_fixup(ip->ip_sum, ip->ip_off,
874 	    ip->ip_off & ~(IP_MF|IP_OFFMASK), 0);
875 	ip->ip_off &= ~(IP_MF|IP_OFFMASK);
876 
877 	if (hdrlen + total > IP_MAXPACKET) {
878 		DPFPRINTF(PF_DEBUG_MISC, "drop: too big: %d", total);
879 		ip->ip_len = 0;
880 		REASON_SET(reason, PFRES_SHORT);
881 		/* PF_DROP requires a valid mbuf *m0 in pf_test() */
882 		return (PF_DROP);
883 	}
884 
885 	DPFPRINTF(PF_DEBUG_MISC, "complete: %p(%d)", m, ntohs(ip->ip_len));
886 	return (PF_PASS);
887 }
888 #endif	/* INET */
889 
890 #ifdef INET6
891 static int
pf_reassemble6(struct mbuf ** m0,struct ip6_frag * fraghdr,uint16_t hdrlen,uint16_t extoff,u_short * reason)892 pf_reassemble6(struct mbuf **m0, struct ip6_frag *fraghdr,
893     uint16_t hdrlen, uint16_t extoff, u_short *reason)
894 {
895 	struct mbuf		*m = *m0;
896 	struct ip6_hdr		*ip6 = mtod(m, struct ip6_hdr *);
897 	struct pf_frent		*frent;
898 	struct pf_fragment	*frag;
899 	struct pf_frnode	 key;
900 	struct m_tag		*mtag;
901 	struct pf_fragment_tag	*ftag;
902 	int			 off;
903 	uint32_t		 frag_id;
904 	uint16_t		 total, maxlen;
905 	uint8_t			 proto;
906 
907 	PF_FRAG_LOCK();
908 
909 	/* Get an entry for the fragment queue. */
910 	if ((frent = pf_create_fragment(reason)) == NULL) {
911 		PF_FRAG_UNLOCK();
912 		return (PF_DROP);
913 	}
914 
915 	frent->fe_m = m;
916 	frent->fe_hdrlen = hdrlen;
917 	frent->fe_extoff = extoff;
918 	frent->fe_len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - hdrlen;
919 	frent->fe_off = ntohs(fraghdr->ip6f_offlg & IP6F_OFF_MASK);
920 	frent->fe_mff = fraghdr->ip6f_offlg & IP6F_MORE_FRAG;
921 
922 	key.fn_src.v6 = ip6->ip6_src;
923 	key.fn_dst.v6 = ip6->ip6_dst;
924 	key.fn_af = AF_INET6;
925 	/* Only the first fragment's protocol is relevant. */
926 	key.fn_proto = 0;
927 
928 	if ((frag = pf_fillup_fragment(&key, fraghdr->ip6f_ident, frent, reason)) == NULL) {
929 		PF_FRAG_UNLOCK();
930 		return (PF_DROP);
931 	}
932 
933 	/* The mbuf is part of the fragment entry, no direct free or access. */
934 	m = *m0 = NULL;
935 
936 	if (frag->fr_holes) {
937 		DPFPRINTF(PF_DEBUG_MISC, "frag %d, holes %d", frag->fr_id,
938 		    frag->fr_holes);
939 		PF_FRAG_UNLOCK();
940 		return (PF_PASS);  /* Drop because *m0 is NULL, no error. */
941 	}
942 
943 	/* We have all the data. */
944 	frent = TAILQ_FIRST(&frag->fr_queue);
945 	KASSERT(frent != NULL, ("frent != NULL"));
946 	extoff = frent->fe_extoff;
947 	maxlen = frag->fr_maxlen;
948 	frag_id = frag->fr_id;
949 	total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
950 		TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
951 	hdrlen = frent->fe_hdrlen - sizeof(struct ip6_frag);
952 
953 	m = *m0 = pf_join_fragment(frag);
954 	frag = NULL;
955 
956 	PF_FRAG_UNLOCK();
957 
958 	/* Take protocol from first fragment header. */
959 	m = m_getptr(m, hdrlen + offsetof(struct ip6_frag, ip6f_nxt), &off);
960 	KASSERT(m, ("%s: short mbuf chain", __func__));
961 	proto = *(mtod(m, uint8_t *) + off);
962 	m = *m0;
963 
964 	/* Delete frag6 header */
965 	if (ip6_deletefraghdr(m, hdrlen, M_NOWAIT) != 0)
966 		goto fail;
967 
968 	if (m->m_flags & M_PKTHDR) {
969 		int plen = 0;
970 		for (m = *m0; m; m = m->m_next)
971 			plen += m->m_len;
972 		m = *m0;
973 		m->m_pkthdr.len = plen;
974 	}
975 
976 	if ((mtag = m_tag_get(PACKET_TAG_PF_REASSEMBLED,
977 	    sizeof(struct pf_fragment_tag), M_NOWAIT)) == NULL)
978 		goto fail;
979 	ftag = (struct pf_fragment_tag *)(mtag + 1);
980 	ftag->ft_hdrlen = hdrlen;
981 	ftag->ft_extoff = extoff;
982 	ftag->ft_maxlen = maxlen;
983 	ftag->ft_id = frag_id;
984 	m_tag_prepend(m, mtag);
985 
986 	ip6 = mtod(m, struct ip6_hdr *);
987 	ip6->ip6_plen = htons(hdrlen - sizeof(struct ip6_hdr) + total);
988 	if (extoff) {
989 		/* Write protocol into next field of last extension header. */
990 		m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt),
991 		    &off);
992 		KASSERT(m, ("%s: short mbuf chain", __func__));
993 		*(mtod(m, char *) + off) = proto;
994 		m = *m0;
995 	} else
996 		ip6->ip6_nxt = proto;
997 
998 	if (hdrlen - sizeof(struct ip6_hdr) + total > IPV6_MAXPACKET) {
999 		DPFPRINTF(PF_DEBUG_MISC, "drop: too big: %d", total);
1000 		ip6->ip6_plen = 0;
1001 		REASON_SET(reason, PFRES_SHORT);
1002 		/* PF_DROP requires a valid mbuf *m0 in pf_test6(). */
1003 		return (PF_DROP);
1004 	}
1005 
1006 	DPFPRINTF(PF_DEBUG_MISC, "complete: %p(%d)", m,
1007 	    ntohs(ip6->ip6_plen));
1008 	return (PF_PASS);
1009 
1010 fail:
1011 	REASON_SET(reason, PFRES_MEMORY);
1012 	/* PF_DROP requires a valid mbuf *m0 in pf_test6(), will free later. */
1013 	return (PF_DROP);
1014 }
1015 #endif	/* INET6 */
1016 
1017 #ifdef INET6
1018 int
pf_max_frag_size(struct mbuf * m)1019 pf_max_frag_size(struct mbuf *m)
1020 {
1021 	struct m_tag *tag;
1022 	struct pf_fragment_tag *ftag;
1023 
1024 	tag = m_tag_find(m, PACKET_TAG_PF_REASSEMBLED, NULL);
1025 	if (tag == NULL)
1026 		return (m->m_pkthdr.len);
1027 
1028 	ftag = (struct pf_fragment_tag *)(tag + 1);
1029 
1030 	return (ftag->ft_maxlen);
1031 }
1032 
1033 int
pf_refragment6(struct ifnet * ifp,struct mbuf ** m0,struct m_tag * mtag,struct ifnet * rt,bool forward)1034 pf_refragment6(struct ifnet *ifp, struct mbuf **m0, struct m_tag *mtag,
1035     struct ifnet *rt, bool forward)
1036 {
1037 	struct mbuf		*m = *m0, *t;
1038 	struct ip6_hdr		*hdr;
1039 	struct pf_fragment_tag	*ftag = (struct pf_fragment_tag *)(mtag + 1);
1040 	struct pf_pdesc		 pd;
1041 	uint32_t		 frag_id;
1042 	uint16_t		 hdrlen, extoff, maxlen;
1043 	uint8_t			 proto;
1044 	int			 error, action;
1045 
1046 	hdrlen = ftag->ft_hdrlen;
1047 	extoff = ftag->ft_extoff;
1048 	maxlen = ftag->ft_maxlen;
1049 	frag_id = ftag->ft_id;
1050 	m_tag_delete(m, mtag);
1051 	mtag = NULL;
1052 	ftag = NULL;
1053 
1054 	if (extoff) {
1055 		int off;
1056 
1057 		/* Use protocol from next field of last extension header */
1058 		m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt),
1059 		    &off);
1060 		KASSERT((m != NULL), ("pf_refragment6: short mbuf chain"));
1061 		proto = *(mtod(m, uint8_t *) + off);
1062 		*(mtod(m, char *) + off) = IPPROTO_FRAGMENT;
1063 		m = *m0;
1064 	} else {
1065 		hdr = mtod(m, struct ip6_hdr *);
1066 		proto = hdr->ip6_nxt;
1067 		hdr->ip6_nxt = IPPROTO_FRAGMENT;
1068 	}
1069 
1070 	/* In case of link-local traffic we'll need a scope set. */
1071 	hdr = mtod(m, struct ip6_hdr *);
1072 
1073 	in6_setscope(&hdr->ip6_src, ifp, NULL);
1074 	in6_setscope(&hdr->ip6_dst, ifp, NULL);
1075 
1076 	/* The MTU must be a multiple of 8 bytes, or we risk doing the
1077 	 * fragmentation wrong. */
1078 	maxlen = maxlen & ~7;
1079 
1080 	/*
1081 	 * Maxlen may be less than 8 if there was only a single
1082 	 * fragment.  As it was fragmented before, add a fragment
1083 	 * header also for a single fragment.  If total or maxlen
1084 	 * is less than 8, ip6_fragment() will return EMSGSIZE and
1085 	 * we drop the packet.
