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/vnet.h>
47 #include <net/pfvar.h>
48 #include <net/if_pflog.h>
49
50 #include <netinet/in.h>
51 #include <netinet/ip.h>
52 #include <netinet/ip_var.h>
53 #include <netinet6/in6_var.h>
54 #include <netinet6/nd6.h>
55 #include <netinet6/ip6_var.h>
56 #include <netinet6/scope6_var.h>
57 #include <netinet/tcp.h>
58 #include <netinet/tcp_fsm.h>
59 #include <netinet/tcp_seq.h>
60 #include <netinet/sctp_constants.h>
61 #include <netinet/sctp_header.h>
62
63 #ifdef INET6
64 #include <netinet/ip6.h>
65 #endif /* INET6 */
66
67 struct pf_frent {
68 TAILQ_ENTRY(pf_frent) fr_next;
69 struct mbuf *fe_m;
70 uint16_t fe_hdrlen; /* ipv4 header length with ip options
71 ipv6, extension, fragment header */
72 uint16_t fe_extoff; /* last extension header offset or 0 */
73 uint16_t fe_len; /* fragment length */
74 uint16_t fe_off; /* fragment offset */
75 uint16_t fe_mff; /* more fragment flag */
76 };
77
78 struct pf_fragment_cmp {
79 struct pf_addr frc_src;
80 struct pf_addr frc_dst;
81 uint32_t frc_id;
82 sa_family_t frc_af;
83 uint8_t frc_proto;
84 };
85
86 struct pf_fragment {
87 struct pf_fragment_cmp fr_key;
88 #define fr_src fr_key.frc_src
89 #define fr_dst fr_key.frc_dst
90 #define fr_id fr_key.frc_id
91 #define fr_af fr_key.frc_af
92 #define fr_proto fr_key.frc_proto
93
94 /* pointers to queue element */
95 struct pf_frent *fr_firstoff[PF_FRAG_ENTRY_POINTS];
96 /* count entries between pointers */
97 uint8_t fr_entries[PF_FRAG_ENTRY_POINTS];
98 RB_ENTRY(pf_fragment) fr_entry;
99 TAILQ_ENTRY(pf_fragment) frag_next;
100 uint32_t fr_timeout;
101 TAILQ_HEAD(pf_fragq, pf_frent) fr_queue;
102 uint16_t fr_maxlen; /* maximum length of single fragment */
103 u_int16_t fr_holes; /* number of holes in the queue */
104 };
105
106 VNET_DEFINE_STATIC(struct mtx, pf_frag_mtx);
107 #define V_pf_frag_mtx VNET(pf_frag_mtx)
108 #define PF_FRAG_LOCK() mtx_lock(&V_pf_frag_mtx)
109 #define PF_FRAG_UNLOCK() mtx_unlock(&V_pf_frag_mtx)
110 #define PF_FRAG_ASSERT() mtx_assert(&V_pf_frag_mtx, MA_OWNED)
111
112 VNET_DEFINE(uma_zone_t, pf_state_scrub_z); /* XXX: shared with pfsync */
113
114 VNET_DEFINE_STATIC(uma_zone_t, pf_frent_z);
115 #define V_pf_frent_z VNET(pf_frent_z)
116 VNET_DEFINE_STATIC(uma_zone_t, pf_frag_z);
117 #define V_pf_frag_z VNET(pf_frag_z)
118
119 TAILQ_HEAD(pf_fragqueue, pf_fragment);
120 TAILQ_HEAD(pf_cachequeue, pf_fragment);
121 VNET_DEFINE_STATIC(struct pf_fragqueue, pf_fragqueue);
122 #define V_pf_fragqueue VNET(pf_fragqueue)
123 RB_HEAD(pf_frag_tree, pf_fragment);
124 VNET_DEFINE_STATIC(struct pf_frag_tree, pf_frag_tree);
125 #define V_pf_frag_tree VNET(pf_frag_tree)
126 static int pf_frag_compare(struct pf_fragment *,
127 struct pf_fragment *);
128 static RB_PROTOTYPE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
129 static RB_GENERATE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
130
131 static void pf_flush_fragments(void);
132 static void pf_free_fragment(struct pf_fragment *);
133
134 static struct pf_frent *pf_create_fragment(u_short *);
135 static int pf_frent_holes(struct pf_frent *frent);
136 static struct pf_fragment *pf_find_fragment(struct pf_fragment_cmp *key,
137 struct pf_frag_tree *tree);
138 static inline int pf_frent_index(struct pf_frent *);
139 static int pf_frent_insert(struct pf_fragment *,
140 struct pf_frent *, struct pf_frent *);
141 void pf_frent_remove(struct pf_fragment *,
142 struct pf_frent *);
143 struct pf_frent *pf_frent_previous(struct pf_fragment *,
144 struct pf_frent *);
145 static struct pf_fragment *pf_fillup_fragment(struct pf_fragment_cmp *,
146 struct pf_frent *, u_short *);
147 static struct mbuf *pf_join_fragment(struct pf_fragment *);
148 #ifdef INET
149 static int pf_reassemble(struct mbuf **, int, u_short *);
150 #endif /* INET */
151 #ifdef INET6
152 static int pf_reassemble6(struct mbuf **,
153 struct ip6_frag *, uint16_t, uint16_t, u_short *);
154 #endif /* INET6 */
155
156 #define DPFPRINTF(x) do { \
157 if (V_pf_status.debug >= PF_DEBUG_MISC) { \
158 printf("%s: ", __func__); \
159 printf x ; \
160 } \
161 } while(0)
162
163 #ifdef INET
164 static void
pf_ip2key(struct ip * ip,int dir,struct pf_fragment_cmp * key)165 pf_ip2key(struct ip *ip, int dir, struct pf_fragment_cmp *key)
166 {
167
168 key->frc_src.v4 = ip->ip_src;
169 key->frc_dst.v4 = ip->ip_dst;
170 key->frc_af = AF_INET;
171 key->frc_proto = ip->ip_p;
172 key->frc_id = ip->ip_id;
173 }
174 #endif /* INET */
175
176 void
pf_normalize_init(void)177 pf_normalize_init(void)
178 {
179
180 V_pf_frag_z = uma_zcreate("pf frags", sizeof(struct pf_fragment),
181 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
182 V_pf_frent_z = uma_zcreate("pf frag entries", sizeof(struct pf_frent),
183 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
184 V_pf_state_scrub_z = uma_zcreate("pf state scrubs",
185 sizeof(struct pf_state_scrub), NULL, NULL, NULL, NULL,
186 UMA_ALIGN_PTR, 0);
187
188 mtx_init(&V_pf_frag_mtx, "pf fragments", NULL, MTX_DEF);
189
190 V_pf_limits[PF_LIMIT_FRAGS].zone = V_pf_frent_z;
191 V_pf_limits[PF_LIMIT_FRAGS].limit = PFFRAG_FRENT_HIWAT;
192 uma_zone_set_max(V_pf_frent_z, PFFRAG_FRENT_HIWAT);
193 uma_zone_set_warning(V_pf_frent_z, "PF frag entries limit reached");
194
195 TAILQ_INIT(&V_pf_fragqueue);
196 }
197
198 void
pf_normalize_cleanup(void)199 pf_normalize_cleanup(void)
200 {
201
202 uma_zdestroy(V_pf_state_scrub_z);
203 uma_zdestroy(V_pf_frent_z);
204 uma_zdestroy(V_pf_frag_z);
205
206 mtx_destroy(&V_pf_frag_mtx);
207 }
208
209 static int
pf_frag_compare(struct pf_fragment * a,struct pf_fragment * b)210 pf_frag_compare(struct pf_fragment *a, struct pf_fragment *b)
211 {
212 int diff;
213
214 if ((diff = a->fr_id - b->fr_id) != 0)
215 return (diff);
216 if ((diff = a->fr_proto - b->fr_proto) != 0)
217 return (diff);
218 if ((diff = a->fr_af - b->fr_af) != 0)
219 return (diff);
220 if ((diff = pf_addr_cmp(&a->fr_src, &b->fr_src, a->fr_af)) != 0)
221 return (diff);
222 if ((diff = pf_addr_cmp(&a->fr_dst, &b->fr_dst, a->fr_af)) != 0)
223 return (diff);
224 return (0);
225 }
226
227 void
pf_purge_expired_fragments(void)228 pf_purge_expired_fragments(void)
229 {
230 u_int32_t expire = time_uptime -
231 V_pf_default_rule.timeout[PFTM_FRAG];
232
233 pf_purge_fragments(expire);
234 }
235
236 void
pf_purge_fragments(uint32_t expire)237 pf_purge_fragments(uint32_t expire)
238 {
239 struct pf_fragment *frag;
240
241 PF_FRAG_LOCK();
242 while ((frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue)) != NULL) {
243 if (frag->fr_timeout > expire)
244 break;
245
246 DPFPRINTF(("expiring %d(%p)\n", frag->fr_id, frag));
247 pf_free_fragment(frag);
248 }
249
250 PF_FRAG_UNLOCK();
251 }
252
253 /*
254 * Try to flush old fragments to make space for new ones
255 */
256 static void
pf_flush_fragments(void)257 pf_flush_fragments(void)
258 {
259 struct pf_fragment *frag;
260 int goal;
261
262 PF_FRAG_ASSERT();
263
264 goal = uma_zone_get_cur(V_pf_frent_z) * 9 / 10;
265 DPFPRINTF(("trying to free %d frag entriess\n", goal));
266 while (goal < uma_zone_get_cur(V_pf_frent_z)) {
267 frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue);
268 if (frag)
269 pf_free_fragment(frag);
270 else
271 break;
272 }
273 }
274
275 /*
276 * Remove a fragment from the fragment queue, free its fragment entries,
277 * and free the fragment itself.
278 */
279 static void
pf_free_fragment(struct pf_fragment * frag)280 pf_free_fragment(struct pf_fragment *frag)
281 {
282 struct pf_frent *frent;
283
284 PF_FRAG_ASSERT();
285
286 RB_REMOVE(pf_frag_tree, &V_pf_frag_tree, frag);
287 TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next);
288
289 /* Free all fragment entries */
290 while ((frent = TAILQ_FIRST(&frag->fr_queue)) != NULL) {
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 uma_zfree(V_pf_frag_z, 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 TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next);
310 TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next);
311 }
312
313 return (frag);
314 }
315
316 static struct pf_frent *
pf_create_fragment(u_short * reason)317 pf_create_fragment(u_short *reason)
318 {
319 struct pf_frent *frent;
320
321 PF_FRAG_ASSERT();
322
323 frent = uma_zalloc(V_pf_frent_z, M_NOWAIT);
324 if (frent == NULL) {
325 pf_flush_fragments();
326 frent = uma_zalloc(V_pf_frent_z, M_NOWAIT);
327 if (frent == NULL) {
328 REASON_SET(reason, PFRES_MEMORY);
329 return (NULL);
330 }
331 }
332
333 return (frent);
334 }
335
336 /*
337 * Calculate the additional holes that were created in the fragment
338 * queue by inserting this fragment. A fragment in the middle
339 * creates one more hole by splitting. For each connected side,
340 * it loses one hole.
341 * Fragment entry must be in the queue when calling this function.
