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 <netinet/tcp.h>
54 #include <netinet/tcp_fsm.h>
55 #include <netinet/tcp_seq.h>
56 #include <netinet/sctp_constants.h>
57 #include <netinet/sctp_header.h>
58
59 #ifdef INET6
60 #include <netinet/ip6.h>
61 #endif /* INET6 */
62
63 struct pf_frent {
64 TAILQ_ENTRY(pf_frent) fr_next;
65 struct mbuf *fe_m;
66 uint16_t fe_hdrlen; /* ipv4 header length with ip options
67 ipv6, extension, fragment header */
68 uint16_t fe_extoff; /* last extension header offset or 0 */
69 uint16_t fe_len; /* fragment length */
70 uint16_t fe_off; /* fragment offset */
71 uint16_t fe_mff; /* more fragment flag */
72 };
73
74 struct pf_fragment_cmp {
75 struct pf_addr frc_src;
76 struct pf_addr frc_dst;
77 uint32_t frc_id;
78 sa_family_t frc_af;
79 uint8_t frc_proto;
80 };
81
82 struct pf_fragment {
83 struct pf_fragment_cmp fr_key;
84 #define fr_src fr_key.frc_src
85 #define fr_dst fr_key.frc_dst
86 #define fr_id fr_key.frc_id
87 #define fr_af fr_key.frc_af
88 #define fr_proto fr_key.frc_proto
89
90 /* pointers to queue element */
91 struct pf_frent *fr_firstoff[PF_FRAG_ENTRY_POINTS];
92 /* count entries between pointers */
93 uint8_t fr_entries[PF_FRAG_ENTRY_POINTS];
94 RB_ENTRY(pf_fragment) fr_entry;
95 TAILQ_ENTRY(pf_fragment) frag_next;
96 uint32_t fr_timeout;
97 TAILQ_HEAD(pf_fragq, pf_frent) fr_queue;
98 uint16_t fr_maxlen; /* maximum length of single fragment */
99 u_int16_t fr_holes; /* number of holes in the queue */
100 };
101
102 struct pf_fragment_tag {
103 uint16_t ft_hdrlen; /* header length of reassembled pkt */
104 uint16_t ft_extoff; /* last extension header offset or 0 */
105 uint16_t ft_maxlen; /* maximum fragment payload length */
106 uint32_t ft_id; /* fragment id */
107 };
108
109 VNET_DEFINE_STATIC(struct mtx, pf_frag_mtx);
110 #define V_pf_frag_mtx VNET(pf_frag_mtx)
111 #define PF_FRAG_LOCK() mtx_lock(&V_pf_frag_mtx)
112 #define PF_FRAG_UNLOCK() mtx_unlock(&V_pf_frag_mtx)
113 #define PF_FRAG_ASSERT() mtx_assert(&V_pf_frag_mtx, MA_OWNED)
114
115 VNET_DEFINE(uma_zone_t, pf_state_scrub_z); /* XXX: shared with pfsync */
116
117 VNET_DEFINE_STATIC(uma_zone_t, pf_frent_z);
118 #define V_pf_frent_z VNET(pf_frent_z)
119 VNET_DEFINE_STATIC(uma_zone_t, pf_frag_z);
120 #define V_pf_frag_z VNET(pf_frag_z)
121
122 TAILQ_HEAD(pf_fragqueue, pf_fragment);
123 TAILQ_HEAD(pf_cachequeue, pf_fragment);
124 VNET_DEFINE_STATIC(struct pf_fragqueue, pf_fragqueue);
125 #define V_pf_fragqueue VNET(pf_fragqueue)
126 RB_HEAD(pf_frag_tree, pf_fragment);
127 VNET_DEFINE_STATIC(struct pf_frag_tree, pf_frag_tree);
128 #define V_pf_frag_tree VNET(pf_frag_tree)
129 static int pf_frag_compare(struct pf_fragment *,
130 struct pf_fragment *);
131 static RB_PROTOTYPE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
132 static RB_GENERATE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
133
134 static void pf_flush_fragments(void);
135 static void pf_free_fragment(struct pf_fragment *);
136 static void pf_remove_fragment(struct pf_fragment *);
137 static int pf_normalize_tcpopt(struct pf_krule *, struct mbuf *,
138 struct tcphdr *, int, sa_family_t);
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 void pf_scrub_ip(struct mbuf **, uint32_t, uint8_t, uint8_t);
155 static int pf_reassemble(struct mbuf **, struct ip *, int, u_short *);
156 #endif /* INET */
157 #ifdef INET6
158 static int pf_reassemble6(struct mbuf **, struct ip6_hdr *,
159 struct ip6_frag *, uint16_t, uint16_t, u_short *);
160 static void pf_scrub_ip6(struct mbuf **, uint8_t);
161 #endif /* INET6 */
162
163 #define DPFPRINTF(x) do { \
164 if (V_pf_status.debug >= PF_DEBUG_MISC) { \
165 printf("%s: ", __func__); \
166 printf x ; \
167 } \
168 } while(0)
169
170 #ifdef INET
171 static void
pf_ip2key(struct ip * ip,int dir,struct pf_fragment_cmp * key)172 pf_ip2key(struct ip *ip, int dir, struct pf_fragment_cmp *key)
173 {
174
175 key->frc_src.v4 = ip->ip_src;
176 key->frc_dst.v4 = ip->ip_dst;
177 key->frc_af = AF_INET;
178 key->frc_proto = ip->ip_p;
179 key->frc_id = ip->ip_id;
180 }
181 #endif /* INET */
182
183 void
pf_normalize_init(void)184 pf_normalize_init(void)
185 {
186
187 V_pf_frag_z = uma_zcreate("pf frags", sizeof(struct pf_fragment),
188 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
189 V_pf_frent_z = uma_zcreate("pf frag entries", sizeof(struct pf_frent),
190 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
191 V_pf_state_scrub_z = uma_zcreate("pf state scrubs",
192 sizeof(struct pf_state_scrub), NULL, NULL, NULL, NULL,
193 UMA_ALIGN_PTR, 0);
194
195 mtx_init(&V_pf_frag_mtx, "pf fragments", NULL, MTX_DEF);
196
197 V_pf_limits[PF_LIMIT_FRAGS].zone = V_pf_frent_z;
198 V_pf_limits[PF_LIMIT_FRAGS].limit = PFFRAG_FRENT_HIWAT;
199 uma_zone_set_max(V_pf_frent_z, PFFRAG_FRENT_HIWAT);
200 uma_zone_set_warning(V_pf_frent_z, "PF frag entries limit reached");
201
202 TAILQ_INIT(&V_pf_fragqueue);
203 }
204
205 void
pf_normalize_cleanup(void)206 pf_normalize_cleanup(void)
207 {
208
209 uma_zdestroy(V_pf_state_scrub_z);
210 uma_zdestroy(V_pf_frent_z);
211 uma_zdestroy(V_pf_frag_z);
212
213 mtx_destroy(&V_pf_frag_mtx);
214 }
215
216 static int
pf_frag_compare(struct pf_fragment * a,struct pf_fragment * b)217 pf_frag_compare(struct pf_fragment *a, struct pf_fragment *b)
218 {
219 int diff;
220
221 if ((diff = a->fr_id - b->fr_id) != 0)
222 return (diff);
223 if ((diff = a->fr_proto - b->fr_proto) != 0)
224 return (diff);
225 if ((diff = a->fr_af - b->fr_af) != 0)
226 return (diff);
227 if ((diff = pf_addr_cmp(&a->fr_src, &b->fr_src, a->fr_af)) != 0)
228 return (diff);
229 if ((diff = pf_addr_cmp(&a->fr_dst, &b->fr_dst, a->fr_af)) != 0)
230 return (diff);
231 return (0);
232 }
233
234 void
pf_purge_expired_fragments(void)235 pf_purge_expired_fragments(void)
236 {
237 u_int32_t expire = time_uptime -
238 V_pf_default_rule.timeout[PFTM_FRAG];
239
240 pf_purge_fragments(expire);
241 }
242
243 void
pf_purge_fragments(uint32_t expire)244 pf_purge_fragments(uint32_t expire)
245 {
246 struct pf_fragment *frag;
247
248 PF_FRAG_LOCK();
249 while ((frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue)) != NULL) {
250 if (frag->fr_timeout > expire)
251 break;
252
253 DPFPRINTF(("expiring %d(%p)\n", frag->fr_id, frag));
254 pf_free_fragment(frag);
255 }
256
257 PF_FRAG_UNLOCK();
258 }
259
260 /*
261 * Try to flush old fragments to make space for new ones
262 */
263 static void
pf_flush_fragments(void)264 pf_flush_fragments(void)
265 {
266 struct pf_fragment *frag;
267 int goal;
268
269 PF_FRAG_ASSERT();
270
271 goal = uma_zone_get_cur(V_pf_frent_z) * 9 / 10;
272 DPFPRINTF(("trying to free %d frag entriess\n", goal));
273 while (goal < uma_zone_get_cur(V_pf_frent_z)) {
274 frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue);
275 if (frag)
276 pf_free_fragment(frag);
277 else
278 break;
279 }
280 }
281
282 /* Frees the fragments and all associated entries */
283 static void
pf_free_fragment(struct pf_fragment * frag)284 pf_free_fragment(struct pf_fragment *frag)
285 {
286 struct pf_frent *frent;
287
288 PF_FRAG_ASSERT();
289
290 /* Free all fragments */
291 for (frent = TAILQ_FIRST(&frag->fr_queue); frent;
292 frent = TAILQ_FIRST(&frag->fr_queue)) {
293 TAILQ_REMOVE(&frag->fr_queue, frent, fr_next);
294
295 m_freem(frent->fe_m);
296 uma_zfree(V_pf_frent_z, frent);
297 }
298
299 pf_remove_fragment(frag);
300 }
301
302 static struct pf_fragment *
pf_find_fragment(struct pf_fragment_cmp * key,struct pf_frag_tree * tree)303 pf_find_fragment(struct pf_fragment_cmp *key, struct pf_frag_tree *tree)
304 {
305 struct pf_fragment *frag;
306
307 PF_FRAG_ASSERT();
308
309 frag = RB_FIND(pf_frag_tree, tree, (struct pf_fragment *)key);
310 if (frag != NULL) {
311 /* XXX Are we sure we want to update the timeout? */
312 frag->fr_timeout = time_uptime;
313 TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next);
314 TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next);
315 }
316
317 return (frag);
318 }
319
320 /* Removes a fragment from the fragment queue and frees the fragment */
321 static void
pf_remove_fragment(struct pf_fragment * frag)322 pf_remove_fragment(struct pf_fragment *frag)
323 {
324
325 PF_FRAG_ASSERT();
326 KASSERT(frag, ("frag != NULL"));
327
328 RB_REMOVE(pf_frag_tree, &V_pf_frag_tree, frag);
329 TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next);
330 uma_zfree(V_pf_frag_z, frag);
331 }
332
333 static struct pf_frent *
pf_create_fragment(u_short * reason)334 pf_create_fragment(u_short *reason)
335 {
336 struct pf_frent *frent;
337
338 PF_FRAG_ASSERT();
339
340 frent = uma_zalloc(V_pf_frent_z, M_NOWAIT);
341 if (frent == NULL) {
342 pf_flush_fragments();
343 frent = uma_zalloc(V_pf_frent_z, M_NOWAIT);
344 if (frent == NULL) {
345 REASON_SET(reason, PFRES_MEMORY);
346 return (NULL);
347 }
348 }
349
350 return (frent);
351 }
352
353 /*
354 * Calculate the additional holes that were created in the fragment
355 * queue by inserting this fragment. A fragment in the middle
356 * creates one more hole by splitting. For each connected side,
357 * it loses one hole.
