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_int8_t fn_direction; /* pf packet direction */
86 u_int32_t fn_fragments; /* number of entries in fn_tree */
87
88 RB_ENTRY(pf_frnode) fn_entry;
89 struct pf_frag_tree fn_tree; /* matching fragments, lookup by id */
90 };
91
92 struct pf_fragment {
93 uint32_t fr_id; /* fragment id for reassemble */
94
95 /* pointers to queue element */
96 struct pf_frent *fr_firstoff[PF_FRAG_ENTRY_POINTS];
97 /* count entries between pointers */
98 uint8_t fr_entries[PF_FRAG_ENTRY_POINTS];
99 RB_ENTRY(pf_fragment) fr_entry;
100 TAILQ_ENTRY(pf_fragment) frag_next;
101 uint32_t fr_timeout;
102 TAILQ_HEAD(pf_fragq, pf_frent) fr_queue;
103 uint16_t fr_maxlen; /* maximum length of single fragment */
104 u_int16_t fr_holes; /* number of holes in the queue */
105 struct pf_frnode *fr_node; /* ip src/dst/proto/af for fragments */
106 };
107
108 VNET_DEFINE_STATIC(struct mtx, pf_frag_mtx);
109 #define V_pf_frag_mtx VNET(pf_frag_mtx)
110 #define PF_FRAG_LOCK() mtx_lock(&V_pf_frag_mtx)
111 #define PF_FRAG_UNLOCK() mtx_unlock(&V_pf_frag_mtx)
112 #define PF_FRAG_ASSERT() mtx_assert(&V_pf_frag_mtx, MA_OWNED)
113
114 VNET_DEFINE(uma_zone_t, pf_state_scrub_z); /* XXX: shared with pfsync */
115
116 VNET_DEFINE_STATIC(uma_zone_t, pf_frent_z);
117 #define V_pf_frent_z VNET(pf_frent_z)
118 VNET_DEFINE_STATIC(uma_zone_t, pf_frnode_z);
119 #define V_pf_frnode_z VNET(pf_frnode_z)
120 VNET_DEFINE_STATIC(uma_zone_t, pf_frag_z);
121 #define V_pf_frag_z VNET(pf_frag_z)
122 VNET_DEFINE(uma_zone_t, pf_anchor_z);
123 VNET_DEFINE(uma_zone_t, pf_eth_anchor_z);
124
125 TAILQ_HEAD(pf_fragqueue, pf_fragment);
126 TAILQ_HEAD(pf_cachequeue, pf_fragment);
127 RB_HEAD(pf_frnode_tree, pf_frnode);
128 VNET_DEFINE_STATIC(struct pf_fragqueue, pf_fragqueue);
129 #define V_pf_fragqueue VNET(pf_fragqueue)
130 static __inline int pf_frnode_compare(struct pf_frnode *,
131 struct pf_frnode *);
132 VNET_DEFINE_STATIC(struct pf_frnode_tree, pf_frnode_tree);
133 #define V_pf_frnode_tree VNET(pf_frnode_tree)
134 RB_PROTOTYPE(pf_frnode_tree, pf_frnode, fn_entry, pf_frnode_compare);
135 RB_GENERATE(pf_frnode_tree, pf_frnode, fn_entry, pf_frnode_compare);
136
137 static int pf_frag_compare(struct pf_fragment *,
138 struct pf_fragment *);
139 static RB_PROTOTYPE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
140 static RB_GENERATE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
141
142 static void pf_flush_fragments(void);
143 static void pf_free_fragment(struct pf_fragment *);
144
145 static struct pf_frent *pf_create_fragment(u_short *);
146 static int pf_frent_holes(struct pf_frent *frent);
147 static struct pf_fragment *pf_find_fragment(struct pf_frnode *, u_int32_t);
148 static inline int pf_frent_index(struct pf_frent *);
149 static int pf_frent_insert(struct pf_fragment *,
150 struct pf_frent *, struct pf_frent *);
151 void pf_frent_remove(struct pf_fragment *,
152 struct pf_frent *);
153 struct pf_frent *pf_frent_previous(struct pf_fragment *,
154 struct pf_frent *);
155 static struct pf_fragment *pf_fillup_fragment(struct pf_frnode *, u_int32_t,
156 struct pf_frent *, u_short *);
157 static struct mbuf *pf_join_fragment(struct pf_fragment *);
158 #ifdef INET
159 static int pf_reassemble(struct mbuf **, uint8_t, u_short *);
160 #endif /* INET */
161 #ifdef INET6
162 static int pf_reassemble6(struct mbuf **,
163 struct ip6_frag *, uint16_t, uint16_t, uint8_t, u_short *);
164 #endif /* INET6 */
165
166 #ifdef INET
167 static void
pf_ip2key(struct ip * ip,struct pf_frnode * key,uint8_t dir)168 pf_ip2key(struct ip *ip, struct pf_frnode *key, uint8_t dir)
169 {
170
171 key->fn_src.v4 = ip->ip_src;
172 key->fn_dst.v4 = ip->ip_dst;
173 key->fn_af = AF_INET;
174 key->fn_proto = ip->ip_p;
175 key->fn_direction = dir;
176 }
177 #endif /* INET */
178
179 void
pf_normalize_init(void)180 pf_normalize_init(void)
181 {
182
183 V_pf_frag_z = uma_zcreate("pf frags", sizeof(struct pf_fragment),
184 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
185 V_pf_frnode_z = uma_zcreate("pf fragment node",
186 sizeof(struct pf_frnode), NULL, NULL, NULL, NULL,
187 UMA_ALIGN_PTR, 0);
188 V_pf_frent_z = uma_zcreate("pf frag entries", sizeof(struct pf_frent),
189 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
190 V_pf_state_scrub_z = uma_zcreate("pf state scrubs",
191 sizeof(struct pf_state_scrub), NULL, NULL, NULL, NULL,
192 UMA_ALIGN_PTR, 0);
193
194 mtx_init(&V_pf_frag_mtx, "pf fragments", NULL, MTX_DEF);
195
196 V_pf_limits[PF_LIMIT_FRAGS].zone = V_pf_frent_z;
197 V_pf_limits[PF_LIMIT_FRAGS].limit = PFFRAG_FRENT_HIWAT;
198 uma_zone_set_max(V_pf_frent_z, PFFRAG_FRENT_HIWAT);
199 uma_zone_set_warning(V_pf_frent_z, "PF frag entries limit reached");
200
201 TAILQ_INIT(&V_pf_fragqueue);
202 }
203
204 void
pf_normalize_cleanup(void)205 pf_normalize_cleanup(void)
206 {
207
208 uma_zdestroy(V_pf_state_scrub_z);
209 uma_zdestroy(V_pf_frent_z);
210 uma_zdestroy(V_pf_frnode_z);
211 uma_zdestroy(V_pf_frag_z);
212
213 mtx_destroy(&V_pf_frag_mtx);
214 }
215
216 uint64_t
pf_normalize_get_frag_count(void)217 pf_normalize_get_frag_count(void)
218 {
219 return (uma_zone_get_cur(V_pf_frent_z));
220 }
221
222 static int
pf_frnode_compare(struct pf_frnode * a,struct pf_frnode * b)223 pf_frnode_compare(struct pf_frnode *a, struct pf_frnode *b)
224 {
225 int diff;
226
227 if ((diff = a->fn_proto - b->fn_proto) != 0)
228 return (diff);
229 if ((diff = a->fn_af - b->fn_af) != 0)
230 return (diff);
231 if ((diff = a->fn_direction - b->fn_direction) != 0)
232 return (diff);
233 if ((diff = pf_addr_cmp(&a->fn_src, &b->fn_src, a->fn_af)) != 0)
234 return (diff);
235 if ((diff = pf_addr_cmp(&a->fn_dst, &b->fn_dst, a->fn_af)) != 0)
236 return (diff);
237 return (0);
238 }
239
240 static __inline int
pf_frag_compare(struct pf_fragment * a,struct pf_fragment * b)241 pf_frag_compare(struct pf_fragment *a, struct pf_fragment *b)
242 {
243 if (a->fr_id > b->fr_id)
244 return (1);
245 if (a->fr_id < b->fr_id)
246 return (-1);
247
248 return (0);
249 }
250
251 void
pf_purge_expired_fragments(void)252 pf_purge_expired_fragments(void)
253 {
254 u_int32_t expire = time_uptime -
255 V_pf_default_rule.timeout[PFTM_FRAG];
256
257 pf_purge_fragments(expire);
258 }
259
260 void
pf_purge_fragments(uint32_t expire)261 pf_purge_fragments(uint32_t expire)
262 {
263 struct pf_fragment *frag;
264
265 PF_FRAG_LOCK();
266 while ((frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue)) != NULL) {
267 if (frag->fr_timeout > expire)
268 break;
269
270 DPFPRINTF(PF_DEBUG_MISC, "expiring %d(%p)",
271 frag->fr_id, frag);
272 pf_free_fragment(frag);
273 }
274
275 PF_FRAG_UNLOCK();
276 }
277
278 /*
279 * Try to flush old fragments to make space for new ones
280 */
281 static void
pf_flush_fragments(void)282 pf_flush_fragments(void)
283 {
284 struct pf_fragment *frag;
285 int goal;
286
287 PF_FRAG_ASSERT();
288
289 goal = uma_zone_get_cur(V_pf_frent_z) * 9 / 10;
290 DPFPRINTF(PF_DEBUG_MISC, "trying to free %d frag entriess", goal);
291 while (goal < uma_zone_get_cur(V_pf_frent_z)) {
292 frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue);
293 if (frag)
294 pf_free_fragment(frag);
295 else
296 break;
297 }
298 }
299
300 /*
301 * Remove a fragment from the fragment queue, free its fragment entries,
302 * and free the fragment itself.
303 */
304 static void
pf_free_fragment(struct pf_fragment * frag)305 pf_free_fragment(struct pf_fragment *frag)
306 {
307 struct pf_frent *frent;
308 struct pf_frnode *frnode;
309
310 PF_FRAG_ASSERT();
311
312 frnode = frag->fr_node;
313 RB_REMOVE(pf_frag_tree, &frnode->fn_tree, frag);
314 MPASS(frnode->fn_fragments >= 1);
315 frnode->fn_fragments--;
316 if (frnode->fn_fragments == 0) {
317 MPASS(RB_EMPTY(&frnode->fn_tree));
318 RB_REMOVE(pf_frnode_tree, &V_pf_frnode_tree, frnode);
319 uma_zfree(V_pf_frnode_z, frnode);
320 }
321
322 TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next);
323
324 /* Free all fragment entries */
325 while ((frent = TAILQ_FIRST(&frag->fr_queue)) != NULL) {
326 TAILQ_REMOVE(&frag->fr_queue, frent, fr_next);
327 counter_u64_add(V_pf_status.ncounters[NCNT_FRAG_REMOVALS], 1);
328
329 m_freem(frent->fe_m);
330 uma_zfree(V_pf_frent_z, frent);
331 }
332
333 uma_zfree(V_pf_frag_z, frag);
334 }
335
336 static struct pf_fragment *
pf_find_fragment(struct pf_frnode * key,uint32_t id)337 pf_find_fragment(struct pf_frnode *key, uint32_t id)
338 {
339 struct pf_fragment *frag, idkey;
340 struct pf_frnode *frnode;
341
342 PF_FRAG_ASSERT();
343
344 frnode = RB_FIND(pf_frnode_tree, &V_pf_frnode_tree, key);
345 counter_u64_add(V_pf_status.ncounters[NCNT_FRAG_SEARCH], 1);
346 if (frnode == NULL)
347 return (NULL);
348 MPASS(frnode->fn_fragments >= 1);
349 idkey.fr_id = id;
350 frag = RB_FIND(pf_frag_tree, &frnode->fn_tree, &idkey);
351 if (frag == NULL)
352 return (NULL);
353 TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next);
354 TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next);
355
356 return (frag);
357 }
358
359 static struct pf_frent *
pf_create_fragment(u_short * reason)360 pf_create_fragment(u_short *reason)
361 {
362 struct pf_frent *frent;
363
364 PF_FRAG_ASSERT();
365
366 frent = uma_zalloc(V_pf_frent_z, M_NOWAIT);
367 if (frent == NULL) {
368 pf_flush_fragments();
369 frent = uma_zalloc(V_pf_frent_z, M_NOWAIT);
370 if (frent == NULL) {
371 REASON_SET(reason, PFRES_MEMORY);
372 return (NULL);
373 }
374 }
375
376 return (frent);
377 }
378
379 /*
380 * Calculate the additional holes that were created in the fragment
381 * queue by inserting this fragment. A fragment in the middle
382 * creates one more hole by splitting. For each connected side,
383 * it loses one hole.
