1 | /* |
2 | * Copyright (c) 2007-2016 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 | #if INET6 |
86 | #include <netinet/ip6.h> |
87 | #endif /* INET6 */ |
88 | |
89 | #include <net/pfvar.h> |
90 | |
91 | struct pf_frent { |
92 | LIST_ENTRY(pf_frent) fr_next; |
93 | struct mbuf *fr_m; |
94 | #define fr_ip fr_u.fru_ipv4 |
95 | #define fr_ip6 fr_u.fru_ipv6 |
96 | union { |
97 | struct ip *fru_ipv4; |
98 | struct ip6_hdr *fru_ipv6; |
99 | } fr_u; |
100 | struct ip6_frag fr_ip6f_opt; |
101 | int fr_ip6f_hlen; |
102 | }; |
103 | |
104 | struct 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 | |
115 | struct 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 | }; |
140 | |
141 | static TAILQ_HEAD(pf_fragqueue, pf_fragment) pf_fragqueue; |
142 | static TAILQ_HEAD(pf_cachequeue, pf_fragment) pf_cachequeue; |
143 | |
144 | static __inline int pf_frag_compare(struct pf_fragment *, |
145 | struct pf_fragment *); |
146 | static RB_HEAD(pf_frag_tree, pf_fragment) pf_frag_tree, pf_cache_tree; |
147 | RB_PROTOTYPE_SC(__private_extern__, pf_frag_tree, pf_fragment, fr_entry, |
148 | pf_frag_compare); |
149 | RB_GENERATE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare); |
150 | |
151 | /* Private prototypes */ |
152 | static void pf_ip6hdr2key(struct pf_fragment *, struct ip6_hdr *, |
153 | struct ip6_frag *); |
154 | static void pf_ip2key(struct pf_fragment *, struct ip *); |
155 | static void pf_remove_fragment(struct pf_fragment *); |
156 | static void pf_flush_fragments(void); |
157 | static void pf_free_fragment(struct pf_fragment *); |
158 | static struct pf_fragment *pf_find_fragment_by_key(struct pf_fragment *, |
159 | struct pf_frag_tree *); |
160 | static __inline struct pf_fragment * |
161 | pf_find_fragment_by_ipv4_header(struct ip *, struct pf_frag_tree *); |
162 | static __inline struct pf_fragment * |
163 | pf_find_fragment_by_ipv6_header(struct ip6_hdr *, struct ip6_frag *, |
164 | struct pf_frag_tree *); |
165 | static struct mbuf *pf_reassemble(struct mbuf *, struct pf_fragment **, |
166 | struct pf_frent *, int); |
167 | static struct mbuf *pf_fragcache(struct mbuf **, struct ip *, |
168 | struct pf_fragment **, int, int, int *); |
169 | static struct mbuf *pf_reassemble6(struct mbuf **, struct pf_fragment **, |
170 | struct pf_frent *, int); |
171 | static struct mbuf *pf_frag6cache(struct mbuf **, struct ip6_hdr*, |
172 | struct ip6_frag *, struct pf_fragment **, int, int, int, int *); |
173 | static int pf_normalize_tcpopt(struct pf_rule *, int, struct pfi_kif *, |
174 | struct pf_pdesc *, pbuf_t *, struct tcphdr *, int, int *); |
175 | |
176 | #define DPFPRINTF(x) do { \ |
177 | if (pf_status.debug >= PF_DEBUG_MISC) { \ |
178 | printf("%s: ", __func__); \ |
179 | printf x ; \ |
180 | } \ |
181 | } while (0) |
182 | |
183 | /* Globals */ |
184 | struct pool pf_frent_pl, pf_frag_pl; |
185 | static struct pool pf_cache_pl, pf_cent_pl; |
186 | struct pool pf_state_scrub_pl; |
187 | |
188 | static int pf_nfrents, pf_ncache; |
189 | |
190 | void |
191 | pf_normalize_init(void) |
192 | { |
193 | pool_init(&pf_frent_pl, sizeof (struct pf_frent), 0, 0, 0, "pffrent" , |
194 | NULL); |
195 | pool_init(&pf_frag_pl, sizeof (struct pf_fragment), 0, 0, 0, "pffrag" , |
196 | NULL); |
197 | pool_init(&pf_cache_pl, sizeof (struct pf_fragment), 0, 0, 0, |
198 | "pffrcache" , NULL); |
199 | pool_init(&pf_cent_pl, sizeof (struct pf_frcache), 0, 0, 0, "pffrcent" , |
200 | NULL); |
201 | pool_init(&pf_state_scrub_pl, sizeof (struct pf_state_scrub), 0, 0, 0, |
202 | "pfstscr" , NULL); |
203 | |
204 | pool_sethiwat(&pf_frag_pl, PFFRAG_FRAG_HIWAT); |
205 | pool_sethardlimit(&pf_frent_pl, PFFRAG_FRENT_HIWAT, NULL, 0); |
206 | pool_sethardlimit(&pf_cache_pl, PFFRAG_FRCACHE_HIWAT, NULL, 0); |
207 | pool_sethardlimit(&pf_cent_pl, PFFRAG_FRCENT_HIWAT, NULL, 0); |
208 | |
209 | TAILQ_INIT(&pf_fragqueue); |
210 | TAILQ_INIT(&pf_cachequeue); |
211 | } |
212 | |
213 | #if 0 |
214 | void |
215 | pf_normalize_destroy(void) |
216 | { |
217 | pool_destroy(&pf_state_scrub_pl); |
218 | pool_destroy(&pf_cent_pl); |
219 | pool_destroy(&pf_cache_pl); |
220 | pool_destroy(&pf_frag_pl); |
221 | pool_destroy(&pf_frent_pl); |
222 | } |
223 | #endif |
224 | |
225 | int |
226 | pf_normalize_isempty(void) |
227 | { |
228 | return (TAILQ_EMPTY(&pf_fragqueue) && TAILQ_EMPTY(&pf_cachequeue)); |
229 | } |
230 | |
231 | static __inline int |
232 | pf_frag_compare(struct pf_fragment *a, struct pf_fragment *b) |
233 | { |
234 | int diff; |
235 | |
236 | if ((diff = a->fr_af - b->fr_af)) |
237 | return (diff); |
238 | else if ((diff = a->fr_p - b->fr_p)) |
239 | return (diff); |
240 | else { |
241 | struct pf_addr *sa = &a->fr_srcx; |
242 | struct pf_addr *sb = &b->fr_srcx; |
243 | struct pf_addr *da = &a->fr_dstx; |
244 | struct pf_addr *db = &b->fr_dstx; |
245 | |
246 | switch (a->fr_af) { |
247 | #ifdef INET |
248 | case AF_INET: |
249 | if ((diff = a->fr_id - b->fr_id)) |
250 | return (diff); |
251 | else if (sa->v4addr.s_addr < sb->v4addr.s_addr) |
252 | return (-1); |
253 | else if (sa->v4addr.s_addr > sb->v4addr.s_addr) |
254 | return (1); |
255 | else if (da->v4addr.s_addr < db->v4addr.s_addr) |
256 | return (-1); |
257 | else if (da->v4addr.s_addr > db->v4addr.s_addr) |