358 * Fragment entry must be in the queue when calling this function.
359 */
360 static int
pf_frent_holes(struct pf_frent * frent)361 pf_frent_holes(struct pf_frent *frent)
362 {
363 struct pf_frent *prev = TAILQ_PREV(frent, pf_fragq, fr_next);
364 struct pf_frent *next = TAILQ_NEXT(frent, fr_next);
365 int holes = 1;
366
367 if (prev == NULL) {
368 if (frent->fe_off == 0)
369 holes--;
370 } else {
371 KASSERT(frent->fe_off != 0, ("frent->fe_off != 0"));
372 if (frent->fe_off == prev->fe_off + prev->fe_len)
373 holes--;
374 }
375 if (next == NULL) {
376 if (!frent->fe_mff)
377 holes--;
378 } else {
379 KASSERT(frent->fe_mff, ("frent->fe_mff"));
380 if (next->fe_off == frent->fe_off + frent->fe_len)
381 holes--;
382 }
383 return holes;
384 }
385
386 static inline int
pf_frent_index(struct pf_frent * frent)387 pf_frent_index(struct pf_frent *frent)
388 {
389 /*
390 * We have an array of 16 entry points to the queue. A full size
391 * 65535 octet IP packet can have 8192 fragments. So the queue
392 * traversal length is at most 512 and at most 16 entry points are
393 * checked. We need 128 additional bytes on a 64 bit architecture.
394 */
395 CTASSERT(((u_int16_t)0xffff &~ 7) / (0x10000 / PF_FRAG_ENTRY_POINTS) ==
396 16 - 1);
397 CTASSERT(((u_int16_t)0xffff >> 3) / PF_FRAG_ENTRY_POINTS == 512 - 1);
398
399 return frent->fe_off / (0x10000 / PF_FRAG_ENTRY_POINTS);
400 }
401
402 static int
pf_frent_insert(struct pf_fragment * frag,struct pf_frent * frent,struct pf_frent * prev)403 pf_frent_insert(struct pf_fragment *frag, struct pf_frent *frent,
404 struct pf_frent *prev)
405 {
406 int index;
407
408 CTASSERT(PF_FRAG_ENTRY_LIMIT <= 0xff);
409
410 /*
411 * A packet has at most 65536 octets. With 16 entry points, each one
412 * spawns 4096 octets. We limit these to 64 fragments each, which
413 * means on average every fragment must have at least 64 octets.
414 */
415 index = pf_frent_index(frent);
416 if (frag->fr_entries[index] >= PF_FRAG_ENTRY_LIMIT)
417 return ENOBUFS;
418 frag->fr_entries[index]++;
419
420 if (prev == NULL) {
421 TAILQ_INSERT_HEAD(&frag->fr_queue, frent, fr_next);
422 } else {
423 KASSERT(prev->fe_off + prev->fe_len <= frent->fe_off,
424 ("overlapping fragment"));
425 TAILQ_INSERT_AFTER(&frag->fr_queue, prev, frent, fr_next);
426 }
427
428 if (frag->fr_firstoff[index] == NULL) {
429 KASSERT(prev == NULL || pf_frent_index(prev) < index,
430 ("prev == NULL || pf_frent_index(pref) < index"));
431 frag->fr_firstoff[index] = frent;
432 } else {
433 if (frent->fe_off < frag->fr_firstoff[index]->fe_off) {
434 KASSERT(prev == NULL || pf_frent_index(prev) < index,
435 ("prev == NULL || pf_frent_index(pref) < index"));
436 frag->fr_firstoff[index] = frent;
437 } else {
438 KASSERT(prev != NULL, ("prev != NULL"));
439 KASSERT(pf_frent_index(prev) == index,
440 ("pf_frent_index(prev) == index"));
441 }
442 }
443
444 frag->fr_holes += pf_frent_holes(frent);
445
446 return 0;
447 }
448
449 void
pf_frent_remove(struct pf_fragment * frag,struct pf_frent * frent)450 pf_frent_remove(struct pf_fragment *frag, struct pf_frent *frent)
451 {
452 #ifdef INVARIANTS
453 struct pf_frent *prev = TAILQ_PREV(frent, pf_fragq, fr_next);
454 #endif
455 struct pf_frent *next = TAILQ_NEXT(frent, fr_next);
456 int index;
457
458 frag->fr_holes -= pf_frent_holes(frent);
459
460 index = pf_frent_index(frent);
461 KASSERT(frag->fr_firstoff[index] != NULL, ("frent not found"));
462 if (frag->fr_firstoff[index]->fe_off == frent->fe_off) {
463 if (next == NULL) {
464 frag->fr_firstoff[index] = NULL;
465 } else {
466 KASSERT(frent->fe_off + frent->fe_len <= next->fe_off,
467 ("overlapping fragment"));
468 if (pf_frent_index(next) == index) {
469 frag->fr_firstoff[index] = next;
470 } else {
471 frag->fr_firstoff[index] = NULL;
472 }
473 }
474 } else {
475 KASSERT(frag->fr_firstoff[index]->fe_off < frent->fe_off,
476 ("frag->fr_firstoff[index]->fe_off < frent->fe_off"));
477 KASSERT(prev != NULL, ("prev != NULL"));
478 KASSERT(prev->fe_off + prev->fe_len <= frent->fe_off,
479 ("overlapping fragment"));
480 KASSERT(pf_frent_index(prev) == index,
481 ("pf_frent_index(prev) == index"));
482 }
483
484 TAILQ_REMOVE(&frag->fr_queue, frent, fr_next);
485
486 KASSERT(frag->fr_entries[index] > 0, ("No fragments remaining"));
487 frag->fr_entries[index]--;
488 }
489
490 struct pf_frent *
pf_frent_previous(struct pf_fragment * frag,struct pf_frent * frent)491 pf_frent_previous(struct pf_fragment *frag, struct pf_frent *frent)
492 {
493 struct pf_frent *prev, *next;
494 int index;
495
496 /*
497 * If there are no fragments after frag, take the final one. Assume
498 * that the global queue is not empty.
499 */
500 prev = TAILQ_LAST(&frag->fr_queue, pf_fragq);
501 KASSERT(prev != NULL, ("prev != NULL"));
502 if (prev->fe_off <= frent->fe_off)
503 return prev;
504 /*
505 * We want to find a fragment entry that is before frag, but still
506 * close to it. Find the first fragment entry that is in the same
507 * entry point or in the first entry point after that. As we have
508 * already checked that there are entries behind frag, this will
509 * succeed.
510 */
511 for (index = pf_frent_index(frent); index < PF_FRAG_ENTRY_POINTS;
512 index++) {
513 prev = frag->fr_firstoff[index];
514 if (prev != NULL)
515 break;
516 }
517 KASSERT(prev != NULL, ("prev != NULL"));
518 /*
519 * In prev we may have a fragment from the same entry point that is
520 * before frent, or one that is just one position behind frent.
521 * In the latter case, we go back one step and have the predecessor.
522 * There may be none if the new fragment will be the first one.
523 */
524 if (prev->fe_off > frent->fe_off) {
525 prev = TAILQ_PREV(prev, pf_fragq, fr_next);
526 if (prev == NULL)
527 return NULL;
528 KASSERT(prev->fe_off <= frent->fe_off,
529 ("prev->fe_off <= frent->fe_off"));
530 return prev;
531 }
532 /*
533 * In prev is the first fragment of the entry point. The offset
534 * of frag is behind it. Find the closest previous fragment.
535 */
536 for (next = TAILQ_NEXT(prev, fr_next); next != NULL;
537 next = TAILQ_NEXT(next, fr_next)) {
538 if (next->fe_off > frent->fe_off)
539 break;
540 prev = next;
541 }
542 return prev;
543 }
544
545 static struct pf_fragment *
pf_fillup_fragment(struct pf_fragment_cmp * key,struct pf_frent * frent,u_short * reason)546 pf_fillup_fragment(struct pf_fragment_cmp *key, struct pf_frent *frent,
547 u_short *reason)
548 {
549 struct pf_frent *after, *next, *prev;
550 struct pf_fragment *frag;
551 uint16_t total;
552
553 PF_FRAG_ASSERT();
554
555 /* No empty fragments. */
556 if (frent->fe_len == 0) {
557 DPFPRINTF(("bad fragment: len 0\n"));
558 goto bad_fragment;
559 }
560
561 /* All fragments are 8 byte aligned. */
562 if (frent->fe_mff && (frent->fe_len & 0x7)) {
563 DPFPRINTF(("bad fragment: mff and len %d\n", frent->fe_len));
564 goto bad_fragment;
565 }
566
567 /* Respect maximum length, IP_MAXPACKET == IPV6_MAXPACKET. */
568 if (frent->fe_off + frent->fe_len > IP_MAXPACKET) {
569 DPFPRINTF(("bad fragment: max packet %d\n",
570 frent->fe_off + frent->fe_len));
571 goto bad_fragment;
