xref: /netbsd-src/sys/dist/pf/net/pf_norm.c (revision de1dfb1250df962f1ff3a011772cf58e605aed11)
1 /*	$NetBSD: pf_norm.c,v 1.4 2004/09/08 12:11:25 yamt Exp $	*/
2 /*	$OpenBSD: pf_norm.c,v 1.80 2004/03/09 21:44:41 mcbride Exp $ */
3 
4 /*
5  * Copyright 2001 Niels Provos <provos@citi.umich.edu>
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 
29 #ifdef _KERNEL_OPT
30 #include "opt_inet.h"
31 #endif
32 
33 #include "pflog.h"
34 
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/mbuf.h>
38 #include <sys/filio.h>
39 #include <sys/fcntl.h>
40 #include <sys/socket.h>
41 #include <sys/kernel.h>
42 #include <sys/time.h>
43 #include <sys/pool.h>
44 
45 #ifdef __OpenBSD__
46 #include <dev/rndvar.h>
47 #else
48 #include <sys/rnd.h>
49 #endif
50 #include <net/if.h>
51 #include <net/if_types.h>
52 #include <net/bpf.h>
53 #include <net/route.h>
54 #include <net/if_pflog.h>
55 
56 #include <netinet/in.h>
57 #include <netinet/in_var.h>
58 #include <netinet/in_systm.h>
59 #include <netinet/ip.h>
60 #include <netinet/ip_var.h>
61 #include <netinet/tcp.h>
62 #include <netinet/tcp_seq.h>
63 #include <netinet/udp.h>
64 #include <netinet/ip_icmp.h>
65 
66 #ifdef INET6
67 #include <netinet/ip6.h>
68 #endif /* INET6 */
69 
70 #include <net/pfvar.h>
71 
72 struct pf_frent {
73 	LIST_ENTRY(pf_frent) fr_next;
74 	struct ip *fr_ip;
75 	struct mbuf *fr_m;
76 };
77 
78 struct pf_frcache {
79 	LIST_ENTRY(pf_frcache) fr_next;
80 	uint16_t	fr_off;
81 	uint16_t	fr_end;
82 };
83 
84 #define PFFRAG_SEENLAST	0x0001		/* Seen the last fragment for this */
85 #define PFFRAG_NOBUFFER	0x0002		/* Non-buffering fragment cache */
86 #define PFFRAG_DROP	0x0004		/* Drop all fragments */
87 #define BUFFER_FRAGMENTS(fr)	(!((fr)->fr_flags & PFFRAG_NOBUFFER))
88 
89 struct pf_fragment {
90 	RB_ENTRY(pf_fragment) fr_entry;
91 	TAILQ_ENTRY(pf_fragment) frag_next;
92 	struct in_addr	fr_src;
93 	struct in_addr	fr_dst;
94 	u_int8_t	fr_p;		/* protocol of this fragment */
95 	u_int8_t	fr_flags;	/* status flags */
96 	u_int16_t	fr_id;		/* fragment id for reassemble */
97 	u_int16_t	fr_max;		/* fragment data max */
98 	u_int32_t	fr_timeout;
99 #define fr_queue	fr_u.fru_queue
100 #define fr_cache	fr_u.fru_cache
101 	union {
102 		LIST_HEAD(pf_fragq, pf_frent) fru_queue;	/* buffering */
103 		LIST_HEAD(pf_cacheq, pf_frcache) fru_cache;	/* non-buf */
104 	} fr_u;
105 };
106 
107 TAILQ_HEAD(pf_fragqueue, pf_fragment)	pf_fragqueue;
108 TAILQ_HEAD(pf_cachequeue, pf_fragment)	pf_cachequeue;
109 
110 static __inline int	 pf_frag_compare(struct pf_fragment *,
111 			    struct pf_fragment *);
112 RB_HEAD(pf_frag_tree, pf_fragment)	pf_frag_tree, pf_cache_tree;
113 RB_PROTOTYPE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
114 RB_GENERATE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
115 
116 /* Private prototypes */
117 void			 pf_ip2key(struct pf_fragment *, struct ip *);
118 void			 pf_remove_fragment(struct pf_fragment *);
119 void			 pf_flush_fragments(void);
120 void			 pf_free_fragment(struct pf_fragment *);
121 struct pf_fragment	*pf_find_fragment(struct ip *, struct pf_frag_tree *);
122 struct mbuf		*pf_reassemble(struct mbuf **, struct pf_fragment **,
123 			    struct pf_frent *, int);
124 struct mbuf		*pf_fragcache(struct mbuf **, struct ip*,
125 			    struct pf_fragment **, int, int, int *);
126 u_int16_t		 pf_cksum_fixup(u_int16_t, u_int16_t, u_int16_t);
127 int			 pf_normalize_tcpopt(struct pf_rule *, struct mbuf *,
128 			    struct tcphdr *, int);
129 
130 #define	DPFPRINTF(x)	if (pf_status.debug >= PF_DEBUG_MISC) \
131 			    { printf("%s: ", __func__); printf x ;}
132 
133 /* Globals */
134 struct pool		 pf_frent_pl, pf_frag_pl, pf_cache_pl, pf_cent_pl;
135 struct pool		 pf_state_scrub_pl;
136 int			 pf_nfrents, pf_ncache;
137 
138 
139 void
140 pf_normalize_init(void)
141 {
142 	pool_init(&pf_frent_pl, sizeof(struct pf_frent), 0, 0, 0, "pffrent",
143 	    NULL);
144 	pool_init(&pf_frag_pl, sizeof(struct pf_fragment), 0, 0, 0, "pffrag",
145 	    NULL);
146 	pool_init(&pf_cache_pl, sizeof(struct pf_fragment), 0, 0, 0,
147 	    "pffrcache", NULL);
148 	pool_init(&pf_cent_pl, sizeof(struct pf_frcache), 0, 0, 0, "pffrcent",
149 	    NULL);
150 	pool_init(&pf_state_scrub_pl, sizeof(struct pf_state_scrub), 0, 0, 0,
151 	    "pfstscr", NULL);
152 
153 	pool_sethiwat(&pf_frag_pl, PFFRAG_FRAG_HIWAT);
154 	pool_sethardlimit(&pf_frent_pl, PFFRAG_FRENT_HIWAT, NULL, 0);
155 	pool_sethardlimit(&pf_cache_pl, PFFRAG_FRCACHE_HIWAT, NULL, 0);
156 	pool_sethardlimit(&pf_cent_pl, PFFRAG_FRCENT_HIWAT, NULL, 0);
157 
158 	TAILQ_INIT(&pf_fragqueue);
159 	TAILQ_INIT(&pf_cachequeue);
160 }
161 
162 #ifdef _LKM
163 #define TAILQ_DRAIN(list, element)				\
164 	do {							\
165 		while ((element = TAILQ_FIRST(list)) != NULL)	\
166 			TAILQ_REMOVE(list, element, frag_next);	\
167 	} while (0)
168 
169 void
170 pf_normalize_destroy(void)
171 {
172 	struct pf_fragment *fragment_e;
173 
174 	TAILQ_DRAIN(&pf_fragqueue, fragment_e);
175 	TAILQ_DRAIN(&pf_cachequeue, fragment_e);
176 
177 	pool_destroy(&pf_state_scrub_pl);
178 	pool_destroy(&pf_cent_pl);
179 	pool_destroy(&pf_cache_pl);
180 	pool_destroy(&pf_frag_pl);
181 	pool_destroy(&pf_frent_pl);
182 }
183 #endif
184 
185 static __inline int
186 pf_frag_compare(struct pf_fragment *a, struct pf_fragment *b)
187 {
188 	int	diff;
189 
190 	if ((diff = a->fr_id - b->fr_id))
191 		return (diff);
192 	else if ((diff = a->fr_p - b->fr_p))
193 		return (diff);
194 	else if (a->fr_src.s_addr < b->fr_src.s_addr)
195 		return (-1);
196 	else if (a->fr_src.s_addr > b->fr_src.s_addr)
197 		return (1);
198 	else if (a->fr_dst.s_addr < b->fr_dst.s_addr)
