1/*
2 * Copyright (c) 2007-2021 Apple Inc. All rights reserved.
3 *
4 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5 *
6 * This file contains Original Code and/or Modifications of Original Code
7 * as defined in and that are subject to the Apple Public Source License
8 * Version 2.0 (the 'License'). You may not use this file except in
9 * compliance with the License. The rights granted to you under the License
10 * may not be used to create, or enable the creation or redistribution of,
11 * unlawful or unlicensed copies of an Apple operating system, or to
12 * circumvent, violate, or enable the circumvention or violation of, any
13 * terms of an Apple operating system software license agreement.
14 *
15 * Please obtain a copy of the License at
16 * http://www.opensource.apple.com/apsl/ and read it before using this file.
17 *
18 * The Original Code and all software distributed under the License are
19 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23 * Please see the License for the specific language governing rights and
24 * limitations under the License.
25 *
26 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27 */
28
29/* $apfw: pf_norm.c,v 1.10 2008/08/28 19:10:53 jhw Exp $ */
30/* $OpenBSD: pf_norm.c,v 1.107 2006/04/16 00:59:52 pascoe Exp $ */
31
32/*
33 * Copyright 2001 Niels Provos <provos@citi.umich.edu>
34 * All rights reserved.
35 *
36 * Redistribution and use in source and binary forms, with or without
37 * modification, are permitted provided that the following conditions
38 * are met:
39 * 1. Redistributions of source code must retain the above copyright
40 * notice, this list of conditions and the following disclaimer.
41 * 2. Redistributions in binary form must reproduce the above copyright
42 * notice, this list of conditions and the following disclaimer in the
43 * documentation and/or other materials provided with the distribution.
44 *
45 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
46 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
47 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
48 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
49 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
50 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
51 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
52 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
53 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
54 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
55 */
56
57#include <sys/param.h>
58#include <sys/systm.h>
59#include <sys/mbuf.h>
60#include <sys/filio.h>
61#include <sys/fcntl.h>
62#include <sys/socket.h>
63#include <sys/kernel.h>
64#include <sys/time.h>
65#include <sys/random.h>
66#include <sys/mcache.h>
67
68#include <net/if.h>
69#include <net/if_types.h>
70#include <net/bpf.h>
71#include <net/route.h>
72#include <net/if_pflog.h>
73
74#include <netinet/in.h>
75#include <netinet/in_var.h>
76#include <netinet/in_systm.h>
77#include <netinet/ip.h>
78#include <netinet/ip_var.h>
79#include <netinet/tcp.h>
80#include <netinet/tcp_seq.h>
81#include <netinet/tcp_fsm.h>
82#include <netinet/udp.h>
83#include <netinet/ip_icmp.h>
84
85#include <netinet/ip6.h>
86#include <netinet6/ip6_var.h>
87
88#include <net/pfvar.h>
89
90struct pf_frent {
91 LIST_ENTRY(pf_frent) fr_next;
92 struct mbuf *fr_m;
93#define fr_ip fr_u.fru_ipv4
94#define fr_ip6 fr_u.fru_ipv6
95 union {
96 struct ip *fru_ipv4;
97 struct ip6_hdr *fru_ipv6;
98 } fr_u;
99 struct ip6_frag fr_ip6f_opt;
100 uint16_t fr_ip6f_hlen; /* total header length */
101 uint16_t fr_ip6f_extoff; /* last extension header offset or 0 */
102};
103
104struct pf_frcache {
105 LIST_ENTRY(pf_frcache) fr_next;
106 uint16_t fr_off;
107 uint16_t fr_end;
108};
109
110#define PFFRAG_SEENLAST 0x0001 /* Seen the last fragment for this */
111#define PFFRAG_NOBUFFER 0x0002 /* Non-buffering fragment cache */
112#define PFFRAG_DROP 0x0004 /* Drop all fragments */
113#define BUFFER_FRAGMENTS(fr) (!((fr)->fr_flags & PFFRAG_NOBUFFER))
114
115struct pf_fragment {
116 RB_ENTRY(pf_fragment) fr_entry;
117 TAILQ_ENTRY(pf_fragment) frag_next;
118 struct pf_addr fr_srcx;
119 struct pf_addr fr_dstx;
120 u_int8_t fr_p; /* protocol of this fragment */
121 u_int8_t fr_flags; /* status flags */
122 u_int16_t fr_max; /* fragment data max */
123#define fr_id fr_uid.fru_id4
124#define fr_id6 fr_uid.fru_id6
125 union {
126 u_int16_t fru_id4;
127 u_int32_t fru_id6;
128 } fr_uid;
129 int fr_af;
130 u_int32_t fr_timeout;
131#define fr_queue fr_u.fru_queue
132#define fr_cache fr_u.fru_cache
133 union {
134 LIST_HEAD(pf_fragq, pf_frent) fru_queue; /* buffering */
135 LIST_HEAD(pf_cacheq, pf_frcache) fru_cache; /* non-buf */
136 } fr_u;
137 uint32_t fr_csum_flags; /* checksum flags */
138 uint32_t fr_csum; /* partial checksum value */
139 uint16_t fr_ip6_maxlen; /* maximum length of a single fragment in IPv6 */
140};
141
142static TAILQ_HEAD(pf_fragqueue, pf_fragment) pf_fragqueue;
143static TAILQ_HEAD(pf_cachequeue, pf_fragment) pf_cachequeue;
144
145static __inline int pf_frag_compare(struct pf_fragment *,
146 struct pf_fragment *);
147static RB_HEAD(pf_frag_tree, pf_fragment) pf_frag_tree, pf_cache_tree;
148RB_PROTOTYPE_SC(__private_extern__, pf_frag_tree, pf_fragment, fr_entry,
149 pf_frag_compare);
150RB_GENERATE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
151
152/* Private prototypes */
153static void pf_ip6hdr2key(struct pf_fragment *, struct ip6_hdr *,
154 struct ip6_frag *);
155static void pf_ip2key(struct pf_fragment *, struct ip *);
156static void pf_remove_fragment(struct pf_fragment *);
157static void pf_flush_fragments(void);
158static void pf_free_fragment(struct pf_fragment *);
159static struct pf_fragment *pf_find_fragment_by_key(struct pf_fragment *,
160 struct pf_frag_tree *);
161static __inline struct pf_fragment *
162pf_find_fragment_by_ipv4_header(struct ip *, struct pf_frag_tree *);
163static struct mbuf *pf_reassemble(struct mbuf *, struct pf_fragment **,
164 struct pf_frent *, int);
165static struct mbuf *pf_fragcache(struct mbuf **, struct ip *,
166 struct pf_fragment **, int, int, int *);
167static int pf_normalize_tcpopt(struct pf_rule *, int, struct pfi_kif *,
168 struct pf_pdesc *, pbuf_t *, struct tcphdr *, int, int *);
169static __inline struct pf_fragment *
170pf_find_fragment_by_ipv6_header(struct ip6_hdr *, struct ip6_frag *,
171 struct pf_frag_tree *);
172static struct mbuf *pf_reassemble6(struct mbuf **, struct pf_fragment **,
173 struct pf_frent *, int);
174static struct mbuf *pf_frag6cache(struct mbuf **, struct ip6_hdr*,
175 struct ip6_frag *, struct pf_fragment **, int, int, int, int *);
176
177#define DPFPRINTF(x) do { \
178 if (pf_status.debug >= PF_DEBUG_MISC) { \
179 printf("%s: ", __func__); \
180 printf x ; \
181 } \
182} while (0)
183
184/* Globals */
185struct pool pf_frent_pl, pf_frag_pl;
186static struct pool pf_cache_pl, pf_cent_pl;
187struct pool pf_state_scrub_pl;
188
189static int pf_nfrents, pf_ncache;
190
191void
192pf_normalize_init(void)
193{
194 pool_init(&pf_frent_pl, sizeof(struct pf_frent), 0, 0, 0, "pffrent",
195 NULL);
196 pool_init(&pf_frag_pl, sizeof(struct pf_fragment), 0, 0, 0, "pffrag",
197 NULL);
198 pool_init(&pf_cache_pl, sizeof(struct pf_fragment), 0, 0, 0,
199 "pffrcache", NULL);
200 pool_init(&pf_cent_pl, sizeof(struct pf_frcache), 0, 0, 0, "pffrcent",
201 NULL);
202 pool_init(&pf_state_scrub_pl, sizeof(struct pf_state_scrub), 0, 0, 0,
203 "pfstscr", NULL);
204
205 pool_sethiwat(&pf_frag_pl, PFFRAG_FRAG_HIWAT);
206 pool_sethardlimit(&pf_frent_pl, PFFRAG_FRENT_HIWAT, NULL, 0);
207 pool_sethardlimit(&pf_cache_pl, PFFRAG_FRCACHE_HIWAT, NULL, 0);
208 pool_sethardlimit(&pf_cent_pl, PFFRAG_FRCENT_HIWAT, NULL, 0);
209
210 TAILQ_INIT(&pf_fragqueue);
211 TAILQ_INIT(&pf_cachequeue);
212}
213
214#if 0
215void
216pf_normalize_destroy(void)
217{
218 pool_destroy(&pf_state_scrub_pl);
219 pool_destroy(&pf_cent_pl);
220 pool_destroy(&pf_cache_pl);
221 pool_destroy(&pf_frag_pl);
222 pool_destroy(&pf_frent_pl);
223}
224#endif
225
226int
227pf_normalize_isempty(void)
228{
229 return TAILQ_EMPTY(&pf_fragqueue) && TAILQ_EMPTY(&pf_cachequeue);
230}
231
232static __inline int
233pf_frag_compare(struct pf_fragment *a, struct pf_fragment *b)
234{
235 int diff;
236
237 if ((diff = a->fr_af - b->fr_af)) {
238 return diff;
239 } else if ((diff = a->fr_p - b->fr_p)) {
240 return diff;
241 } else {
242 struct pf_addr *sa = &a->fr_srcx;
243 struct pf_addr *sb = &b->fr_srcx;
244 struct pf_addr *da = &a->fr_dstx;
245 struct pf_addr *db = &b->fr_dstx;
246
247 switch (a->fr_af) {
248#ifdef INET
249 case AF_INET:
250 if ((diff = a->fr_id - b->fr_id)) {
251 return diff;
252 } else if (sa->v4addr.s_addr < sb->v4addr.s_addr) {
253 return -1;
254 } else if (sa->v4addr.s_addr > sb->v4addr.s_addr) {
255 return 1;
256 } else if (da->v4addr.s_addr < db->v4addr.s_addr) {
257 return -1;
258 } else if (da->v4addr.s_addr > db->v4addr.s_addr) {
259 return 1;
260 }
261 break;
262#endif
263 case AF_INET6:
264 if ((diff = a->fr_id6 - b->fr_id6)) {
265 return diff;
266 } else if (sa->addr32[3] < sb->addr32[3]) {
267 return -1;
268 } else if (sa->addr32[3] > sb->addr32[3]) {
269 return 1;
270 } else if (sa->addr32[2] < sb->addr32[2]) {
271 return -1;
272 } else if (sa->addr32[2] > sb->addr32[2]) {
273 return 1;
274 } else if (sa->addr32[1] < sb->addr32[1]) {
275 return -1;
276 } else if (sa->addr32[1] > sb->addr32[1]) {
277 return 1;
278 } else if (sa->addr32[0] < sb->addr32[0]) {
279 return -1;
280 } else if (sa->addr32[0] > sb->addr32[0]) {
281 return 1;
282 } else if (da->addr32[3] < db->addr32[3]) {
283 return -1;
284 } else if (da->addr32[3] > db->addr32[3]) {
285 return 1;
286 } else if (da->addr32[2] < db->addr32[2]) {
287 return -1;
288 } else if (da->addr32[2] > db->addr32[2]) {
289 return 1;
290 } else if (da->addr32[1] < db->addr32[1]) {
291 return -1;
292 } else if (da->addr32[1] > db->addr32[1]) {
293 return 1;
294 } else if (da->addr32[0] < db->addr32[0]) {
295 return -1;
296 } else if (da->addr32[0] > db->addr32[0]) {
297 return 1;
298 }
299 break;
300 default:
301 VERIFY(!0 && "only IPv4 and IPv6 supported!");
302 break;
303 }
304 }
305 return 0;
306}
307
308void
309pf_purge_expired_fragments(void)
310{
311 struct pf_fragment *frag;
312 u_int32_t expire = pf_time_second() -
313 pf_default_rule.timeout[PFTM_FRAG];
314
315 while ((frag = TAILQ_LAST(&pf_fragqueue, pf_fragqueue)) != NULL) {
316 VERIFY(BUFFER_FRAGMENTS(frag));
317 if (frag->fr_timeout > expire) {
318 break;
319 }
320
321 switch (frag->fr_af) {
322 case AF_INET:
323 DPFPRINTF(("expiring IPv4 %d(0x%llx) from queue.\n",
324 ntohs(frag->fr_id),
325 (uint64_t)VM_KERNEL_ADDRHASH(frag)));
326 break;
327 case AF_INET6:
328 DPFPRINTF(("expiring IPv6 %d(0x%llx) from queue.\n",
329 ntohl(frag->fr_id6),
330 (uint64_t)VM_KERNEL_ADDRHASH(frag)));
331 break;
332 default:
333 VERIFY(0 && "only IPv4 and IPv6 supported");
334 break;
335 }
336 pf_free_fragment(frag);
337 }
338
339 while ((frag = TAILQ_LAST(&pf_cachequeue, pf_cachequeue)) != NULL) {
340 VERIFY(!BUFFER_FRAGMENTS(frag));
341 if (frag->fr_timeout > expire) {
342 break;
343 }
344
345 switch (frag->fr_af) {
346 case AF_INET:
347 DPFPRINTF(("expiring IPv4 %d(0x%llx) from cache.\n",
348 ntohs(frag->fr_id),
349 (uint64_t)VM_KERNEL_ADDRHASH(frag)));
350 break;
351 case AF_INET6:
352 DPFPRINTF(("expiring IPv6 %d(0x%llx) from cache.\n",
353 ntohl(frag->fr_id6),
354 (uint64_t)VM_KERNEL_ADDRHASH(frag)));
355 break;
356 default:
357 VERIFY(0 && "only IPv4 and IPv6 supported");
358 break;
359 }
360 pf_free_fragment(frag);
361 VERIFY(TAILQ_EMPTY(&pf_cachequeue) ||
362 TAILQ_LAST(&pf_cachequeue, pf_cachequeue) != frag);
363 }
364}
365
366/*
367 * Try to flush old fragments to make space for new ones
368 */
369
370static void
371pf_flush_fragments(void)
372{
373 struct pf_fragment *frag;
374 int goal;
375
376 goal = pf_nfrents * 9 / 10;
377 DPFPRINTF(("trying to free > %d frents\n",
378 pf_nfrents - goal));
379 while (goal < pf_nfrents) {
380 frag = TAILQ_LAST(&pf_fragqueue, pf_fragqueue);
381 if (frag == NULL) {
382 break;
383 }
384 pf_free_fragment(frag);
385 }
386
387
388 goal = pf_ncache * 9 / 10;
389 DPFPRINTF(("trying to free > %d cache entries\n",
390 pf_ncache - goal));
391 while (goal < pf_ncache) {
392 frag = TAILQ_LAST(&pf_cachequeue, pf_cachequeue);
393 if (frag == NULL) {
394 break;
395 }
396 pf_free_fragment(frag);
397 }
398}
399
400/* Frees the fragments and all associated entries */
401
402static void
403pf_free_fragment(struct pf_fragment *frag)
404{
405 struct pf_frent *frent;
406 struct pf_frcache *frcache;
407
408 /* Free all fragments */
409 if (BUFFER_FRAGMENTS(frag)) {
410 for (frent = LIST_FIRST(&frag->fr_queue); frent;
411 frent = LIST_FIRST(&frag->fr_queue)) {
412 LIST_REMOVE(frent, fr_next);
413
414 m_freem(frent->fr_m);
415 pool_put(&pf_frent_pl, frent);
416 pf_nfrents--;
417 }
418 } else {
419 for (frcache = LIST_FIRST(&frag->fr_cache); frcache;
420 frcache = LIST_FIRST(&frag->fr_cache)) {
421 LIST_REMOVE(frcache, fr_next);
422
423 VERIFY(LIST_EMPTY(&frag->fr_cache) ||
424 LIST_FIRST(&frag->fr_cache)->fr_off >
425 frcache->fr_end);
426
427 pool_put(&pf_cent_pl, frcache);
428 pf_ncache--;
429 }
430 }
431
432 pf_remove_fragment(frag);
433}
434
435static void
436pf_ip6hdr2key(struct pf_fragment *key, struct ip6_hdr *ip6,
437 struct ip6_frag *fh)
438{
439 key->fr_p = fh->ip6f_nxt;
440 key->fr_id6 = fh->ip6f_ident;
441 key->fr_af = AF_INET6;
442 key->fr_srcx.v6addr = ip6->ip6_src;
443 key->fr_dstx.v6addr = ip6->ip6_dst;
444}
445
446static void
447pf_ip2key(struct pf_fragment *key, struct ip *ip)
448{
449 key->fr_p = ip->ip_p;
450 key->fr_id = ip->ip_id;
451 key->fr_af = AF_INET;
452 key->fr_srcx.v4addr.s_addr = ip->ip_src.s_addr;
453 key->fr_dstx.v4addr.s_addr = ip->ip_dst.s_addr;
454}
455
456static struct pf_fragment *
457pf_find_fragment_by_key(struct pf_fragment *key, struct pf_frag_tree *tree)
458{
459 struct pf_fragment *frag;
460
461 frag = RB_FIND(pf_frag_tree, tree, key);
462 if (frag != NULL) {
463 /* XXX Are we sure we want to update the timeout? */
464 frag->fr_timeout = pf_time_second();
465 if (BUFFER_FRAGMENTS(frag)) {
466 TAILQ_REMOVE(&pf_fragqueue, frag, frag_next);
467 TAILQ_INSERT_HEAD(&pf_fragqueue, frag, frag_next);
468 } else {
469 TAILQ_REMOVE(&pf_cachequeue, frag, frag_next);
470 TAILQ_INSERT_HEAD(&pf_cachequeue, frag, frag_next);
471 }
472 }
473
474 return frag;
475}
476
477static __attribute__((noinline)) struct pf_fragment *
478pf_find_fragment_by_ipv4_header(struct ip *ip, struct pf_frag_tree *tree)
479{
480 struct pf_fragment key;
481 pf_ip2key(key: &key, ip);
482 return pf_find_fragment_by_key(key: &key, tree);
483}
484
485/* Removes a fragment from the fragment queue and frees the fragment */
486static void
487pf_remove_fragment(struct pf_fragment *frag)
488{
489 if (BUFFER_FRAGMENTS(frag)) {
490 RB_REMOVE(pf_frag_tree, &pf_frag_tree, frag);
491 TAILQ_REMOVE(&pf_fragqueue, frag, frag_next);
492 pool_put(&pf_frag_pl, frag);
493 } else {
494 RB_REMOVE(pf_frag_tree, &pf_cache_tree, frag);
495 TAILQ_REMOVE(&pf_cachequeue, frag, frag_next);
496 pool_put(&pf_cache_pl, frag);
497 }
498}
499
500#define FR_IP_OFF(fr) ((ntohs((fr)->fr_ip->ip_off) & IP_OFFMASK) << 3)
501static struct mbuf *
502pf_reassemble(struct mbuf *m0, struct pf_fragment **frag,
503 struct pf_frent *frent, int mff)
504{
505 struct mbuf *m = m0, *m2;
506 struct pf_frent *frea, *next;
507 struct pf_frent *frep = NULL;
508 struct ip *ip = frent->fr_ip;
509 uint32_t hlen = ip->ip_hl << 2;
510 u_int16_t off = (ntohs(ip->ip_off) & IP_OFFMASK) << 3;
511 u_int16_t ip_len = ntohs(ip->ip_len) - ip->ip_hl * 4;
512 u_int16_t fr_max = ip_len + off;
513 uint32_t csum, csum_flags;
514
515 VERIFY(*frag == NULL || BUFFER_FRAGMENTS(*frag));
516
517 /*
518 * Leverage partial checksum offload for IP fragments. Narrow down
519 * the scope to cover only UDP without IP options, as that is the
520 * most common case.