342 */
343 static int
pf_frent_holes(struct pf_frent * frent)344 pf_frent_holes(struct pf_frent *frent)
345 {
346 struct pf_frent *prev = TAILQ_PREV(frent, pf_fragq, fr_next);
347 struct pf_frent *next = TAILQ_NEXT(frent, fr_next);
348 int holes = 1;
349
350 if (prev == NULL) {
351 if (frent->fe_off == 0)
352 holes--;
353 } else {
354 KASSERT(frent->fe_off != 0, ("frent->fe_off != 0"));
355 if (frent->fe_off == prev->fe_off + prev->fe_len)
356 holes--;
357 }
358 if (next == NULL) {
359 if (!frent->fe_mff)
360 holes--;
361 } else {
362 KASSERT(frent->fe_mff, ("frent->fe_mff"));
363 if (next->fe_off == frent->fe_off + frent->fe_len)
364 holes--;
365 }
366 return holes;
367 }
368
369 static inline int
pf_frent_index(struct pf_frent * frent)370 pf_frent_index(struct pf_frent *frent)
371 {
372 /*
373 * We have an array of 16 entry points to the queue. A full size
374 * 65535 octet IP packet can have 8192 fragments. So the queue
375 * traversal length is at most 512 and at most 16 entry points are
376 * checked. We need 128 additional bytes on a 64 bit architecture.
377 */
378 CTASSERT(((u_int16_t)0xffff &~ 7) / (0x10000 / PF_FRAG_ENTRY_POINTS) ==
379 16 - 1);
380 CTASSERT(((u_int16_t)0xffff >> 3) / PF_FRAG_ENTRY_POINTS == 512 - 1);
381
382 return frent->fe_off / (0x10000 / PF_FRAG_ENTRY_POINTS);
383 }
384
385 static int
pf_frent_insert(struct pf_fragment * frag,struct pf_frent * frent,struct pf_frent * prev)386 pf_frent_insert(struct pf_fragment *frag, struct pf_frent *frent,
387 struct pf_frent *prev)
388 {
389 int index;
390
391 CTASSERT(PF_FRAG_ENTRY_LIMIT <= 0xff);
392
393 /*
394 * A packet has at most 65536 octets. With 16 entry points, each one
395 * spawns 4096 octets. We limit these to 64 fragments each, which
396 * means on average every fragment must have at least 64 octets.
397 */
398 index = pf_frent_index(frent);
399 if (frag->fr_entries[index] >= PF_FRAG_ENTRY_LIMIT)
400 return ENOBUFS;
401 frag->fr_entries[index]++;
402
403 if (prev == NULL) {
404 TAILQ_INSERT_HEAD(&frag->fr_queue, frent, fr_next);
405 } else {
406 KASSERT(prev->fe_off + prev->fe_len <= frent->fe_off,
407 ("overlapping fragment"));
408 TAILQ_INSERT_AFTER(&frag->fr_queue, prev, frent, fr_next);
409 }
410
411 if (frag->fr_firstoff[index] == NULL) {
412 KASSERT(prev == NULL || pf_frent_index(prev) < index,
413 ("prev == NULL || pf_frent_index(pref) < index"));
414 frag->fr_firstoff[index] = frent;
415 } else {
416 if (frent->fe_off < frag->fr_firstoff[index]->fe_off) {
417 KASSERT(prev == NULL || pf_frent_index(prev) < index,
418 ("prev == NULL || pf_frent_index(pref) < index"));
419 frag->fr_firstoff[index] = frent;
420 } else {
421 KASSERT(prev != NULL, ("prev != NULL"));
422 KASSERT(pf_frent_index(prev) == index,
423 ("pf_frent_index(prev) == index"));
424 }
425 }
426
427 frag->fr_holes += pf_frent_holes(frent);
428
429 return 0;
430 }
431
432 void
pf_frent_remove(struct pf_fragment * frag,struct pf_frent * frent)433 pf_frent_remove(struct pf_fragment *frag, struct pf_frent *frent)
434 {
435 #ifdef INVARIANTS
436 struct pf_frent *prev = TAILQ_PREV(frent, pf_fragq, fr_next);
437 #endif
438 struct pf_frent *next = TAILQ_NEXT(frent, fr_next);
439 int index;
440
441 frag->fr_holes -= pf_frent_holes(frent);
442
443 index = pf_frent_index(frent);
444 KASSERT(frag->fr_firstoff[index] != NULL, ("frent not found"));
445 if (frag->fr_firstoff[index]->fe_off == frent->fe_off) {
446 if (next == NULL) {
447 frag->fr_firstoff[index] = NULL;
448 } else {
449 KASSERT(frent->fe_off + frent->fe_len <= next->fe_off,
450 ("overlapping fragment"));
451 if (pf_frent_index(next) == index) {
452 frag->fr_firstoff[index] = next;
453 } else {
454 frag->fr_firstoff[index] = NULL;
455 }
456 }
457 } else {
458 KASSERT(frag->fr_firstoff[index]->fe_off < frent->fe_off,
459 ("frag->fr_firstoff[index]->fe_off < frent->fe_off"));
460 KASSERT(prev != NULL, ("prev != NULL"));
461 KASSERT(prev->fe_off + prev->fe_len <= frent->fe_off,
462 ("overlapping fragment"));
463 KASSERT(pf_frent_index(prev) == index,
464 ("pf_frent_index(prev) == index"));
465 }
466
467 TAILQ_REMOVE(&frag->fr_queue, frent, fr_next);
468
469 KASSERT(frag->fr_entries[index] > 0, ("No fragments remaining"));
470 frag->fr_entries[index]--;
471 }
472
473 struct pf_frent *
pf_frent_previous(struct pf_fragment * frag,struct pf_frent * frent)474 pf_frent_previous(struct pf_fragment *frag, struct pf_frent *frent)
475 {
476 struct pf_frent *prev, *next;
477 int index;
478
479 /*
480 * If there are no fragments after frag, take the final one. Assume
481 * that the global queue is not empty.
482 */
483 prev = TAILQ_LAST(&frag->fr_queue, pf_fragq);
484 KASSERT(prev != NULL, ("prev != NULL"));
485 if (prev->fe_off <= frent->fe_off)
486 return prev;
487 /*
488 * We want to find a fragment entry that is before frag, but still
489 * close to it. Find the first fragment entry that is in the same
490 * entry point or in the first entry point after that. As we have
491 * already checked that there are entries behind frag, this will
492 * succeed.
493 */
494 for (index = pf_frent_index(frent); index < PF_FRAG_ENTRY_POINTS;
495 index++) {
496 prev = frag->fr_firstoff[index];
497 if (prev != NULL)
498 break;
499 }
500 KASSERT(prev != NULL, ("prev != NULL"));
501 /*
502 * In prev we may have a fragment from the same entry point that is
503 * before frent, or one that is just one position behind frent.
504 * In the latter case, we go back one step and have the predecessor.
505 * There may be none if the new fragment will be the first one.
506 */
507 if (prev->fe_off > frent->fe_off) {
508 prev = TAILQ_PREV(prev, pf_fragq, fr_next);
509 if (prev == NULL)
510 return NULL;
511 KASSERT(prev->fe_off <= frent->fe_off,
512 ("prev->fe_off <= frent->fe_off"));
513 return prev;
514 }
515 /*
516 * In prev is the first fragment of the entry point. The offset
517 * of frag is behind it. Find the closest previous fragment.
518 */
519 for (next = TAILQ_NEXT(prev, fr_next); next != NULL;
520 next = TAILQ_NEXT(next, fr_next)) {
521 if (next->fe_off > frent->fe_off)
522 break;
523 prev = next;
524 }
525 return prev;
526 }
527
528 static struct pf_fragment *
pf_fillup_fragment(struct pf_fragment_cmp * key,struct pf_frent * frent,u_short * reason)529 pf_fillup_fragment(struct pf_fragment_cmp *key, struct pf_frent *frent,
530 u_short *reason)
531 {
532 struct pf_frent *after, *next, *prev;
533 struct pf_fragment *frag;
534 uint16_t total;
535
536 PF_FRAG_ASSERT();
537
538 /* No empty fragments. */
539 if (frent->fe_len == 0) {
540 DPFPRINTF(("bad fragment: len 0\n"));
541 goto bad_fragment;
542 }
543
544 /* All fragments are 8 byte aligned. */
545 if (frent->fe_mff && (frent->fe_len & 0x7)) {
546 DPFPRINTF(("bad fragment: mff and len %d\n", frent->fe_len));
547 goto bad_fragment;
548 }
549
550 /* Respect maximum length, IP_MAXPACKET == IPV6_MAXPACKET. */
551 if (frent->fe_off + frent->fe_len > IP_MAXPACKET) {
552 DPFPRINTF(("bad fragment: max packet %d\n",
553 frent->fe_off + frent->fe_len));
554 goto bad_fragment;
555 }
556
557 DPFPRINTF((key->frc_af == AF_INET ?
558 "reass frag %d @ %d-%d\n" : "reass frag %#08x @ %d-%d\n",
559 key->frc_id, frent->fe_off, frent->fe_off + frent->fe_len));
560
561 /* Fully buffer all of the fragments in this fragment queue. */
562 frag = pf_find_fragment(key, &V_pf_frag_tree);
563
564 /* Create a new reassembly queue for this packet. */
565 if (frag == NULL) {
566 frag = uma_zalloc(V_pf_frag_z, M_NOWAIT);
567 if (frag == NULL) {
568 pf_flush_fragments();
569 frag = uma_zalloc(V_pf_frag_z, M_NOWAIT);
570 if (frag == NULL) {
571 REASON_SET(reason, PFRES_MEMORY);
572 goto drop_fragment;
573 }
574 }
575
576 *(struct pf_fragment_cmp *)frag = *key;
577 memset(frag->fr_firstoff, 0, sizeof(frag->fr_firstoff));
578 memset(frag->fr_entries, 0, sizeof(frag->fr_entries));
579 frag->fr_timeout = time_uptime;
580 TAILQ_INIT(&frag->fr_queue);
581 frag->fr_maxlen = frent->fe_len;
582 frag->fr_holes = 1;
583
584 RB_INSERT(pf_frag_tree, &V_pf_frag_tree, frag);
585 TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next);
586
587 /* We do not have a previous fragment, cannot fail. */
588 pf_frent_insert(frag, frent, NULL);
589
590 return (frag);
591 }
592
593 KASSERT(!TAILQ_EMPTY(&frag->fr_queue), ("!TAILQ_EMPTY()->fr_queue"));
594
595 /* Remember maximum fragment len for refragmentation. */
596 if (frent->fe_len > frag->fr_maxlen)
597 frag->fr_maxlen = frent->fe_len;
598
599 /* Maximum data we have seen already. */
600 total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
601 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
602
603 /* Non terminal fragments must have more fragments flag. */
604 if (frent->fe_off + frent->fe_len < total && !frent->fe_mff)
605 goto bad_fragment;
606
607 /* Check if we saw the last fragment already. */
608 if (!TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_mff) {
609 if (frent->fe_off + frent->fe_len > total ||
610 (frent->fe_off + frent->fe_len == total && frent->fe_mff))
611 goto bad_fragment;
612 } else {
613 if (frent->fe_off + frent->fe_len == total && !frent->fe_mff)
614 goto bad_fragment;
615 }
616
617 /* Find neighbors for newly inserted fragment */
618 prev = pf_frent_previous(frag, frent);
619 if (prev == NULL) {
620 after = TAILQ_FIRST(&frag->fr_queue);
621 KASSERT(after != NULL, ("after != NULL"));
622 } else {
623 after = TAILQ_NEXT(prev, fr_next);
624 }
625
626 if (prev != NULL && prev->fe_off + prev->fe_len > frent->fe_off) {
627 uint16_t precut;
628
629 if (frag->fr_af == AF_INET6)
630 goto free_fragment;
631
632 precut = prev->fe_off + prev->fe_len - frent->fe_off;
633 if (precut >= frent->fe_len) {
634 DPFPRINTF(("new frag overlapped\n"));
635 goto drop_fragment;
636 }
637 DPFPRINTF(("frag head overlap %d\n", precut));
638 m_adj(frent->fe_m, precut);
639 frent->fe_off += precut;
640 frent->fe_len -= precut;
641 }
642
643 for (; after != NULL && frent->fe_off + frent->fe_len > after->fe_off;
644 after = next) {
645 uint16_t aftercut;
646
647 aftercut = frent->fe_off + frent->fe_len - after->fe_off;
648 if (aftercut < after->fe_len) {
649 DPFPRINTF(("frag tail overlap %d", aftercut));
650 m_adj(after->fe_m, aftercut);
651 /* Fragment may switch queue as fe_off changes */
652 pf_frent_remove(frag, after);
653 after->fe_off += aftercut;
654 after->fe_len -= aftercut;
655 /* Insert into correct queue */
656 if (pf_frent_insert(frag, after, prev)) {
657 DPFPRINTF(("fragment requeue limit exceeded"));
658 m_freem(after->fe_m);
659 uma_zfree(V_pf_frent_z, after);
660 /* There is not way to recover */
661 goto free_fragment;
662 }
663 break;
664 }
665
666 /* This fragment is completely overlapped, lose it. */
667 DPFPRINTF(("old frag overlapped\n"));
668 next = TAILQ_NEXT(after, fr_next);
669 pf_frent_remove(frag, after);
670 m_freem(after->fe_m);
671 uma_zfree(V_pf_frent_z, after);
672 }
673
674 /* If part of the queue gets too long, there is not way to recover. */
675 if (pf_frent_insert(frag, frent, prev)) {
676 DPFPRINTF(("fragment queue limit exceeded\n"));
677 goto bad_fragment;
678 }
679
680 return (frag);
681
682 free_fragment:
683 /*
684 * RFC 5722, Errata 3089: When reassembling an IPv6 datagram, if one
685 * or more its constituent fragments is determined to be an overlapping
686 * fragment, the entire datagram (and any constituent fragments) MUST
687 * be silently discarded.