199 		return (-1);
200 	else if (a->fr_dst.s_addr > b->fr_dst.s_addr)
201 		return (1);
202 	return (0);
203 }
204 
205 void
206 pf_purge_expired_fragments(void)
207 {
208 	struct pf_fragment	*frag;
209 	u_int32_t		 expire = time.tv_sec -
210 				    pf_default_rule.timeout[PFTM_FRAG];
211 
212 	while ((frag = TAILQ_LAST(&pf_fragqueue, pf_fragqueue)) != NULL) {
213 		KASSERT(BUFFER_FRAGMENTS(frag));
214 		if (frag->fr_timeout > expire)
215 			break;
216 
217 		DPFPRINTF(("expiring %d(%p)\n", frag->fr_id, frag));
218 		pf_free_fragment(frag);
219 	}
220 
221 	while ((frag = TAILQ_LAST(&pf_cachequeue, pf_cachequeue)) != NULL) {
222 		KASSERT(!BUFFER_FRAGMENTS(frag));
223 		if (frag->fr_timeout > expire)
224 			break;
225 
226 		DPFPRINTF(("expiring %d(%p)\n", frag->fr_id, frag));
227 		pf_free_fragment(frag);
228 		KASSERT(TAILQ_EMPTY(&pf_cachequeue) ||
229 		    TAILQ_LAST(&pf_cachequeue, pf_cachequeue) != frag);
230 	}
231 }
232 
233 /*
234  * Try to flush old fragments to make space for new ones
235  */
236 
237 void
238 pf_flush_fragments(void)
239 {
240 	struct pf_fragment	*frag;
241 	int			 goal;
242 
243 	goal = pf_nfrents * 9 / 10;
244 	DPFPRINTF(("trying to free > %d frents\n",
245 	    pf_nfrents - goal));
246 	while (goal < pf_nfrents) {
247 		frag = TAILQ_LAST(&pf_fragqueue, pf_fragqueue);
248 		if (frag == NULL)
249 			break;
250 		pf_free_fragment(frag);
251 	}
252 
253 
254 	goal = pf_ncache * 9 / 10;
255 	DPFPRINTF(("trying to free > %d cache entries\n",
256 	    pf_ncache - goal));
257 	while (goal < pf_ncache) {
258 		frag = TAILQ_LAST(&pf_cachequeue, pf_cachequeue);
259 		if (frag == NULL)
260 			break;
261 		pf_free_fragment(frag);
262 	}
263 }
264 
265 /* Frees the fragments and all associated entries */
266 
267 void
268 pf_free_fragment(struct pf_fragment *frag)
269 {
270 	struct pf_frent		*frent;
271 	struct pf_frcache	*frcache;
272 
273 	/* Free all fragments */
274 	if (BUFFER_FRAGMENTS(frag)) {
275 		for (frent = LIST_FIRST(&frag->fr_queue); frent;
276 		    frent = LIST_FIRST(&frag->fr_queue)) {
277 			LIST_REMOVE(frent, fr_next);
278 
279 			m_freem(frent->fr_m);
280 			pool_put(&pf_frent_pl, frent);
281 			pf_nfrents--;
282 		}
283 	} else {
284 		for (frcache = LIST_FIRST(&frag->fr_cache); frcache;
285 		    frcache = LIST_FIRST(&frag->fr_cache)) {
286 			LIST_REMOVE(frcache, fr_next);
287 
288 			KASSERT(LIST_EMPTY(&frag->fr_cache) ||
289 			    LIST_FIRST(&frag->fr_cache)->fr_off >
290 			    frcache->fr_end);
291 
292 			pool_put(&pf_cent_pl, frcache);
293 			pf_ncache--;
294 		}
295 	}
296 
297 	pf_remove_fragment(frag);
298 }
299 
300 void
301 pf_ip2key(struct pf_fragment *key, struct ip *ip)
302 {
303 	key->fr_p = ip->ip_p;
304 	key->fr_id = ip->ip_id;
305 	key->fr_src.s_addr = ip->ip_src.s_addr;
306 	key->fr_dst.s_addr = ip->ip_dst.s_addr;
307 }
308 
309 struct pf_fragment *
310 pf_find_fragment(struct ip *ip, struct pf_frag_tree *tree)
311 {
312 	struct pf_fragment	 key;
313 	struct pf_fragment	*frag;
314 
315 	pf_ip2key(&key, ip);
316 
317 	frag = RB_FIND(pf_frag_tree, tree, &key);
318 	if (frag != NULL) {
319 		/* XXX Are we sure we want to update the timeout? */
320 		frag->fr_timeout = time.tv_sec;
321 		if (BUFFER_FRAGMENTS(frag)) {
322 			TAILQ_REMOVE(&pf_fragqueue, frag, frag_next);
323 			TAILQ_INSERT_HEAD(&pf_fragqueue, frag, frag_next);
324 		} else {
325 			TAILQ_REMOVE(&pf_cachequeue, frag, frag_next);
326 			TAILQ_INSERT_HEAD(&pf_cachequeue, frag, frag_next);
327 		}
328 	}
329 
330 	return (frag);
331 }
332 
333 /* Removes a fragment from the fragment queue and frees the fragment */
334 
335 void
336 pf_remove_fragment(struct pf_fragment *frag)
337 {
338 	if (BUFFER_FRAGMENTS(frag)) {
339 		RB_REMOVE(pf_frag_tree, &pf_frag_tree, frag);
340 		TAILQ_REMOVE(&pf_fragqueue, frag, frag_next);
341 		pool_put(&pf_frag_pl, frag);
342 	} else {
343 		RB_REMOVE(pf_frag_tree, &pf_cache_tree, frag);
344 		TAILQ_REMOVE(&pf_cachequeue, frag, frag_next);
345 		pool_put(&pf_cache_pl, frag);
346 	}
347 }
348 
349 #define FR_IP_OFF(fr)	((ntohs((fr)->fr_ip->ip_off) & IP_OFFMASK) << 3)
350 struct mbuf *
351 pf_reassemble(struct mbuf **m0, struct pf_fragment **frag,
352     struct pf_frent *frent, int mff)
353 {
354 	struct mbuf	*m = *m0, *m2;
355 	struct pf_frent	*frea, *next;
356 	struct pf_frent	*frep = NULL;
357 	struct ip	*ip = frent->fr_ip;
358 	int		 hlen = ip->ip_hl << 2;
359 	u_int16_t	 off = (ntohs(ip->ip_off) & IP_OFFMASK) << 3;
360 	u_int16_t	 ip_len = ntohs(ip->ip_len) - ip->ip_hl * 4;
361 	u_int16_t	 max = ip_len + off;
362 
363 	KASSERT(*frag == NULL || BUFFER_FRAGMENTS(*frag));
364 
365 	/* Strip off ip header */
366 	m->m_data += hlen;
367 	m->m_len -= hlen;
368 
369 	/* Create a new reassembly queue for this packet */
370 	if (*frag == NULL) {
371 		*frag = pool_get(&pf_frag_pl, PR_NOWAIT);
372 		if (*frag == NULL) {
373 			pf_flush_fragments();
374 			*frag = pool_get(&pf_frag_pl, PR_NOWAIT);
375 			if (*frag == NULL)
376 				goto drop_fragment;
377 		}
378 
379 		(*frag)->fr_flags = 0;
380 		(*frag)->fr_max = 0;
381 		(*frag)->fr_src = frent->fr_ip->ip_src;
382 		(*frag)->fr_dst = frent->fr_ip->ip_dst;
383 		(*frag)->fr_p = frent->fr_ip->ip_p;
384 		(*frag)->fr_id = frent->fr_ip->ip_id;
385 		(*frag)->fr_timeout = time.tv_sec;
386 		LIST_INIT(&(*frag)->fr_queue);
387 
388 		RB_INSERT(pf_frag_tree, &pf_frag_tree, *frag);
389 		TAILQ_INSERT_HEAD(&pf_fragqueue, *frag, frag_next);
390 
391 		/* We do not have a previous fragment */
392 		frep = NULL;
393 		goto insert;
394 	}
395 
396 	/*
397 	 * Find a fragment after the current one:
398 	 *  - off contains the real shifted offset.