521 *
522 * Perform 1's complement adjustment of octets that got included/
523 * excluded in the hardware-calculated checksum value. Ignore cases
524 * where the value includes the entire IPv4 header span, as the sum
525 * for those octets would already be 0 by the time we get here; IP
526 * has already performed its header checksum validation. Also take
527 * care of any trailing bytes and subtract out their partial sum.
528 */
529 if (ip->ip_p == IPPROTO_UDP && hlen == sizeof(struct ip) &&
530 (m->m_pkthdr.csum_flags &
531 (CSUM_DATA_VALID | CSUM_PARTIAL | CSUM_PSEUDO_HDR)) ==
532 (CSUM_DATA_VALID | CSUM_PARTIAL)) {
533 uint32_t start = m->m_pkthdr.csum_rx_start;
534 int32_t trailer = (m_pktlen(m) - ntohs(ip->ip_len));
535 uint32_t swbytes = (uint32_t)trailer;
536
537 csum = m->m_pkthdr.csum_rx_val;
538
539 ASSERT(trailer >= 0);
540 if ((start != 0 && start != hlen) || trailer != 0) {
541#if BYTE_ORDER != BIG_ENDIAN
542 if (start < hlen) {
543 HTONS(ip->ip_len);
544 HTONS(ip->ip_off);
545 }
546#endif /* BYTE_ORDER != BIG_ENDIAN */
547 /* callee folds in sum */
548 csum = m_adj_sum16(m, start, hlen,
549 (ip->ip_len - hlen), csum);
550 if (hlen > start) {
551 swbytes += (hlen - start);
552 } else {
553 swbytes += (start - hlen);
554 }
555#if BYTE_ORDER != BIG_ENDIAN
556 if (start < hlen) {
557 NTOHS(ip->ip_off);
558 NTOHS(ip->ip_len);
559 }
560#endif /* BYTE_ORDER != BIG_ENDIAN */
561 }
562 csum_flags = m->m_pkthdr.csum_flags;
563
564 if (swbytes != 0) {
565 udp_in_cksum_stats(swbytes);
566 }
567 if (trailer != 0) {
568 m_adj(m, -trailer);
569 }
570 } else {
571 csum = 0;
572 csum_flags = 0;
573 }
574
575 /* Invalidate checksum */
576 m->m_pkthdr.csum_flags &= ~CSUM_DATA_VALID;
577
578 /* Strip off ip header */
579 m->m_data += hlen;
580 m->m_len -= hlen;
581
582 /* Create a new reassembly queue for this packet */
583 if (*frag == NULL) {
584 *frag = pool_get(&pf_frag_pl, PR_NOWAIT);
585 if (*frag == NULL) {
586 pf_flush_fragments();
587 *frag = pool_get(&pf_frag_pl, PR_NOWAIT);
588 if (*frag == NULL) {
589 goto drop_fragment;
590 }
591 }
592
593 (*frag)->fr_flags = 0;
594 (*frag)->fr_max = 0;
595 (*frag)->fr_af = AF_INET;
596 (*frag)->fr_srcx.v4addr = frent->fr_ip->ip_src;
597 (*frag)->fr_dstx.v4addr = frent->fr_ip->ip_dst;
598 (*frag)->fr_p = frent->fr_ip->ip_p;
599 (*frag)->fr_id = frent->fr_ip->ip_id;
600 (*frag)->fr_timeout = pf_time_second();
601 if (csum_flags != 0) {
602 (*frag)->fr_csum_flags = csum_flags;
603 (*frag)->fr_csum = csum;
604 }
605 LIST_INIT(&(*frag)->fr_queue);
606
607 RB_INSERT(pf_frag_tree, &pf_frag_tree, *frag);
608 TAILQ_INSERT_HEAD(&pf_fragqueue, *frag, frag_next);
609
610 /* We do not have a previous fragment */
611 frep = NULL;
612 goto insert;
613 }
614
615 /*
616 * If this fragment contains similar checksum offload info
617 * as that of the existing ones, accumulate checksum. Otherwise,
618 * invalidate checksum offload info for the entire datagram.
619 */
620 if (csum_flags != 0 && csum_flags == (*frag)->fr_csum_flags) {
621 (*frag)->fr_csum += csum;
622 } else if ((*frag)->fr_csum_flags != 0) {
623 (*frag)->fr_csum_flags = 0;
624 }
625
626 /*
627 * Find a fragment after the current one:
628 * - off contains the real shifted offset.
629 */
630 LIST_FOREACH(frea, &(*frag)->fr_queue, fr_next) {
631 if (FR_IP_OFF(frea) > off) {
632 break;
633 }
634 frep = frea;
635 }
636
637 VERIFY(frep != NULL || frea != NULL);
638
639 if (frep != NULL &&
640 FR_IP_OFF(frep) + ntohs(frep->fr_ip->ip_len) - frep->fr_ip->ip_hl *
641 4 > off) {
642 u_int16_t precut;
643
644 precut = FR_IP_OFF(frep) + ntohs(frep->fr_ip->ip_len) -
645 frep->fr_ip->ip_hl * 4 - off;
646 if (precut >= ip_len) {
647 goto drop_fragment;
648 }
649 m_adj(frent->fr_m, precut);
650 DPFPRINTF(("overlap -%d\n", precut));
651 /* Enforce 8 byte boundaries */
652 ip->ip_off = htons(ntohs(ip->ip_off) + (precut >> 3));
653 off = (ntohs(ip->ip_off) & IP_OFFMASK) << 3;
654 ip_len -= precut;
655 ip->ip_len = htons(ip_len);
656 }
657
658 for (; frea != NULL && ip_len + off > FR_IP_OFF(frea);
659 frea = next) {
660 u_int16_t aftercut;
661
662 aftercut = ip_len + off - FR_IP_OFF(frea);
663 DPFPRINTF(("adjust overlap %d\n", aftercut));
664 if (aftercut < ntohs(frea->fr_ip->ip_len) - frea->fr_ip->ip_hl
665 * 4) {
666 frea->fr_ip->ip_len =
667 htons(ntohs(frea->fr_ip->ip_len) - aftercut);
668 frea->fr_ip->ip_off = htons(ntohs(frea->fr_ip->ip_off) +
669 (aftercut >> 3));
670 m_adj(frea->fr_m, aftercut);
671 break;
672 }
673
674 /* This fragment is completely overlapped, lose it */
675 next = LIST_NEXT(frea, fr_next);
676 m_freem(frea->fr_m);
677 LIST_REMOVE(frea, fr_next);
678 pool_put(&pf_frent_pl, frea);
679 pf_nfrents--;
680 }
681
682insert:
683 /* Update maximum data size */
684 if ((*frag)->fr_max < fr_max) {
685 (*frag)->fr_max = fr_max;
686 }
687 /* This is the last segment */
688 if (!mff) {
689 (*frag)->fr_flags |= PFFRAG_SEENLAST;
690 }
691
692 if (frep == NULL) {
693 LIST_INSERT_HEAD(&(*frag)->fr_queue, frent, fr_next);
694 } else {
695 LIST_INSERT_AFTER(frep, frent, fr_next);
696 }
697
698 /* Check if we are completely reassembled */
699 if (!((*frag)->fr_flags & PFFRAG_SEENLAST)) {
700 return NULL;
701 }
702
703 /* Check if we have all the data */
704 off = 0;
705 for (frep = LIST_FIRST(&(*frag)->fr_queue); frep; frep = next) {
706 next = LIST_NEXT(frep, fr_next);
707
708 off += ntohs(frep->fr_ip->ip_len) - frep->fr_ip->ip_hl * 4;
709 if (off < (*frag)->fr_max &&
710 (next == NULL || FR_IP_OFF(next) != off)) {
711 DPFPRINTF(("missing fragment at %d, next %d, max %d\n",
712 off, next == NULL ? -1 : FR_IP_OFF(next),
713 (*frag)->fr_max));
714 return NULL;
715 }
716 }
717 DPFPRINTF(("%d < %d?\n", off, (*frag)->fr_max));
718 if (off < (*frag)->fr_max) {
719 return NULL;
720 }
721
722 /* We have all the data */
723 frent = LIST_FIRST(&(*frag)->fr_queue);
724 VERIFY(frent != NULL);
725 if ((frent->fr_ip->ip_hl << 2) + off > IP_MAXPACKET) {
726 DPFPRINTF(("drop: too big: %d\n", off));
727 pf_free_fragment(frag: *frag);
728 *frag = NULL;
729 return NULL;
730 }
731 next = LIST_NEXT(frent, fr_next);
732
733 /* Magic from ip_input */
734 ip = frent->fr_ip;
735 m = frent->fr_m;
736 m2 = m->m_next;
737 m->m_next = NULL;
738 m_cat(m, m2);
739 pool_put(&pf_frent_pl, frent);
740 pf_nfrents--;
741 for (frent = next; frent != NULL; frent = next) {
742 next = LIST_NEXT(frent, fr_next);
743
744 m2 = frent->fr_m;
745 pool_put(&pf_frent_pl, frent);
746 pf_nfrents--;
747 m_cat(m, m2);
748 }
749
750 ip->ip_src = (*frag)->fr_srcx.v4addr;
751 ip->ip_dst = (*frag)->fr_dstx.v4addr;
752
753 if ((*frag)->fr_csum_flags != 0) {
754 csum = (*frag)->fr_csum;
755
756 ADDCARRY(csum);
757
758 m->m_pkthdr.csum_rx_val = csum;
759 m->m_pkthdr.csum_rx_start = sizeof(struct ip);
760 m->m_pkthdr.csum_flags = (*frag)->fr_csum_flags;
761 } else if ((m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) ||
762 (m->m_pkthdr.pkt_flags & PKTF_LOOP)) {
763 /* loopback checksums are always OK */
764 m->m_pkthdr.csum_data = 0xffff;
765 m->m_pkthdr.csum_flags =
766 CSUM_DATA_VALID | CSUM_PSEUDO_HDR |
767 CSUM_IP_CHECKED | CSUM_IP_VALID;
768 }
769
770 /* Remove from fragment queue */
771 pf_remove_fragment(frag: *frag);
772 *frag = NULL;
773
774 hlen = ip->ip_hl << 2;
775 ip->ip_len = htons(off + hlen);
776 m->m_len += hlen;
777 m->m_data -= hlen;
778
779 /* some debugging cruft by sklower, below, will go away soon */
780 /* XXX this should be done elsewhere */
781 if (m->m_flags & M_PKTHDR) {
782 int plen = 0;
783 for (m2 = m; m2; m2 = m2->m_next) {
784 plen += m2->m_len;
785 }
786 m->m_pkthdr.len = plen;
787 }
788
789 DPFPRINTF(("complete: 0x%llx(%d)\n",
790 (uint64_t)VM_KERNEL_ADDRPERM(m), ntohs(ip->ip_len)));
791 return m;
792
793drop_fragment:
794 /* Oops - fail safe - drop packet */
795 pool_put(&pf_frent_pl, frent);
796 pf_nfrents--;
797 m_freem(m);
798 return NULL;
799}
800
801static __attribute__((noinline)) struct mbuf *
802pf_fragcache(struct mbuf **m0, struct ip *h, struct pf_fragment **frag, int mff,
803 int drop, int *nomem)
804{
805 struct mbuf *m = *m0;
806 struct pf_frcache *frp, *fra, *cur = NULL;
807 int ip_len = ntohs(h->ip_len) - (h->ip_hl << 2);
808 u_int16_t off = ntohs(h->ip_off) << 3;
809 u_int16_t fr_max = ip_len + off;
810 int hosed = 0;
811
812 VERIFY(*frag == NULL || !BUFFER_FRAGMENTS(*frag));
813
814 /* Create a new range queue for this packet */
815 if (*frag == NULL) {
816 *frag = pool_get(&pf_cache_pl, PR_NOWAIT);
817 if (*frag == NULL) {
818 pf_flush_fragments();
819 *frag = pool_get(&pf_cache_pl, PR_NOWAIT);
820 if (*frag == NULL) {
821 goto no_mem;
822 }
823 }
824
825 /* Get an entry for the queue */
826 cur = pool_get(&pf_cent_pl, PR_NOWAIT);
827 if (cur == NULL) {
828 pool_put(&pf_cache_pl, *frag);
829 *frag = NULL;
830 goto no_mem;
831 }
832 pf_ncache++;
833
834 (*frag)->fr_flags = PFFRAG_NOBUFFER;
835 (*frag)->fr_max = 0;
836 (*frag)->fr_af = AF_INET;
837 (*frag)->fr_srcx.v4addr = h->ip_src;
838 (*frag)->fr_dstx.v4addr = h->ip_dst;
839 (*frag)->fr_p = h->ip_p;
840 (*frag)->fr_id = h->ip_id;
841 (*frag)->fr_timeout = pf_time_second();
842
843 cur->fr_off = off;
844 cur->fr_end = fr_max;
845 LIST_INIT(&(*frag)->fr_cache);
846 LIST_INSERT_HEAD(&(*frag)->fr_cache, cur, fr_next);
847
848 RB_INSERT(pf_frag_tree, &pf_cache_tree, *frag);
849 TAILQ_INSERT_HEAD(&pf_cachequeue, *frag, frag_next);
850
851 DPFPRINTF(("fragcache[%d]: new %d-%d\n", h->ip_id, off,
852 fr_max));
853
854 goto pass;
855 }
856
857 /*
858 * Find a fragment after the current one:
859 * - off contains the real shifted offset.