688 */
689 DPFPRINTF(("flush overlapping fragments\n"));
690 pf_free_fragment(frag);
691
692 bad_fragment:
693 REASON_SET(reason, PFRES_FRAG);
694 drop_fragment:
695 uma_zfree(V_pf_frent_z, frent);
696 return (NULL);
697 }
698
699 static struct mbuf *
pf_join_fragment(struct pf_fragment * frag)700 pf_join_fragment(struct pf_fragment *frag)
701 {
702 struct mbuf *m, *m2;
703 struct pf_frent *frent;
704
705 frent = TAILQ_FIRST(&frag->fr_queue);
706 TAILQ_REMOVE(&frag->fr_queue, frent, fr_next);
707
708 m = frent->fe_m;
709 if ((frent->fe_hdrlen + frent->fe_len) < m->m_pkthdr.len)
710 m_adj(m, (frent->fe_hdrlen + frent->fe_len) - m->m_pkthdr.len);
711 uma_zfree(V_pf_frent_z, frent);
712 while ((frent = TAILQ_FIRST(&frag->fr_queue)) != NULL) {
713 TAILQ_REMOVE(&frag->fr_queue, frent, fr_next);
714
715 m2 = frent->fe_m;
716 /* Strip off ip header. */
717 m_adj(m2, frent->fe_hdrlen);
718 /* Strip off any trailing bytes. */
719 if (frent->fe_len < m2->m_pkthdr.len)
720 m_adj(m2, frent->fe_len - m2->m_pkthdr.len);
721
722 uma_zfree(V_pf_frent_z, frent);
723 m_cat(m, m2);
724 }
725
726 /* Remove from fragment queue. */
727 pf_free_fragment(frag);
728
729 return (m);
730 }
731
732 #ifdef INET
733 static int
pf_reassemble(struct mbuf ** m0,int dir,u_short * reason)734 pf_reassemble(struct mbuf **m0, int dir, u_short *reason)
735 {
736 struct mbuf *m = *m0;
737 struct ip *ip = mtod(m, struct ip *);
738 struct pf_frent *frent;
739 struct pf_fragment *frag;
740 struct m_tag *mtag;
741 struct pf_fragment_tag *ftag;
742 struct pf_fragment_cmp key;
743 uint16_t total, hdrlen;
744 uint32_t frag_id;
745 uint16_t maxlen;
746
747 /* Get an entry for the fragment queue */
748 if ((frent = pf_create_fragment(reason)) == NULL)
749 return (PF_DROP);
750
751 frent->fe_m = m;
752 frent->fe_hdrlen = ip->ip_hl << 2;
753 frent->fe_extoff = 0;
754 frent->fe_len = ntohs(ip->ip_len) - (ip->ip_hl << 2);
755 frent->fe_off = (ntohs(ip->ip_off) & IP_OFFMASK) << 3;
756 frent->fe_mff = ntohs(ip->ip_off) & IP_MF;
757
758 pf_ip2key(ip, dir, &key);
759
760 if ((frag = pf_fillup_fragment(&key, frent, reason)) == NULL)
761 return (PF_DROP);
762
763 /* The mbuf is part of the fragment entry, no direct free or access */
764 m = *m0 = NULL;
765
766 if (frag->fr_holes) {
767 DPFPRINTF(("frag %d, holes %d\n", frag->fr_id, frag->fr_holes));
768 return (PF_PASS); /* drop because *m0 is NULL, no error */
769 }
770
771 /* We have all the data */
772 frent = TAILQ_FIRST(&frag->fr_queue);
773 KASSERT(frent != NULL, ("frent != NULL"));
774 total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
775 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
776 hdrlen = frent->fe_hdrlen;
777
778 maxlen = frag->fr_maxlen;
779 frag_id = frag->fr_id;
780 m = *m0 = pf_join_fragment(frag);
781 frag = NULL;
782
783 if (m->m_flags & M_PKTHDR) {
784 int plen = 0;
785 for (m = *m0; m; m = m->m_next)
786 plen += m->m_len;
787 m = *m0;
788 m->m_pkthdr.len = plen;
789 }
790
791 if ((mtag = m_tag_get(PACKET_TAG_PF_REASSEMBLED,
792 sizeof(struct pf_fragment_tag), M_NOWAIT)) == NULL) {
793 REASON_SET(reason, PFRES_SHORT);
794 /* PF_DROP requires a valid mbuf *m0 in pf_test() */
795 return (PF_DROP);
796 }
797 ftag = (struct pf_fragment_tag *)(mtag + 1);
798 ftag->ft_hdrlen = hdrlen;
799 ftag->ft_extoff = 0;
800 ftag->ft_maxlen = maxlen;
801 ftag->ft_id = frag_id;
802 m_tag_prepend(m, mtag);
803
804 ip = mtod(m, struct ip *);
805 ip->ip_sum = pf_cksum_fixup(ip->ip_sum, ip->ip_len,
806 htons(hdrlen + total), 0);
807 ip->ip_len = htons(hdrlen + total);
808 ip->ip_sum = pf_cksum_fixup(ip->ip_sum, ip->ip_off,
809 ip->ip_off & ~(IP_MF|IP_OFFMASK), 0);
810 ip->ip_off &= ~(IP_MF|IP_OFFMASK);
811
812 if (hdrlen + total > IP_MAXPACKET) {
813 DPFPRINTF(("drop: too big: %d\n", total));
814 ip->ip_len = 0;
815 REASON_SET(reason, PFRES_SHORT);
816 /* PF_DROP requires a valid mbuf *m0 in pf_test() */
817 return (PF_DROP);
818 }
819
820 DPFPRINTF(("complete: %p(%d)\n", m, ntohs(ip->ip_len)));
821 return (PF_PASS);
822 }
823 #endif /* INET */
824
825 #ifdef INET6
826 static int
pf_reassemble6(struct mbuf ** m0,struct ip6_frag * fraghdr,uint16_t hdrlen,uint16_t extoff,u_short * reason)827 pf_reassemble6(struct mbuf **m0, struct ip6_frag *fraghdr,
828 uint16_t hdrlen, uint16_t extoff, u_short *reason)
829 {
830 struct mbuf *m = *m0;
831 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
832 struct pf_frent *frent;
833 struct pf_fragment *frag;
834 struct pf_fragment_cmp key;
835 struct m_tag *mtag;
836 struct pf_fragment_tag *ftag;
837 int off;
838 uint32_t frag_id;
839 uint16_t total, maxlen;
840 uint8_t proto;
841
842 PF_FRAG_LOCK();
843
844 /* Get an entry for the fragment queue. */
845 if ((frent = pf_create_fragment(reason)) == NULL) {
846 PF_FRAG_UNLOCK();
847 return (PF_DROP);
848 }
849
850 frent->fe_m = m;
851 frent->fe_hdrlen = hdrlen;
852 frent->fe_extoff = extoff;
853 frent->fe_len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - hdrlen;
854 frent->fe_off = ntohs(fraghdr->ip6f_offlg & IP6F_OFF_MASK);
855 frent->fe_mff = fraghdr->ip6f_offlg & IP6F_MORE_FRAG;
856
857 key.frc_src.v6 = ip6->ip6_src;
858 key.frc_dst.v6 = ip6->ip6_dst;
859 key.frc_af = AF_INET6;
860 /* Only the first fragment's protocol is relevant. */
861 key.frc_proto = 0;
862 key.frc_id = fraghdr->ip6f_ident;
863
864 if ((frag = pf_fillup_fragment(&key, frent, reason)) == NULL) {
865 PF_FRAG_UNLOCK();
866 return (PF_DROP);
867 }
868
869 /* The mbuf is part of the fragment entry, no direct free or access. */
870 m = *m0 = NULL;
871
872 if (frag->fr_holes) {
873 DPFPRINTF(("frag %d, holes %d\n", frag->fr_id,
874 frag->fr_holes));
875 PF_FRAG_UNLOCK();
876 return (PF_PASS); /* Drop because *m0 is NULL, no error. */
877 }
878
879 /* We have all the data. */
880 frent = TAILQ_FIRST(&frag->fr_queue);
881 KASSERT(frent != NULL, ("frent != NULL"));
882 extoff = frent->fe_extoff;
883 maxlen = frag->fr_maxlen;
884 frag_id = frag->fr_id;
885 total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
886 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
887 hdrlen = frent->fe_hdrlen - sizeof(struct ip6_frag);
888
889 m = *m0 = pf_join_fragment(frag);
890 frag = NULL;
891
892 PF_FRAG_UNLOCK();
893
894 /* Take protocol from first fragment header. */
895 m = m_getptr(m, hdrlen + offsetof(struct ip6_frag, ip6f_nxt), &off);
896 KASSERT(m, ("%s: short mbuf chain", __func__));
897 proto = *(mtod(m, uint8_t *) + off);
898 m = *m0;
899
900 /* Delete frag6 header */
901 if (ip6_deletefraghdr(m, hdrlen, M_NOWAIT) != 0)
902 goto fail;
903
904 if (m->m_flags & M_PKTHDR) {
905 int plen = 0;
906 for (m = *m0; m; m = m->m_next)
907 plen += m->m_len;
908 m = *m0;
909 m->m_pkthdr.len = plen;
910 }
911
912 if ((mtag = m_tag_get(PACKET_TAG_PF_REASSEMBLED,
913 sizeof(struct pf_fragment_tag), M_NOWAIT)) == NULL)
914 goto fail;
915 ftag = (struct pf_fragment_tag *)(mtag + 1);
916 ftag->ft_hdrlen = hdrlen;
917 ftag->ft_extoff = extoff;
918 ftag->ft_maxlen = maxlen;
919 ftag->ft_id = frag_id;
920 m_tag_prepend(m, mtag);
921
922 ip6 = mtod(m, struct ip6_hdr *);
923 ip6->ip6_plen = htons(hdrlen - sizeof(struct ip6_hdr) + total);
924 if (extoff) {
925 /* Write protocol into next field of last extension header. */
926 m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt),
927 &off);
