1 /*-
2 * SPDX-License-Identifier: BSD-2-Clause
3 *
4 * Copyright 2001 Niels Provos <provos@citi.umich.edu>
5 * Copyright 2011-2018 Alexander Bluhm <bluhm@openbsd.org>
6 * All rights reserved.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27 *
28 * $OpenBSD: pf_norm.c,v 1.114 2009/01/29 14:11:45 henning Exp $
29 */
30
31 #include <sys/cdefs.h>
32 #include "opt_inet.h"
33 #include "opt_inet6.h"
34 #include "opt_pf.h"
35
36 #include <sys/param.h>
37 #include <sys/kernel.h>
38 #include <sys/lock.h>
39 #include <sys/mbuf.h>
40 #include <sys/mutex.h>
41 #include <sys/refcount.h>
42 #include <sys/socket.h>
43
44 #include <net/if.h>
45 #include <net/vnet.h>
46 #include <net/pfvar.h>
47 #include <net/if_pflog.h>
48
49 #include <netinet/in.h>
50 #include <netinet/ip.h>
51 #include <netinet/ip_var.h>
52 #include <netinet6/ip6_var.h>
53 #include <netinet6/scope6_var.h>
54 #include <netinet/tcp.h>
55 #include <netinet/tcp_fsm.h>
56 #include <netinet/tcp_seq.h>
57 #include <netinet/sctp_constants.h>
58 #include <netinet/sctp_header.h>
59
60 #ifdef INET6
61 #include <netinet/ip6.h>
62 #endif /* INET6 */
63
64 struct pf_frent {
65 TAILQ_ENTRY(pf_frent) fr_next;
66 struct mbuf *fe_m;
67 uint16_t fe_hdrlen; /* ipv4 header length with ip options
68 ipv6, extension, fragment header */
69 uint16_t fe_extoff; /* last extension header offset or 0 */
70 uint16_t fe_len; /* fragment length */
71 uint16_t fe_off; /* fragment offset */
72 uint16_t fe_mff; /* more fragment flag */
73 };
74
75 struct pf_fragment_cmp {
76 struct pf_addr frc_src;
77 struct pf_addr frc_dst;
78 uint32_t frc_id;
79 sa_family_t frc_af;
80 uint8_t frc_proto;
81 };
82
83 struct pf_fragment {
84 struct pf_fragment_cmp fr_key;
85 #define fr_src fr_key.frc_src
86 #define fr_dst fr_key.frc_dst
87 #define fr_id fr_key.frc_id
88 #define fr_af fr_key.frc_af
89 #define fr_proto fr_key.frc_proto
90
91 /* pointers to queue element */
92 struct pf_frent *fr_firstoff[PF_FRAG_ENTRY_POINTS];
93 /* count entries between pointers */
94 uint8_t fr_entries[PF_FRAG_ENTRY_POINTS];
95 RB_ENTRY(pf_fragment) fr_entry;
96 TAILQ_ENTRY(pf_fragment) frag_next;
97 uint32_t fr_timeout;
98 uint16_t fr_maxlen; /* maximum length of single fragment */
99 u_int16_t fr_holes; /* number of holes in the queue */
100 TAILQ_HEAD(pf_fragq, pf_frent) fr_queue;
101 };
102
103 struct pf_fragment_tag {
104 uint16_t ft_hdrlen; /* header length of reassembled pkt */
105 uint16_t ft_extoff; /* last extension header offset or 0 */
106 uint16_t ft_maxlen; /* maximum fragment payload length */
107 uint32_t ft_id; /* fragment id */
108 };
109
110 VNET_DEFINE_STATIC(struct mtx, pf_frag_mtx);
111 #define V_pf_frag_mtx VNET(pf_frag_mtx)
112 #define PF_FRAG_LOCK() mtx_lock(&V_pf_frag_mtx)
113 #define PF_FRAG_UNLOCK() mtx_unlock(&V_pf_frag_mtx)
114 #define PF_FRAG_ASSERT() mtx_assert(&V_pf_frag_mtx, MA_OWNED)
115
116 VNET_DEFINE(uma_zone_t, pf_state_scrub_z); /* XXX: shared with pfsync */
117
118 VNET_DEFINE_STATIC(uma_zone_t, pf_frent_z);
119 #define V_pf_frent_z VNET(pf_frent_z)
120 VNET_DEFINE_STATIC(uma_zone_t, pf_frag_z);
121 #define V_pf_frag_z VNET(pf_frag_z)
122
123 TAILQ_HEAD(pf_fragqueue, pf_fragment);
124 TAILQ_HEAD(pf_cachequeue, pf_fragment);
125 VNET_DEFINE_STATIC(struct pf_fragqueue, pf_fragqueue);
126 #define V_pf_fragqueue VNET(pf_fragqueue)
127 RB_HEAD(pf_frag_tree, pf_fragment);
128 VNET_DEFINE_STATIC(struct pf_frag_tree, pf_frag_tree);
129 #define V_pf_frag_tree VNET(pf_frag_tree)
130 static int pf_frag_compare(struct pf_fragment *,
131 struct pf_fragment *);
132 static RB_PROTOTYPE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
133 static RB_GENERATE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
134
135 static void pf_flush_fragments(void);
136 static void pf_free_fragment(struct pf_fragment *);
137 static void pf_remove_fragment(struct pf_fragment *);
138
139 static struct pf_frent *pf_create_fragment(u_short *);
140 static int pf_frent_holes(struct pf_frent *frent);
141 static struct pf_fragment *pf_find_fragment(struct pf_fragment_cmp *key,
142 struct pf_frag_tree *tree);
143 static inline int pf_frent_index(struct pf_frent *);
144 static int pf_frent_insert(struct pf_fragment *,
145 struct pf_frent *, struct pf_frent *);
146 void pf_frent_remove(struct pf_fragment *,
147 struct pf_frent *);
148 struct pf_frent *pf_frent_previous(struct pf_fragment *,
149 struct pf_frent *);
150 static struct pf_fragment *pf_fillup_fragment(struct pf_fragment_cmp *,
151 struct pf_frent *, u_short *);
152 static struct mbuf *pf_join_fragment(struct pf_fragment *);
153 #ifdef INET
154 static int pf_reassemble(struct mbuf **, int, u_short *);
155 #endif /* INET */
156 #ifdef INET6
157 static int pf_reassemble6(struct mbuf **,
158 struct ip6_frag *, uint16_t, uint16_t, u_short *);
159 #endif /* INET6 */
160
161 #define DPFPRINTF(x) do { \
162 if (V_pf_status.debug >= PF_DEBUG_MISC) { \
163 printf("%s: ", __func__); \
164 printf x ; \
165 } \
166 } while(0)
167
168 #ifdef INET
169 static void
pf_ip2key(struct ip * ip,int dir,struct pf_fragment_cmp * key)170 pf_ip2key(struct ip *ip, int dir, struct pf_fragment_cmp *key)
171 {
172
173 key->frc_src.v4 = ip->ip_src;
174 key->frc_dst.v4 = ip->ip_dst;
175 key->frc_af = AF_INET;
176 key->frc_proto = ip->ip_p;
177 key->frc_id = ip->ip_id;
178 }
179 #endif /* INET */
180
181 void
pf_normalize_init(void)182 pf_normalize_init(void)
183 {
184
185 V_pf_frag_z = uma_zcreate("pf frags", sizeof(struct pf_fragment),
186 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
187 V_pf_frent_z = uma_zcreate("pf frag entries", sizeof(struct pf_frent),
188 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
189 V_pf_state_scrub_z = uma_zcreate("pf state scrubs",
190 sizeof(struct pf_state_scrub), NULL, NULL, NULL, NULL,
191 UMA_ALIGN_PTR, 0);
192
193 mtx_init(&V_pf_frag_mtx, "pf fragments", NULL, MTX_DEF);
194
195 V_pf_limits[PF_LIMIT_FRAGS].zone = V_pf_frent_z;
196 V_pf_limits[PF_LIMIT_FRAGS].limit = PFFRAG_FRENT_HIWAT;
197 uma_zone_set_max(V_pf_frent_z, PFFRAG_FRENT_HIWAT);
198 uma_zone_set_warning(V_pf_frent_z, "PF frag entries limit reached");
199
200 TAILQ_INIT(&V_pf_fragqueue);
201 }
202
203 void
pf_normalize_cleanup(void)204 pf_normalize_cleanup(void)
205 {
206
207 uma_zdestroy(V_pf_state_scrub_z);
208 uma_zdestroy(V_pf_frent_z);
209 uma_zdestroy(V_pf_frag_z);
210
211 mtx_destroy(&V_pf_frag_mtx);
212 }
213
214 static int
pf_frag_compare(struct pf_fragment * a,struct pf_fragment * b)215 pf_frag_compare(struct pf_fragment *a, struct pf_fragment *b)
216 {
217 int diff;
218
219 if ((diff = a->fr_id - b->fr_id) != 0)
220 return (diff);
221 if ((diff = a->fr_proto - b->fr_proto) != 0)
222 return (diff);
223 if ((diff = a->fr_af - b->fr_af) != 0)
224 return (diff);
225 if ((diff = pf_addr_cmp(&a->fr_src, &b->fr_src, a->fr_af)) != 0)
226 return (diff);
227 if ((diff = pf_addr_cmp(&a->fr_dst, &b->fr_dst, a->fr_af)) != 0)
228 return (diff);
229 return (0);
230 }
231
232 void
pf_purge_expired_fragments(void)233 pf_purge_expired_fragments(void)
234 {
235 u_int32_t expire = time_uptime -
236 V_pf_default_rule.timeout[PFTM_FRAG];
237
238 pf_purge_fragments(expire);
239 }
240
241 void
pf_purge_fragments(uint32_t expire)242 pf_purge_fragments(uint32_t expire)
243 {
244 struct pf_fragment *frag;
245
246 PF_FRAG_LOCK();
247 while ((frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue)) != NULL) {
248 if (frag->fr_timeout > expire)
249 break;
250
251 DPFPRINTF(("expiring %d(%p)\n", frag->fr_id, frag));
252 pf_free_fragment(frag);
253 }
254
255 PF_FRAG_UNLOCK();
256 }
257
258 /*
259 * Try to flush old fragments to make space for new ones
260 */
261 static void
pf_flush_fragments(void)262 pf_flush_fragments(void)
263 {
264 struct pf_fragment *frag;
265 int goal;
266
267 PF_FRAG_ASSERT();
268
269 goal = uma_zone_get_cur(V_pf_frent_z) * 9 / 10;
270 DPFPRINTF(("trying to free %d frag entriess\n", goal));
271 while (goal < uma_zone_get_cur(V_pf_frent_z)) {
272 frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue);
273 if (frag)
274 pf_free_fragment(frag);
275 else
276 break;
277 }
278 }
279
280 /* Frees the fragments and all associated entries */
281 static void
pf_free_fragment(struct pf_fragment * frag)282 pf_free_fragment(struct pf_fragment *frag)
283 {
284 struct pf_frent *frent;
285
286 PF_FRAG_ASSERT();
287
288 /* Free all fragments */
289 for (frent = TAILQ_FIRST(&frag->fr_queue); frent;
290 frent = TAILQ_FIRST(&frag->fr_queue)) {
291 TAILQ_REMOVE(&frag->fr_queue, frent, fr_next);
292
293 m_freem(frent->fe_m);
294 uma_zfree(V_pf_frent_z, frent);
295 }
296
297 pf_remove_fragment(frag);
298 }
299
300 static struct pf_fragment *
pf_find_fragment(struct pf_fragment_cmp * key,struct pf_frag_tree * tree)301 pf_find_fragment(struct pf_fragment_cmp *key, struct pf_frag_tree *tree)
302 {
303 struct pf_fragment *frag;
304
305 PF_FRAG_ASSERT();
306
307 frag = RB_FIND(pf_frag_tree, tree, (struct pf_fragment *)key);
308 if (frag != NULL) {
309 /* XXX Are we sure we want to update the timeout? */
310 frag->fr_timeout = time_uptime;
311 TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next);
312 TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next);
313 }
314
315 return (frag);
316 }
317
318 /* Removes a fragment from the fragment queue and frees the fragment */
319 static void
pf_remove_fragment(struct pf_fragment * frag)320 pf_remove_fragment(struct pf_fragment *frag)
321 {
322
323 PF_FRAG_ASSERT();
324 KASSERT(frag, ("frag != NULL"));
325
326 RB_REMOVE(pf_frag_tree, &V_pf_frag_tree, frag);
327 TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next);
328 uma_zfree(V_pf_frag_z, frag);
329 }
330
331 static struct pf_frent *
pf_create_fragment(u_short * reason)332 pf_create_fragment(u_short *reason)
333 {
334 struct pf_frent *frent;
335
336 PF_FRAG_ASSERT();
337
338 frent = uma_zalloc(V_pf_frent_z, M_NOWAIT);
339 if (frent == NULL) {
340 pf_flush_fragments();
341 frent = uma_zalloc(V_pf_frent_z, M_NOWAIT);
342 if (frent == NULL) {
343 REASON_SET(reason, PFRES_MEMORY);
344 return (NULL);
345 }
346 }
347
348 return (frent);
349 }
350
351 /*
352 * Calculate the additional holes that were created in the fragment
353 * queue by inserting this fragment. A fragment in the middle
354 * creates one more hole by splitting. For each connected side,
355 * it loses one hole.