1086 	 */
1087 	error = ip6_fragment(ifp, m, hdrlen, proto, maxlen, frag_id);
1088 	m = (*m0)->m_nextpkt;
1089 	(*m0)->m_nextpkt = NULL;
1090 	if (error == 0) {
1091 		/* The first mbuf contains the unfragmented packet. */
1092 		m_freem(*m0);
1093 		*m0 = NULL;
1094 		action = PF_PASS;
1095 	} else {
1096 		/* Drop expects an mbuf to free. */
1097 		DPFPRINTF(PF_DEBUG_MISC, "refragment error %d", error);
1098 		action = PF_DROP;
1099 	}
1100 	for (; m; m = t) {
1101 		t = m->m_nextpkt;
1102 		m->m_nextpkt = NULL;
1103 		m->m_flags |= M_SKIP_FIREWALL;
1104 		memset(&pd, 0, sizeof(pd));
1105 		pd.pf_mtag = pf_find_mtag(m);
1106 		if (error != 0) {
1107 			m_freem(m);
1108 			continue;
1109 		}
1110 		if (rt != NULL) {
1111 			struct sockaddr_in6	dst;
1112 			hdr = mtod(m, struct ip6_hdr *);
1113 
1114 			bzero(&dst, sizeof(dst));
1115 			dst.sin6_family = AF_INET6;
1116 			dst.sin6_len = sizeof(dst);
1117 			dst.sin6_addr = hdr->ip6_dst;
1118 
1119 			if (m->m_pkthdr.len <= if_getmtu(ifp)) {
1120 				nd6_output_ifp(rt, rt, m, &dst, NULL);
1121 			} else {
1122 				in6_ifstat_inc(ifp, ifs6_in_toobig);
1123 				icmp6_error(m, ICMP6_PACKET_TOO_BIG, 0,
1124 				    if_getmtu(ifp));
1125 			}
1126 		} else if (forward) {
1127 			MPASS(m->m_pkthdr.rcvif != NULL);
1128 			ip6_forward(m, 0);
1129 		} else {
1130 			(void)ip6_output(m, NULL, NULL, 0, NULL, NULL,
1131 			    NULL);
1132 		}
1133 	}
1134 
1135 	return (action);
1136 }
1137 #endif /* INET6 */
1138 
1139 #ifdef INET
1140 int
pf_normalize_ip(u_short * reason,struct pf_pdesc * pd)1141 pf_normalize_ip(u_short *reason, struct pf_pdesc *pd)
1142 {
1143 	struct pf_krule		*r;
1144 	struct ip		*h = mtod(pd->m, struct ip *);
1145 	int			 mff = (ntohs(h->ip_off) & IP_MF);
1146 	int			 hlen = h->ip_hl << 2;
1147 	u_int16_t		 fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
1148 	u_int16_t		 max;
1149 	int			 ip_len;
1150 	int			 tag = -1;
1151 	int			 verdict;
1152 	bool			 scrub_compat;
1153 
1154 	PF_RULES_RASSERT();
1155 
1156 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1157 	/*
1158 	 * Check if there are any scrub rules, matching or not.
1159 	 * Lack of scrub rules means:
1160 	 *  - enforced packet normalization operation just like in OpenBSD
1161 	 *  - fragment reassembly depends on V_pf_status.reass
1162 	 * With scrub rules:
1163 	 *  - packet normalization is performed if there is a matching scrub rule
1164 	 *  - fragment reassembly is performed if the matching rule has no
1165 	 *    PFRULE_FRAGMENT_NOREASS flag
1166 	 */
1167 	scrub_compat = (r != NULL);
1168 	while (r != NULL) {
1169 		pf_counter_u64_add(&r->evaluations, 1);
1170 		if (pfi_kkif_match(r->kif, pd->kif) == r->ifnot)
1171 			r = r->skip[PF_SKIP_IFP];
1172 		else if (r->direction && r->direction != pd->dir)
1173 			r = r->skip[PF_SKIP_DIR];
1174 		else if (r->af && r->af != AF_INET)
1175 			r = r->skip[PF_SKIP_AF];
1176 		else if (r->proto && r->proto != h->ip_p)
1177 			r = r->skip[PF_SKIP_PROTO];
1178 		else if (PF_MISMATCHAW(&r->src.addr,
1179 		    (struct pf_addr *)&h->ip_src.s_addr, AF_INET,
1180 		    r->src.neg, pd->kif, M_GETFIB(pd->m)))
1181 			r = r->skip[PF_SKIP_SRC_ADDR];
1182 		else if (PF_MISMATCHAW(&r->dst.addr,
1183 		    (struct pf_addr *)&h->ip_dst.s_addr, AF_INET,
1184 		    r->dst.neg, NULL, M_GETFIB(pd->m)))
1185 			r = r->skip[PF_SKIP_DST_ADDR];
1186 		else if (r->match_tag && !pf_match_tag(pd->m, r, &tag,
1187 		    pd->pf_mtag ? pd->pf_mtag->tag : 0))
1188 			r = TAILQ_NEXT(r, entries);
1189 		else
1190 			break;
1191 	}
1192 
1193 	if (scrub_compat) {
1194 		/* With scrub rules present IPv4 normalization happens only
1195 		 * if one of rules has matched and it's not a "no scrub" rule */
1196 		if (r == NULL || r->action == PF_NOSCRUB)
1197 			return (PF_PASS);
1198 
1199 		pf_counter_u64_critical_enter();
1200 		pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1);
1201 		pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len);
1202 		pf_counter_u64_critical_exit();
1203 		pf_rule_to_actions(r, &pd->act);
1204 	}
1205 
1206 	/* Check for illegal packets */
1207 	if (hlen < (int)sizeof(struct ip)) {
1208 		REASON_SET(reason, PFRES_NORM);
1209 		goto drop;
1210 	}
1211 
1212 	if (hlen > ntohs(h->ip_len)) {
1213 		REASON_SET(reason, PFRES_NORM);
1214 		goto drop;
1215 	}
1216 
1217 	/* Clear IP_DF if the rule uses the no-df option or we're in no-df mode */
1218 	if (((!scrub_compat && V_pf_status.reass & PF_REASS_NODF) ||
1219 	    (r != NULL && r->rule_flag & PFRULE_NODF)) &&
1220 	    (h->ip_off & htons(IP_DF))
1221 	) {
1222 		u_int16_t ip_off = h->ip_off;
1223 
1224 		h->ip_off &= htons(~IP_DF);
1225 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
1226 	}
1227 
1228 	/* We will need other tests here */
1229 	if (!fragoff && !mff)
1230 		goto no_fragment;
1231 
1232 	/* We're dealing with a fragment now. Don't allow fragments
1233 	 * with IP_DF to enter the cache. If the flag was cleared by
1234 	 * no-df above, fine. Otherwise drop it.