928 KASSERT(m, ("%s: short mbuf chain", __func__));
929 *(mtod(m, char *) + off) = proto;
930 m = *m0;
931 } else
932 ip6->ip6_nxt = proto;
933
934 if (hdrlen - sizeof(struct ip6_hdr) + total > IPV6_MAXPACKET) {
935 DPFPRINTF(("drop: too big: %d\n", total));
936 ip6->ip6_plen = 0;
937 REASON_SET(reason, PFRES_SHORT);
938 /* PF_DROP requires a valid mbuf *m0 in pf_test6(). */
939 return (PF_DROP);
940 }
941
942 DPFPRINTF(("complete: %p(%d)\n", m, ntohs(ip6->ip6_plen)));
943 return (PF_PASS);
944
945 fail:
946 REASON_SET(reason, PFRES_MEMORY);
947 /* PF_DROP requires a valid mbuf *m0 in pf_test6(), will free later. */
948 return (PF_DROP);
949 }
950 #endif /* INET6 */
951
952 #ifdef INET6
953 int
pf_max_frag_size(struct mbuf * m)954 pf_max_frag_size(struct mbuf *m)
955 {
956 struct m_tag *tag;
957 struct pf_fragment_tag *ftag;
958
959 tag = m_tag_find(m, PACKET_TAG_PF_REASSEMBLED, NULL);
960 if (tag == NULL)
961 return (m->m_pkthdr.len);
962
963 ftag = (struct pf_fragment_tag *)(tag + 1);
964
965 return (ftag->ft_maxlen);
966 }
967
968 int
pf_refragment6(struct ifnet * ifp,struct mbuf ** m0,struct m_tag * mtag,struct ifnet * rt,bool forward)969 pf_refragment6(struct ifnet *ifp, struct mbuf **m0, struct m_tag *mtag,
970 struct ifnet *rt, bool forward)
971 {
972 struct mbuf *m = *m0, *t;
973 struct ip6_hdr *hdr;
974 struct pf_fragment_tag *ftag = (struct pf_fragment_tag *)(mtag + 1);
975 struct pf_pdesc pd;
976 uint32_t frag_id;
977 uint16_t hdrlen, extoff, maxlen;
978 uint8_t proto;
979 int error, action;
980
981 hdrlen = ftag->ft_hdrlen;
982 extoff = ftag->ft_extoff;
983 maxlen = ftag->ft_maxlen;
984 frag_id = ftag->ft_id;
985 m_tag_delete(m, mtag);
986 mtag = NULL;
987 ftag = NULL;
988
989 if (extoff) {
990 int off;
991
992 /* Use protocol from next field of last extension header */
993 m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt),
994 &off);
995 KASSERT((m != NULL), ("pf_refragment6: short mbuf chain"));
996 proto = *(mtod(m, uint8_t *) + off);
997 *(mtod(m, char *) + off) = IPPROTO_FRAGMENT;
998 m = *m0;
999 } else {
1000 hdr = mtod(m, struct ip6_hdr *);
1001 proto = hdr->ip6_nxt;
1002 hdr->ip6_nxt = IPPROTO_FRAGMENT;
1003 }
1004
1005 /* In case of link-local traffic we'll need a scope set. */
1006 hdr = mtod(m, struct ip6_hdr *);
1007
1008 in6_setscope(&hdr->ip6_src, ifp, NULL);
1009 in6_setscope(&hdr->ip6_dst, ifp, NULL);
1010
1011 /* The MTU must be a multiple of 8 bytes, or we risk doing the
1012 * fragmentation wrong. */
1013 maxlen = maxlen & ~7;
1014
1015 /*
1016 * Maxlen may be less than 8 if there was only a single
1017 * fragment. As it was fragmented before, add a fragment
1018 * header also for a single fragment. If total or maxlen
1019 * is less than 8, ip6_fragment() will return EMSGSIZE and
1020 * we drop the packet.
1021 */
1022 error = ip6_fragment(ifp, m, hdrlen, proto, maxlen, frag_id);
1023 m = (*m0)->m_nextpkt;
1024 (*m0)->m_nextpkt = NULL;
1025 if (error == 0) {
1026 /* The first mbuf contains the unfragmented packet. */
1027 m_freem(*m0);
1028 *m0 = NULL;
1029 action = PF_PASS;
1030 } else {
1031 /* Drop expects an mbuf to free. */
1032 DPFPRINTF(("refragment error %d\n", error));
1033 action = PF_DROP;
1034 }
1035 for (; m; m = t) {
1036 t = m->m_nextpkt;
1037 m->m_nextpkt = NULL;
1038 m->m_flags |= M_SKIP_FIREWALL;
1039 memset(&pd, 0, sizeof(pd));
1040 pd.pf_mtag = pf_find_mtag(m);
1041 if (error != 0) {
1042 m_freem(m);
1043 continue;
1044 }
1045 if (rt != NULL) {
1046 struct sockaddr_in6 dst;
1047 hdr = mtod(m, struct ip6_hdr *);
1048
1049 bzero(&dst, sizeof(dst));
1050 dst.sin6_family = AF_INET6;
1051 dst.sin6_len = sizeof(dst);
1052 dst.sin6_addr = hdr->ip6_dst;
1053
1054 nd6_output_ifp(rt, rt, m, &dst, NULL);
1055 } else if (forward) {
1056 MPASS(m->m_pkthdr.rcvif != NULL);
1057 ip6_forward(m, 0);
1058 } else {
1059 (void)ip6_output(m, NULL, NULL, 0, NULL, NULL,
1060 NULL);
1061 }
1062 }
1063
1064 return (action);
1065 }
1066 #endif /* INET6 */
1067
1068 #ifdef INET
1069 int
pf_normalize_ip(u_short * reason,struct pf_pdesc * pd)1070 pf_normalize_ip(u_short *reason, struct pf_pdesc *pd)
1071 {
1072 struct pf_krule *r;
1073 struct ip *h = mtod(pd->m, struct ip *);
1074 int mff = (ntohs(h->ip_off) & IP_MF);
1075 int hlen = h->ip_hl << 2;
1076 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
1077 u_int16_t max;
1078 int ip_len;
1079 int tag = -1;
1080 int verdict;
1081 bool scrub_compat;
1082
1083 PF_RULES_RASSERT();
1084
1085 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1086 /*
1087 * Check if there are any scrub rules, matching or not.
1088 * Lack of scrub rules means:
1089 * - enforced packet normalization operation just like in OpenBSD
1090 * - fragment reassembly depends on V_pf_status.reass
1091 * With scrub rules:
1092 * - packet normalization is performed if there is a matching scrub rule
1093 * - fragment reassembly is performed if the matching rule has no
1094 * PFRULE_FRAGMENT_NOREASS flag
1095 */
1096 scrub_compat = (r != NULL);
1097 while (r != NULL) {
1098 pf_counter_u64_add(&r->evaluations, 1);
1099 if (pfi_kkif_match(r->kif, pd->kif) == r->ifnot)
1100 r = r->skip[PF_SKIP_IFP];
1101 else if (r->direction && r->direction != pd->dir)
1102 r = r->skip[PF_SKIP_DIR];
1103 else if (r->af && r->af != AF_INET)
1104 r = r->skip[PF_SKIP_AF];
1105 else if (r->proto && r->proto != h->ip_p)
1106 r = r->skip[PF_SKIP_PROTO];
1107 else if (PF_MISMATCHAW(&r->src.addr,
1108 (struct pf_addr *)&h->ip_src.s_addr, AF_INET,
1109 r->src.neg, pd->kif, M_GETFIB(pd->m)))
1110 r = r->skip[PF_SKIP_SRC_ADDR];
1111 else if (PF_MISMATCHAW(&r->dst.addr,
1112 (struct pf_addr *)&h->ip_dst.s_addr, AF_INET,
1113 r->dst.neg, NULL, M_GETFIB(pd->m)))
1114 r = r->skip[PF_SKIP_DST_ADDR];
1115 else if (r->match_tag && !pf_match_tag(pd->m, r, &tag,
1116 pd->pf_mtag ? pd->pf_mtag->tag : 0))
1117 r = TAILQ_NEXT(r, entries);
1118 else
1119 break;
1120 }
1121
1122 if (scrub_compat) {
1123 /* With scrub rules present IPv4 normalization happens only
1124 * if one of rules has matched and it's not a "no scrub" rule */
1125 if (r == NULL || r->action == PF_NOSCRUB)
1126 return (PF_PASS);
1127
1128 pf_counter_u64_critical_enter();
1129 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1);
1130 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len);
1131 pf_counter_u64_critical_exit();
1132 pf_rule_to_actions(r, &pd->act);
1133 }
1134
1135 /* Check for illegal packets */
1136 if (hlen < (int)sizeof(struct ip)) {
1137 REASON_SET(reason, PFRES_NORM);
1138 goto drop;
1139 }
1140
1141 if (hlen > ntohs(h->ip_len)) {
1142 REASON_SET(reason, PFRES_NORM);
1143 goto drop;
1144 }
1145
1146 /* Clear IP_DF if the rule uses the no-df option or we're in no-df mode */
1147 if (((!scrub_compat && V_pf_status.reass & PF_REASS_NODF) ||
1148 (r != NULL && r->rule_flag & PFRULE_NODF)) &&
1149 (h->ip_off & htons(IP_DF))
1150 ) {
1151 u_int16_t ip_off = h->ip_off;
1152
1153 h->ip_off &= htons(~IP_DF);
1154 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
1155 }
1156
1157 /* We will need other tests here */
1158 if (!fragoff && !mff)
1159 goto no_fragment;
1160
1161 /* We're dealing with a fragment now. Don't allow fragments
1162 * with IP_DF to enter the cache. If the flag was cleared by
1163 * no-df above, fine. Otherwise drop it.