384 * Fragment entry must be in the queue when calling this function.
385 */
386 static int
pf_frent_holes(struct pf_frent * frent)387 pf_frent_holes(struct pf_frent *frent)
388 {
389 struct pf_frent *prev = TAILQ_PREV(frent, pf_fragq, fr_next);
390 struct pf_frent *next = TAILQ_NEXT(frent, fr_next);
391 int holes = 1;
392
393 if (prev == NULL) {
394 if (frent->fe_off == 0)
395 holes--;
396 } else {
397 KASSERT(frent->fe_off != 0, ("frent->fe_off != 0"));
398 if (frent->fe_off == prev->fe_off + prev->fe_len)
399 holes--;
400 }
401 if (next == NULL) {
402 if (!frent->fe_mff)
403 holes--;
404 } else {
405 KASSERT(frent->fe_mff, ("frent->fe_mff"));
406 if (next->fe_off == frent->fe_off + frent->fe_len)
407 holes--;
408 }
409 return holes;
410 }
411
412 static inline int
pf_frent_index(struct pf_frent * frent)413 pf_frent_index(struct pf_frent *frent)
414 {
415 /*
416 * We have an array of 16 entry points to the queue. A full size
417 * 65535 octet IP packet can have 8192 fragments. So the queue
418 * traversal length is at most 512 and at most 16 entry points are
419 * checked. We need 128 additional bytes on a 64 bit architecture.
420 */
421 CTASSERT(((u_int16_t)0xffff &~ 7) / (0x10000 / PF_FRAG_ENTRY_POINTS) ==
422 16 - 1);
423 CTASSERT(((u_int16_t)0xffff >> 3) / PF_FRAG_ENTRY_POINTS == 512 - 1);
424
425 return frent->fe_off / (0x10000 / PF_FRAG_ENTRY_POINTS);
426 }
427
428 static int
pf_frent_insert(struct pf_fragment * frag,struct pf_frent * frent,struct pf_frent * prev)429 pf_frent_insert(struct pf_fragment *frag, struct pf_frent *frent,
430 struct pf_frent *prev)
431 {
432 int index;
433
434 CTASSERT(PF_FRAG_ENTRY_LIMIT <= 0xff);
435
436 /*
437 * A packet has at most 65536 octets. With 16 entry points, each one
438 * spawns 4096 octets. We limit these to 64 fragments each, which
439 * means on average every fragment must have at least 64 octets.
440 */
441 index = pf_frent_index(frent);
442 if (frag->fr_entries[index] >= PF_FRAG_ENTRY_LIMIT)
443 return ENOBUFS;
444 frag->fr_entries[index]++;
445
446 if (prev == NULL) {
447 TAILQ_INSERT_HEAD(&frag->fr_queue, frent, fr_next);
448 } else {
449 KASSERT(prev->fe_off + prev->fe_len <= frent->fe_off,
450 ("overlapping fragment"));
451 TAILQ_INSERT_AFTER(&frag->fr_queue, prev, frent, fr_next);
452 }
453 counter_u64_add(V_pf_status.ncounters[NCNT_FRAG_INSERT], 1);
454
455 if (frag->fr_firstoff[index] == NULL) {
456 KASSERT(prev == NULL || pf_frent_index(prev) < index,
457 ("prev == NULL || pf_frent_index(pref) < index"));
458 frag->fr_firstoff[index] = frent;
459 } else {
460 if (frent->fe_off < frag->fr_firstoff[index]->fe_off) {
461 KASSERT(prev == NULL || pf_frent_index(prev) < index,
462 ("prev == NULL || pf_frent_index(pref) < index"));
463 frag->fr_firstoff[index] = frent;
464 } else {
465 KASSERT(prev != NULL, ("prev != NULL"));
466 KASSERT(pf_frent_index(prev) == index,
467 ("pf_frent_index(prev) == index"));
468 }
469 }
470
471 frag->fr_holes += pf_frent_holes(frent);
472
473 return 0;
474 }
475
476 void
pf_frent_remove(struct pf_fragment * frag,struct pf_frent * frent)477 pf_frent_remove(struct pf_fragment *frag, struct pf_frent *frent)
478 {
479 #ifdef INVARIANTS
480 struct pf_frent *prev = TAILQ_PREV(frent, pf_fragq, fr_next);
481 #endif /* INVARIANTS */
482 struct pf_frent *next = TAILQ_NEXT(frent, fr_next);
483 int index;
484
485 frag->fr_holes -= pf_frent_holes(frent);
486
487 index = pf_frent_index(frent);
488 KASSERT(frag->fr_firstoff[index] != NULL, ("frent not found"));
489 if (frag->fr_firstoff[index]->fe_off == frent->fe_off) {
490 if (next == NULL) {
491 frag->fr_firstoff[index] = NULL;
492 } else {
493 KASSERT(frent->fe_off + frent->fe_len <= next->fe_off,
494 ("overlapping fragment"));
495 if (pf_frent_index(next) == index) {
496 frag->fr_firstoff[index] = next;
497 } else {
498 frag->fr_firstoff[index] = NULL;
499 }
500 }
501 } else {
502 KASSERT(frag->fr_firstoff[index]->fe_off < frent->fe_off,
503 ("frag->fr_firstoff[index]->fe_off < frent->fe_off"));
504 KASSERT(prev != NULL, ("prev != NULL"));
505 KASSERT(prev->fe_off + prev->fe_len <= frent->fe_off,
506 ("overlapping fragment"));
507 KASSERT(pf_frent_index(prev) == index,
508 ("pf_frent_index(prev) == index"));
509 }
510
511 TAILQ_REMOVE(&frag->fr_queue, frent, fr_next);
512 counter_u64_add(V_pf_status.ncounters[NCNT_FRAG_REMOVALS], 1);
513
514 KASSERT(frag->fr_entries[index] > 0, ("No fragments remaining"));
515 frag->fr_entries[index]--;
516 }
517
518 struct pf_frent *
pf_frent_previous(struct pf_fragment * frag,struct pf_frent * frent)519 pf_frent_previous(struct pf_fragment *frag, struct pf_frent *frent)
520 {
521 struct pf_frent *prev, *next;
522 int index;
523
524 /*
525 * If there are no fragments after frag, take the final one. Assume
526 * that the global queue is not empty.
527 */
528 prev = TAILQ_LAST(&frag->fr_queue, pf_fragq);
529 KASSERT(prev != NULL, ("prev != NULL"));
530 if (prev->fe_off <= frent->fe_off)
531 return prev;
532 /*
533 * We want to find a fragment entry that is before frag, but still
534 * close to it. Find the first fragment entry that is in the same
535 * entry point or in the first entry point after that. As we have
536 * already checked that there are entries behind frag, this will
537 * succeed.
538 */
539 for (index = pf_frent_index(frent); index < PF_FRAG_ENTRY_POINTS;
540 index++) {
541 prev = frag->fr_firstoff[index];
542 if (prev != NULL)
543 break;
544 }
545 KASSERT(prev != NULL, ("prev != NULL"));
546 /*
547 * In prev we may have a fragment from the same entry point that is
548 * before frent, or one that is just one position behind frent.
549 * In the latter case, we go back one step and have the predecessor.
550 * There may be none if the new fragment will be the first one.
551 */
552 if (prev->fe_off > frent->fe_off) {
553 prev = TAILQ_PREV(prev, pf_fragq, fr_next);
554 if (prev == NULL)
555 return NULL;
556 KASSERT(prev->fe_off <= frent->fe_off,
557 ("prev->fe_off <= frent->fe_off"));
558 return prev;
559 }
560 /*
561 * In prev is the first fragment of the entry point. The offset
562 * of frag is behind it. Find the closest previous fragment.