399 	 */
400 	LIST_FOREACH(frea, &(*frag)->fr_queue, fr_next) {
401 		if (FR_IP_OFF(frea) > off)
402 			break;
403 		frep = frea;
404 	}
405 
406 	KASSERT(frep != NULL || frea != NULL);
407 
408 	if (frep != NULL &&
409 	    FR_IP_OFF(frep) + ntohs(frep->fr_ip->ip_len) - frep->fr_ip->ip_hl *
410 	    4 > off)
411 	{
412 		u_int16_t	precut;
413 
414 		precut = FR_IP_OFF(frep) + ntohs(frep->fr_ip->ip_len) -
415 		    frep->fr_ip->ip_hl * 4 - off;
416 		if (precut >= ip_len)
417 			goto drop_fragment;
418 		m_adj(frent->fr_m, precut);
419 		DPFPRINTF(("overlap -%d\n", precut));
420 		/* Enforce 8 byte boundaries */
421 		ip->ip_off = htons(ntohs(ip->ip_off) + (precut >> 3));
422 		off = (ntohs(ip->ip_off) & IP_OFFMASK) << 3;
423 		ip_len -= precut;
424 		ip->ip_len = htons(ip_len);
425 	}
426 
427 	for (; frea != NULL && ip_len + off > FR_IP_OFF(frea);
428 	    frea = next)
429 	{
430 		u_int16_t	aftercut;
431 
432 		aftercut = ip_len + off - FR_IP_OFF(frea);
433 		DPFPRINTF(("adjust overlap %d\n", aftercut));
434 		if (aftercut < ntohs(frea->fr_ip->ip_len) - frea->fr_ip->ip_hl
435 		    * 4)
436 		{
437 			frea->fr_ip->ip_len =
438 			    htons(ntohs(frea->fr_ip->ip_len) - aftercut);
439 			frea->fr_ip->ip_off = htons(ntohs(frea->fr_ip->ip_off) +
440 			    (aftercut >> 3));
441 			m_adj(frea->fr_m, aftercut);
442 			break;
443 		}
444 
445 		/* This fragment is completely overlapped, loose it */
446 		next = LIST_NEXT(frea, fr_next);
447 		m_freem(frea->fr_m);
448 		LIST_REMOVE(frea, fr_next);
449 		pool_put(&pf_frent_pl, frea);
450 		pf_nfrents--;
451 	}
452 
453  insert:
454 	/* Update maximum data size */
455 	if ((*frag)->fr_max < max)
456 		(*frag)->fr_max = max;
457 	/* This is the last segment */
458 	if (!mff)
459 		(*frag)->fr_flags |= PFFRAG_SEENLAST;
460 
461 	if (frep == NULL)
462 		LIST_INSERT_HEAD(&(*frag)->fr_queue, frent, fr_next);
463 	else
464 		LIST_INSERT_AFTER(frep, frent, fr_next);
465 
466 	/* Check if we are completely reassembled */
467 	if (!((*frag)->fr_flags & PFFRAG_SEENLAST))
468 		return (NULL);
469 
470 	/* Check if we have all the data */
471 	off = 0;
472 	for (frep = LIST_FIRST(&(*frag)->fr_queue); frep; frep = next) {
473 		next = LIST_NEXT(frep, fr_next);
474 
475 		off += ntohs(frep->fr_ip->ip_len) - frep->fr_ip->ip_hl * 4;
476 		if (off < (*frag)->fr_max &&
477 		    (next == NULL || FR_IP_OFF(next) != off))
478 		{
479 			DPFPRINTF(("missing fragment at %d, next %d, max %d\n",
480 			    off, next == NULL ? -1 : FR_IP_OFF(next),
481 			    (*frag)->fr_max));
482 			return (NULL);
483 		}
484 	}
485 	DPFPRINTF(("%d < %d?\n", off, (*frag)->fr_max));
486 	if (off < (*frag)->fr_max)
487 		return (NULL);
488 
489 	/* We have all the data */
490 	frent = LIST_FIRST(&(*frag)->fr_queue);
491 	KASSERT(frent != NULL);
492 	if ((frent->fr_ip->ip_hl << 2) + off > IP_MAXPACKET) {
493 		DPFPRINTF(("drop: too big: %d\n", off));
494 		pf_free_fragment(*frag);
495 		*frag = NULL;
496 		return (NULL);
497 	}
498 	next = LIST_NEXT(frent, fr_next);
499 
500 	/* Magic from ip_input */
501 	ip = frent->fr_ip;
502 	m = frent->fr_m;
503 	m2 = m->m_next;
504 	m->m_next = NULL;
505 	m_cat(m, m2);
506 	pool_put(&pf_frent_pl, frent);
507 	pf_nfrents--;
508 	for (frent = next; frent != NULL; frent = next) {
509 		next = LIST_NEXT(frent, fr_next);
510 
511 		m2 = frent->fr_m;
512 		pool_put(&pf_frent_pl, frent);
513 		pf_nfrents--;
514 		m_cat(m, m2);
515 	}
516 
517 	ip->ip_src = (*frag)->fr_src;
518 	ip->ip_dst = (*frag)->fr_dst;
519 
520 	/* Remove from fragment queue */
521 	pf_remove_fragment(*frag);
522 	*frag = NULL;
523 
524 	hlen = ip->ip_hl << 2;
525 	ip->ip_len = htons(off + hlen);
526 	m->m_len += hlen;
527 	m->m_data -= hlen;
528 
529 	/* some debugging cruft by sklower, below, will go away soon */
530 	/* XXX this should be done elsewhere */
531 	if (m->m_flags & M_PKTHDR) {
532 		int plen = 0;
533 		for (m2 = m; m2; m2 = m2->m_next)
534 			plen += m2->m_len;
535 		m->m_pkthdr.len = plen;
536 	}
537 
538 	DPFPRINTF(("complete: %p(%d)\n", m, ntohs(ip->ip_len)));
539 	return (m);
540 
541  drop_fragment:
542 	/* Oops - fail safe - drop packet */
543 	pool_put(&pf_frent_pl, frent);
544 	pf_nfrents--;
545 	m_freem(m);
546 	return (NULL);
547 }
548 
549 struct mbuf *
550 pf_fragcache(struct mbuf **m0, struct ip *h, struct pf_fragment **frag, int mff,
551     int drop, int *nomem)
552 {
553 	struct mbuf		*m = *m0;
554 	struct pf_frcache	*frp, *fra, *cur = NULL;
555 	int			 ip_len = ntohs(h->ip_len) - (h->ip_hl << 2);
556 	u_int16_t		 off = ntohs(h->ip_off) << 3;
557 	u_int16_t		 max = ip_len + off;
558 	int			 hosed = 0;
559 
560 	KASSERT(*frag == NULL || !BUFFER_FRAGMENTS(*frag));
561 
562 	/* Create a new range queue for this packet */
563 	if (*frag == NULL) {
564 		*frag = pool_get(&pf_cache_pl, PR_NOWAIT);
565 		if (*frag == NULL) {
566 			pf_flush_fragments();
567 			*frag = pool_get(&pf_cache_pl, PR_NOWAIT);
568 			if (*frag == NULL)
569 				goto no_mem;
570 		}
571 
572 		/* Get an entry for the queue */
573 		cur = pool_get(&pf_cent_pl, PR_NOWAIT);
574 		if (cur == NULL) {
575 			pool_put(&pf_cache_pl, *frag);
576 			*frag = NULL;
577 			goto no_mem;
578 		}
579 		pf_ncache++;
580 
581 		(*frag)->fr_flags = PFFRAG_NOBUFFER;
582 		(*frag)->fr_max = 0;
583 		(*frag)->fr_src = h->ip_src;
584 		(*frag)->fr_dst = h->ip_dst;
585 		(*frag)->fr_p = h->ip_p;
586 		(*frag)->fr_id = h->ip_id;
587 		(*frag)->fr_timeout = time.tv_sec;
588 
589 		cur->fr_off = off;
590 		cur->fr_end = max;
591 		LIST_INIT(&(*frag)->fr_cache);
592 		LIST_INSERT_HEAD(&(*frag)->fr_cache, cur, fr_next);
593 
594 		RB_INSERT(pf_frag_tree, &pf_cache_tree, *frag);
595 		TAILQ_INSERT_HEAD(&pf_cachequeue, *frag, frag_next);
596 
597 		DPFPRINTF(("fragcache[%d]: new %d-%d\n", h->ip_id, off, max));
598 
599 		goto pass;
600 	}
601 
602 	/*
603 	 * Find a fragment after the current one:
604 	 *  - off contains the real shifted offset.