356 * Fragment entry must be in the queue when calling this function.
357 */
358 static int
pf_frent_holes(struct pf_frent * frent)359 pf_frent_holes(struct pf_frent *frent)
360 {
361 struct pf_frent *prev = TAILQ_PREV(frent, pf_fragq, fr_next);
362 struct pf_frent *next = TAILQ_NEXT(frent, fr_next);
363 int holes = 1;
364
365 if (prev == NULL) {
366 if (frent->fe_off == 0)
367 holes--;
368 } else {
369 KASSERT(frent->fe_off != 0, ("frent->fe_off != 0"));
370 if (frent->fe_off == prev->fe_off + prev->fe_len)
371 holes--;
372 }
373 if (next == NULL) {
374 if (!frent->fe_mff)
375 holes--;
376 } else {
377 KASSERT(frent->fe_mff, ("frent->fe_mff"));
378 if (next->fe_off == frent->fe_off + frent->fe_len)
379 holes--;
380 }
381 return holes;
382 }
383
384 static inline int
pf_frent_index(struct pf_frent * frent)385 pf_frent_index(struct pf_frent *frent)
386 {
387 /*
388 * We have an array of 16 entry points to the queue. A full size
389 * 65535 octet IP packet can have 8192 fragments. So the queue
390 * traversal length is at most 512 and at most 16 entry points are
391 * checked. We need 128 additional bytes on a 64 bit architecture.
392 */
393 CTASSERT(((u_int16_t)0xffff &~ 7) / (0x10000 / PF_FRAG_ENTRY_POINTS) ==
394 16 - 1);
395 CTASSERT(((u_int16_t)0xffff >> 3) / PF_FRAG_ENTRY_POINTS == 512 - 1);
396
397 return frent->fe_off / (0x10000 / PF_FRAG_ENTRY_POINTS);
398 }
399
400 static int
pf_frent_insert(struct pf_fragment * frag,struct pf_frent * frent,struct pf_frent * prev)401 pf_frent_insert(struct pf_fragment *frag, struct pf_frent *frent,
402 struct pf_frent *prev)
403 {
404 int index;
405
406 CTASSERT(PF_FRAG_ENTRY_LIMIT <= 0xff);
407
408 /*
409 * A packet has at most 65536 octets. With 16 entry points, each one
410 * spawns 4096 octets. We limit these to 64 fragments each, which
411 * means on average every fragment must have at least 64 octets.
412 */
413 index = pf_frent_index(frent);
414 if (frag->fr_entries[index] >= PF_FRAG_ENTRY_LIMIT)
415 return ENOBUFS;
416 frag->fr_entries[index]++;
417
418 if (prev == NULL) {
419 TAILQ_INSERT_HEAD(&frag->fr_queue, frent, fr_next);
420 } else {
421 KASSERT(prev->fe_off + prev->fe_len <= frent->fe_off,
422 ("overlapping fragment"));
423 TAILQ_INSERT_AFTER(&frag->fr_queue, prev, frent, fr_next);
424 }
425
426 if (frag->fr_firstoff[index] == NULL) {
427 KASSERT(prev == NULL || pf_frent_index(prev) < index,
428 ("prev == NULL || pf_frent_index(pref) < index"));
429 frag->fr_firstoff[index] = frent;
430 } else {
431 if (frent->fe_off < frag->fr_firstoff[index]->fe_off) {
432 KASSERT(prev == NULL || pf_frent_index(prev) < index,
433 ("prev == NULL || pf_frent_index(pref) < index"));
434 frag->fr_firstoff[index] = frent;
435 } else {
436 KASSERT(prev != NULL, ("prev != NULL"));
437 KASSERT(pf_frent_index(prev) == index,
438 ("pf_frent_index(prev) == index"));
439 }
440 }
441
442 frag->fr_holes += pf_frent_holes(frent);
443
444 return 0;
445 }
446
447 void
pf_frent_remove(struct pf_fragment * frag,struct pf_frent * frent)448 pf_frent_remove(struct pf_fragment *frag, struct pf_frent *frent)
449 {
450 #ifdef INVARIANTS
451 struct pf_frent *prev = TAILQ_PREV(frent, pf_fragq, fr_next);
452 #endif
453 struct pf_frent *next = TAILQ_NEXT(frent, fr_next);
454 int index;
455
456 frag->fr_holes -= pf_frent_holes(frent);
457
458 index = pf_frent_index(frent);
459 KASSERT(frag->fr_firstoff[index] != NULL, ("frent not found"));
460 if (frag->fr_firstoff[index]->fe_off == frent->fe_off) {
461 if (next == NULL) {
462 frag->fr_firstoff[index] = NULL;
463 } else {
464 KASSERT(frent->fe_off + frent->fe_len <= next->fe_off,
465 ("overlapping fragment"));
466 if (pf_frent_index(next) == index) {
467 frag->fr_firstoff[index] = next;
468 } else {
469 frag->fr_firstoff[index] = NULL;
470 }
471 }
472 } else {
473 KASSERT(frag->fr_firstoff[index]->fe_off < frent->fe_off,
474 ("frag->fr_firstoff[index]->fe_off < frent->fe_off"));
475 KASSERT(prev != NULL, ("prev != NULL"));
476 KASSERT(prev->fe_off + prev->fe_len <= frent->fe_off,
477 ("overlapping fragment"));
478 KASSERT(pf_frent_index(prev) == index,
479 ("pf_frent_index(prev) == index"));
480 }
481
482 TAILQ_REMOVE(&frag->fr_queue, frent, fr_next);
483
484 KASSERT(frag->fr_entries[index] > 0, ("No fragments remaining"));
485 frag->fr_entries[index]--;
486 }
487
488 struct pf_frent *
pf_frent_previous(struct pf_fragment * frag,struct pf_frent * frent)489 pf_frent_previous(struct pf_fragment *frag, struct pf_frent *frent)
490 {
491 struct pf_frent *prev, *next;
492 int index;
493
494 /*
495 * If there are no fragments after frag, take the final one. Assume
496 * that the global queue is not empty.
497 */
498 prev = TAILQ_LAST(&frag->fr_queue, pf_fragq);
499 KASSERT(prev != NULL, ("prev != NULL"));
500 if (prev->fe_off <= frent->fe_off)
501 return prev;
502 /*
503 * We want to find a fragment entry that is before frag, but still
504 * close to it. Find the first fragment entry that is in the same
505 * entry point or in the first entry point after that. As we have
506 * already checked that there are entries behind frag, this will
507 * succeed.
508 */
509 for (index = pf_frent_index(frent); index < PF_FRAG_ENTRY_POINTS;
510 index++) {
511 prev = frag->fr_firstoff[index];
512 if (prev != NULL)
513 break;
514 }
515 KASSERT(prev != NULL, ("prev != NULL"));
516 /*
517 * In prev we may have a fragment from the same entry point that is
518 * before frent, or one that is just one position behind frent.
519 * In the latter case, we go back one step and have the predecessor.
520 * There may be none if the new fragment will be the first one.
521 */
522 if (prev->fe_off > frent->fe_off) {
523 prev = TAILQ_PREV(prev, pf_fragq, fr_next);
524 if (prev == NULL)
525 return NULL;
526 KASSERT(prev->fe_off <= frent->fe_off,
527 ("prev->fe_off <= frent->fe_off"));
528 return prev;
529 }
530 /*
531 * In prev is the first fragment of the entry point. The offset
532 * of frag is behind it. Find the closest previous fragment.
533 */
534 for (next = TAILQ_NEXT(prev, fr_next); next != NULL;
535 next = TAILQ_NEXT(next, fr_next)) {
536 if (next->fe_off > frent->fe_off)
537 break;
538 prev = next;
539 }
540 return prev;
541 }
542
543 static struct pf_fragment *
pf_fillup_fragment(struct pf_fragment_cmp * key,struct pf_frent * frent,u_short * reason)544 pf_fillup_fragment(struct pf_fragment_cmp *key, struct pf_frent *frent,
545 u_short *reason)
546 {
547 struct pf_frent *after, *next, *prev;
548 struct pf_fragment *frag;
549 uint16_t total;
550 int old_index, new_index;
551
552 PF_FRAG_ASSERT();
553
554 /* No empty fragments. */
555 if (frent->fe_len == 0) {
556 DPFPRINTF(("bad fragment: len 0\n"));
557 goto bad_fragment;
558 }
559
560 /* All fragments are 8 byte aligned. */
561 if (frent->fe_mff && (frent->fe_len & 0x7)) {
562 DPFPRINTF(("bad fragment: mff and len %d\n", frent->fe_len));
563 goto bad_fragment;
564 }
565
566 /* Respect maximum length, IP_MAXPACKET == IPV6_MAXPACKET. */
567 if (frent->fe_off + frent->fe_len > IP_MAXPACKET) {
568 DPFPRINTF(("bad fragment: max packet %d\n",
569 frent->fe_off + frent->fe_len));
570 goto bad_fragment;