563 */
564 for (next = TAILQ_NEXT(prev, fr_next); next != NULL;
565 next = TAILQ_NEXT(next, fr_next)) {
566 if (next->fe_off > frent->fe_off)
567 break;
568 prev = next;
569 }
570 return prev;
571 }
572
573 static struct pf_fragment *
pf_fillup_fragment(struct pf_frnode * key,uint32_t id,struct pf_frent * frent,u_short * reason)574 pf_fillup_fragment(struct pf_frnode *key, uint32_t id,
575 struct pf_frent *frent, u_short *reason)
576 {
577 struct pf_frent *after, *next, *prev;
578 struct pf_fragment *frag;
579 struct pf_frnode *frnode;
580 uint16_t total;
581
582 PF_FRAG_ASSERT();
583
584 /* No empty fragments. */
585 if (frent->fe_len == 0) {
586 DPFPRINTF(PF_DEBUG_MISC, "bad fragment: len 0");
587 goto bad_fragment;
588 }
589
590 /* All fragments are 8 byte aligned. */
591 if (frent->fe_mff && (frent->fe_len & 0x7)) {
592 DPFPRINTF(PF_DEBUG_MISC, "bad fragment: mff and len %d",
593 frent->fe_len);
594 goto bad_fragment;
595 }
596
597 /* Respect maximum length, IP_MAXPACKET == IPV6_MAXPACKET. */
598 if (frent->fe_off + frent->fe_len > IP_MAXPACKET) {
599 DPFPRINTF(PF_DEBUG_MISC, "bad fragment: max packet %d",
600 frent->fe_off + frent->fe_len);
601 goto bad_fragment;
602 }
603
604 if (key->fn_af == AF_INET)
605 DPFPRINTF(PF_DEBUG_MISC, "reass frag %d @ %d-%d\n",
606 id, frent->fe_off, frent->fe_off + frent->fe_len);
607 else
608 DPFPRINTF(PF_DEBUG_MISC, "reass frag %#08x @ %d-%d",
609 id, frent->fe_off, frent->fe_off + frent->fe_len);
610
611 /* Fully buffer all of the fragments in this fragment queue. */
612 frag = pf_find_fragment(key, id);
613
614 /* Create a new reassembly queue for this packet. */
615 if (frag == NULL) {
616 frag = uma_zalloc(V_pf_frag_z, M_NOWAIT);
617 if (frag == NULL) {
618 pf_flush_fragments();
619 frag = uma_zalloc(V_pf_frag_z, M_NOWAIT);
620 if (frag == NULL) {
621 REASON_SET(reason, PFRES_MEMORY);
622 goto drop_fragment;
623 }
624 }
625
626 frnode = RB_FIND(pf_frnode_tree, &V_pf_frnode_tree, key);
627 if (frnode == NULL) {
628 frnode = uma_zalloc(V_pf_frnode_z, M_NOWAIT);
629 if (frnode == NULL) {
630 pf_flush_fragments();
631 frnode = uma_zalloc(V_pf_frnode_z, M_NOWAIT);
632 if (frnode == NULL) {
633 REASON_SET(reason, PFRES_MEMORY);
634 uma_zfree(V_pf_frag_z, frag);
635 goto drop_fragment;
636 }
637 }
638 *frnode = *key;
639 RB_INIT(&frnode->fn_tree);
640 frnode->fn_fragments = 0;
641 }
642 memset(frag->fr_firstoff, 0, sizeof(frag->fr_firstoff));
643 memset(frag->fr_entries, 0, sizeof(frag->fr_entries));
644 frag->fr_timeout = time_uptime;
645 TAILQ_INIT(&frag->fr_queue);
646 frag->fr_maxlen = frent->fe_len;
647 frag->fr_holes = 1;
648
649 frag->fr_id = id;
650 frag->fr_node = frnode;
651 /* RB_INSERT cannot fail as pf_find_fragment() found nothing */
652 RB_INSERT(pf_frag_tree, &frnode->fn_tree, frag);
653 frnode->fn_fragments++;
654 if (frnode->fn_fragments == 1)
655 RB_INSERT(pf_frnode_tree, &V_pf_frnode_tree, frnode);
656
657 TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next);
658
659 /* We do not have a previous fragment, cannot fail. */
660 pf_frent_insert(frag, frent, NULL);
661
662 return (frag);
663 }
664
665 KASSERT(!TAILQ_EMPTY(&frag->fr_queue), ("!TAILQ_EMPTY()->fr_queue"));
666 MPASS(frag->fr_node);
667
668 /* Remember maximum fragment len for refragmentation. */
669 if (frent->fe_len > frag->fr_maxlen)
670 frag->fr_maxlen = frent->fe_len;
671
672 /* Maximum data we have seen already. */
673 total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
674 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
675
676 /* Non terminal fragments must have more fragments flag. */
677 if (frent->fe_off + frent->fe_len < total && !frent->fe_mff)
678 goto free_ipv6_fragment;
679
680 /* Check if we saw the last fragment already. */
681 if (!TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_mff) {
682 if (frent->fe_off + frent->fe_len > total ||
683 (frent->fe_off + frent->fe_len == total && frent->fe_mff))
684 goto free_ipv6_fragment;
685 } else {
686 if (frent->fe_off + frent->fe_len == total && !frent->fe_mff)
687 goto free_ipv6_fragment;
688 }
689
690 /* Find neighbors for newly inserted fragment */
691 prev = pf_frent_previous(frag, frent);
692 if (prev == NULL) {
693 after = TAILQ_FIRST(&frag->fr_queue);
694 KASSERT(after != NULL, ("after != NULL"));
695 } else {
696 after = TAILQ_NEXT(prev, fr_next);
697 }
698
699 if (prev != NULL && prev->fe_off + prev->fe_len > frent->fe_off) {
700 uint16_t precut;
701
702 if (frag->fr_node->fn_af == AF_INET6)
703 goto free_fragment;
704
705 precut = prev->fe_off + prev->fe_len - frent->fe_off;
706 if (precut >= frent->fe_len) {
707 DPFPRINTF(PF_DEBUG_MISC, "new frag overlapped");
708 goto drop_fragment;
709 }
710 DPFPRINTF(PF_DEBUG_MISC, "frag head overlap %d", precut);
711 m_adj(frent->fe_m, precut);
712 frent->fe_off += precut;
713 frent->fe_len -= precut;
714 }
715
716 for (; after != NULL && frent->fe_off + frent->fe_len > after->fe_off;
717 after = next) {
718 uint16_t aftercut;
719
720 aftercut = frent->fe_off + frent->fe_len - after->fe_off;
721 if (aftercut < after->fe_len) {
722 DPFPRINTF(PF_DEBUG_MISC, "frag tail overlap %d",
723 aftercut);
724 m_adj(after->fe_m, aftercut);
725 /* Fragment may switch queue as fe_off changes */
726 pf_frent_remove(frag, after);
727 after->fe_off += aftercut;
728 after->fe_len -= aftercut;
729 /* Insert into correct queue */
730 if (pf_frent_insert(frag, after, prev)) {
731 DPFPRINTF(PF_DEBUG_MISC,
732 "fragment requeue limit exceeded");
733 m_freem(after->fe_m);
734 uma_zfree(V_pf_frent_z, after);
735 /* There is not way to recover */
736 goto free_fragment;
737 }
738 break;
739 }
740
741 /* This fragment is completely overlapped, lose it. */
742 DPFPRINTF(PF_DEBUG_MISC, "old frag overlapped");
743 next = TAILQ_NEXT(after, fr_next);
744 pf_frent_remove(frag, after);
745 m_freem(after->fe_m);
746 uma_zfree(V_pf_frent_z, after);
747 }
748
749 /* If part of the queue gets too long, there is not way to recover. */
750 if (pf_frent_insert(frag, frent, prev)) {
751 DPFPRINTF(PF_DEBUG_MISC, "fragment queue limit exceeded");
752 goto bad_fragment;
753 }
754
755 return (frag);
756
757 free_ipv6_fragment:
758 if (frag->fr_node->fn_af == AF_INET)
759 goto bad_fragment;
760 free_fragment:
761 /*
762 * RFC 5722, Errata 3089: When reassembling an IPv6 datagram, if one
763 * or more its constituent fragments is determined to be an overlapping
764 * fragment, the entire datagram (and any constituent fragments) MUST
765 * be silently discarded.
766 */
767 DPFPRINTF(PF_DEBUG_MISC, "flush overlapping fragments");
768 pf_free_fragment(frag);
769
770 bad_fragment:
771 REASON_SET(reason, PFRES_FRAG);
772 drop_fragment:
773 uma_zfree(V_pf_frent_z, frent);
774 return (NULL);
775 }
776
777 static struct mbuf *
pf_join_fragment(struct pf_fragment * frag)778 pf_join_fragment(struct pf_fragment *frag)
779 {
780 struct mbuf *m, *m2;
781 struct pf_frent *frent;
782
783 frent = TAILQ_FIRST(&frag->fr_queue);
784 TAILQ_REMOVE(&frag->fr_queue, frent, fr_next);
785 counter_u64_add(V_pf_status.ncounters[NCNT_FRAG_REMOVALS], 1);
786
787 m = frent->fe_m;
788 if ((frent->fe_hdrlen + frent->fe_len) < m->m_pkthdr.len)
789 m_adj(m, (frent->fe_hdrlen + frent->fe_len) - m->m_pkthdr.len);
790 uma_zfree(V_pf_frent_z, frent);
791 while ((frent = TAILQ_FIRST(&frag->fr_queue)) != NULL) {
792 TAILQ_REMOVE(&frag->fr_queue, frent, fr_next);
793 counter_u64_add(V_pf_status.ncounters[NCNT_FRAG_REMOVALS], 1);
794
795 m2 = frent->fe_m;
796 /* Strip off ip header. */
797 m_adj(m2, frent->fe_hdrlen);
798 /* Strip off any trailing bytes. */
799 if (frent->fe_len < m2->m_pkthdr.len)
800 m_adj(m2, frent->fe_len - m2->m_pkthdr.len);
801
802 uma_zfree(V_pf_frent_z, frent);
803 m_cat(m, m2);
804 }
805
806 /* Remove from fragment queue. */
807 pf_free_fragment(frag);
808
809 return (m);
810 }
811
812 #ifdef INET
813 static int
pf_reassemble(struct mbuf ** m0,uint8_t dir,u_short * reason)814 pf_reassemble(struct mbuf **m0, uint8_t dir, u_short *reason)
815 {
816 struct mbuf *m = *m0;
817 struct ip *ip = mtod(m, struct ip *);
818 struct pf_frent *frent;
819 struct pf_fragment *frag;
820 struct m_tag *mtag;
821 struct pf_fragment_tag *ftag;
822 struct pf_frnode key;
823 uint16_t total, hdrlen;
824 uint32_t frag_id;
825 uint16_t maxlen;
826
827 /* Get an entry for the fragment queue */
828 if ((frent = pf_create_fragment(reason)) == NULL)
829 return (PF_DROP);
830
831 frent->fe_m = m;
832 frent->fe_hdrlen = ip->ip_hl << 2;
833 frent->fe_extoff = 0;
834 frent->fe_len = ntohs(ip->ip_len) - (ip->ip_hl << 2);
835 frent->fe_off = (ntohs(ip->ip_off) & IP_OFFMASK) << 3;
836 frent->fe_mff = ntohs(ip->ip_off) & IP_MF;
837
838 pf_ip2key(ip, &key, dir);
839
840 if ((frag = pf_fillup_fragment(&key, ip->ip_id, frent, reason)) == NULL)
841 return (PF_DROP);
842
843 /* The mbuf is part of the fragment entry, no direct free or access */
844 m = *m0 = NULL;
845
846 if (frag->fr_holes) {
847 DPFPRINTF(PF_DEBUG_MISC, "frag %d, holes %d",
848 frag->fr_id, frag->fr_holes);
849 return (PF_PASS); /* drop because *m0 is NULL, no error */
850 }
851
852 /* We have all the data */
853 frent = TAILQ_FIRST(&frag->fr_queue);
854 KASSERT(frent != NULL, ("frent != NULL"));
855 total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
856 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
857 hdrlen = frent->fe_hdrlen;
858
859 maxlen = frag->fr_maxlen;
860 frag_id = frag->fr_id;
861 m = *m0 = pf_join_fragment(frag);
862 frag = NULL;
863
864 if (m->m_flags & M_PKTHDR) {
865 int plen = 0;
866 for (m = *m0; m; m = m->m_next)
867 plen += m->m_len;
868 m = *m0;
869 m->m_pkthdr.len = plen;
870 }
871
872 if ((mtag = m_tag_get(PACKET_TAG_PF_REASSEMBLED,
873 sizeof(struct pf_fragment_tag), M_NOWAIT)) == NULL) {
874 REASON_SET(reason, PFRES_SHORT);
875 /* PF_DROP requires a valid mbuf *m0 in pf_test() */
876 return (PF_DROP);
877 }
878 ftag = (struct pf_fragment_tag *)(mtag + 1);
879 ftag->ft_hdrlen = hdrlen;
880 ftag->ft_extoff = 0;
881 ftag->ft_maxlen = maxlen;
882 ftag->ft_id = frag_id;
883 m_tag_prepend(m, mtag);
884
885 ip = mtod(m, struct ip *);
886 ip->ip_sum = pf_cksum_fixup(ip->ip_sum, ip->ip_len,
887 htons(hdrlen + total), 0);
888 ip->ip_len = htons(hdrlen + total);
889 ip->ip_sum = pf_cksum_fixup(ip->ip_sum, ip->ip_off,
890 ip->ip_off & ~(IP_MF|IP_OFFMASK), 0);
891 ip->ip_off &= ~(IP_MF|IP_OFFMASK);
892
893 if (hdrlen + total > IP_MAXPACKET) {
894 DPFPRINTF(PF_DEBUG_MISC, "drop: too big: %d", total);
895 ip->ip_len = 0;
896 REASON_SET(reason, PFRES_SHORT);
897 /* PF_DROP requires a valid mbuf *m0 in pf_test() */
898 return (PF_DROP);
899 }
900
901 DPFPRINTF(PF_DEBUG_MISC, "complete: %p(%d)", m, ntohs(ip->ip_len));
902 return (PF_PASS);
903 }
904 #endif /* INET */
905
906 #ifdef INET6
907 static int