605 	 */
606 	frp = NULL;
607 	LIST_FOREACH(fra, &(*frag)->fr_cache, fr_next) {
608 		if (fra->fr_off > off)
609 			break;
610 		frp = fra;
611 	}
612 
613 	KASSERT(frp != NULL || fra != NULL);
614 
615 	if (frp != NULL) {
616 		int	precut;
617 
618 		precut = frp->fr_end - off;
619 		if (precut >= ip_len) {
620 			/* Fragment is entirely a duplicate */
621 			DPFPRINTF(("fragcache[%d]: dead (%d-%d) %d-%d\n",
622 			    h->ip_id, frp->fr_off, frp->fr_end, off, max));
623 			goto drop_fragment;
624 		}
625 		if (precut == 0) {
626 			/* They are adjacent.  Fixup cache entry */
627 			DPFPRINTF(("fragcache[%d]: adjacent (%d-%d) %d-%d\n",
628 			    h->ip_id, frp->fr_off, frp->fr_end, off, max));
629 			frp->fr_end = max;
630 		} else if (precut > 0) {
631 			/* The first part of this payload overlaps with a
632 			 * fragment that has already been passed.
633 			 * Need to trim off the first part of the payload.
634 			 * But to do so easily, we need to create another
635 			 * mbuf to throw the original header into.
636 			 */
637 
638 			DPFPRINTF(("fragcache[%d]: chop %d (%d-%d) %d-%d\n",
639 			    h->ip_id, precut, frp->fr_off, frp->fr_end, off,
640 			    max));
641 
642 			off += precut;
643 			max -= precut;
644 			/* Update the previous frag to encompass this one */
645 			frp->fr_end = max;
646 
647 			if (!drop) {
648 				/* XXX Optimization opportunity
649 				 * This is a very heavy way to trim the payload.
650 				 * we could do it much faster by diddling mbuf
651 				 * internals but that would be even less legible
652 				 * than this mbuf magic.  For my next trick,
653 				 * I'll pull a rabbit out of my laptop.
654 				 */
655 #ifdef __OpenBSD__
656 				*m0 = m_copym2(m, 0, h->ip_hl << 2, M_NOWAIT);
657 #else
658 				*m0 = m_dup(m, 0, h->ip_hl << 2, M_NOWAIT);
659 #endif
660 				if (*m0 == NULL)
661 					goto no_mem;
662 				KASSERT((*m0)->m_next == NULL);
663 				m_adj(m, precut + (h->ip_hl << 2));
664 				m_cat(*m0, m);
665 				m = *m0;
666 				if (m->m_flags & M_PKTHDR) {
667 					int plen = 0;
668 					struct mbuf *t;
669 					for (t = m; t; t = t->m_next)
670 						plen += t->m_len;
671 					m->m_pkthdr.len = plen;
672 				}
673 
674 
675 				h = mtod(m, struct ip *);
676 
677 
678 				KASSERT((int)m->m_len ==
679 				    ntohs(h->ip_len) - precut);
680 				h->ip_off = htons(ntohs(h->ip_off) +
681 				    (precut >> 3));
682 				h->ip_len = htons(ntohs(h->ip_len) - precut);
683 			} else {
684 				hosed++;
685 			}
686 		} else {
687 			/* There is a gap between fragments */
688 
689 			DPFPRINTF(("fragcache[%d]: gap %d (%d-%d) %d-%d\n",
690 			    h->ip_id, -precut, frp->fr_off, frp->fr_end, off,
691 			    max));
692 
693 			cur = pool_get(&pf_cent_pl, PR_NOWAIT);
694 			if (cur == NULL)
695 				goto no_mem;
696 			pf_ncache++;
697 
698 			cur->fr_off = off;
699 			cur->fr_end = max;
700 			LIST_INSERT_AFTER(frp, cur, fr_next);
701 		}
702 	}
703 
704 	if (fra != NULL) {
705 		int	aftercut;
706 		int	merge = 0;
707 
708 		aftercut = max - fra->fr_off;
709 		if (aftercut == 0) {
710 			/* Adjacent fragments */
711 			DPFPRINTF(("fragcache[%d]: adjacent %d-%d (%d-%d)\n",
712 			    h->ip_id, off, max, fra->fr_off, fra->fr_end));
713 			fra->fr_off = off;
714 			merge = 1;
715 		} else if (aftercut > 0) {
716 			/* Need to chop off the tail of this fragment */
717 			DPFPRINTF(("fragcache[%d]: chop %d %d-%d (%d-%d)\n",
718 			    h->ip_id, aftercut, off, max, fra->fr_off,
719 			    fra->fr_end));
720 			fra->fr_off = off;
721 			max -= aftercut;
722 
723 			merge = 1;
724 
725 			if (!drop) {
726 				m_adj(m, -aftercut);
727 				if (m->m_flags & M_PKTHDR) {
728 					int plen = 0;
729 					struct mbuf *t;
730 					for (t = m; t; t = t->m_next)
731 						plen += t->m_len;
732 					m->m_pkthdr.len = plen;
733 				}
734 				h = mtod(m, struct ip *);
735 				KASSERT((int)m->m_len ==
736 				    ntohs(h->ip_len) - aftercut);
737 				h->ip_len = htons(ntohs(h->ip_len) - aftercut);
738 			} else {
739 				hosed++;
740 			}
741 		} else {
742 			/* There is a gap between fragments */
743 			DPFPRINTF(("fragcache[%d]: gap %d %d-%d (%d-%d)\n",
744 			    h->ip_id, -aftercut, off, max, fra->fr_off,
745 			    fra->fr_end));
746 
747 			cur = pool_get(&pf_cent_pl, PR_NOWAIT);
748 			if (cur == NULL)
749 				goto no_mem;
750 			pf_ncache++;
751 
752 			cur->fr_off = off;
753 			cur->fr_end = max;
754 			LIST_INSERT_BEFORE(fra, cur, fr_next);
755 		}
756 
757 
758 		/* Need to glue together two separate fragment descriptors */
759 		if (merge) {
760 			if (cur && fra->fr_off <= cur->fr_end) {
761 				/* Need to merge in a previous 'cur' */
762 				DPFPRINTF(("fragcache[%d]: adjacent(merge "
763 				    "%d-%d) %d-%d (%d-%d)\n",
764 				    h->ip_id, cur->fr_off, cur->fr_end, off,
765 				    max, fra->fr_off, fra->fr_end));
766 				fra->fr_off = cur->fr_off;
767 				LIST_REMOVE(cur, fr_next);
768 				pool_put(&pf_cent_pl, cur);
769 				pf_ncache--;
770 				cur = NULL;
771 
772 			} else if (frp && fra->fr_off <= frp->fr_end) {
773 				/* Need to merge in a modified 'frp' */
774 				KASSERT(cur == NULL);
775 				DPFPRINTF(("fragcache[%d]: adjacent(merge "
776 				    "%d-%d) %d-%d (%d-%d)\n",
777 				    h->ip_id, frp->fr_off, frp->fr_end, off,
778 				    max, fra->fr_off, fra->fr_end));
779 				fra->fr_off = frp->fr_off;
780 				LIST_REMOVE(frp, fr_next);
781 				pool_put(&pf_cent_pl, frp);
782 				pf_ncache--;
783 				frp = NULL;
784 
785 			}
786 		}
787 	}
788 
789 	if (hosed) {
790 		/*
791 		 * We must keep tracking the overall fragment even when
792 		 * we're going to drop it anyway so that we know when to
793 		 * free the overall descriptor.  Thus we drop the frag late.