571 }
572
573 DPFPRINTF((key->frc_af == AF_INET ?
574 "reass frag %d @ %d-%d\n" : "reass frag %#08x @ %d-%d\n",
575 key->frc_id, frent->fe_off, frent->fe_off + frent->fe_len));
576
577 /* Fully buffer all of the fragments in this fragment queue. */
578 frag = pf_find_fragment(key, &V_pf_frag_tree);
579
580 /* Create a new reassembly queue for this packet. */
581 if (frag == NULL) {
582 frag = uma_zalloc(V_pf_frag_z, M_NOWAIT);
583 if (frag == NULL) {
584 pf_flush_fragments();
585 frag = uma_zalloc(V_pf_frag_z, M_NOWAIT);
586 if (frag == NULL) {
587 REASON_SET(reason, PFRES_MEMORY);
588 goto drop_fragment;
589 }
590 }
591
592 *(struct pf_fragment_cmp *)frag = *key;
593 memset(frag->fr_firstoff, 0, sizeof(frag->fr_firstoff));
594 memset(frag->fr_entries, 0, sizeof(frag->fr_entries));
595 frag->fr_timeout = time_uptime;
596 frag->fr_maxlen = frent->fe_len;
597 frag->fr_holes = 1;
598 TAILQ_INIT(&frag->fr_queue);
599
600 RB_INSERT(pf_frag_tree, &V_pf_frag_tree, frag);
601 TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next);
602
603 /* We do not have a previous fragment, cannot fail. */
604 pf_frent_insert(frag, frent, NULL);
605
606 return (frag);
607 }
608
609 KASSERT(!TAILQ_EMPTY(&frag->fr_queue), ("!TAILQ_EMPTY()->fr_queue"));
610
611 /* Remember maximum fragment len for refragmentation. */
612 if (frent->fe_len > frag->fr_maxlen)
613 frag->fr_maxlen = frent->fe_len;
614
615 /* Maximum data we have seen already. */
616 total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
617 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
618
619 /* Non terminal fragments must have more fragments flag. */
620 if (frent->fe_off + frent->fe_len < total && !frent->fe_mff)
621 goto bad_fragment;
622
623 /* Check if we saw the last fragment already. */
624 if (!TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_mff) {
625 if (frent->fe_off + frent->fe_len > total ||
626 (frent->fe_off + frent->fe_len == total && frent->fe_mff))
627 goto bad_fragment;
628 } else {
629 if (frent->fe_off + frent->fe_len == total && !frent->fe_mff)
630 goto bad_fragment;
631 }
632
633 /* Find neighbors for newly inserted fragment */
634 prev = pf_frent_previous(frag, frent);
635 if (prev == NULL) {
636 after = TAILQ_FIRST(&frag->fr_queue);
637 KASSERT(after != NULL, ("after != NULL"));
638 } else {
639 after = TAILQ_NEXT(prev, fr_next);
640 }
641
642 if (prev != NULL && prev->fe_off + prev->fe_len > frent->fe_off) {
643 uint16_t precut;
644
645 precut = prev->fe_off + prev->fe_len - frent->fe_off;
646 if (precut >= frent->fe_len)
647 goto bad_fragment;
648 DPFPRINTF(("overlap -%d\n", precut));
649 m_adj(frent->fe_m, precut);
650 frent->fe_off += precut;
651 frent->fe_len -= precut;
652 }
653
654 for (; after != NULL && frent->fe_off + frent->fe_len > after->fe_off;
655 after = next) {
656 uint16_t aftercut;
657
658 aftercut = frent->fe_off + frent->fe_len - after->fe_off;
659 DPFPRINTF(("adjust overlap %d\n", aftercut));
660 if (aftercut < after->fe_len) {
661 m_adj(after->fe_m, aftercut);
662 old_index = pf_frent_index(after);
663 after->fe_off += aftercut;
664 after->fe_len -= aftercut;
665 new_index = pf_frent_index(after);
666 if (old_index != new_index) {
667 DPFPRINTF(("frag index %d, new %d",
668 old_index, new_index));
669 /* Fragment switched queue as fe_off changed */
670 after->fe_off -= aftercut;
671 after->fe_len += aftercut;
672 /* Remove restored fragment from old queue */
673 pf_frent_remove(frag, after);
674 after->fe_off += aftercut;
675 after->fe_len -= aftercut;
676 /* Insert into correct queue */
677 if (pf_frent_insert(frag, after, prev)) {
678 DPFPRINTF(
679 ("fragment requeue limit exceeded"));
680 m_freem(after->fe_m);
681 uma_zfree(V_pf_frent_z, after);
682 /* There is not way to recover */
683 goto bad_fragment;
684 }
685 }
686 break;
687 }
688
689 /* This fragment is completely overlapped, lose it. */
690 next = TAILQ_NEXT(after, fr_next);
691 pf_frent_remove(frag, after);
692 m_freem(after->fe_m);
693 uma_zfree(V_pf_frent_z, after);
694 }
695
696 /* If part of the queue gets too long, there is not way to recover. */
697 if (pf_frent_insert(frag, frent, prev)) {
698 DPFPRINTF(("fragment queue limit exceeded\n"));
699 goto bad_fragment;
700 }
701
702 return (frag);
703
704 bad_fragment:
705 REASON_SET(reason, PFRES_FRAG);
706 drop_fragment:
707 uma_zfree(V_pf_frent_z, frent);
708 return (NULL);
709 }
710
711 static struct mbuf *
pf_join_fragment(struct pf_fragment * frag)712 pf_join_fragment(struct pf_fragment *frag)
713 {
714 struct mbuf *m, *m2;
715 struct pf_frent *frent, *next;
716
717 frent = TAILQ_FIRST(&frag->fr_queue);
718 next = TAILQ_NEXT(frent, fr_next);
719
720 m = frent->fe_m;
721 m_adj(m, (frent->fe_hdrlen + frent->fe_len) - m->m_pkthdr.len);
722 uma_zfree(V_pf_frent_z, frent);
723 for (frent = next; frent != NULL; frent = next) {
724 next = TAILQ_NEXT(frent, fr_next);
725
726 m2 = frent->fe_m;
727 /* Strip off ip header. */
728 m_adj(m2, frent->fe_hdrlen);
729 /* Strip off any trailing bytes. */
730 m_adj(m2, frent->fe_len - m2->m_pkthdr.len);
731
732 uma_zfree(V_pf_frent_z, frent);
733 m_cat(m, m2);
734 }
735
736 /* Remove from fragment queue. */
737 pf_remove_fragment(frag);
738
739 return (m);
740 }
741
742 #ifdef INET
743 static int
pf_reassemble(struct mbuf ** m0,int dir,u_short * reason)744 pf_reassemble(struct mbuf **m0, int dir, u_short *reason)
745 {
746 struct mbuf *m = *m0;
747 struct ip *ip = mtod(m, struct ip *);
748 struct pf_frent *frent;
749 struct pf_fragment *frag;
750 struct pf_fragment_cmp key;
751 uint16_t total, hdrlen;
752
753 /* Get an entry for the fragment queue */
754 if ((frent = pf_create_fragment(reason)) == NULL)
755 return (PF_DROP);
756
757 frent->fe_m = m;
758 frent->fe_hdrlen = ip->ip_hl << 2;
759 frent->fe_extoff = 0;
760 frent->fe_len = ntohs(ip->ip_len) - (ip->ip_hl << 2);
761 frent->fe_off = (ntohs(ip->ip_off) & IP_OFFMASK) << 3;
762 frent->fe_mff = ntohs(ip->ip_off) & IP_MF;
763
764 pf_ip2key(ip, dir, &key);
765
766 if ((frag = pf_fillup_fragment(&key, frent, reason)) == NULL)
767 return (PF_DROP);
768
769 /* The mbuf is part of the fragment entry, no direct free or access */
770 m = *m0 = NULL;
771
772 if (frag->fr_holes) {
773 DPFPRINTF(("frag %d, holes %d\n", frag->fr_id, frag->fr_holes));
774 return (PF_PASS); /* drop because *m0 is NULL, no error */
775 }
776
777 /* We have all the data */
778 frent = TAILQ_FIRST(&frag->fr_queue);
779 KASSERT(frent != NULL, ("frent != NULL"));
780 total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
781 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
782 hdrlen = frent->fe_hdrlen;
783
784 m = *m0 = pf_join_fragment(frag);
785 frag = NULL;
786
787 if (m->m_flags & M_PKTHDR) {
788 int plen = 0;
789 for (m = *m0; m; m = m->m_next)
790 plen += m->m_len;
791 m = *m0;
792 m->m_pkthdr.len = plen;
793 }
794
795 ip = mtod(m, struct ip *);
796 ip->ip_sum = pf_cksum_fixup(ip->ip_sum, ip->ip_len,
797 htons(hdrlen + total), 0);
798 ip->ip_len = htons(hdrlen + total);
799 ip->ip_sum = pf_cksum_fixup(ip->ip_sum, ip->ip_off,
800 ip->ip_off & ~(IP_MF|IP_OFFMASK), 0);
801 ip->ip_off &= ~(IP_MF|IP_OFFMASK);
802
803 if (hdrlen + total > IP_MAXPACKET) {
804 DPFPRINTF(("drop: too big: %d\n", total));
805 ip->ip_len = 0;
806 REASON_SET(reason, PFRES_SHORT);
807 /* PF_DROP requires a valid mbuf *m0 in pf_test() */
808 return (PF_DROP);
809 }
810
811 DPFPRINTF(("complete: %p(%d)\n", m, ntohs(ip->ip_len)));
812 return (PF_PASS);
813 }
814 #endif /* INET */
815
816 #ifdef INET6
817 static int
pf_reassemble6(struct mbuf ** m0,struct ip6_frag * fraghdr,uint16_t hdrlen,uint16_t extoff,u_short * reason)818 pf_reassemble6(struct mbuf **m0, struct ip6_frag *fraghdr,
819 uint16_t hdrlen, uint16_t extoff, u_short *reason)
820 {
821 struct mbuf *m = *m0;
822 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
823 struct pf_frent *frent;
824 struct pf_fragment *frag;
825 struct pf_fragment_cmp key;
826 struct m_tag *mtag;
827 struct pf_fragment_tag *ftag;
828 int off;
829 uint32_t frag_id;
830 uint16_t total, maxlen;
831 uint8_t proto;
832
833 PF_FRAG_LOCK();
834
835 /* Get an entry for the fragment queue. */
836 if ((frent = pf_create_fragment(reason)) == NULL) {
837 PF_FRAG_UNLOCK();
838 return (PF_DROP);
839 }
840
841 frent->fe_m = m;
842 frent->fe_hdrlen = hdrlen;
843 frent->fe_extoff = extoff;
844 frent->fe_len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - hdrlen;
845 frent->fe_off = ntohs(fraghdr->ip6f_offlg & IP6F_OFF_MASK);
846 frent->fe_mff = fraghdr->ip6f_offlg & IP6F_MORE_FRAG;
847
848 key.frc_src.v6 = ip6->ip6_src;
849 key.frc_dst.v6 = ip6->ip6_dst;
850 key.frc_af = AF_INET6;
851 /* Only the first fragment's protocol is relevant. */
852 key.frc_proto = 0;
853 key.frc_id = fraghdr->ip6f_ident;
854
855 if ((frag = pf_fillup_fragment(&key, frent, reason)) == NULL) {
856 PF_FRAG_UNLOCK();
857 return (PF_DROP);