1235 	 */
1236 	if (h->ip_off & htons(IP_DF)) {
1237 		DPFPRINTF(PF_DEBUG_MISC, "IP_DF");
1238 		goto bad;
1239 	}
1240 
1241 	ip_len = ntohs(h->ip_len) - hlen;
1242 
1243 	/* All fragments are 8 byte aligned */
1244 	if (mff && (ip_len & 0x7)) {
1245 		DPFPRINTF(PF_DEBUG_MISC, "mff and %d", ip_len);
1246 		goto bad;
1247 	}
1248 
1249 	/* Respect maximum length */
1250 	if (fragoff + ip_len > IP_MAXPACKET) {
1251 		DPFPRINTF(PF_DEBUG_MISC, "max packet %d", fragoff + ip_len);
1252 		goto bad;
1253 	}
1254 
1255 	if ((!scrub_compat && V_pf_status.reass) ||
1256 	    (r != NULL && !(r->rule_flag & PFRULE_FRAGMENT_NOREASS))
1257 	) {
1258 		max = fragoff + ip_len;
1259 
1260 		/* Fully buffer all of the fragments
1261 		 * Might return a completely reassembled mbuf, or NULL */
1262 		PF_FRAG_LOCK();
1263 		DPFPRINTF(PF_DEBUG_MISC, "reass frag %d @ %d-%d",
1264 		    h->ip_id, fragoff, max);
1265 		verdict = pf_reassemble(&pd->m, reason);
1266 		PF_FRAG_UNLOCK();
1267 
1268 		if (verdict != PF_PASS)
1269 			return (PF_DROP);
1270 
1271 		if (pd->m == NULL)
1272 			return (PF_DROP);
1273 
1274 		h = mtod(pd->m, struct ip *);
1275 		pd->tot_len = htons(h->ip_len);
1276 
1277  no_fragment:
1278 		/* At this point, only IP_DF is allowed in ip_off */
1279 		if (h->ip_off & ~htons(IP_DF)) {
1280 			u_int16_t ip_off = h->ip_off;
1281 
1282 			h->ip_off &= htons(IP_DF);
1283 			h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
1284 		}
1285 	}
1286 
1287 	return (PF_PASS);
1288 
1289  bad:
1290 	DPFPRINTF(PF_DEBUG_MISC, "dropping bad fragment");
1291 	REASON_SET(reason, PFRES_FRAG);
1292  drop:
1293 	if (r != NULL && r->log)
1294 		PFLOG_PACKET(PF_DROP, *reason, r, NULL, NULL, pd, 1, NULL);
1295 
1296 	return (PF_DROP);
1297 }
1298 #endif
1299 
1300 #ifdef INET6
1301 int
pf_normalize_ip6(int off,u_short * reason,struct pf_pdesc * pd)1302 pf_normalize_ip6(int off, u_short *reason,
1303     struct pf_pdesc *pd)
1304 {
1305 	struct pf_krule		*r;
1306 	struct ip6_hdr		*h;
1307 	struct ip6_frag		 frag;
1308 	bool			 scrub_compat;
1309 
1310 	PF_RULES_RASSERT();
1311 
1312 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1313 	/*
1314 	 * Check if there are any scrub rules, matching or not.
1315 	 * Lack of scrub rules means:
1316 	 *  - enforced packet normalization operation just like in OpenBSD
1317 	 * With scrub rules:
1318 	 *  - packet normalization is performed if there is a matching scrub rule
1319 	 * XXX: Fragment reassembly always performed for IPv6!
1320 	 */
1321 	scrub_compat = (r != NULL);
1322 	while (r != NULL) {
1323 		pf_counter_u64_add(&r->evaluations, 1);
1324 		if (pfi_kkif_match(r->kif, pd->kif) == r->ifnot)
1325 			r = r->skip[PF_SKIP_IFP];
1326 		else if (r->direction && r->direction != pd->dir)
1327 			r = r->skip[PF_SKIP_DIR];
1328 		else if (r->af && r->af != AF_INET6)
1329 			r = r->skip[PF_SKIP_AF];
1330 		else if (r->proto && r->proto != pd->proto)
1331 			r = r->skip[PF_SKIP_PROTO];
1332 		else if (PF_MISMATCHAW(&r->src.addr,
1333 		    (struct pf_addr *)&pd->src, AF_INET6,
1334 		    r->src.neg, pd->kif, M_GETFIB(pd->m)))
1335 			r = r->skip[PF_SKIP_SRC_ADDR];
1336 		else if (PF_MISMATCHAW(&r->dst.addr,
1337 		    (struct pf_addr *)&pd->dst, AF_INET6,
1338 		    r->dst.neg, NULL, M_GETFIB(pd->m)))
1339 			r = r->skip[PF_SKIP_DST_ADDR];
1340 		else
1341 			break;
1342 	}
1343 
1344 	if (scrub_compat) {
1345 		/* With scrub rules present IPv6 normalization happens only
1346 		 * if one of rules has matched and it's not a "no scrub" rule */
1347 		if (r == NULL || r->action == PF_NOSCRUB)
1348 			return (PF_PASS);
1349 
1350 		pf_counter_u64_critical_enter();
1351 		pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1);
1352 		pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len);
1353 		pf_counter_u64_critical_exit();
1354 		pf_rule_to_actions(r, &pd->act);
1355 	}
1356 
1357 	if (!pf_pull_hdr(pd->m, off, &frag, sizeof(frag), NULL, reason, AF_INET6))
1358 		return (PF_DROP);
1359 
1360 	/* Offset now points to data portion. */
1361 	off += sizeof(frag);
1362 
1363 	if (pd->virtual_proto == PF_VPROTO_FRAGMENT) {
1364 		/* Returns PF_DROP or *m0 is NULL or completely reassembled
1365 		 * mbuf. */
1366 		if (pf_reassemble6(&pd->m, &frag, off, pd->extoff, reason) != PF_PASS)
1367 			return (PF_DROP);
1368 		if (pd->m == NULL)
1369 			return (PF_DROP);
1370 		h = mtod(pd->m, struct ip6_hdr *);
1371 		pd->tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr);
1372 	}
1373 
1374 	return (PF_PASS);
1375 }
1376 #endif /* INET6 */
1377 
1378 int
pf_normalize_tcp(struct pf_pdesc * pd)1379 pf_normalize_tcp(struct pf_pdesc *pd)
1380 {
1381 	struct pf_krule	*r, *rm = NULL;
1382 	struct tcphdr	*th = &pd->hdr.tcp;
1383 	int		 rewrite = 0;
1384 	u_short		 reason;
1385 	u_int16_t	 flags;
1386 	sa_family_t	 af = pd->af;
1387 	int		 srs;
1388 
1389 	PF_RULES_RASSERT();
1390 
1391 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1392 	/* Check if there any scrub rules. Lack of scrub rules means enforced
1393 	 * packet normalization operation just like in OpenBSD. */
1394 	srs = (r != NULL);
1395 	while (r != NULL) {
1396 		pf_counter_u64_add(&r->evaluations, 1);
1397 		if (pfi_kkif_match(r->kif, pd->kif) == r->ifnot)
1398 			r = r->skip[PF_SKIP_IFP];
1399 		else if (r->direction && r->direction != pd->dir)
1400 			r = r->skip[PF_SKIP_DIR];
1401 		else if (r->af && r->af != af)
1402 			r = r->skip[PF_SKIP_AF];
1403 		else if (r->proto && r->proto != pd->proto)
1404 			r = r->skip[PF_SKIP_PROTO];
1405 		else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
1406 		    r->src.neg, pd->kif, M_GETFIB(pd->m)))
1407 			r = r->skip[PF_SKIP_SRC_ADDR];
1408 		else if (r->src.port_op && !pf_match_port(r->src.port_op,
1409 			    r->src.port[0], r->src.port[1], th->th_sport))
1410 			r = r->skip[PF_SKIP_SRC_PORT];
1411 		else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
1412 		    r->dst.neg, NULL, M_GETFIB(pd->m)))
1413 			r = r->skip[PF_SKIP_DST_ADDR];
1414 		else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
1415 			    r->dst.port[0], r->dst.port[1], th->th_dport))
1416 			r = r->skip[PF_SKIP_DST_PORT];
1417 		else if (r->os_fingerprint != PF_OSFP_ANY && !pf_osfp_match(
1418 			    pf_osfp_fingerprint(pd, th),