794 		 */
795 		goto drop_fragment;
796 	}
797 
798 
799  pass:
800 	/* Update maximum data size */
801 	if ((*frag)->fr_max < max)
802 		(*frag)->fr_max = max;
803 
804 	/* This is the last segment */
805 	if (!mff)
806 		(*frag)->fr_flags |= PFFRAG_SEENLAST;
807 
808 	/* Check if we are completely reassembled */
809 	if (((*frag)->fr_flags & PFFRAG_SEENLAST) &&
810 	    LIST_FIRST(&(*frag)->fr_cache)->fr_off == 0 &&
811 	    LIST_FIRST(&(*frag)->fr_cache)->fr_end == (*frag)->fr_max) {
812 		/* Remove from fragment queue */
813 		DPFPRINTF(("fragcache[%d]: done 0-%d\n", h->ip_id,
814 		    (*frag)->fr_max));
815 		pf_free_fragment(*frag);
816 		*frag = NULL;
817 	}
818 
819 	return (m);
820 
821  no_mem:
822 	*nomem = 1;
823 
824 	/* Still need to pay attention to !IP_MF */
825 	if (!mff && *frag != NULL)
826 		(*frag)->fr_flags |= PFFRAG_SEENLAST;
827 
828 	m_freem(m);
829 	return (NULL);
830 
831  drop_fragment:
832 
833 	/* Still need to pay attention to !IP_MF */
834 	if (!mff && *frag != NULL)
835 		(*frag)->fr_flags |= PFFRAG_SEENLAST;
836 
837 	if (drop) {
838 		/* This fragment has been deemed bad.  Don't reass */
839 		if (((*frag)->fr_flags & PFFRAG_DROP) == 0)
840 			DPFPRINTF(("fragcache[%d]: dropping overall fragment\n",
841 			    h->ip_id));
842 		(*frag)->fr_flags |= PFFRAG_DROP;
843 	}
844 
845 	m_freem(m);
846 	return (NULL);
847 }
848 
849 int
850 pf_normalize_ip(struct mbuf **m0, int dir, struct pfi_kif *kif, u_short *reason)
851 {
852 	struct mbuf		*m = *m0;
853 	struct pf_rule		*r;
854 	struct pf_frent		*frent;
855 	struct pf_fragment	*frag = NULL;
856 	struct ip		*h = mtod(m, struct ip *);
857 	int			 mff = (ntohs(h->ip_off) & IP_MF);
858 	int			 hlen = h->ip_hl << 2;
859 	u_int16_t		 fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
860 	u_int16_t		 max;
861 	int			 ip_len;
862 	int			 ip_off;
863 
864 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
865 	while (r != NULL) {
866 		r->evaluations++;
867 		if (r->kif != NULL &&
868 		    (r->kif != kif && r->kif != kif->pfik_parent) == !r->ifnot)
869 			r = r->skip[PF_SKIP_IFP].ptr;
870 		else if (r->direction && r->direction != dir)
871 			r = r->skip[PF_SKIP_DIR].ptr;
872 		else if (r->af && r->af != AF_INET)
873 			r = r->skip[PF_SKIP_AF].ptr;
874 		else if (r->proto && r->proto != h->ip_p)
875 			r = r->skip[PF_SKIP_PROTO].ptr;
876 		else if (PF_MISMATCHAW(&r->src.addr,
877 		    (struct pf_addr *)&h->ip_src.s_addr, AF_INET, r->src.not))
878 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
879 		else if (PF_MISMATCHAW(&r->dst.addr,
880 		    (struct pf_addr *)&h->ip_dst.s_addr, AF_INET, r->dst.not))
881 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
882 		else
883 			break;
884 	}
885 
886 	if (r == NULL)
887 		return (PF_PASS);
888 	else
889 		r->packets++;
890 
891 	/* Check for illegal packets */
892 	if (hlen < (int)sizeof(struct ip))
893 		goto drop;
894 
895 	if (hlen > ntohs(h->ip_len))
896 		goto drop;
897 
898 	/* Clear IP_DF if the rule uses the no-df option */
899 	if (r->rule_flag & PFRULE_NODF)
900 		h->ip_off &= htons(~IP_DF);
901 
902 	/* We will need other tests here */
903 	if (!fragoff && !mff)
904 		goto no_fragment;
905 
906 	/* We're dealing with a fragment now. Don't allow fragments
907 	 * with IP_DF to enter the cache. If the flag was cleared by
908 	 * no-df above, fine. Otherwise drop it.
909 	 */
910 	if (h->ip_off & htons(IP_DF)) {
911 		DPFPRINTF(("IP_DF\n"));
912 		goto bad;
913 	}
914 
915 	ip_len = ntohs(h->ip_len) - hlen;
916 	ip_off = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
917 
918 	/* All fragments are 8 byte aligned */
919 	if (mff && (ip_len & 0x7)) {
920 		DPFPRINTF(("mff and %d\n", ip_len));
921 		goto bad;
922 	}
923 
924 	/* Respect maximum length */
925 	if (fragoff + ip_len > IP_MAXPACKET) {
926 		DPFPRINTF(("max packet %d\n", fragoff + ip_len));
927 		goto bad;
928 	}
929 	max = fragoff + ip_len;
930 
931 	if ((r->rule_flag & (PFRULE_FRAGCROP|PFRULE_FRAGDROP)) == 0) {
932 		/* Fully buffer all of the fragments */
933 
934 		frag = pf_find_fragment(h, &pf_frag_tree);
935 
936 		/* Check if we saw the last fragment already */
937 		if (frag != NULL && (frag->fr_flags & PFFRAG_SEENLAST) &&
938 		    max > frag->fr_max)
939 			goto bad;
940 
941 		/* Get an entry for the fragment queue */
942 		frent = pool_get(&pf_frent_pl, PR_NOWAIT);
943 		if (frent == NULL) {
944 			REASON_SET(reason, PFRES_MEMORY);
945 			return (PF_DROP);
946 		}
947 		pf_nfrents++;
948 		frent->fr_ip = h;
949 		frent->fr_m = m;
950 
951 		/* Might return a completely reassembled mbuf, or NULL */
952 		DPFPRINTF(("reass frag %d @ %d-%d\n", h->ip_id, fragoff, max));
953 		*m0 = m = pf_reassemble(m0, &frag, frent, mff);
954 
955 		if (m == NULL)
956 			return (PF_DROP);
957 
958 		if (frag != NULL && (frag->fr_flags & PFFRAG_DROP))
959 			goto drop;
960 
961 		h = mtod(m, struct ip *);
962 	} else {
963 		/* non-buffering fragment cache (drops or masks overlaps) */
964 		int	nomem = 0;
965 
966 		if (dir == PF_OUT) {
967 			if (m_tag_find(m, PACKET_TAG_PF_FRAGCACHE, NULL) !=
968 			    NULL) {
969 				/* Already passed the fragment cache in the
970 				 * input direction.  If we continued, it would
971 				 * appear to be a dup and would be dropped.