858 }
859
860 /* The mbuf is part of the fragment entry, no direct free or access. */
861 m = *m0 = NULL;
862
863 if (frag->fr_holes) {
864 DPFPRINTF(("frag %d, holes %d\n", frag->fr_id,
865 frag->fr_holes));
866 PF_FRAG_UNLOCK();
867 return (PF_PASS); /* Drop because *m0 is NULL, no error. */
868 }
869
870 /* We have all the data. */
871 frent = TAILQ_FIRST(&frag->fr_queue);
872 KASSERT(frent != NULL, ("frent != NULL"));
873 extoff = frent->fe_extoff;
874 maxlen = frag->fr_maxlen;
875 frag_id = frag->fr_id;
876 total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
877 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
878 hdrlen = frent->fe_hdrlen - sizeof(struct ip6_frag);
879
880 m = *m0 = pf_join_fragment(frag);
881 frag = NULL;
882
883 PF_FRAG_UNLOCK();
884
885 /* Take protocol from first fragment header. */
886 m = m_getptr(m, hdrlen + offsetof(struct ip6_frag, ip6f_nxt), &off);
887 KASSERT(m, ("%s: short mbuf chain", __func__));
888 proto = *(mtod(m, uint8_t *) + off);
889 m = *m0;
890
891 /* Delete frag6 header */
892 if (ip6_deletefraghdr(m, hdrlen, M_NOWAIT) != 0)
893 goto fail;
894
895 if (m->m_flags & M_PKTHDR) {
896 int plen = 0;
897 for (m = *m0; m; m = m->m_next)
898 plen += m->m_len;
899 m = *m0;
900 m->m_pkthdr.len = plen;
901 }
902
903 if ((mtag = m_tag_get(PACKET_TAG_PF_REASSEMBLED,
904 sizeof(struct pf_fragment_tag), M_NOWAIT)) == NULL)
905 goto fail;
906 ftag = (struct pf_fragment_tag *)(mtag + 1);
907 ftag->ft_hdrlen = hdrlen;
908 ftag->ft_extoff = extoff;
909 ftag->ft_maxlen = maxlen;
910 ftag->ft_id = frag_id;
911 m_tag_prepend(m, mtag);
912
913 ip6 = mtod(m, struct ip6_hdr *);
914 ip6->ip6_plen = htons(hdrlen - sizeof(struct ip6_hdr) + total);
915 if (extoff) {
916 /* Write protocol into next field of last extension header. */
917 m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt),
918 &off);
919 KASSERT(m, ("%s: short mbuf chain", __func__));
920 *(mtod(m, char *) + off) = proto;
921 m = *m0;
922 } else
923 ip6->ip6_nxt = proto;
924
925 if (hdrlen - sizeof(struct ip6_hdr) + total > IPV6_MAXPACKET) {
926 DPFPRINTF(("drop: too big: %d\n", total));
927 ip6->ip6_plen = 0;
928 REASON_SET(reason, PFRES_SHORT);
929 /* PF_DROP requires a valid mbuf *m0 in pf_test6(). */
930 return (PF_DROP);
931 }
932
933 DPFPRINTF(("complete: %p(%d)\n", m, ntohs(ip6->ip6_plen)));
934 return (PF_PASS);
935
936 fail:
937 REASON_SET(reason, PFRES_MEMORY);
938 /* PF_DROP requires a valid mbuf *m0 in pf_test6(), will free later. */
939 return (PF_DROP);
940 }
941 #endif /* INET6 */
942
943 #ifdef INET6
944 int
pf_max_frag_size(struct mbuf * m)945 pf_max_frag_size(struct mbuf *m)
946 {
947 struct m_tag *tag;
948 struct pf_fragment_tag *ftag;
949
950 tag = m_tag_find(m, PACKET_TAG_PF_REASSEMBLED, NULL);
951 if (tag == NULL)
952 return (m->m_pkthdr.len);
953
954 ftag = (struct pf_fragment_tag *)(tag + 1);
955
956 return (ftag->ft_maxlen);
957 }
958
959 int
pf_refragment6(struct ifnet * ifp,struct mbuf ** m0,struct m_tag * mtag,bool forward)960 pf_refragment6(struct ifnet *ifp, struct mbuf **m0, struct m_tag *mtag,
961 bool forward)
962 {
963 struct mbuf *m = *m0, *t;
964 struct ip6_hdr *hdr;
965 struct pf_fragment_tag *ftag = (struct pf_fragment_tag *)(mtag + 1);
966 struct pf_pdesc pd;
967 uint32_t frag_id;
968 uint16_t hdrlen, extoff, maxlen;
969 uint8_t proto;
970 int error, action;
971
972 hdrlen = ftag->ft_hdrlen;
973 extoff = ftag->ft_extoff;
974 maxlen = ftag->ft_maxlen;
975 frag_id = ftag->ft_id;
976 m_tag_delete(m, mtag);
977 mtag = NULL;
978 ftag = NULL;
979
980 if (extoff) {
981 int off;
982
983 /* Use protocol from next field of last extension header */
984 m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt),
985 &off);
986 KASSERT((m != NULL), ("pf_refragment6: short mbuf chain"));
987 proto = *(mtod(m, uint8_t *) + off);
988 *(mtod(m, char *) + off) = IPPROTO_FRAGMENT;
989 m = *m0;
990 } else {
991 hdr = mtod(m, struct ip6_hdr *);
992 proto = hdr->ip6_nxt;
993 hdr->ip6_nxt = IPPROTO_FRAGMENT;
994 }
995
996 /* In case of link-local traffic we'll need a scope set. */
997 hdr = mtod(m, struct ip6_hdr *);
998
999 in6_setscope(&hdr->ip6_src, ifp, NULL);
1000 in6_setscope(&hdr->ip6_dst, ifp, NULL);
1001
1002 /* The MTU must be a multiple of 8 bytes, or we risk doing the
1003 * fragmentation wrong. */
1004 maxlen = maxlen & ~7;
1005
1006 /*
1007 * Maxlen may be less than 8 if there was only a single
1008 * fragment. As it was fragmented before, add a fragment
1009 * header also for a single fragment. If total or maxlen
1010 * is less than 8, ip6_fragment() will return EMSGSIZE and
1011 * we drop the packet.
1012 */
1013 error = ip6_fragment(ifp, m, hdrlen, proto, maxlen, frag_id);
1014 m = (*m0)->m_nextpkt;
1015 (*m0)->m_nextpkt = NULL;
1016 if (error == 0) {
1017 /* The first mbuf contains the unfragmented packet. */
1018 m_freem(*m0);
1019 *m0 = NULL;
1020 action = PF_PASS;
1021 } else {
1022 /* Drop expects an mbuf to free. */
1023 DPFPRINTF(("refragment error %d\n", error));
1024 action = PF_DROP;
1025 }
1026 for (; m; m = t) {
1027 t = m->m_nextpkt;
1028 m->m_nextpkt = NULL;
1029 m->m_flags |= M_SKIP_FIREWALL;
1030 memset(&pd, 0, sizeof(pd));
1031 pd.pf_mtag = pf_find_mtag(m);
1032 if (error == 0)
1033 if (forward) {
1034 MPASS(m->m_pkthdr.rcvif != NULL);
1035 ip6_forward(m, 0);
1036 } else {
1037 (void)ip6_output(m, NULL, NULL, 0, NULL, NULL,
1038 NULL);
1039 }
1040 else
1041 m_freem(m);
1042 }
1043
1044 return (action);
1045 }
1046 #endif /* INET6 */
1047
1048 #ifdef INET
1049 int
pf_normalize_ip(struct mbuf ** m0,u_short * reason,struct pf_pdesc * pd)1050 pf_normalize_ip(struct mbuf **m0, u_short *reason,
1051 struct pf_pdesc *pd)
1052 {
1053 struct pf_krule *r;
1054 struct ip *h = mtod(*m0, struct ip *);
1055 int mff = (ntohs(h->ip_off) & IP_MF);
1056 int hlen = h->ip_hl << 2;
1057 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
1058 u_int16_t max;
1059 int ip_len;
1060 int tag = -1;
1061 int verdict;
1062 bool scrub_compat;
1063
1064 PF_RULES_RASSERT();
1065
1066 MPASS(pd->m == *m0);
1067
1068 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1069 /*
1070 * Check if there are any scrub rules, matching or not.
1071 * Lack of scrub rules means:
1072 * - enforced packet normalization operation just like in OpenBSD
1073 * - fragment reassembly depends on V_pf_status.reass
1074 * With scrub rules:
1075 * - packet normalization is performed if there is a matching scrub rule
1076 * - fragment reassembly is performed if the matching rule has no
1077 * PFRULE_FRAGMENT_NOREASS flag
1078 */
1079 scrub_compat = (r != NULL);
1080 while (r != NULL) {
1081 pf_counter_u64_add(&r->evaluations, 1);
1082 if (pfi_kkif_match(r->kif, pd->kif) == r->ifnot)
1083 r = r->skip[PF_SKIP_IFP];
1084 else if (r->direction && r->direction != pd->dir)
1085 r = r->skip[PF_SKIP_DIR];
1086 else if (r->af && r->af != AF_INET)
1087 r = r->skip[PF_SKIP_AF];
1088 else if (r->proto && r->proto != h->ip_p)
1089 r = r->skip[PF_SKIP_PROTO];
1090 else if (PF_MISMATCHAW(&r->src.addr,
1091 (struct pf_addr *)&h->ip_src.s_addr, AF_INET,
1092 r->src.neg, pd->kif, M_GETFIB(pd->m)))
1093 r = r->skip[PF_SKIP_SRC_ADDR];
1094 else if (PF_MISMATCHAW(&r->dst.addr,
1095 (struct pf_addr *)&h->ip_dst.s_addr, AF_INET,
1096 r->dst.neg, NULL, M_GETFIB(pd->m)))
1097 r = r->skip[PF_SKIP_DST_ADDR];
1098 else if (r->match_tag && !pf_match_tag(pd->m, r, &tag,
1099 pd->pf_mtag ? pd->pf_mtag->tag : 0))
1100 r = TAILQ_NEXT(r, entries);
1101 else
1102 break;
1103 }
1104
1105 if (scrub_compat) {
1106 /* With scrub rules present IPv4 normalization happens only
1107 * if one of rules has matched and it's not a "no scrub" rule */
1108 if (r == NULL || r->action == PF_NOSCRUB)
1109 return (PF_PASS);
1110
1111 pf_counter_u64_critical_enter();
1112 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1);
1113 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len);
1114 pf_counter_u64_critical_exit();
1115 pf_rule_to_actions(r, &pd->act);
1116 }
1117
1118 /* Check for illegal packets */
1119 if (hlen < (int)sizeof(struct ip)) {
1120 REASON_SET(reason, PFRES_NORM);
1121 goto drop;
1122 }
1123
1124 if (hlen > ntohs(h->ip_len)) {
1125 REASON_SET(reason, PFRES_NORM);
1126 goto drop;
1127 }
1128
1129 /* Clear IP_DF if the rule uses the no-df option or we're in no-df mode */
1130 if (((!scrub_compat && V_pf_status.reass & PF_REASS_NODF) ||
1131 (r != NULL && r->rule_flag & PFRULE_NODF)) &&
1132 (h->ip_off & htons(IP_DF))
1133 ) {
1134 u_int16_t ip_off = h->ip_off;
1135
1136 h->ip_off &= htons(~IP_DF);
1137 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
1138 }
1139
1140 /* We will need other tests here */
1141 if (!fragoff && !mff)
1142 goto no_fragment;
1143
1144 /* We're dealing with a fragment now. Don't allow fragments
1145 * with IP_DF to enter the cache. If the flag was cleared by
1146 * no-df above, fine. Otherwise drop it.