572 }
573
574 DPFPRINTF((key->frc_af == AF_INET ?
575 "reass frag %d @ %d-%d\n" : "reass frag %#08x @ %d-%d\n",
576 key->frc_id, frent->fe_off, frent->fe_off + frent->fe_len));
577
578 /* Fully buffer all of the fragments in this fragment queue. */
579 frag = pf_find_fragment(key, &V_pf_frag_tree);
580
581 /* Create a new reassembly queue for this packet. */
582 if (frag == NULL) {
583 frag = uma_zalloc(V_pf_frag_z, M_NOWAIT);
584 if (frag == NULL) {
585 pf_flush_fragments();
586 frag = uma_zalloc(V_pf_frag_z, M_NOWAIT);
587 if (frag == NULL) {
588 REASON_SET(reason, PFRES_MEMORY);
589 goto drop_fragment;
590 }
591 }
592
593 *(struct pf_fragment_cmp *)frag = *key;
594 memset(frag->fr_firstoff, 0, sizeof(frag->fr_firstoff));
595 memset(frag->fr_entries, 0, sizeof(frag->fr_entries));
596 frag->fr_timeout = time_uptime;
597 TAILQ_INIT(&frag->fr_queue);
598 frag->fr_maxlen = frent->fe_len;
599 frag->fr_holes = 1;
600
601 RB_INSERT(pf_frag_tree, &V_pf_frag_tree, frag);
602 TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next);
603
604 /* We do not have a previous fragment, cannot fail. */
605 pf_frent_insert(frag, frent, NULL);
606
607 return (frag);
608 }
609
610 KASSERT(!TAILQ_EMPTY(&frag->fr_queue), ("!TAILQ_EMPTY()->fr_queue"));
611
612 /* Remember maximum fragment len for refragmentation. */
613 if (frent->fe_len > frag->fr_maxlen)
614 frag->fr_maxlen = frent->fe_len;
615
616 /* Maximum data we have seen already. */
617 total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
618 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
619
620 /* Non terminal fragments must have more fragments flag. */
621 if (frent->fe_off + frent->fe_len < total && !frent->fe_mff)
622 goto bad_fragment;
623
624 /* Check if we saw the last fragment already. */
625 if (!TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_mff) {
626 if (frent->fe_off + frent->fe_len > total ||
627 (frent->fe_off + frent->fe_len == total && frent->fe_mff))
628 goto bad_fragment;
629 } else {
630 if (frent->fe_off + frent->fe_len == total && !frent->fe_mff)
631 goto bad_fragment;
632 }
633
634 /* Find neighbors for newly inserted fragment */
635 prev = pf_frent_previous(frag, frent);
636 if (prev == NULL) {
637 after = TAILQ_FIRST(&frag->fr_queue);
638 KASSERT(after != NULL, ("after != NULL"));
639 } else {
640 after = TAILQ_NEXT(prev, fr_next);
641 }
642
643 if (prev != NULL && prev->fe_off + prev->fe_len > frent->fe_off) {
644 uint16_t precut;
645
646 if (frag->fr_af == AF_INET6)
647 goto free_fragment;
648
649 precut = prev->fe_off + prev->fe_len - frent->fe_off;
650 if (precut >= frent->fe_len) {
651 DPFPRINTF(("new frag overlapped\n"));
652 goto drop_fragment;
653 }
654 DPFPRINTF(("frag head overlap %d\n", precut));
655 m_adj(frent->fe_m, precut);
656 frent->fe_off += precut;
657 frent->fe_len -= precut;
658 }
659
660 for (; after != NULL && frent->fe_off + frent->fe_len > after->fe_off;
661 after = next) {
662 uint16_t aftercut;
663
664 aftercut = frent->fe_off + frent->fe_len - after->fe_off;
665 if (aftercut < after->fe_len) {
666 DPFPRINTF(("frag tail overlap %d", aftercut));
667 m_adj(after->fe_m, aftercut);
668 /* Fragment may switch queue as fe_off changes */
669 pf_frent_remove(frag, after);
670 after->fe_off += aftercut;
671 after->fe_len -= aftercut;
672 /* Insert into correct queue */
673 if (pf_frent_insert(frag, after, prev)) {
674 DPFPRINTF(("fragment requeue limit exceeded"));
675 m_freem(after->fe_m);
676 uma_zfree(V_pf_frent_z, after);
677 /* There is not way to recover */
678 goto free_fragment;
679 }
680 break;
681 }
682
683 /* This fragment is completely overlapped, lose it. */
684 DPFPRINTF(("old frag overlapped\n"));
685 next = TAILQ_NEXT(after, fr_next);
686 pf_frent_remove(frag, after);
687 m_freem(after->fe_m);
688 uma_zfree(V_pf_frent_z, after);
689 }
690
691 /* If part of the queue gets too long, there is not way to recover. */
692 if (pf_frent_insert(frag, frent, prev)) {
693 DPFPRINTF(("fragment queue limit exceeded\n"));
694 goto bad_fragment;
695 }
696
697 return (frag);
698
699 free_fragment:
700 /*
701 * RFC 5722, Errata 3089: When reassembling an IPv6 datagram, if one
702 * or more its constituent fragments is determined to be an overlapping
703 * fragment, the entire datagram (and any constituent fragments) MUST
704 * be silently discarded.
705 */
706 DPFPRINTF(("flush overlapping fragments\n"));
707 pf_free_fragment(frag);
708
709 bad_fragment:
710 REASON_SET(reason, PFRES_FRAG);
711 drop_fragment:
712 uma_zfree(V_pf_frent_z, frent);
713 return (NULL);
714 }
715
716 static struct mbuf *
pf_join_fragment(struct pf_fragment * frag)717 pf_join_fragment(struct pf_fragment *frag)
718 {
719 struct mbuf *m, *m2;
720 struct pf_frent *frent, *next;
721
722 frent = TAILQ_FIRST(&frag->fr_queue);
723 next = TAILQ_NEXT(frent, fr_next);
724
725 m = frent->fe_m;
726 m_adj(m, (frent->fe_hdrlen + frent->fe_len) - m->m_pkthdr.len);
727 uma_zfree(V_pf_frent_z, frent);
728 for (frent = next; frent != NULL; frent = next) {
729 next = TAILQ_NEXT(frent, fr_next);
730
731 m2 = frent->fe_m;
732 /* Strip off ip header. */
733 m_adj(m2, frent->fe_hdrlen);
734 /* Strip off any trailing bytes. */
735 m_adj(m2, frent->fe_len - m2->m_pkthdr.len);
736
737 uma_zfree(V_pf_frent_z, frent);
738 m_cat(m, m2);
739 }
740
741 /* Remove from fragment queue. */
742 pf_remove_fragment(frag);
743
744 return (m);
745 }
746
747 #ifdef INET
748 static int
pf_reassemble(struct mbuf ** m0,struct ip * ip,int dir,u_short * reason)749 pf_reassemble(struct mbuf **m0, struct ip *ip, int dir, u_short *reason)
750 {
751 struct mbuf *m = *m0;
752 struct pf_frent *frent;
753 struct pf_fragment *frag;
754 struct pf_fragment_cmp key;
755 uint16_t total, hdrlen;
756
757 /* Get an entry for the fragment queue */
758 if ((frent = pf_create_fragment(reason)) == NULL)
759 return (PF_DROP);
760
761 frent->fe_m = m;
762 frent->fe_hdrlen = ip->ip_hl << 2;
763 frent->fe_extoff = 0;
764 frent->fe_len = ntohs(ip->ip_len) - (ip->ip_hl << 2);
765 frent->fe_off = (ntohs(ip->ip_off) & IP_OFFMASK) << 3;
766 frent->fe_mff = ntohs(ip->ip_off) & IP_MF;
767
768 pf_ip2key(ip, dir, &key);
769
770 if ((frag = pf_fillup_fragment(&key, frent, reason)) == NULL)
771 return (PF_DROP);
772
773 /* The mbuf is part of the fragment entry, no direct free or access */
774 m = *m0 = NULL;
775
776 if (frag->fr_holes) {
777 DPFPRINTF(("frag %d, holes %d\n", frag->fr_id, frag->fr_holes));
778 return (PF_PASS); /* drop because *m0 is NULL, no error */
779 }
780
781 /* We have all the data */
782 frent = TAILQ_FIRST(&frag->fr_queue);
783 KASSERT(frent != NULL, ("frent != NULL"));
784 total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
785 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
786 hdrlen = frent->fe_hdrlen;
787
788 m = *m0 = pf_join_fragment(frag);
789 frag = NULL;
790
791 if (m->m_flags & M_PKTHDR) {
792 int plen = 0;
793 for (m = *m0; m; m = m->m_next)
794 plen += m->m_len;
795 m = *m0;
796 m->m_pkthdr.len = plen;
797 }
798
799 ip = mtod(m, struct ip *);
800 ip->ip_sum = pf_cksum_fixup(ip->ip_sum, ip->ip_len,
801 htons(hdrlen + total), 0);
802 ip->ip_len = htons(hdrlen + total);
803 ip->ip_sum = pf_cksum_fixup(ip->ip_sum, ip->ip_off,
804 ip->ip_off & ~(IP_MF|IP_OFFMASK), 0);
805 ip->ip_off &= ~(IP_MF|IP_OFFMASK);
806
807 if (hdrlen + total > IP_MAXPACKET) {
808 DPFPRINTF(("drop: too big: %d\n", total));
809 ip->ip_len = 0;
810 REASON_SET(reason, PFRES_SHORT);
811 /* PF_DROP requires a valid mbuf *m0 in pf_test() */
812 return (PF_DROP);
813 }
814
815 DPFPRINTF(("complete: %p(%d)\n", m, ntohs(ip->ip_len)));
816 return (PF_PASS);
817 }
818 #endif /* INET */
819
820 #ifdef INET6
821 static int
pf_reassemble6(struct mbuf ** m0,struct ip6_hdr * ip6,struct ip6_frag * fraghdr,uint16_t hdrlen,uint16_t extoff,u_short * reason)822 pf_reassemble6(struct mbuf **m0, struct ip6_hdr *ip6, struct ip6_frag *fraghdr,
823 uint16_t hdrlen, uint16_t extoff, u_short *reason)
824 {
825 struct mbuf *m = *m0;
826 struct pf_frent *frent;
827 struct pf_fragment *frag;
828 struct pf_fragment_cmp key;
829 struct m_tag *mtag;
830 struct pf_fragment_tag *ftag;
831 int off;
832 uint32_t frag_id;
833 uint16_t total, maxlen;
834 uint8_t proto;
835
836 PF_FRAG_LOCK();
837
838 /* Get an entry for the fragment queue. */
839 if ((frent = pf_create_fragment(reason)) == NULL) {
840 PF_FRAG_UNLOCK();
841 return (PF_DROP);
842 }
843
844 frent->fe_m = m;
845 frent->fe_hdrlen = hdrlen;
846 frent->fe_extoff = extoff;
847 frent->fe_len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - hdrlen;
848 frent->fe_off = ntohs(fraghdr->ip6f_offlg & IP6F_OFF_MASK);
849 frent->fe_mff = fraghdr->ip6f_offlg & IP6F_MORE_FRAG;
850
851 key.frc_src.v6 = ip6->ip6_src;
852 key.frc_dst.v6 = ip6->ip6_dst;
853 key.frc_af = AF_INET6;
854 /* Only the first fragment's protocol is relevant. */
855 key.frc_proto = 0;
856 key.frc_id = fraghdr->ip6f_ident;
857
858 if ((frag = pf_fillup_fragment(&key, frent, reason)) == NULL) {
859 PF_FRAG_UNLOCK();
860 return (PF_DROP);
861 }
862
863 /* The mbuf is part of the fragment entry, no direct free or access. */
864 m = *m0 = NULL;
865
866 if (frag->fr_holes) {
867 DPFPRINTF(("frag %d, holes %d\n", frag->fr_id,
868 frag->fr_holes));
869 PF_FRAG_UNLOCK();
870 return (PF_PASS); /* Drop because *m0 is NULL, no error. */
871 }
872
873 /* We have all the data. */
874 frent = TAILQ_FIRST(&frag->fr_queue);
875 KASSERT(frent != NULL, ("frent != NULL"));
876 extoff = frent->fe_extoff;
877 maxlen = frag->fr_maxlen;
878 frag_id = frag->fr_id;
879 total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
880 TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
881 hdrlen = frent->fe_hdrlen - sizeof(struct ip6_frag);
882
883 m = *m0 = pf_join_fragment(frag);
884 frag = NULL;
885
886 PF_FRAG_UNLOCK();
887
888 /* Take protocol from first fragment header. */
889 m = m_getptr(m, hdrlen + offsetof(struct ip6_frag, ip6f_nxt), &off);
890 KASSERT(m, ("%s: short mbuf chain", __func__));
891 proto = *(mtod(m, caddr_t) + off);
892 m = *m0;
893
894 /* Delete frag6 header */
895 if (ip6_deletefraghdr(m, hdrlen, M_NOWAIT) != 0)
896 goto fail;
897
898 if (m->m_flags & M_PKTHDR) {
899 int plen = 0;
900 for (m = *m0; m; m = m->m_next)
901 plen += m->m_len;
902 m = *m0;
903 m->m_pkthdr.len = plen;
904 }
905
906 if ((mtag = m_tag_get(PF_REASSEMBLED, sizeof(struct pf_fragment_tag),
907 M_NOWAIT)) == NULL)
908 goto fail;
909 ftag = (struct pf_fragment_tag *)(mtag + 1);
910 ftag->ft_hdrlen = hdrlen;
911 ftag->ft_extoff = extoff;
912 ftag->ft_maxlen = maxlen;
913 ftag->ft_id = frag_id;
914 m_tag_prepend(m, mtag);
915
916 ip6 = mtod(m, struct ip6_hdr *);
917 ip6->ip6_plen = htons(hdrlen - sizeof(struct ip6_hdr) + total);
918 if (extoff) {
919 /* Write protocol into next field of last extension header. */
920 m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt),
921 &off);
922 KASSERT(m, ("%s: short mbuf chain", __func__));
923 *(mtod(m, char *) + off) = proto;
924 m = *m0;
925 } else
926 ip6->ip6_nxt = proto;
927
928 if (hdrlen - sizeof(struct ip6_hdr) + total > IPV6_MAXPACKET) {
929 DPFPRINTF(("drop: too big: %d\n", total));
930 ip6->ip6_plen = 0;
931 REASON_SET(reason, PFRES_SHORT);
932 /* PF_DROP requires a valid mbuf *m0 in pf_test6(). */
933 return (PF_DROP);
934 }
935
936 DPFPRINTF(("complete: %p(%d)\n", m, ntohs(ip6->ip6_plen)));
937 return (PF_PASS);
938
939 fail:
940 REASON_SET(reason, PFRES_MEMORY);
941 /* PF_DROP requires a valid mbuf *m0 in pf_test6(), will free later. */
942 return (PF_DROP);
943 }
944 #endif /* INET6 */
945
946 #ifdef INET6
947 int
pf_refragment6(struct ifnet * ifp,struct mbuf ** m0,struct m_tag * mtag)948 pf_refragment6(struct ifnet *ifp, struct mbuf **m0, struct m_tag *mtag)
949 {
950 struct mbuf *m = *m0, *t;
951 struct pf_fragment_tag *ftag = (struct pf_fragment_tag *)(mtag + 1);
952 struct pf_pdesc pd;
953 uint32_t frag_id;
954 uint16_t hdrlen, extoff, maxlen;
955 uint8_t proto;
956 int error, action;
957
958 hdrlen = ftag->ft_hdrlen;
959 extoff = ftag->ft_extoff;
960 maxlen = ftag->ft_maxlen;
961 frag_id = ftag->ft_id;
962 m_tag_delete(m, mtag);
963 mtag = NULL;
964 ftag = NULL;
965
966 if (extoff) {
967 int off;
968
969 /* Use protocol from next field of last extension header */
970 m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt),
971 &off);
972 KASSERT((m != NULL), ("pf_refragment6: short mbuf chain"));
973 proto = *(mtod(m, caddr_t) + off);
974 *(mtod(m, char *) + off) = IPPROTO_FRAGMENT;
975 m = *m0;
976 } else {
977 struct ip6_hdr *hdr;
978
979 hdr = mtod(m, struct ip6_hdr *);
980 proto = hdr->ip6_nxt;
981 hdr->ip6_nxt = IPPROTO_FRAGMENT;
982 }
983
984 /* The MTU must be a multiple of 8 bytes, or we risk doing the
985 * fragmentation wrong. */
986 maxlen = maxlen & ~7;
987
988 /*
989 * Maxlen may be less than 8 if there was only a single
990 * fragment. As it was fragmented before, add a fragment
991 * header also for a single fragment. If total or maxlen
992 * is less than 8, ip6_fragment() will return EMSGSIZE and
993 * we drop the packet.