258 | return (1); |
259 | break; |
260 | #endif |
261 | #ifdef INET6 |
262 | case AF_INET6: |
263 | if ((diff = a->fr_id6 - b->fr_id6)) |
264 | return (diff); |
265 | else if (sa->addr32[3] < sb->addr32[3]) |
266 | return (-1); |
267 | else if (sa->addr32[3] > sb->addr32[3]) |
268 | return (1); |
269 | else if (sa->addr32[2] < sb->addr32[2]) |
270 | return (-1); |
271 | else if (sa->addr32[2] > sb->addr32[2]) |
272 | return (1); |
273 | else if (sa->addr32[1] < sb->addr32[1]) |
274 | return (-1); |
275 | else if (sa->addr32[1] > sb->addr32[1]) |
276 | return (1); |
277 | else if (sa->addr32[0] < sb->addr32[0]) |
278 | return (-1); |
279 | else if (sa->addr32[0] > sb->addr32[0]) |
280 | return (1); |
281 | else if (da->addr32[3] < db->addr32[3]) |
282 | return (-1); |
283 | else if (da->addr32[3] > db->addr32[3]) |
284 | return (1); |
285 | else if (da->addr32[2] < db->addr32[2]) |
286 | return (-1); |
287 | else if (da->addr32[2] > db->addr32[2]) |
288 | return (1); |
289 | else if (da->addr32[1] < db->addr32[1]) |
290 | return (-1); |
291 | else if (da->addr32[1] > db->addr32[1]) |
292 | return (1); |
293 | else if (da->addr32[0] < db->addr32[0]) |
294 | return (-1); |
295 | else if (da->addr32[0] > db->addr32[0]) |
296 | return (1); |
297 | break; |
298 | #endif |
299 | default: |
300 | VERIFY(!0 && "only IPv4 and IPv6 supported!" ); |
301 | break; |
302 | } |
303 | } |
304 | return (0); |
305 | } |
306 | |
307 | void |
308 | pf_purge_expired_fragments(void) |
309 | { |
310 | struct pf_fragment *frag; |
311 | u_int32_t expire = pf_time_second() - |
312 | pf_default_rule.timeout[PFTM_FRAG]; |
313 | |
314 | while ((frag = TAILQ_LAST(&pf_fragqueue, pf_fragqueue)) != NULL) { |
315 | VERIFY(BUFFER_FRAGMENTS(frag)); |
316 | if (frag->fr_timeout > expire) |
317 | break; |
318 | |
319 | switch (frag->fr_af) { |
320 | case AF_INET: |
321 | DPFPRINTF(("expiring IPv4 %d(0x%llx) from queue.\n" , |
322 | ntohs(frag->fr_id), |
323 | (uint64_t)VM_KERNEL_ADDRPERM(frag))); |
324 | break; |
325 | case AF_INET6: |
326 | DPFPRINTF(("expiring IPv6 %d(0x%llx) from queue.\n" , |
327 | ntohl(frag->fr_id6), |
328 | (uint64_t)VM_KERNEL_ADDRPERM(frag))); |
329 | break; |
330 | default: |
331 | VERIFY(0 && "only IPv4 and IPv6 supported" ); |
332 | break; |
333 | } |
334 | pf_free_fragment(frag); |
335 | } |
336 | |
337 | while ((frag = TAILQ_LAST(&pf_cachequeue, pf_cachequeue)) != NULL) { |
338 | VERIFY(!BUFFER_FRAGMENTS(frag)); |
339 | if (frag->fr_timeout > expire) |
340 | break; |
341 | |
342 | switch (frag->fr_af) { |
343 | case AF_INET: |
344 | DPFPRINTF(("expiring IPv4 %d(0x%llx) from cache.\n" , |
345 | ntohs(frag->fr_id), |
346 | (uint64_t)VM_KERNEL_ADDRPERM(frag))); |
347 | break; |
348 | case AF_INET6: |
349 | DPFPRINTF(("expiring IPv6 %d(0x%llx) from cache.\n" , |
350 | ntohl(frag->fr_id6), |
351 | (uint64_t)VM_KERNEL_ADDRPERM(frag))); |
352 | break; |
353 | default: |
354 | VERIFY(0 && "only IPv4 and IPv6 supported" ); |
355 | break; |
356 | } |
357 | pf_free_fragment(frag); |
358 | VERIFY(TAILQ_EMPTY(&pf_cachequeue) || |
359 | TAILQ_LAST(&pf_cachequeue, pf_cachequeue) != frag); |
360 | } |
361 | } |
362 | |
363 | /* |
364 | * Try to flush old fragments to make space for new ones |
365 | */ |
366 | |
367 | static void |
368 | pf_flush_fragments(void) |
369 | { |
370 | struct pf_fragment *frag; |
371 | int goal; |
372 | |
373 | goal = pf_nfrents * 9 / 10; |
374 | DPFPRINTF(("trying to free > %d frents\n" , |
375 | pf_nfrents - goal)); |
376 | while (goal < pf_nfrents) { |
377 | frag = TAILQ_LAST(&pf_fragqueue, pf_fragqueue); |
378 | if (frag == NULL) |
379 | break; |
380 | pf_free_fragment(frag); |
381 | } |
382 | |
383 | |
384 | goal = pf_ncache * 9 / 10; |
385 | DPFPRINTF(("trying to free > %d cache entries\n" , |
386 | pf_ncache - goal)); |
387 | while (goal < pf_ncache) { |
388 | frag = TAILQ_LAST(&pf_cachequeue, pf_cachequeue); |
389 | if (frag == NULL) |
390 | break; |
391 | pf_free_fragment(frag); |
392 | } |
393 | } |
394 | |
395 | /* Frees the fragments and all associated entries */ |
396 | |
397 | static void |
398 | pf_free_fragment(struct pf_fragment *frag) |
399 | { |
400 | struct pf_frent *frent; |
401 | struct pf_frcache *frcache; |
402 | |
403 | /* Free all fragments */ |
404 | if (BUFFER_FRAGMENTS(frag)) { |
405 | for (frent = LIST_FIRST(&frag->fr_queue); frent; |
406 | frent = LIST_FIRST(&frag->fr_queue)) { |
407 | LIST_REMOVE(frent, fr_next); |
408 | |
409 | m_freem(frent->fr_m); |
410 | pool_put(&pf_frent_pl, frent); |
411 | pf_nfrents--; |
412 | } |
413 | } else { |
414 | for (frcache = LIST_FIRST(&frag->fr_cache); frcache; |
415 | frcache = LIST_FIRST(&frag->fr_cache)) { |
416 | LIST_REMOVE(frcache, fr_next); |
417 | |
418 | VERIFY(LIST_EMPTY(&frag->fr_cache) || |
419 | LIST_FIRST(&frag->fr_cache)->fr_off > |
420 | frcache->fr_end); |
421 | |
422 | pool_put(&pf_cent_pl, frcache); |
423 | pf_ncache--; |
424 | } |
425 | } |
426 | |
427 | pf_remove_fragment(frag); |
428 | } |
429 | |
430 | static void |
431 | pf_ip6hdr2key(struct pf_fragment *key, struct ip6_hdr *ip6, |
432 | struct ip6_frag *fh) |
433 | { |
434 | key->fr_p = fh->ip6f_nxt; |
435 | key->fr_id6 = fh->ip6f_ident; |
436 | key->fr_af = AF_INET6; |