860 */
861 frp = NULL;
862 LIST_FOREACH(fra, &(*frag)->fr_cache, fr_next) {
863 if (fra->fr_off > off) {
864 break;
865 }
866 frp = fra;
867 }
868
869 VERIFY(frp != NULL || fra != NULL);
870
871 if (frp != NULL) {
872 int precut;
873
874 precut = frp->fr_end - off;
875 if (precut >= ip_len) {
876 /* Fragment is entirely a duplicate */
877 DPFPRINTF(("fragcache[%d]: dead (%d-%d) %d-%d\n",
878 h->ip_id, frp->fr_off, frp->fr_end, off, fr_max));
879 goto drop_fragment;
880 }
881 if (precut == 0) {
882 /* They are adjacent. Fixup cache entry */
883 DPFPRINTF(("fragcache[%d]: adjacent (%d-%d) %d-%d\n",
884 h->ip_id, frp->fr_off, frp->fr_end, off, fr_max));
885 frp->fr_end = fr_max;
886 } else if (precut > 0) {
887 /*
888 * The first part of this payload overlaps with a
889 * fragment that has already been passed.
890 * Need to trim off the first part of the payload.
891 * But to do so easily, we need to create another
892 * mbuf to throw the original header into.
893 */
894
895 DPFPRINTF(("fragcache[%d]: chop %d (%d-%d) %d-%d\n",
896 h->ip_id, precut, frp->fr_off, frp->fr_end, off,
897 fr_max));
898
899 off += precut;
900 fr_max -= precut;
901 /* Update the previous frag to encompass this one */
902 frp->fr_end = fr_max;
903
904 if (!drop) {
905 /*
906 * XXX Optimization opportunity
907 * This is a very heavy way to trim the payload.
908 * we could do it much faster by diddling mbuf
909 * internals but that would be even less legible
910 * than this mbuf magic. For my next trick,
911 * I'll pull a rabbit out of my laptop.
912 */
913 *m0 = m_copym(m, 0, h->ip_hl << 2, M_NOWAIT);
914 if (*m0 == NULL) {
915 goto no_mem;
916 }
917 VERIFY((*m0)->m_next == NULL);
918 m_adj(m, precut + (h->ip_hl << 2));
919 m_cat(*m0, m);
920 m = *m0;
921 if (m->m_flags & M_PKTHDR) {
922 int plen = 0;
923 struct mbuf *t;
924 for (t = m; t; t = t->m_next) {
925 plen += t->m_len;
926 }
927 m->m_pkthdr.len = plen;
928 }
929
930
931 h = mtod(m, struct ip *);
932
933
934 VERIFY((int)m->m_len ==
935 ntohs(h->ip_len) - precut);
936 h->ip_off = htons(ntohs(h->ip_off) +
937 (precut >> 3));
938 h->ip_len = htons(ntohs(h->ip_len) - precut);
939 } else {
940 hosed++;
941 }
942 } else {
943 /* There is a gap between fragments */
944
945 DPFPRINTF(("fragcache[%d]: gap %d (%d-%d) %d-%d\n",
946 h->ip_id, -precut, frp->fr_off, frp->fr_end, off,
947 fr_max));
948
949 cur = pool_get(&pf_cent_pl, PR_NOWAIT);
950 if (cur == NULL) {
951 goto no_mem;
952 }
953 pf_ncache++;
954
955 cur->fr_off = off;
956 cur->fr_end = fr_max;
957 LIST_INSERT_AFTER(frp, cur, fr_next);
958 }
959 }
960
961 if (fra != NULL) {
962 int aftercut;
963 int merge = 0;
964
965 aftercut = fr_max - fra->fr_off;
966 if (aftercut == 0) {
967 /* Adjacent fragments */
968 DPFPRINTF(("fragcache[%d]: adjacent %d-%d (%d-%d)\n",
969 h->ip_id, off, fr_max, fra->fr_off, fra->fr_end));
970 fra->fr_off = off;
971 merge = 1;
972 } else if (aftercut > 0) {
973 /* Need to chop off the tail of this fragment */
974 DPFPRINTF(("fragcache[%d]: chop %d %d-%d (%d-%d)\n",
975 h->ip_id, aftercut, off, fr_max, fra->fr_off,
976 fra->fr_end));
977 fra->fr_off = off;
978 fr_max -= aftercut;
979
980 merge = 1;
981
982 if (!drop) {
983 m_adj(m, -aftercut);
984 if (m->m_flags & M_PKTHDR) {
985 int plen = 0;
986 struct mbuf *t;
987 for (t = m; t; t = t->m_next) {
988 plen += t->m_len;
989 }
990 m->m_pkthdr.len = plen;
991 }
992 h = mtod(m, struct ip *);
993 VERIFY((int)m->m_len ==
994 ntohs(h->ip_len) - aftercut);
995 h->ip_len = htons(ntohs(h->ip_len) - aftercut);
996 } else {
997 hosed++;
998 }
999 } else if (frp == NULL) {
1000 /* There is a gap between fragments */
1001 DPFPRINTF(("fragcache[%d]: gap %d %d-%d (%d-%d)\n",
1002 h->ip_id, -aftercut, off, fr_max, fra->fr_off,
1003 fra->fr_end));
1004
1005 cur = pool_get(&pf_cent_pl, PR_NOWAIT);
1006 if (cur == NULL) {
1007 goto no_mem;
1008 }
1009 pf_ncache++;
1010
1011 cur->fr_off = off;
1012 cur->fr_end = fr_max;
1013 LIST_INSERT_BEFORE(fra, cur, fr_next);
1014 }
1015
1016
1017 /* Need to glue together two separate fragment descriptors */
1018 if (merge) {
1019 if (cur && fra->fr_off <= cur->fr_end) {
1020 /* Need to merge in a previous 'cur' */
1021 DPFPRINTF(("fragcache[%d]: adjacent(merge "
1022 "%d-%d) %d-%d (%d-%d)\n",
1023 h->ip_id, cur->fr_off, cur->fr_end, off,
1024 fr_max, fra->fr_off, fra->fr_end));
1025 fra->fr_off = cur->fr_off;
1026 LIST_REMOVE(cur, fr_next);
1027 pool_put(&pf_cent_pl, cur);
1028 pf_ncache--;
1029 cur = NULL;
1030 } else if (frp && fra->fr_off <= frp->fr_end) {
1031 /* Need to merge in a modified 'frp' */
1032 VERIFY(cur == NULL);
1033 DPFPRINTF(("fragcache[%d]: adjacent(merge "
1034 "%d-%d) %d-%d (%d-%d)\n",
1035 h->ip_id, frp->fr_off, frp->fr_end, off,
1036 fr_max, fra->fr_off, fra->fr_end));
1037 fra->fr_off = frp->fr_off;
1038 LIST_REMOVE(frp, fr_next);
1039 pool_put(&pf_cent_pl, frp);
1040 pf_ncache--;
1041 frp = NULL;
1042 }
1043 }
1044 }
1045
1046 if (hosed) {
1047 /*
1048 * We must keep tracking the overall fragment even when
1049 * we're going to drop it anyway so that we know when to
1050 * free the overall descriptor. Thus we drop the frag late.
1051 */
1052 goto drop_fragment;
1053 }
1054
1055
1056pass:
1057 /* Update maximum data size */
1058 if ((*frag)->fr_max < fr_max) {
1059 (*frag)->fr_max = fr_max;
1060 }
1061
1062 /* This is the last segment */
1063 if (!mff) {
1064 (*frag)->fr_flags |= PFFRAG_SEENLAST;
1065 }
1066
1067 /* Check if we are completely reassembled */
1068 if (((*frag)->fr_flags & PFFRAG_SEENLAST) &&
1069 LIST_FIRST(&(*frag)->fr_cache)->fr_off == 0 &&
1070 LIST_FIRST(&(*frag)->fr_cache)->fr_end == (*frag)->fr_max) {
1071 /* Remove from fragment queue */
1072 DPFPRINTF(("fragcache[%d]: done 0-%d\n", h->ip_id,
1073 (*frag)->fr_max));
1074 pf_free_fragment(frag: *frag);
1075 *frag = NULL;
1076 }
1077
1078 return m;
1079
1080no_mem:
1081 *nomem = 1;
1082
1083 /* Still need to pay attention to !IP_MF */
1084 if (!mff && *frag != NULL) {
1085 (*frag)->fr_flags |= PFFRAG_SEENLAST;
1086 }
1087
1088 m_freem(m);
1089 return NULL;
1090
1091drop_fragment:
1092
1093 /* Still need to pay attention to !IP_MF */
1094 if (!mff && *frag != NULL) {
1095 (*frag)->fr_flags |= PFFRAG_SEENLAST;
1096 }
1097
1098 if (drop) {
1099 /* This fragment has been deemed bad. Don't reass */
1100 if (((*frag)->fr_flags & PFFRAG_DROP) == 0) {
1101 DPFPRINTF(("fragcache[%d]: dropping overall fragment\n",
1102 h->ip_id));
1103 }
1104 (*frag)->fr_flags |= PFFRAG_DROP;
1105 }
1106
1107 m_freem(m);
1108 return NULL;
1109}
1110
1111#define FR_IP6_OFF(fr) \
1112 (ntohs((fr)->fr_ip6f_opt.ip6f_offlg & IP6F_OFF_MASK))
1113#define FR_IP6_PLEN(fr) (ntohs((fr)->fr_ip6->ip6_plen))
1114struct mbuf *
1115pf_reassemble6(struct mbuf **m0, struct pf_fragment **frag,
1116 struct pf_frent *frent, int mff)
1117{
1118 struct mbuf *m, *m2;
1119 struct pf_frent *frea, *frep, *next;
1120 struct ip6_hdr *ip6;
1121 struct ip6_frag *ip6f;
1122 int plen, off, fr_max, pktlen;
1123 uint32_t uoff, csum, csum_flags;
1124
1125 VERIFY(*frag == NULL || BUFFER_FRAGMENTS(*frag));
1126 m = *m0;
1127 frep = NULL;
1128 ip6 = frent->fr_ip6;
1129 ip6f = &frent->fr_ip6f_opt;
1130 off = FR_IP6_OFF(frent);
1131 uoff = frent->fr_ip6f_hlen;
1132 plen = FR_IP6_PLEN(frent);
1133 fr_max = off + plen - (frent->fr_ip6f_hlen - sizeof(*ip6));
1134 pktlen = plen + sizeof(*ip6);
1135
1136 DPFPRINTF(("0x%llx IPv6 frag plen %u off %u fr_ip6f_hlen %u "
1137 "fr_max %u m_len %u\n", (uint64_t)VM_KERNEL_ADDRPERM(m), plen, off,
1138 frent->fr_ip6f_hlen, fr_max, m->m_len));
1139
1140 /*
1141 * Leverage partial checksum offload for simple UDP/IP fragments,
1142 * as that is the most common case.
1143 *
1144 * Perform 1's complement adjustment of octets that got included/
1145 * excluded in the hardware-calculated checksum value. Also take
1146 * care of any trailing bytes and subtract out their partial sum.
1147 */
1148 if (ip6f->ip6f_nxt == IPPROTO_UDP &&
1149 uoff == (sizeof(*ip6) + sizeof(*ip6f)) &&
1150 (m->m_pkthdr.csum_flags &
1151 (CSUM_DATA_VALID | CSUM_PARTIAL | CSUM_PSEUDO_HDR)) ==
1152 (CSUM_DATA_VALID | CSUM_PARTIAL)) {
1153 uint32_t start = m->m_pkthdr.csum_rx_start;
1154 uint32_t ip_len = (sizeof(*ip6) + ntohs(ip6->ip6_plen));
1155 int32_t trailer = (m_pktlen(m) - ip_len);
1156 uint32_t swbytes = (uint32_t)trailer;
1157
1158 csum = m->m_pkthdr.csum_rx_val;
1159
1160 ASSERT(trailer >= 0);
1161 if (start != uoff || trailer != 0) {
1162 uint16_t s = 0, d = 0;
1163
1164 if (IN6_IS_SCOPE_EMBED(&ip6->ip6_src)) {
1165 s = ip6->ip6_src.s6_addr16[1];
1166 ip6->ip6_src.s6_addr16[1] = 0;
1167 }
1168 if (IN6_IS_SCOPE_EMBED(&ip6->ip6_dst)) {
1169 d = ip6->ip6_dst.s6_addr16[1];
1170 ip6->ip6_dst.s6_addr16[1] = 0;
1171 }
1172
1173 /* callee folds in sum */
1174 csum = m_adj_sum16(m, start, uoff,
1175 (ip_len - uoff), csum);
1176 if (uoff > start) {
1177 swbytes += (uoff - start);
1178 } else {
1179 swbytes += (start - uoff);
1180 }
1181
1182 if (IN6_IS_SCOPE_EMBED(&ip6->ip6_src)) {
1183 ip6->ip6_src.s6_addr16[1] = s;
1184 }
1185 if (IN6_IS_SCOPE_EMBED(&ip6->ip6_dst)) {
1186 ip6->ip6_dst.s6_addr16[1] = d;
1187 }
1188 }
1189 csum_flags = m->m_pkthdr.csum_flags;
1190
1191 if (swbytes != 0) {
1192 udp_in6_cksum_stats(swbytes);
1193 }
1194 if (trailer != 0) {
1195 m_adj(m, -trailer);
1196 }
1197 } else {
1198 csum = 0;
1199 csum_flags = 0;
1200 }
1201
1202 /* Invalidate checksum */
1203 m->m_pkthdr.csum_flags &= ~CSUM_DATA_VALID;
1204
1205 /* strip off headers up to the fragment payload */
1206 m->m_data += frent->fr_ip6f_hlen;
1207 m->m_len -= frent->fr_ip6f_hlen;
1208
1209 /* Create a new reassembly queue for this packet */
1210 if (*frag == NULL) {
1211 *frag = pool_get(&pf_frag_pl, PR_NOWAIT);
1212 if (*frag == NULL) {
1213 pf_flush_fragments();
1214 *frag = pool_get(&pf_frag_pl, PR_NOWAIT);
1215 if (*frag == NULL) {
1216 goto drop_fragment;
1217 }
1218 }
1219
1220 (*frag)->fr_flags = 0;
1221 (*frag)->fr_max = 0;
1222 (*frag)->fr_ip6_maxlen = pktlen;
1223 (*frag)->fr_af = AF_INET6;
1224 (*frag)->fr_srcx.v6addr = frent->fr_ip6->ip6_src;
1225 (*frag)->fr_dstx.v6addr = frent->fr_ip6->ip6_dst;
1226 (*frag)->fr_p = frent->fr_ip6f_opt.ip6f_nxt;
1227 (*frag)->fr_id6 = frent->fr_ip6f_opt.ip6f_ident;
1228 (*frag)->fr_timeout = pf_time_second();
1229 if (csum_flags != 0) {
1230 (*frag)->fr_csum_flags = csum_flags;
1231 (*frag)->fr_csum = csum;
1232 }
1233 LIST_INIT(&(*frag)->fr_queue);
1234
1235 RB_INSERT(pf_frag_tree, &pf_frag_tree, *frag);
1236 TAILQ_INSERT_HEAD(&pf_fragqueue, *frag, frag_next);
1237
1238 /* We do not have a previous fragment */
1239 frep = NULL;
1240 goto insert;
1241 }
1242
1243 /* Remember maximum fragment len for refragmentation */
1244 if (pktlen > (*frag)->fr_ip6_maxlen) {
1245 (*frag)->fr_ip6_maxlen = pktlen;
1246 }
1247 /*
1248 * If this fragment contains similar checksum offload info
1249 * as that of the existing ones, accumulate checksum. Otherwise,
1250 * invalidate checksum offload info for the entire datagram.
1251 */
1252 if (csum_flags != 0 && csum_flags == (*frag)->fr_csum_flags) {
1253 (*frag)->fr_csum += csum;
1254 } else if ((*frag)->fr_csum_flags != 0) {
1255 (*frag)->fr_csum_flags = 0;
1256 }
1257
1258 /*
1259 * Find a fragment after the current one:
1260 * - off contains the real shifted offset.