1164 */
1165 if (h->ip_off & htons(IP_DF)) {
1166 DPFPRINTF(("IP_DF\n"));
1167 goto bad;
1168 }
1169
1170 ip_len = ntohs(h->ip_len) - hlen;
1171
1172 /* All fragments are 8 byte aligned */
1173 if (mff && (ip_len & 0x7)) {
1174 DPFPRINTF(("mff and %d\n", ip_len));
1175 goto bad;
1176 }
1177
1178 /* Respect maximum length */
1179 if (fragoff + ip_len > IP_MAXPACKET) {
1180 DPFPRINTF(("max packet %d\n", fragoff + ip_len));
1181 goto bad;
1182 }
1183
1184 if ((!scrub_compat && V_pf_status.reass) ||
1185 (r != NULL && !(r->rule_flag & PFRULE_FRAGMENT_NOREASS))
1186 ) {
1187 max = fragoff + ip_len;
1188
1189 /* Fully buffer all of the fragments
1190 * Might return a completely reassembled mbuf, or NULL */
1191 PF_FRAG_LOCK();
1192 DPFPRINTF(("reass frag %d @ %d-%d\n", h->ip_id, fragoff, max));
1193 verdict = pf_reassemble(&pd->m, pd->dir, reason);
1194 PF_FRAG_UNLOCK();
1195
1196 if (verdict != PF_PASS)
1197 return (PF_DROP);
1198
1199 if (pd->m == NULL)
1200 return (PF_DROP);
1201
1202 h = mtod(pd->m, struct ip *);
1203 pd->tot_len = htons(h->ip_len);
1204
1205 no_fragment:
1206 /* At this point, only IP_DF is allowed in ip_off */
1207 if (h->ip_off & ~htons(IP_DF)) {
1208 u_int16_t ip_off = h->ip_off;
1209
1210 h->ip_off &= htons(IP_DF);
1211 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
1212 }
1213 }
1214
1215 return (PF_PASS);
1216
1217 bad:
1218 DPFPRINTF(("dropping bad fragment\n"));
1219 REASON_SET(reason, PFRES_FRAG);
1220 drop:
1221 if (r != NULL && r->log)
1222 PFLOG_PACKET(PF_DROP, *reason, r, NULL, NULL, pd, 1, NULL);
1223
1224 return (PF_DROP);
1225 }
1226 #endif
1227
1228 #ifdef INET6
1229 int
pf_normalize_ip6(int off,u_short * reason,struct pf_pdesc * pd)1230 pf_normalize_ip6(int off, u_short *reason,
1231 struct pf_pdesc *pd)
1232 {
1233 struct pf_krule *r;
1234 struct ip6_hdr *h;
1235 struct ip6_frag frag;
1236 bool scrub_compat;
1237
1238 PF_RULES_RASSERT();
1239
1240 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1241 /*
1242 * Check if there are any scrub rules, matching or not.
1243 * Lack of scrub rules means:
1244 * - enforced packet normalization operation just like in OpenBSD
1245 * With scrub rules:
1246 * - packet normalization is performed if there is a matching scrub rule
1247 * XXX: Fragment reassembly always performed for IPv6!
1248 */
1249 scrub_compat = (r != NULL);
1250 while (r != NULL) {
1251 pf_counter_u64_add(&r->evaluations, 1);
1252 if (pfi_kkif_match(r->kif, pd->kif) == r->ifnot)
1253 r = r->skip[PF_SKIP_IFP];
1254 else if (r->direction && r->direction != pd->dir)
1255 r = r->skip[PF_SKIP_DIR];
1256 else if (r->af && r->af != AF_INET6)
1257 r = r->skip[PF_SKIP_AF];
1258 else if (r->proto && r->proto != pd->proto)
1259 r = r->skip[PF_SKIP_PROTO];
1260 else if (PF_MISMATCHAW(&r->src.addr,
1261 (struct pf_addr *)&pd->src, AF_INET6,
1262 r->src.neg, pd->kif, M_GETFIB(pd->m)))
1263 r = r->skip[PF_SKIP_SRC_ADDR];
1264 else if (PF_MISMATCHAW(&r->dst.addr,
1265 (struct pf_addr *)&pd->dst, AF_INET6,
1266 r->dst.neg, NULL, M_GETFIB(pd->m)))
1267 r = r->skip[PF_SKIP_DST_ADDR];
1268 else
1269 break;
1270 }
1271
1272 if (scrub_compat) {
1273 /* With scrub rules present IPv6 normalization happens only
1274 * if one of rules has matched and it's not a "no scrub" rule */
1275 if (r == NULL || r->action == PF_NOSCRUB)
1276 return (PF_PASS);
1277
1278 pf_counter_u64_critical_enter();
1279 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1);
1280 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len);
1281 pf_counter_u64_critical_exit();
1282 pf_rule_to_actions(r, &pd->act);
1283 }
1284
1285 if (!pf_pull_hdr(pd->m, off, &frag, sizeof(frag), NULL, reason, AF_INET6))
1286 return (PF_DROP);
1287
1288 /* Offset now points to data portion. */
1289 off += sizeof(frag);
1290
1291 if (pd->virtual_proto == PF_VPROTO_FRAGMENT) {
1292 /* Returns PF_DROP or *m0 is NULL or completely reassembled
1293 * mbuf. */
1294 if (pf_reassemble6(&pd->m, &frag, off, pd->extoff, reason) != PF_PASS)
1295 return (PF_DROP);
1296 if (pd->m == NULL)
1297 return (PF_DROP);
1298 h = mtod(pd->m, struct ip6_hdr *);
1299 pd->tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr);
1300 }
1301
1302 return (PF_PASS);
1303 }
1304 #endif /* INET6 */
1305
1306 int
pf_normalize_tcp(struct pf_pdesc * pd)1307 pf_normalize_tcp(struct pf_pdesc *pd)
1308 {
1309 struct pf_krule *r, *rm = NULL;
1310 struct tcphdr *th = &pd->hdr.tcp;
1311 int rewrite = 0;
1312 u_short reason;
1313 u_int16_t flags;
1314 sa_family_t af = pd->af;
1315 int srs;
1316
1317 PF_RULES_RASSERT();
1318
1319 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1320 /* Check if there any scrub rules. Lack of scrub rules means enforced
1321 * packet normalization operation just like in OpenBSD. */
1322 srs = (r != NULL);
1323 while (r != NULL) {
1324 pf_counter_u64_add(&r->evaluations, 1);
1325 if (pfi_kkif_match(r->kif, pd->kif) == r->ifnot)
1326 r = r->skip[PF_SKIP_IFP];
1327 else if (r->direction && r->direction != pd->dir)
1328 r = r->skip[PF_SKIP_DIR];
1329 else if (r->af && r->af != af)
1330 r = r->skip[PF_SKIP_AF];
1331 else if (r->proto && r->proto != pd->proto)
1332 r = r->skip[PF_SKIP_PROTO];
1333 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
1334 r->src.neg, pd->kif, M_GETFIB(pd->m)))
1335 r = r->skip[PF_SKIP_SRC_ADDR];
1336 else if (r->src.port_op && !pf_match_port(r->src.port_op,
1337 r->src.port[0], r->src.port[1], th->th_sport))
1338 r = r->skip[PF_SKIP_SRC_PORT];
1339 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
1340 r->dst.neg, NULL, M_GETFIB(pd->m)))
1341 r = r->skip[PF_SKIP_DST_ADDR];
1342 else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
1343 r->dst.port[0], r->dst.port[1], th->th_dport))
1344 r = r->skip[PF_SKIP_DST_PORT];
1345 else if (r->os_fingerprint != PF_OSFP_ANY && !pf_osfp_match(
1346 pf_osfp_fingerprint(pd, th),
1347 r->os_fingerprint))
1348 r = TAILQ_NEXT(r, entries);
1349 else {
1350 rm = r;
1351 break;
1352 }
1353 }
1354
1355 if (srs) {
1356 /* With scrub rules present TCP normalization happens only
1357 * if one of rules has matched and it's not a "no scrub" rule */
1358 if (rm == NULL || rm->action == PF_NOSCRUB)
1359 return (PF_PASS);
1360
1361 pf_counter_u64_critical_enter();
1362 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1);
1363 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len);
1364 pf_counter_u64_critical_exit();
1365 pf_rule_to_actions(rm, &pd->act);
1366 }
1367
1368 if (rm && rm->rule_flag & PFRULE_REASSEMBLE_TCP)
1369 pd->flags |= PFDESC_TCP_NORM;
1370
1371 flags = tcp_get_flags(th);
1372 if (flags & TH_SYN) {
1373 /* Illegal packet */
1374 if (flags & TH_RST)
1375 goto tcp_drop;
1376
1377 if (flags & TH_FIN)
1378 goto tcp_drop;
1379 } else {
1380 /* Illegal packet */
1381 if (!(flags & (TH_ACK|TH_RST)))
1382 goto tcp_drop;
1383 }
1384
1385 if (!(flags & TH_ACK)) {
1386 /* These flags are only valid if ACK is set */
1387 if ((flags & TH_FIN) || (flags & TH_PUSH) || (flags & TH_URG))
1388 goto tcp_drop;
1389 }
1390
1391 /* Check for illegal header length */
1392 if (th->th_off < (sizeof(struct tcphdr) >> 2))
1393 goto tcp_drop;
1394
1395 /* If flags changed, or reserved data set, then adjust */
1396 if (flags != tcp_get_flags(th) ||
1397 (tcp_get_flags(th) & (TH_RES1|TH_RES2|TH_RES2)) != 0) {
1398 u_int16_t ov, nv;
1399
1400 ov = *(u_int16_t *)(&th->th_ack + 1);
1401 flags &= ~(TH_RES1 | TH_RES2 | TH_RES3);
1402 tcp_set_flags(th, flags);
1403 nv = *(u_int16_t *)(&th->th_ack + 1);
1404
1405 th->th_sum = pf_proto_cksum_fixup(pd->m, th->th_sum, ov, nv, 0);
1406 rewrite = 1;
1407 }
1408
1409 /* Remove urgent pointer, if TH_URG is not set */
1410 if (!(flags & TH_URG) && th->th_urp) {
1411 th->th_sum = pf_proto_cksum_fixup(pd->m, th->th_sum, th->th_urp,
1412 0, 0);
1413 th->th_urp = 0;
1414 rewrite = 1;
1415 }
1416
1417 /* copy back packet headers if we sanitized */
1418 if (rewrite)
1419 m_copyback(pd->m, pd->off, sizeof(*th), (caddr_t)th);
1420
1421 return (PF_PASS);
1422
1423 tcp_drop:
1424 REASON_SET(&reason, PFRES_NORM);
1425 if (rm != NULL && r->log)
1426 PFLOG_PACKET(PF_DROP, reason, r, NULL, NULL, pd, 1, NULL);
1427 return (PF_DROP);
1428 }
1429
1430 int
pf_normalize_tcp_init(struct pf_pdesc * pd,struct tcphdr * th,struct pf_state_peer * src,struct pf_state_peer * dst)1431 pf_normalize_tcp_init(struct pf_pdesc *pd, struct tcphdr *th,
1432 struct pf_state_peer *src, struct pf_state_peer *dst)
1433 {
1434 u_int32_t tsval, tsecr;
1435 u_int8_t hdr[60];
1436 u_int8_t *opt;
1437
1438 KASSERT((src->scrub == NULL),
1439 ("pf_normalize_tcp_init: src->scrub != NULL"));
1440
1441 src->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT);
1442 if (src->scrub == NULL)
1443 return (1);
1444
1445 switch (pd->af) {
1446 #ifdef INET
1447 case AF_INET: {
1448 struct ip *h = mtod(pd->m, struct ip *);
1449 src->scrub->pfss_ttl = h->ip_ttl;
1450 break;
1451 }
1452 #endif /* INET */
1453 #ifdef INET6
1454 case AF_INET6: {
1455 struct ip6_hdr *h = mtod(pd->m, struct ip6_hdr *);
1456 src->scrub->pfss_ttl = h->ip6_hlim;
1457 break;
1458 }
1459 #endif /* INET6 */
1460 }
1461
1462 /*
1463 * All normalizations below are only begun if we see the start of
1464 * the connections. They must all set an enabled bit in pfss_flags
1465 */
1466 if ((tcp_get_flags(th) & TH_SYN) == 0)
1467 return (0);
1468
1469 if (th->th_off > (sizeof(struct tcphdr) >> 2) && src->scrub &&
1470 pf_pull_hdr(pd->m, pd->off, hdr, th->th_off << 2, NULL, NULL, pd->af)) {
1471 /* Diddle with TCP options */
1472 int hlen;
1473 opt = hdr + sizeof(struct tcphdr);
1474 hlen = (th->th_off << 2) - sizeof(struct tcphdr);
1475 while (hlen >= TCPOLEN_TIMESTAMP) {
1476 switch (*opt) {
1477 case TCPOPT_EOL: /* FALLTHROUGH */
1478 case TCPOPT_NOP:
1479 opt++;
1480 hlen--;
1481 break;
1482 case TCPOPT_TIMESTAMP:
1483 if (opt[1] >= TCPOLEN_TIMESTAMP) {
1484 src->scrub->pfss_flags |=
1485 PFSS_TIMESTAMP;
1486 src->scrub->pfss_ts_mod =
1487 htonl(arc4random());
1488
1489 /* note PFSS_PAWS not set yet */
1490 memcpy(&tsval, &opt[2],
1491 sizeof(u_int32_t));
1492 memcpy(&tsecr, &opt[6],
1493 sizeof(u_int32_t));
1494 src->scrub->pfss_tsval0 = ntohl(tsval);
1495 src->scrub->pfss_tsval = ntohl(tsval);
1496 src->scrub->pfss_tsecr = ntohl(tsecr);
1497 getmicrouptime(&src->scrub->pfss_last);
1498 }
1499 /* FALLTHROUGH */
1500 default:
1501 hlen -= MAX(opt[1], 2);
1502 opt += MAX(opt[1], 2);
1503 break;
1504 }
1505 }
1506 }
1507
1508 return (0);
1509 }
1510
1511 void
pf_normalize_tcp_cleanup(struct pf_kstate * state)1512 pf_normalize_tcp_cleanup(struct pf_kstate *state)
1513 {
1514 /* XXX Note: this also cleans up SCTP. */
1515 uma_zfree(V_pf_state_scrub_z, state->src.scrub);
1516 uma_zfree(V_pf_state_scrub_z, state->dst.scrub);
1517
1518 /* Someday... flush the TCP segment reassembly descriptors. */
1519 }
1520 int
pf_normalize_sctp_init(struct pf_pdesc * pd,struct pf_state_peer * src,struct pf_state_peer * dst)1521 pf_normalize_sctp_init(struct pf_pdesc *pd, struct pf_state_peer *src,
1522 struct pf_state_peer *dst)
1523 {
1524 src->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT);
1525 if (src->scrub == NULL)
1526 return (1);
1527
1528 dst->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT);
1529 if (dst->scrub == NULL) {
1530 uma_zfree(V_pf_state_scrub_z, src);
1531 return (1);
1532 }
1533
1534 dst->scrub->pfss_v_tag = pd->sctp_initiate_tag;
1535
1536 return (0);
1537 }
1538
1539 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)1540 pf_normalize_tcp_stateful(struct pf_pdesc *pd,
1541 u_short *reason, struct tcphdr *th, struct pf_kstate *state,
1542 struct pf_state_peer *src, struct pf_state_peer *dst, int *writeback)
1543 {
1544 struct timeval uptime;
1545 u_int32_t tsval, tsecr;
1546 u_int tsval_from_last;
1547 u_int8_t hdr[60];
1548 u_int8_t *opt;
1549 int copyback = 0;
1550 int got_ts = 0;
1551 size_t startoff;
1552
1553 KASSERT((src->scrub || dst->scrub),
1554 ("%s: src->scrub && dst->scrub!", __func__));
1555
1556 /*
1557 * Enforce the minimum TTL seen for this connection. Negate a common
1558 * technique to evade an intrusion detection system and confuse
1559 * firewall state code.