pf_reassemble6(struct mbuf ** m0,struct ip6_frag * fraghdr,uint16_t hdrlen,uint16_t extoff,uint8_t dir,u_short * reason)908 pf_reassemble6(struct mbuf **m0, struct ip6_frag *fraghdr,
909 uint16_t hdrlen, uint16_t extoff, uint8_t dir, u_short *reason)
910 {
911 struct mbuf *m = *m0;
912 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
913 struct pf_frent *frent;
914 struct pf_fragment *frag;
915 struct pf_frnode key;
916 struct m_tag *mtag;
917 struct pf_fragment_tag *ftag;
918 int off;
919 uint32_t frag_id;
920 uint16_t total, maxlen;
921 uint8_t proto;
922
923 PF_FRAG_LOCK();
924
925 /* Get an entry for the fragment queue. */
926 if ((frent = pf_create_fragment(reason)) == NULL) {
927 PF_FRAG_UNLOCK();
928 return (PF_DROP);
929 }
930
931 frent->fe_m = m;
932 frent->fe_hdrlen = hdrlen;
933 frent->fe_extoff = extoff;
934 frent->fe_len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - hdrlen;
935 frent->fe_off = ntohs(fraghdr->ip6f_offlg & IP6F_OFF_MASK);
936 frent->fe_mff = fraghdr->ip6f_offlg & IP6F_MORE_FRAG;
937
938 key.fn_src.v6 = ip6->ip6_src;
939 key.fn_dst.v6 = ip6->ip6_dst;
940 key.fn_af = AF_INET6;
941 /* Only the first fragment's protocol is relevant. */
942 key.fn_proto = 0;
943 key.fn_direction = dir;
944
945 if ((frag = pf_fillup_fragment(&key, fraghdr->ip6f_ident, frent, reason)) == NULL) {
946 PF_FRAG_UNLOCK();
947 return (PF_DROP);
948 }
949
950 /* The mbuf is part of the fragment entry, no direct free or access. */
951 m = *m0 = NULL;
952
953 if (frag->fr_holes) {
954 DPFPRINTF(PF_DEBUG_MISC, "frag %d, holes %d", frag->fr_id,
955 frag->fr_holes);
956 PF_FRAG_UNLOCK();
957 return (PF_PASS); /* Drop because *m0 is NULL, no error. */
958 }
959
960 /* We have all the data. */
961 frent = TAILQ_FIRST(&frag->fr_queue);
962 KASSERT(frent != NULL, ("frent != NULL"));
963 extoff = frent->fe_extoff;
964 maxlen = frag->fr_maxlen;
965 frag_id = frag->fr_id;
966 total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
967 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
968 hdrlen = frent->fe_hdrlen - sizeof(struct ip6_frag);
969
970 m = *m0 = pf_join_fragment(frag);
971 frag = NULL;
972
973 PF_FRAG_UNLOCK();
974
975 /* Take protocol from first fragment header. */
976 m = m_getptr(m, hdrlen + offsetof(struct ip6_frag, ip6f_nxt), &off);
977 KASSERT(m, ("%s: short mbuf chain", __func__));
978 proto = *(mtod(m, uint8_t *) + off);
979 m = *m0;
980
981 /* Delete frag6 header */
982 if (ip6_deletefraghdr(m, hdrlen, M_NOWAIT) != 0)
983 goto fail;
984
985 if (m->m_flags & M_PKTHDR) {
986 int plen = 0;
987 for (m = *m0; m; m = m->m_next)
988 plen += m->m_len;
989 m = *m0;
990 m->m_pkthdr.len = plen;
991 }
992
993 if ((mtag = m_tag_get(PACKET_TAG_PF_REASSEMBLED,
994 sizeof(struct pf_fragment_tag), M_NOWAIT)) == NULL)
995 goto fail;
996 ftag = (struct pf_fragment_tag *)(mtag + 1);
997 ftag->ft_hdrlen = hdrlen;
998 ftag->ft_extoff = extoff;
999 ftag->ft_maxlen = maxlen;
1000 ftag->ft_id = frag_id;
1001 m_tag_prepend(m, mtag);
1002
1003 ip6 = mtod(m, struct ip6_hdr *);
1004 ip6->ip6_plen = htons(hdrlen - sizeof(struct ip6_hdr) + total);
1005 if (extoff) {
1006 /* Write protocol into next field of last extension header. */
1007 m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt),
1008 &off);
1009 KASSERT(m, ("%s: short mbuf chain", __func__));
1010 *(mtod(m, char *) + off) = proto;
1011 m = *m0;
1012 } else
1013 ip6->ip6_nxt = proto;
1014
1015 if (hdrlen - sizeof(struct ip6_hdr) + total > IPV6_MAXPACKET) {
1016 DPFPRINTF(PF_DEBUG_MISC, "drop: too big: %d", total);
1017 ip6->ip6_plen = 0;
1018 REASON_SET(reason, PFRES_SHORT);
1019 /* PF_DROP requires a valid mbuf *m0 in pf_test6(). */
1020 return (PF_DROP);
1021 }
1022
1023 DPFPRINTF(PF_DEBUG_MISC, "complete: %p(%d)", m,
1024 ntohs(ip6->ip6_plen));
1025 return (PF_PASS);
1026
1027 fail:
1028 REASON_SET(reason, PFRES_MEMORY);
1029 /* PF_DROP requires a valid mbuf *m0 in pf_test6(), will free later. */
1030 return (PF_DROP);
1031 }
1032 #endif /* INET6 */
1033
1034 #ifdef INET6
1035 int
pf_max_frag_size(struct mbuf * m)1036 pf_max_frag_size(struct mbuf *m)
1037 {
1038 struct m_tag *tag;
1039 struct pf_fragment_tag *ftag;
1040
1041 tag = m_tag_find(m, PACKET_TAG_PF_REASSEMBLED, NULL);
1042 if (tag == NULL)
1043 return (m->m_pkthdr.len);
1044
1045 ftag = (struct pf_fragment_tag *)(tag + 1);
1046
1047 return (ftag->ft_maxlen);
1048 }
1049
1050 int
pf_refragment6(struct ifnet * ifp,struct mbuf ** m0,struct m_tag * mtag,struct ifnet * rt,bool forward)1051 pf_refragment6(struct ifnet *ifp, struct mbuf **m0, struct m_tag *mtag,
1052 struct ifnet *rt, bool forward)
1053 {
1054 struct mbuf *m = *m0, *t;
1055 struct ip6_hdr *hdr;
1056 struct pf_fragment_tag *ftag = (struct pf_fragment_tag *)(mtag + 1);
1057 struct pf_pdesc pd;
1058 uint32_t frag_id;
1059 uint16_t hdrlen, extoff, maxlen;
1060 uint8_t proto;
1061 int error, action;
1062
1063 hdrlen = ftag->ft_hdrlen;
1064 extoff = ftag->ft_extoff;
1065 maxlen = ftag->ft_maxlen;
1066 frag_id = ftag->ft_id;
1067 m_tag_delete(m, mtag);
1068 mtag = NULL;
1069 ftag = NULL;
1070
1071 if (extoff) {
1072 int off;
1073
1074 /* Use protocol from next field of last extension header */
1075 m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt),
1076 &off);
1077 KASSERT((m != NULL), ("pf_refragment6: short mbuf chain"));
1078 proto = *(mtod(m, uint8_t *) + off);
1079 *(mtod(m, char *) + off) = IPPROTO_FRAGMENT;
1080 m = *m0;
1081 } else {
1082 hdr = mtod(m, struct ip6_hdr *);
1083 proto = hdr->ip6_nxt;
1084 hdr->ip6_nxt = IPPROTO_FRAGMENT;
1085 }
1086
1087 /* In case of link-local traffic we'll need a scope set. */
1088 hdr = mtod(m, struct ip6_hdr *);
1089
1090 in6_setscope(&hdr->ip6_src, ifp, NULL);
1091 in6_setscope(&hdr->ip6_dst, ifp, NULL);
1092
1093 /* The MTU must be a multiple of 8 bytes, or we risk doing the
1094 * fragmentation wrong. */
1095 maxlen = maxlen & ~7;
1096
1097 /*
1098 * Maxlen may be less than 8 if there was only a single
1099 * fragment. As it was fragmented before, add a fragment
1100 * header also for a single fragment. If total or maxlen
1101 * is less than 8, ip6_fragment() will return EMSGSIZE and
1102 * we drop the packet.
1103 */
1104 error = ip6_fragment(ifp, m, hdrlen, proto, maxlen, frag_id);
1105 m = (*m0)->m_nextpkt;
1106 (*m0)->m_nextpkt = NULL;
1107 if (error == 0) {
1108 /* The first mbuf contains the unfragmented packet. */
1109 m_freem(*m0);
1110 *m0 = NULL;
1111 action = PF_PASS;
1112 } else {
1113 /* Drop expects an mbuf to free. */
1114 DPFPRINTF(PF_DEBUG_MISC, "refragment error %d", error);
1115 action = PF_DROP;
1116 }
1117 for (; m; m = t) {
1118 t = m->m_nextpkt;
1119 m->m_nextpkt = NULL;
1120 m->m_flags |= M_SKIP_FIREWALL;
1121 memset(&pd, 0, sizeof(pd));
1122 pd.pf_mtag = pf_find_mtag(m);
1123 if (error != 0) {
1124 m_freem(m);
1125 continue;
1126 }
1127 if (rt != NULL) {
1128 struct sockaddr_in6 dst;
1129 hdr = mtod(m, struct ip6_hdr *);
1130
1131 bzero(&dst, sizeof(dst));
1132 dst.sin6_family = AF_INET6;
1133 dst.sin6_len = sizeof(dst);
1134 dst.sin6_addr = hdr->ip6_dst;
1135
1136 if (m->m_pkthdr.len <= if_getmtu(ifp)) {
1137 nd6_output_ifp(rt, rt, m, &dst, NULL);
1138 } else {
1139 in6_ifstat_inc(ifp, ifs6_in_toobig);
1140 icmp6_error(m, ICMP6_PACKET_TOO_BIG, 0,
1141 if_getmtu(ifp));
1142 }
1143 } else if (forward) {
1144 MPASS(m->m_pkthdr.rcvif != NULL);
1145 ip6_forward(m, 0);
1146 } else {
1147 (void)ip6_output(m, NULL, NULL, 0, NULL, NULL,
1148 NULL);
1149 }
1150 }
1151
1152 return (action);
1153 }
1154 #endif /* INET6 */
1155
1156 #ifdef INET
1157 int
pf_normalize_ip(u_short * reason,struct pf_pdesc * pd)1158 pf_normalize_ip(u_short *reason, struct pf_pdesc *pd)
1159 {
1160 struct pf_krule *r;
1161 struct ip *h = mtod(pd->m, struct ip *);
1162 int mff = (ntohs(h->ip_off) & IP_MF);
1163 int hlen = h->ip_hl << 2;
1164 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
1165 u_int16_t max;
1166 int ip_len;
1167 int tag = -1;
1168 int verdict;
1169 bool scrub_compat;
1170
1171 PF_RULES_RASSERT();
1172
1173 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1174 /*
1175 * Check if there are any scrub rules, matching or not.
1176 * Lack of scrub rules means:
1177 * - enforced packet normalization operation just like in OpenBSD
1178 * - fragment reassembly depends on V_pf_status.reass
1179 * With scrub rules:
1180 * - packet normalization is performed if there is a matching scrub rule
1181 * - fragment reassembly is performed if the matching rule has no
1182 * PFRULE_FRAGMENT_NOREASS flag
1183 */
1184 scrub_compat = (r != NULL);
1185 while (r != NULL) {
1186 pf_counter_u64_add(&r->evaluations, 1);
1187 if (pfi_kkif_match(r->kif, pd->kif) == r->ifnot)
1188 r = r->skip[PF_SKIP_IFP];
1189 else if (r->direction && r->direction != pd->dir)
1190 r = r->skip[PF_SKIP_DIR];
1191 else if (r->af && r->af != AF_INET)
1192 r = r->skip[PF_SKIP_AF];
1193 else if (r->proto && r->proto != h->ip_p)
1194 r = r->skip[PF_SKIP_PROTO];
1195 else if (PF_MISMATCHAW(&r->src.addr,
1196 (struct pf_addr *)&h->ip_src.s_addr, AF_INET,
1197 r->src.neg, pd->kif, M_GETFIB(pd->m)))
1198 r = r->skip[PF_SKIP_SRC_ADDR];
1199 else if (PF_MISMATCHAW(&r->dst.addr,
1200 (struct pf_addr *)&h->ip_dst.s_addr, AF_INET,
1201 r->dst.neg, NULL, M_GETFIB(pd->m)))
1202 r = r->skip[PF_SKIP_DST_ADDR];
1203 else if (r->match_tag && !pf_match_tag(pd->m, r, &tag,
1204 pd->pf_mtag ? pd->pf_mtag->tag : 0))
1205 r = TAILQ_NEXT(r, entries);
1206 else
1207 break;
1208 }
1209
1210 if (scrub_compat) {
1211 /* With scrub rules present IPv4 normalization happens only
1212 * if one of rules has matched and it's not a "no scrub" rule */
1213 if (r == NULL || r->action == PF_NOSCRUB)
1214 return (PF_PASS);
1215
1216 pf_counter_u64_critical_enter();
1217 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1);
1218 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len);
1219 pf_counter_u64_critical_exit();
1220 pf_rule_to_actions(r, &pd->act);
1221 }
1222
1223 /* Check for illegal packets */
1224 if (hlen < (int)sizeof(struct ip)) {
1225 REASON_SET(reason, PFRES_NORM);
1226 goto drop;
1227 }
1228
1229 if (hlen > ntohs(h->ip_len)) {
1230 REASON_SET(reason, PFRES_NORM);
1231 goto drop;
1232 }
1233
1234 /* Clear IP_DF if the rule uses the no-df option or we're in no-df mode */
1235 if (((!scrub_compat && V_pf_status.reass & PF_REASS_NODF) ||
1236 (r != NULL && r->rule_flag & PFRULE_NODF)) &&
1237 (h->ip_off & htons(IP_DF))
1238 ) {
1239 u_int16_t ip_off = h->ip_off;
1240
1241 h->ip_off &= htons(~IP_DF);
1242 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
1243 }
1244
1245 /* We will need other tests here */
1246 if (!fragoff && !mff)
1247 goto no_fragment;
1248
1249 /* We're dealing with a fragment now. Don't allow fragments
1250 * with IP_DF to enter the cache. If the flag was cleared by
1251 * no-df above, fine. Otherwise drop it.