1419 			    r->os_fingerprint))
1420 			r = TAILQ_NEXT(r, entries);
1421 		else {
1422 			rm = r;
1423 			break;
1424 		}
1425 	}
1426 
1427 	if (srs) {
1428 		/* With scrub rules present TCP normalization happens only
1429 		 * if one of rules has matched and it's not a "no scrub" rule */
1430 		if (rm == NULL || rm->action == PF_NOSCRUB)
1431 			return (PF_PASS);
1432 
1433 		pf_counter_u64_critical_enter();
1434 		pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1);
1435 		pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len);
1436 		pf_counter_u64_critical_exit();
1437 		pf_rule_to_actions(rm, &pd->act);
1438 	}
1439 
1440 	if (rm && rm->rule_flag & PFRULE_REASSEMBLE_TCP)
1441 		pd->flags |= PFDESC_TCP_NORM;
1442 
1443 	flags = tcp_get_flags(th);
1444 	if (flags & TH_SYN) {
1445 		/* Illegal packet */
1446 		if (flags & TH_RST)
1447 			goto tcp_drop;
1448 
1449 		if (flags & TH_FIN)
1450 			goto tcp_drop;
1451 	} else {
1452 		/* Illegal packet */
1453 		if (!(flags & (TH_ACK|TH_RST)))
1454 			goto tcp_drop;
1455 	}
1456 
1457 	if (!(flags & TH_ACK)) {
1458 		/* These flags are only valid if ACK is set */
1459 		if ((flags & TH_FIN) || (flags & TH_PUSH) || (flags & TH_URG))
1460 			goto tcp_drop;
1461 	}
1462 
1463 	/* Check for illegal header length */
1464 	if (th->th_off < (sizeof(struct tcphdr) >> 2))
1465 		goto tcp_drop;
1466 
1467 	/* If flags changed, or reserved data set, then adjust */
1468 	if (flags != tcp_get_flags(th) ||
1469 	    (tcp_get_flags(th) & (TH_RES1|TH_RES2|TH_RES2)) != 0) {
1470 		u_int16_t	ov, nv;
1471 
1472 		ov = *(u_int16_t *)(&th->th_ack + 1);
1473 		flags &= ~(TH_RES1 | TH_RES2 | TH_RES3);
1474 		tcp_set_flags(th, flags);
1475 		nv = *(u_int16_t *)(&th->th_ack + 1);
1476 
1477 		th->th_sum = pf_proto_cksum_fixup(pd->m, th->th_sum, ov, nv, 0);
1478 		rewrite = 1;
1479 	}
1480 
1481 	/* Remove urgent pointer, if TH_URG is not set */
1482 	if (!(flags & TH_URG) && th->th_urp) {
1483 		th->th_sum = pf_proto_cksum_fixup(pd->m, th->th_sum, th->th_urp,
1484 		    0, 0);
1485 		th->th_urp = 0;
1486 		rewrite = 1;
1487 	}
1488 
1489 	/* copy back packet headers if we sanitized */
1490 	if (rewrite)
1491 		m_copyback(pd->m, pd->off, sizeof(*th), (caddr_t)th);
1492 
1493 	return (PF_PASS);
1494 
1495  tcp_drop:
1496 	REASON_SET(&reason, PFRES_NORM);
1497 	if (rm != NULL && r->log)
1498 		PFLOG_PACKET(PF_DROP, reason, r, NULL, NULL, pd, 1, NULL);
1499 	return (PF_DROP);
1500 }
1501 
1502 int
pf_normalize_tcp_init(struct pf_pdesc * pd,struct tcphdr * th,struct pf_state_peer * src)1503 pf_normalize_tcp_init(struct pf_pdesc *pd, struct tcphdr *th,
1504     struct pf_state_peer *src)
1505 {
1506 	u_int32_t tsval, tsecr;
1507 	int		 olen;
1508 	uint8_t		 opts[MAX_TCPOPTLEN], *opt;
1509 
1510 	KASSERT((src->scrub == NULL),
1511 	    ("pf_normalize_tcp_init: src->scrub != NULL"));
1512 
1513 	src->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT);
1514 	if (src->scrub == NULL)
1515 		return (1);
1516 
1517 	switch (pd->af) {
1518 #ifdef INET
1519 	case AF_INET: {
1520 		struct ip *h = mtod(pd->m, struct ip *);
1521 		src->scrub->pfss_ttl = h->ip_ttl;
1522 		break;
1523 	}
1524 #endif /* INET */
1525 #ifdef INET6
1526 	case AF_INET6: {
1527 		struct ip6_hdr *h = mtod(pd->m, struct ip6_hdr *);
1528 		src->scrub->pfss_ttl = h->ip6_hlim;
1529 		break;
1530 	}
1531 #endif /* INET6 */
1532 	default:
1533 		unhandled_af(pd->af);
1534 	}
1535 
1536 	/*
1537 	 * All normalizations below are only begun if we see the start of
1538 	 * the connections.  They must all set an enabled bit in pfss_flags
1539 	 */
1540 	if ((tcp_get_flags(th) & TH_SYN) == 0)
1541 		return (0);
1542 
1543 	olen = (th->th_off << 2) - sizeof(*th);
1544 	if (olen < TCPOLEN_TIMESTAMP || !pf_pull_hdr(pd->m,
1545 	    pd->off + sizeof(*th), opts, olen, NULL, NULL, pd->af))
1546 		return (0);
1547 
1548 	opt = opts;
1549 	while ((opt = pf_find_tcpopt(opt, opts, olen,
1550 	    TCPOPT_TIMESTAMP, TCPOLEN_TIMESTAMP)) != NULL) {
1551 		src->scrub->pfss_flags |= PFSS_TIMESTAMP;
1552 		src->scrub->pfss_ts_mod = arc4random();
1553 		/* note PFSS_PAWS not set yet */
1554 		memcpy(&tsval, &opt[2], sizeof(u_int32_t));
1555 		memcpy(&tsecr, &opt[6], sizeof(u_int32_t));
1556 		src->scrub->pfss_tsval0 = ntohl(tsval);
1557 		src->scrub->pfss_tsval = ntohl(tsval);
1558 		src->scrub->pfss_tsecr = ntohl(tsecr);
1559 		getmicrouptime(&src->scrub->pfss_last);
1560 
1561 		opt += opt[1];
1562 	}
1563 
1564 	return (0);
1565 }
1566 
1567 void
pf_normalize_tcp_cleanup(struct pf_kstate * state)1568 pf_normalize_tcp_cleanup(struct pf_kstate *state)
1569 {
1570 	/* XXX Note: this also cleans up SCTP. */
1571 	uma_zfree(V_pf_state_scrub_z, state->src.scrub);
1572 	uma_zfree(V_pf_state_scrub_z, state->dst.scrub);
1573 
1574 	/* Someday... flush the TCP segment reassembly descriptors. */
1575 }
1576 int
pf_normalize_sctp_init(struct pf_pdesc * pd,struct pf_state_peer * src,struct pf_state_peer * dst)1577 pf_normalize_sctp_init(struct pf_pdesc *pd, struct pf_state_peer *src,
1578     struct pf_state_peer *dst)
1579 {
1580 	src->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT);
1581 	if (src->scrub == NULL)
1582 		return (1);
1583 
1584 	dst->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT);
1585 	if (dst->scrub == NULL) {
1586 		uma_zfree(V_pf_state_scrub_z, src);
1587 		return (1);
1588 	}
1589 
1590 	dst->scrub->pfss_v_tag = pd->sctp_initiate_tag;
1591 
1592 	return (0);
1593 }
1594 
1595 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)1596 pf_normalize_tcp_stateful(struct pf_pdesc *pd,
1597     u_short *reason, struct tcphdr *th, struct pf_kstate *state,
1598     struct pf_state_peer *src, struct pf_state_peer *dst, int *writeback)
1599 {
1600 	struct timeval uptime;
1601 	u_int tsval_from_last;
1602 	uint32_t tsval, tsecr;
1603 	int copyback = 0;
1604 	int got_ts = 0;
1605 	int olen;
1606 	uint8_t opts[MAX_TCPOPTLEN], *opt;
1607 
1608 	KASSERT((src->scrub || dst->scrub),
1609 	    ("%s: src->scrub && dst->scrub!", __func__));
1610 
1611 	/*
1612 	 * Enforce the minimum TTL seen for this connection.  Negate a common
1613 	 * technique to evade an intrusion detection system and confuse
1614 	 * firewall state code.