1261 */
1262 LIST_FOREACH(frea, &(*frag)->fr_queue, fr_next) {
1263 if (FR_IP6_OFF(frea) > off) {
1264 break;
1265 }
1266 frep = frea;
1267 }
1268
1269 VERIFY(frep != NULL || frea != NULL);
1270
1271 if (frep != NULL &&
1272 FR_IP6_OFF(frep) + FR_IP6_PLEN(frep) - frep->fr_ip6f_hlen > off) {
1273 u_int16_t precut;
1274
1275 precut = FR_IP6_OFF(frep) + FR_IP6_PLEN(frep) -
1276 frep->fr_ip6f_hlen - off;
1277 if (precut >= plen) {
1278 goto drop_fragment;
1279 }
1280 m_adj(frent->fr_m, precut);
1281 DPFPRINTF(("overlap -%d\n", precut));
1282 /* Enforce 8 byte boundaries */
1283 frent->fr_ip6f_opt.ip6f_offlg =
1284 htons(ntohs(frent->fr_ip6f_opt.ip6f_offlg) +
1285 (precut >> 3));
1286 off = FR_IP6_OFF(frent);
1287 plen -= precut;
1288 ip6->ip6_plen = htons(plen);
1289 }
1290
1291 for (; frea != NULL && plen + off > FR_IP6_OFF(frea); frea = next) {
1292 u_int16_t aftercut;
1293
1294 aftercut = plen + off - FR_IP6_OFF(frea);
1295 DPFPRINTF(("adjust overlap %d\n", aftercut));
1296 if (aftercut < FR_IP6_PLEN(frea) - frea->fr_ip6f_hlen) {
1297 frea->fr_ip6->ip6_plen = htons(FR_IP6_PLEN(frea) -
1298 aftercut);
1299 frea->fr_ip6f_opt.ip6f_offlg =
1300 htons(ntohs(frea->fr_ip6f_opt.ip6f_offlg) +
1301 (aftercut >> 3));
1302 m_adj(frea->fr_m, aftercut);
1303 break;
1304 }
1305
1306 /* This fragment is completely overlapped, lose it */
1307 next = LIST_NEXT(frea, fr_next);
1308 m_freem(frea->fr_m);
1309 LIST_REMOVE(frea, fr_next);
1310 pool_put(&pf_frent_pl, frea);
1311 pf_nfrents--;
1312 }
1313
1314insert:
1315 /* Update maximum data size */
1316 if ((*frag)->fr_max < fr_max) {
1317 (*frag)->fr_max = fr_max;
1318 }
1319 /* This is the last segment */
1320 if (!mff) {
1321 (*frag)->fr_flags |= PFFRAG_SEENLAST;
1322 }
1323
1324 if (frep == NULL) {
1325 LIST_INSERT_HEAD(&(*frag)->fr_queue, frent, fr_next);
1326 } else {
1327 LIST_INSERT_AFTER(frep, frent, fr_next);
1328 }
1329
1330 /* Check if we are completely reassembled */
1331 if (!((*frag)->fr_flags & PFFRAG_SEENLAST)) {
1332 return NULL;
1333 }
1334
1335 /* Check if we have all the data */
1336 off = 0;
1337 for (frep = LIST_FIRST(&(*frag)->fr_queue); frep; frep = next) {
1338 next = LIST_NEXT(frep, fr_next);
1339 off += FR_IP6_PLEN(frep) - (frent->fr_ip6f_hlen - sizeof *ip6);
1340 DPFPRINTF(("frep at %d, next %d, max %d\n",
1341 off, next == NULL ? -1 : FR_IP6_OFF(next),
1342 (*frag)->fr_max));
1343 if (off < (*frag)->fr_max &&
1344 (next == NULL || FR_IP6_OFF(next) != off)) {
1345 DPFPRINTF(("missing fragment at %d, next %d, max %d\n",
1346 off, next == NULL ? -1 : FR_IP6_OFF(next),
1347 (*frag)->fr_max));
1348 return NULL;
1349 }
1350 }
1351 DPFPRINTF(("%d < %d?\n", off, (*frag)->fr_max));
1352 if (off < (*frag)->fr_max) {
1353 return NULL;
1354 }
1355
1356 /* We have all the data */
1357 frent = LIST_FIRST(&(*frag)->fr_queue);
1358 VERIFY(frent != NULL);
1359 if (frent->fr_ip6f_hlen + off > IP_MAXPACKET) {
1360 DPFPRINTF(("drop: too big: %d\n", off));
1361 pf_free_fragment(frag: *frag);
1362 *frag = NULL;
1363 return NULL;
1364 }
1365
1366 ASSERT(*frag != NULL);
1367 ASSERT(frent != NULL);
1368 next = LIST_NEXT(frent, fr_next);
1369 if (next == NULL) {
1370 DPFPRINTF(("drop: atomic fragment\n"));
1371 pf_free_fragment(frag: *frag);
1372 *frag = NULL;
1373 return NULL;
1374 }
1375
1376 /* retrieve the values to be filled in to reassembled tag */
1377 uint16_t hdrlen, unfragpartlen, extoff, maxlen;
1378 uint32_t id;
1379
1380 /* Get total extension header length from the first fragment */
1381 hdrlen = frent->fr_ip6f_hlen - sizeof(struct ip6_frag);
1382 /*
1383 * Get total extension header length of per-fragment headers from the
1384 * subsequent fragment.
1385 */
1386 unfragpartlen = next->fr_ip6f_hlen - sizeof(struct ip6_frag);
1387 extoff = frent->fr_ip6f_extoff;
1388 maxlen = (*frag)->fr_ip6_maxlen;
1389 id = (*frag)->fr_id6;
1390
1391 ip6 = frent->fr_ip6;
1392 ip6->ip6_nxt = (*frag)->fr_p;
1393 ip6->ip6_plen = htons(off);
1394 ip6->ip6_src = (*frag)->fr_srcx.v6addr;
1395 ip6->ip6_dst = (*frag)->fr_dstx.v6addr;
1396
1397 if ((*frag)->fr_csum_flags != 0) {
1398 csum = (*frag)->fr_csum;
1399
1400 ADDCARRY(csum);
1401
1402 m->m_pkthdr.csum_rx_val = csum;
1403 m->m_pkthdr.csum_rx_start = sizeof(struct ip6_hdr);
1404 m->m_pkthdr.csum_flags = (*frag)->fr_csum_flags;
1405 } else if ((m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) ||
1406 (m->m_pkthdr.pkt_flags & PKTF_LOOP)) {
1407 /* loopback checksums are always OK */
1408 m->m_pkthdr.csum_data = 0xffff;
1409 m->m_pkthdr.csum_flags = CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1410 }
1411
1412 /* Remove from fragment queue */
1413 pf_remove_fragment(frag: *frag);
1414 *frag = NULL;
1415
1416 m = frent->fr_m;
1417 m->m_len += sizeof(struct ip6_hdr);
1418 m->m_data -= sizeof(struct ip6_hdr);
1419 memmove(dst: m_mtod_current(m), src: ip6, n: sizeof(struct ip6_hdr));
1420
1421 next = LIST_NEXT(frent, fr_next);
1422 pool_put(&pf_frent_pl, frent);
1423 pf_nfrents--;
1424 for (frent = next; next != NULL; frent = next) {
1425 m2 = frent->fr_m;
1426
1427 m_cat(m, m2);
1428 next = LIST_NEXT(frent, fr_next);
1429 pool_put(&pf_frent_pl, frent);
1430 pf_nfrents--;
1431 }
1432
1433 /* XXX this should be done elsewhere */
1434 if (m->m_flags & M_PKTHDR) {
1435 int len = 0;
1436 for (m2 = m; m2; m2 = m2->m_next) {
1437 len += m2->m_len;
1438 }
1439 m->m_pkthdr.len = len;
1440 }
1441
1442 DPFPRINTF(("complete: 0x%llx ip6_plen %d m_pkthdr.len %d\n",
1443 (uint64_t)VM_KERNEL_ADDRHASH(m), ntohs(ip6->ip6_plen),
1444 m->m_pkthdr.len));
1445
1446 /* Add the reassembled tag */
1447 struct m_tag *mtag;
1448 struct pf_fragment_tag *ftag;
1449 mtag = m_tag_create(KERNEL_MODULE_TAG_ID, KERNEL_TAG_TYPE_PF_REASS,
1450 sizeof(*ftag), M_NOWAIT, m);
1451 if (mtag == NULL) {
1452 /* XXX: add stats */
1453 m_freem(m);
1454 return NULL;
1455 }
1456 ftag = (struct pf_fragment_tag *)mtag->m_tag_data;
1457 ftag->ft_hdrlen = hdrlen;
1458 ftag->ft_unfragpartlen = unfragpartlen;
1459 ftag->ft_extoff = extoff;
1460 ftag->ft_maxlen = maxlen;
1461 ftag->ft_id = id;
1462 m_tag_prepend(m, mtag);
1463
1464 struct pf_mtag *pftag = pf_get_mtag(m);
1465 ASSERT(pftag != NULL);
1466 pftag->pftag_flags |= PF_TAG_REASSEMBLED;
1467 return m;
1468
1469drop_fragment:
1470 /* Oops - fail safe - drop packet */
1471 pool_put(&pf_frent_pl, frent);
1472 --pf_nfrents;
1473 m_freem(m);
1474 return NULL;
1475}
1476
1477static __attribute__((noinline)) struct mbuf *
1478pf_frag6cache(struct mbuf **m0, struct ip6_hdr *h, struct ip6_frag *fh,
1479 struct pf_fragment **frag, int hlen, int mff, int drop, int *nomem)
1480{
1481 struct mbuf *m = *m0;
1482 u_int16_t plen, off, fr_max;
1483 struct pf_frcache *frp, *fra, *cur = NULL;
1484 int hosed = 0;
1485
1486 VERIFY(*frag == NULL || !BUFFER_FRAGMENTS(*frag));
1487 m = *m0;
1488 off = ntohs(fh->ip6f_offlg & IP6F_OFF_MASK);
1489 plen = ntohs(h->ip6_plen) - (hlen - sizeof *h);
1490
1491 /*
1492 * Apple Modification: dimambro@apple.com. The hlen, being passed
1493 * into this function Includes all the headers associated with
1494 * the packet, and may include routing headers, so to get to
1495 * the data payload as stored in the original IPv6 header we need
1496 * to subtract al those headers and the IP header.
1497 *
1498 * The 'max' local variable should also contain the offset from the start
1499 * of the reassembled packet to the octet just past the end of the octets
1500 * in the current fragment where:
1501 * - 'off' is the offset from the start of the reassembled packet to the
1502 * first octet in the fragment,
1503 * - 'plen' is the length of the "payload data length" Excluding all the
1504 * IPv6 headers of the fragment.
1505 * - 'hlen' is computed in pf_normalize_ip6() as the offset from the start
1506 * of the IPv6 packet to the beginning of the data.
1507 */
1508 fr_max = off + plen;
1509
1510 DPFPRINTF(("0x%llx plen %u off %u fr_max %u\n",
1511 (uint64_t)VM_KERNEL_ADDRHASH(m), plen, off, fr_max));
1512
1513 /* Create a new range queue for this packet */
1514 if (*frag == NULL) {
1515 *frag = pool_get(&pf_cache_pl, PR_NOWAIT);
1516 if (*frag == NULL) {
1517 pf_flush_fragments();
1518 *frag = pool_get(&pf_cache_pl, PR_NOWAIT);
1519 if (*frag == NULL) {
1520 goto no_mem;
1521 }
1522 }
1523
1524 /* Get an entry for the queue */
1525 cur = pool_get(&pf_cent_pl, PR_NOWAIT);
1526 if (cur == NULL) {
1527 pool_put(&pf_cache_pl, *frag);
1528 *frag = NULL;
1529 goto no_mem;
1530 }
1531 pf_ncache++;
1532
1533 (*frag)->fr_flags = PFFRAG_NOBUFFER;
1534 (*frag)->fr_max = 0;
1535 (*frag)->fr_af = AF_INET6;
1536 (*frag)->fr_srcx.v6addr = h->ip6_src;
1537 (*frag)->fr_dstx.v6addr = h->ip6_dst;
1538 (*frag)->fr_p = fh->ip6f_nxt;
1539 (*frag)->fr_id6 = fh->ip6f_ident;
1540 (*frag)->fr_timeout = pf_time_second();
1541
1542 cur->fr_off = off;
1543 cur->fr_end = fr_max;
1544 LIST_INIT(&(*frag)->fr_cache);
1545 LIST_INSERT_HEAD(&(*frag)->fr_cache, cur, fr_next);
1546
1547 RB_INSERT(pf_frag_tree, &pf_cache_tree, *frag);
1548 TAILQ_INSERT_HEAD(&pf_cachequeue, *frag, frag_next);
1549
1550 DPFPRINTF(("frag6cache[%d]: new %d-%d\n", ntohl(fh->ip6f_ident),
1551 off, fr_max));
1552
1553 goto pass;
1554 }
1555
1556 /*
1557 * Find a fragment after the current one:
1558 * - off contains the real shifted offset.
1559 */
1560 frp = NULL;
1561 LIST_FOREACH(fra, &(*frag)->fr_cache, fr_next) {
1562 if (fra->fr_off > off) {
1563 break;
1564 }
1565 frp = fra;
1566 }
1567
1568 VERIFY(frp != NULL || fra != NULL);
1569
1570 if (frp != NULL) {
1571 int precut;
1572
1573 precut = frp->fr_end - off;
1574 if (precut >= plen) {
1575 /* Fragment is entirely a duplicate */
1576 DPFPRINTF(("frag6cache[%u]: dead (%d-%d) %d-%d\n",
1577 ntohl(fh->ip6f_ident), frp->fr_off, frp->fr_end,
1578 off, fr_max));
1579 goto drop_fragment;
1580 }
1581 if (precut == 0) {
1582 /* They are adjacent. Fixup cache entry */
1583 DPFPRINTF(("frag6cache[%u]: adjacent (%d-%d) %d-%d\n",
1584 ntohl(fh->ip6f_ident), frp->fr_off, frp->fr_end,
1585 off, fr_max));
1586 frp->fr_end = fr_max;
1587 } else if (precut > 0) {
1588 /* The first part of this payload overlaps with a
1589 * fragment that has already been passed.
1590 * Need to trim off the first part of the payload.
1591 * But to do so easily, we need to create another
1592 * mbuf to throw the original header into.
1593 */
1594
1595 DPFPRINTF(("frag6cache[%u]: chop %d (%d-%d) %d-%d\n",
1596 ntohl(fh->ip6f_ident), precut, frp->fr_off,
1597 frp->fr_end, off, fr_max));
1598
1599 off += precut;
1600 fr_max -= precut;
1601 /* Update the previous frag to encompass this one */
1602 frp->fr_end = fr_max;
1603
1604 if (!drop) {
1605 /* XXX Optimization opportunity
1606 * This is a very heavy way to trim the payload.
1607 * we could do it much faster by diddling mbuf
1608 * internals but that would be even less legible
1609 * than this mbuf magic. For my next trick,
1610 * I'll pull a rabbit out of my laptop.
1611 */
1612 *m0 = m_copym(m, 0, hlen, M_NOWAIT);
1613 if (*m0 == NULL) {
1614 goto no_mem;
1615 }
1616 VERIFY((*m0)->m_next == NULL);
1617 m_adj(m, precut + hlen);
1618 m_cat(*m0, m);
1619 m = *m0;
1620 if (m->m_flags & M_PKTHDR) {
1621 int pktlen = 0;
1622 struct mbuf *t;
1623 for (t = m; t; t = t->m_next) {
1624 pktlen += t->m_len;
1625 }
1626 m->m_pkthdr.len = pktlen;
1627 }
1628
1629 h = mtod(m, struct ip6_hdr *);
1630
1631 VERIFY((int)m->m_len ==
1632 ntohs(h->ip6_plen) - precut);
1633 fh->ip6f_offlg &= ~IP6F_OFF_MASK;
1634 fh->ip6f_offlg |=
1635 htons(ntohs(fh->ip6f_offlg & IP6F_OFF_MASK)
1636 + (precut >> 3));
1637 h->ip6_plen = htons(ntohs(h->ip6_plen) -
1638 precut);
1639 } else {
1640 hosed++;
1641 }
1642 } else {
1643 /* There is a gap between fragments */
1644
1645 DPFPRINTF(("frag6cache[%u]: gap %d (%d-%d) %d-%d\n",
1646 ntohl(fh->ip6f_ident), -precut, frp->fr_off,
1647 frp->fr_end, off, fr_max));
1648
1649 cur = pool_get(&pf_cent_pl, PR_NOWAIT);
1650 if (cur == NULL) {
1651 goto no_mem;
1652 }
1653 pf_ncache++;
1654
1655 cur->fr_off = off;
1656 cur->fr_end = fr_max;
1657 LIST_INSERT_AFTER(frp, cur, fr_next);
1658 }
1659 }
1660
1661 if (fra != NULL) {
1662 int aftercut;
1663 int merge = 0;
1664
1665 aftercut = fr_max - fra->fr_off;
1666 if (aftercut == 0) {
1667 /* Adjacent fragments */
1668 DPFPRINTF(("frag6cache[%u]: adjacent %d-%d (%d-%d)\n",
1669 ntohl(fh->ip6f_ident), off, fr_max, fra->fr_off,
1670 fra->fr_end));
1671 fra->fr_off = off;
1672 merge = 1;
1673 } else if (aftercut > 0) {
1674 /* Need to chop off the tail of this fragment */
1675 DPFPRINTF(("frag6cache[%u]: chop %d %d-%d (%d-%d)\n",
1676 ntohl(fh->ip6f_ident), aftercut, off, fr_max,
1677 fra->fr_off, fra->fr_end));
1678 fra->fr_off = off;
1679 fr_max -= aftercut;
1680
1681 merge = 1;
1682
1683 if (!drop) {
1684 m_adj(m, -aftercut);
1685 if (m->m_flags & M_PKTHDR) {
1686 int pktlen = 0;
1687 struct mbuf *t;
1688 for (t = m; t; t = t->m_next) {
1689 pktlen += t->m_len;
1690 }
1691 m->m_pkthdr.len = pktlen;
1692 }
1693 h = mtod(m, struct ip6_hdr *);
1694 VERIFY((int)m->m_len ==
1695 ntohs(h->ip6_plen) - aftercut);
1696 h->ip6_plen =
1697 htons(ntohs(h->ip6_plen) - aftercut);
1698 } else {
1699 hosed++;
1700 }
1701 } else if (frp == NULL) {
1702 /* There is a gap between fragments */
1703 DPFPRINTF(("frag6cache[%u]: gap %d %d-%d (%d-%d)\n",
1704 ntohl(fh->ip6f_ident), -aftercut, off, fr_max,
1705 fra->fr_off, fra->fr_end));
1706
1707 cur = pool_get(&pf_cent_pl, PR_NOWAIT);
1708 if (cur == NULL) {
1709 goto no_mem;
1710 }
1711 pf_ncache++;
1712
1713 cur->fr_off = off;
1714 cur->fr_end = fr_max;
1715 LIST_INSERT_BEFORE(fra, cur, fr_next);
1716 }
1717
1718 /* Need to glue together two separate fragment descriptors */
1719 if (merge) {
1720 if (cur && fra->fr_off <= cur->fr_end) {
1721 /* Need to merge in a previous 'cur' */
1722 DPFPRINTF(("frag6cache[%u]: adjacent(merge "
1723 "%d-%d) %d-%d (%d-%d)\n",
1724 ntohl(fh->ip6f_ident), cur->fr_off,
1725 cur->fr_end, off, fr_max, fra->fr_off,
1726 fra->fr_end));
1727 fra->fr_off = cur->fr_off;
1728 LIST_REMOVE(cur, fr_next);
1729 pool_put(&pf_cent_pl, cur);
1730 pf_ncache--;
1731 cur = NULL;
1732 } else if (frp && fra->fr_off <= frp->fr_end) {
1733 /* Need to merge in a modified 'frp' */
1734 VERIFY(cur == NULL);
1735 DPFPRINTF(("frag6cache[%u]: adjacent(merge "
1736 "%d-%d) %d-%d (%d-%d)\n",
1737 ntohl(fh->ip6f_ident), frp->fr_off,
1738 frp->fr_end, off, fr_max, fra->fr_off,
1739 fra->fr_end));
1740 fra->fr_off = frp->fr_off;
1741 LIST_REMOVE(frp, fr_next);
1742 pool_put(&pf_cent_pl, frp);
1743 pf_ncache--;
1744 frp = NULL;
1745 }
1746 }
1747 }
1748
1749 if (hosed) {
1750 /*
1751 * We must keep tracking the overall fragment even when
1752 * we're going to drop it anyway so that we know when to
1753 * free the overall descriptor. Thus we drop the frag late.