972 				 */
973 				goto fragment_pass;
974 			}
975 		}
976 
977 		frag = pf_find_fragment(h, &pf_cache_tree);
978 
979 		/* Check if we saw the last fragment already */
980 		if (frag != NULL && (frag->fr_flags & PFFRAG_SEENLAST) &&
981 		    max > frag->fr_max) {
982 			if (r->rule_flag & PFRULE_FRAGDROP)
983 				frag->fr_flags |= PFFRAG_DROP;
984 			goto bad;
985 		}
986 
987 		*m0 = m = pf_fragcache(m0, h, &frag, mff,
988 		    (r->rule_flag & PFRULE_FRAGDROP) ? 1 : 0, &nomem);
989 		if (m == NULL) {
990 			if (nomem)
991 				goto no_mem;
992 			goto drop;
993 		}
994 
995 		if (dir == PF_IN) {
996 			struct m_tag	*mtag;
997 
998 			mtag = m_tag_get(PACKET_TAG_PF_FRAGCACHE, 0, M_NOWAIT);
999 			if (mtag == NULL)
1000 				goto no_mem;
1001 			m_tag_prepend(m, mtag);
1002 		}
1003 		if (frag != NULL && (frag->fr_flags & PFFRAG_DROP))
1004 			goto drop;
1005 		goto fragment_pass;
1006 	}
1007 
1008  no_fragment:
1009 	/* At this point, only IP_DF is allowed in ip_off */
1010 	h->ip_off &= htons(IP_DF);
1011 
1012 	/* Enforce a minimum ttl, may cause endless packet loops */
1013 	if (r->min_ttl && h->ip_ttl < r->min_ttl)
1014 		h->ip_ttl = r->min_ttl;
1015 
1016 	if (r->rule_flag & PFRULE_RANDOMID)
1017 		h->ip_id = ip_randomid();
1018 
1019 	return (PF_PASS);
1020 
1021  fragment_pass:
1022 	/* Enforce a minimum ttl, may cause endless packet loops */
1023 	if (r->min_ttl && h->ip_ttl < r->min_ttl)
1024 		h->ip_ttl = r->min_ttl;
1025 
1026 	return (PF_PASS);
1027 
1028  no_mem:
1029 	REASON_SET(reason, PFRES_MEMORY);
1030 	if (r != NULL && r->log)
1031 		PFLOG_PACKET(kif, h, m, AF_INET, dir, *reason, r, NULL, NULL);
1032 	return (PF_DROP);
1033 
1034  drop:
1035 	REASON_SET(reason, PFRES_NORM);
1036 	if (r != NULL && r->log)
1037 		PFLOG_PACKET(kif, h, m, AF_INET, dir, *reason, r, NULL, NULL);
1038 	return (PF_DROP);
1039 
1040  bad:
1041 	DPFPRINTF(("dropping bad fragment\n"));
1042 
1043 	/* Free associated fragments */
1044 	if (frag != NULL)
1045 		pf_free_fragment(frag);
1046 
1047 	REASON_SET(reason, PFRES_FRAG);
1048 	if (r != NULL && r->log)
1049 		PFLOG_PACKET(kif, h, m, AF_INET, dir, *reason, r, NULL, NULL);
1050 
1051 	return (PF_DROP);
1052 }
1053 
1054 #ifdef INET6
1055 int
1056 pf_normalize_ip6(struct mbuf **m0, int dir, struct pfi_kif *kif,
1057     u_short *reason)
1058 {
1059 	struct mbuf		*m = *m0;
1060 	struct pf_rule		*r;
1061 	struct ip6_hdr		*h = mtod(m, struct ip6_hdr *);
1062 	int			 off;
1063 	struct ip6_ext		 ext;
1064 	struct ip6_opt		 opt;
1065 	struct ip6_opt_jumbo	 jumbo;
1066 	struct ip6_frag		 frag;
1067 	u_int32_t		 jumbolen = 0, plen;
1068 	u_int16_t		 fragoff = 0;
1069 	int			 optend;
1070 	int			 ooff;
1071 	u_int8_t		 proto;
1072 	int			 terminal;
1073 
1074 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1075 	while (r != NULL) {
1076 		r->evaluations++;
1077 		if (r->kif != NULL &&
1078 		    (r->kif != kif && r->kif != kif->pfik_parent) == !r->ifnot)
1079 			r = r->skip[PF_SKIP_IFP].ptr;
1080 		else if (r->direction && r->direction != dir)
1081 			r = r->skip[PF_SKIP_DIR].ptr;
1082 		else if (r->af && r->af != AF_INET6)
1083 			r = r->skip[PF_SKIP_AF].ptr;
1084 #if 0 /* header chain! */
1085 		else if (r->proto && r->proto != h->ip6_nxt)
1086 			r = r->skip[PF_SKIP_PROTO].ptr;
1087 #endif
1088 		else if (PF_MISMATCHAW(&r->src.addr,
1089 		    (struct pf_addr *)&h->ip6_src, AF_INET6, r->src.not))
1090 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
1091 		else if (PF_MISMATCHAW(&r->dst.addr,
1092 		    (struct pf_addr *)&h->ip6_dst, AF_INET6, r->dst.not))
1093 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
1094 		else
1095 			break;
1096 	}
1097 
1098 	if (r == NULL)
1099 		return (PF_PASS);
1100 	else
1101 		r->packets++;
1102 
1103 	/* Check for illegal packets */
1104 	if (sizeof(struct ip6_hdr) + IPV6_MAXPACKET < m->m_pkthdr.len)
1105 		goto drop;
1106 
1107 	off = sizeof(struct ip6_hdr);
1108 	proto = h->ip6_nxt;
1109 	terminal = 0;
1110 	do {
1111 		switch (proto) {
1112 		case IPPROTO_FRAGMENT:
1113 			goto fragment;
1114 			break;
1115 		case IPPROTO_AH:
1116 		case IPPROTO_ROUTING:
1117 		case IPPROTO_DSTOPTS:
1118 			if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL,
1119 			    NULL, AF_INET6))
1120 				goto shortpkt;
1121 			if (proto == IPPROTO_AH)
1122 				off += (ext.ip6e_len + 2) * 4;
1123 			else
1124 				off += (ext.ip6e_len + 1) * 8;
1125 			proto = ext.ip6e_nxt;
1126 			break;
1127 		case IPPROTO_HOPOPTS:
1128 			if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL,
1129 			    NULL, AF_INET6))
1130 				goto shortpkt;
1131 			optend = off + (ext.ip6e_len + 1) * 8;
1132 			ooff = off + sizeof(ext);
1133 			do {
1134 				if (!pf_pull_hdr(m, ooff, &opt.ip6o_type,
1135 				    sizeof(opt.ip6o_type), NULL, NULL,
1136 				    AF_INET6))
1137 					goto shortpkt;
1138 				if (opt.ip6o_type == IP6OPT_PAD1) {
1139 					ooff++;
1140 					continue;
1141 				}
1142 				if (!pf_pull_hdr(m, ooff, &opt, sizeof(opt),
1143 				    NULL, NULL, AF_INET6))
1144 					goto shortpkt;
1145 				if (ooff + sizeof(opt) + opt.ip6o_len > optend)
1146 					goto drop;
1147 				switch (opt.ip6o_type) {
1148 				case IP6OPT_JUMBO:
1149 					if (h->ip6_plen != 0)
1150 						goto drop;
1151 					if (!pf_pull_hdr(m, ooff, &jumbo,
1152 					    sizeof(jumbo), NULL, NULL,
1153 					    AF_INET6))
1154 						goto shortpkt;
1155 					memcpy(&jumbolen, jumbo.ip6oj_jumbo_len,
1156 					    sizeof(jumbolen));
1157 					jumbolen = ntohl(jumbolen);
1158 					if (jumbolen <= IPV6_MAXPACKET)
1159 						goto drop;
1160 					if (sizeof(struct ip6_hdr) + jumbolen !=
1161 					    m->m_pkthdr.len)
1162 						goto drop;
1163 					break;
1164 				default:
1165 					break;
1166 				}
1167 				ooff += sizeof(opt) + opt.ip6o_len;
1168 			} while (ooff < optend);
1169 
1170 			off = optend;
1171 			proto = ext.ip6e_nxt;
1172 			break;
1173 		default:
1174 			terminal = 1;
1175 			break;
1176 		}
1177 	} while (!terminal);
1178 
1179 	/* jumbo payload option must be present, or plen > 0 */
1180 	if (ntohs(h->ip6_plen) == 0)
1181 		plen = jumbolen;
1182 	else
1183 		plen = ntohs(h->ip6_plen);
1184 	if (plen == 0)
1185 		goto drop;
1186 	if (sizeof(struct ip6_hdr) + plen > m->m_pkthdr.len)
1187 		goto shortpkt;
1188 
1189 	/* Enforce a minimum ttl, may cause endless packet loops */
1190 	if (r->min_ttl && h->ip6_hlim < r->min_ttl)
1191 		h->ip6_hlim = r->min_ttl;
1192 
1193 	return (PF_PASS);
1194 
1195  fragment:
1196 	if (ntohs(h->ip6_plen) == 0 || jumbolen)
1197 		goto drop;
1198 	plen = ntohs(h->ip6_plen);
1199 
1200 	if (!pf_pull_hdr(m, off, &frag, sizeof(frag), NULL, NULL, AF_INET6))
1201 		goto shortpkt;
1202 	fragoff = ntohs(frag.ip6f_offlg & IP6F_OFF_MASK);
1203 	if (fragoff + (plen - off - sizeof(frag)) > IPV6_MAXPACKET)
1204 		goto badfrag;
1205 
1206 	/* do something about it */
1207 	return (PF_PASS);
1208 
1209  shortpkt:
1210 	REASON_SET(reason, PFRES_SHORT);
1211 	if (r != NULL && r->log)
1212 		PFLOG_PACKET(kif, h, m, AF_INET6, dir, *reason, r, NULL, NULL);
1213 	return (PF_DROP);
1214 
1215  drop:
1216 	REASON_SET(reason, PFRES_NORM);
1217 	if (r != NULL && r->log)
1218 		PFLOG_PACKET(kif, h, m, AF_INET6, dir, *reason, r, NULL, NULL);
1219 	return (PF_DROP);
1220 
1221  badfrag:
1222 	REASON_SET(reason, PFRES_FRAG);
1223 	if (r != NULL && r->log)
1224 		PFLOG_PACKET(kif, h, m, AF_INET6, dir, *reason, r, NULL, NULL);
1225 	return (PF_DROP);
1226 }
1227 #endif
1228 
1229 int
1230 pf_normalize_tcp(int dir, struct pfi_kif *kif, struct mbuf *m, int ipoff,
1231     int off, void *h, struct pf_pdesc *pd)
1232 {
1233 	struct pf_rule	*r, *rm = NULL;
1234 	struct tcphdr	*th = pd->hdr.tcp;
1235 	int		 rewrite = 0;
1236 	u_short		 reason;
1237 	u_int8_t	 flags;
1238 	sa_family_t	 af = pd->af;
1239 
1240 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1241 	while (r != NULL) {
1242 		r->evaluations++;
1243 		if (r->kif != NULL &&
1244 		    (r->kif != kif && r->kif != kif->pfik_parent) == !r->ifnot)
1245 			r = r->skip[PF_SKIP_IFP].ptr;
1246 		else if (r->direction && r->direction != dir)
1247 			r = r->skip[PF_SKIP_DIR].ptr;
1248 		else if (r->af && r->af != af)
1249 			r = r->skip[PF_SKIP_AF].ptr;
1250 		else if (r->proto && r->proto != pd->proto)
1251 			r = r->skip[PF_SKIP_PROTO].ptr;
1252 		else if (PF_MISMATCHAW(&r->src.addr, pd->src, af, r->src.not))
1253 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
1254 		else if (r->src.port_op && !pf_match_port(r->src.port_op,
1255 			    r->src.port[0], r->src.port[1], th->th_sport))
1256 			r = r->skip[PF_SKIP_SRC_PORT].ptr;
1257 		else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af, r->dst.not))
1258 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
1259 		else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
1260 			    r->dst.port[0], r->dst.port[1], th->th_dport))
1261 			r = r->skip[PF_SKIP_DST_PORT].ptr;
1262 		else if (r->os_fingerprint != PF_OSFP_ANY && !pf_osfp_match(
1263 			    pf_osfp_fingerprint(pd, m, off, th),
1264 			    r->os_fingerprint))
1265 			r = TAILQ_NEXT(r, entries);
1266 		else {
1267 			rm = r;
1268 			break;
1269 		}
1270 	}
1271 
1272 	if (rm == NULL)
1273 		return (PF_PASS);
1274 	else
1275 		r->packets++;
1276 
1277 	if (rm->rule_flag & PFRULE_REASSEMBLE_TCP)
1278 		pd->flags |= PFDESC_TCP_NORM;
1279 
1280 	flags = th->th_flags;
1281 	if (flags & TH_SYN) {
1282 		/* Illegal packet */
1283 		if (flags & TH_RST)
1284 			goto tcp_drop;
1285 
1286 		if (flags & TH_FIN)
1287 			flags &= ~TH_FIN;
1288 	} else {
1289 		/* Illegal packet */
1290 		if (!(flags & (TH_ACK|TH_RST)))
1291 			goto tcp_drop;
1292 	}
1293 
1294 	if (!(flags & TH_ACK)) {
1295 		/* These flags are only valid if ACK is set */
1296 		if ((flags & TH_FIN) || (flags & TH_PUSH) || (flags & TH_URG))
1297 			goto tcp_drop;
1298 	}
1299 
1300 	/* Check for illegal header length */
1301 	if (th->th_off < (sizeof(struct tcphdr) >> 2))
1302 		goto tcp_drop;
1303 
1304 	/* If flags changed, or reserved data set, then adjust */
1305 	if (flags != th->th_flags || th->th_x2 != 0) {
1306 		u_int16_t	ov, nv;
1307 
1308 		ov = *(u_int16_t *)(&th->th_ack + 1);
1309 		th->th_flags = flags;
1310 		th->th_x2 = 0;
1311 		nv = *(u_int16_t *)(&th->th_ack + 1);
1312 
1313 		th->th_sum = pf_cksum_fixup(th->th_sum, ov, nv);
1314 		rewrite = 1;
1315 	}
1316 
1317 	/* Remove urgent pointer, if TH_URG is not set */
1318 	if (!(flags & TH_URG) && th->th_urp) {
1319 		th->th_sum = pf_cksum_fixup(th->th_sum, th->th_urp, 0);
1320 		th->th_urp = 0;
1321 		rewrite = 1;
1322 	}
1323 
1324 	/* Process options */
1325 	if (r->max_mss && pf_normalize_tcpopt(r, m, th, off))
1326 		rewrite = 1;
1327 
1328 	/* copy back packet headers if we sanitized */
1329 	if (rewrite)
1330 		m_copyback(m, off, sizeof(*th), th);
1331 
1332 	return (PF_PASS);
1333 
1334  tcp_drop:
1335 	REASON_SET(&reason, PFRES_NORM);
1336 	if (rm != NULL && r->log)
1337 		PFLOG_PACKET(kif, h, m, AF_INET, dir, reason, r, NULL, NULL);
1338 	return (PF_DROP);
1339 }
1340 
1341 int
1342 pf_normalize_tcp_init(struct mbuf *m, int off, struct pf_pdesc *pd,
1343     struct tcphdr *th, struct pf_state_peer *src, struct pf_state_peer *dst)
1344 {
1345 	u_int8_t hdr[60];
1346 	u_int8_t *opt;
1347 
1348 	KASSERT(src->scrub == NULL);
1349 
1350 	src->scrub = pool_get(&pf_state_scrub_pl, PR_NOWAIT);
1351 	if (src->scrub == NULL)
1352 		return (1);
1353 	bzero(src->scrub, sizeof(*src->scrub));
1354 
1355 	switch (pd->af) {
1356 #ifdef INET
1357 	case AF_INET: {
1358 		struct ip *h = mtod(m, struct ip *);
1359 		src->scrub->pfss_ttl = h->ip_ttl;
1360 		break;
1361 	}
1362 #endif /* INET */
1363 #ifdef INET6
1364 	case AF_INET6: {
1365 		struct ip6_hdr *h = mtod(m, struct ip6_hdr *);
1366 		src->scrub->pfss_ttl = h->ip6_hlim;
1367 		break;
1368 	}