994 */
995 error = ip6_fragment(ifp, m, hdrlen, proto, maxlen, frag_id);
996 m = (*m0)->m_nextpkt;
997 (*m0)->m_nextpkt = NULL;
998 if (error == 0) {
999 /* The first mbuf contains the unfragmented packet. */
1000 m_freem(*m0);
1001 *m0 = NULL;
1002 action = PF_PASS;
1003 } else {
1004 /* Drop expects an mbuf to free. */
1005 DPFPRINTF(("refragment error %d\n", error));
1006 action = PF_DROP;
1007 }
1008 for (; m; m = t) {
1009 t = m->m_nextpkt;
1010 m->m_nextpkt = NULL;
1011 m->m_flags |= M_SKIP_FIREWALL;
1012 memset(&pd, 0, sizeof(pd));
1013 pd.pf_mtag = pf_find_mtag(m);
1014 if (error == 0)
1015 ip6_forward(m, 0);
1016 else
1017 m_freem(m);
1018 }
1019
1020 return (action);
1021 }
1022 #endif /* INET6 */
1023
1024 #ifdef INET
1025 int
pf_normalize_ip(struct mbuf ** m0,int dir,struct pfi_kkif * kif,u_short * reason,struct pf_pdesc * pd)1026 pf_normalize_ip(struct mbuf **m0, int dir, struct pfi_kkif *kif, u_short *reason,
1027 struct pf_pdesc *pd)
1028 {
1029 struct mbuf *m = *m0;
1030 struct pf_krule *r;
1031 struct ip *h = mtod(m, struct ip *);
1032 int mff = (ntohs(h->ip_off) & IP_MF);
1033 int hlen = h->ip_hl << 2;
1034 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
1035 u_int16_t max;
1036 int ip_len;
1037 int tag = -1;
1038 int verdict;
1039
1040 PF_RULES_RASSERT();
1041
1042 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1043 while (r != NULL) {
1044 pf_counter_u64_add(&r->evaluations, 1);
1045 if (pfi_kkif_match(r->kif, kif) == r->ifnot)
1046 r = r->skip[PF_SKIP_IFP].ptr;
1047 else if (r->direction && r->direction != dir)
1048 r = r->skip[PF_SKIP_DIR].ptr;
1049 else if (r->af && r->af != AF_INET)
1050 r = r->skip[PF_SKIP_AF].ptr;
1051 else if (r->proto && r->proto != h->ip_p)
1052 r = r->skip[PF_SKIP_PROTO].ptr;
1053 else if (PF_MISMATCHAW(&r->src.addr,
1054 (struct pf_addr *)&h->ip_src.s_addr, AF_INET,
1055 r->src.neg, kif, M_GETFIB(m)))
1056 r = r->skip[PF_SKIP_SRC_ADDR].ptr;
1057 else if (PF_MISMATCHAW(&r->dst.addr,
1058 (struct pf_addr *)&h->ip_dst.s_addr, AF_INET,
1059 r->dst.neg, NULL, M_GETFIB(m)))
1060 r = r->skip[PF_SKIP_DST_ADDR].ptr;
1061 else if (r->match_tag && !pf_match_tag(m, r, &tag,
1062 pd->pf_mtag ? pd->pf_mtag->tag : 0))
1063 r = TAILQ_NEXT(r, entries);
1064 else
1065 break;
1066 }
1067
1068 if (r == NULL || r->action == PF_NOSCRUB)
1069 return (PF_PASS);
1070
1071 pf_counter_u64_critical_enter();
1072 pf_counter_u64_add_protected(&r->packets[dir == PF_OUT], 1);
1073 pf_counter_u64_add_protected(&r->bytes[dir == PF_OUT], pd->tot_len);
1074 pf_counter_u64_critical_exit();
1075
1076 /* Check for illegal packets */
1077 if (hlen < (int)sizeof(struct ip)) {
1078 REASON_SET(reason, PFRES_NORM);
1079 goto drop;
1080 }
1081
1082 if (hlen > ntohs(h->ip_len)) {
1083 REASON_SET(reason, PFRES_NORM);
1084 goto drop;
1085 }
1086
1087 /* Clear IP_DF if the rule uses the no-df option */
1088 if (r->rule_flag & PFRULE_NODF && h->ip_off & htons(IP_DF)) {
1089 u_int16_t ip_off = h->ip_off;
1090
1091 h->ip_off &= htons(~IP_DF);
1092 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
1093 }
1094
1095 /* We will need other tests here */
1096 if (!fragoff && !mff)
1097 goto no_fragment;
1098
1099 /* We're dealing with a fragment now. Don't allow fragments
1100 * with IP_DF to enter the cache. If the flag was cleared by
1101 * no-df above, fine. Otherwise drop it.
1102 */
1103 if (h->ip_off & htons(IP_DF)) {
1104 DPFPRINTF(("IP_DF\n"));
1105 goto bad;
1106 }
1107
1108 ip_len = ntohs(h->ip_len) - hlen;
1109
1110 /* All fragments are 8 byte aligned */
1111 if (mff && (ip_len & 0x7)) {
1112 DPFPRINTF(("mff and %d\n", ip_len));
1113 goto bad;
1114 }
1115
1116 /* Respect maximum length */
1117 if (fragoff + ip_len > IP_MAXPACKET) {
1118 DPFPRINTF(("max packet %d\n", fragoff + ip_len));
1119 goto bad;
1120 }
1121 max = fragoff + ip_len;
1122
1123 /* Fully buffer all of the fragments
1124 * Might return a completely reassembled mbuf, or NULL */
1125 PF_FRAG_LOCK();
1126 DPFPRINTF(("reass frag %d @ %d-%d\n", h->ip_id, fragoff, max));
1127 verdict = pf_reassemble(m0, h, dir, reason);
1128 PF_FRAG_UNLOCK();
1129
1130 if (verdict != PF_PASS)
1131 return (PF_DROP);
1132
1133 m = *m0;
1134 if (m == NULL)
1135 return (PF_DROP);
1136
1137 h = mtod(m, struct ip *);
1138
1139 no_fragment:
1140 /* At this point, only IP_DF is allowed in ip_off */
1141 if (h->ip_off & ~htons(IP_DF)) {
1142 u_int16_t ip_off = h->ip_off;
1143
1144 h->ip_off &= htons(IP_DF);
1145 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
1146 }
1147
1148 pf_scrub_ip(&m, r->rule_flag, r->min_ttl, r->set_tos);
1149
1150 return (PF_PASS);
1151
1152 bad:
1153 DPFPRINTF(("dropping bad fragment\n"));
1154 REASON_SET(reason, PFRES_FRAG);
1155 drop:
1156 if (r != NULL && r->log)
1157 PFLOG_PACKET(kif, m, AF_INET, dir, *reason, r, NULL, NULL, pd,
1158 1);
1159
1160 return (PF_DROP);
1161 }
1162 #endif
1163
1164 #ifdef INET6
1165 int
pf_normalize_ip6(struct mbuf ** m0,int dir,struct pfi_kkif * kif,u_short * reason,struct pf_pdesc * pd)1166 pf_normalize_ip6(struct mbuf **m0, int dir, struct pfi_kkif *kif,
1167 u_short *reason, struct pf_pdesc *pd)
1168 {
1169 struct mbuf *m = *m0;
1170 struct pf_krule *r;
1171 struct ip6_hdr *h = mtod(m, struct ip6_hdr *);
1172 int extoff;
1173 int off;
1174 struct ip6_ext ext;
1175 struct ip6_opt opt;
1176 struct ip6_frag frag;
1177 u_int32_t plen;
1178 int optend;
1179 int ooff;
1180 u_int8_t proto;
1181 int terminal;
1182
1183 PF_RULES_RASSERT();
1184
1185 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1186 while (r != NULL) {
1187 pf_counter_u64_add(&r->evaluations, 1);
1188 if (pfi_kkif_match(r->kif, kif) == r->ifnot)
1189 r = r->skip[PF_SKIP_IFP].ptr;
1190 else if (r->direction && r->direction != dir)
1191 r = r->skip[PF_SKIP_DIR].ptr;
1192 else if (r->af && r->af != AF_INET6)
1193 r = r->skip[PF_SKIP_AF].ptr;
1194 #if 0 /* header chain! */
1195 else if (r->proto && r->proto != h->ip6_nxt)
1196 r = r->skip[PF_SKIP_PROTO].ptr;
1197 #endif
1198 else if (PF_MISMATCHAW(&r->src.addr,
1199 (struct pf_addr *)&h->ip6_src, AF_INET6,
1200 r->src.neg, kif, M_GETFIB(m)))
1201 r = r->skip[PF_SKIP_SRC_ADDR].ptr;
1202 else if (PF_MISMATCHAW(&r->dst.addr,
1203 (struct pf_addr *)&h->ip6_dst, AF_INET6,
1204 r->dst.neg, NULL, M_GETFIB(m)))
1205 r = r->skip[PF_SKIP_DST_ADDR].ptr;
1206 else
1207 break;
1208 }
1209
1210 if (r == NULL || r->action == PF_NOSCRUB)
1211 return (PF_PASS);
1212
1213 pf_counter_u64_critical_enter();
1214 pf_counter_u64_add_protected(&r->packets[dir == PF_OUT], 1);
1215 pf_counter_u64_add_protected(&r->bytes[dir == PF_OUT], pd->tot_len);
1216 pf_counter_u64_critical_exit();
1217
1218 /* Check for illegal packets */
1219 if (sizeof(struct ip6_hdr) + IPV6_MAXPACKET < m->m_pkthdr.len)
1220 goto drop;
1221
1222 again:
1223 h = mtod(m, struct ip6_hdr *);
1224 plen = ntohs(h->ip6_plen);
1225 /* jumbo payload option not supported */
1226 if (plen == 0)
1227 goto drop;
1228
1229 extoff = 0;
1230 off = sizeof(struct ip6_hdr);
1231 proto = h->ip6_nxt;
1232 terminal = 0;
1233 do {
1234 switch (proto) {
1235 case IPPROTO_FRAGMENT:
1236 goto fragment;
1237 break;
1238 case IPPROTO_AH:
1239 case IPPROTO_ROUTING:
1240 case IPPROTO_DSTOPTS:
1241 if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL,
1242 NULL, AF_INET6))
1243 goto shortpkt;
1244 extoff = off;
1245 if (proto == IPPROTO_AH)
1246 off += (ext.ip6e_len + 2) * 4;
1247 else
1248 off += (ext.ip6e_len + 1) * 8;
1249 proto = ext.ip6e_nxt;
1250 break;
1251 case IPPROTO_HOPOPTS:
1252 if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL,
1253 NULL, AF_INET6))
1254 goto shortpkt;
1255 extoff = off;
1256 optend = off + (ext.ip6e_len + 1) * 8;
1257 ooff = off + sizeof(ext);
1258 do {
1259 if (!pf_pull_hdr(m, ooff, &opt.ip6o_type,
1260 sizeof(opt.ip6o_type), NULL, NULL,
1261 AF_INET6))
1262 goto shortpkt;
1263 if (opt.ip6o_type == IP6OPT_PAD1) {
1264 ooff++;
1265 continue;
1266 }
1267 if (!pf_pull_hdr(m, ooff, &opt, sizeof(opt),