437 | key->fr_srcx.v6addr = ip6->ip6_src; |
438 | key->fr_dstx.v6addr = ip6->ip6_dst; |
439 | } |
440 | |
441 | static void |
442 | pf_ip2key(struct pf_fragment *key, struct ip *ip) |
443 | { |
444 | key->fr_p = ip->ip_p; |
445 | key->fr_id = ip->ip_id; |
446 | key->fr_af = AF_INET; |
447 | key->fr_srcx.v4addr.s_addr = ip->ip_src.s_addr; |
448 | key->fr_dstx.v4addr.s_addr = ip->ip_dst.s_addr; |
449 | } |
450 | |
451 | static struct pf_fragment * |
452 | pf_find_fragment_by_key(struct pf_fragment *key, struct pf_frag_tree *tree) |
453 | { |
454 | struct pf_fragment *frag; |
455 | |
456 | frag = RB_FIND(pf_frag_tree, tree, key); |
457 | if (frag != NULL) { |
458 | /* XXX Are we sure we want to update the timeout? */ |
459 | frag->fr_timeout = pf_time_second(); |
460 | if (BUFFER_FRAGMENTS(frag)) { |
461 | TAILQ_REMOVE(&pf_fragqueue, frag, frag_next); |
462 | TAILQ_INSERT_HEAD(&pf_fragqueue, frag, frag_next); |
463 | } else { |
464 | TAILQ_REMOVE(&pf_cachequeue, frag, frag_next); |
465 | TAILQ_INSERT_HEAD(&pf_cachequeue, frag, frag_next); |
466 | } |
467 | } |
468 | |
469 | return (frag); |
470 | } |
471 | |
472 | static __inline struct pf_fragment * |
473 | (struct ip *ip, struct pf_frag_tree *tree) |
474 | { |
475 | struct pf_fragment key; |
476 | pf_ip2key(&key, ip); |
477 | return pf_find_fragment_by_key(&key, tree); |
478 | } |
479 | |
480 | static __inline struct pf_fragment * |
481 | (struct ip6_hdr *ip6, struct ip6_frag *fh, |
482 | struct pf_frag_tree *tree) |
483 | { |
484 | struct pf_fragment key; |
485 | pf_ip6hdr2key(&key, ip6, fh); |
486 | return pf_find_fragment_by_key(&key, tree); |
487 | } |
488 | |
489 | /* Removes a fragment from the fragment queue and frees the fragment */ |
490 | |
491 | static void |
492 | pf_remove_fragment(struct pf_fragment *frag) |
493 | { |
494 | if (BUFFER_FRAGMENTS(frag)) { |
495 | RB_REMOVE(pf_frag_tree, &pf_frag_tree, frag); |
496 | TAILQ_REMOVE(&pf_fragqueue, frag, frag_next); |
497 | pool_put(&pf_frag_pl, frag); |
498 | } else { |
499 | RB_REMOVE(pf_frag_tree, &pf_cache_tree, frag); |
500 | TAILQ_REMOVE(&pf_cachequeue, frag, frag_next); |
501 | pool_put(&pf_cache_pl, frag); |
502 | } |
503 | } |
504 | |
505 | #define FR_IP_OFF(fr) ((ntohs((fr)->fr_ip->ip_off) & IP_OFFMASK) << 3) |
506 | static struct mbuf * |
507 | pf_reassemble(struct mbuf *m0, struct pf_fragment **frag, |
508 | struct pf_frent *frent, int mff) |
509 | { |
510 | struct mbuf *m = m0, *m2; |
511 | struct pf_frent *frea, *next; |
512 | struct pf_frent *frep = NULL; |
513 | struct ip *ip = frent->fr_ip; |
514 | uint32_t hlen = ip->ip_hl << 2; |
515 | u_int16_t off = (ntohs(ip->ip_off) & IP_OFFMASK) << 3; |
516 | u_int16_t ip_len = ntohs(ip->ip_len) - ip->ip_hl * 4; |
517 | u_int16_t fr_max = ip_len + off; |
518 | uint32_t csum, csum_flags; |
519 | |
520 | VERIFY(*frag == NULL || BUFFER_FRAGMENTS(*frag)); |
521 | |
522 | /* |
523 | * Leverage partial checksum offload for IP fragments. Narrow down |
524 | * the scope to cover only UDP without IP options, as that is the |
525 | * most common case. |
526 | * |
527 | * Perform 1's complement adjustment of octets that got included/ |
528 | * excluded in the hardware-calculated checksum value. Ignore cases |
529 | * where the value includes the entire IPv4 header span, as the sum |
530 | * for those octets would already be 0 by the time we get here; IP |
531 | * has already performed its header checksum validation. Also take |
532 | * care of any trailing bytes and subtract out their partial sum. |
533 | */ |
534 | if (ip->ip_p == IPPROTO_UDP && hlen == sizeof (struct ip) && |
535 | (m->m_pkthdr.csum_flags & |
536 | (CSUM_DATA_VALID | CSUM_PARTIAL | CSUM_PSEUDO_HDR)) == |
537 | (CSUM_DATA_VALID | CSUM_PARTIAL)) { |
538 | uint32_t start = m->m_pkthdr.csum_rx_start; |
539 | int32_t trailer = (m_pktlen(m) - ntohs(ip->ip_len)); |
540 | uint32_t swbytes = (uint32_t)trailer; |
541 | |
542 | csum = m->m_pkthdr.csum_rx_val; |
543 | |
544 | ASSERT(trailer >= 0); |
545 | if ((start != 0 && start != hlen) || trailer != 0) { |
546 | #if BYTE_ORDER != BIG_ENDIAN |
547 | if (start < hlen) { |
548 | HTONS(ip->ip_len); |
549 | HTONS(ip->ip_off); |
550 | } |
551 | #endif /* BYTE_ORDER != BIG_ENDIAN */ |
552 | /* callee folds in sum */ |
553 | csum = m_adj_sum16(m, start, hlen, |
554 | (ip->ip_len - hlen), csum); |
555 | if (hlen > start) |
556 | swbytes += (hlen - start); |
557 | else |
558 | swbytes += (start - hlen); |
559 | #if BYTE_ORDER != BIG_ENDIAN |
560 | if (start < hlen) { |
561 | NTOHS(ip->ip_off); |
562 | NTOHS(ip->ip_len); |
563 | } |
564 | #endif /* BYTE_ORDER != BIG_ENDIAN */ |
565 | } |
566 | csum_flags = m->m_pkthdr.csum_flags; |
567 | |
568 | if (swbytes != 0) |
569 | udp_in_cksum_stats(swbytes); |
570 | if (trailer != 0) |
571 | m_adj(m, -trailer); |
572 | } else { |
573 | csum = 0; |
574 | csum_flags = 0; |
575 | } |
576 | |
577 | /* Invalidate checksum */ |
578 | m->m_pkthdr.csum_flags &= ~CSUM_DATA_VALID; |
579 | |
580 | /* Strip off ip header */ |
581 | m->m_data += hlen; |
582 | m->m_len -= hlen; |
583 | |
584 | /* Create a new reassembly queue for this packet */ |
585 | if (*frag == NULL) { |
586 | *frag = pool_get(&pf_frag_pl, PR_NOWAIT); |
587 | if (*frag == NULL) { |