1369 #endif /* INET6 */
1370 	}
1371 
1372 
1373 	/*
1374 	 * All normalizations below are only begun if we see the start of
1375 	 * the connections.  They must all set an enabled bit in pfss_flags
1376 	 */
1377 	if ((th->th_flags & TH_SYN) == 0)
1378 		return (0);
1379 
1380 
1381 	if (th->th_off > (sizeof(struct tcphdr) >> 2) && src->scrub &&
1382 	    pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) {
1383 		/* Diddle with TCP options */
1384 		int hlen;
1385 		opt = hdr + sizeof(struct tcphdr);
1386 		hlen = (th->th_off << 2) - sizeof(struct tcphdr);
1387 		while (hlen >= TCPOLEN_TIMESTAMP) {
1388 			switch (*opt) {
1389 			case TCPOPT_EOL:	/* FALLTHROUGH */
1390 			case TCPOPT_NOP:
1391 				opt++;
1392 				hlen--;
1393 				break;
1394 			case TCPOPT_TIMESTAMP:
1395 				if (opt[1] >= TCPOLEN_TIMESTAMP) {
1396 					src->scrub->pfss_flags |=
1397 					    PFSS_TIMESTAMP;
1398 					src->scrub->pfss_ts_mod = arc4random();
1399 				}
1400 				/* FALLTHROUGH */
1401 			default:
1402 				hlen -= opt[1];
1403 				opt += opt[1];
1404 				break;
1405 			}
1406 		}
1407 	}
1408 
1409 	return (0);
1410 }
1411 
1412 void
1413 pf_normalize_tcp_cleanup(struct pf_state *state)
1414 {
1415 	if (state->src.scrub)
1416 		pool_put(&pf_state_scrub_pl, state->src.scrub);
1417 	if (state->dst.scrub)
1418 		pool_put(&pf_state_scrub_pl, state->dst.scrub);
1419 
1420 	/* Someday... flush the TCP segment reassembly descriptors. */
1421 }
1422 
1423 int
1424 pf_normalize_tcp_stateful(struct mbuf *m, int off, struct pf_pdesc *pd,
1425     u_short *reason, struct tcphdr *th, struct pf_state_peer *src,
1426     struct pf_state_peer *dst, int *writeback)
1427 {
1428 	u_int8_t hdr[60];
1429 	u_int8_t *opt;
1430 	int copyback = 0;
1431 
1432 	KASSERT(src->scrub || dst->scrub);
1433 
1434 	/*
1435 	 * Enforce the minimum TTL seen for this connection.  Negate a common
1436 	 * technique to evade an intrusion detection system and confuse
1437 	 * firewall state code.
1438 	 */
1439 	switch (pd->af) {
1440 #ifdef INET
1441 	case AF_INET: {
1442 		if (src->scrub) {
1443 			struct ip *h = mtod(m, struct ip *);
1444 			if (h->ip_ttl > src->scrub->pfss_ttl)
1445 				src->scrub->pfss_ttl = h->ip_ttl;
1446 			h->ip_ttl = src->scrub->pfss_ttl;
1447 		}
1448 		break;
1449 	}
1450 #endif /* INET */
1451 #ifdef INET6
1452 	case AF_INET6: {
1453 		if (src->scrub) {
1454 			struct ip6_hdr *h = mtod(m, struct ip6_hdr *);
1455 			if (h->ip6_hlim > src->scrub->pfss_ttl)
1456 				src->scrub->pfss_ttl = h->ip6_hlim;
1457 			h->ip6_hlim = src->scrub->pfss_ttl;
1458 		}
1459 		break;
1460 	}
1461 #endif /* INET6 */
1462 	}
1463 
1464 	if (th->th_off > (sizeof(struct tcphdr) >> 2) &&
1465 	    ((src->scrub && (src->scrub->pfss_flags & PFSS_TIMESTAMP)) ||
1466 	    (dst->scrub && (dst->scrub->pfss_flags & PFSS_TIMESTAMP))) &&
1467 	    pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) {
1468 		/* Diddle with TCP options */
1469 		int hlen;
1470 		opt = hdr + sizeof(struct tcphdr);
1471 		hlen = (th->th_off << 2) - sizeof(struct tcphdr);
1472 		while (hlen >= TCPOLEN_TIMESTAMP) {
1473 			switch (*opt) {
1474 			case TCPOPT_EOL:	/* FALLTHROUGH */
1475 			case TCPOPT_NOP:
1476 				opt++;
1477 				hlen--;
1478 				break;
1479 			case TCPOPT_TIMESTAMP:
1480 				/* Modulate the timestamps.  Can be used for
1481 				 * NAT detection, OS uptime determination or
1482 				 * reboot detection.
1483 				 */
1484 				if (opt[1] >= TCPOLEN_TIMESTAMP) {
1485 					u_int32_t ts_value;
1486 					if (src->scrub &&
1487 					    (src->scrub->pfss_flags &
1488 					    PFSS_TIMESTAMP)) {
1489 						memcpy(&ts_value, &opt[2],
1490 						    sizeof(u_int32_t));
1491 						ts_value = htonl(ntohl(ts_value)
1492 						    + src->scrub->pfss_ts_mod);
1493 						pf_change_a(&opt[2],
1494 						    &th->th_sum, ts_value, 0);
1495 						copyback = 1;
1496 					}
1497 
1498 					/* Modulate TS reply iff valid (!0) */
1499 					memcpy(&ts_value, &opt[6],
1500 					    sizeof(u_int32_t));
1501 					if (ts_value && dst->scrub &&
1502 					    (dst->scrub->pfss_flags &
1503 					    PFSS_TIMESTAMP)) {
1504 						ts_value = htonl(ntohl(ts_value)
1505 						    - dst->scrub->pfss_ts_mod);
1506 						pf_change_a(&opt[6],
1507 						    &th->th_sum, ts_value, 0);
1508 						copyback = 1;
1509 					}
1510 				}
1511 				/* FALLTHROUGH */
1512 			default:
1513 				hlen -= opt[1];
1514 				opt += opt[1];
1515 				break;
1516 			}
1517 		}
1518 		if (copyback) {
1519 			/* Copyback the options, caller copys back header */
1520 			*writeback = 1;
1521 			m_copyback(m, off + sizeof(struct tcphdr),
1522 			    (th->th_off << 2) - sizeof(struct tcphdr), hdr +
1523 			    sizeof(struct tcphdr));
1524 		}
1525 	}
1526 
1527 
1528 	/* I have a dream....  TCP segment reassembly.... */
1529 	return (0);
1530 }
1531 int
1532 pf_normalize_tcpopt(struct pf_rule *r, struct mbuf *m, struct tcphdr *th,
1533     int off)
1534 {
1535 	u_int16_t	*mss;
1536 	int		 thoff;
1537 	int		 opt, cnt, optlen = 0;
1538 	int		 rewrite = 0;
1539 	u_char		*optp;
1540 
1541 	thoff = th->th_off << 2;
1542 	cnt = thoff - sizeof(struct tcphdr);
1543 	optp = mtod(m, caddr_t) + off + sizeof(struct tcphdr);
1544 
1545 	for (; cnt > 0; cnt -= optlen, optp += optlen) {
1546 		opt = optp[0];
1547 		if (opt == TCPOPT_EOL)
1548 			break;
1549 		if (opt == TCPOPT_NOP)
1550 			optlen = 1;
1551 		else {
1552 			if (cnt < 2)
1553 				break;
1554 			optlen = optp[1];
1555 			if (optlen < 2 || optlen > cnt)
1556 				break;
1557 		}
1558 		switch (opt) {
1559 		case TCPOPT_MAXSEG:
1560 			mss = (u_int16_t *)(optp + 2);
1561 			if ((ntohs(*mss)) > r->max_mss) {
1562 				th->th_sum = pf_cksum_fixup(th->th_sum,
1563 				    *mss, htons(r->max_mss));
1564 				*mss = htons(r->max_mss);
1565 				rewrite = 1;
1566 			}
1567 			break;
1568 		default:
1569 			break;
1570 		}
1571 	}
1572 
1573 	return (rewrite);
1574 }
1575