1147 */
1148 if (h->ip_off & htons(IP_DF)) {
1149 DPFPRINTF(("IP_DF\n"));
1150 goto bad;
1151 }
1152
1153 ip_len = ntohs(h->ip_len) - hlen;
1154
1155 /* All fragments are 8 byte aligned */
1156 if (mff && (ip_len & 0x7)) {
1157 DPFPRINTF(("mff and %d\n", ip_len));
1158 goto bad;
1159 }
1160
1161 /* Respect maximum length */
1162 if (fragoff + ip_len > IP_MAXPACKET) {
1163 DPFPRINTF(("max packet %d\n", fragoff + ip_len));
1164 goto bad;
1165 }
1166
1167 if ((!scrub_compat && V_pf_status.reass) ||
1168 (r != NULL && !(r->rule_flag & PFRULE_FRAGMENT_NOREASS))
1169 ) {
1170 max = fragoff + ip_len;
1171
1172 /* Fully buffer all of the fragments
1173 * Might return a completely reassembled mbuf, or NULL */
1174 PF_FRAG_LOCK();
1175 DPFPRINTF(("reass frag %d @ %d-%d\n", h->ip_id, fragoff, max));
1176 verdict = pf_reassemble(m0, pd->dir, reason);
1177 PF_FRAG_UNLOCK();
1178
1179 if (verdict != PF_PASS)
1180 return (PF_DROP);
1181
1182 pd->m = *m0;
1183 if (pd->m == NULL)
1184 return (PF_DROP);
1185
1186 h = mtod(pd->m, struct ip *);
1187
1188 no_fragment:
1189 /* At this point, only IP_DF is allowed in ip_off */
1190 if (h->ip_off & ~htons(IP_DF)) {
1191 u_int16_t ip_off = h->ip_off;
1192
1193 h->ip_off &= htons(IP_DF);
1194 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
1195 }
1196 }
1197
1198 return (PF_PASS);
1199
1200 bad:
1201 DPFPRINTF(("dropping bad fragment\n"));
1202 REASON_SET(reason, PFRES_FRAG);
1203 drop:
1204 if (r != NULL && r->log)
1205 PFLOG_PACKET(PF_DROP, *reason, r, NULL, NULL, pd, 1);
1206
1207 return (PF_DROP);
1208 }
1209 #endif
1210
1211 #ifdef INET6
1212 int
pf_normalize_ip6(struct mbuf ** m0,int off,u_short * reason,struct pf_pdesc * pd)1213 pf_normalize_ip6(struct mbuf **m0, int off, u_short *reason,
1214 struct pf_pdesc *pd)
1215 {
1216 struct pf_krule *r;
1217 struct ip6_frag frag;
1218 bool scrub_compat;
1219
1220 PF_RULES_RASSERT();
1221
1222 pd->m = *m0;
1223
1224 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1225 /*
1226 * Check if there are any scrub rules, matching or not.
1227 * Lack of scrub rules means:
1228 * - enforced packet normalization operation just like in OpenBSD
1229 * With scrub rules:
1230 * - packet normalization is performed if there is a matching scrub rule
1231 * XXX: Fragment reassembly always performed for IPv6!
1232 */
1233 scrub_compat = (r != NULL);
1234 while (r != NULL) {
1235 pf_counter_u64_add(&r->evaluations, 1);
1236 if (pfi_kkif_match(r->kif, pd->kif) == r->ifnot)
1237 r = r->skip[PF_SKIP_IFP];
1238 else if (r->direction && r->direction != pd->dir)
1239 r = r->skip[PF_SKIP_DIR];
1240 else if (r->af && r->af != AF_INET6)
1241 r = r->skip[PF_SKIP_AF];
1242 else if (r->proto && r->proto != pd->proto)
1243 r = r->skip[PF_SKIP_PROTO];
1244 else if (PF_MISMATCHAW(&r->src.addr,
1245 (struct pf_addr *)&pd->src, AF_INET6,
1246 r->src.neg, pd->kif, M_GETFIB(pd->m)))
1247 r = r->skip[PF_SKIP_SRC_ADDR];
1248 else if (PF_MISMATCHAW(&r->dst.addr,
1249 (struct pf_addr *)&pd->dst, AF_INET6,
1250 r->dst.neg, NULL, M_GETFIB(pd->m)))
1251 r = r->skip[PF_SKIP_DST_ADDR];
1252 else
1253 break;
1254 }
1255
1256 if (scrub_compat) {
1257 /* With scrub rules present IPv6 normalization happens only
1258 * if one of rules has matched and it's not a "no scrub" rule */
1259 if (r == NULL || r->action == PF_NOSCRUB)
1260 return (PF_PASS);
1261
1262 pf_counter_u64_critical_enter();
1263 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1);
1264 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len);
1265 pf_counter_u64_critical_exit();
1266 pf_rule_to_actions(r, &pd->act);
1267 }
1268
1269 if (!pf_pull_hdr(pd->m, off, &frag, sizeof(frag), NULL, reason, AF_INET6))
1270 return (PF_DROP);
1271
1272 /* Offset now points to data portion. */
1273 off += sizeof(frag);
1274
1275 if (pd->virtual_proto == PF_VPROTO_FRAGMENT) {
1276 /* Returns PF_DROP or *m0 is NULL or completely reassembled
1277 * mbuf. */
1278 if (pf_reassemble6(m0, &frag, off, pd->extoff, reason) != PF_PASS)
1279 return (PF_DROP);
1280 pd->m = *m0;
1281 if (pd->m == NULL)
1282 return (PF_DROP);
1283 }
1284
1285 return (PF_PASS);
1286 }
1287 #endif /* INET6 */
1288
1289 int
pf_normalize_tcp(struct pf_pdesc * pd)1290 pf_normalize_tcp(struct pf_pdesc *pd)
1291 {
1292 struct pf_krule *r, *rm = NULL;
1293 struct tcphdr *th = &pd->hdr.tcp;
1294 int rewrite = 0;
1295 u_short reason;
1296 u_int16_t flags;
1297 sa_family_t af = pd->af;
1298 int srs;
1299
1300 PF_RULES_RASSERT();
1301
1302 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1303 /* Check if there any scrub rules. Lack of scrub rules means enforced
1304 * packet normalization operation just like in OpenBSD. */
1305 srs = (r != NULL);
1306 while (r != NULL) {
1307 pf_counter_u64_add(&r->evaluations, 1);
1308 if (pfi_kkif_match(r->kif, pd->kif) == r->ifnot)
1309 r = r->skip[PF_SKIP_IFP];
1310 else if (r->direction && r->direction != pd->dir)
1311 r = r->skip[PF_SKIP_DIR];
1312 else if (r->af && r->af != af)
1313 r = r->skip[PF_SKIP_AF];
1314 else if (r->proto && r->proto != pd->proto)
1315 r = r->skip[PF_SKIP_PROTO];
1316 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
1317 r->src.neg, pd->kif, M_GETFIB(pd->m)))
1318 r = r->skip[PF_SKIP_SRC_ADDR];
1319 else if (r->src.port_op && !pf_match_port(r->src.port_op,
1320 r->src.port[0], r->src.port[1], th->th_sport))
1321 r = r->skip[PF_SKIP_SRC_PORT];
1322 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
1323 r->dst.neg, NULL, M_GETFIB(pd->m)))
1324 r = r->skip[PF_SKIP_DST_ADDR];
1325 else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
1326 r->dst.port[0], r->dst.port[1], th->th_dport))
1327 r = r->skip[PF_SKIP_DST_PORT];
1328 else if (r->os_fingerprint != PF_OSFP_ANY && !pf_osfp_match(
1329 pf_osfp_fingerprint(pd, th),
1330 r->os_fingerprint))
1331 r = TAILQ_NEXT(r, entries);
1332 else {
1333 rm = r;
1334 break;
1335 }
1336 }
1337
1338 if (srs) {
1339 /* With scrub rules present TCP normalization happens only
1340 * if one of rules has matched and it's not a "no scrub" rule */
1341 if (rm == NULL || rm->action == PF_NOSCRUB)
1342 return (PF_PASS);
1343
1344 pf_counter_u64_critical_enter();
1345 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1);
1346 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len);
1347 pf_counter_u64_critical_exit();
1348 pf_rule_to_actions(rm, &pd->act);
1349 }
1350
1351 if (rm && rm->rule_flag & PFRULE_REASSEMBLE_TCP)
1352 pd->flags |= PFDESC_TCP_NORM;
1353
1354 flags = tcp_get_flags(th);
1355 if (flags & TH_SYN) {
1356 /* Illegal packet */
1357 if (flags & TH_RST)
1358 goto tcp_drop;
1359
1360 if (flags & TH_FIN)
1361 goto tcp_drop;
1362 } else {
1363 /* Illegal packet */
1364 if (!(flags & (TH_ACK|TH_RST)))
1365 goto tcp_drop;
1366 }
1367
1368 if (!(flags & TH_ACK)) {
1369 /* These flags are only valid if ACK is set */
1370 if ((flags & TH_FIN) || (flags & TH_PUSH) || (flags & TH_URG))
1371 goto tcp_drop;
1372 }
1373
1374 /* Check for illegal header length */
1375 if (th->th_off < (sizeof(struct tcphdr) >> 2))
1376 goto tcp_drop;
1377
1378 /* If flags changed, or reserved data set, then adjust */
1379 if (flags != tcp_get_flags(th) ||
1380 (tcp_get_flags(th) & (TH_RES1|TH_RES2|TH_RES2)) != 0) {
1381 u_int16_t ov, nv;
1382
1383 ov = *(u_int16_t *)(&th->th_ack + 1);
1384 flags &= ~(TH_RES1 | TH_RES2 | TH_RES3);
1385 tcp_set_flags(th, flags);
1386 nv = *(u_int16_t *)(&th->th_ack + 1);
1387
1388 th->th_sum = pf_proto_cksum_fixup(pd->m, th->th_sum, ov, nv, 0);
1389 rewrite = 1;
1390 }
1391
1392 /* Remove urgent pointer, if TH_URG is not set */
1393 if (!(flags & TH_URG) && th->th_urp) {
1394 th->th_sum = pf_proto_cksum_fixup(pd->m, th->th_sum, th->th_urp,
1395 0, 0);
1396 th->th_urp = 0;
1397 rewrite = 1;
1398 }
1399
1400 /* copy back packet headers if we sanitized */
1401 if (rewrite)
1402 m_copyback(pd->m, pd->off, sizeof(*th), (caddr_t)th);
1403
1404 return (PF_PASS);
1405
1406 tcp_drop:
1407 REASON_SET(&reason, PFRES_NORM);
1408 if (rm != NULL && r->log)
1409 PFLOG_PACKET(PF_DROP, reason, r, NULL, NULL, pd, 1);
1410 return (PF_DROP);
1411 }
1412
1413 int
pf_normalize_tcp_init(struct pf_pdesc * pd,struct tcphdr * th,struct pf_state_peer * src,struct pf_state_peer * dst)1414 pf_normalize_tcp_init(struct pf_pdesc *pd, struct tcphdr *th,
1415 struct pf_state_peer *src, struct pf_state_peer *dst)
1416 {
1417 u_int32_t tsval, tsecr;
1418 u_int8_t hdr[60];
1419 u_int8_t *opt;
1420
1421 KASSERT((src->scrub == NULL),
1422 ("pf_normalize_tcp_init: src->scrub != NULL"));
1423
1424 src->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT);
1425 if (src->scrub == NULL)
1426 return (1);
1427
1428 switch (pd->af) {
1429 #ifdef INET
1430 case AF_INET: {
1431 struct ip *h = mtod(pd->m, struct ip *);
1432 src->scrub->pfss_ttl = h->ip_ttl;
1433 break;
1434 }
1435 #endif /* INET */
1436 #ifdef INET6
1437 case AF_INET6: {
1438 struct ip6_hdr *h = mtod(pd->m, struct ip6_hdr *);
1439 src->scrub->pfss_ttl = h->ip6_hlim;
1440 break;
1441 }
1442 #endif /* INET6 */
1443 }
1444
1445 /*
1446 * All normalizations below are only begun if we see the start of