1268 NULL, NULL, AF_INET6))
1269 goto shortpkt;
1270 if (ooff + sizeof(opt) + opt.ip6o_len > optend)
1271 goto drop;
1272 if (opt.ip6o_type == IP6OPT_JUMBO)
1273 goto drop;
1274 ooff += sizeof(opt) + opt.ip6o_len;
1275 } while (ooff < optend);
1276
1277 off = optend;
1278 proto = ext.ip6e_nxt;
1279 break;
1280 default:
1281 terminal = 1;
1282 break;
1283 }
1284 } while (!terminal);
1285
1286 if (sizeof(struct ip6_hdr) + plen > m->m_pkthdr.len)
1287 goto shortpkt;
1288
1289 pf_scrub_ip6(&m, r->min_ttl);
1290
1291 return (PF_PASS);
1292
1293 fragment:
1294 if (pd->flags & PFDESC_IP_REAS)
1295 return (PF_DROP);
1296 if (sizeof(struct ip6_hdr) + plen > m->m_pkthdr.len)
1297 goto shortpkt;
1298
1299 if (!pf_pull_hdr(m, off, &frag, sizeof(frag), NULL, NULL, AF_INET6))
1300 goto shortpkt;
1301
1302 /* Offset now points to data portion. */
1303 off += sizeof(frag);
1304
1305 /* Returns PF_DROP or *m0 is NULL or completely reassembled mbuf. */
1306 if (pf_reassemble6(m0, h, &frag, off, extoff, reason) != PF_PASS)
1307 return (PF_DROP);
1308 m = *m0;
1309 if (m == NULL)
1310 return (PF_DROP);
1311
1312 pd->flags |= PFDESC_IP_REAS;
1313 goto again;
1314
1315 shortpkt:
1316 REASON_SET(reason, PFRES_SHORT);
1317 if (r != NULL && r->log)
1318 PFLOG_PACKET(kif, m, AF_INET6, dir, *reason, r, NULL, NULL, pd,
1319 1);
1320 return (PF_DROP);
1321
1322 drop:
1323 REASON_SET(reason, PFRES_NORM);
1324 if (r != NULL && r->log)
1325 PFLOG_PACKET(kif, m, AF_INET6, dir, *reason, r, NULL, NULL, pd,
1326 1);
1327 return (PF_DROP);
1328 }
1329 #endif /* INET6 */
1330
1331 int
pf_normalize_tcp(int dir,struct pfi_kkif * kif,struct mbuf * m,int ipoff,int off,void * h,struct pf_pdesc * pd)1332 pf_normalize_tcp(int dir, struct pfi_kkif *kif, struct mbuf *m, int ipoff,
1333 int off, void *h, struct pf_pdesc *pd)
1334 {
1335 struct pf_krule *r, *rm = NULL;
1336 struct tcphdr *th = &pd->hdr.tcp;
1337 int rewrite = 0;
1338 u_short reason;
1339 u_int8_t flags;
1340 sa_family_t af = pd->af;
1341
1342 PF_RULES_RASSERT();
1343
1344 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1345 while (r != NULL) {
1346 pf_counter_u64_add(&r->evaluations, 1);
1347 if (pfi_kkif_match(r->kif, kif) == r->ifnot)
1348 r = r->skip[PF_SKIP_IFP].ptr;
1349 else if (r->direction && r->direction != dir)
1350 r = r->skip[PF_SKIP_DIR].ptr;
1351 else if (r->af && r->af != af)
1352 r = r->skip[PF_SKIP_AF].ptr;
1353 else if (r->proto && r->proto != pd->proto)
1354 r = r->skip[PF_SKIP_PROTO].ptr;
1355 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
1356 r->src.neg, kif, M_GETFIB(m)))
1357 r = r->skip[PF_SKIP_SRC_ADDR].ptr;
1358 else if (r->src.port_op && !pf_match_port(r->src.port_op,
1359 r->src.port[0], r->src.port[1], th->th_sport))
1360 r = r->skip[PF_SKIP_SRC_PORT].ptr;
1361 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
1362 r->dst.neg, NULL, M_GETFIB(m)))
1363 r = r->skip[PF_SKIP_DST_ADDR].ptr;
1364 else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
1365 r->dst.port[0], r->dst.port[1], th->th_dport))
1366 r = r->skip[PF_SKIP_DST_PORT].ptr;
1367 else if (r->os_fingerprint != PF_OSFP_ANY && !pf_osfp_match(
1368 pf_osfp_fingerprint(pd, m, off, th),
1369 r->os_fingerprint))
1370 r = TAILQ_NEXT(r, entries);
1371 else {
1372 rm = r;
1373 break;
1374 }
1375 }
1376
1377 if (rm == NULL || rm->action == PF_NOSCRUB)
1378 return (PF_PASS);
1379
1380 pf_counter_u64_critical_enter();
1381 pf_counter_u64_add_protected(&r->packets[dir == PF_OUT], 1);
1382 pf_counter_u64_add_protected(&r->bytes[dir == PF_OUT], pd->tot_len);
1383 pf_counter_u64_critical_exit();
1384
1385 if (rm->rule_flag & PFRULE_REASSEMBLE_TCP)
1386 pd->flags |= PFDESC_TCP_NORM;
1387
1388 flags = th->th_flags;
1389 if (flags & TH_SYN) {
1390 /* Illegal packet */
1391 if (flags & TH_RST)
1392 goto tcp_drop;
1393
1394 if (flags & TH_FIN)
1395 goto tcp_drop;
1396 } else {
1397 /* Illegal packet */
1398 if (!(flags & (TH_ACK|TH_RST)))
1399 goto tcp_drop;
1400 }
1401
1402 if (!(flags & TH_ACK)) {
1403 /* These flags are only valid if ACK is set */
1404 if ((flags & TH_FIN) || (flags & TH_PUSH) || (flags & TH_URG))
1405 goto tcp_drop;
1406 }
1407
1408 /* Check for illegal header length */
1409 if (th->th_off < (sizeof(struct tcphdr) >> 2))
1410 goto tcp_drop;
1411
1412 /* If flags changed, or reserved data set, then adjust */
1413 if (flags != th->th_flags || th->th_x2 != 0) {
1414 u_int16_t ov, nv;
1415
1416 ov = *(u_int16_t *)(&th->th_ack + 1);
1417 th->th_flags = flags;
1418 th->th_x2 = 0;
1419 nv = *(u_int16_t *)(&th->th_ack + 1);
1420
1421 th->th_sum = pf_proto_cksum_fixup(m, th->th_sum, ov, nv, 0);
1422 rewrite = 1;
1423 }
1424
1425 /* Remove urgent pointer, if TH_URG is not set */
1426 if (!(flags & TH_URG) && th->th_urp) {
1427 th->th_sum = pf_proto_cksum_fixup(m, th->th_sum, th->th_urp,
1428 0, 0);
1429 th->th_urp = 0;
1430 rewrite = 1;
1431 }
1432
1433 /* Process options */
1434 if (r->max_mss && pf_normalize_tcpopt(r, m, th, off, pd->af))
1435 rewrite = 1;
1436
1437 /* copy back packet headers if we sanitized */
1438 if (rewrite)
1439 m_copyback(m, off, sizeof(*th), (caddr_t)th);
1440
1441 return (PF_PASS);
1442
1443 tcp_drop:
1444 REASON_SET(&reason, PFRES_NORM);
1445 if (rm != NULL && r->log)
1446 PFLOG_PACKET(kif, m, AF_INET, dir, reason, r, NULL, NULL, pd,
1447 1);
1448 return (PF_DROP);
1449 }
1450
1451 int
pf_normalize_tcp_init(struct mbuf * m,int off,struct pf_pdesc * pd,struct tcphdr * th,struct pf_state_peer * src,struct pf_state_peer * dst)1452 pf_normalize_tcp_init(struct mbuf *m, int off, struct pf_pdesc *pd,
1453 struct tcphdr *th, struct pf_state_peer *src, struct pf_state_peer *dst)
1454 {
1455 u_int32_t tsval, tsecr;
1456 u_int8_t hdr[60];
1457 u_int8_t *opt;
1458
1459 KASSERT((src->scrub == NULL),
1460 ("pf_normalize_tcp_init: src->scrub != NULL"));
1461
1462 src->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT);
1463 if (src->scrub == NULL)
1464 return (1);
1465
1466 switch (pd->af) {
1467 #ifdef INET
1468 case AF_INET: {
1469 struct ip *h = mtod(m, struct ip *);
1470 src->scrub->pfss_ttl = h->ip_ttl;
1471 break;
1472 }
1473 #endif /* INET */
1474 #ifdef INET6
1475 case AF_INET6: {
1476 struct ip6_hdr *h = mtod(m, struct ip6_hdr *);
1477 src->scrub->pfss_ttl = h->ip6_hlim;
1478 break;
1479 }
1480 #endif /* INET6 */
1481 }
1482
1483 /*
1484 * All normalizations below are only begun if we see the start of
1485 * the connections. They must all set an enabled bit in pfss_flags
1486 */
1487 if ((th->th_flags & TH_SYN) == 0)
1488 return (0);
1489
1490 if (th->th_off > (sizeof(struct tcphdr) >> 2) && src->scrub &&
1491 pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) {
1492 /* Diddle with TCP options */
1493 int hlen;
1494 opt = hdr + sizeof(struct tcphdr);
1495 hlen = (th->th_off << 2) - sizeof(struct tcphdr);
1496 while (hlen >= TCPOLEN_TIMESTAMP) {
1497 switch (*opt) {
1498 case TCPOPT_EOL: /* FALLTHROUGH */
1499 case TCPOPT_NOP:
1500 opt++;
1501 hlen--;
1502 break;
1503 case TCPOPT_TIMESTAMP:
1504 if (opt[1] >= TCPOLEN_TIMESTAMP) {
1505 src->scrub->pfss_flags |=
1506 PFSS_TIMESTAMP;
1507 src->scrub->pfss_ts_mod =
1508 htonl(arc4random());
1509
1510 /* note PFSS_PAWS not set yet */
1511 memcpy(&tsval, &opt[2],
1512 sizeof(u_int32_t));
1513 memcpy(&tsecr, &opt[6],
1514 sizeof(u_int32_t));
1515 src->scrub->pfss_tsval0 = ntohl(tsval);
1516 src->scrub->pfss_tsval = ntohl(tsval);
1517 src->scrub->pfss_tsecr = ntohl(tsecr);
1518 getmicrouptime(&src->scrub->pfss_last);
1519 }
1520 /* FALLTHROUGH */
1521 default:
1522 hlen -= MAX(opt[1], 2);
1523 opt += MAX(opt[1], 2);
1524 break;
1525 }
1526 }
1527 }
1528
1529 return (0);
1530 }
1531
1532 void
pf_normalize_tcp_cleanup(struct pf_kstate * state)1533 pf_normalize_tcp_cleanup(struct pf_kstate *state)
1534 {
1535 /* XXX Note: this also cleans up SCTP. */
1536 if (state->src.scrub)
1537 uma_zfree(V_pf_state_scrub_z, state->src.scrub);
1538 if (state->dst.scrub)
1539 uma_zfree(V_pf_state_scrub_z, state->dst.scrub);