588 | pf_flush_fragments(); |
589 | *frag = pool_get(&pf_frag_pl, PR_NOWAIT); |
590 | if (*frag == NULL) |
591 | goto drop_fragment; |
592 | } |
593 | |
594 | (*frag)->fr_flags = 0; |
595 | (*frag)->fr_max = 0; |
596 | (*frag)->fr_af = AF_INET; |
597 | (*frag)->fr_srcx.v4addr = frent->fr_ip->ip_src; |
598 | (*frag)->fr_dstx.v4addr = frent->fr_ip->ip_dst; |
599 | (*frag)->fr_p = frent->fr_ip->ip_p; |
600 | (*frag)->fr_id = frent->fr_ip->ip_id; |
601 | (*frag)->fr_timeout = pf_time_second(); |
602 | if (csum_flags != 0) { |
603 | (*frag)->fr_csum_flags = csum_flags; |
604 | (*frag)->fr_csum = csum; |
605 | } |
606 | LIST_INIT(&(*frag)->fr_queue); |
607 | |
608 | RB_INSERT(pf_frag_tree, &pf_frag_tree, *frag); |
609 | TAILQ_INSERT_HEAD(&pf_fragqueue, *frag, frag_next); |
610 | |
611 | /* We do not have a previous fragment */ |
612 | frep = NULL; |
613 | goto insert; |
614 | } |
615 | |
616 | /* |
617 | * If this fragment contains similar checksum offload info |
618 | * as that of the existing ones, accumulate checksum. Otherwise, |
619 | * invalidate checksum offload info for the entire datagram. |
620 | */ |
621 | if (csum_flags != 0 && csum_flags == (*frag)->fr_csum_flags) |
622 | (*frag)->fr_csum += csum; |
623 | else if ((*frag)->fr_csum_flags != 0) |
624 | (*frag)->fr_csum_flags = 0; |
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 | frep = frea; |
634 | } |
635 | |
636 | VERIFY(frep != NULL || frea != NULL); |
637 | |
638 | if (frep != NULL && |
639 | FR_IP_OFF(frep) + ntohs(frep->fr_ip->ip_len) - frep->fr_ip->ip_hl * |
640 | 4 > off) { |
641 | u_int16_t precut; |
642 | |
643 | precut = FR_IP_OFF(frep) + ntohs(frep->fr_ip->ip_len) - |
644 | frep->fr_ip->ip_hl * 4 - off; |
645 | if (precut >= ip_len) |
646 | goto drop_fragment; |
647 | m_adj(frent->fr_m, precut); |
648 | DPFPRINTF(("overlap -%d\n" , precut)); |
649 | /* Enforce 8 byte boundaries */ |
650 | ip->ip_off = htons(ntohs(ip->ip_off) + (precut >> 3)); |
651 | off = (ntohs(ip->ip_off) & IP_OFFMASK) << 3; |
652 | ip_len -= precut; |
653 | ip->ip_len = htons(ip_len); |
654 | } |
655 | |
656 | for (; frea != NULL && ip_len + off > FR_IP_OFF(frea); |
657 | frea = next) { |
658 | u_int16_t aftercut; |
659 | |
660 | aftercut = ip_len + off - FR_IP_OFF(frea); |
661 | DPFPRINTF(("adjust overlap %d\n" , aftercut)); |
662 | if (aftercut < ntohs(frea->fr_ip->ip_len) - frea->fr_ip->ip_hl |
663 | * 4) { |
664 | frea->fr_ip->ip_len = |
665 | htons(ntohs(frea->fr_ip->ip_len) - aftercut); |
666 | frea->fr_ip->ip_off = htons(ntohs(frea->fr_ip->ip_off) + |
667 | (aftercut >> 3)); |
668 | m_adj(frea->fr_m, aftercut); |
669 | break; |
670 | } |
671 | |
672 | /* This fragment is completely overlapped, lose it */ |
673 | next = LIST_NEXT(frea, fr_next); |
674 | m_freem(frea->fr_m); |
675 | LIST_REMOVE(frea, fr_next); |
676 | pool_put(&pf_frent_pl, frea); |
677 | pf_nfrents--; |
678 | } |
679 | |
680 | insert: |
681 | /* Update maximum data size */ |
682 | if ((*frag)->fr_max < fr_max) |
683 | (*frag)->fr_max = fr_max; |
684 | /* This is the last segment */ |
685 | if (!mff) |
686 | (*frag)->fr_flags |= PFFRAG_SEENLAST; |
687 | |
688 | if (frep == NULL) |
689 | LIST_INSERT_HEAD(&(*frag)->fr_queue, frent, fr_next); |
690 | else |
691 | LIST_INSERT_AFTER(frep, frent, fr_next); |
692 | |
693 | /* Check if we are completely reassembled */ |
694 | if (!((*frag)->fr_flags & PFFRAG_SEENLAST)) |
695 | return (NULL); |
696 | |
697 | /* Check if we have all the data */ |
698 | off = 0; |
699 | for (frep = LIST_FIRST(&(*frag)->fr_queue); frep; frep = next) { |
700 | next = LIST_NEXT(frep, fr_next); |
701 | |
702 | off += ntohs(frep->fr_ip->ip_len) - frep->fr_ip->ip_hl * 4; |
703 | if (off < (*frag)->fr_max && |
704 | (next == NULL || FR_IP_OFF(next) != off)) { |
705 | DPFPRINTF(("missing fragment at %d, next %d, max %d\n" , |
706 | off, next == NULL ? -1 : FR_IP_OFF(next), |
707 | (*frag)->fr_max)); |
708 | return (NULL); |
709 | } |
710 | } |
711 | DPFPRINTF(("%d < %d?\n" , off, (*frag)->fr_max)); |
712 | if (off < (*frag)->fr_max) |
713 | return (NULL); |
714 | |
715 | /* We have all the data */ |
716 | frent = LIST_FIRST(&(*frag)->fr_queue); |
717 | VERIFY(frent != NULL); |
718 | if ((frent->fr_ip->ip_hl << 2) + off > IP_MAXPACKET) { |
719 | DPFPRINTF(("drop: too big: %d\n" , off)); |
720 | pf_free_fragment(*frag); |
721 | *frag = NULL; |
722 | return (NULL); |
723 | } |
724 | next = LIST_NEXT(frent, fr_next); |
725 | |
726 | /* Magic from ip_input */ |
727 | ip = frent->fr_ip; |
728 | m = frent->fr_m; |
729 | m2 = m->m_next; |
730 | m->m_next = NULL; |
731 | m_cat(m, m2); |
732 | pool_put(&pf_frent_pl, frent); |
733 | pf_nfrents--; |
734 | for (frent = next; frent != NULL; frent = next) { |
735 | next = LIST_NEXT(frent, fr_next); |
736 | |
737 | m2 = frent->fr_m; |
738 | pool_put(&pf_frent_pl, frent); |
739 | pf_nfrents--; |
740 | m_cat(m, m2); |
741 | } |
742 | |
743 | ip->ip_src = (*frag)->fr_srcx.v4addr; |
744 | ip->ip_dst = (*frag)->fr_dstx.v4addr; |
745 | |
746 | if ((*frag)->fr_csum_flags != 0) { |
747 | csum = (*frag)->fr_csum; |
748 | |
749 | ADDCARRY(csum); |
750 | |
751 | m->m_pkthdr.csum_rx_val = csum; |
752 | m->m_pkthdr.csum_rx_start = sizeof (struct ip); |