1560 */
1561 switch (pd->af) {
1562 #ifdef INET
1563 case AF_INET: {
1564 if (src->scrub) {
1565 struct ip *h = mtod(pd->m, struct ip *);
1566 if (h->ip_ttl > src->scrub->pfss_ttl)
1567 src->scrub->pfss_ttl = h->ip_ttl;
1568 h->ip_ttl = src->scrub->pfss_ttl;
1569 }
1570 break;
1571 }
1572 #endif /* INET */
1573 #ifdef INET6
1574 case AF_INET6: {
1575 if (src->scrub) {
1576 struct ip6_hdr *h = mtod(pd->m, struct ip6_hdr *);
1577 if (h->ip6_hlim > src->scrub->pfss_ttl)
1578 src->scrub->pfss_ttl = h->ip6_hlim;
1579 h->ip6_hlim = src->scrub->pfss_ttl;
1580 }
1581 break;
1582 }
1583 #endif /* INET6 */
1584 }
1585
1586 if (th->th_off > (sizeof(struct tcphdr) >> 2) &&
1587 ((src->scrub && (src->scrub->pfss_flags & PFSS_TIMESTAMP)) ||
1588 (dst->scrub && (dst->scrub->pfss_flags & PFSS_TIMESTAMP))) &&
1589 pf_pull_hdr(pd->m, pd->off, hdr, th->th_off << 2, NULL, NULL, pd->af)) {
1590 /* Diddle with TCP options */
1591 int hlen;
1592 opt = hdr + sizeof(struct tcphdr);
1593 hlen = (th->th_off << 2) - sizeof(struct tcphdr);
1594 while (hlen >= TCPOLEN_TIMESTAMP) {
1595 startoff = opt - (hdr + sizeof(struct tcphdr));
1596 switch (*opt) {
1597 case TCPOPT_EOL: /* FALLTHROUGH */
1598 case TCPOPT_NOP:
1599 opt++;
1600 hlen--;
1601 break;
1602 case TCPOPT_TIMESTAMP:
1603 /* Modulate the timestamps. Can be used for
1604 * NAT detection, OS uptime determination or
1605 * reboot detection.
1606 */
1607
1608 if (got_ts) {
1609 /* Huh? Multiple timestamps!? */
1610 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1611 DPFPRINTF(("multiple TS??\n"));
1612 pf_print_state(state);
1613 printf("\n");
1614 }
1615 REASON_SET(reason, PFRES_TS);
1616 return (PF_DROP);
1617 }
1618 if (opt[1] >= TCPOLEN_TIMESTAMP) {
1619 memcpy(&tsval, &opt[2],
1620 sizeof(u_int32_t));
1621 if (tsval && src->scrub &&
1622 (src->scrub->pfss_flags &
1623 PFSS_TIMESTAMP)) {
1624 tsval = ntohl(tsval);
1625 pf_patch_32_unaligned(pd->m,
1626 &th->th_sum,
1627 &opt[2],
1628 htonl(tsval +
1629 src->scrub->pfss_ts_mod),
1630 PF_ALGNMNT(startoff),
1631 0);
1632 copyback = 1;
1633 }
1634
1635 /* Modulate TS reply iff valid (!0) */
1636 memcpy(&tsecr, &opt[6],
1637 sizeof(u_int32_t));
1638 if (tsecr && dst->scrub &&
1639 (dst->scrub->pfss_flags &
1640 PFSS_TIMESTAMP)) {
1641 tsecr = ntohl(tsecr)
1642 - dst->scrub->pfss_ts_mod;
1643 pf_patch_32_unaligned(pd->m,
1644 &th->th_sum,
1645 &opt[6],
1646 htonl(tsecr),
1647 PF_ALGNMNT(startoff),
1648 0);
1649 copyback = 1;
1650 }
1651 got_ts = 1;
1652 }
1653 /* FALLTHROUGH */
1654 default:
1655 hlen -= MAX(opt[1], 2);
1656 opt += MAX(opt[1], 2);
1657 break;
1658 }
1659 }
1660 if (copyback) {
1661 /* Copyback the options, caller copys back header */
1662 *writeback = 1;
1663 m_copyback(pd->m, pd->off + sizeof(struct tcphdr),
1664 (th->th_off << 2) - sizeof(struct tcphdr), hdr +
1665 sizeof(struct tcphdr));
1666 }
1667 }
1668
1669 /*
1670 * Must invalidate PAWS checks on connections idle for too long.
1671 * The fastest allowed timestamp clock is 1ms. That turns out to
1672 * be about 24 days before it wraps. XXX Right now our lowerbound
1673 * TS echo check only works for the first 12 days of a connection
1674 * when the TS has exhausted half its 32bit space
1675 */
1676 #define TS_MAX_IDLE (24*24*60*60)
1677 #define TS_MAX_CONN (12*24*60*60) /* XXX remove when better tsecr check */
1678
1679 getmicrouptime(&uptime);
1680 if (src->scrub && (src->scrub->pfss_flags & PFSS_PAWS) &&
1681 (uptime.tv_sec - src->scrub->pfss_last.tv_sec > TS_MAX_IDLE ||
1682 time_uptime - (state->creation / 1000) > TS_MAX_CONN)) {
1683 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1684 DPFPRINTF(("src idled out of PAWS\n"));
1685 pf_print_state(state);
1686 printf("\n");
1687 }
1688 src->scrub->pfss_flags = (src->scrub->pfss_flags & ~PFSS_PAWS)
1689 | PFSS_PAWS_IDLED;
1690 }
1691 if (dst->scrub && (dst->scrub->pfss_flags & PFSS_PAWS) &&
1692 uptime.tv_sec - dst->scrub->pfss_last.tv_sec > TS_MAX_IDLE) {
1693 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1694 DPFPRINTF(("dst idled out of PAWS\n"));
1695 pf_print_state(state);
1696 printf("\n");
1697 }
1698 dst->scrub->pfss_flags = (dst->scrub->pfss_flags & ~PFSS_PAWS)
1699 | PFSS_PAWS_IDLED;
1700 }
1701
1702 if (got_ts && src->scrub && dst->scrub &&
1703 (src->scrub->pfss_flags & PFSS_PAWS) &&
1704 (dst->scrub->pfss_flags & PFSS_PAWS)) {
1705 /* Validate that the timestamps are "in-window".
1706 * RFC1323 describes TCP Timestamp options that allow
1707 * measurement of RTT (round trip time) and PAWS
1708 * (protection against wrapped sequence numbers). PAWS
1709 * gives us a set of rules for rejecting packets on
1710 * long fat pipes (packets that were somehow delayed
1711 * in transit longer than the time it took to send the
1712 * full TCP sequence space of 4Gb). We can use these
1713 * rules and infer a few others that will let us treat
1714 * the 32bit timestamp and the 32bit echoed timestamp
1715 * as sequence numbers to prevent a blind attacker from
1716 * inserting packets into a connection.
1717 *
1718 * RFC1323 tells us:
1719 * - The timestamp on this packet must be greater than
1720 * or equal to the last value echoed by the other
1721 * endpoint. The RFC says those will be discarded
1722 * since it is a dup that has already been acked.
1723 * This gives us a lowerbound on the timestamp.
1724 * timestamp >= other last echoed timestamp
1725 * - The timestamp will be less than or equal to
1726 * the last timestamp plus the time between the
1727 * last packet and now. The RFC defines the max
1728 * clock rate as 1ms. We will allow clocks to be
1729 * up to 10% fast and will allow a total difference
1730 * or 30 seconds due to a route change. And this
1731 * gives us an upperbound on the timestamp.
1732 * timestamp <= last timestamp + max ticks
1733 * We have to be careful here. Windows will send an
1734 * initial timestamp of zero and then initialize it
1735 * to a random value after the 3whs; presumably to
1736 * avoid a DoS by having to call an expensive RNG
1737 * during a SYN flood. Proof MS has at least one
1738 * good security geek.
1739 *
1740 * - The TCP timestamp option must also echo the other
1741 * endpoints timestamp. The timestamp echoed is the
1742 * one carried on the earliest unacknowledged segment
1743 * on the left edge of the sequence window. The RFC
1744 * states that the host will reject any echoed
1745 * timestamps that were larger than any ever sent.