1540
1541 /* Someday... flush the TCP segment reassembly descriptors. */
1542 }
1543
1544 int
pf_normalize_sctp_init(struct mbuf * m,int off,struct pf_pdesc * pd,struct pf_state_peer * src,struct pf_state_peer * dst)1545 pf_normalize_sctp_init(struct mbuf *m, int off, struct pf_pdesc *pd,
1546 struct pf_state_peer *src, struct pf_state_peer *dst)
1547 {
1548 src->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT);
1549 if (src->scrub == NULL)
1550 return (1);
1551
1552 dst->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT);
1553 if (dst->scrub == NULL) {
1554 uma_zfree(V_pf_state_scrub_z, src);
1555 return (1);
1556 }
1557
1558 dst->scrub->pfss_v_tag = pd->sctp_initiate_tag;
1559
1560 return (0);
1561 }
1562
1563 int
pf_normalize_tcp_stateful(struct mbuf * m,int off,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)1564 pf_normalize_tcp_stateful(struct mbuf *m, int off, struct pf_pdesc *pd,
1565 u_short *reason, struct tcphdr *th, struct pf_kstate *state,
1566 struct pf_state_peer *src, struct pf_state_peer *dst, int *writeback)
1567 {
1568 struct timeval uptime;
1569 u_int32_t tsval, tsecr;
1570 u_int tsval_from_last;
1571 u_int8_t hdr[60];
1572 u_int8_t *opt;
1573 int copyback = 0;
1574 int got_ts = 0;
1575 size_t startoff;
1576
1577 KASSERT((src->scrub || dst->scrub),
1578 ("%s: src->scrub && dst->scrub!", __func__));
1579
1580 /*
1581 * Enforce the minimum TTL seen for this connection. Negate a common
1582 * technique to evade an intrusion detection system and confuse
1583 * firewall state code.
1584 */
1585 switch (pd->af) {
1586 #ifdef INET
1587 case AF_INET: {
1588 if (src->scrub) {
1589 struct ip *h = mtod(m, struct ip *);
1590 if (h->ip_ttl > src->scrub->pfss_ttl)
1591 src->scrub->pfss_ttl = h->ip_ttl;
1592 h->ip_ttl = src->scrub->pfss_ttl;
1593 }
1594 break;
1595 }
1596 #endif /* INET */
1597 #ifdef INET6
1598 case AF_INET6: {
1599 if (src->scrub) {
1600 struct ip6_hdr *h = mtod(m, struct ip6_hdr *);
1601 if (h->ip6_hlim > src->scrub->pfss_ttl)
1602 src->scrub->pfss_ttl = h->ip6_hlim;
1603 h->ip6_hlim = src->scrub->pfss_ttl;
1604 }
1605 break;
1606 }
1607 #endif /* INET6 */
1608 }
1609
1610 if (th->th_off > (sizeof(struct tcphdr) >> 2) &&
1611 ((src->scrub && (src->scrub->pfss_flags & PFSS_TIMESTAMP)) ||
1612 (dst->scrub && (dst->scrub->pfss_flags & PFSS_TIMESTAMP))) &&
1613 pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) {
1614 /* Diddle with TCP options */
1615 int hlen;
1616 opt = hdr + sizeof(struct tcphdr);
1617 hlen = (th->th_off << 2) - sizeof(struct tcphdr);
1618 while (hlen >= TCPOLEN_TIMESTAMP) {
1619 startoff = opt - (hdr + sizeof(struct tcphdr));
1620 switch (*opt) {
1621 case TCPOPT_EOL: /* FALLTHROUGH */
1622 case TCPOPT_NOP:
1623 opt++;
1624 hlen--;
1625 break;
1626 case TCPOPT_TIMESTAMP:
1627 /* Modulate the timestamps. Can be used for
1628 * NAT detection, OS uptime determination or
1629 * reboot detection.
1630 */
1631
1632 if (got_ts) {
1633 /* Huh? Multiple timestamps!? */
1634 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1635 DPFPRINTF(("multiple TS??\n"));
1636 pf_print_state(state);
1637 printf("\n");
1638 }
1639 REASON_SET(reason, PFRES_TS);
1640 return (PF_DROP);
1641 }
1642 if (opt[1] >= TCPOLEN_TIMESTAMP) {
1643 memcpy(&tsval, &opt[2],
1644 sizeof(u_int32_t));
1645 if (tsval && src->scrub &&
1646 (src->scrub->pfss_flags &
1647 PFSS_TIMESTAMP)) {
1648 tsval = ntohl(tsval);
1649 pf_patch_32_unaligned(m,
1650 &th->th_sum,
1651 &opt[2],
1652 htonl(tsval +
1653 src->scrub->pfss_ts_mod),
1654 PF_ALGNMNT(startoff),
1655 0);
1656 copyback = 1;
1657 }
1658
1659 /* Modulate TS reply iff valid (!0) */
1660 memcpy(&tsecr, &opt[6],
1661 sizeof(u_int32_t));
1662 if (tsecr && dst->scrub &&
1663 (dst->scrub->pfss_flags &
1664 PFSS_TIMESTAMP)) {
1665 tsecr = ntohl(tsecr)
1666 - dst->scrub->pfss_ts_mod;
1667 pf_patch_32_unaligned(m,
1668 &th->th_sum,
1669 &opt[6],
1670 htonl(tsecr),
1671 PF_ALGNMNT(startoff),
1672 0);
1673 copyback = 1;
1674 }
1675 got_ts = 1;
1676 }
1677 /* FALLTHROUGH */
1678 default:
1679 hlen -= MAX(opt[1], 2);
1680 opt += MAX(opt[1], 2);
1681 break;
1682 }
1683 }
1684 if (copyback) {
1685 /* Copyback the options, caller copys back header */
1686 *writeback = 1;
1687 m_copyback(m, off + sizeof(struct tcphdr),
1688 (th->th_off << 2) - sizeof(struct tcphdr), hdr +
1689 sizeof(struct tcphdr));
1690 }
1691 }
1692
1693 /*
1694 * Must invalidate PAWS checks on connections idle for too long.
1695 * The fastest allowed timestamp clock is 1ms. That turns out to
1696 * be about 24 days before it wraps. XXX Right now our lowerbound
1697 * TS echo check only works for the first 12 days of a connection
1698 * when the TS has exhausted half its 32bit space
1699 */
1700 #define TS_MAX_IDLE (24*24*60*60)
1701 #define TS_MAX_CONN (12*24*60*60) /* XXX remove when better tsecr check */
1702
1703 getmicrouptime(&uptime);
1704 if (src->scrub && (src->scrub->pfss_flags & PFSS_PAWS) &&
1705 (uptime.tv_sec - src->scrub->pfss_last.tv_sec > TS_MAX_IDLE ||
1706 time_uptime - state->creation > TS_MAX_CONN)) {
1707 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1708 DPFPRINTF(("src idled out of PAWS\n"));
1709 pf_print_state(state);
1710 printf("\n");
1711 }
1712 src->scrub->pfss_flags = (src->scrub->pfss_flags & ~PFSS_PAWS)
1713 | PFSS_PAWS_IDLED;
1714 }
1715 if (dst->scrub && (dst->scrub->pfss_flags & PFSS_PAWS) &&
1716 uptime.tv_sec - dst->scrub->pfss_last.tv_sec > TS_MAX_IDLE) {
1717 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1718 DPFPRINTF(("dst idled out of PAWS\n"));
1719 pf_print_state(state);
1720 printf("\n");
1721 }
1722 dst->scrub->pfss_flags = (dst->scrub->pfss_flags & ~PFSS_PAWS)
1723 | PFSS_PAWS_IDLED;
1724 }
1725
1726 if (got_ts && src->scrub && dst->scrub &&
1727 (src->scrub->pfss_flags & PFSS_PAWS) &&
1728 (dst->scrub->pfss_flags & PFSS_PAWS)) {
1729 /* Validate that the timestamps are "in-window".
1730 * RFC1323 describes TCP Timestamp options that allow
1731 * measurement of RTT (round trip time) and PAWS
1732 * (protection against wrapped sequence numbers). PAWS
1733 * gives us a set of rules for rejecting packets on
1734 * long fat pipes (packets that were somehow delayed
1735 * in transit longer than the time it took to send the
1736 * full TCP sequence space of 4Gb). We can use these
1737 * rules and infer a few others that will let us treat
1738 * the 32bit timestamp and the 32bit echoed timestamp
1739 * as sequence numbers to prevent a blind attacker from
1740 * inserting packets into a connection.
1741 *
1742 * RFC1323 tells us:
1743 * - The timestamp on this packet must be greater than
1744 * or equal to the last value echoed by the other
1745 * endpoint. The RFC says those will be discarded
1746 * since it is a dup that has already been acked.
1747 * This gives us a lowerbound on the timestamp.
1748 * timestamp >= other last echoed timestamp
1749 * - The timestamp will be less than or equal to
1750 * the last timestamp plus the time between the
1751 * last packet and now. The RFC defines the max
1752 * clock rate as 1ms. We will allow clocks to be
1753 * up to 10% fast and will allow a total difference
1754 * or 30 seconds due to a route change. And this
1755 * gives us an upperbound on the timestamp.
1756 * timestamp <= last timestamp + max ticks
1757 * We have to be careful here. Windows will send an
1758 * initial timestamp of zero and then initialize it
1759 * to a random value after the 3whs; presumably to
1760 * avoid a DoS by having to call an expensive RNG
1761 * during a SYN flood. Proof MS has at least one
1762 * good security geek.
1763 *
1764 * - The TCP timestamp option must also echo the other
1765 * endpoints timestamp. The timestamp echoed is the
1766 * one carried on the earliest unacknowledged segment
1767 * on the left edge of the sequence window. The RFC
1768 * states that the host will reject any echoed
1769 * timestamps that were larger than any ever sent.