753 | m->m_pkthdr.csum_flags = (*frag)->fr_csum_flags; |
754 | } else if ((m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) || |
755 | (m->m_pkthdr.pkt_flags & PKTF_LOOP)) { |
756 | /* loopback checksums are always OK */ |
757 | m->m_pkthdr.csum_data = 0xffff; |
758 | m->m_pkthdr.csum_flags &= ~CSUM_PARTIAL; |
759 | m->m_pkthdr.csum_flags = |
760 | CSUM_DATA_VALID | CSUM_PSEUDO_HDR | |
761 | CSUM_IP_CHECKED | CSUM_IP_VALID; |
762 | } |
763 | |
764 | /* Remove from fragment queue */ |
765 | pf_remove_fragment(*frag); |
766 | *frag = NULL; |
767 | |
768 | hlen = ip->ip_hl << 2; |
769 | ip->ip_len = htons(off + hlen); |
770 | m->m_len += hlen; |
771 | m->m_data -= hlen; |
772 | |
773 | /* some debugging cruft by sklower, below, will go away soon */ |
774 | /* XXX this should be done elsewhere */ |
775 | if (m->m_flags & M_PKTHDR) { |
776 | int plen = 0; |
777 | for (m2 = m; m2; m2 = m2->m_next) |
778 | plen += m2->m_len; |
779 | m->m_pkthdr.len = plen; |
780 | } |
781 | |
782 | DPFPRINTF(("complete: 0x%llx(%d)\n" , |
783 | (uint64_t)VM_KERNEL_ADDRPERM(m), ntohs(ip->ip_len))); |
784 | return (m); |
785 | |
786 | drop_fragment: |
787 | /* Oops - fail safe - drop packet */ |
788 | pool_put(&pf_frent_pl, frent); |
789 | pf_nfrents--; |
790 | m_freem(m); |
791 | return (NULL); |
792 | } |
793 | |
794 | static struct mbuf * |
795 | pf_fragcache(struct mbuf **m0, struct ip *h, struct pf_fragment **frag, int mff, |
796 | int drop, int *nomem) |
797 | { |
798 | struct mbuf *m = *m0; |
799 | struct pf_frcache *frp, *fra, *cur = NULL; |
800 | int ip_len = ntohs(h->ip_len) - (h->ip_hl << 2); |
801 | u_int16_t off = ntohs(h->ip_off) << 3; |
802 | u_int16_t fr_max = ip_len + off; |
803 | int hosed = 0; |
804 | |
805 | VERIFY(*frag == NULL || !BUFFER_FRAGMENTS(*frag)); |
806 | |
807 | /* Create a new range queue for this packet */ |
808 | if (*frag == NULL) { |
809 | *frag = pool_get(&pf_cache_pl, PR_NOWAIT); |
810 | if (*frag == NULL) { |
811 | pf_flush_fragments(); |
812 | *frag = pool_get(&pf_cache_pl, PR_NOWAIT); |
813 | if (*frag == NULL) |
814 | goto no_mem; |
815 | } |
816 | |
817 | /* Get an entry for the queue */ |
818 | cur = pool_get(&pf_cent_pl, PR_NOWAIT); |
819 | if (cur == NULL) { |
820 | pool_put(&pf_cache_pl, *frag); |
821 | *frag = NULL; |
822 | goto no_mem; |
823 | } |
824 | pf_ncache++; |
825 | |
826 | (*frag)->fr_flags = PFFRAG_NOBUFFER; |
827 | (*frag)->fr_max = 0; |
828 | (*frag)->fr_af = AF_INET; |
829 | (*frag)->fr_srcx.v4addr = h->ip_src; |
830 | (*frag)->fr_dstx.v4addr = h->ip_dst; |
831 | (*frag)->fr_p = h->ip_p; |
832 | (*frag)->fr_id = h->ip_id; |
833 | (*frag)->fr_timeout = pf_time_second(); |
834 | |
835 | cur->fr_off = off; |
836 | cur->fr_end = fr_max; |
837 | LIST_INIT(&(*frag)->fr_cache); |
838 | LIST_INSERT_HEAD(&(*frag)->fr_cache, cur, fr_next); |
839 | |
840 | RB_INSERT(pf_frag_tree, &pf_cache_tree, *frag); |
841 | TAILQ_INSERT_HEAD(&pf_cachequeue, *frag, frag_next); |
842 | |
843 | DPFPRINTF(("fragcache[%d]: new %d-%d\n" , h->ip_id, off, |
844 | fr_max)); |
845 | |
846 | goto pass; |
847 | } |
848 | |
849 | /* |
850 | * Find a fragment after the current one: |
851 | * - off contains the real shifted offset. |
852 | */ |
853 | frp = NULL; |
854 | LIST_FOREACH(fra, &(*frag)->fr_cache, fr_next) { |
855 | if (fra->fr_off > off) |
856 | break; |
857 | frp = fra; |
858 | } |
859 | |
860 | VERIFY(frp != NULL || fra != NULL); |
861 | |
862 | if (frp != NULL) { |
863 | int precut; |
864 | |
865 | precut = frp->fr_end - off; |
866 | if (precut >= ip_len) { |
867 | /* Fragment is entirely a duplicate */ |
868 | DPFPRINTF(("fragcache[%d]: dead (%d-%d) %d-%d\n" , |
869 | h->ip_id, frp->fr_off, frp->fr_end, off, fr_max)); |
870 | goto drop_fragment; |
871 | } |
872 | if (precut == 0) { |
873 | /* They are adjacent. Fixup cache entry */ |
874 | DPFPRINTF(("fragcache[%d]: adjacent (%d-%d) %d-%d\n" , |
875 | h->ip_id, frp->fr_off, frp->fr_end, off, fr_max)); |
876 | frp->fr_end = fr_max; |
877 | } else if (precut > 0) { |
878 | /* |
879 | * The first part of this payload overlaps with a |
880 | * fragment that has already been passed. |
881 | * Need to trim off the first part of the payload. |
882 | * But to do so easily, we need to create another |
883 | * mbuf to throw the original header into. |
884 | */ |
885 | |
886 | DPFPRINTF(("fragcache[%d]: chop %d (%d-%d) %d-%d\n" , |
887 | h->ip_id, precut, frp->fr_off, frp->fr_end, off, |
888 | fr_max)); |
889 | |
890 | off += precut; |
891 | fr_max -= precut; |
892 | /* Update the previous frag to encompass this one */ |
893 | frp->fr_end = fr_max; |
894 | |
895 | if (!drop) { |
896 | /* |
897 | * XXX Optimization opportunity |
898 | * This is a very heavy way to trim the payload. |
899 | * we could do it much faster by diddling mbuf |
900 | * internals but that would be even less legible |
901 | * than this mbuf magic. For my next trick, |
902 | * I'll pull a rabbit out of my laptop. |
903 | */ |
904 | *m0 = m_copym(m, 0, h->ip_hl << 2, M_NOWAIT); |
905 | if (*m0 == NULL) |
906 | goto no_mem; |
907 | VERIFY((*m0)->m_next == NULL); |
908 | m_adj(m, precut + (h->ip_hl << 2)); |
909 | m_cat(*m0, m); |
910 | m = *m0; |
911 | if (m->m_flags & M_PKTHDR) { |
912 | int plen = 0; |
913 | struct mbuf *t; |