1447 * the connections. They must all set an enabled bit in pfss_flags
1448 */
1449 if ((th->th_flags & TH_SYN) == 0)
1450 return (0);
1451
1452 if (th->th_off > (sizeof(struct tcphdr) >> 2) && src->scrub &&
1453 pf_pull_hdr(pd->m, pd->off, hdr, th->th_off << 2, NULL, NULL, pd->af)) {
1454 /* Diddle with TCP options */
1455 int hlen;
1456 opt = hdr + sizeof(struct tcphdr);
1457 hlen = (th->th_off << 2) - sizeof(struct tcphdr);
1458 while (hlen >= TCPOLEN_TIMESTAMP) {
1459 switch (*opt) {
1460 case TCPOPT_EOL: /* FALLTHROUGH */
1461 case TCPOPT_NOP:
1462 opt++;
1463 hlen--;
1464 break;
1465 case TCPOPT_TIMESTAMP:
1466 if (opt[1] >= TCPOLEN_TIMESTAMP) {
1467 src->scrub->pfss_flags |=
1468 PFSS_TIMESTAMP;
1469 src->scrub->pfss_ts_mod =
1470 htonl(arc4random());
1471
1472 /* note PFSS_PAWS not set yet */
1473 memcpy(&tsval, &opt[2],
1474 sizeof(u_int32_t));
1475 memcpy(&tsecr, &opt[6],
1476 sizeof(u_int32_t));
1477 src->scrub->pfss_tsval0 = ntohl(tsval);
1478 src->scrub->pfss_tsval = ntohl(tsval);
1479 src->scrub->pfss_tsecr = ntohl(tsecr);
1480 getmicrouptime(&src->scrub->pfss_last);
1481 }
1482 /* FALLTHROUGH */
1483 default:
1484 hlen -= MAX(opt[1], 2);
1485 opt += MAX(opt[1], 2);
1486 break;
1487 }
1488 }
1489 }
1490
1491 return (0);
1492 }
1493
1494 void
pf_normalize_tcp_cleanup(struct pf_kstate * state)1495 pf_normalize_tcp_cleanup(struct pf_kstate *state)
1496 {
1497 /* XXX Note: this also cleans up SCTP. */
1498 uma_zfree(V_pf_state_scrub_z, state->src.scrub);
1499 uma_zfree(V_pf_state_scrub_z, state->dst.scrub);
1500
1501 /* Someday... flush the TCP segment reassembly descriptors. */
1502 }
1503 int
pf_normalize_sctp_init(struct pf_pdesc * pd,struct pf_state_peer * src,struct pf_state_peer * dst)1504 pf_normalize_sctp_init(struct pf_pdesc *pd, struct pf_state_peer *src,
1505 struct pf_state_peer *dst)
1506 {
1507 src->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT);
1508 if (src->scrub == NULL)
1509 return (1);
1510
1511 dst->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT);
1512 if (dst->scrub == NULL) {
1513 uma_zfree(V_pf_state_scrub_z, src);
1514 return (1);
1515 }
1516
1517 dst->scrub->pfss_v_tag = pd->sctp_initiate_tag;
1518
1519 return (0);
1520 }
1521
1522 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)1523 pf_normalize_tcp_stateful(struct pf_pdesc *pd,
1524 u_short *reason, struct tcphdr *th, struct pf_kstate *state,
1525 struct pf_state_peer *src, struct pf_state_peer *dst, int *writeback)
1526 {
1527 struct timeval uptime;
1528 u_int32_t tsval, tsecr;
1529 u_int tsval_from_last;
1530 u_int8_t hdr[60];
1531 u_int8_t *opt;
1532 int copyback = 0;
1533 int got_ts = 0;
1534 size_t startoff;
1535
1536 KASSERT((src->scrub || dst->scrub),
1537 ("%s: src->scrub && dst->scrub!", __func__));
1538
1539 /*
1540 * Enforce the minimum TTL seen for this connection. Negate a common
1541 * technique to evade an intrusion detection system and confuse
1542 * firewall state code.
1543 */
1544 switch (pd->af) {
1545 #ifdef INET
1546 case AF_INET: {
1547 if (src->scrub) {
1548 struct ip *h = mtod(pd->m, struct ip *);
1549 if (h->ip_ttl > src->scrub->pfss_ttl)
1550 src->scrub->pfss_ttl = h->ip_ttl;
1551 h->ip_ttl = src->scrub->pfss_ttl;
1552 }
1553 break;
1554 }
1555 #endif /* INET */
1556 #ifdef INET6
1557 case AF_INET6: {
1558 if (src->scrub) {
1559 struct ip6_hdr *h = mtod(pd->m, struct ip6_hdr *);
1560 if (h->ip6_hlim > src->scrub->pfss_ttl)
1561 src->scrub->pfss_ttl = h->ip6_hlim;
1562 h->ip6_hlim = src->scrub->pfss_ttl;
1563 }
1564 break;
1565 }
1566 #endif /* INET6 */
1567 }
1568
1569 if (th->th_off > (sizeof(struct tcphdr) >> 2) &&
1570 ((src->scrub && (src->scrub->pfss_flags & PFSS_TIMESTAMP)) ||
1571 (dst->scrub && (dst->scrub->pfss_flags & PFSS_TIMESTAMP))) &&
1572 pf_pull_hdr(pd->m, pd->off, hdr, th->th_off << 2, NULL, NULL, pd->af)) {
1573 /* Diddle with TCP options */
1574 int hlen;
1575 opt = hdr + sizeof(struct tcphdr);
1576 hlen = (th->th_off << 2) - sizeof(struct tcphdr);
1577 while (hlen >= TCPOLEN_TIMESTAMP) {
1578 startoff = opt - (hdr + sizeof(struct tcphdr));
1579 switch (*opt) {
1580 case TCPOPT_EOL: /* FALLTHROUGH */
1581 case TCPOPT_NOP:
1582 opt++;
1583 hlen--;
1584 break;
1585 case TCPOPT_TIMESTAMP:
1586 /* Modulate the timestamps. Can be used for
1587 * NAT detection, OS uptime determination or
1588 * reboot detection.
1589 */
1590
1591 if (got_ts) {
1592 /* Huh? Multiple timestamps!? */
1593 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1594 DPFPRINTF(("multiple TS??\n"));
1595 pf_print_state(state);
1596 printf("\n");
1597 }
1598 REASON_SET(reason, PFRES_TS);
1599 return (PF_DROP);
1600 }
1601 if (opt[1] >= TCPOLEN_TIMESTAMP) {
1602 memcpy(&tsval, &opt[2],
1603 sizeof(u_int32_t));
1604 if (tsval && src->scrub &&
1605 (src->scrub->pfss_flags &
1606 PFSS_TIMESTAMP)) {
1607 tsval = ntohl(tsval);
1608 pf_patch_32_unaligned(pd->m,
1609 &th->th_sum,
1610 &opt[2],
1611 htonl(tsval +
1612 src->scrub->pfss_ts_mod),
1613 PF_ALGNMNT(startoff),
1614 0);
1615 copyback = 1;
1616 }
1617
1618 /* Modulate TS reply iff valid (!0) */
1619 memcpy(&tsecr, &opt[6],
1620 sizeof(u_int32_t));
1621 if (tsecr && dst->scrub &&
1622 (dst->scrub->pfss_flags &
1623 PFSS_TIMESTAMP)) {
1624 tsecr = ntohl(tsecr)
1625 - dst->scrub->pfss_ts_mod;
1626 pf_patch_32_unaligned(pd->m,
1627 &th->th_sum,
1628 &opt[6],
1629 htonl(tsecr),
1630 PF_ALGNMNT(startoff),
1631 0);
1632 copyback = 1;
1633 }
1634 got_ts = 1;
1635 }
1636 /* FALLTHROUGH */
1637 default:
1638 hlen -= MAX(opt[1], 2);
1639 opt += MAX(opt[1], 2);
1640 break;
1641 }
1642 }
1643 if (copyback) {
1644 /* Copyback the options, caller copys back header */
1645 *writeback = 1;
1646 m_copyback(pd->m, pd->off + sizeof(struct tcphdr),
1647 (th->th_off << 2) - sizeof(struct tcphdr), hdr +
1648 sizeof(struct tcphdr));
1649 }
1650 }
1651
1652 /*
1653 * Must invalidate PAWS checks on connections idle for too long.
1654 * The fastest allowed timestamp clock is 1ms. That turns out to
1655 * be about 24 days before it wraps. XXX Right now our lowerbound
1656 * TS echo check only works for the first 12 days of a connection
1657 * when the TS has exhausted half its 32bit space
1658 */
1659 #define TS_MAX_IDLE (24*24*60*60)
1660 #define TS_MAX_CONN (12*24*60*60) /* XXX remove when better tsecr check */
1661
1662 getmicrouptime(&uptime);
1663 if (src->scrub && (src->scrub->pfss_flags & PFSS_PAWS) &&
1664 (uptime.tv_sec - src->scrub->pfss_last.tv_sec > TS_MAX_IDLE ||
1665 time_uptime - (state->creation / 1000) > TS_MAX_CONN)) {
1666 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1667 DPFPRINTF(("src idled out of PAWS\n"));
1668 pf_print_state(state);
1669 printf("\n");
1670 }
1671 src->scrub->pfss_flags = (src->scrub->pfss_flags & ~PFSS_PAWS)
1672 | PFSS_PAWS_IDLED;
1673 }
1674 if (dst->scrub && (dst->scrub->pfss_flags & PFSS_PAWS) &&
1675 uptime.tv_sec - dst->scrub->pfss_last.tv_sec > TS_MAX_IDLE) {
1676 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1677 DPFPRINTF(("dst idled out of PAWS\n"));
1678 pf_print_state(state);
1679 printf("\n");
1680 }
1681 dst->scrub->pfss_flags = (dst->scrub->pfss_flags & ~PFSS_PAWS)
1682 | PFSS_PAWS_IDLED;
1683 }
1684
1685 if (got_ts && src->scrub && dst->scrub &&
1686 (src->scrub->pfss_flags & PFSS_PAWS) &&
1687 (dst->scrub->pfss_flags & PFSS_PAWS)) {
1688 /* Validate that the timestamps are "in-window".
1689 * RFC1323 describes TCP Timestamp options that allow
1690 * measurement of RTT (round trip time) and PAWS
1691 * (protection against wrapped sequence numbers). PAWS
1692 * gives us a set of rules for rejecting packets on
1693 * long fat pipes (packets that were somehow delayed
1694 * in transit longer than the time it took to send the
1695 * full TCP sequence space of 4Gb). We can use these
1696 * rules and infer a few others that will let us treat
1697 * the 32bit timestamp and the 32bit echoed timestamp
1698 * as sequence numbers to prevent a blind attacker from
1699 * inserting packets into a connection.
1700 *
1701 * RFC1323 tells us:
1702 * - The timestamp on this packet must be greater than
1703 * or equal to the last value echoed by the other
1704 * endpoint. The RFC says those will be discarded
1705 * since it is a dup that has already been acked.
1706 * This gives us a lowerbound on the timestamp.
1707 * timestamp >= other last echoed timestamp
1708 * - The timestamp will be less than or equal to
1709 * the last timestamp plus the time between the
1710 * last packet and now. The RFC defines the max
1711 * clock rate as 1ms. We will allow clocks to be
1712 * up to 10% fast and will allow a total difference
1713 * or 30 seconds due to a route change. And this
1714 * gives us an upperbound on the timestamp.
1715 * timestamp <= last timestamp + max ticks
1716 * We have to be careful here. Windows will send an
1717 * initial timestamp of zero and then initialize it
1718 * to a random value after the 3whs; presumably to
1719 * avoid a DoS by having to call an expensive RNG
1720 * during a SYN flood. Proof MS has at least one
1721 * good security geek.