1770 * This gives us an upperbound on the TS echo.
1771 * tescr <= largest_tsval
1772 * - The lowerbound on the TS echo is a little more
1773 * tricky to determine. The other endpoint's echoed
1774 * values will not decrease. But there may be
1775 * network conditions that re-order packets and
1776 * cause our view of them to decrease. For now the
1777 * only lowerbound we can safely determine is that
1778 * the TS echo will never be less than the original
1779 * TS. XXX There is probably a better lowerbound.
1780 * Remove TS_MAX_CONN with better lowerbound check.
1781 * tescr >= other original TS
1782 *
1783 * It is also important to note that the fastest
1784 * timestamp clock of 1ms will wrap its 32bit space in
1785 * 24 days. So we just disable TS checking after 24
1786 * days of idle time. We actually must use a 12d
1787 * connection limit until we can come up with a better
1788 * lowerbound to the TS echo check.
1789 */
1790 struct timeval delta_ts;
1791 int ts_fudge;
1792
1793 /*
1794 * PFTM_TS_DIFF is how many seconds of leeway to allow
1795 * a host's timestamp. This can happen if the previous
1796 * packet got delayed in transit for much longer than
1797 * this packet.
1798 */
1799 if ((ts_fudge = state->rule.ptr->timeout[PFTM_TS_DIFF]) == 0)
1800 ts_fudge = V_pf_default_rule.timeout[PFTM_TS_DIFF];
1801
1802 /* Calculate max ticks since the last timestamp */
1803 #define TS_MAXFREQ 1100 /* RFC max TS freq of 1Khz + 10% skew */
1804 #define TS_MICROSECS 1000000 /* microseconds per second */
1805 delta_ts = uptime;
1806 timevalsub(&delta_ts, &src->scrub->pfss_last);
1807 tsval_from_last = (delta_ts.tv_sec + ts_fudge) * TS_MAXFREQ;
1808 tsval_from_last += delta_ts.tv_usec / (TS_MICROSECS/TS_MAXFREQ);
1809
1810 if ((src->state >= TCPS_ESTABLISHED &&
1811 dst->state >= TCPS_ESTABLISHED) &&
1812 (SEQ_LT(tsval, dst->scrub->pfss_tsecr) ||
1813 SEQ_GT(tsval, src->scrub->pfss_tsval + tsval_from_last) ||
1814 (tsecr && (SEQ_GT(tsecr, dst->scrub->pfss_tsval) ||
1815 SEQ_LT(tsecr, dst->scrub->pfss_tsval0))))) {
1816 /* Bad RFC1323 implementation or an insertion attack.
1817 *
1818 * - Solaris 2.6 and 2.7 are known to send another ACK
1819 * after the FIN,FIN|ACK,ACK closing that carries
1820 * an old timestamp.
1821 */
1822
1823 DPFPRINTF(("Timestamp failed %c%c%c%c\n",
1824 SEQ_LT(tsval, dst->scrub->pfss_tsecr) ? '0' : ' ',
1825 SEQ_GT(tsval, src->scrub->pfss_tsval +
1826 tsval_from_last) ? '1' : ' ',
1827 SEQ_GT(tsecr, dst->scrub->pfss_tsval) ? '2' : ' ',
1828 SEQ_LT(tsecr, dst->scrub->pfss_tsval0)? '3' : ' '));
1829 DPFPRINTF((" tsval: %u tsecr: %u +ticks: %u "
1830 "idle: %jus %lums\n",
1831 tsval, tsecr, tsval_from_last,
1832 (uintmax_t)delta_ts.tv_sec,
1833 delta_ts.tv_usec / 1000));
1834 DPFPRINTF((" src->tsval: %u tsecr: %u\n",
1835 src->scrub->pfss_tsval, src->scrub->pfss_tsecr));
1836 DPFPRINTF((" dst->tsval: %u tsecr: %u tsval0: %u"
1837 "\n", dst->scrub->pfss_tsval,
1838 dst->scrub->pfss_tsecr, dst->scrub->pfss_tsval0));
1839 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1840 pf_print_state(state);
1841 pf_print_flags(th->th_flags);
1842 printf("\n");
1843 }
1844 REASON_SET(reason, PFRES_TS);
1845 return (PF_DROP);
1846 }
1847
1848 /* XXX I'd really like to require tsecr but it's optional */
1849
1850 } else if (!got_ts && (th->th_flags & TH_RST) == 0 &&
1851 ((src->state == TCPS_ESTABLISHED && dst->state == TCPS_ESTABLISHED)
1852 || pd->p_len > 0 || (th->th_flags & TH_SYN)) &&
1853 src->scrub && dst->scrub &&
1854 (src->scrub->pfss_flags & PFSS_PAWS) &&
1855 (dst->scrub->pfss_flags & PFSS_PAWS)) {
1856 /* Didn't send a timestamp. Timestamps aren't really useful
1857 * when:
1858 * - connection opening or closing (often not even sent).
1859 * but we must not let an attacker to put a FIN on a
1860 * data packet to sneak it through our ESTABLISHED check.
1861 * - on a TCP reset. RFC suggests not even looking at TS.
1862 * - on an empty ACK. The TS will not be echoed so it will
1863 * probably not help keep the RTT calculation in sync and
1864 * there isn't as much danger when the sequence numbers
1865 * got wrapped. So some stacks don't include TS on empty
1866 * ACKs :-(
1867 *
1868 * To minimize the disruption to mostly RFC1323 conformant
1869 * stacks, we will only require timestamps on data packets.
1870 *
1871 * And what do ya know, we cannot require timestamps on data
1872 * packets. There appear to be devices that do legitimate
1873 * TCP connection hijacking. There are HTTP devices that allow
1874 * a 3whs (with timestamps) and then buffer the HTTP request.
1875 * If the intermediate device has the HTTP response cache, it
1876 * will spoof the response but not bother timestamping its
1877 * packets. So we can look for the presence of a timestamp in
1878 * the first data packet and if there, require it in all future
1879 * packets.
1880 */
1881
1882 if (pd->p_len > 0 && (src->scrub->pfss_flags & PFSS_DATA_TS)) {
1883 /*
1884 * Hey! Someone tried to sneak a packet in. Or the
1885 * stack changed its RFC1323 behavior?!?!
1886 */
1887 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1888 DPFPRINTF(("Did not receive expected RFC1323 "
1889 "timestamp\n"));
1890 pf_print_state(state);
1891 pf_print_flags(th->th_flags);
1892 printf("\n");
1893 }
1894 REASON_SET(reason, PFRES_TS);
1895 return (PF_DROP);
1896 }
1897 }
1898
1899 /*
1900 * We will note if a host sends his data packets with or without
1901 * timestamps. And require all data packets to contain a timestamp
1902 * if the first does. PAWS implicitly requires that all data packets be
1903 * timestamped. But I think there are middle-man devices that hijack
1904 * TCP streams immediately after the 3whs and don't timestamp their
1905 * packets (seen in a WWW accelerator or cache).
1906 */
1907 if (pd->p_len > 0 && src->scrub && (src->scrub->pfss_flags &
1908 (PFSS_TIMESTAMP|PFSS_DATA_TS|PFSS_DATA_NOTS)) == PFSS_TIMESTAMP) {
1909 if (got_ts)
1910 src->scrub->pfss_flags |= PFSS_DATA_TS;
1911 else {
1912 src->scrub->pfss_flags |= PFSS_DATA_NOTS;
1913 if (V_pf_status.debug >= PF_DEBUG_MISC && dst->scrub &&
1914 (dst->scrub->pfss_flags & PFSS_TIMESTAMP)) {
1915 /* Don't warn if other host rejected RFC1323 */
1916 DPFPRINTF(("Broken RFC1323 stack did not "
1917 "timestamp data packet. Disabled PAWS "
1918 "security.\n"));
1919 pf_print_state(state);
1920 pf_print_flags(th->th_flags);
1921 printf("\n");
1922 }
1923 }
1924 }
1925
1926 /*
1927 * Update PAWS values
1928 */
1929 if (got_ts && src->scrub && PFSS_TIMESTAMP == (src->scrub->pfss_flags &
1930 (PFSS_PAWS_IDLED|PFSS_TIMESTAMP))) {
1931 getmicrouptime(&src->scrub->pfss_last);
1932 if (SEQ_GEQ(tsval, src->scrub->pfss_tsval) ||
1933 (src->scrub->pfss_flags & PFSS_PAWS) == 0)
1934 src->scrub->pfss_tsval = tsval;
1935
1936 if (tsecr) {
1937 if (SEQ_GEQ(tsecr, src->scrub->pfss_tsecr) ||
1938 (src->scrub->pfss_flags & PFSS_PAWS) == 0)
1939 src->scrub->pfss_tsecr = tsecr;
1940
1941 if ((src->scrub->pfss_flags & PFSS_PAWS) == 0 &&
1942 (SEQ_LT(tsval, src->scrub->pfss_tsval0) ||
1943 src->scrub->pfss_tsval0 == 0)) {
1944 /* tsval0 MUST be the lowest timestamp */
1945 src->scrub->pfss_tsval0 = tsval;
1946 }
1947
1948 /* Only fully initialized after a TS gets echoed */
1949 if ((src->scrub->pfss_flags & PFSS_PAWS) == 0)
1950 src->scrub->pfss_flags |= PFSS_PAWS;
1951 }
1952 }
1953
1954 /* I have a dream.... TCP segment reassembly.... */
1955 return (0);
1956 }
1957
1958 static int
pf_normalize_tcpopt(struct pf_krule * r,struct mbuf * m,struct tcphdr * th,int off,sa_family_t af)1959 pf_normalize_tcpopt(struct pf_krule *r, struct mbuf *m, struct tcphdr *th,
1960 int off, sa_family_t af)
1961 {
1962 u_int16_t *mss;
1963 int thoff;
1964 int opt, cnt, optlen = 0;
1965 int rewrite = 0;
1966 u_char opts[TCP_MAXOLEN];
1967 u_char *optp = opts;
1968 size_t startoff;
1969
1970 thoff = th->th_off << 2;
1971 cnt = thoff - sizeof(struct tcphdr);
1972
1973 if (cnt > 0 && !pf_pull_hdr(m, off + sizeof(*th), opts, cnt,
1974 NULL, NULL, af))
1975 return (rewrite);
1976
1977 for (; cnt > 0; cnt -= optlen, optp += optlen) {
1978 startoff = optp - opts;
1979 opt = optp[0];
1980 if (opt == TCPOPT_EOL)
1981 break;
1982 if (opt == TCPOPT_NOP)
1983 optlen = 1;
1984 else {
1985 if (cnt < 2)
1986 break;
1987 optlen = optp[1];
1988 if (optlen < 2 || optlen > cnt)
1989 break;
1990 }
1991 switch (opt) {
1992 case TCPOPT_MAXSEG:
1993 mss = (u_int16_t *)(optp + 2);
1994 if ((ntohs(*mss)) > r->max_mss) {
1995 pf_patch_16_unaligned(m,
1996 &th->th_sum,
1997 mss, htons(r->max_mss),
1998 PF_ALGNMNT(startoff),
1999 0);
2000 rewrite = 1;
2001 }
2002 break;
2003 default:
2004 break;
2005 }
2006 }
2007
2008 if (rewrite)
2009 m_copyback(m, off + sizeof(*th), thoff - sizeof(*th), opts);
2010
2011 return (rewrite);
2012 }
2013
2014 static int
pf_scan_sctp(struct mbuf * m,int ipoff,int off,struct pf_pdesc * pd,struct pfi_kkif * kif)2015 pf_scan_sctp(struct mbuf *m, int ipoff, int off, struct pf_pdesc *pd,
2016 struct pfi_kkif *kif)
2017 {
2018 struct sctp_chunkhdr ch = { };
2019 int chunk_off = sizeof(struct sctphdr);
2020 int chunk_start;
2021 int ret;
2022
2023 while (off + chunk_off < pd->tot_len) {
2024 if (!pf_pull_hdr(m, off + chunk_off, &ch, sizeof(ch), NULL,
2025 NULL, pd->af))
2026 return (PF_DROP);
2027
2028 /* Length includes the header, this must be at least 4. */
2029 if (ntohs(ch.chunk_length) < 4)
2030 return (PF_DROP);
2031
2032 chunk_start = chunk_off;
2033 chunk_off += roundup(ntohs(ch.chunk_length), 4);
2034
2035 switch (ch.chunk_type) {
2036 case SCTP_INITIATION:
2037 case SCTP_INITIATION_ACK: {
2038 struct sctp_init_chunk init;
2039
2040 if (!pf_pull_hdr(m, off + chunk_start, &init,
2041 sizeof(init), NULL, NULL, pd->af))
2042 return (PF_DROP);
2043
2044 /*
2045 * RFC 9620, Section 3.3.2, "The Initiate Tag is allowed to have
2046 * any value except 0."