1746 * This gives us an upperbound on the TS echo.
1747 * tescr <= largest_tsval
1748 * - The lowerbound on the TS echo is a little more
1749 * tricky to determine. The other endpoint's echoed
1750 * values will not decrease. But there may be
1751 * network conditions that re-order packets and
1752 * cause our view of them to decrease. For now the
1753 * only lowerbound we can safely determine is that
1754 * the TS echo will never be less than the original
1755 * TS. XXX There is probably a better lowerbound.
1756 * Remove TS_MAX_CONN with better lowerbound check.
1757 * tescr >= other original TS
1758 *
1759 * It is also important to note that the fastest
1760 * timestamp clock of 1ms will wrap its 32bit space in
1761 * 24 days. So we just disable TS checking after 24
1762 * days of idle time. We actually must use a 12d
1763 * connection limit until we can come up with a better
1764 * lowerbound to the TS echo check.
1765 */
1766 struct timeval delta_ts;
1767 int ts_fudge;
1768
1769 /*
1770 * PFTM_TS_DIFF is how many seconds of leeway to allow
1771 * a host's timestamp. This can happen if the previous
1772 * packet got delayed in transit for much longer than
1773 * this packet.
1774 */
1775 if ((ts_fudge = state->rule->timeout[PFTM_TS_DIFF]) == 0)
1776 ts_fudge = V_pf_default_rule.timeout[PFTM_TS_DIFF];
1777
1778 /* Calculate max ticks since the last timestamp */
1779 #define TS_MAXFREQ 1100 /* RFC max TS freq of 1Khz + 10% skew */
1780 #define TS_MICROSECS 1000000 /* microseconds per second */
1781 delta_ts = uptime;
1782 timevalsub(&delta_ts, &src->scrub->pfss_last);
1783 tsval_from_last = (delta_ts.tv_sec + ts_fudge) * TS_MAXFREQ;
1784 tsval_from_last += delta_ts.tv_usec / (TS_MICROSECS/TS_MAXFREQ);
1785
1786 if ((src->state >= TCPS_ESTABLISHED &&
1787 dst->state >= TCPS_ESTABLISHED) &&
1788 (SEQ_LT(tsval, dst->scrub->pfss_tsecr) ||
1789 SEQ_GT(tsval, src->scrub->pfss_tsval + tsval_from_last) ||
1790 (tsecr && (SEQ_GT(tsecr, dst->scrub->pfss_tsval) ||
1791 SEQ_LT(tsecr, dst->scrub->pfss_tsval0))))) {
1792 /* Bad RFC1323 implementation or an insertion attack.
1793 *
1794 * - Solaris 2.6 and 2.7 are known to send another ACK
1795 * after the FIN,FIN|ACK,ACK closing that carries
1796 * an old timestamp.
1797 */
1798
1799 DPFPRINTF(("Timestamp failed %c%c%c%c\n",
1800 SEQ_LT(tsval, dst->scrub->pfss_tsecr) ? '0' : ' ',
1801 SEQ_GT(tsval, src->scrub->pfss_tsval +
1802 tsval_from_last) ? '1' : ' ',
1803 SEQ_GT(tsecr, dst->scrub->pfss_tsval) ? '2' : ' ',
1804 SEQ_LT(tsecr, dst->scrub->pfss_tsval0)? '3' : ' '));
1805 DPFPRINTF((" tsval: %u tsecr: %u +ticks: %u "
1806 "idle: %jus %lums\n",
1807 tsval, tsecr, tsval_from_last,
1808 (uintmax_t)delta_ts.tv_sec,
1809 delta_ts.tv_usec / 1000));
1810 DPFPRINTF((" src->tsval: %u tsecr: %u\n",
1811 src->scrub->pfss_tsval, src->scrub->pfss_tsecr));
1812 DPFPRINTF((" dst->tsval: %u tsecr: %u tsval0: %u"
1813 "\n", dst->scrub->pfss_tsval,
1814 dst->scrub->pfss_tsecr, dst->scrub->pfss_tsval0));
1815 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1816 pf_print_state(state);
1817 pf_print_flags(tcp_get_flags(th));
1818 printf("\n");
1819 }
1820 REASON_SET(reason, PFRES_TS);
1821 return (PF_DROP);
1822 }
1823
1824 /* XXX I'd really like to require tsecr but it's optional */
1825
1826 } else if (!got_ts && (tcp_get_flags(th) & TH_RST) == 0 &&
1827 ((src->state == TCPS_ESTABLISHED && dst->state == TCPS_ESTABLISHED)
1828 || pd->p_len > 0 || (tcp_get_flags(th) & TH_SYN)) &&
1829 src->scrub && dst->scrub &&
1830 (src->scrub->pfss_flags & PFSS_PAWS) &&
1831 (dst->scrub->pfss_flags & PFSS_PAWS)) {
1832 /* Didn't send a timestamp. Timestamps aren't really useful
1833 * when:
1834 * - connection opening or closing (often not even sent).
1835 * but we must not let an attacker to put a FIN on a
1836 * data packet to sneak it through our ESTABLISHED check.
1837 * - on a TCP reset. RFC suggests not even looking at TS.
1838 * - on an empty ACK. The TS will not be echoed so it will
1839 * probably not help keep the RTT calculation in sync and
1840 * there isn't as much danger when the sequence numbers
1841 * got wrapped. So some stacks don't include TS on empty
1842 * ACKs :-(
1843 *
1844 * To minimize the disruption to mostly RFC1323 conformant
1845 * stacks, we will only require timestamps on data packets.
1846 *
1847 * And what do ya know, we cannot require timestamps on data
1848 * packets. There appear to be devices that do legitimate
1849 * TCP connection hijacking. There are HTTP devices that allow
1850 * a 3whs (with timestamps) and then buffer the HTTP request.
1851 * If the intermediate device has the HTTP response cache, it
1852 * will spoof the response but not bother timestamping its
1853 * packets. So we can look for the presence of a timestamp in
1854 * the first data packet and if there, require it in all future
1855 * packets.
1856 */
1857
1858 if (pd->p_len > 0 && (src->scrub->pfss_flags & PFSS_DATA_TS)) {
1859 /*
1860 * Hey! Someone tried to sneak a packet in. Or the
1861 * stack changed its RFC1323 behavior?!?!
1862 */
1863 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1864 DPFPRINTF(("Did not receive expected RFC1323 "
1865 "timestamp\n"));
1866 pf_print_state(state);
1867 pf_print_flags(tcp_get_flags(th));
1868 printf("\n");
1869 }
1870 REASON_SET(reason, PFRES_TS);
1871 return (PF_DROP);
1872 }
1873 }
1874
1875 /*
1876 * We will note if a host sends his data packets with or without
1877 * timestamps. And require all data packets to contain a timestamp
1878 * if the first does. PAWS implicitly requires that all data packets be
1879 * timestamped. But I think there are middle-man devices that hijack
1880 * TCP streams immediately after the 3whs and don't timestamp their
1881 * packets (seen in a WWW accelerator or cache).
1882 */
1883 if (pd->p_len > 0 && src->scrub && (src->scrub->pfss_flags &
1884 (PFSS_TIMESTAMP|PFSS_DATA_TS|PFSS_DATA_NOTS)) == PFSS_TIMESTAMP) {
1885 if (got_ts)
1886 src->scrub->pfss_flags |= PFSS_DATA_TS;
1887 else {
1888 src->scrub->pfss_flags |= PFSS_DATA_NOTS;
1889 if (V_pf_status.debug >= PF_DEBUG_MISC && dst->scrub &&
1890 (dst->scrub->pfss_flags & PFSS_TIMESTAMP)) {
1891 /* Don't warn if other host rejected RFC1323 */
1892 DPFPRINTF(("Broken RFC1323 stack did not "
1893 "timestamp data packet. Disabled PAWS "
1894 "security.\n"));
1895 pf_print_state(state);
1896 pf_print_flags(tcp_get_flags(th));
1897 printf("\n");
1898 }
1899 }
1900 }
1901
1902 /*
1903 * Update PAWS values
1904 */
1905 if (got_ts && src->scrub && PFSS_TIMESTAMP == (src->scrub->pfss_flags &
1906 (PFSS_PAWS_IDLED|PFSS_TIMESTAMP))) {
1907 getmicrouptime(&src->scrub->pfss_last);
1908 if (SEQ_GEQ(tsval, src->scrub->pfss_tsval) ||
1909 (src->scrub->pfss_flags & PFSS_PAWS) == 0)
1910 src->scrub->pfss_tsval = tsval;
1911
1912 if (tsecr) {
1913 if (SEQ_GEQ(tsecr, src->scrub->pfss_tsecr) ||
1914 (src->scrub->pfss_flags & PFSS_PAWS) == 0)
1915 src->scrub->pfss_tsecr = tsecr;
1916
1917 if ((src->scrub->pfss_flags & PFSS_PAWS) == 0 &&
1918 (SEQ_LT(tsval, src->scrub->pfss_tsval0) ||
1919 src->scrub->pfss_tsval0 == 0)) {
1920 /* tsval0 MUST be the lowest timestamp */
1921 src->scrub->pfss_tsval0 = tsval;
1922 }
1923
1924 /* Only fully initialized after a TS gets echoed */
1925 if ((src->scrub->pfss_flags & PFSS_PAWS) == 0)
1926 src->scrub->pfss_flags |= PFSS_PAWS;
1927 }
1928 }
1929
1930 /* I have a dream.... TCP segment reassembly.... */
1931 return (0);
1932 }
1933
1934 int
pf_normalize_mss(struct pf_pdesc * pd)1935 pf_normalize_mss(struct pf_pdesc *pd)
1936 {
1937 struct tcphdr *th = &pd->hdr.tcp;
1938 u_int16_t *mss;
1939 int thoff;
1940 int opt, cnt, optlen = 0;
1941 u_char opts[TCP_MAXOLEN];
1942 u_char *optp = opts;
1943 size_t startoff;
1944
1945 thoff = th->th_off << 2;
1946 cnt = thoff - sizeof(struct tcphdr);
1947
1948 if (cnt <= 0 || cnt > MAX_TCPOPTLEN || !pf_pull_hdr(pd->m,
1949 pd->off + sizeof(*th), opts, cnt, NULL, NULL, pd->af))
1950 return (0);
1951
1952 for (; cnt > 0; cnt -= optlen, optp += optlen) {
1953 startoff = optp - opts;
1954 opt = optp[0];
1955 if (opt == TCPOPT_EOL)
1956 break;
1957 if (opt == TCPOPT_NOP)
1958 optlen = 1;
1959 else {
1960 if (cnt < 2)
1961 break;
1962 optlen = optp[1];
1963 if (optlen < 2 || optlen > cnt)
1964 break;
1965 }
1966 switch (opt) {
1967 case TCPOPT_MAXSEG:
1968 mss = (u_int16_t *)(optp + 2);
1969 if ((ntohs(*mss)) > pd->act.max_mss) {
1970 pf_patch_16_unaligned(pd->m,
1971 &th->th_sum,
1972 mss, htons(pd->act.max_mss),
1973 PF_ALGNMNT(startoff),
1974 0);
1975 m_copyback(pd->m, pd->off + sizeof(*th),
1976 thoff - sizeof(*th), opts);
1977 m_copyback(pd->m, pd->off, sizeof(*th), (caddr_t)th);
1978 }
1979 break;
1980 default:
1981 break;
1982 }
1983 }
1984
1985 return (0);
1986 }
1987
1988 int
pf_scan_sctp(struct pf_pdesc * pd)1989 pf_scan_sctp(struct pf_pdesc *pd)
1990 {
1991 struct sctp_chunkhdr ch = { };
1992 int chunk_off = sizeof(struct sctphdr);
1993 int chunk_start;
1994 int ret;
1995
1996 while (pd->off + chunk_off < pd->tot_len) {
1997 if (!pf_pull_hdr(pd->m, pd->off + chunk_off, &ch, sizeof(ch), NULL,
1998 NULL, pd->af))
1999 return (PF_DROP);
2000
2001 /* Length includes the header, this must be at least 4. */
2002 if (ntohs(ch.chunk_length) < 4)
2003 return (PF_DROP);
2004
2005 chunk_start = chunk_off;
2006 chunk_off += roundup(ntohs(ch.chunk_length), 4);
2007
2008 switch (ch.chunk_type) {
2009 case SCTP_INITIATION:
2010 case SCTP_INITIATION_ACK: {
2011 struct sctp_init_chunk init;
2012
2013 if (!pf_pull_hdr(pd->m, pd->off + chunk_start, &init,
2014 sizeof(init), NULL, NULL, pd->af))
2015 return (PF_DROP);
2016
2017 /*
2018 * RFC 9620, Section 3.3.2, "The Initiate Tag is allowed to have
2019 * any value except 0."