1722 *
1723 * - The TCP timestamp option must also echo the other
1724 * endpoints timestamp. The timestamp echoed is the
1725 * one carried on the earliest unacknowledged segment
1726 * on the left edge of the sequence window. The RFC
1727 * states that the host will reject any echoed
1728 * timestamps that were larger than any ever sent.
1729 * This gives us an upperbound on the TS echo.
1730 * tescr <= largest_tsval
1731 * - The lowerbound on the TS echo is a little more
1732 * tricky to determine. The other endpoint's echoed
1733 * values will not decrease. But there may be
1734 * network conditions that re-order packets and
1735 * cause our view of them to decrease. For now the
1736 * only lowerbound we can safely determine is that
1737 * the TS echo will never be less than the original
1738 * TS. XXX There is probably a better lowerbound.
1739 * Remove TS_MAX_CONN with better lowerbound check.
1740 * tescr >= other original TS
1741 *
1742 * It is also important to note that the fastest
1743 * timestamp clock of 1ms will wrap its 32bit space in
1744 * 24 days. So we just disable TS checking after 24
1745 * days of idle time. We actually must use a 12d
1746 * connection limit until we can come up with a better
1747 * lowerbound to the TS echo check.
1748 */
1749 struct timeval delta_ts;
1750 int ts_fudge;
1751
1752 /*
1753 * PFTM_TS_DIFF is how many seconds of leeway to allow
1754 * a host's timestamp. This can happen if the previous
1755 * packet got delayed in transit for much longer than
1756 * this packet.
1757 */
1758 if ((ts_fudge = state->rule->timeout[PFTM_TS_DIFF]) == 0)
1759 ts_fudge = V_pf_default_rule.timeout[PFTM_TS_DIFF];
1760
1761 /* Calculate max ticks since the last timestamp */
1762 #define TS_MAXFREQ 1100 /* RFC max TS freq of 1Khz + 10% skew */
1763 #define TS_MICROSECS 1000000 /* microseconds per second */
1764 delta_ts = uptime;
1765 timevalsub(&delta_ts, &src->scrub->pfss_last);
1766 tsval_from_last = (delta_ts.tv_sec + ts_fudge) * TS_MAXFREQ;
1767 tsval_from_last += delta_ts.tv_usec / (TS_MICROSECS/TS_MAXFREQ);
1768
1769 if ((src->state >= TCPS_ESTABLISHED &&
1770 dst->state >= TCPS_ESTABLISHED) &&
1771 (SEQ_LT(tsval, dst->scrub->pfss_tsecr) ||
1772 SEQ_GT(tsval, src->scrub->pfss_tsval + tsval_from_last) ||
1773 (tsecr && (SEQ_GT(tsecr, dst->scrub->pfss_tsval) ||
1774 SEQ_LT(tsecr, dst->scrub->pfss_tsval0))))) {
1775 /* Bad RFC1323 implementation or an insertion attack.
1776 *
1777 * - Solaris 2.6 and 2.7 are known to send another ACK
1778 * after the FIN,FIN|ACK,ACK closing that carries
1779 * an old timestamp.
1780 */
1781
1782 DPFPRINTF(("Timestamp failed %c%c%c%c\n",
1783 SEQ_LT(tsval, dst->scrub->pfss_tsecr) ? '0' : ' ',
1784 SEQ_GT(tsval, src->scrub->pfss_tsval +
1785 tsval_from_last) ? '1' : ' ',
1786 SEQ_GT(tsecr, dst->scrub->pfss_tsval) ? '2' : ' ',
1787 SEQ_LT(tsecr, dst->scrub->pfss_tsval0)? '3' : ' '));
1788 DPFPRINTF((" tsval: %u tsecr: %u +ticks: %u "
1789 "idle: %jus %lums\n",
1790 tsval, tsecr, tsval_from_last,
1791 (uintmax_t)delta_ts.tv_sec,
1792 delta_ts.tv_usec / 1000));
1793 DPFPRINTF((" src->tsval: %u tsecr: %u\n",
1794 src->scrub->pfss_tsval, src->scrub->pfss_tsecr));
1795 DPFPRINTF((" dst->tsval: %u tsecr: %u tsval0: %u"
1796 "\n", dst->scrub->pfss_tsval,
1797 dst->scrub->pfss_tsecr, dst->scrub->pfss_tsval0));
1798 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1799 pf_print_state(state);
1800 pf_print_flags(th->th_flags);
1801 printf("\n");
1802 }
1803 REASON_SET(reason, PFRES_TS);
1804 return (PF_DROP);
1805 }
1806
1807 /* XXX I'd really like to require tsecr but it's optional */
1808
1809 } else if (!got_ts && (th->th_flags & TH_RST) == 0 &&
1810 ((src->state == TCPS_ESTABLISHED && dst->state == TCPS_ESTABLISHED)
1811 || pd->p_len > 0 || (th->th_flags & TH_SYN)) &&
1812 src->scrub && dst->scrub &&
1813 (src->scrub->pfss_flags & PFSS_PAWS) &&
1814 (dst->scrub->pfss_flags & PFSS_PAWS)) {
1815 /* Didn't send a timestamp. Timestamps aren't really useful
1816 * when:
1817 * - connection opening or closing (often not even sent).
1818 * but we must not let an attacker to put a FIN on a
1819 * data packet to sneak it through our ESTABLISHED check.
1820 * - on a TCP reset. RFC suggests not even looking at TS.
1821 * - on an empty ACK. The TS will not be echoed so it will
1822 * probably not help keep the RTT calculation in sync and
1823 * there isn't as much danger when the sequence numbers
1824 * got wrapped. So some stacks don't include TS on empty
1825 * ACKs :-(
1826 *
1827 * To minimize the disruption to mostly RFC1323 conformant
1828 * stacks, we will only require timestamps on data packets.
1829 *
1830 * And what do ya know, we cannot require timestamps on data
1831 * packets. There appear to be devices that do legitimate
1832 * TCP connection hijacking. There are HTTP devices that allow
1833 * a 3whs (with timestamps) and then buffer the HTTP request.
1834 * If the intermediate device has the HTTP response cache, it
1835 * will spoof the response but not bother timestamping its
1836 * packets. So we can look for the presence of a timestamp in
1837 * the first data packet and if there, require it in all future
1838 * packets.
1839 */
1840
1841 if (pd->p_len > 0 && (src->scrub->pfss_flags & PFSS_DATA_TS)) {
1842 /*
1843 * Hey! Someone tried to sneak a packet in. Or the
1844 * stack changed its RFC1323 behavior?!?!
1845 */
1846 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1847 DPFPRINTF(("Did not receive expected RFC1323 "
1848 "timestamp\n"));
1849 pf_print_state(state);
1850 pf_print_flags(th->th_flags);
1851 printf("\n");
1852 }
1853 REASON_SET(reason, PFRES_TS);
1854 return (PF_DROP);
1855 }
1856 }
1857
1858 /*
1859 * We will note if a host sends his data packets with or without
1860 * timestamps. And require all data packets to contain a timestamp
1861 * if the first does. PAWS implicitly requires that all data packets be
1862 * timestamped. But I think there are middle-man devices that hijack
1863 * TCP streams immediately after the 3whs and don't timestamp their
1864 * packets (seen in a WWW accelerator or cache).
1865 */
1866 if (pd->p_len > 0 && src->scrub && (src->scrub->pfss_flags &
1867 (PFSS_TIMESTAMP|PFSS_DATA_TS|PFSS_DATA_NOTS)) == PFSS_TIMESTAMP) {
1868 if (got_ts)
1869 src->scrub->pfss_flags |= PFSS_DATA_TS;
1870 else {
1871 src->scrub->pfss_flags |= PFSS_DATA_NOTS;
1872 if (V_pf_status.debug >= PF_DEBUG_MISC && dst->scrub &&
1873 (dst->scrub->pfss_flags & PFSS_TIMESTAMP)) {
1874 /* Don't warn if other host rejected RFC1323 */
1875 DPFPRINTF(("Broken RFC1323 stack did not "
1876 "timestamp data packet. Disabled PAWS "
1877 "security.\n"));
1878 pf_print_state(state);
1879 pf_print_flags(th->th_flags);
1880 printf("\n");
1881 }
1882 }
1883 }
1884
1885 /*
1886 * Update PAWS values
1887 */
1888 if (got_ts && src->scrub && PFSS_TIMESTAMP == (src->scrub->pfss_flags &
1889 (PFSS_PAWS_IDLED|PFSS_TIMESTAMP))) {
1890 getmicrouptime(&src->scrub->pfss_last);
1891 if (SEQ_GEQ(tsval, src->scrub->pfss_tsval) ||
1892 (src->scrub->pfss_flags & PFSS_PAWS) == 0)
1893 src->scrub->pfss_tsval = tsval;
1894
1895 if (tsecr) {
1896 if (SEQ_GEQ(tsecr, src->scrub->pfss_tsecr) ||
1897 (src->scrub->pfss_flags & PFSS_PAWS) == 0)
1898 src->scrub->pfss_tsecr = tsecr;
1899
1900 if ((src->scrub->pfss_flags & PFSS_PAWS) == 0 &&
1901 (SEQ_LT(tsval, src->scrub->pfss_tsval0) ||
1902 src->scrub->pfss_tsval0 == 0)) {
1903 /* tsval0 MUST be the lowest timestamp */
1904 src->scrub->pfss_tsval0 = tsval;
1905 }
1906
1907 /* Only fully initialized after a TS gets echoed */
1908 if ((src->scrub->pfss_flags & PFSS_PAWS) == 0)
1909 src->scrub->pfss_flags |= PFSS_PAWS;
1910 }
1911 }
1912
1913 /* I have a dream.... TCP segment reassembly.... */
1914 return (0);
1915 }
1916
1917 int
pf_normalize_mss(struct pf_pdesc * pd)1918 pf_normalize_mss(struct pf_pdesc *pd)
1919 {
1920 struct tcphdr *th = &pd->hdr.tcp;
1921 u_int16_t *mss;
1922 int thoff;
1923 int opt, cnt, optlen = 0;
1924 u_char opts[TCP_MAXOLEN];
1925 u_char *optp = opts;
1926 size_t startoff;
1927
1928 thoff = th->th_off << 2;
1929 cnt = thoff - sizeof(struct tcphdr);
1930
1931 if (cnt > 0 && !pf_pull_hdr(pd->m, pd->off + sizeof(*th), opts, cnt,
1932 NULL, NULL, pd->af))
1933 return (0);
1934
1935 for (; cnt > 0; cnt -= optlen, optp += optlen) {
1936 startoff = optp - opts;
1937 opt = optp[0];
1938 if (opt == TCPOPT_EOL)
1939 break;
1940 if (opt == TCPOPT_NOP)
1941 optlen = 1;
1942 else {
1943 if (cnt < 2)
1944 break;
1945 optlen = optp[1];
1946 if (optlen < 2 || optlen > cnt)
1947 break;
1948 }
1949 switch (opt) {
1950 case TCPOPT_MAXSEG:
1951 mss = (u_int16_t *)(optp + 2);
1952 if ((ntohs(*mss)) > pd->act.max_mss) {
1953 pf_patch_16_unaligned(pd->m,
1954 &th->th_sum,
1955 mss, htons(pd->act.max_mss),
1956 PF_ALGNMNT(startoff),
1957 0);
1958 m_copyback(pd->m, pd->off + sizeof(*th),
1959 thoff - sizeof(*th), opts);
1960 m_copyback(pd->m, pd->off, sizeof(*th), (caddr_t)th);
1961 }
1962 break;
1963 default:
1964 break;
1965 }
1966 }
1967
1968 return (0);
1969 }
1970
1971 int
pf_scan_sctp(struct pf_pdesc * pd)1972 pf_scan_sctp(struct pf_pdesc *pd)
1973 {
1974 struct sctp_chunkhdr ch = { };
1975 int chunk_off = sizeof(struct sctphdr);
1976 int chunk_start;
1977 int ret;
1978
1979 while (pd->off + chunk_off < pd->tot_len) {
1980 if (!pf_pull_hdr(pd->m, pd->off + chunk_off, &ch, sizeof(ch), NULL,
1981 NULL, pd->af))
1982 return (PF_DROP);
1983
1984 /* Length includes the header, this must be at least 4. */
1985 if (ntohs(ch.chunk_length) < 4)
1986 return (PF_DROP);
1987
1988 chunk_start = chunk_off;
1989 chunk_off += roundup(ntohs(ch.chunk_length), 4);
1990
1991 switch (ch.chunk_type) {
1992 case SCTP_INITIATION:
1993 case SCTP_INITIATION_ACK: {
1994 struct sctp_init_chunk init;
1995
1996 if (!pf_pull_hdr(pd->m, pd->off + chunk_start, &init,
1997 sizeof(init), NULL, NULL, pd->af))
1998 return (PF_DROP);
1999
2000 /*
2001 * RFC 9620, Section 3.3.2, "The Initiate Tag is allowed to have
2002 * any value except 0."