1615 	 */
1616 	switch (pd->af) {
1617 #ifdef INET
1618 	case AF_INET: {
1619 		if (src->scrub) {
1620 			struct ip *h = mtod(pd->m, struct ip *);
1621 			if (h->ip_ttl > src->scrub->pfss_ttl)
1622 				src->scrub->pfss_ttl = h->ip_ttl;
1623 			h->ip_ttl = src->scrub->pfss_ttl;
1624 		}
1625 		break;
1626 	}
1627 #endif /* INET */
1628 #ifdef INET6
1629 	case AF_INET6: {
1630 		if (src->scrub) {
1631 			struct ip6_hdr *h = mtod(pd->m, struct ip6_hdr *);
1632 			if (h->ip6_hlim > src->scrub->pfss_ttl)
1633 				src->scrub->pfss_ttl = h->ip6_hlim;
1634 			h->ip6_hlim = src->scrub->pfss_ttl;
1635 		}
1636 		break;
1637 	}
1638 #endif /* INET6 */
1639 	default:
1640 		unhandled_af(pd->af);
1641 	}
1642 
1643 	olen = (th->th_off << 2) - sizeof(*th);
1644 
1645 	if (olen >= TCPOLEN_TIMESTAMP &&
1646 	    ((src->scrub && (src->scrub->pfss_flags & PFSS_TIMESTAMP)) ||
1647 	    (dst->scrub && (dst->scrub->pfss_flags & PFSS_TIMESTAMP))) &&
1648 	    pf_pull_hdr(pd->m, pd->off + sizeof(*th), opts, olen, NULL, NULL, pd->af)) {
1649 		/* Modulate the timestamps.  Can be used for NAT detection, OS
1650 		 * uptime determination or reboot detection.
1651 		 */
1652 		opt = opts;
1653 		while ((opt = pf_find_tcpopt(opt, opts, olen,
1654 		    TCPOPT_TIMESTAMP, TCPOLEN_TIMESTAMP)) != NULL) {
1655 			uint8_t *ts = opt + 2;
1656 			uint8_t *tsr = opt + 6;
1657 
1658 			if (got_ts) {
1659 				/* Huh?  Multiple timestamps!? */
1660 				if (V_pf_status.debug >= PF_DEBUG_MISC) {
1661 					printf("pf: %s: multiple TS??", __func__);
1662 					pf_print_state(state);
1663 					printf("\n");
1664 				}
1665 				REASON_SET(reason, PFRES_TS);
1666 				return (PF_DROP);
1667 			}
1668 
1669 			memcpy(&tsval, ts, sizeof(u_int32_t));
1670 			memcpy(&tsecr, tsr, sizeof(u_int32_t));
1671 
1672 			/* modulate TS */
1673 			if (tsval && src->scrub &&
1674 			    (src->scrub->pfss_flags & PFSS_TIMESTAMP)) {
1675 				/* tsval used further on */
1676 				tsval = ntohl(tsval);
1677 				pf_patch_32(pd,
1678 				    ts, htonl(tsval + src->scrub->pfss_ts_mod),
1679 				    PF_ALGNMNT(ts - opts));
1680 				copyback = 1;
1681 			}
1682 
1683 			/* modulate TS reply if any (!0) */
1684 			if (tsecr && dst->scrub &&
1685 			    (dst->scrub->pfss_flags & PFSS_TIMESTAMP)) {
1686 				/* tsecr used further on */
1687 				tsecr = ntohl(tsecr) - dst->scrub->pfss_ts_mod;
1688 				pf_patch_32(pd, tsr, htonl(tsecr),
1689 				    PF_ALGNMNT(tsr - opts));
1690 				copyback = 1;
1691 			}
1692 
1693 			got_ts = 1;
1694 			opt += opt[1];
1695 		}
1696 
1697 		if (copyback) {
1698 			/* Copyback the options, caller copys back header */
1699 			*writeback = 1;
1700 			m_copyback(pd->m, pd->off + sizeof(*th), olen, opts);
1701 		}
1702 	}
1703 
1704 	/*
1705 	 * Must invalidate PAWS checks on connections idle for too long.
1706 	 * The fastest allowed timestamp clock is 1ms.  That turns out to
1707 	 * be about 24 days before it wraps.  XXX Right now our lowerbound
1708 	 * TS echo check only works for the first 12 days of a connection
1709 	 * when the TS has exhausted half its 32bit space
1710 	 */
1711 #define TS_MAX_IDLE	(24*24*60*60)
1712 #define TS_MAX_CONN	(12*24*60*60)	/* XXX remove when better tsecr check */
1713 
1714 	getmicrouptime(&uptime);
1715 	if (src->scrub && (src->scrub->pfss_flags & PFSS_PAWS) &&
1716 	    (uptime.tv_sec - src->scrub->pfss_last.tv_sec > TS_MAX_IDLE ||
1717 	    time_uptime - (state->creation / 1000) > TS_MAX_CONN))  {
1718 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
1719 			DPFPRINTF(PF_DEBUG_MISC, "src idled out of PAWS");
1720 			pf_print_state(state);
1721 			printf("\n");
1722 		}
1723 		src->scrub->pfss_flags = (src->scrub->pfss_flags & ~PFSS_PAWS)
1724 		    | PFSS_PAWS_IDLED;
1725 	}
1726 	if (dst->scrub && (dst->scrub->pfss_flags & PFSS_PAWS) &&
1727 	    uptime.tv_sec - dst->scrub->pfss_last.tv_sec > TS_MAX_IDLE) {
1728 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
1729 			DPFPRINTF(PF_DEBUG_MISC, "dst idled out of PAWS");
1730 			pf_print_state(state);
1731 			printf("\n");
1732 		}
1733 		dst->scrub->pfss_flags = (dst->scrub->pfss_flags & ~PFSS_PAWS)
1734 		    | PFSS_PAWS_IDLED;
1735 	}
1736 
1737 	if (got_ts && src->scrub && dst->scrub &&
1738 	    (src->scrub->pfss_flags & PFSS_PAWS) &&
1739 	    (dst->scrub->pfss_flags & PFSS_PAWS)) {
1740 		/* Validate that the timestamps are "in-window".
1741 		 * RFC1323 describes TCP Timestamp options that allow
1742 		 * measurement of RTT (round trip time) and PAWS
1743 		 * (protection against wrapped sequence numbers).  PAWS
1744 		 * gives us a set of rules for rejecting packets on
1745 		 * long fat pipes (packets that were somehow delayed
1746 		 * in transit longer than the time it took to send the
1747 		 * full TCP sequence space of 4Gb).  We can use these
1748 		 * rules and infer a few others that will let us treat
1749 		 * the 32bit timestamp and the 32bit echoed timestamp
1750 		 * as sequence numbers to prevent a blind attacker from
1751 		 * inserting packets into a connection.
1752 		 *
1753 		 * RFC1323 tells us:
1754 		 *  - The timestamp on this packet must be greater than
1755 		 *    or equal to the last value echoed by the other
1756 		 *    endpoint.  The RFC says those will be discarded
1757 		 *    since it is a dup that has already been acked.
1758 		 *    This gives us a lowerbound on the timestamp.
1759 		 *        timestamp >= other last echoed timestamp
1760 		 *  - The timestamp will be less than or equal to
1761 		 *    the last timestamp plus the time between the
1762 		 *    last packet and now.  The RFC defines the max
1763 		 *    clock rate as 1ms.  We will allow clocks to be
1764 		 *    up to 10% fast and will allow a total difference
1765 		 *    or 30 seconds due to a route change.  And this
1766 		 *    gives us an upperbound on the timestamp.
1767 		 *        timestamp <= last timestamp + max ticks
1768 		 *    We have to be careful here.  Windows will send an
1769 		 *    initial timestamp of zero and then initialize it
1770 		 *    to a random value after the 3whs; presumably to
1771 		 *    avoid a DoS by having to call an expensive RNG
1772 		 *    during a SYN flood.  Proof MS has at least one
1773 		 *    good security geek.