2020 */
2021 if (init.init.initiate_tag == 0)
2022 return (PF_DROP);
2023 if (init.init.num_inbound_streams == 0)
2024 return (PF_DROP);
2025 if (init.init.num_outbound_streams == 0)
2026 return (PF_DROP);
2027 if (ntohl(init.init.a_rwnd) < SCTP_MIN_RWND)
2028 return (PF_DROP);
2029
2030 /*
2031 * RFC 9260, Section 3.1, INIT chunks MUST have zero
2032 * verification tag.
2033 */
2034 if (ch.chunk_type == SCTP_INITIATION &&
2035 pd->hdr.sctp.v_tag != 0)
2036 return (PF_DROP);
2037
2038 pd->sctp_initiate_tag = init.init.initiate_tag;
2039
2040 if (ch.chunk_type == SCTP_INITIATION)
2041 pd->sctp_flags |= PFDESC_SCTP_INIT;
2042 else
2043 pd->sctp_flags |= PFDESC_SCTP_INIT_ACK;
2044
2045 ret = pf_multihome_scan_init(pd->off + chunk_start,
2046 ntohs(init.ch.chunk_length), pd);
2047 if (ret != PF_PASS)
2048 return (ret);
2049
2050 break;
2051 }
2052 case SCTP_ABORT_ASSOCIATION:
2053 pd->sctp_flags |= PFDESC_SCTP_ABORT;
2054 break;
2055 case SCTP_SHUTDOWN:
2056 case SCTP_SHUTDOWN_ACK:
2057 pd->sctp_flags |= PFDESC_SCTP_SHUTDOWN;
2058 break;
2059 case SCTP_SHUTDOWN_COMPLETE:
2060 pd->sctp_flags |= PFDESC_SCTP_SHUTDOWN_COMPLETE;
2061 break;
2062 case SCTP_COOKIE_ECHO:
2063 pd->sctp_flags |= PFDESC_SCTP_COOKIE;
2064 break;
2065 case SCTP_COOKIE_ACK:
2066 pd->sctp_flags |= PFDESC_SCTP_COOKIE_ACK;
2067 break;
2068 case SCTP_DATA:
2069 pd->sctp_flags |= PFDESC_SCTP_DATA;
2070 break;
2071 case SCTP_HEARTBEAT_REQUEST:
2072 pd->sctp_flags |= PFDESC_SCTP_HEARTBEAT;
2073 break;
2074 case SCTP_HEARTBEAT_ACK:
2075 pd->sctp_flags |= PFDESC_SCTP_HEARTBEAT_ACK;
2076 break;
2077 case SCTP_ASCONF:
2078 pd->sctp_flags |= PFDESC_SCTP_ASCONF;
2079
2080 ret = pf_multihome_scan_asconf(pd->off + chunk_start,
2081 ntohs(ch.chunk_length), pd);
2082 if (ret != PF_PASS)
2083 return (ret);
2084 break;
2085 default:
2086 pd->sctp_flags |= PFDESC_SCTP_OTHER;
2087 break;
2088 }
2089 }
2090
2091 /* Validate chunk lengths vs. packet length. */
2092 if (pd->off + chunk_off != pd->tot_len)
2093 return (PF_DROP);
2094
2095 /*
2096 * INIT, INIT_ACK or SHUTDOWN_COMPLETE chunks must always be the only
2097 * one in a packet.
2098 */
2099 if ((pd->sctp_flags & PFDESC_SCTP_INIT) &&
2100 (pd->sctp_flags & ~PFDESC_SCTP_INIT))
2101 return (PF_DROP);
2102 if ((pd->sctp_flags & PFDESC_SCTP_INIT_ACK) &&
2103 (pd->sctp_flags & ~PFDESC_SCTP_INIT_ACK))
2104 return (PF_DROP);
2105 if ((pd->sctp_flags & PFDESC_SCTP_SHUTDOWN_COMPLETE) &&
2106 (pd->sctp_flags & ~PFDESC_SCTP_SHUTDOWN_COMPLETE))
2107 return (PF_DROP);
2108 if ((pd->sctp_flags & PFDESC_SCTP_ABORT) &&
2109 (pd->sctp_flags & PFDESC_SCTP_DATA)) {
2110 /*
2111 * RFC4960 3.3.7: DATA chunks MUST NOT be
2112 * bundled with ABORT.
2113 */
2114 return (PF_DROP);
2115 }
2116
2117 return (PF_PASS);
2118 }
2119
2120 int
pf_normalize_sctp(struct pf_pdesc * pd)2121 pf_normalize_sctp(struct pf_pdesc *pd)
2122 {
2123 struct pf_krule *r, *rm = NULL;
2124 struct sctphdr *sh = &pd->hdr.sctp;
2125 u_short reason;
2126 sa_family_t af = pd->af;
2127 int srs;
2128
2129 PF_RULES_RASSERT();
2130
2131 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
2132 /* Check if there any scrub rules. Lack of scrub rules means enforced
2133 * packet normalization operation just like in OpenBSD. */
2134 srs = (r != NULL);
2135 while (r != NULL) {
2136 pf_counter_u64_add(&r->evaluations, 1);
2137 if (pfi_kkif_match(r->kif, pd->kif) == r->ifnot)
2138 r = r->skip[PF_SKIP_IFP];
2139 else if (r->direction && r->direction != pd->dir)
2140 r = r->skip[PF_SKIP_DIR];
2141 else if (r->af && r->af != af)
2142 r = r->skip[PF_SKIP_AF];
2143 else if (r->proto && r->proto != pd->proto)
2144 r = r->skip[PF_SKIP_PROTO];
2145 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
2146 r->src.neg, pd->kif, M_GETFIB(pd->m)))
2147 r = r->skip[PF_SKIP_SRC_ADDR];
2148 else if (r->src.port_op && !pf_match_port(r->src.port_op,
2149 r->src.port[0], r->src.port[1], sh->src_port))
2150 r = r->skip[PF_SKIP_SRC_PORT];
2151 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
2152 r->dst.neg, NULL, M_GETFIB(pd->m)))
2153 r = r->skip[PF_SKIP_DST_ADDR];
2154 else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
2155 r->dst.port[0], r->dst.port[1], sh->dest_port))
2156 r = r->skip[PF_SKIP_DST_PORT];
2157 else {
2158 rm = r;
2159 break;
2160 }
2161 }
2162
2163 if (srs) {
2164 /* With scrub rules present SCTP normalization happens only
2165 * if one of rules has matched and it's not a "no scrub" rule */
2166 if (rm == NULL || rm->action == PF_NOSCRUB)
2167 return (PF_PASS);
2168
2169 pf_counter_u64_critical_enter();
2170 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1);
2171 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len);
2172 pf_counter_u64_critical_exit();
2173 }
2174
2175 /* Verify we're a multiple of 4 bytes long */
2176 if ((pd->tot_len - pd->off - sizeof(struct sctphdr)) % 4)
2177 goto sctp_drop;
2178
2179 /* INIT chunk needs to be the only chunk */
2180 if (pd->sctp_flags & PFDESC_SCTP_INIT)
2181 if (pd->sctp_flags & ~PFDESC_SCTP_INIT)
2182 goto sctp_drop;
2183
2184 return (PF_PASS);
2185
2186 sctp_drop:
2187 REASON_SET(&reason, PFRES_NORM);
2188 if (rm != NULL && r->log)
2189 PFLOG_PACKET(PF_DROP, reason, r, NULL, NULL, pd,
2190 1, NULL);
2191
2192 return (PF_DROP);
2193 }
2194
2195 #if defined(INET) || defined(INET6)
2196 void
pf_scrub(struct pf_pdesc * pd)2197 pf_scrub(struct pf_pdesc *pd)
2198 {
2199
2200 struct ip *h = mtod(pd->m, struct ip *);
2201 #ifdef INET6
2202 struct ip6_hdr *h6 = mtod(pd->m, struct ip6_hdr *);
2203 #endif
2204
2205 /* Clear IP_DF if no-df was requested */
2206 if (pd->af == AF_INET && pd->act.flags & PFSTATE_NODF &&
2207 h->ip_off & htons(IP_DF))
2208 {
2209 u_int16_t ip_off = h->ip_off;
2210
2211 h->ip_off &= htons(~IP_DF);
2212 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
2213 }
2214
2215 /* Enforce a minimum ttl, may cause endless packet loops */
2216 if (pd->af == AF_INET && pd->act.min_ttl &&
2217 h->ip_ttl < pd->act.min_ttl) {
2218 u_int16_t ip_ttl = h->ip_ttl;
2219
2220 h->ip_ttl = pd->act.min_ttl;
2221 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_ttl, h->ip_ttl, 0);
2222 }
2223 #ifdef INET6
2224 /* Enforce a minimum ttl, may cause endless packet loops */
2225 if (pd->af == AF_INET6 && pd->act.min_ttl &&
2226 h6->ip6_hlim < pd->act.min_ttl)
2227 h6->ip6_hlim = pd->act.min_ttl;
2228 #endif
2229 /* Enforce tos */
2230 if (pd->act.flags & PFSTATE_SETTOS) {
2231 switch (pd->af) {
2232 case AF_INET: {
2233 u_int16_t ov, nv;
2234
2235 ov = *(u_int16_t *)h;
2236 h->ip_tos = pd->act.set_tos | (h->ip_tos & IPTOS_ECN_MASK);
2237 nv = *(u_int16_t *)h;
2238
2239 h->ip_sum = pf_cksum_fixup(h->ip_sum, ov, nv, 0);
2240 break;
2241 }
2242 #ifdef INET6
2243 case AF_INET6:
2244 h6->ip6_flow &= IPV6_FLOWLABEL_MASK | IPV6_VERSION_MASK;
2245 h6->ip6_flow |= htonl((pd->act.set_tos | IPV6_ECN(h6)) << 20);
2246 break;
2247 #endif
2248 }
2249 }
2250
2251 /* random-id, but not for fragments */
2252 #ifdef INET
2253 if (pd->af == AF_INET &&
2254 pd->act.flags & PFSTATE_RANDOMID && !(h->ip_off & ~htons(IP_DF))) {
2255 uint16_t ip_id = h->ip_id;
2256
2257 ip_fillid(h, V_ip_random_id);
2258 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_id, h->ip_id, 0);
2259 }
2260 #endif
2261 }
2262 #endif
2263