1754 */
1755 goto drop_fragment;
1756 }
1757
1758pass:
1759 /* Update maximum data size */
1760 if ((*frag)->fr_max < fr_max) {
1761 (*frag)->fr_max = fr_max;
1762 }
1763
1764 /* This is the last segment */
1765 if (!mff) {
1766 (*frag)->fr_flags |= PFFRAG_SEENLAST;
1767 }
1768
1769 /* Check if we are completely reassembled */
1770 if (((*frag)->fr_flags & PFFRAG_SEENLAST) &&
1771 LIST_FIRST(&(*frag)->fr_cache)->fr_off == 0 &&
1772 LIST_FIRST(&(*frag)->fr_cache)->fr_end == (*frag)->fr_max) {
1773 /* Remove from fragment queue */
1774 DPFPRINTF(("frag6cache[%u]: done 0-%d\n",
1775 ntohl(fh->ip6f_ident), (*frag)->fr_max));
1776 pf_free_fragment(frag: *frag);
1777 *frag = NULL;
1778 }
1779
1780 return m;
1781
1782no_mem:
1783 *nomem = 1;
1784
1785 /* Still need to pay attention to !IP_MF */
1786 if (!mff && *frag != NULL) {
1787 (*frag)->fr_flags |= PFFRAG_SEENLAST;
1788 }
1789
1790 m_freem(m);
1791 return NULL;
1792
1793drop_fragment:
1794
1795 /* Still need to pay attention to !IP_MF */
1796 if (!mff && *frag != NULL) {
1797 (*frag)->fr_flags |= PFFRAG_SEENLAST;
1798 }
1799
1800 if (drop) {
1801 /* This fragment has been deemed bad. Don't reass */
1802 if (((*frag)->fr_flags & PFFRAG_DROP) == 0) {
1803 DPFPRINTF(("frag6cache[%u]: dropping overall fragment\n",
1804 ntohl(fh->ip6f_ident)));
1805 }
1806 (*frag)->fr_flags |= PFFRAG_DROP;
1807 }
1808
1809 m_freem(m);
1810 return NULL;
1811}
1812
1813int
1814pf_refragment6(struct ifnet *ifp, pbuf_t **pbufp, struct pf_fragment_tag *ftag)
1815{
1816 struct mbuf *m;
1817 uint32_t frag_id;
1818 uint16_t hdrlen, extoff, maxlen, unfragpartlen;
1819 uint8_t proto;
1820 int error, action;
1821 uint8_t *lexthdrsp;
1822 struct route_in6 ip6route;
1823 struct route_in6 *ro;
1824 struct sockaddr_in6 *dst;
1825 struct ip6_hdr *hdr;
1826 struct pf_mtag *mtag;
1827 struct m_tag *tag;
1828
1829 if (pbufp == NULL || !pbuf_is_valid(*pbufp) || ftag == NULL) {
1830 panic("pf_route6: invalid parameters");
1831 /* NOT REACHED */
1832 }
1833 m = pbuf_to_mbuf(*pbufp, FALSE);
1834 hdr = mtod(m, struct ip6_hdr *);
1835 mtag = pf_find_mtag(m);
1836 hdrlen = ftag->ft_hdrlen - sizeof(struct ip6_hdr);
1837 extoff = ftag->ft_extoff;
1838 maxlen = ftag->ft_maxlen;
1839 frag_id = ftag->ft_id;
1840 unfragpartlen = ftag->ft_unfragpartlen;
1841 tag = m_tag_locate(m, KERNEL_MODULE_TAG_ID, KERNEL_TAG_TYPE_PF_REASS);
1842 m_tag_delete(m, tag);
1843 ftag = NULL;
1844 tag = NULL;
1845 mtag->pftag_flags &= ~PF_TAG_REASSEMBLED;
1846 ro = &ip6route;
1847 bzero(s: (caddr_t)ro, n: sizeof(*ro));
1848 dst = (struct sockaddr_in6 *)&ro->ro_dst;
1849 dst->sin6_family = AF_INET6;
1850 dst->sin6_len = sizeof(*dst);
1851 dst->sin6_addr = hdr->ip6_dst;
1852
1853 if (extoff) {
1854 int off;
1855 struct mbuf *mexthdr;
1856
1857 /* Use protocol from next field of last extension header */
1858 mexthdr = m_getptr(m, extoff +
1859 offsetof(struct ip6_ext, ip6e_nxt), &off);
1860 ASSERT(mexthdr != NULL);
1861 lexthdrsp = (mtod(mexthdr, uint8_t *) + off);
1862 proto = *lexthdrsp;
1863 if (proto == IPPROTO_DSTOPTS) {
1864 struct ip6_ext ext;
1865 if (!pf_pull_hdr(*pbufp, off, &ext, sizeof(ext), NULL,
1866 NULL, AF_INET6)) {
1867 DPFPRINTF(("pkt too short"));
1868 action = PF_DROP;
1869 goto done;
1870 }
1871 proto = ext.ip6e_nxt;
1872 }
1873 } else {
1874 lexthdrsp = NULL;
1875 proto = hdr->ip6_nxt;
1876 }
1877
1878 /*
1879 * The MTU must be a multiple of 8 bytes, or we risk doing the
1880 * fragmentation wrong.
1881 */
1882 maxlen = maxlen & ~7;
1883
1884 error = ip6_do_fragmentation(&m, hdrlen, NULL, unfragpartlen,
1885 hdr, lexthdrsp, maxlen, proto, frag_id);
1886
1887 if (error == 0) {
1888 /*
1889 * PF_TAG_REFRAGMENTED flag set to indicate ip6_forward()
1890 * and pf_route6() that the mbuf contains a chain of fragments.
1891 */
1892 mtag->pftag_flags |= PF_TAG_REFRAGMENTED;
1893 action = PF_PASS;
1894 pbuf_init_mbuf(*pbufp, m, ifp);
1895 } else {
1896 DPFPRINTF(("refragment error %d", error));
1897 action = PF_DROP;
1898 goto done;
1899 }
1900done:
1901 return action;
1902}
1903
1904int
1905pf_normalize_ip(pbuf_t *pbuf, int dir, struct pfi_kif *kif, u_short *reason,
1906 struct pf_pdesc *pd)
1907{
1908 struct mbuf *m;
1909 struct pf_rule *r;
1910 struct pf_frent *frent;
1911 struct pf_fragment *frag = NULL;
1912 struct ip *h = pbuf->pb_data;
1913 int mff = (ntohs(h->ip_off) & IP_MF);
1914 int hlen = h->ip_hl << 2;
1915 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
1916 u_int16_t fr_max;
1917 int ip_len;
1918 int ip_off;
1919 int asd = 0;
1920 struct pf_ruleset *ruleset = NULL;
1921 struct ifnet *ifp = pbuf->pb_ifp;
1922 uint64_t ipid_salt = (uint64_t)pbuf_get_packet_buffer_address(pbuf);
1923
1924 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1925 while (r != NULL) {
1926 r->evaluations++;
1927 if (pfi_kif_match(r->kif, kif) == r->ifnot) {
1928 r = r->skip[PF_SKIP_IFP].ptr;
1929 } else if (r->direction && r->direction != dir) {
1930 r = r->skip[PF_SKIP_DIR].ptr;
1931 } else if (r->af && r->af != AF_INET) {
1932 r = r->skip[PF_SKIP_AF].ptr;
1933 } else if (r->proto && r->proto != h->ip_p) {
1934 r = r->skip[PF_SKIP_PROTO].ptr;
1935 } else if (PF_MISMATCHAW(&r->src.addr,
1936 (struct pf_addr *)&h->ip_src.s_addr, AF_INET,
1937 r->src.neg, kif)) {
1938 r = r->skip[PF_SKIP_SRC_ADDR].ptr;
1939 } else if (PF_MISMATCHAW(&r->dst.addr,
1940 (struct pf_addr *)&h->ip_dst.s_addr, AF_INET,
1941 r->dst.neg, NULL)) {
1942 r = r->skip[PF_SKIP_DST_ADDR].ptr;
1943 } else {
1944 if (r->anchor == NULL) {
1945 break;
1946 } else {
1947 pf_step_into_anchor(&asd, &ruleset,
1948 PF_RULESET_SCRUB, &r, NULL, NULL);
1949 }
1950 }
1951 if (r == NULL && pf_step_out_of_anchor(&asd, &ruleset,
1952 PF_RULESET_SCRUB, &r, NULL, NULL)) {
1953 break;
1954 }
1955 }
1956
1957 if (r == NULL || r->action == PF_NOSCRUB) {
1958 return PF_PASS;
1959 } else {
1960 r->packets[dir == PF_OUT]++;
1961 r->bytes[dir == PF_OUT] += pd->tot_len;
1962 }
1963
1964 /* Check for illegal packets */
1965 if (hlen < (int)sizeof(struct ip)) {
1966 goto drop;
1967 }
1968
1969 if (hlen > ntohs(h->ip_len)) {
1970 goto drop;
1971 }
1972
1973 /* Clear IP_DF if the rule uses the no-df option */
1974 if (r->rule_flag & PFRULE_NODF && h->ip_off & htons(IP_DF)) {
1975 u_int16_t ipoff = h->ip_off;
1976
1977 h->ip_off &= htons(~IP_DF);
1978 h->ip_sum = pf_cksum_fixup(h->ip_sum, ipoff, h->ip_off, 0);
1979 }
1980
1981 /* We will need other tests here */
1982 if (!fragoff && !mff) {
1983 goto no_fragment;
1984 }
1985
1986 /*
1987 * We're dealing with a fragment now. Don't allow fragments
1988 * with IP_DF to enter the cache. If the flag was cleared by
1989 * no-df above, fine. Otherwise drop it.
1990 */
1991 if (h->ip_off & htons(IP_DF)) {
1992 DPFPRINTF(("IP_DF\n"));
1993 goto bad;
1994 }
1995
1996 ip_len = ntohs(h->ip_len) - hlen;
1997 ip_off = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
1998
1999 /* All fragments are 8 byte aligned */
2000 if (mff && (ip_len & 0x7)) {
2001 DPFPRINTF(("mff and %d\n", ip_len));
2002 goto bad;
2003 }
2004
2005 /* Respect maximum length */
2006 if (fragoff + ip_len > IP_MAXPACKET) {
2007 DPFPRINTF(("max packet %d\n", fragoff + ip_len));
2008 goto bad;
2009 }
2010 fr_max = fragoff + ip_len;
2011
2012 if ((r->rule_flag & (PFRULE_FRAGCROP | PFRULE_FRAGDROP)) == 0) {
2013 /* Fully buffer all of the fragments */
2014
2015 frag = pf_find_fragment_by_ipv4_header(ip: h, tree: &pf_frag_tree);
2016 /* Check if we saw the last fragment already */
2017 if (frag != NULL && (frag->fr_flags & PFFRAG_SEENLAST) &&
2018 fr_max > frag->fr_max) {
2019 goto bad;
2020 }
2021
2022 if ((m = pbuf_to_mbuf(pbuf, TRUE)) == NULL) {
2023 REASON_SET(reason, PFRES_MEMORY);
2024 return PF_DROP;
2025 }
2026
2027 VERIFY(!pbuf_is_valid(pbuf));
2028
2029 /* Restore iph pointer after pbuf_to_mbuf() */
2030 h = mtod(m, struct ip *);
2031
2032 /* Get an entry for the fragment queue */
2033 frent = pool_get(&pf_frent_pl, PR_NOWAIT);
2034 if (frent == NULL) {
2035 REASON_SET(reason, PFRES_MEMORY);
2036 m_freem(m);
2037 return PF_DROP;
2038 }
2039 pf_nfrents++;
2040 frent->fr_ip = h;
2041 frent->fr_m = m;
2042
2043 /* Might return a completely reassembled mbuf, or NULL */
2044 DPFPRINTF(("reass IPv4 frag %d @ %d-%d\n", ntohs(h->ip_id),
2045 fragoff, fr_max));
2046 m = pf_reassemble(m0: m, frag: &frag, frent, mff);
2047
2048 if (m == NULL) {
2049 return PF_DROP;
2050 }
2051
2052 VERIFY(m->m_flags & M_PKTHDR);
2053 pbuf_init_mbuf(pbuf, m, ifp);
2054
2055 /* use mtag from concatenated mbuf chain */
2056 pd->pf_mtag = pf_find_mtag_pbuf(pbuf);
2057#if 0
2058// SCW: This check is superfluous
2059#if DIAGNOSTIC
2060 if (pd->pf_mtag == NULL) {
2061 printf("%s: pf_find_mtag returned NULL(1)\n", __func__);
2062 if ((pd->pf_mtag = pf_get_mtag(m)) == NULL) {
2063 m_freem(m);
2064 m = NULL;
2065 goto no_mem;
2066 }
2067 }
2068#endif
2069#endif
2070
2071 h = mtod(m, struct ip *);
2072
2073 if (frag != NULL && (frag->fr_flags & PFFRAG_DROP)) {
2074 goto drop;
2075 }
2076 } else {
2077 /* non-buffering fragment cache (drops or masks overlaps) */
2078 int nomem = 0;
2079
2080 if (dir == PF_OUT && (pd->pf_mtag->pftag_flags & PF_TAG_FRAGCACHE)) {
2081 /*
2082 * Already passed the fragment cache in the
2083 * input direction. If we continued, it would
2084 * appear to be a dup and would be dropped.