914 | for (t = m; t; t = t->m_next) |
915 | plen += t->m_len; |
916 | m->m_pkthdr.len = plen; |
917 | } |
918 | |
919 | |
920 | h = mtod(m, struct ip *); |
921 | |
922 | |
923 | VERIFY((int)m->m_len == |
924 | ntohs(h->ip_len) - precut); |
925 | h->ip_off = htons(ntohs(h->ip_off) + |
926 | (precut >> 3)); |
927 | h->ip_len = htons(ntohs(h->ip_len) - precut); |
928 | } else { |
929 | hosed++; |
930 | } |
931 | } else { |
932 | /* There is a gap between fragments */ |
933 | |
934 | DPFPRINTF(("fragcache[%d]: gap %d (%d-%d) %d-%d\n" , |
935 | h->ip_id, -precut, frp->fr_off, frp->fr_end, off, |
936 | fr_max)); |
937 | |
938 | cur = pool_get(&pf_cent_pl, PR_NOWAIT); |
939 | if (cur == NULL) |
940 | goto no_mem; |
941 | pf_ncache++; |
942 | |
943 | cur->fr_off = off; |
944 | cur->fr_end = fr_max; |
945 | LIST_INSERT_AFTER(frp, cur, fr_next); |
946 | } |
947 | } |
948 | |
949 | if (fra != NULL) { |
950 | int aftercut; |
951 | int merge = 0; |
952 | |
953 | aftercut = fr_max - fra->fr_off; |
954 | if (aftercut == 0) { |
955 | /* Adjacent fragments */ |
956 | DPFPRINTF(("fragcache[%d]: adjacent %d-%d (%d-%d)\n" , |
957 | h->ip_id, off, fr_max, fra->fr_off, fra->fr_end)); |
958 | fra->fr_off = off; |
959 | merge = 1; |
960 | } else if (aftercut > 0) { |
961 | /* Need to chop off the tail of this fragment */ |
962 | DPFPRINTF(("fragcache[%d]: chop %d %d-%d (%d-%d)\n" , |
963 | h->ip_id, aftercut, off, fr_max, fra->fr_off, |
964 | fra->fr_end)); |
965 | fra->fr_off = off; |
966 | fr_max -= aftercut; |
967 | |
968 | merge = 1; |
969 | |
970 | if (!drop) { |
971 | m_adj(m, -aftercut); |
972 | if (m->m_flags & M_PKTHDR) { |
973 | int plen = 0; |
974 | struct mbuf *t; |
975 | for (t = m; t; t = t->m_next) |
976 | plen += t->m_len; |
977 | m->m_pkthdr.len = plen; |
978 | } |
979 | h = mtod(m, struct ip *); |
980 | VERIFY((int)m->m_len == |
981 | ntohs(h->ip_len) - aftercut); |
982 | h->ip_len = htons(ntohs(h->ip_len) - aftercut); |
983 | } else { |
984 | hosed++; |
985 | } |
986 | } else if (frp == NULL) { |
987 | /* There is a gap between fragments */ |
988 | DPFPRINTF(("fragcache[%d]: gap %d %d-%d (%d-%d)\n" , |
989 | h->ip_id, -aftercut, off, fr_max, fra->fr_off, |
990 | fra->fr_end)); |
991 | |
992 | cur = pool_get(&pf_cent_pl, PR_NOWAIT); |
993 | if (cur == NULL) |
994 | goto no_mem; |
995 | pf_ncache++; |
996 | |
997 | cur->fr_off = off; |
998 | cur->fr_end = fr_max; |
999 | LIST_INSERT_BEFORE(fra, cur, fr_next); |
1000 | } |
1001 | |
1002 | |
1003 | /* Need to glue together two separate fragment descriptors */ |
1004 | if (merge) { |
1005 | if (cur && fra->fr_off <= cur->fr_end) { |
1006 | /* Need to merge in a previous 'cur' */ |
1007 | DPFPRINTF(("fragcache[%d]: adjacent(merge " |
1008 | "%d-%d) %d-%d (%d-%d)\n" , |
1009 | h->ip_id, cur->fr_off, cur->fr_end, off, |
1010 | fr_max, fra->fr_off, fra->fr_end)); |
1011 | fra->fr_off = cur->fr_off; |
1012 | LIST_REMOVE(cur, fr_next); |
1013 | pool_put(&pf_cent_pl, cur); |
1014 | pf_ncache--; |
1015 | cur = NULL; |
1016 | |
1017 | } else if (frp && fra->fr_off <= frp->fr_end) { |
1018 | /* Need to merge in a modified 'frp' */ |
1019 | VERIFY(cur == NULL); |
1020 | DPFPRINTF(("fragcache[%d]: adjacent(merge " |
1021 | "%d-%d) %d-%d (%d-%d)\n" , |
1022 | h->ip_id, frp->fr_off, frp->fr_end, off, |
1023 | fr_max, fra->fr_off, fra->fr_end)); |
1024 | fra->fr_off = frp->fr_off; |
1025 | LIST_REMOVE(frp, fr_next); |
1026 | pool_put(&pf_cent_pl, frp); |
1027 | pf_ncache--; |
1028 | frp = NULL; |
1029 | |
1030 | } |
1031 | } |
1032 | } |
1033 | |
1034 | if (hosed) { |
1035 | /* |
1036 | * We must keep tracking the overall fragment even when |
1037 | * we're going to drop it anyway so that we know when to |
1038 | * free the overall descriptor. Thus we drop the frag late. |
1039 | */ |
1040 | goto drop_fragment; |
1041 | } |
1042 | |
1043 | |
1044 | pass: |
1045 | /* Update maximum data size */ |
1046 | if ((*frag)->fr_max < fr_max) |
1047 | (*frag)->fr_max = fr_max; |
1048 | |
1049 | /* This is the last segment */ |
1050 | if (!mff) |
1051 | (*frag)->fr_flags |= PFFRAG_SEENLAST; |
1052 | |
1053 | /* Check if we are completely reassembled */ |
1054 | if (((*frag)->fr_flags & PFFRAG_SEENLAST) && |
1055 | LIST_FIRST(&(*frag)->fr_cache)->fr_off == 0 && |
1056 | LIST_FIRST(&(*frag)->fr_cache)->fr_end == (*frag)->fr_max) { |
1057 | /* Remove from fragment queue */ |
1058 | DPFPRINTF(("fragcache[%d]: done 0-%d\n" , h->ip_id, |
1059 | (*frag)->fr_max)); |
1060 | pf_free_fragment(*frag); |
1061 | *frag = NULL; |
1062 | } |
1063 | |
1064 | return (m); |
1065 | |
1066 | no_mem: |
1067 | *nomem = 1; |
1068 | |
1069 | /* Still need to pay attention to !IP_MF */ |
1070 | if (!mff && *frag != NULL) |
1071 | (*frag)->fr_flags |= PFFRAG_SEENLAST; |
1072 | |
1073 | m_freem(m); |
1074 | return (NULL); |
1075 | |
1076 | drop_fragment: |
1077 | |
1078 | /* Still need to pay attention to !IP_MF */ |
1079 | if (!mff && *frag != NULL) |
1080 | (*frag)->fr_flags |= PFFRAG_SEENLAST; |
1081 | |
1082 | if (drop) { |
1083 | /* This fragment has been deemed bad. Don't reass */ |
1084 | if (((*frag)->fr_flags & PFFRAG_DROP) == 0) |
1085 | DPFPRINTF(("fragcache[%d]: dropping overall fragment\n" , |
1086 | h->ip_id)); |