1774 		 *
1775 		 *  - The TCP timestamp option must also echo the other
1776 		 *    endpoints timestamp.  The timestamp echoed is the
1777 		 *    one carried on the earliest unacknowledged segment
1778 		 *    on the left edge of the sequence window.  The RFC
1779 		 *    states that the host will reject any echoed
1780 		 *    timestamps that were larger than any ever sent.
1781 		 *    This gives us an upperbound on the TS echo.
1782 		 *        tescr <= largest_tsval
1783 		 *  - The lowerbound on the TS echo is a little more
1784 		 *    tricky to determine.  The other endpoint's echoed
1785 		 *    values will not decrease.  But there may be
1786 		 *    network conditions that re-order packets and
1787 		 *    cause our view of them to decrease.  For now the
1788 		 *    only lowerbound we can safely determine is that
1789 		 *    the TS echo will never be less than the original
1790 		 *    TS.  XXX There is probably a better lowerbound.
1791 		 *    Remove TS_MAX_CONN with better lowerbound check.
1792 		 *        tescr >= other original TS
1793 		 *
1794 		 * It is also important to note that the fastest
1795 		 * timestamp clock of 1ms will wrap its 32bit space in
1796 		 * 24 days.  So we just disable TS checking after 24
1797 		 * days of idle time.  We actually must use a 12d
1798 		 * connection limit until we can come up with a better
1799 		 * lowerbound to the TS echo check.
1800 		 */
1801 		struct timeval delta_ts;
1802 		int ts_fudge;
1803 
1804 		/*
1805 		 * PFTM_TS_DIFF is how many seconds of leeway to allow
1806 		 * a host's timestamp.  This can happen if the previous
1807 		 * packet got delayed in transit for much longer than
1808 		 * this packet.
1809 		 */
1810 		if ((ts_fudge = state->rule->timeout[PFTM_TS_DIFF]) == 0)
1811 			ts_fudge = V_pf_default_rule.timeout[PFTM_TS_DIFF];
1812 
1813 		/* Calculate max ticks since the last timestamp */
1814 #define TS_MAXFREQ	1100		/* RFC max TS freq of 1Khz + 10% skew */
1815 #define TS_MICROSECS	1000000		/* microseconds per second */
1816 		delta_ts = uptime;
1817 		timevalsub(&delta_ts, &src->scrub->pfss_last);
1818 		tsval_from_last = (delta_ts.tv_sec + ts_fudge) * TS_MAXFREQ;
1819 		tsval_from_last += delta_ts.tv_usec / (TS_MICROSECS/TS_MAXFREQ);
1820 
1821 		if ((src->state >= TCPS_ESTABLISHED &&
1822 		    dst->state >= TCPS_ESTABLISHED) &&
1823 		    (SEQ_LT(tsval, dst->scrub->pfss_tsecr) ||
1824 		    SEQ_GT(tsval, src->scrub->pfss_tsval + tsval_from_last) ||
1825 		    (tsecr && (SEQ_GT(tsecr, dst->scrub->pfss_tsval) ||
1826 		    SEQ_LT(tsecr, dst->scrub->pfss_tsval0))))) {
1827 			/* Bad RFC1323 implementation or an insertion attack.
1828 			 *
1829 			 * - Solaris 2.6 and 2.7 are known to send another ACK
1830 			 *   after the FIN,FIN|ACK,ACK closing that carries
1831 			 *   an old timestamp.
1832 			 */
1833 
1834 			DPFPRINTF(PF_DEBUG_MISC, "Timestamp failed %c%c%c%c",
1835 			    SEQ_LT(tsval, dst->scrub->pfss_tsecr) ? '0' : ' ',
1836 			    SEQ_GT(tsval, src->scrub->pfss_tsval +
1837 			    tsval_from_last) ? '1' : ' ',
1838 			    SEQ_GT(tsecr, dst->scrub->pfss_tsval) ? '2' : ' ',
1839 			    SEQ_LT(tsecr, dst->scrub->pfss_tsval0)? '3' : ' ');
1840 			DPFPRINTF(PF_DEBUG_MISC, " tsval: %u  tsecr: %u  +ticks: "
1841 			    "%u  idle: %jus %lums",
1842 			    tsval, tsecr, tsval_from_last,
1843 			    (uintmax_t)delta_ts.tv_sec,
1844 			    delta_ts.tv_usec / 1000);
1845 			DPFPRINTF(PF_DEBUG_MISC, " src->tsval: %u  tsecr: %u",
1846 			    src->scrub->pfss_tsval, src->scrub->pfss_tsecr);
1847 			DPFPRINTF(PF_DEBUG_MISC, " dst->tsval: %u  tsecr: %u  "
1848 			    "tsval0: %u", dst->scrub->pfss_tsval,
1849 			    dst->scrub->pfss_tsecr, dst->scrub->pfss_tsval0);
1850 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
1851 				pf_print_state(state);
1852 				pf_print_flags(tcp_get_flags(th));
1853 				printf("\n");
1854 			}
1855 			REASON_SET(reason, PFRES_TS);
1856 			return (PF_DROP);
1857 		}
1858 
1859 		/* XXX I'd really like to require tsecr but it's optional */
1860 
1861 	} else if (!got_ts && (tcp_get_flags(th) & TH_RST) == 0 &&
1862 	    ((src->state == TCPS_ESTABLISHED && dst->state == TCPS_ESTABLISHED)
1863 	    || pd->p_len > 0 || (tcp_get_flags(th) & TH_SYN)) &&
1864 	    src->scrub && dst->scrub &&
1865 	    (src->scrub->pfss_flags & PFSS_PAWS) &&
1866 	    (dst->scrub->pfss_flags & PFSS_PAWS)) {
1867 		/* Didn't send a timestamp.  Timestamps aren't really useful
1868 		 * when:
1869 		 *  - connection opening or closing (often not even sent).
1870 		 *    but we must not let an attacker to put a FIN on a
1871 		 *    data packet to sneak it through our ESTABLISHED check.
1872 		 *  - on a TCP reset.  RFC suggests not even looking at TS.
1873 		 *  - on an empty ACK.  The TS will not be echoed so it will
1874 		 *    probably not help keep the RTT calculation in sync and
1875 		 *    there isn't as much danger when the sequence numbers
1876 		 *    got wrapped.  So some stacks don't include TS on empty
1877 		 *    ACKs :-(
1878 		 *
1879 		 * To minimize the disruption to mostly RFC1323 conformant
1880 		 * stacks, we will only require timestamps on data packets.
1881 		 *
1882 		 * And what do ya know, we cannot require timestamps on data
1883 		 * packets.  There appear to be devices that do legitimate
1884 		 * TCP connection hijacking.  There are HTTP devices that allow
1885 		 * a 3whs (with timestamps) and then buffer the HTTP request.
1886 		 * If the intermediate device has the HTTP response cache, it
1887 		 * will spoof the response but not bother timestamping its
1888 		 * packets.  So we can look for the presence of a timestamp in
1889 		 * the first data packet and if there, require it in all future
1890 		 * packets.
1891 		 */
1892 
1893 		if (pd->p_len > 0 && (src->scrub->pfss_flags & PFSS_DATA_TS)) {
1894 			/*
1895 			 * Hey!  Someone tried to sneak a packet in.  Or the
1896 			 * stack changed its RFC1323 behavior?!?!
1897 			 */
1898 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
1899 				DPFPRINTF(PF_DEBUG_MISC, "Did not receive expected "
1900 				    "RFC1323 timestamp");
1901 				pf_print_state(state);
1902 				pf_print_flags(tcp_get_flags(th));
1903 				printf("\n");
1904 			}
1905 			REASON_SET(reason, PFRES_TS);
1906 			return (PF_DROP);
1907 		}
1908 	}
1909 
1910 	/*
1911 	 * We will note if a host sends his data packets with or without
1912 	 * timestamps.  And require all data packets to contain a timestamp
1913 	 * if the first does.  PAWS implicitly requires that all data packets be
1914 	 * timestamped.  But I think there are middle-man devices that hijack
1915 	 * TCP streams immediately after the 3whs and don't timestamp their
1916 	 * packets (seen in a WWW accelerator or cache).