2085 */
2086 goto fragment_pass;
2087 }
2088
2089 frag = pf_find_fragment_by_ipv4_header(ip: h, tree: &pf_cache_tree);
2090
2091 /* Check if we saw the last fragment already */
2092 if (frag != NULL && (frag->fr_flags & PFFRAG_SEENLAST) &&
2093 fr_max > frag->fr_max) {
2094 if (r->rule_flag & PFRULE_FRAGDROP) {
2095 frag->fr_flags |= PFFRAG_DROP;
2096 }
2097 goto bad;
2098 }
2099
2100 if ((m = pbuf_to_mbuf(pbuf, TRUE)) == NULL) {
2101 REASON_SET(reason, PFRES_MEMORY);
2102 goto bad;
2103 }
2104
2105 VERIFY(!pbuf_is_valid(pbuf));
2106
2107 /* Restore iph pointer after pbuf_to_mbuf() */
2108 h = mtod(m, struct ip *);
2109
2110 m = pf_fragcache(m0: &m, h, frag: &frag, mff,
2111 drop: (r->rule_flag & PFRULE_FRAGDROP) ? 1 : 0, nomem: &nomem);
2112 if (m == NULL) {
2113 // Note: pf_fragcache() has already m_freem'd the mbuf
2114 if (nomem) {
2115 goto no_mem;
2116 }
2117 goto drop;
2118 }
2119
2120 VERIFY(m->m_flags & M_PKTHDR);
2121 pbuf_init_mbuf(pbuf, m, ifp);
2122
2123 /* use mtag from copied and trimmed mbuf chain */
2124 pd->pf_mtag = pf_find_mtag_pbuf(pbuf);
2125#if 0
2126// SCW: This check is superfluous
2127#if DIAGNOSTIC
2128 if (pd->pf_mtag == NULL) {
2129 printf("%s: pf_find_mtag returned NULL(2)\n", __func__);
2130 if ((pd->pf_mtag = pf_get_mtag(m)) == NULL) {
2131 m_freem(m);
2132 m = NULL;
2133 goto no_mem;
2134 }
2135 }
2136#endif
2137#endif
2138 if (dir == PF_IN) {
2139 pd->pf_mtag->pftag_flags |= PF_TAG_FRAGCACHE;
2140 }
2141
2142 if (frag != NULL && (frag->fr_flags & PFFRAG_DROP)) {
2143 goto drop;
2144 }
2145
2146 goto fragment_pass;
2147 }
2148
2149no_fragment:
2150 /* At this point, only IP_DF is allowed in ip_off */
2151 if (h->ip_off & ~htons(IP_DF)) {
2152 u_int16_t ipoff = h->ip_off;
2153
2154 h->ip_off &= htons(IP_DF);
2155 h->ip_sum = pf_cksum_fixup(h->ip_sum, ipoff, h->ip_off, 0);
2156 }
2157
2158 /* Enforce a minimum ttl, may cause endless packet loops */
2159 if (r->min_ttl && h->ip_ttl < r->min_ttl) {
2160 u_int16_t ip_ttl = h->ip_ttl;
2161
2162 h->ip_ttl = r->min_ttl;
2163 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_ttl, h->ip_ttl, 0);
2164 }
2165 if (r->rule_flag & PFRULE_RANDOMID) {
2166 u_int16_t oip_id = h->ip_id;
2167
2168 if (rfc6864 && IP_OFF_IS_ATOMIC(ntohs(h->ip_off))) {
2169 h->ip_id = 0;
2170 } else {
2171 h->ip_id = ip_randomid(ipid_salt);
2172 }
2173 h->ip_sum = pf_cksum_fixup(h->ip_sum, oip_id, h->ip_id, 0);
2174 }
2175 if ((r->rule_flag & (PFRULE_FRAGCROP | PFRULE_FRAGDROP)) == 0) {
2176 pd->flags |= PFDESC_IP_REAS;
2177 }
2178
2179 return PF_PASS;
2180
2181fragment_pass:
2182 /* Enforce a minimum ttl, may cause endless packet loops */
2183 if (r->min_ttl && h->ip_ttl < r->min_ttl) {
2184 u_int16_t ip_ttl = h->ip_ttl;
2185
2186 h->ip_ttl = r->min_ttl;
2187 h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_ttl, h->ip_ttl, 0);
2188 }
2189 if ((r->rule_flag & (PFRULE_FRAGCROP | PFRULE_FRAGDROP)) == 0) {
2190 pd->flags |= PFDESC_IP_REAS;
2191 }
2192 return PF_PASS;
2193
2194no_mem:
2195 REASON_SET(reason, PFRES_MEMORY);
2196 if (r != NULL && r->log && pbuf_is_valid(pbuf)) {
2197 PFLOG_PACKET(kif, h, pbuf, AF_INET, dir, *reason, r,
2198 NULL, NULL, pd);
2199 }
2200 return PF_DROP;
2201
2202drop:
2203 REASON_SET(reason, PFRES_NORM);
2204 if (r != NULL && r->log && pbuf_is_valid(pbuf)) {
2205 PFLOG_PACKET(kif, h, pbuf, AF_INET, dir, *reason, r,
2206 NULL, NULL, pd);
2207 }
2208 return PF_DROP;
2209
2210bad:
2211 DPFPRINTF(("dropping bad IPv4 fragment\n"));
2212
2213 /* Free associated fragments */
2214 if (frag != NULL) {
2215 pf_free_fragment(frag);
2216 }
2217
2218 REASON_SET(reason, PFRES_FRAG);
2219 if (r != NULL && r->log && pbuf_is_valid(pbuf)) {
2220 PFLOG_PACKET(kif, h, pbuf, AF_INET, dir, *reason, r, NULL, NULL, pd);
2221 }
2222
2223 return PF_DROP;
2224}
2225
2226static __attribute__((noinline)) struct pf_fragment *
2227pf_find_fragment_by_ipv6_header(struct ip6_hdr *ip6, struct ip6_frag *fh,
2228 struct pf_frag_tree *tree)
2229{
2230 struct pf_fragment key;
2231 pf_ip6hdr2key(key: &key, ip6, fh);
2232 return pf_find_fragment_by_key(key: &key, tree);
2233}
2234
2235int
2236pf_normalize_ip6(pbuf_t *pbuf, int dir, struct pfi_kif *kif,
2237 u_short *reason, struct pf_pdesc *pd)
2238{
2239 struct mbuf *m = NULL;
2240 struct pf_rule *r;
2241 struct ip6_hdr *h = pbuf->pb_data;
2242 int extoff;
2243 int off;
2244 struct ip6_ext ext;
2245 struct ip6_opt opt;
2246 struct ip6_opt_jumbo jumbo;
2247 int optend;
2248 int ooff;
2249 struct ip6_frag frag;
2250 u_int32_t jumbolen = 0, plen;
2251 u_int16_t fragoff = 0;
2252 u_int8_t proto;
2253 int terminal;
2254 struct pf_frent *frent;
2255 struct pf_fragment *pff = NULL;
2256 int mff = 0, rh_cnt = 0;
2257 u_int16_t fr_max;
2258 int asd = 0;
2259 struct pf_ruleset *ruleset = NULL;
2260 struct ifnet *ifp = pbuf->pb_ifp;
2261
2262 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
2263 while (r != NULL) {
2264 r->evaluations++;
2265 if (pfi_kif_match(r->kif, kif) == r->ifnot) {
2266 r = r->skip[PF_SKIP_IFP].ptr;
2267 } else if (r->direction && r->direction != dir) {
2268 r = r->skip[PF_SKIP_DIR].ptr;
2269 } else if (r->af && r->af != AF_INET6) {
2270 r = r->skip[PF_SKIP_AF].ptr;
2271 }
2272#if 0 /* header chain! */
2273 else if (r->proto && r->proto != h->ip6_nxt) {
2274 r = r->skip[PF_SKIP_PROTO].ptr;
2275 }
2276#endif
2277 else if (PF_MISMATCHAW(&r->src.addr,
2278 (struct pf_addr *)(uintptr_t)&h->ip6_src, AF_INET6,
2279 r->src.neg, kif)) {
2280 r = r->skip[PF_SKIP_SRC_ADDR].ptr;
2281 } else if (PF_MISMATCHAW(&r->dst.addr,
2282 (struct pf_addr *)(uintptr_t)&h->ip6_dst, AF_INET6,
2283 r->dst.neg, NULL)) {
2284 r = r->skip[PF_SKIP_DST_ADDR].ptr;
2285 } else {
2286 if (r->anchor == NULL) {
2287 break;
2288 } else {
2289 pf_step_into_anchor(&asd, &ruleset,
2290 PF_RULESET_SCRUB, &r, NULL, NULL);
2291 }
2292 }
2293 if (r == NULL && pf_step_out_of_anchor(&asd, &ruleset,
2294 PF_RULESET_SCRUB, &r, NULL, NULL)) {
2295 break;
2296 }
2297 }
2298
2299 if (r == NULL || r->action == PF_NOSCRUB) {
2300 return PF_PASS;
2301 } else {
2302 r->packets[dir == PF_OUT]++;
2303 r->bytes[dir == PF_OUT] += pd->tot_len;
2304 }
2305
2306 /* Check for illegal packets */
2307 if ((uint32_t)(sizeof(struct ip6_hdr) + IPV6_MAXPACKET) <
2308 pbuf->pb_packet_len) {
2309 goto drop;
2310 }
2311
2312 extoff = 0;
2313 off = sizeof(struct ip6_hdr);
2314 proto = h->ip6_nxt;
2315 terminal = 0;
2316 do {
2317 pd->proto = proto;
2318 switch (proto) {
2319 case IPPROTO_FRAGMENT:
2320 goto fragment;
2321 case IPPROTO_AH:
2322 case IPPROTO_ROUTING:
2323 case IPPROTO_DSTOPTS:
2324 if (!pf_pull_hdr(pbuf, off, &ext, sizeof(ext), NULL,
2325 NULL, AF_INET6)) {
2326 goto shortpkt;
2327 }
2328 extoff = off;
2329 /*
2330 * <jhw@apple.com>
2331 * Multiple routing headers not allowed.
2332 * Routing header type zero considered harmful.
2333 */
2334 if (proto == IPPROTO_ROUTING) {
2335 const struct ip6_rthdr *rh =
2336 (const struct ip6_rthdr *)&ext;
2337 if (rh_cnt++) {
2338 goto drop;
2339 }
2340 if (rh->ip6r_type == IPV6_RTHDR_TYPE_0) {
2341 goto drop;
2342 }
2343 } else if (proto == IPPROTO_AH) {
2344 off += (ext.ip6e_len + 2) * 4;
2345 } else {
2346 off += (ext.ip6e_len + 1) * 8;
2347 }
2348 proto = ext.ip6e_nxt;
2349 break;
2350 case IPPROTO_HOPOPTS:
2351 if (!pf_pull_hdr(pbuf, off, &ext, sizeof(ext), NULL,
2352 NULL, AF_INET6)) {
2353 goto shortpkt;
2354 }
2355 extoff = off;
2356 optend = off + (ext.ip6e_len + 1) * 8;
2357 ooff = off + sizeof(ext);
2358 do {
2359 if (!pf_pull_hdr(pbuf, ooff, &opt.ip6o_type,
2360 sizeof(opt.ip6o_type), NULL, NULL,
2361 AF_INET6)) {
2362 goto shortpkt;
2363 }
2364 if (opt.ip6o_type == IP6OPT_PAD1) {
2365 ooff++;
2366 continue;
2367 }
2368 if (!pf_pull_hdr(pbuf, ooff, &opt, sizeof(opt),
2369 NULL, NULL, AF_INET6)) {
2370 goto shortpkt;
2371 }
2372 if ((ooff + (int) sizeof(opt) + opt.ip6o_len) >
2373 optend) {
2374 goto drop;
2375 }
2376 switch (opt.ip6o_type) {
2377 case IP6OPT_JUMBO:
2378 if (h->ip6_plen != 0) {
2379 goto drop;
2380 }
2381 if (!pf_pull_hdr(pbuf, ooff, &jumbo,
2382 sizeof(jumbo), NULL, NULL,
2383 AF_INET6)) {
2384 goto shortpkt;
2385 }
2386 memcpy(dst: &jumbolen, src: jumbo.ip6oj_jumbo_len,
2387 n: sizeof(jumbolen));
2388 jumbolen = ntohl(jumbolen);
2389 if (jumbolen <= IPV6_MAXPACKET) {
2390 goto drop;
2391 }
2392 if ((sizeof(struct ip6_hdr) +
2393 jumbolen) != pbuf->pb_packet_len) {
2394 goto drop;
2395 }
2396 break;
2397 default:
2398 break;
2399 }
2400 ooff += sizeof(opt) + opt.ip6o_len;
2401 } while (ooff < optend);
2402
2403 off = optend;
2404 proto = ext.ip6e_nxt;
2405 break;
2406 default:
2407 terminal = 1;
2408 break;
2409 }
2410 } while (!terminal);
2411
2412 /* jumbo payload option must be present, or plen > 0 */
2413 if (ntohs(h->ip6_plen) == 0) {
2414 plen = jumbolen;
2415 } else {
2416 plen = ntohs(h->ip6_plen);
2417 }
2418 if (plen == 0) {
2419 goto drop;
2420 }
2421 if ((uint32_t)(sizeof(struct ip6_hdr) + plen) > pbuf->pb_packet_len) {
2422 goto shortpkt;
2423 }
2424
2425 /* Enforce a minimum ttl, may cause endless packet loops */
2426 if (r->min_ttl && h->ip6_hlim < r->min_ttl) {
2427 h->ip6_hlim = r->min_ttl;
2428 }
2429
2430 return PF_PASS;
2431
2432fragment:
2433 plen = ntohs(h->ip6_plen);
2434 /* Jumbo payload packets cannot be fragmented */
2435 if (plen == 0 || jumbolen) {
2436 goto drop;
2437 }
2438
2439 if (!pf_pull_hdr(pbuf, off, &frag, sizeof(frag), NULL, NULL, AF_INET6)) {
2440 goto shortpkt;
2441 }
2442 fragoff = ntohs(frag.ip6f_offlg & IP6F_OFF_MASK);
2443 pd->proto = frag.ip6f_nxt;
2444 mff = ntohs(frag.ip6f_offlg & IP6F_MORE_FRAG);
2445 off += sizeof(frag);
2446 if (fragoff + (plen - off) > IPV6_MAXPACKET) {
2447 goto badfrag;
2448 }
2449
2450 fr_max = fragoff + plen - (off - sizeof(struct ip6_hdr));
2451// XXX SCW: mbuf-specific
2452// DPFPRINTF(("0x%llx IPv6 frag plen %u mff %d off %u fragoff %u "
2453// "fr_max %u\n", (uint64_t)VM_KERNEL_ADDRPERM(m), plen, mff, off,
2454// fragoff, fr_max));
2455
2456 if ((r->rule_flag & (PFRULE_FRAGCROP | PFRULE_FRAGDROP)) == 0) {
2457 /* Fully buffer all of the fragments */
2458 pd->flags |= PFDESC_IP_REAS;
2459
2460 pff = pf_find_fragment_by_ipv6_header(ip6: h, fh: &frag,
2461 tree: &pf_frag_tree);
2462
2463 /* Check if we saw the last fragment already */
2464 if (pff != NULL && (pff->fr_flags & PFFRAG_SEENLAST) &&
2465 fr_max > pff->fr_max) {
2466 goto badfrag;
2467 }
2468
2469 if ((m = pbuf_to_mbuf(pbuf, TRUE)) == NULL) {
2470 REASON_SET(reason, PFRES_MEMORY);
2471 return PF_DROP;
2472 }
2473
2474 /* Restore iph pointer after pbuf_to_mbuf() */
2475 h = mtod(m, struct ip6_hdr *);
2476
2477 /* Get an entry for the fragment queue */
2478 frent = pool_get(&pf_frent_pl, PR_NOWAIT);
2479 if (frent == NULL) {
2480 REASON_SET(reason, PFRES_MEMORY);
2481 return PF_DROP;
2482 }
2483
2484 pf_nfrents++;
2485 frent->fr_ip6 = h;
2486 frent->fr_m = m;
2487 frent->fr_ip6f_opt = frag;
2488 frent->fr_ip6f_extoff = extoff;
2489 frent->fr_ip6f_hlen = off;
2490 /* account for 2nd Destination Options header if present */
2491 if (pd->proto == IPPROTO_DSTOPTS) {
2492 if (!pf_pull_hdr(pbuf, off, &ext, sizeof(ext), NULL,
2493 NULL, AF_INET6)) {
2494 goto shortpkt;
2495 }
2496 frent->fr_ip6f_hlen += (ext.ip6e_len + 1) * 8;
2497 }
2498
2499 /* Might return a completely reassembled mbuf, or NULL */
2500 DPFPRINTF(("reass IPv6 frag %d @ %d-%d\n",
2501 ntohl(frag.ip6f_ident), fragoff, fr_max));
2502 m = pf_reassemble6(m0: &m, frag: &pff, frent, mff);
2503
2504 if (m == NULL) {
2505 return PF_DROP;
2506 }
2507
2508 pbuf_init_mbuf(pbuf, m, ifp);
2509 h = pbuf->pb_data;
2510
2511 if (pff != NULL && (pff->fr_flags & PFFRAG_DROP)) {
2512 goto drop;
2513 }
2514 } else if (dir == PF_IN ||
2515 !(pd->pf_mtag->pftag_flags & PF_TAG_FRAGCACHE)) {
2516 /* non-buffering fragment cache (overlaps: see RFC 5722) */
2517 int nomem = 0;
2518
2519 pff = pf_find_fragment_by_ipv6_header(ip6: h, fh: &frag,
2520 tree: &pf_cache_tree);
2521
2522 /* Check if we saw the last fragment already */
2523 if (pff != NULL && (pff->fr_flags & PFFRAG_SEENLAST) &&
2524 fr_max > pff->fr_max) {