2047 */
2048 if (init.init.initiate_tag == 0)
2049 return (PF_DROP);
2050 if (init.init.num_inbound_streams == 0)
2051 return (PF_DROP);
2052 if (init.init.num_outbound_streams == 0)
2053 return (PF_DROP);
2054 if (ntohl(init.init.a_rwnd) < SCTP_MIN_RWND)
2055 return (PF_DROP);
2056
2057 /*
2058 * RFC 9260, Section 3.1, INIT chunks MUST have zero
2059 * verification tag.
2060 */
2061 if (ch.chunk_type == SCTP_INITIATION &&
2062 pd->hdr.sctp.v_tag != 0)
2063 return (PF_DROP);
2064
2065 pd->sctp_initiate_tag = init.init.initiate_tag;
2066
2067 if (ch.chunk_type == SCTP_INITIATION)
2068 pd->sctp_flags |= PFDESC_SCTP_INIT;
2069 else
2070 pd->sctp_flags |= PFDESC_SCTP_INIT_ACK;
2071
2072 ret = pf_multihome_scan_init(m, off + chunk_start,
2073 ntohs(init.ch.chunk_length), pd, kif);
2074 if (ret != PF_PASS)
2075 return (ret);
2076
2077 break;
2078 }
2079 case SCTP_ABORT_ASSOCIATION:
2080 pd->sctp_flags |= PFDESC_SCTP_ABORT;
2081 break;
2082 case SCTP_SHUTDOWN:
2083 case SCTP_SHUTDOWN_ACK:
2084 pd->sctp_flags |= PFDESC_SCTP_SHUTDOWN;
2085 break;
2086 case SCTP_SHUTDOWN_COMPLETE:
2087 pd->sctp_flags |= PFDESC_SCTP_SHUTDOWN_COMPLETE;
2088 break;
2089 case SCTP_COOKIE_ECHO:
2090 pd->sctp_flags |= PFDESC_SCTP_COOKIE;
2091 break;
2092 case SCTP_COOKIE_ACK:
2093 pd->sctp_flags |= PFDESC_SCTP_COOKIE_ACK;
2094 break;
2095 case SCTP_DATA:
2096 pd->sctp_flags |= PFDESC_SCTP_DATA;
2097 break;
2098 case SCTP_HEARTBEAT_REQUEST:
2099 pd->sctp_flags |= PFDESC_SCTP_HEARTBEAT;
2100 break;
2101 case SCTP_HEARTBEAT_ACK:
2102 pd->sctp_flags |= PFDESC_SCTP_HEARTBEAT_ACK;
2103 break;
2104 case SCTP_ASCONF:
2105 pd->sctp_flags |= PFDESC_SCTP_ASCONF;
2106
2107 ret = pf_multihome_scan_asconf(m, off + chunk_start,
2108 ntohs(ch.chunk_length), pd, kif);
2109 if (ret != PF_PASS)
2110 return (ret);
2111 break;
2112 default:
2113 pd->sctp_flags |= PFDESC_SCTP_OTHER;
2114 break;
2115 }
2116 }
2117
2118 /* Validate chunk lengths vs. packet length. */
2119 if (off + chunk_off != pd->tot_len)
2120 return (PF_DROP);
2121
2122 /*
2123 * INIT, INIT_ACK or SHUTDOWN_COMPLETE chunks must always be the only
2124 * one in a packet.
2125 */
2126 if ((pd->sctp_flags & PFDESC_SCTP_INIT) &&
2127 (pd->sctp_flags & ~PFDESC_SCTP_INIT))
2128 return (PF_DROP);
2129 if ((pd->sctp_flags & PFDESC_SCTP_INIT_ACK) &&
2130 (pd->sctp_flags & ~PFDESC_SCTP_INIT_ACK))
2131 return (PF_DROP);
2132 if ((pd->sctp_flags & PFDESC_SCTP_SHUTDOWN_COMPLETE) &&
2133 (pd->sctp_flags & ~PFDESC_SCTP_SHUTDOWN_COMPLETE))
2134 return (PF_DROP);
2135
2136 return (PF_PASS);
2137 }
2138
2139 int
pf_normalize_sctp(int dir,struct pfi_kkif * kif,struct mbuf * m,int ipoff,int off,void * h,struct pf_pdesc * pd)2140 pf_normalize_sctp(int dir, struct pfi_kkif *kif, struct mbuf *m, int ipoff,
2141 int off, void *h, struct pf_pdesc *pd)
2142 {
2143 struct pf_krule *r, *rm = NULL;
2144 struct sctphdr *sh = &pd->hdr.sctp;
2145 u_short reason;
2146 sa_family_t af = pd->af;
2147 int srs;
2148
2149 PF_RULES_RASSERT();
2150
2151 /* Unconditionally scan the SCTP packet, because we need to look for
2152 * things like shutdown and asconf chunks. */
2153 if (pf_scan_sctp(m, ipoff, off, pd, kif) != PF_PASS)
2154 goto sctp_drop;
2155
2156 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
2157 /* Check if there any scrub rules. Lack of scrub rules means enforced
2158 * packet normalization operation just like in OpenBSD. */
2159 srs = (r != NULL);
2160 while (r != NULL) {
2161 pf_counter_u64_add(&r->evaluations, 1);
2162 if (pfi_kkif_match(r->kif, kif) == r->ifnot)
2163 r = r->skip[PF_SKIP_IFP].ptr;
2164 else if (r->direction && r->direction != dir)
2165 r = r->skip[PF_SKIP_DIR].ptr;
2166 else if (r->af && r->af != af)
2167 r = r->skip[PF_SKIP_AF].ptr;
2168 else if (r->proto && r->proto != pd->proto)
2169 r = r->skip[PF_SKIP_PROTO].ptr;
2170 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
2171 r->src.neg, kif, M_GETFIB(m)))
2172 r = r->skip[PF_SKIP_SRC_ADDR].ptr;
2173 else if (r->src.port_op && !pf_match_port(r->src.port_op,
2174 r->src.port[0], r->src.port[1], sh->src_port))
2175 r = r->skip[PF_SKIP_SRC_PORT].ptr;
2176 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
2177 r->dst.neg, NULL, M_GETFIB(m)))
2178 r = r->skip[PF_SKIP_DST_ADDR].ptr;
2179 else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
2180 r->dst.port[0], r->dst.port[1], sh->dest_port))
2181 r = r->skip[PF_SKIP_DST_PORT].ptr;
2182 else {
2183 rm = r;
2184 break;
2185 }
2186 }
2187
2188 if (srs) {
2189 /* With scrub rules present SCTP normalization happens only
2190 * if one of rules has matched and it's not a "no scrub" rule */
2191 if (rm == NULL || rm->action == PF_NOSCRUB)
2192 return (PF_PASS);
2193
2194 pf_counter_u64_critical_enter();
2195 pf_counter_u64_add_protected(&r->packets[dir == PF_OUT], 1);
2196 pf_counter_u64_add_protected(&r->bytes[dir == PF_OUT], pd->tot_len);
2197 pf_counter_u64_critical_exit();
2198 }
2199
2200 /* Verify we're a multiple of 4 bytes long */
2201 if ((pd->tot_len - off - sizeof(struct sctphdr)) % 4)
2202 goto sctp_drop;
2203
2204 /* INIT chunk needs to be the only chunk */
2205 if (pd->sctp_flags & PFDESC_SCTP_INIT)
2206 if (pd->sctp_flags & ~PFDESC_SCTP_INIT)
2207 goto sctp_drop;
2208
2209 return (PF_PASS);
2210
2211 sctp_drop:
2212 REASON_SET(&reason, PFRES_NORM);
2213 if (rm != NULL && r->log)
2214 PFLOG_PACKET(kif, m, AF_INET, pd->dir, reason, r, NULL, NULL, pd,
2215 1);
2216
2217 return (PF_DROP);
2218 }
2219
2220 #ifdef INET
2221 static void
pf_scrub_ip(struct mbuf ** m0,u_int32_t flags,u_int8_t min_ttl,u_int8_t tos)2222 pf_scrub_ip(struct mbuf **m0, u_int32_t flags, u_int8_t min_ttl, u_int8_t tos)
2223 {
2224 struct mbuf *m = *m0;
2225 struct ip *h = mtod(m, struct ip *);
2226
2227 /* Clear IP_DF if no-df was requested */
2228 if (flags & PFRULE_NODF && h->ip_off & htons(IP_DF)) {
2229 u_int16_t ip_off = h->ip_off;
2230
2231 h->ip_off &= htons(~IP_DF);
2232 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
2233 }
2234
2235 /* Enforce a minimum ttl, may cause endless packet loops */
2236 if (min_ttl && h->ip_ttl < min_ttl) {
2237 u_int16_t ip_ttl = h->ip_ttl;
2238
2239 h->ip_ttl = min_ttl;
2240 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_ttl, h->ip_ttl, 0);
2241 }
2242
2243 /* Enforce tos */
2244 if (flags & PFRULE_SET_TOS) {
2245 u_int16_t ov, nv;
2246
2247 ov = *(u_int16_t *)h;
2248 h->ip_tos = tos | (h->ip_tos & IPTOS_ECN_MASK);
2249 nv = *(u_int16_t *)h;
2250
2251 h->ip_sum = pf_cksum_fixup(h->ip_sum, ov, nv, 0);
2252 }
2253
2254 /* random-id, but not for fragments */
2255 if (flags & PFRULE_RANDOMID && !(h->ip_off & ~htons(IP_DF))) {
2256 uint16_t ip_id = h->ip_id;
2257
2258 ip_fillid(h);
2259 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_id, h->ip_id, 0);
2260 }
2261 }
2262 #endif /* INET */
2263
2264 #ifdef INET6
2265 static void
pf_scrub_ip6(struct mbuf ** m0,u_int8_t min_ttl)2266 pf_scrub_ip6(struct mbuf **m0, u_int8_t min_ttl)
2267 {
2268 struct mbuf *m = *m0;
2269 struct ip6_hdr *h = mtod(m, struct ip6_hdr *);
2270
2271 /* Enforce a minimum ttl, may cause endless packet loops */
2272 if (min_ttl && h->ip6_hlim < min_ttl)
2273 h->ip6_hlim = min_ttl;
2274 }
2275 #endif
2276