1917 	 */
1918 	if (pd->p_len > 0 && src->scrub && (src->scrub->pfss_flags &
1919 	    (PFSS_TIMESTAMP|PFSS_DATA_TS|PFSS_DATA_NOTS)) == PFSS_TIMESTAMP) {
1920 		if (got_ts)
1921 			src->scrub->pfss_flags |= PFSS_DATA_TS;
1922 		else {
1923 			src->scrub->pfss_flags |= PFSS_DATA_NOTS;
1924 			if (V_pf_status.debug >= PF_DEBUG_MISC && dst->scrub &&
1925 			    (dst->scrub->pfss_flags & PFSS_TIMESTAMP)) {
1926 				/* Don't warn if other host rejected RFC1323 */
1927 				DPFPRINTF(PF_DEBUG_MISC, "Broken RFC1323 stack did "
1928 				    "not timestamp data packet. Disabled PAWS "
1929 				    "security.");
1930 				pf_print_state(state);
1931 				pf_print_flags(tcp_get_flags(th));
1932 				printf("\n");
1933 			}
1934 		}
1935 	}
1936 
1937 	/*
1938 	 * Update PAWS values
1939 	 */
1940 	if (got_ts && src->scrub && PFSS_TIMESTAMP == (src->scrub->pfss_flags &
1941 	    (PFSS_PAWS_IDLED|PFSS_TIMESTAMP))) {
1942 		getmicrouptime(&src->scrub->pfss_last);
1943 		if (SEQ_GEQ(tsval, src->scrub->pfss_tsval) ||
1944 		    (src->scrub->pfss_flags & PFSS_PAWS) == 0)
1945 			src->scrub->pfss_tsval = tsval;
1946 
1947 		if (tsecr) {
1948 			if (SEQ_GEQ(tsecr, src->scrub->pfss_tsecr) ||
1949 			    (src->scrub->pfss_flags & PFSS_PAWS) == 0)
1950 				src->scrub->pfss_tsecr = tsecr;
1951 
1952 			if ((src->scrub->pfss_flags & PFSS_PAWS) == 0 &&
1953 			    (SEQ_LT(tsval, src->scrub->pfss_tsval0) ||
1954 			    src->scrub->pfss_tsval0 == 0)) {
1955 				/* tsval0 MUST be the lowest timestamp */
1956 				src->scrub->pfss_tsval0 = tsval;
1957 			}
1958 
1959 			/* Only fully initialized after a TS gets echoed */
1960 			if ((src->scrub->pfss_flags & PFSS_PAWS) == 0)
1961 				src->scrub->pfss_flags |= PFSS_PAWS;
1962 		}
1963 	}
1964 
1965 	/* I have a dream....  TCP segment reassembly.... */
1966 	return (0);
1967 }
1968 
1969 int
pf_normalize_mss(struct pf_pdesc * pd)1970 pf_normalize_mss(struct pf_pdesc *pd)
1971 {
1972 	int		 olen, optsoff;
1973 	uint8_t		 opts[MAX_TCPOPTLEN], *opt;
1974 
1975 	olen = (pd->hdr.tcp.th_off << 2) - sizeof(struct tcphdr);
1976 	optsoff = pd->off + sizeof(struct tcphdr);
1977 	if (olen < TCPOLEN_MAXSEG ||
1978 	    !pf_pull_hdr(pd->m, optsoff, opts, olen, NULL, NULL, pd->af))
1979 		return (0);
1980 
1981 	opt = opts;
1982 	while ((opt = pf_find_tcpopt(opt, opts, olen,
1983 	    TCPOPT_MAXSEG, TCPOLEN_MAXSEG)) != NULL) {
1984 		uint16_t	 mss;
1985 		uint8_t		*mssp = opt + 2;
1986 		memcpy(&mss, mssp, sizeof(mss));
1987 		if (ntohs(mss) > pd->act.max_mss) {
1988 			size_t mssoffopts = mssp - opts;
1989 			pf_patch_16(pd, &mss,
1990 			    htons(pd->act.max_mss), PF_ALGNMNT(mssoffopts));
1991 			m_copyback(pd->m, optsoff + mssoffopts,
1992 			    sizeof(mss), (caddr_t)&mss);
1993 			m_copyback(pd->m, pd->off,
1994 			    sizeof(struct tcphdr), (caddr_t)&pd->hdr.tcp);
1995 		}
1996 
1997 		opt += opt[1];
1998 	}
1999 
2000 	return (0);
2001 }
2002 
2003 int
pf_scan_sctp(struct pf_pdesc * pd)2004 pf_scan_sctp(struct pf_pdesc *pd)
2005 {
2006 	struct sctp_chunkhdr ch = { };
2007 	int chunk_off = sizeof(struct sctphdr);
2008 	int chunk_start;
2009 	int ret;
2010 
2011 	while (pd->off + chunk_off < pd->tot_len) {
2012 		if (!pf_pull_hdr(pd->m, pd->off + chunk_off, &ch, sizeof(ch), NULL,
2013 		    NULL, pd->af))
2014 			return (PF_DROP);
2015 
2016 		/* Length includes the header, this must be at least 4. */
2017 		if (ntohs(ch.chunk_length) < 4)
2018 			return (PF_DROP);
2019 
2020 		chunk_start = chunk_off;
2021 		chunk_off += roundup(ntohs(ch.chunk_length), 4);
2022 
2023 		switch (ch.chunk_type) {
2024 		case SCTP_INITIATION:
2025 		case SCTP_INITIATION_ACK: {
2026 			struct sctp_init_chunk init;
2027 
2028 			if (!pf_pull_hdr(pd->m, pd->off + chunk_start, &init,
2029 			    sizeof(init), NULL, NULL, pd->af))
2030 				return (PF_DROP);
2031 
2032 			/*
2033 			 * RFC 9620, Section 3.3.2, "The Initiate Tag is allowed to have
2034 			 * any value except 0."
2035 			 */
2036 			if (init.init.initiate_tag == 0)
2037 				return (PF_DROP);
2038 			if (init.init.num_inbound_streams == 0)
2039 				return (PF_DROP);
2040 			if (init.init.num_outbound_streams == 0)
2041 				return (PF_DROP);
2042 			if (ntohl(init.init.a_rwnd) < SCTP_MIN_RWND)
2043 				return (PF_DROP);
2044 
2045 			/*
2046 			 * RFC 9260, Section 3.1, INIT chunks MUST have zero
2047 			 * verification tag.
2048 			 */
2049 			if (ch.chunk_type == SCTP_INITIATION &&
2050 			    pd->hdr.sctp.v_tag != 0)
2051 				return (PF_DROP);
2052 
2053 			pd->sctp_initiate_tag = init.init.initiate_tag;
2054 
2055 			if (ch.chunk_type == SCTP_INITIATION)
2056 				pd->sctp_flags |= PFDESC_SCTP_INIT;
2057 			else
2058 				pd->sctp_flags |= PFDESC_SCTP_INIT_ACK;
2059 
2060 			ret = pf_multihome_scan_init(pd->off + chunk_start,
2061 			    ntohs(init.ch.chunk_length), pd);
2062 			if (ret != PF_PASS)
2063 				return (ret);
2064 
2065 			break;
2066 		}
2067 		case SCTP_ABORT_ASSOCIATION:
2068 			pd->sctp_flags |= PFDESC_SCTP_ABORT;
2069 			break;
2070 		case SCTP_SHUTDOWN:
2071 		case SCTP_SHUTDOWN_ACK:
2072 			pd->sctp_flags |= PFDESC_SCTP_SHUTDOWN;
2073 			break;
2074 		case SCTP_SHUTDOWN_COMPLETE:
2075 			pd->sctp_flags |= PFDESC_SCTP_SHUTDOWN_COMPLETE;
2076 			break;
2077 		case SCTP_COOKIE_ECHO:
2078 			pd->sctp_flags |= PFDESC_SCTP_COOKIE;
2079 			break;
2080 		case SCTP_COOKIE_ACK:
2081 			pd->sctp_flags |= PFDESC_SCTP_COOKIE_ACK;
2082 			break;
2083 		case SCTP_DATA:
2084 			pd->sctp_flags |= PFDESC_SCTP_DATA;
2085 			break;
2086 		case SCTP_HEARTBEAT_REQUEST:
2087 			pd->sctp_flags |= PFDESC_SCTP_HEARTBEAT;
2088 			break;
2089 		case SCTP_HEARTBEAT_ACK:
2090 			pd->sctp_flags |= PFDESC_SCTP_HEARTBEAT_ACK;
2091 			break;
2092 		case SCTP_ASCONF:
2093 			pd->sctp_flags |= PFDESC_SCTP_ASCONF;
2094 
2095 			ret = pf_multihome_scan_asconf(pd->off + chunk_start,
2096 			    ntohs(ch.chunk_length), pd);
2097 			if (ret != PF_PASS)
2098 				return (ret);
2099 			break;
2100 		default:
2101 			pd->sctp_flags |= PFDESC_SCTP_OTHER;
2102 			break;
2103 		}
2104 	}
2105 
2106 	/* Validate chunk lengths vs. packet length. */
2107 	if (pd->off + chunk_off != pd->tot_len)
2108 		return (PF_DROP);
2109 
2110 	/*
2111 	 * INIT, INIT_ACK or SHUTDOWN_COMPLETE chunks must always be the only
2112 	 * one in a packet.