2003 */
2004 if (init.init.initiate_tag == 0)
2005 return (PF_DROP);
2006 if (init.init.num_inbound_streams == 0)
2007 return (PF_DROP);
2008 if (init.init.num_outbound_streams == 0)
2009 return (PF_DROP);
2010 if (ntohl(init.init.a_rwnd) < SCTP_MIN_RWND)
2011 return (PF_DROP);
2012
2013 /*
2014 * RFC 9260, Section 3.1, INIT chunks MUST have zero
2015 * verification tag.
2016 */
2017 if (ch.chunk_type == SCTP_INITIATION &&
2018 pd->hdr.sctp.v_tag != 0)
2019 return (PF_DROP);
2020
2021 pd->sctp_initiate_tag = init.init.initiate_tag;
2022
2023 if (ch.chunk_type == SCTP_INITIATION)
2024 pd->sctp_flags |= PFDESC_SCTP_INIT;
2025 else
2026 pd->sctp_flags |= PFDESC_SCTP_INIT_ACK;
2027
2028 ret = pf_multihome_scan_init(pd->off + chunk_start,
2029 ntohs(init.ch.chunk_length), pd);
2030 if (ret != PF_PASS)
2031 return (ret);
2032
2033 break;
2034 }
2035 case SCTP_ABORT_ASSOCIATION:
2036 pd->sctp_flags |= PFDESC_SCTP_ABORT;
2037 break;
2038 case SCTP_SHUTDOWN:
2039 case SCTP_SHUTDOWN_ACK:
2040 pd->sctp_flags |= PFDESC_SCTP_SHUTDOWN;
2041 break;
2042 case SCTP_SHUTDOWN_COMPLETE:
2043 pd->sctp_flags |= PFDESC_SCTP_SHUTDOWN_COMPLETE;
2044 break;
2045 case SCTP_COOKIE_ECHO:
2046 pd->sctp_flags |= PFDESC_SCTP_COOKIE;
2047 break;
2048 case SCTP_COOKIE_ACK:
2049 pd->sctp_flags |= PFDESC_SCTP_COOKIE_ACK;
2050 break;
2051 case SCTP_DATA:
2052 pd->sctp_flags |= PFDESC_SCTP_DATA;
2053 break;
2054 case SCTP_HEARTBEAT_REQUEST:
2055 pd->sctp_flags |= PFDESC_SCTP_HEARTBEAT;
2056 break;
2057 case SCTP_HEARTBEAT_ACK:
2058 pd->sctp_flags |= PFDESC_SCTP_HEARTBEAT_ACK;
2059 break;
2060 case SCTP_ASCONF:
2061 pd->sctp_flags |= PFDESC_SCTP_ASCONF;
2062
2063 ret = pf_multihome_scan_asconf(pd->off + chunk_start,
2064 ntohs(ch.chunk_length), pd);
2065 if (ret != PF_PASS)
2066 return (ret);
2067 break;
2068 default:
2069 pd->sctp_flags |= PFDESC_SCTP_OTHER;
2070 break;
2071 }
2072 }
2073
2074 /* Validate chunk lengths vs. packet length. */
2075 if (pd->off + chunk_off != pd->tot_len)
2076 return (PF_DROP);
2077
2078 /*
2079 * INIT, INIT_ACK or SHUTDOWN_COMPLETE chunks must always be the only
2080 * one in a packet.
2081 */
2082 if ((pd->sctp_flags & PFDESC_SCTP_INIT) &&
2083 (pd->sctp_flags & ~PFDESC_SCTP_INIT))
2084 return (PF_DROP);
2085 if ((pd->sctp_flags & PFDESC_SCTP_INIT_ACK) &&
2086 (pd->sctp_flags & ~PFDESC_SCTP_INIT_ACK))
2087 return (PF_DROP);
2088 if ((pd->sctp_flags & PFDESC_SCTP_SHUTDOWN_COMPLETE) &&
2089 (pd->sctp_flags & ~PFDESC_SCTP_SHUTDOWN_COMPLETE))
2090 return (PF_DROP);
2091
2092 return (PF_PASS);
2093 }
2094
2095 int
pf_normalize_sctp(struct pf_pdesc * pd)2096 pf_normalize_sctp(struct pf_pdesc *pd)
2097 {
2098 struct pf_krule *r, *rm = NULL;
2099 struct sctphdr *sh = &pd->hdr.sctp;
2100 u_short reason;
2101 sa_family_t af = pd->af;
2102 int srs;
2103
2104 PF_RULES_RASSERT();
2105
2106 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
2107 /* Check if there any scrub rules. Lack of scrub rules means enforced
2108 * packet normalization operation just like in OpenBSD. */
2109 srs = (r != NULL);
2110 while (r != NULL) {
2111 pf_counter_u64_add(&r->evaluations, 1);
2112 if (pfi_kkif_match(r->kif, pd->kif) == r->ifnot)
2113 r = r->skip[PF_SKIP_IFP];
2114 else if (r->direction && r->direction != pd->dir)
2115 r = r->skip[PF_SKIP_DIR];
2116 else if (r->af && r->af != af)
2117 r = r->skip[PF_SKIP_AF];
2118 else if (r->proto && r->proto != pd->proto)
2119 r = r->skip[PF_SKIP_PROTO];
2120 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
2121 r->src.neg, pd->kif, M_GETFIB(pd->m)))
2122 r = r->skip[PF_SKIP_SRC_ADDR];
2123 else if (r->src.port_op && !pf_match_port(r->src.port_op,
2124 r->src.port[0], r->src.port[1], sh->src_port))
2125 r = r->skip[PF_SKIP_SRC_PORT];
2126 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
2127 r->dst.neg, NULL, M_GETFIB(pd->m)))
2128 r = r->skip[PF_SKIP_DST_ADDR];
2129 else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
2130 r->dst.port[0], r->dst.port[1], sh->dest_port))
2131 r = r->skip[PF_SKIP_DST_PORT];
2132 else {
2133 rm = r;
2134 break;
2135 }
2136 }
2137
2138 if (srs) {
2139 /* With scrub rules present SCTP normalization happens only
2140 * if one of rules has matched and it's not a "no scrub" rule */
2141 if (rm == NULL || rm->action == PF_NOSCRUB)
2142 return (PF_PASS);
2143
2144 pf_counter_u64_critical_enter();
2145 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1);
2146 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len);
2147 pf_counter_u64_critical_exit();
2148 }
2149
2150 /* Verify we're a multiple of 4 bytes long */
2151 if ((pd->tot_len - pd->off - sizeof(struct sctphdr)) % 4)
2152 goto sctp_drop;
2153
2154 /* INIT chunk needs to be the only chunk */
2155 if (pd->sctp_flags & PFDESC_SCTP_INIT)
2156 if (pd->sctp_flags & ~PFDESC_SCTP_INIT)
2157 goto sctp_drop;
2158
2159 return (PF_PASS);
2160
2161 sctp_drop:
2162 REASON_SET(&reason, PFRES_NORM);
2163 if (rm != NULL && r->log)
2164 PFLOG_PACKET(PF_DROP, reason, r, NULL, NULL, pd,
2165 1);
2166
2167 return (PF_DROP);
2168 }
2169
2170 #if defined(INET) || defined(INET6)
2171 void
pf_scrub(struct pf_pdesc * pd)2172 pf_scrub(struct pf_pdesc *pd)
2173 {
2174
2175 struct ip *h = mtod(pd->m, struct ip *);
2176 #ifdef INET6
2177 struct ip6_hdr *h6 = mtod(pd->m, struct ip6_hdr *);
2178 #endif
2179
2180 /* Clear IP_DF if no-df was requested */
2181 if (pd->af == AF_INET && pd->act.flags & PFSTATE_NODF &&
2182 h->ip_off & htons(IP_DF))
2183 {
2184 u_int16_t ip_off = h->ip_off;
2185
2186 h->ip_off &= htons(~IP_DF);
2187 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
2188 }
2189
2190 /* Enforce a minimum ttl, may cause endless packet loops */
2191 if (pd->af == AF_INET && pd->act.min_ttl &&
2192 h->ip_ttl < pd->act.min_ttl) {
2193 u_int16_t ip_ttl = h->ip_ttl;
2194
2195 h->ip_ttl = pd->act.min_ttl;
2196 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_ttl, h->ip_ttl, 0);
2197 }
2198 #ifdef INET6
2199 /* Enforce a minimum ttl, may cause endless packet loops */
2200 if (pd->af == AF_INET6 && pd->act.min_ttl &&
2201 h6->ip6_hlim < pd->act.min_ttl)
2202 h6->ip6_hlim = pd->act.min_ttl;
2203 #endif
2204 /* Enforce tos */
2205 if (pd->act.flags & PFSTATE_SETTOS) {
2206 switch (pd->af) {
2207 case AF_INET: {
2208 u_int16_t ov, nv;
2209
2210 ov = *(u_int16_t *)h;
2211 h->ip_tos = pd->act.set_tos | (h->ip_tos & IPTOS_ECN_MASK);
2212 nv = *(u_int16_t *)h;
2213
2214 h->ip_sum = pf_cksum_fixup(h->ip_sum, ov, nv, 0);
2215 break;
2216 }
2217 #ifdef INET6
2218 case AF_INET6:
2219 h6->ip6_flow &= IPV6_FLOWLABEL_MASK | IPV6_VERSION_MASK;
2220 h6->ip6_flow |= htonl((pd->act.set_tos | IPV6_ECN(h6)) << 20);
2221 break;
2222 #endif
2223 }
2224 }
2225
2226 /* random-id, but not for fragments */
2227 #ifdef INET
2228 if (pd->af == AF_INET &&
2229 pd->act.flags & PFSTATE_RANDOMID && !(h->ip_off & ~htons(IP_DF))) {
2230 uint16_t ip_id = h->ip_id;
2231
2232 ip_fillid(h);
2233 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_id, h->ip_id, 0);
2234 }
2235 #endif
2236 }
2237 #endif
2238