1087 | (*frag)->fr_flags |= PFFRAG_DROP; |
1088 | } |
1089 | |
1090 | m_freem(m); |
1091 | return (NULL); |
1092 | } |
1093 | |
1094 | #define FR_IP6_OFF(fr) \ |
1095 | (ntohs((fr)->fr_ip6f_opt.ip6f_offlg & IP6F_OFF_MASK)) |
1096 | #define FR_IP6_PLEN(fr) (ntohs((fr)->fr_ip6->ip6_plen)) |
1097 | struct mbuf * |
1098 | pf_reassemble6(struct mbuf **m0, struct pf_fragment **frag, |
1099 | struct pf_frent *frent, int mff) |
1100 | { |
1101 | struct mbuf *m, *m2; |
1102 | struct pf_frent *frea, *frep, *next; |
1103 | struct ip6_hdr *ip6; |
1104 | struct ip6_frag *ip6f; |
1105 | int plen, off, fr_max; |
1106 | uint32_t uoff, csum, csum_flags; |
1107 | |
1108 | VERIFY(*frag == NULL || BUFFER_FRAGMENTS(*frag)); |
1109 | m = *m0; |
1110 | frep = NULL; |
1111 | ip6 = frent->fr_ip6; |
1112 | ip6f = &frent->fr_ip6f_opt; |
1113 | off = FR_IP6_OFF(frent); |
1114 | uoff = frent->fr_ip6f_hlen; |
1115 | plen = FR_IP6_PLEN(frent); |
1116 | fr_max = off + plen - (frent->fr_ip6f_hlen - sizeof *ip6); |
1117 | |
1118 | DPFPRINTF(("0x%llx IPv6 frag plen %u off %u fr_ip6f_hlen %u " |
1119 | "fr_max %u m_len %u\n" , (uint64_t)VM_KERNEL_ADDRPERM(m), plen, off, |
1120 | frent->fr_ip6f_hlen, fr_max, m->m_len)); |
1121 | |
1122 | /* |
1123 | * Leverage partial checksum offload for simple UDP/IP fragments, |
1124 | * as that is the most common case. |
1125 | * |
1126 | * Perform 1's complement adjustment of octets that got included/ |
1127 | * excluded in the hardware-calculated checksum value. Also take |
1128 | * care of any trailing bytes and subtract out their partial sum. |
1129 | */ |
1130 | if (ip6f->ip6f_nxt == IPPROTO_UDP && |
1131 | uoff == (sizeof (*ip6) + sizeof (*ip6f)) && |
1132 | (m->m_pkthdr.csum_flags & |
1133 | (CSUM_DATA_VALID | CSUM_PARTIAL | CSUM_PSEUDO_HDR)) == |
1134 | (CSUM_DATA_VALID | CSUM_PARTIAL)) { |
1135 | uint32_t start = m->m_pkthdr.csum_rx_start; |
1136 | uint32_t ip_len = (sizeof (*ip6) + ntohs(ip6->ip6_plen)); |
1137 | int32_t trailer = (m_pktlen(m) - ip_len); |
1138 | uint32_t swbytes = (uint32_t)trailer; |
1139 | |
1140 | csum = m->m_pkthdr.csum_rx_val; |
1141 | |
1142 | ASSERT(trailer >= 0); |
1143 | if (start != uoff || trailer != 0) { |
1144 | uint16_t s = 0, d = 0; |
1145 | |
1146 | if (IN6_IS_SCOPE_EMBED(&ip6->ip6_src)) { |
1147 | s = ip6->ip6_src.s6_addr16[1]; |
1148 | ip6->ip6_src.s6_addr16[1] = 0 ; |
1149 | } |
1150 | if (IN6_IS_SCOPE_EMBED(&ip6->ip6_dst)) { |
1151 | d = ip6->ip6_dst.s6_addr16[1]; |
1152 | ip6->ip6_dst.s6_addr16[1] = 0; |
1153 | } |
1154 | |
1155 | /* callee folds in sum */ |
1156 | csum = m_adj_sum16(m, start, uoff, |
1157 | (ip_len - uoff), csum); |
1158 | if (uoff > start) |
1159 | swbytes += (uoff - start); |
1160 | else |
1161 | swbytes += (start - uoff); |
1162 | |
1163 | if (IN6_IS_SCOPE_EMBED(&ip6->ip6_src)) |
1164 | ip6->ip6_src.s6_addr16[1] = s; |
1165 | if (IN6_IS_SCOPE_EMBED(&ip6->ip6_dst)) |
1166 | ip6->ip6_dst.s6_addr16[1] = d; |
1167 | |
1168 | } |
1169 | csum_flags = m->m_pkthdr.csum_flags; |
1170 | |
1171 | if (swbytes != 0) |
1172 | udp_in6_cksum_stats(swbytes); |
1173 | if (trailer != 0) |
1174 | m_adj(m, -trailer); |
1175 | } else { |
1176 | csum = 0; |
1177 | csum_flags = 0; |
1178 | } |
1179 | |
1180 | /* Invalidate checksum */ |
1181 | m->m_pkthdr.csum_flags &= ~CSUM_DATA_VALID; |
1182 | |
1183 | /* strip off headers up to the fragment payload */ |
1184 | m->m_data += frent->fr_ip6f_hlen; |
1185 | m->m_len -= frent->fr_ip6f_hlen; |
1186 | |
1187 | /* Create a new reassembly queue for this packet */ |
1188 | if (*frag == NULL) { |
1189 | *frag = pool_get(&pf_frag_pl, PR_NOWAIT); |
1190 | if (*frag == NULL) { |
1191 | pf_flush_fragments(); |
1192 | *frag = pool_get(&pf_frag_pl, PR_NOWAIT); |
1193 | if (*frag == NULL) |
1194 | goto drop_fragment; |
1195 | } |
1196 | |
1197 | (*frag)->fr_flags = 0; |
1198 | (*frag)->fr_max = 0; |
1199 | (*frag)->fr_af = AF_INET6; |
1200 | (*frag)->fr_srcx.v6addr = frent->fr_ip6->ip6_src; |
1201 | (*frag)->fr_dstx.v6addr = frent->fr_ip6->ip6_dst; |
1202 | (*frag)->fr_p = frent->fr_ip6f_opt.ip6f_nxt; |
1203 | (*frag)->fr_id6 = frent->fr_ip6f_opt.ip6f_ident; |
1204 | (*frag)->fr_timeout = pf_time_second(); |
1205 | if (csum_flags != 0) { |
1206 | (*frag)->fr_csum_flags = csum_flags; |
1207 | (*frag)->fr_csum = csum; |
1208 | } |
1209 | LIST_INIT(&(*frag)->fr_queue); |
1210 | |
1211 | RB_INSERT(pf_frag_tree, &pf_frag_tree, *frag); |
1212 | TAILQ_INSERT_HEAD(&pf_fragqueue, *frag, frag_next); |
1213 | |
1214 | /* We do not have a previous fragment */ |
1215 | frep = NULL; |
1216 | goto insert; |
1217 | } |
1218 | |
1219 | /* |
1220 | * If this fragment contains similar checksum offload info |
1221 | * as that of the existing ones, accumulate checksum. Otherwise, |
1222 | * invalidate checksum offload info for the entire datagram. |
1223 | */ |
1224 | if (csum_flags != 0 && csum_flags == (*frag)->fr_csum_flags) |
1225 | (*frag)->fr_csum += csum; |
1226 | else if ((*frag)->fr_csum_flags != 0) |
1227 | (*frag)->fr_csum_flags = 0; |
1228 | |
1229 | /* |
1230 | * Find a fragment after the current one: |
1231 | * - off contains the real shifted offset. |
1232 | */ |
1233 | LIST_FOREACH(frea, &(*frag)->fr_queue, fr_next) { |
1234 | if (FR_IP6_OFF(frea) > off) |
1235 | break; |