2525 if (r->rule_flag & PFRULE_FRAGDROP) {
2526 pff->fr_flags |= PFFRAG_DROP;
2527 }
2528 goto badfrag;
2529 }
2530
2531 if ((m = pbuf_to_mbuf(pbuf, TRUE)) == NULL) {
2532 goto no_mem;
2533 }
2534
2535 /* Restore iph pointer after pbuf_to_mbuf() */
2536 h = mtod(m, struct ip6_hdr *);
2537
2538 m = pf_frag6cache(m0: &m, h, fh: &frag, frag: &pff, hlen: off, mff,
2539 drop: (r->rule_flag & PFRULE_FRAGDROP) ? 1 : 0, nomem: &nomem);
2540 if (m == NULL) {
2541 // Note: pf_frag6cache() has already m_freem'd the mbuf
2542 if (nomem) {
2543 goto no_mem;
2544 }
2545 goto drop;
2546 }
2547
2548 pbuf_init_mbuf(pbuf, m, ifp);
2549 pd->pf_mtag = pf_find_mtag_pbuf(pbuf);
2550 h = pbuf->pb_data;
2551
2552 if (dir == PF_IN) {
2553 pd->pf_mtag->pftag_flags |= PF_TAG_FRAGCACHE;
2554 }
2555
2556 if (pff != NULL && (pff->fr_flags & PFFRAG_DROP)) {
2557 goto drop;
2558 }
2559 }
2560
2561 /* Enforce a minimum ttl, may cause endless packet loops */
2562 if (r->min_ttl && h->ip6_hlim < r->min_ttl) {
2563 h->ip6_hlim = r->min_ttl;
2564 }
2565 return PF_PASS;
2566
2567no_mem:
2568 REASON_SET(reason, PFRES_MEMORY);
2569 goto dropout;
2570
2571shortpkt:
2572 REASON_SET(reason, PFRES_SHORT);
2573 goto dropout;
2574
2575drop:
2576 REASON_SET(reason, PFRES_NORM);
2577 goto dropout;
2578
2579badfrag:
2580 DPFPRINTF(("dropping bad IPv6 fragment\n"));
2581 REASON_SET(reason, PFRES_FRAG);
2582 goto dropout;
2583
2584dropout:
2585 if (pff != NULL) {
2586 pf_free_fragment(frag: pff);
2587 }
2588 if (r != NULL && r->log && pbuf_is_valid(pbuf)) {
2589 PFLOG_PACKET(kif, h, pbuf, AF_INET6, dir, *reason, r, NULL, NULL, pd);
2590 }
2591 return PF_DROP;
2592}
2593
2594int
2595pf_normalize_tcp(int dir, struct pfi_kif *kif, pbuf_t *pbuf, int ipoff,
2596 int off, void *h, struct pf_pdesc *pd)
2597{
2598#pragma unused(ipoff, h)
2599 struct pf_rule *r, *rm = NULL;
2600 struct tcphdr *th = pd->hdr.tcp;
2601 int rewrite = 0;
2602 int asd = 0;
2603 u_short reason;
2604 u_int8_t flags;
2605 sa_family_t af = pd->af;
2606 struct pf_ruleset *ruleset = NULL;
2607 union pf_state_xport sxport, dxport;
2608
2609 sxport.port = th->th_sport;
2610 dxport.port = th->th_dport;
2611
2612 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
2613 while (r != NULL) {
2614 r->evaluations++;
2615 if (pfi_kif_match(r->kif, kif) == r->ifnot) {
2616 r = r->skip[PF_SKIP_IFP].ptr;
2617 } else if (r->direction && r->direction != dir) {
2618 r = r->skip[PF_SKIP_DIR].ptr;
2619 } else if (r->af && r->af != af) {
2620 r = r->skip[PF_SKIP_AF].ptr;
2621 } else if (r->proto && r->proto != pd->proto) {
2622 r = r->skip[PF_SKIP_PROTO].ptr;
2623 } else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
2624 r->src.neg, kif)) {
2625 r = r->skip[PF_SKIP_SRC_ADDR].ptr;
2626 } else if (r->src.xport.range.op &&
2627 !pf_match_xport(r->src.xport.range.op, r->proto_variant,
2628 &r->src.xport, &sxport)) {
2629 r = r->skip[PF_SKIP_SRC_PORT].ptr;
2630 } else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
2631 r->dst.neg, NULL)) {
2632 r = r->skip[PF_SKIP_DST_ADDR].ptr;
2633 } else if (r->dst.xport.range.op &&
2634 !pf_match_xport(r->dst.xport.range.op, r->proto_variant,
2635 &r->dst.xport, &dxport)) {
2636 r = r->skip[PF_SKIP_DST_PORT].ptr;
2637 } else if (r->os_fingerprint != PF_OSFP_ANY &&
2638 !pf_osfp_match(pf_osfp_fingerprint(pd, pbuf, off, th),
2639 r->os_fingerprint)) {
2640 r = TAILQ_NEXT(r, entries);
2641 } else {
2642 if (r->anchor == NULL) {
2643 rm = r;
2644 break;
2645 } else {
2646 pf_step_into_anchor(&asd, &ruleset,
2647 PF_RULESET_SCRUB, &r, NULL, NULL);
2648 }
2649 }
2650 if (r == NULL && pf_step_out_of_anchor(&asd, &ruleset,
2651 PF_RULESET_SCRUB, &r, NULL, NULL)) {
2652 break;
2653 }
2654 }
2655
2656 if (rm == NULL || rm->action == PF_NOSCRUB) {
2657 return PF_PASS;
2658 } else {
2659 r->packets[dir == PF_OUT]++;
2660 r->bytes[dir == PF_OUT] += pd->tot_len;
2661 }
2662
2663 if (rm->rule_flag & PFRULE_REASSEMBLE_TCP) {
2664 pd->flags |= PFDESC_TCP_NORM;
2665 }
2666
2667 flags = th->th_flags;
2668 if (flags & TH_SYN) {
2669 /* Illegal packet */
2670 if (flags & TH_RST) {
2671 goto tcp_drop;
2672 }
2673
2674 if (flags & TH_FIN) {
2675 flags &= ~TH_FIN;
2676 }
2677 } else {
2678 /* Illegal packet */
2679 if (!(flags & (TH_ACK | TH_RST))) {
2680 goto tcp_drop;
2681 }
2682 }
2683
2684 if (!(flags & TH_ACK)) {
2685 /* These flags are only valid if ACK is set */
2686 if ((flags & TH_FIN) || (flags & TH_PUSH) || (flags & TH_URG)) {
2687 goto tcp_drop;
2688 }
2689 }
2690
2691 /* Check for illegal header length */
2692 if (th->th_off < (sizeof(struct tcphdr) >> 2)) {
2693 goto tcp_drop;
2694 }
2695
2696 /* If flags changed, or reserved data set, then adjust */
2697 if (flags != th->th_flags || th->th_x2 != 0) {
2698 u_int16_t ov, nv;
2699
2700 ov = *(u_int16_t *)(&th->th_ack + 1);
2701 th->th_flags = flags;
2702 th->th_x2 = 0;
2703 nv = *(u_int16_t *)(&th->th_ack + 1);
2704
2705 th->th_sum = pf_cksum_fixup(th->th_sum, ov, nv, 0);
2706 rewrite = 1;
2707 }
2708
2709 /* Remove urgent pointer, if TH_URG is not set */
2710 if (!(flags & TH_URG) && th->th_urp) {
2711 th->th_sum = pf_cksum_fixup(th->th_sum, th->th_urp, 0, 0);
2712 th->th_urp = 0;
2713 rewrite = 1;
2714 }
2715
2716 /* copy back packet headers if we sanitized */
2717 /* Process options */
2718 if (r->max_mss) {
2719 int rv = pf_normalize_tcpopt(r, dir, kif, pd, pbuf, th, off,
2720 &rewrite);
2721 if (rv == PF_DROP) {
2722 return rv;
2723 }
2724 pbuf = pd->mp;
2725 }
2726
2727 if (rewrite) {
2728 if (pf_lazy_makewritable(pd, pbuf,
2729 off + sizeof(*th)) == NULL) {
2730 REASON_SET(&reason, PFRES_MEMORY);
2731 if (r->log) {
2732 PFLOG_PACKET(kif, h, pbuf, AF_INET, dir, reason,
2733 r, 0, 0, pd);
2734 }
2735 return PF_DROP;
2736 }
2737
2738 pbuf_copy_back(pbuf, off, sizeof(*th), th);
2739 }
2740
2741 return PF_PASS;
2742
2743tcp_drop:
2744 REASON_SET(&reason, PFRES_NORM);
2745 if (rm != NULL && r->log) {
2746 PFLOG_PACKET(kif, h, pbuf, AF_INET, dir, reason, r, NULL, NULL, pd);
2747 }
2748 return PF_DROP;
2749}
2750
2751int
2752pf_normalize_tcp_init(pbuf_t *pbuf, int off, struct pf_pdesc *pd,
2753 struct tcphdr *th, struct pf_state_peer *src, struct pf_state_peer *dst)
2754{
2755#pragma unused(dst)
2756 u_int32_t tsval, tsecr;
2757 u_int8_t hdr[60];
2758 u_int8_t *opt;
2759
2760 VERIFY(src->scrub == NULL);
2761
2762 src->scrub = pool_get(&pf_state_scrub_pl, PR_NOWAIT);
2763 if (src->scrub == NULL) {
2764 return 1;
2765 }
2766 bzero(s: src->scrub, n: sizeof(*src->scrub));
2767
2768 switch (pd->af) {
2769#if INET
2770 case AF_INET: {
2771 struct ip *h = pbuf->pb_data;
2772 src->scrub->pfss_ttl = h->ip_ttl;
2773 break;
2774 }
2775#endif /* INET */
2776 case AF_INET6: {
2777 struct ip6_hdr *h = pbuf->pb_data;
2778 src->scrub->pfss_ttl = h->ip6_hlim;
2779 break;
2780 }
2781 }
2782
2783
2784 /*
2785 * All normalizations below are only begun if we see the start of
2786 * the connections. They must all set an enabled bit in pfss_flags
2787 */
2788 if ((th->th_flags & TH_SYN) == 0) {
2789 return 0;
2790 }
2791
2792
2793 if (th->th_off > (sizeof(struct tcphdr) >> 2) && src->scrub &&
2794 pf_pull_hdr(pbuf, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) {
2795 /* Diddle with TCP options */
2796 int hlen;
2797 opt = hdr + sizeof(struct tcphdr);
2798 hlen = (th->th_off << 2) - sizeof(struct tcphdr);
2799 while (hlen >= TCPOLEN_TIMESTAMP) {
2800 switch (*opt) {
2801 case TCPOPT_EOL: /* FALLTHROUGH */
2802 case TCPOPT_NOP:
2803 opt++;
2804 hlen--;
2805 break;
2806 case TCPOPT_TIMESTAMP:
2807 if (opt[1] >= TCPOLEN_TIMESTAMP) {
2808 src->scrub->pfss_flags |=
2809 PFSS_TIMESTAMP;
2810 src->scrub->pfss_ts_mod =
2811 htonl(random());
2812
2813 /* note PFSS_PAWS not set yet */
2814 memcpy(dst: &tsval, src: &opt[2],
2815 n: sizeof(u_int32_t));
2816 memcpy(dst: &tsecr, src: &opt[6],
2817 n: sizeof(u_int32_t));
2818 src->scrub->pfss_tsval0 = ntohl(tsval);
2819 src->scrub->pfss_tsval = ntohl(tsval);
2820 src->scrub->pfss_tsecr = ntohl(tsecr);
2821 getmicrouptime(&src->scrub->pfss_last);
2822 }
2823 OS_FALLTHROUGH;
2824 default:
2825 hlen -= MAX(opt[1], 2);
2826 opt += MAX(opt[1], 2);
2827 break;
2828 }
2829 }
2830 }
2831
2832 return 0;
2833}
2834
2835void
2836pf_normalize_tcp_cleanup(struct pf_state *state)
2837{
2838 if (state->src.scrub) {
2839 pool_put(&pf_state_scrub_pl, state->src.scrub);
2840 }
2841 if (state->dst.scrub) {
2842 pool_put(&pf_state_scrub_pl, state->dst.scrub);
2843 }
2844
2845 /* Someday... flush the TCP segment reassembly descriptors. */
2846}
2847
2848int
2849pf_normalize_tcp_stateful(pbuf_t *pbuf, int off, struct pf_pdesc *pd,
2850 u_short *reason, struct tcphdr *th, struct pf_state *state,
2851 struct pf_state_peer *src, struct pf_state_peer *dst, int *writeback)
2852{
2853 struct timeval uptime;
2854 u_int32_t tsval = 0, tsecr = 0;
2855 u_int tsval_from_last;
2856 u_int8_t hdr[60];
2857 u_int8_t *opt;
2858 int copyback = 0;
2859 int got_ts = 0;
2860
2861 VERIFY(src->scrub || dst->scrub);
2862
2863 /*
2864 * Enforce the minimum TTL seen for this connection. Negate a common
2865 * technique to evade an intrusion detection system and confuse
2866 * firewall state code.
2867 */
2868 switch (pd->af) {
2869#if INET
2870 case AF_INET: {
2871 if (src->scrub) {
2872 struct ip *h = pbuf->pb_data;
2873 if (h->ip_ttl > src->scrub->pfss_ttl) {
2874 src->scrub->pfss_ttl = h->ip_ttl;
2875 }
2876 h->ip_ttl = src->scrub->pfss_ttl;
2877 }
2878 break;
2879 }
2880#endif /* INET */
2881 case AF_INET6: {
2882 if (src->scrub) {
2883 struct ip6_hdr *h = pbuf->pb_data;
2884 if (h->ip6_hlim > src->scrub->pfss_ttl) {
2885 src->scrub->pfss_ttl = h->ip6_hlim;
2886 }
2887 h->ip6_hlim = src->scrub->pfss_ttl;
2888 }
2889 break;
2890 }
2891 }
2892
2893 if (th->th_off > (sizeof(struct tcphdr) >> 2) &&
2894 ((src->scrub && (src->scrub->pfss_flags & PFSS_TIMESTAMP)) ||
2895 (dst->scrub && (dst->scrub->pfss_flags & PFSS_TIMESTAMP))) &&
2896 pf_pull_hdr(pbuf, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) {
2897 /* Diddle with TCP options */
2898 int hlen;
2899 opt = hdr + sizeof(struct tcphdr);
2900 hlen = (th->th_off << 2) - sizeof(struct tcphdr);
2901 while (hlen >= TCPOLEN_TIMESTAMP) {
2902 switch (*opt) {
2903 case TCPOPT_EOL: /* FALLTHROUGH */
2904 case TCPOPT_NOP:
2905 opt++;
2906 hlen--;
2907 break;
2908 case TCPOPT_TIMESTAMP:
2909 /*
2910 * Modulate the timestamps. Can be used for
2911 * NAT detection, OS uptime determination or
2912 * reboot detection.
2913 */
2914
2915 if (got_ts) {
2916 /* Huh? Multiple timestamps!? */
2917 if (pf_status.debug >= PF_DEBUG_MISC) {
2918 DPFPRINTF(("multiple TS??"));
2919 pf_print_state(state);
2920 printf("\n");
2921 }
2922 REASON_SET(reason, PFRES_TS);
2923 return PF_DROP;
2924 }
2925 if (opt[1] >= TCPOLEN_TIMESTAMP) {
2926 memcpy(dst: &tsval, src: &opt[2],
2927 n: sizeof(u_int32_t));
2928 if (tsval && src->scrub &&
2929 (src->scrub->pfss_flags &
2930 PFSS_TIMESTAMP)) {
2931 tsval = ntohl(tsval);
2932 pf_change_a(&opt[2],
2933 &th->th_sum,
2934 htonl(tsval +
2935 src->scrub->pfss_ts_mod),
2936 0);
2937 copyback = 1;
2938 }
2939
2940 /* Modulate TS reply iff valid (!0) */
2941 memcpy(dst: &tsecr, src: &opt[6],
2942 n: sizeof(u_int32_t));
2943 if (tsecr && dst->scrub &&
2944 (dst->scrub->pfss_flags &
2945 PFSS_TIMESTAMP)) {
2946 tsecr = ntohl(tsecr)
2947 - dst->scrub->pfss_ts_mod;
2948 pf_change_a(&opt[6],
2949 &th->th_sum, htonl(tsecr),
2950 0);
2951 copyback = 1;
2952 }
2953 got_ts = 1;
2954 }
2955 OS_FALLTHROUGH;
2956 default:
2957 hlen -= MAX(opt[1], 2);
2958 opt += MAX(opt[1], 2);
2959 break;
2960 }
2961 }
2962 if (copyback) {
2963 /* Copyback the options, caller copys back header */
2964 int optoff = off + sizeof(*th);
2965 int optlen = (th->th_off << 2) - sizeof(*th);
2966 if (pf_lazy_makewritable(pd, pbuf, optoff + optlen) ==
2967 NULL) {
2968 REASON_SET(reason, PFRES_MEMORY);
2969 return PF_DROP;
2970 }
2971 *writeback = optoff + optlen;
2972 pbuf_copy_back(pbuf, optoff, optlen, hdr + sizeof(*th));
2973 }
2974 }
2975
2976
2977 /*
2978 * Must invalidate PAWS checks on connections idle for too long.