1252 */
1253 if (h->ip_off & htons(IP_DF)) {
1254 DPFPRINTF(PF_DEBUG_MISC, "IP_DF");
1255 goto bad;
1256 }
1257
1258 ip_len = ntohs(h->ip_len) - hlen;
1259
1260 /* All fragments are 8 byte aligned */
1261 if (mff && (ip_len & 0x7)) {
1262 DPFPRINTF(PF_DEBUG_MISC, "mff and %d", ip_len);
1263 goto bad;
1264 }
1265
1266 /* Respect maximum length */
1267 if (fragoff + ip_len > IP_MAXPACKET) {
1268 DPFPRINTF(PF_DEBUG_MISC, "max packet %d", fragoff + ip_len);
1269 goto bad;
1270 }
1271
1272 if ((!scrub_compat && V_pf_status.reass) ||
1273 (r != NULL && !(r->rule_flag & PFRULE_FRAGMENT_NOREASS))
1274 ) {
1275 max = fragoff + ip_len;
1276
1277 /* Fully buffer all of the fragments
1278 * Might return a completely reassembled mbuf, or NULL */
1279 PF_FRAG_LOCK();
1280 DPFPRINTF(PF_DEBUG_MISC, "reass frag %d @ %d-%d",
1281 h->ip_id, fragoff, max);
1282 verdict = pf_reassemble(&pd->m, pd->dir, reason);
1283 PF_FRAG_UNLOCK();
1284
1285 if (verdict != PF_PASS)
1286 return (PF_DROP);
1287
1288 if (pd->m == NULL)
1289 return (PF_DROP);
1290
1291 h = mtod(pd->m, struct ip *);
1292 pd->tot_len = htons(h->ip_len);
1293
1294 no_fragment:
1295 /* At this point, only IP_DF is allowed in ip_off */
1296 if (h->ip_off & ~htons(IP_DF)) {
1297 u_int16_t ip_off = h->ip_off;
1298
1299 h->ip_off &= htons(IP_DF);
1300 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
1301 }
1302 }
1303
1304 return (PF_PASS);
1305
1306 bad:
1307 DPFPRINTF(PF_DEBUG_MISC, "dropping bad fragment");
1308 REASON_SET(reason, PFRES_FRAG);
1309 drop:
1310 if (r != NULL && r->log)
1311 PFLOG_PACKET(PF_DROP, *reason, r, NULL, NULL, pd, 1, NULL);
1312
1313 return (PF_DROP);
1314 }
1315 #endif
1316
1317 #ifdef INET6
1318 int
pf_normalize_ip6(int off,u_short * reason,struct pf_pdesc * pd)1319 pf_normalize_ip6(int off, u_short *reason,
1320 struct pf_pdesc *pd)
1321 {
1322 struct pf_krule *r;
1323 struct ip6_hdr *h;
1324 struct ip6_frag frag;
1325 bool scrub_compat;
1326
1327 PF_RULES_RASSERT();
1328
1329 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1330 /*
1331 * Check if there are any scrub rules, matching or not.
1332 * Lack of scrub rules means:
1333 * - enforced packet normalization operation just like in OpenBSD
1334 * With scrub rules:
1335 * - packet normalization is performed if there is a matching scrub rule
1336 * XXX: Fragment reassembly always performed for IPv6!
1337 */
1338 scrub_compat = (r != NULL);
1339 while (r != NULL) {
1340 pf_counter_u64_add(&r->evaluations, 1);
1341 if (pfi_kkif_match(r->kif, pd->kif) == r->ifnot)
1342 r = r->skip[PF_SKIP_IFP];
1343 else if (r->direction && r->direction != pd->dir)
1344 r = r->skip[PF_SKIP_DIR];
1345 else if (r->af && r->af != AF_INET6)
1346 r = r->skip[PF_SKIP_AF];
1347 else if (r->proto && r->proto != pd->proto)
1348 r = r->skip[PF_SKIP_PROTO];
1349 else if (PF_MISMATCHAW(&r->src.addr,
1350 (struct pf_addr *)&pd->src, AF_INET6,
1351 r->src.neg, pd->kif, M_GETFIB(pd->m)))
1352 r = r->skip[PF_SKIP_SRC_ADDR];
1353 else if (PF_MISMATCHAW(&r->dst.addr,
1354 (struct pf_addr *)&pd->dst, AF_INET6,
1355 r->dst.neg, NULL, M_GETFIB(pd->m)))
1356 r = r->skip[PF_SKIP_DST_ADDR];
1357 else
1358 break;
1359 }
1360
1361 if (scrub_compat) {
1362 /* With scrub rules present IPv6 normalization happens only
1363 * if one of rules has matched and it's not a "no scrub" rule */
1364 if (r == NULL || r->action == PF_NOSCRUB)
1365 return (PF_PASS);
1366
1367 pf_counter_u64_critical_enter();
1368 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1);
1369 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len);
1370 pf_counter_u64_critical_exit();
1371 pf_rule_to_actions(r, &pd->act);
1372 }
1373
1374 if (!pf_pull_hdr(pd->m, off, &frag, sizeof(frag), reason, AF_INET6))
1375 return (PF_DROP);
1376
1377 /* Offset now points to data portion. */
1378 off += sizeof(frag);
1379
1380 if (pd->virtual_proto == PF_VPROTO_FRAGMENT) {
1381 /* Returns PF_DROP or *m0 is NULL or completely reassembled
1382 * mbuf. */
1383 if (pf_reassemble6(&pd->m, &frag, off, pd->extoff, pd->dir, reason)
1384 != PF_PASS)
1385 return (PF_DROP);
1386 if (pd->m == NULL)
1387 return (PF_DROP);
1388 h = mtod(pd->m, struct ip6_hdr *);
1389 pd->tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr);
1390 }
1391
1392 return (PF_PASS);
1393 }
1394 #endif /* INET6 */
1395
1396 int
pf_normalize_tcp(struct pf_pdesc * pd)1397 pf_normalize_tcp(struct pf_pdesc *pd)
1398 {
1399 struct pf_krule *r, *rm = NULL;
1400 struct tcphdr *th = &pd->hdr.tcp;
1401 int rewrite = 0;
1402 u_short reason;
1403 u_int16_t flags;
1404 sa_family_t af = pd->af;
1405 int srs;
1406
1407 PF_RULES_RASSERT();
1408
1409 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1410 /* Check if there any scrub rules. Lack of scrub rules means enforced
1411 * packet normalization operation just like in OpenBSD. */
1412 srs = (r != NULL);
1413 while (r != NULL) {
1414 pf_counter_u64_add(&r->evaluations, 1);
1415 if (pfi_kkif_match(r->kif, pd->kif) == r->ifnot)
1416 r = r->skip[PF_SKIP_IFP];
1417 else if (r->direction && r->direction != pd->dir)
1418 r = r->skip[PF_SKIP_DIR];
1419 else if (r->af && r->af != af)
1420 r = r->skip[PF_SKIP_AF];
1421 else if (r->proto && r->proto != pd->proto)
1422 r = r->skip[PF_SKIP_PROTO];
1423 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
1424 r->src.neg, pd->kif, M_GETFIB(pd->m)))
1425 r = r->skip[PF_SKIP_SRC_ADDR];
1426 else if (r->src.port_op && !pf_match_port(r->src.port_op,
1427 r->src.port[0], r->src.port[1], th->th_sport))
1428 r = r->skip[PF_SKIP_SRC_PORT];
1429 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
1430 r->dst.neg, NULL, M_GETFIB(pd->m)))
1431 r = r->skip[PF_SKIP_DST_ADDR];
1432 else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
1433 r->dst.port[0], r->dst.port[1], th->th_dport))
1434 r = r->skip[PF_SKIP_DST_PORT];
1435 else if (r->os_fingerprint != PF_OSFP_ANY && !pf_osfp_match(
1436 pf_osfp_fingerprint(pd, th),
1437 r->os_fingerprint))
1438 r = TAILQ_NEXT(r, entries);
1439 else {
1440 rm = r;
1441 break;
1442 }
1443 }
1444
1445 if (srs) {
1446 /* With scrub rules present TCP normalization happens only
1447 * if one of rules has matched and it's not a "no scrub" rule */
1448 if (rm == NULL || rm->action == PF_NOSCRUB)
1449 return (PF_PASS);
1450
1451 pf_counter_u64_critical_enter();
1452 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1);
1453 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len);
1454 pf_counter_u64_critical_exit();
1455 pf_rule_to_actions(rm, &pd->act);
1456 }
1457
1458 if (rm && rm->rule_flag & PFRULE_REASSEMBLE_TCP)
1459 pd->flags |= PFDESC_TCP_NORM;
1460
1461 flags = tcp_get_flags(th);
1462 if (flags & TH_SYN) {
1463 /* Illegal packet */
1464 if (flags & TH_RST)
1465 goto tcp_drop;
1466
1467 if (flags & TH_FIN)
1468 goto tcp_drop;
1469 } else {
1470 /* Illegal packet */
1471 if (!(flags & (TH_ACK|TH_RST)))
1472 goto tcp_drop;
1473 }
1474
1475 if (!(flags & TH_ACK)) {
1476 /* These flags are only valid if ACK is set */
1477 if ((flags & TH_FIN) || (flags & TH_PUSH) || (flags & TH_URG))
1478 goto tcp_drop;
1479 }
1480
1481 /* Check for illegal header length */
1482 if (th->th_off < (sizeof(struct tcphdr) >> 2))
1483 goto tcp_drop;
1484
1485 /* If flags changed, or reserved data set, then adjust */
1486 if (flags != tcp_get_flags(th) ||
1487 (tcp_get_flags(th) & (TH_RES1|TH_RES2|TH_RES2)) != 0) {
1488 u_int16_t ov, nv;
1489
1490 ov = *(u_int16_t *)(&th->th_ack + 1);
1491 flags &= ~(TH_RES1 | TH_RES2 | TH_RES3);
1492 tcp_set_flags(th, flags);
1493 nv = *(u_int16_t *)(&th->th_ack + 1);
1494
1495 th->th_sum = pf_proto_cksum_fixup(pd->m, th->th_sum, ov, nv, 0);
1496 rewrite = 1;
1497 }
1498
1499 /* Remove urgent pointer, if TH_URG is not set */
1500 if (!(flags & TH_URG) && th->th_urp) {
1501 th->th_sum = pf_proto_cksum_fixup(pd->m, th->th_sum, th->th_urp,
1502 0, 0);
1503 th->th_urp = 0;
1504 rewrite = 1;
1505 }
1506
1507 /* copy back packet headers if we sanitized */
1508 if (rewrite)
1509 m_copyback(pd->m, pd->off, sizeof(*th), (caddr_t)th);
1510
1511 return (PF_PASS);
1512
1513 tcp_drop:
1514 REASON_SET(&reason, PFRES_NORM);
1515 if (rm != NULL && r->log)
1516 PFLOG_PACKET(PF_DROP, reason, r, NULL, NULL, pd, 1, NULL);
1517 return (PF_DROP);
1518 }
1519
1520 int
pf_normalize_tcp_init(struct pf_pdesc * pd,struct tcphdr * th,struct pf_state_peer * src)1521 pf_normalize_tcp_init(struct pf_pdesc *pd, struct tcphdr *th,