2113 	 */
2114 	if ((pd->sctp_flags & PFDESC_SCTP_INIT) &&
2115 	    (pd->sctp_flags & ~PFDESC_SCTP_INIT))
2116 		return (PF_DROP);
2117 	if ((pd->sctp_flags & PFDESC_SCTP_INIT_ACK) &&
2118 	    (pd->sctp_flags & ~PFDESC_SCTP_INIT_ACK))
2119 		return (PF_DROP);
2120 	if ((pd->sctp_flags & PFDESC_SCTP_SHUTDOWN_COMPLETE) &&
2121 	    (pd->sctp_flags & ~PFDESC_SCTP_SHUTDOWN_COMPLETE))
2122 		return (PF_DROP);
2123 	if ((pd->sctp_flags & PFDESC_SCTP_ABORT) &&
2124 	    (pd->sctp_flags & PFDESC_SCTP_DATA)) {
2125 		/*
2126 		 * RFC4960 3.3.7: DATA chunks MUST NOT be
2127 		 * bundled with ABORT.
2128 		 */
2129 		return (PF_DROP);
2130 	}
2131 
2132 	return (PF_PASS);
2133 }
2134 
2135 int
pf_normalize_sctp(struct pf_pdesc * pd)2136 pf_normalize_sctp(struct pf_pdesc *pd)
2137 {
2138 	struct pf_krule	*r, *rm = NULL;
2139 	struct sctphdr	*sh = &pd->hdr.sctp;
2140 	u_short		 reason;
2141 	sa_family_t	 af = pd->af;
2142 	int		 srs;
2143 
2144 	PF_RULES_RASSERT();
2145 
2146 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
2147 	/* Check if there any scrub rules. Lack of scrub rules means enforced
2148 	 * packet normalization operation just like in OpenBSD. */
2149 	srs = (r != NULL);
2150 	while (r != NULL) {
2151 		pf_counter_u64_add(&r->evaluations, 1);
2152 		if (pfi_kkif_match(r->kif, pd->kif) == r->ifnot)
2153 			r = r->skip[PF_SKIP_IFP];
2154 		else if (r->direction && r->direction != pd->dir)
2155 			r = r->skip[PF_SKIP_DIR];
2156 		else if (r->af && r->af != af)
2157 			r = r->skip[PF_SKIP_AF];
2158 		else if (r->proto && r->proto != pd->proto)
2159 			r = r->skip[PF_SKIP_PROTO];
2160 		else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
2161 		    r->src.neg, pd->kif, M_GETFIB(pd->m)))
2162 			r = r->skip[PF_SKIP_SRC_ADDR];
2163 		else if (r->src.port_op && !pf_match_port(r->src.port_op,
2164 			    r->src.port[0], r->src.port[1], sh->src_port))
2165 			r = r->skip[PF_SKIP_SRC_PORT];
2166 		else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
2167 		    r->dst.neg, NULL, M_GETFIB(pd->m)))
2168 			r = r->skip[PF_SKIP_DST_ADDR];
2169 		else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
2170 			    r->dst.port[0], r->dst.port[1], sh->dest_port))
2171 			r = r->skip[PF_SKIP_DST_PORT];
2172 		else {
2173 			rm = r;
2174 			break;
2175 		}
2176 	}
2177 
2178 	if (srs) {
2179 		/* With scrub rules present SCTP normalization happens only
2180 		 * if one of rules has matched and it's not a "no scrub" rule */
2181 		if (rm == NULL || rm->action == PF_NOSCRUB)
2182 			return (PF_PASS);
2183 
2184 		pf_counter_u64_critical_enter();
2185 		pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1);
2186 		pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len);
2187 		pf_counter_u64_critical_exit();
2188 	}
2189 
2190 	/* Verify we're a multiple of 4 bytes long */
2191 	if ((pd->tot_len - pd->off - sizeof(struct sctphdr)) % 4)
2192 		goto sctp_drop;
2193 
2194 	/* INIT chunk needs to be the only chunk */
2195 	if (pd->sctp_flags & PFDESC_SCTP_INIT)
2196 		if (pd->sctp_flags & ~PFDESC_SCTP_INIT)
2197 			goto sctp_drop;
2198 
2199 	return (PF_PASS);
2200 
2201 sctp_drop:
2202 	REASON_SET(&reason, PFRES_NORM);
2203 	if (rm != NULL && r->log)
2204 		PFLOG_PACKET(PF_DROP, reason, r, NULL, NULL, pd,
2205 		    1, NULL);
2206 
2207 	return (PF_DROP);
2208 }
2209 
2210 #if defined(INET) || defined(INET6)
2211 void
pf_scrub(struct pf_pdesc * pd)2212 pf_scrub(struct pf_pdesc *pd)
2213 {
2214 
2215 	struct ip		*h = mtod(pd->m, struct ip *);
2216 #ifdef INET6
2217 	struct ip6_hdr		*h6 = mtod(pd->m, struct ip6_hdr *);
2218 #endif /* INET6 */
2219 
2220 	/* Clear IP_DF if no-df was requested */
2221 	if (pd->af == AF_INET && pd->act.flags & PFSTATE_NODF &&
2222 	    h->ip_off & htons(IP_DF))
2223 	{
2224 		u_int16_t ip_off = h->ip_off;
2225 
2226 		h->ip_off &= htons(~IP_DF);
2227 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
2228 	}
2229 
2230 	/* Enforce a minimum ttl, may cause endless packet loops */
2231 	if (pd->af == AF_INET && pd->act.min_ttl &&
2232 	    h->ip_ttl < pd->act.min_ttl) {
2233 		u_int16_t ip_ttl = h->ip_ttl;
2234 
2235 		h->ip_ttl = pd->act.min_ttl;
2236 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_ttl, h->ip_ttl, 0);
2237 	}
2238 #ifdef INET6
2239 	/* Enforce a minimum ttl, may cause endless packet loops */
2240 	if (pd->af == AF_INET6 && pd->act.min_ttl &&
2241 	    h6->ip6_hlim < pd->act.min_ttl)
2242 		h6->ip6_hlim = pd->act.min_ttl;
2243 #endif /* INET6 */
2244 	/* Enforce tos */
2245 	if (pd->act.flags & PFSTATE_SETTOS) {
2246 		switch (pd->af) {
2247 		case AF_INET: {
2248 			u_int16_t	ov, nv;
2249 
2250 			ov = *(u_int16_t *)h;
2251 			h->ip_tos = pd->act.set_tos | (h->ip_tos & IPTOS_ECN_MASK);
2252 			nv = *(u_int16_t *)h;
2253 
2254 			h->ip_sum = pf_cksum_fixup(h->ip_sum, ov, nv, 0);
2255 			break;
2256 		}
2257 #ifdef INET6
2258 		case AF_INET6:
2259 			h6->ip6_flow &= IPV6_FLOWLABEL_MASK | IPV6_VERSION_MASK;
2260 			h6->ip6_flow |= htonl((pd->act.set_tos | IPV6_ECN(h6)) << 20);
2261 			break;
2262 #endif /* INET6 */
2263 		}
2264 	}
2265 
2266 	/* random-id, but not for fragments */
2267 #ifdef INET
2268 	if (pd->af == AF_INET &&
2269 	    pd->act.flags & PFSTATE_RANDOMID && !(h->ip_off & ~htons(IP_DF))) {
2270 		uint16_t ip_id = h->ip_id;
2271 
2272 		ip_fillid(h, V_ip_random_id);
2273 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_id, h->ip_id, 0);
2274 	}
2275 #endif /* INET */
2276 }
2277 #endif /* INET || INET6 */
2278