2979 * The fastest allowed timestamp clock is 1ms. That turns out to
2980 * be about 24 days before it wraps. XXX Right now our lowerbound
2981 * TS echo check only works for the first 12 days of a connection
2982 * when the TS has exhausted half its 32bit space
2983 */
2984#define TS_MAX_IDLE (24*24*60*60)
2985#define TS_MAX_CONN (12*24*60*60) /* XXX remove when better tsecr check */
2986
2987 getmicrouptime(&uptime);
2988 if (src->scrub && (src->scrub->pfss_flags & PFSS_PAWS) &&
2989 (uptime.tv_sec - src->scrub->pfss_last.tv_sec > TS_MAX_IDLE ||
2990 pf_time_second() - state->creation > TS_MAX_CONN)) {
2991 if (pf_status.debug >= PF_DEBUG_MISC) {
2992 DPFPRINTF(("src idled out of PAWS\n"));
2993 pf_print_state(state);
2994 printf("\n");
2995 }
2996 src->scrub->pfss_flags = (src->scrub->pfss_flags & ~PFSS_PAWS)
2997 | PFSS_PAWS_IDLED;
2998 }
2999 if (dst->scrub && (dst->scrub->pfss_flags & PFSS_PAWS) &&
3000 uptime.tv_sec - dst->scrub->pfss_last.tv_sec > TS_MAX_IDLE) {
3001 if (pf_status.debug >= PF_DEBUG_MISC) {
3002 DPFPRINTF(("dst idled out of PAWS\n"));
3003 pf_print_state(state);
3004 printf("\n");
3005 }
3006 dst->scrub->pfss_flags = (dst->scrub->pfss_flags & ~PFSS_PAWS)
3007 | PFSS_PAWS_IDLED;
3008 }
3009
3010 if (got_ts && src->scrub && dst->scrub &&
3011 (src->scrub->pfss_flags & PFSS_PAWS) &&
3012 (dst->scrub->pfss_flags & PFSS_PAWS)) {
3013 /*
3014 * Validate that the timestamps are "in-window".
3015 * RFC1323 describes TCP Timestamp options that allow
3016 * measurement of RTT (round trip time) and PAWS
3017 * (protection against wrapped sequence numbers). PAWS
3018 * gives us a set of rules for rejecting packets on
3019 * long fat pipes (packets that were somehow delayed
3020 * in transit longer than the time it took to send the
3021 * full TCP sequence space of 4Gb). We can use these
3022 * rules and infer a few others that will let us treat
3023 * the 32bit timestamp and the 32bit echoed timestamp
3024 * as sequence numbers to prevent a blind attacker from
3025 * inserting packets into a connection.
3026 *
3027 * RFC1323 tells us:
3028 * - The timestamp on this packet must be greater than
3029 * or equal to the last value echoed by the other
3030 * endpoint. The RFC says those will be discarded
3031 * since it is a dup that has already been acked.
3032 * This gives us a lowerbound on the timestamp.
3033 * timestamp >= other last echoed timestamp
3034 * - The timestamp will be less than or equal to
3035 * the last timestamp plus the time between the
3036 * last packet and now. The RFC defines the max
3037 * clock rate as 1ms. We will allow clocks to be
3038 * up to 10% fast and will allow a total difference
3039 * or 30 seconds due to a route change. And this
3040 * gives us an upperbound on the timestamp.
3041 * timestamp <= last timestamp + max ticks
3042 * We have to be careful here. Windows will send an
3043 * initial timestamp of zero and then initialize it
3044 * to a random value after the 3whs; presumably to
3045 * avoid a DoS by having to call an expensive RNG
3046 * during a SYN flood. Proof MS has at least one
3047 * good security geek.
3048 *
3049 * - The TCP timestamp option must also echo the other
3050 * endpoints timestamp. The timestamp echoed is the
3051 * one carried on the earliest unacknowledged segment
3052 * on the left edge of the sequence window. The RFC
3053 * states that the host will reject any echoed
3054 * timestamps that were larger than any ever sent.
3055 * This gives us an upperbound on the TS echo.
3056 * tescr <= largest_tsval
3057 * - The lowerbound on the TS echo is a little more
3058 * tricky to determine. The other endpoint's echoed
3059 * values will not decrease. But there may be
3060 * network conditions that re-order packets and
3061 * cause our view of them to decrease. For now the
3062 * only lowerbound we can safely determine is that
3063 * the TS echo will never be less than the original
3064 * TS. XXX There is probably a better lowerbound.
3065 * Remove TS_MAX_CONN with better lowerbound check.
3066 * tescr >= other original TS
3067 *
3068 * It is also important to note that the fastest
3069 * timestamp clock of 1ms will wrap its 32bit space in
3070 * 24 days. So we just disable TS checking after 24
3071 * days of idle time. We actually must use a 12d
3072 * connection limit until we can come up with a better
3073 * lowerbound to the TS echo check.
3074 */
3075 struct timeval delta_ts;
3076 int ts_fudge;
3077
3078
3079 /*
3080 * PFTM_TS_DIFF is how many seconds of leeway to allow
3081 * a host's timestamp. This can happen if the previous
3082 * packet got delayed in transit for much longer than
3083 * this packet.
3084 */
3085 if ((ts_fudge = state->rule.ptr->timeout[PFTM_TS_DIFF]) == 0) {
3086 ts_fudge = pf_default_rule.timeout[PFTM_TS_DIFF];
3087 }
3088
3089
3090 /* Calculate max ticks since the last timestamp */
3091#define TS_MAXFREQ 1100 /* RFC max TS freq of 1Khz + 10% skew */
3092#define TS_MICROSECS 1000000 /* microseconds per second */
3093 timersub(&uptime, &src->scrub->pfss_last, &delta_ts);
3094 tsval_from_last = (delta_ts.tv_sec + ts_fudge) * TS_MAXFREQ;
3095 tsval_from_last += delta_ts.tv_usec / (TS_MICROSECS / TS_MAXFREQ);
3096
3097
3098 if ((src->state >= TCPS_ESTABLISHED &&
3099 dst->state >= TCPS_ESTABLISHED) &&
3100 (SEQ_LT(tsval, dst->scrub->pfss_tsecr) ||
3101 SEQ_GT(tsval, src->scrub->pfss_tsval + tsval_from_last) ||
3102 (tsecr && (SEQ_GT(tsecr, dst->scrub->pfss_tsval) ||
3103 SEQ_LT(tsecr, dst->scrub->pfss_tsval0))))) {
3104 /*
3105 * Bad RFC1323 implementation or an insertion attack.
3106 *
3107 * - Solaris 2.6 and 2.7 are known to send another ACK
3108 * after the FIN,FIN|ACK,ACK closing that carries
3109 * an old timestamp.
3110 */
3111
3112 DPFPRINTF(("Timestamp failed %c%c%c%c\n",
3113 SEQ_LT(tsval, dst->scrub->pfss_tsecr) ? '0' : ' ',
3114 SEQ_GT(tsval, src->scrub->pfss_tsval +
3115 tsval_from_last) ? '1' : ' ',
3116 SEQ_GT(tsecr, dst->scrub->pfss_tsval) ? '2' : ' ',
3117 SEQ_LT(tsecr, dst->scrub->pfss_tsval0)? '3' : ' '));
3118 DPFPRINTF((" tsval: %u tsecr: %u +ticks: %u "
3119 "idle: %lus %ums\n",
3120 tsval, tsecr, tsval_from_last, delta_ts.tv_sec,
3121 delta_ts.tv_usec / 1000));
3122 DPFPRINTF((" src->tsval: %u tsecr: %u\n",
3123 src->scrub->pfss_tsval, src->scrub->pfss_tsecr));
3124 DPFPRINTF((" dst->tsval: %u tsecr: %u tsval0: %u\n",
3125 dst->scrub->pfss_tsval, dst->scrub->pfss_tsecr,
3126 dst->scrub->pfss_tsval0));
3127 if (pf_status.debug >= PF_DEBUG_MISC) {
3128 pf_print_state(state);
3129 pf_print_flags(th->th_flags);
3130 printf("\n");
3131 }
3132 REASON_SET(reason, PFRES_TS);
3133 return PF_DROP;
3134 }
3135
3136 /* XXX I'd really like to require tsecr but it's optional */
3137 } else if (!got_ts && (th->th_flags & TH_RST) == 0 &&
3138 ((src->state == TCPS_ESTABLISHED && dst->state == TCPS_ESTABLISHED)
3139 || pd->p_len > 0 || (th->th_flags & TH_SYN)) &&
3140 src->scrub && dst->scrub &&
3141 (src->scrub->pfss_flags & PFSS_PAWS) &&
3142 (dst->scrub->pfss_flags & PFSS_PAWS)) {
3143 /*
3144 * Didn't send a timestamp. Timestamps aren't really useful
3145 * when:
3146 * - connection opening or closing (often not even sent).
3147 * but we must not let an attacker to put a FIN on a
3148 * data packet to sneak it through our ESTABLISHED check.
3149 * - on a TCP reset. RFC suggests not even looking at TS.
3150 * - on an empty ACK. The TS will not be echoed so it will
3151 * probably not help keep the RTT calculation in sync and
3152 * there isn't as much danger when the sequence numbers
3153 * got wrapped. So some stacks don't include TS on empty
3154 * ACKs :-(
3155 *
3156 * To minimize the disruption to mostly RFC1323 conformant
3157 * stacks, we will only require timestamps on data packets.
3158 *
3159 * And what do ya know, we cannot require timestamps on data
3160 * packets. There appear to be devices that do legitimate
3161 * TCP connection hijacking. There are HTTP devices that allow
3162 * a 3whs (with timestamps) and then buffer the HTTP request.
3163 * If the intermediate device has the HTTP response cache, it
3164 * will spoof the response but not bother timestamping its
3165 * packets. So we can look for the presence of a timestamp in
3166 * the first data packet and if there, require it in all future
3167 * packets.
3168 */
3169
3170 if (pd->p_len > 0 && (src->scrub->pfss_flags & PFSS_DATA_TS)) {
3171 /*
3172 * Hey! Someone tried to sneak a packet in. Or the
3173 * stack changed its RFC1323 behavior?!?!
3174 */
3175 if (pf_status.debug >= PF_DEBUG_MISC) {
3176 DPFPRINTF(("Did not receive expected RFC1323 "
3177 "timestamp\n"));
3178 pf_print_state(state);
3179 pf_print_flags(th->th_flags);
3180 printf("\n");
3181 }
3182 REASON_SET(reason, PFRES_TS);
3183 return PF_DROP;
3184 }
3185 }
3186
3187
3188 /*
3189 * We will note if a host sends his data packets with or without
3190 * timestamps. And require all data packets to contain a timestamp
3191 * if the first does. PAWS implicitly requires that all data packets be
3192 * timestamped. But I think there are middle-man devices that hijack
3193 * TCP streams immediately after the 3whs and don't timestamp their
3194 * packets (seen in a WWW accelerator or cache).
3195 */
3196 if (pd->p_len > 0 && src->scrub && (src->scrub->pfss_flags &
3197 (PFSS_TIMESTAMP | PFSS_DATA_TS | PFSS_DATA_NOTS)) == PFSS_TIMESTAMP) {
3198 if (got_ts) {
3199 src->scrub->pfss_flags |= PFSS_DATA_TS;
3200 } else {
3201 src->scrub->pfss_flags |= PFSS_DATA_NOTS;
3202 if (pf_status.debug >= PF_DEBUG_MISC && dst->scrub &&
3203 (dst->scrub->pfss_flags & PFSS_TIMESTAMP)) {
3204 /* Don't warn if other host rejected RFC1323 */
3205 DPFPRINTF(("Broken RFC1323 stack did not "
3206 "timestamp data packet. Disabled PAWS "
3207 "security.\n"));
3208 pf_print_state(state);
3209 pf_print_flags(th->th_flags);
3210 printf("\n");
3211 }
3212 }
3213 }
3214
3215
3216 /*
3217 * Update PAWS values
3218 */
3219 if (got_ts && src->scrub && PFSS_TIMESTAMP == (src->scrub->pfss_flags &
3220 (PFSS_PAWS_IDLED | PFSS_TIMESTAMP))) {
3221 getmicrouptime(&src->scrub->pfss_last);
3222 if (SEQ_GEQ(tsval, src->scrub->pfss_tsval) ||
3223 (src->scrub->pfss_flags & PFSS_PAWS) == 0) {
3224 src->scrub->pfss_tsval = tsval;
3225 }
3226
3227 if (tsecr) {
3228 if (SEQ_GEQ(tsecr, src->scrub->pfss_tsecr) ||
3229 (src->scrub->pfss_flags & PFSS_PAWS) == 0) {
3230 src->scrub->pfss_tsecr = tsecr;
3231 }
3232
3233 if ((src->scrub->pfss_flags & PFSS_PAWS) == 0 &&
3234 (SEQ_LT(tsval, src->scrub->pfss_tsval0) ||
3235 src->scrub->pfss_tsval0 == 0)) {
3236 /* tsval0 MUST be the lowest timestamp */
3237 src->scrub->pfss_tsval0 = tsval;
3238 }
3239
3240 /* Only fully initialized after a TS gets echoed */
3241 if ((src->scrub->pfss_flags & PFSS_PAWS) == 0) {
3242 src->scrub->pfss_flags |= PFSS_PAWS;
3243 }
3244 }
3245 }
3246
3247 /* I have a dream.... TCP segment reassembly.... */
3248 return 0;
3249}
3250
3251static __attribute__((noinline)) int
3252pf_normalize_tcpopt(struct pf_rule *r, int dir, struct pfi_kif *kif,
3253 struct pf_pdesc *pd, pbuf_t *pbuf, struct tcphdr *th, int off,
3254 int *rewrptr)
3255{
3256#pragma unused(dir, kif)
3257 sa_family_t af = pd->af;
3258 u_int16_t *mss;
3259 int thoff;
3260 int opt, cnt, optlen = 0;
3261 int rewrite = 0;
3262 u_char opts[MAX_TCPOPTLEN];
3263 u_char *optp = opts;
3264
3265 thoff = th->th_off << 2;
3266 cnt = thoff - sizeof(struct tcphdr);
3267
3268 if (cnt > 0 && !pf_pull_hdr(pbuf, off + sizeof(*th), opts, cnt,
3269 NULL, NULL, af)) {
3270 return PF_DROP;
3271 }
3272
3273 for (; cnt > 0; cnt -= optlen, optp += optlen) {
3274 opt = optp[0];
3275 if (opt == TCPOPT_EOL) {
3276 break;
3277 }
3278 if (opt == TCPOPT_NOP) {
3279 optlen = 1;
3280 } else {
3281 if (cnt < 2) {
3282 break;
3283 }
3284 optlen = optp[1];
3285 if (optlen < 2 || optlen > cnt) {
3286 break;
3287 }
3288 }
3289 switch (opt) {
3290 case TCPOPT_MAXSEG:
3291 mss = (u_int16_t *)(void *)(optp + 2);
3292 if ((ntohs(*mss)) > r->max_mss) {
3293 /*
3294 * <jhw@apple.com>
3295 * Only do the TCP checksum fixup if delayed
3296 * checksum calculation will not be performed.
3297 */
3298 if (pbuf->pb_ifp ||
3299 !(*pbuf->pb_csum_flags & CSUM_TCP)) {
3300 th->th_sum = pf_cksum_fixup(th->th_sum,
3301 *mss, htons(r->max_mss), 0);
3302 }
3303 *mss = htons(r->max_mss);
3304 rewrite = 1;
3305 }
3306 break;
3307 default:
3308 break;
3309 }
3310 }
3311
3312 if (rewrite) {
3313 u_short reason;
3314
3315 VERIFY(pbuf == pd->mp);
3316
3317 if (pf_lazy_makewritable(pd, pd->mp,
3318 off + sizeof(*th) + thoff) == NULL) {
3319 REASON_SET(&reason, PFRES_MEMORY);
3320 if (r->log) {
3321 PFLOG_PACKET(kif, h, pbuf, AF_INET, dir, reason,
3322 r, 0, 0, pd);
3323 }
3324 return PF_DROP;
3325 }
3326
3327 *rewrptr = 1;
3328 pbuf_copy_back(pd->mp, off + sizeof(*th), thoff - sizeof(*th), opts);
3329 }
3330
3331 return PF_PASS;
3332}
3333