1522 struct pf_state_peer *src)
1523 {
1524 u_int32_t tsval, tsecr;
1525 int olen;
1526 uint8_t opts[MAX_TCPOPTLEN], *opt;
1527
1528 KASSERT((src->scrub == NULL),
1529 ("pf_normalize_tcp_init: src->scrub != NULL"));
1530
1531 src->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT);
1532 if (src->scrub == NULL)
1533 return (1);
1534
1535 switch (pd->af) {
1536 #ifdef INET
1537 case AF_INET: {
1538 struct ip *h = mtod(pd->m, struct ip *);
1539 src->scrub->pfss_ttl = h->ip_ttl;
1540 break;
1541 }
1542 #endif /* INET */
1543 #ifdef INET6
1544 case AF_INET6: {
1545 struct ip6_hdr *h = mtod(pd->m, struct ip6_hdr *);
1546 src->scrub->pfss_ttl = h->ip6_hlim;
1547 break;
1548 }
1549 #endif /* INET6 */
1550 default:
1551 unhandled_af(pd->af);
1552 }
1553
1554 /*
1555 * All normalizations below are only begun if we see the start of
1556 * the connections. They must all set an enabled bit in pfss_flags
1557 */
1558 if ((tcp_get_flags(th) & TH_SYN) == 0)
1559 return (0);
1560
1561 olen = (th->th_off << 2) - sizeof(*th);
1562 if (olen < TCPOLEN_TIMESTAMP || !pf_pull_hdr(pd->m,
1563 pd->off + sizeof(*th), opts, olen, NULL, pd->af))
1564 return (0);
1565
1566 opt = opts;
1567 while ((opt = pf_find_tcpopt(opt, opts, olen,
1568 TCPOPT_TIMESTAMP, TCPOLEN_TIMESTAMP)) != NULL) {
1569 src->scrub->pfss_flags |= PFSS_TIMESTAMP;
1570 src->scrub->pfss_ts_mod = arc4random();
1571 /* note PFSS_PAWS not set yet */
1572 memcpy(&tsval, &opt[2], sizeof(u_int32_t));
1573 memcpy(&tsecr, &opt[6], sizeof(u_int32_t));
1574 src->scrub->pfss_tsval0 = ntohl(tsval);
1575 src->scrub->pfss_tsval = ntohl(tsval);
1576 src->scrub->pfss_tsecr = ntohl(tsecr);
1577 getmicrouptime(&src->scrub->pfss_last);
1578
1579 opt += opt[1];
1580 }
1581
1582 return (0);
1583 }
1584
1585 void
pf_normalize_tcp_cleanup(struct pf_kstate * state)1586 pf_normalize_tcp_cleanup(struct pf_kstate *state)
1587 {
1588 /* XXX Note: this also cleans up SCTP. */
1589 uma_zfree(V_pf_state_scrub_z, state->src.scrub);
1590 uma_zfree(V_pf_state_scrub_z, state->dst.scrub);
1591
1592 /* Someday... flush the TCP segment reassembly descriptors. */
1593 }
1594 int
pf_normalize_sctp_init(struct pf_pdesc * pd,struct pf_state_peer * src,struct pf_state_peer * dst)1595 pf_normalize_sctp_init(struct pf_pdesc *pd, struct pf_state_peer *src,
1596 struct pf_state_peer *dst)
1597 {
1598 src->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT);
1599 if (src->scrub == NULL)
1600 return (1);
1601
1602 dst->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT);
1603 if (dst->scrub == NULL) {
1604 uma_zfree(V_pf_state_scrub_z, src);
1605 return (1);
1606 }
1607
1608 dst->scrub->pfss_v_tag = pd->sctp_initiate_tag;
1609
1610 return (0);
1611 }
1612
1613 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)1614 pf_normalize_tcp_stateful(struct pf_pdesc *pd,
1615 u_short *reason, struct tcphdr *th, struct pf_kstate *state,
1616 struct pf_state_peer *src, struct pf_state_peer *dst, int *writeback)
1617 {
1618 struct timeval uptime;
1619 u_int tsval_from_last;
1620 uint32_t tsval, tsecr;
1621 int copyback = 0;
1622 int got_ts = 0;
1623 int olen;
1624 uint8_t opts[MAX_TCPOPTLEN], *opt;
1625
1626 KASSERT((src->scrub || dst->scrub),
1627 ("%s: src->scrub && dst->scrub!", __func__));
1628
1629 /*
1630 * Enforce the minimum TTL seen for this connection. Negate a common
1631 * technique to evade an intrusion detection system and confuse
1632 * firewall state code.
1633 */
1634 switch (pd->af) {
1635 #ifdef INET
1636 case AF_INET: {
1637 if (src->scrub) {
1638 struct ip *h = mtod(pd->m, struct ip *);
1639 if (h->ip_ttl > src->scrub->pfss_ttl)
1640 src->scrub->pfss_ttl = h->ip_ttl;
1641 h->ip_ttl = src->scrub->pfss_ttl;
1642 }
1643 break;
1644 }
1645 #endif /* INET */
1646 #ifdef INET6
1647 case AF_INET6: {
1648 if (src->scrub) {
1649 struct ip6_hdr *h = mtod(pd->m, struct ip6_hdr *);
1650 if (h->ip6_hlim > src->scrub->pfss_ttl)
1651 src->scrub->pfss_ttl = h->ip6_hlim;
1652 h->ip6_hlim = src->scrub->pfss_ttl;
1653 }
1654 break;
1655 }
1656 #endif /* INET6 */
1657 default:
1658 unhandled_af(pd->af);
1659 }
1660
1661 olen = (th->th_off << 2) - sizeof(*th);
1662
1663 if (olen >= TCPOLEN_TIMESTAMP &&
1664 ((src->scrub && (src->scrub->pfss_flags & PFSS_TIMESTAMP)) ||
1665 (dst->scrub && (dst->scrub->pfss_flags & PFSS_TIMESTAMP))) &&
1666 pf_pull_hdr(pd->m, pd->off + sizeof(*th), opts, olen, NULL, pd->af)) {
1667 /* Modulate the timestamps. Can be used for NAT detection, OS
1668 * uptime determination or reboot detection.
1669 */
1670 opt = opts;
1671 while ((opt = pf_find_tcpopt(opt, opts, olen,
1672 TCPOPT_TIMESTAMP, TCPOLEN_TIMESTAMP)) != NULL) {
1673 uint8_t *ts = opt + 2;
1674 uint8_t *tsr = opt + 6;
1675
1676 if (got_ts) {
1677 /* Huh? Multiple timestamps!? */
1678 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1679 printf("pf: %s: multiple TS??", __func__);
1680 pf_print_state(state);
1681 printf("\n");
1682 }
1683 REASON_SET(reason, PFRES_TS);
1684 return (PF_DROP);
1685 }
1686
1687 memcpy(&tsval, ts, sizeof(u_int32_t));
1688 memcpy(&tsecr, tsr, sizeof(u_int32_t));
1689
1690 /* modulate TS */
1691 if (tsval && src->scrub &&
1692 (src->scrub->pfss_flags & PFSS_TIMESTAMP)) {
1693 /* tsval used further on */
1694 tsval = ntohl(tsval);
1695 pf_patch_32(pd,
1696 ts, htonl(tsval + src->scrub->pfss_ts_mod),
1697 PF_ALGNMNT(ts - opts));
1698 copyback = 1;
1699 }
1700
1701 /* modulate TS reply if any (!0) */
1702 if (tsecr && dst->scrub &&
1703 (dst->scrub->pfss_flags & PFSS_TIMESTAMP)) {
1704 /* tsecr used further on */
1705 tsecr = ntohl(tsecr) - dst->scrub->pfss_ts_mod;
1706 pf_patch_32(pd, tsr, htonl(tsecr),
1707 PF_ALGNMNT(tsr - opts));
1708 copyback = 1;
1709 }
1710
1711 got_ts = 1;
1712 opt += opt[1];
1713 }
1714
1715 if (copyback) {
1716 /* Copyback the options, caller copys back header */
1717 *writeback = 1;
1718 m_copyback(pd->m, pd->off + sizeof(*th), olen, opts);
1719 }
1720 }
1721
1722 /*
1723 * Must invalidate PAWS checks on connections idle for too long.
1724 * The fastest allowed timestamp clock is 1ms. That turns out to
1725 * be about 24 days before it wraps. XXX Right now our lowerbound
1726 * TS echo check only works for the first 12 days of a connection
1727 * when the TS has exhausted half its 32bit space
1728 */
1729 #define TS_MAX_IDLE (24*24*60*60)
1730 #define TS_MAX_CONN (12*24*60*60) /* XXX remove when better tsecr check */
1731
1732 getmicrouptime(&uptime);
1733 if (src->scrub && (src->scrub->pfss_flags & PFSS_PAWS) &&
1734 (uptime.tv_sec - src->scrub->pfss_last.tv_sec > TS_MAX_IDLE ||
1735 time_uptime - (state->creation / 1000) > TS_MAX_CONN)) {
1736 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1737 DPFPRINTF(PF_DEBUG_MISC, "src idled out of PAWS");
1738 pf_print_state(state);
1739 printf("\n");
1740 }
1741 src->scrub->pfss_flags = (src->scrub->pfss_flags & ~PFSS_PAWS)
1742 | PFSS_PAWS_IDLED;
1743 }
1744 if (dst->scrub && (dst->scrub->pfss_flags & PFSS_PAWS) &&
1745 uptime.tv_sec - dst->scrub->pfss_last.tv_sec > TS_MAX_IDLE) {
1746 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1747 DPFPRINTF(PF_DEBUG_MISC, "dst idled out of PAWS");
1748 pf_print_state(state);
1749 printf("\n");
1750 }
1751 dst->scrub->pfss_flags = (dst->scrub->pfss_flags & ~PFSS_PAWS)
1752 | PFSS_PAWS_IDLED;
1753 }
1754
1755 if (got_ts && src->scrub && dst->scrub &&
1756 (src->scrub->pfss_flags & PFSS_PAWS) &&
1757 (dst->scrub->pfss_flags & PFSS_PAWS)) {
1758 /* Validate that the timestamps are "in-window".
1759 * RFC1323 describes TCP Timestamp options that allow
1760 * measurement of RTT (round trip time) and PAWS
1761 * (protection against wrapped sequence numbers). PAWS
1762 * gives us a set of rules for rejecting packets on
1763 * long fat pipes (packets that were somehow delayed
1764 * in transit longer than the time it took to send the
1765 * full TCP sequence space of 4Gb). We can use these
1766 * rules and infer a few others that will let us treat
1767 * the 32bit timestamp and the 32bit echoed timestamp
1768 * as sequence numbers to prevent a blind attacker from
1769 * inserting packets into a connection.
1770 *
1771 * RFC1323 tells us:
1772 * - The timestamp on this packet must be greater than
1773 * or equal to the last value echoed by the other
1774 * endpoint. The RFC says those will be discarded
1775 * since it is a dup that has already been acked.
1776 * This gives us a lowerbound on the timestamp.
1777 * timestamp >= other last echoed timestamp
1778 * - The timestamp will be less than or equal to
1779 * the last timestamp plus the time between the
1780 * last packet and now. The RFC defines the max
1781 * clock rate as 1ms. We will allow clocks to be
1782 * up to 10% fast and will allow a total difference
1783 * or 30 seconds due to a route change. And this
1784 * gives us an upperbound on the timestamp.
1785 * timestamp <= last timestamp + max ticks
1786 * We have to be careful here. Windows will send an
1787 * initial timestamp of zero and then initialize it
1788 * to a random value after the 3whs; presumably to
1789 * avoid a DoS by having to call an expensive RNG
1790 * during a SYN flood. Proof MS has at least one
1791 * good security geek.
1792 *
1793 * - The TCP timestamp option must also echo the other
1794 * endpoints timestamp. The timestamp echoed is the
1795 * one carried on the earliest unacknowledged segment
1796 * on the left edge of the sequence window. The RFC
1797 * states that the host will reject any echoed
1798 * timestamps that were larger than any ever sent.
1799 * This gives us an upperbound on the TS echo.
1800 * tescr <= largest_tsval
1801 * - The lowerbound on the TS echo is a little more
1802 * tricky to determine. The other endpoint's echoed
1803 * values will not decrease. But there may be
1804 * network conditions that re-order packets and
1805 * cause our view of them to decrease. For now the
1806 * only lowerbound we can safely determine is that
1807 * the TS echo will never be less than the original
1808 * TS. XXX There is probably a better lowerbound.
1809 * Remove TS_MAX_CONN with better lowerbound check.
1810 * tescr >= other original TS
1811 *
1812 * It is also important to note that the fastest
1813 * timestamp clock of 1ms will wrap its 32bit space in
1814 * 24 days. So we just disable TS checking after 24
1815 * days of idle time. We actually must use a 12d
1816 * connection limit until we can come up with a better
1817 * lowerbound to the TS echo check.
1818 */
1819 struct timeval delta_ts;
1820 int ts_fudge;
1821
1822 /*
1823 * PFTM_TS_DIFF is how many seconds of leeway to allow
1824 * a host's timestamp. This can happen if the previous
1825 * packet got delayed in transit for much longer than
1826 * this packet.
1827 */
1828 if ((ts_fudge = state->rule->timeout[PFTM_TS_DIFF]) == 0)
1829 ts_fudge = V_pf_default_rule.timeout[PFTM_TS_DIFF];
1830
1831 /* Calculate max ticks since the last timestamp */
1832 #define TS_MAXFREQ 1100 /* RFC max TS freq of 1Khz + 10% skew */
1833 #define TS_MICROSECS 1000000 /* microseconds per second */
1834 delta_ts = uptime;
1835 timevalsub(&delta_ts, &src->scrub->pfss_last);
1836 tsval_from_last = (delta_ts.tv_sec + ts_fudge) * TS_MAXFREQ;
1837 tsval_from_last += delta_ts.tv_usec / (TS_MICROSECS/TS_MAXFREQ);
1838
1839 if ((src->state >= TCPS_ESTABLISHED &&
1840 dst->state >= TCPS_ESTABLISHED) &&
1841 (SEQ_LT(tsval, dst->scrub->pfss_tsecr) ||
1842 SEQ_GT(tsval, src->scrub->pfss_tsval + tsval_from_last) ||
1843 (tsecr && (SEQ_GT(tsecr, dst->scrub->pfss_tsval) ||
1844 SEQ_LT(tsecr, dst->scrub->pfss_tsval0))))) {
1845 /* Bad RFC1323 implementation or an insertion attack.
1846 *
1847 * - Solaris 2.6 and 2.7 are known to send another ACK
1848 * after the FIN,FIN|ACK,ACK closing that carries
1849 * an old timestamp.
1850 */
1851
1852 DPFPRINTF(PF_DEBUG_MISC, "Timestamp failed %c%c%c%c",
1853 SEQ_LT(tsval, dst->scrub->pfss_tsecr) ? '0' : ' ',
1854 SEQ_GT(tsval, src->scrub->pfss_tsval +
1855 tsval_from_last) ? '1' : ' ',
1856 SEQ_GT(tsecr, dst->scrub->pfss_tsval) ? '2' : ' ',
1857 SEQ_LT(tsecr, dst->scrub->pfss_tsval0)? '3' : ' ');
1858 DPFPRINTF(PF_DEBUG_MISC, " tsval: %u tsecr: %u +ticks: "
1859 "%u idle: %jus %lums",
1860 tsval, tsecr, tsval_from_last,
1861 (uintmax_t)delta_ts.tv_sec,
1862 delta_ts.tv_usec / 1000);
1863 DPFPRINTF(PF_DEBUG_MISC, " src->tsval: %u tsecr: %u",
1864 src->scrub->pfss_tsval, src->scrub->pfss_tsecr);
1865 DPFPRINTF(PF_DEBUG_MISC, " dst->tsval: %u tsecr: %u "
1866 "tsval0: %u", dst->scrub->pfss_tsval,
1867 dst->scrub->pfss_tsecr, dst->scrub->pfss_tsval0);
1868 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1869 pf_print_state(state);
1870 pf_print_flags(tcp_get_flags(th));
1871 printf("\n");
1872 }
1873 REASON_SET(reason, PFRES_TS);
1874 return (PF_DROP);
1875 }
1876
1877 /* XXX I'd really like to require tsecr but it's optional */
1878
1879 } else if (!got_ts && (tcp_get_flags(th) & TH_RST) == 0 &&
1880 ((src->state == TCPS_ESTABLISHED && dst->state == TCPS_ESTABLISHED)
1881 || pd->p_len > 0 || (tcp_get_flags(th) & TH_SYN)) &&
1882 src->scrub && dst->scrub &&
1883 (src->scrub->pfss_flags & PFSS_PAWS) &&
1884 (dst->scrub->pfss_flags & PFSS_PAWS)) {
1885 /* Didn't send a timestamp. Timestamps aren't really useful
1886 * when:
1887 * - connection opening or closing (often not even sent).
1888 * but we must not let an attacker to put a FIN on a
1889 * data packet to sneak it through our ESTABLISHED check.
1890 * - on a TCP reset. RFC suggests not even looking at TS.
1891 * - on an empty ACK. The TS will not be echoed so it will
1892 * probably not help keep the RTT calculation in sync and
1893 * there isn't as much danger when the sequence numbers
1894 * got wrapped. So some stacks don't include TS on empty
1895 * ACKs :-(
1896 *
1897 * To minimize the disruption to mostly RFC1323 conformant
1898 * stacks, we will only require timestamps on data packets.
1899 *
1900 * And what do ya know, we cannot require timestamps on data
1901 * packets. There appear to be devices that do legitimate
1902 * TCP connection hijacking. There are HTTP devices that allow
1903 * a 3whs (with timestamps) and then buffer the HTTP request.
1904 * If the intermediate device has the HTTP response cache, it
1905 * will spoof the response but not bother timestamping its
1906 * packets. So we can look for the presence of a timestamp in
1907 * the first data packet and if there, require it in all future
1908 * packets.
1909 */
1910
1911 if (pd->p_len > 0 && (src->scrub->pfss_flags & PFSS_DATA_TS)) {
1912 /*
1913 * Hey! Someone tried to sneak a packet in. Or the
1914 * stack changed its RFC1323 behavior?!?!
1915 */
1916 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1917 DPFPRINTF(PF_DEBUG_MISC, "Did not receive expected "
1918 "RFC1323 timestamp");
1919 pf_print_state(state);
1920 pf_print_flags(tcp_get_flags(th));
1921 printf("\n");
1922 }
1923 REASON_SET(reason, PFRES_TS);
1924 return (PF_DROP);
1925 }
1926 }
1927
1928 /*
1929 * We will note if a host sends his data packets with or without
1930 * timestamps. And require all data packets to contain a timestamp
1931 * if the first does. PAWS implicitly requires that all data packets be
1932 * timestamped. But I think there are middle-man devices that hijack
1933 * TCP streams immediately after the 3whs and don't timestamp their
1934 * packets (seen in a WWW accelerator or cache).
1935 */
1936 if (pd->p_len > 0 && src->scrub && (src->scrub->pfss_flags &
1937 (PFSS_TIMESTAMP|PFSS_DATA_TS|PFSS_DATA_NOTS)) == PFSS_TIMESTAMP) {
1938 if (got_ts)
1939 src->scrub->pfss_flags |= PFSS_DATA_TS;
1940 else {
1941 src->scrub->pfss_flags |= PFSS_DATA_NOTS;
1942 if (V_pf_status.debug >= PF_DEBUG_MISC && dst->scrub &&
1943 (dst->scrub->pfss_flags & PFSS_TIMESTAMP)) {
1944 /* Don't warn if other host rejected RFC1323 */
1945 DPFPRINTF(PF_DEBUG_MISC, "Broken RFC1323 stack did "
1946 "not timestamp data packet. Disabled PAWS "
1947 "security.");
1948 pf_print_state(state);
1949 pf_print_flags(tcp_get_flags(th));
1950 printf("\n");
1951 }
1952 }
1953 }
1954
1955 /*
1956 * Update PAWS values
1957 */
1958 if (got_ts && src->scrub && PFSS_TIMESTAMP == (src->scrub->pfss_flags &
1959 (PFSS_PAWS_IDLED|PFSS_TIMESTAMP))) {
1960 getmicrouptime(&src->scrub->pfss_last);
1961 if (SEQ_GEQ(tsval, src->scrub->pfss_tsval) ||
1962 (src->scrub->pfss_flags & PFSS_PAWS) == 0)
1963 src->scrub->pfss_tsval = tsval;
1964
1965 if (tsecr) {
1966 if (SEQ_GEQ(tsecr, src->scrub->pfss_tsecr) ||
1967 (src->scrub->pfss_flags & PFSS_PAWS) == 0)
1968 src->scrub->pfss_tsecr = tsecr;
1969
1970 if ((src->scrub->pfss_flags & PFSS_PAWS) == 0 &&
1971 (SEQ_LT(tsval, src->scrub->pfss_tsval0) ||
1972 src->scrub->pfss_tsval0 == 0)) {
1973 /* tsval0 MUST be the lowest timestamp */
1974 src->scrub->pfss_tsval0 = tsval;
1975 }
1976
1977 /* Only fully initialized after a TS gets echoed */
1978 if ((src->scrub->pfss_flags & PFSS_PAWS) == 0)
1979 src->scrub->pfss_flags |= PFSS_PAWS;
1980 }
1981 }
1982
1983 /* I have a dream.... TCP segment reassembly.... */
1984 return (0);
1985 }
1986
1987 int
pf_normalize_mss(struct pf_pdesc * pd)1988 pf_normalize_mss(struct pf_pdesc *pd)
1989 {
1990 int olen, optsoff;
1991 uint8_t opts[MAX_TCPOPTLEN], *opt;
1992
1993 olen = (pd->hdr.tcp.th_off << 2) - sizeof(struct tcphdr);
1994 optsoff = pd->off + sizeof(struct tcphdr);
1995 if (olen < TCPOLEN_MAXSEG ||
1996 !pf_pull_hdr(pd->m, optsoff, opts, olen, NULL, pd->af))
1997 return (0);
1998
1999 opt = opts;
2000 while ((opt = pf_find_tcpopt(opt, opts, olen,
2001 TCPOPT_MAXSEG, TCPOLEN_MAXSEG)) != NULL) {
2002 uint16_t mss;
2003 uint8_t *mssp = opt + 2;
2004 memcpy(&mss, mssp, sizeof(mss));
2005 if (ntohs(mss) > pd->act.max_mss) {
2006 size_t mssoffopts = mssp - opts;
2007 pf_patch_16(pd, &mss,
2008 htons(pd->act.max_mss), PF_ALGNMNT(mssoffopts));
2009 m_copyback(pd->m, optsoff + mssoffopts,
2010 sizeof(mss), (caddr_t)&mss);
2011 m_copyback(pd->m, pd->off,
2012 sizeof(struct tcphdr), (caddr_t)&pd->hdr.tcp);
2013 }
2014
2015 opt += opt[1];
2016 }
2017
2018 return (0);
2019 }
2020
2021 int
pf_scan_sctp(struct pf_pdesc * pd)2022 pf_scan_sctp(struct pf_pdesc *pd)
2023 {
2024 struct sctp_chunkhdr ch = { };
2025 int chunk_off = sizeof(struct sctphdr);
2026 int chunk_start;
2027 int ret;
2028
2029 while (pd->off + chunk_off < pd->tot_len) {
2030 if (!pf_pull_hdr(pd->m, pd->off + chunk_off, &ch, sizeof(ch),
2031 NULL, pd->af))
2032 return (PF_DROP);
2033
2034 /* Length includes the header, this must be at least 4. */
2035 if (ntohs(ch.chunk_length) < 4)
2036 return (PF_DROP);
2037
2038 chunk_start = chunk_off;
2039 chunk_off += roundup(ntohs(ch.chunk_length), 4);
2040
2041 switch (ch.chunk_type) {
2042 case SCTP_INITIATION:
2043 case SCTP_INITIATION_ACK: {
2044 struct sctp_init_chunk init;
2045
2046 if (!pf_pull_hdr(pd->m, pd->off + chunk_start, &init,
2047 sizeof(init), NULL, pd->af))
2048 return (PF_DROP);
2049
2050 /*
2051 * RFC 9620, Section 3.3.2, "The Initiate Tag is allowed to have
2052 * any value except 0."
2053 */
2054 if (init.init.initiate_tag == 0)
2055 return (PF_DROP);
2056 if (init.init.num_inbound_streams == 0)
2057 return (PF_DROP);
2058 if (init.init.num_outbound_streams == 0)
2059 return (PF_DROP);
2060 if (ntohl(init.init.a_rwnd) < SCTP_MIN_RWND)
2061 return (PF_DROP);
2062
2063 /*
2064 * RFC 9260, Section 3.1, INIT chunks MUST have zero
2065 * verification tag.
2066 */
2067 if (ch.chunk_type == SCTP_INITIATION &&
2068 pd->hdr.sctp.v_tag != 0)
2069 return (PF_DROP);
2070
2071 pd->sctp_initiate_tag = init.init.initiate_tag;
2072
2073 if (ch.chunk_type == SCTP_INITIATION)
2074 pd->sctp_flags |= PFDESC_SCTP_INIT;
2075 else
2076 pd->sctp_flags |= PFDESC_SCTP_INIT_ACK;
2077
2078 ret = pf_multihome_scan_init(pd->off + chunk_start,
2079 ntohs(init.ch.chunk_length), pd);
2080 if (ret != PF_PASS)
2081 return (ret);
2082
2083 break;
2084 }
2085 case SCTP_ABORT_ASSOCIATION:
2086 pd->sctp_flags |= PFDESC_SCTP_ABORT;
2087 break;
2088 case SCTP_SHUTDOWN:
2089 case SCTP_SHUTDOWN_ACK:
2090 pd->sctp_flags |= PFDESC_SCTP_SHUTDOWN;
2091 break;
2092 case SCTP_SHUTDOWN_COMPLETE:
2093 pd->sctp_flags |= PFDESC_SCTP_SHUTDOWN_COMPLETE;
2094 break;
2095 case SCTP_COOKIE_ECHO:
2096 pd->sctp_flags |= PFDESC_SCTP_COOKIE;
2097 break;
2098 case SCTP_COOKIE_ACK:
2099 pd->sctp_flags |= PFDESC_SCTP_COOKIE_ACK;
2100 break;
2101 case SCTP_DATA:
2102 pd->sctp_flags |= PFDESC_SCTP_DATA;
2103 break;
2104 case SCTP_HEARTBEAT_REQUEST:
2105 pd->sctp_flags |= PFDESC_SCTP_HEARTBEAT;
2106 break;
2107 case SCTP_HEARTBEAT_ACK:
2108 pd->sctp_flags |= PFDESC_SCTP_HEARTBEAT_ACK;
2109 break;
2110 case SCTP_ASCONF:
2111 pd->sctp_flags |= PFDESC_SCTP_ASCONF;
2112
2113 ret = pf_multihome_scan_asconf(pd->off + chunk_start,
2114 ntohs(ch.chunk_length), pd);
2115 if (ret != PF_PASS)
2116 return (ret);
2117 break;
2118 default:
2119 pd->sctp_flags |= PFDESC_SCTP_OTHER;
2120 break;
2121 }
2122 }
2123
2124 /* Validate chunk lengths vs. packet length. */
2125 if (pd->off + chunk_off != pd->tot_len)
2126 return (PF_DROP);
2127
2128 /*
2129 * INIT, INIT_ACK or SHUTDOWN_COMPLETE chunks must always be the only
2130 * one in a packet.
2131 */
2132 if ((pd->sctp_flags & PFDESC_SCTP_INIT) &&
2133 (pd->sctp_flags & ~PFDESC_SCTP_INIT))
2134 return (PF_DROP);
2135 if ((pd->sctp_flags & PFDESC_SCTP_INIT_ACK) &&
2136 (pd->sctp_flags & ~PFDESC_SCTP_INIT_ACK))
2137 return (PF_DROP);
2138 if ((pd->sctp_flags & PFDESC_SCTP_SHUTDOWN_COMPLETE) &&
2139 (pd->sctp_flags & ~PFDESC_SCTP_SHUTDOWN_COMPLETE))
2140 return (PF_DROP);
2141 if ((pd->sctp_flags & PFDESC_SCTP_ABORT) &&
2142 (pd->sctp_flags & PFDESC_SCTP_DATA)) {
2143 /*
2144 * RFC4960 3.3.7: DATA chunks MUST NOT be
2145 * bundled with ABORT.
2146 */
2147 return (PF_DROP);
2148 }
2149
2150 return (PF_PASS);
2151 }
2152
2153 int
pf_normalize_sctp(struct pf_pdesc * pd)2154 pf_normalize_sctp(struct pf_pdesc *pd)
2155 {
2156 struct pf_krule *r, *rm = NULL;
2157 struct sctphdr *sh = &pd->hdr.sctp;
2158 u_short reason;
2159 sa_family_t af = pd->af;
2160 int srs;
2161
2162 PF_RULES_RASSERT();
2163
2164 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
2165 /* Check if there any scrub rules. Lack of scrub rules means enforced
2166 * packet normalization operation just like in OpenBSD. */
2167 srs = (r != NULL);
2168 while (r != NULL) {
2169 pf_counter_u64_add(&r->evaluations, 1);
2170 if (pfi_kkif_match(r->kif, pd->kif) == r->ifnot)
2171 r = r->skip[PF_SKIP_IFP];
2172 else if (r->direction && r->direction != pd->dir)
2173 r = r->skip[PF_SKIP_DIR];
2174 else if (r->af && r->af != af)
2175 r = r->skip[PF_SKIP_AF];
2176 else if (r->proto && r->proto != pd->proto)
2177 r = r->skip[PF_SKIP_PROTO];
2178 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
2179 r->src.neg, pd->kif, M_GETFIB(pd->m)))
2180 r = r->skip[PF_SKIP_SRC_ADDR];
2181 else if (r->src.port_op && !pf_match_port(r->src.port_op,
2182 r->src.port[0], r->src.port[1], sh->src_port))
2183 r = r->skip[PF_SKIP_SRC_PORT];
2184 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
2185 r->dst.neg, NULL, M_GETFIB(pd->m)))
2186 r = r->skip[PF_SKIP_DST_ADDR];
2187 else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
2188 r->dst.port[0], r->dst.port[1], sh->dest_port))
2189 r = r->skip[PF_SKIP_DST_PORT];
2190 else {
2191 rm = r;
2192 break;
2193 }
2194 }
2195
2196 if (srs) {
2197 /* With scrub rules present SCTP normalization happens only
2198 * if one of rules has matched and it's not a "no scrub" rule */
2199 if (rm == NULL || rm->action == PF_NOSCRUB)
2200 return (PF_PASS);
2201
2202 pf_counter_u64_critical_enter();
2203 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1);
2204 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len);
2205 pf_counter_u64_critical_exit();
2206 }
2207
2208 /* Verify we're a multiple of 4 bytes long */
2209 if ((pd->tot_len - pd->off - sizeof(struct sctphdr)) % 4)
2210 goto sctp_drop;
2211
2212 /* INIT chunk needs to be the only chunk */
2213 if (pd->sctp_flags & PFDESC_SCTP_INIT)
2214 if (pd->sctp_flags & ~PFDESC_SCTP_INIT)
2215 goto sctp_drop;
2216
2217 return (PF_PASS);
2218
2219 sctp_drop:
2220 REASON_SET(&reason, PFRES_NORM);
2221 if (rm != NULL && r->log)
2222 PFLOG_PACKET(PF_DROP, reason, r, NULL, NULL, pd,
2223 1, NULL);
2224
2225 return (PF_DROP);
2226 }
2227
2228 #if defined(INET) || defined(INET6)
2229 void
pf_scrub(struct pf_pdesc * pd)2230 pf_scrub(struct pf_pdesc *pd)
2231 {
2232
2233 struct ip *h = mtod(pd->m, struct ip *);
2234 #ifdef INET6
2235 struct ip6_hdr *h6 = mtod(pd->m, struct ip6_hdr *);
2236 #endif /* INET6 */
2237
2238 /* Clear IP_DF if no-df was requested */
2239 if (pd->af == AF_INET && pd->act.flags & PFSTATE_NODF &&
2240 h->ip_off & htons(IP_DF))
2241 {
2242 u_int16_t ip_off = h->ip_off;
2243
2244 h->ip_off &= htons(~IP_DF);
2245 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
2246 }
2247
2248 /* Enforce a minimum ttl, may cause endless packet loops */
2249 if (pd->af == AF_INET && pd->act.min_ttl &&
2250 h->ip_ttl < pd->act.min_ttl) {
2251 u_int16_t ip_ttl = h->ip_ttl;
2252
2253 pd->ttl = h->ip_ttl = pd->act.min_ttl;
2254 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_ttl, h->ip_ttl, 0);
2255 }
2256 #ifdef INET6
2257 /* Enforce a minimum ttl, may cause endless packet loops */
2258 if (pd->af == AF_INET6 && pd->act.min_ttl &&
2259 h6->ip6_hlim < pd->act.min_ttl)
2260 pd->ttl = h6->ip6_hlim = pd->act.min_ttl;
2261 #endif /* INET6 */
2262 /* Enforce tos */
2263 if (pd->act.flags & PFSTATE_SETTOS) {
2264 switch (pd->af) {
2265 case AF_INET: {
2266 u_int16_t ov, nv;
2267
2268 ov = *(u_int16_t *)h;
2269 h->ip_tos = pd->act.set_tos | (h->ip_tos & IPTOS_ECN_MASK);
2270 pd->tos = h->ip_tos & ~IPTOS_ECN_MASK;
2271 nv = *(u_int16_t *)h;
2272
2273 h->ip_sum = pf_cksum_fixup(h->ip_sum, ov, nv, 0);
2274 break;
2275 }
2276 #ifdef INET6
2277 case AF_INET6:
2278 h6->ip6_flow &= IPV6_FLOWLABEL_MASK | IPV6_VERSION_MASK;
2279 h6->ip6_flow |= htonl((pd->act.set_tos | IPV6_ECN(h6)) << 20);
2280 pd->tos = IPV6_DSCP(h6);
2281 break;
2282 #endif /* INET6 */
2283 }
2284 }
2285
2286 /* random-id, but not for fragments */
2287 #ifdef INET
2288 if (pd->af == AF_INET &&
2289 pd->act.flags & PFSTATE_RANDOMID && !(h->ip_off & ~htons(IP_DF))) {
2290 uint16_t ip_id = h->ip_id;
2291
2292 ip_fillid(h, V_ip_random_id);
2293 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_id, h->ip_id, 0);
2294 }
2295 #endif /* INET */
2296 }
2297 #endif /* INET || INET6 */
2298