1236 | frep = frea; |
1237 | } |
1238 | |
1239 | VERIFY(frep != NULL || frea != NULL); |
1240 | |
1241 | if (frep != NULL && |
1242 | FR_IP6_OFF(frep) + FR_IP6_PLEN(frep) - frep->fr_ip6f_hlen > off) |
1243 | { |
1244 | u_int16_t precut; |
1245 | |
1246 | precut = FR_IP6_OFF(frep) + FR_IP6_PLEN(frep) - |
1247 | frep->fr_ip6f_hlen - off; |
1248 | if (precut >= plen) |
1249 | goto drop_fragment; |
1250 | m_adj(frent->fr_m, precut); |
1251 | DPFPRINTF(("overlap -%d\n" , precut)); |
1252 | /* Enforce 8 byte boundaries */ |
1253 | frent->fr_ip6f_opt.ip6f_offlg = |
1254 | htons(ntohs(frent->fr_ip6f_opt.ip6f_offlg) + |
1255 | (precut >> 3)); |
1256 | off = FR_IP6_OFF(frent); |
1257 | plen -= precut; |
1258 | ip6->ip6_plen = htons(plen); |
1259 | } |
1260 | |
1261 | for (; frea != NULL && plen + off > FR_IP6_OFF(frea); frea = next) { |
1262 | u_int16_t aftercut; |
1263 | |
1264 | aftercut = plen + off - FR_IP6_OFF(frea); |
1265 | DPFPRINTF(("adjust overlap %d\n" , aftercut)); |
1266 | if (aftercut < FR_IP6_PLEN(frea) - frea->fr_ip6f_hlen) { |
1267 | frea->fr_ip6->ip6_plen = htons(FR_IP6_PLEN(frea) - |
1268 | aftercut); |
1269 | frea->fr_ip6f_opt.ip6f_offlg = |
1270 | htons(ntohs(frea->fr_ip6f_opt.ip6f_offlg) + |
1271 | (aftercut >> 3)); |
1272 | m_adj(frea->fr_m, aftercut); |
1273 | break; |
1274 | } |
1275 | |
1276 | /* This fragment is completely overlapped, lose it */ |
1277 | next = LIST_NEXT(frea, fr_next); |
1278 | m_freem(frea->fr_m); |
1279 | LIST_REMOVE(frea, fr_next); |
1280 | pool_put(&pf_frent_pl, frea); |
1281 | pf_nfrents--; |
1282 | } |
1283 | |
1284 | insert: |
1285 | /* Update maximum data size */ |
1286 | if ((*frag)->fr_max < fr_max) |
1287 | (*frag)->fr_max = fr_max; |
1288 | /* This is the last segment */ |
1289 | if (!mff) |
1290 | (*frag)->fr_flags |= PFFRAG_SEENLAST; |
1291 | |
1292 | if (frep == NULL) |
1293 | LIST_INSERT_HEAD(&(*frag)->fr_queue, frent, fr_next); |
1294 | else |
1295 | LIST_INSERT_AFTER(frep, frent, fr_next); |
1296 | |
1297 | /* Check if we are completely reassembled */ |
1298 | if (!((*frag)->fr_flags & PFFRAG_SEENLAST)) |
1299 | return (NULL); |
1300 | |
1301 | /* Check if we have all the data */ |
1302 | off = 0; |
1303 | for (frep = LIST_FIRST(&(*frag)->fr_queue); frep; frep = next) { |
1304 | next = LIST_NEXT(frep, fr_next); |
1305 | off += FR_IP6_PLEN(frep) - (frent->fr_ip6f_hlen - sizeof *ip6); |
1306 | DPFPRINTF(("frep at %d, next %d, max %d\n" , |
1307 | off, next == NULL ? -1 : FR_IP6_OFF(next), |
1308 | (*frag)->fr_max)); |
1309 | if (off < (*frag)->fr_max && |
1310 | (next == NULL || FR_IP6_OFF(next) != off)) { |
1311 | DPFPRINTF(("missing fragment at %d, next %d, max %d\n" , |
1312 | off, next == NULL ? -1 : FR_IP6_OFF(next), |
1313 | (*frag)->fr_max)); |
1314 | return (NULL); |
1315 | } |
1316 | } |
1317 | DPFPRINTF(("%d < %d?\n" , off, (*frag)->fr_max)); |
1318 | if (off < (*frag)->fr_max) |
1319 | return (NULL); |
1320 | |
1321 | /* We have all the data */ |
1322 | frent = LIST_FIRST(&(*frag)->fr_queue); |
1323 | VERIFY(frent != NULL); |
1324 | if (frent->fr_ip6f_hlen + off > IP_MAXPACKET) { |
1325 | DPFPRINTF(("drop: too big: %d\n" , off)); |
1326 | pf_free_fragment(*frag); |
1327 | *frag = NULL; |
1328 | return (NULL); |
1329 | } |
1330 | |
1331 | ip6 = frent->fr_ip6; |
1332 | ip6->ip6_nxt = (*frag)->fr_p; |
1333 | ip6->ip6_plen = htons(off); |
1334 | ip6->ip6_src = (*frag)->fr_srcx.v6addr; |
1335 | ip6->ip6_dst = (*frag)->fr_dstx.v6addr; |
1336 | |
1337 | if ((*frag)->fr_csum_flags != 0) { |
1338 | csum = (*frag)->fr_csum; |
1339 | |
1340 | ADDCARRY(csum); |
1341 | |
1342 | m->m_pkthdr.csum_rx_val = csum; |
1343 | m->m_pkthdr.csum_rx_start = sizeof (struct ip6_hdr); |
1344 | m->m_pkthdr.csum_flags = (*frag)->fr_csum_flags; |
1345 | } else if ((m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) || |
1346 | (m->m_pkthdr.pkt_flags & PKTF_LOOP)) { |
1347 | /* loopback checksums are always OK */ |
1348 | m->m_pkthdr.csum_data = 0xffff; |
1349 | m->m_pkthdr.csum_flags &= ~CSUM_PARTIAL; |
1350 | m->m_pkthdr.csum_flags = CSUM_DATA_VALID | CSUM_PSEUDO_HDR; |
1351 | } |
1352 | |
1353 | /* Remove from fragment queue */ |
1354 | pf_remove_fragment(*frag); |
1355 | *frag = NULL; |
1356 | |
1357 | m = frent->fr_m; |
1358 | m->m_len += sizeof(struct ip6_hdr); |
1359 | m->m_data -= sizeof(struct ip6_hdr); |
1360 | memmove(m->m_data, ip6, sizeof(struct ip6_hdr)); |
1361 | |
1362 | next = LIST_NEXT(frent, fr_next); |
1363 | pool_put(&pf_frent_pl, frent); |
1364 | pf_nfrents--; |
1365 | for (frent = next; next != NULL; frent = next) { |
1366 | m2 = frent->fr_m; |
1367 | |
1368 | m_cat(m, m2); |
1369 | next = LIST_NEXT(frent, fr_next); |
1370 | pool_put(&pf_frent_pl, frent); |
1371 | pf_nfrents--; |
1372 | } |
1373 | |
1374 | /* XXX this should be done elsewhere */ |
1375 | if (m->m_flags & M_PKTHDR) { |
1376 | int pktlen = 0; |
1377 | for (m2 = m; m2; m2 = m2->m_next) |
1378 | pktlen += m2->m_len; |
1379 | m->m_pkthdr.len = pktlen; |
1380 | } |
1381 | |
1382 | DPFPRINTF(("complete: 0x%llx ip6_plen %d m_pkthdr.len %d\n" , |
1383 | (uint64_t)VM_KERNEL_ADDRPERM(m), ntohs(ip6->ip6_plen), |
1384 | m->m_pkthdr.len)); |
1385 | |
1386 | return m; |
1387 | |
1388 | drop_fragment: |
1389 | /* Oops - fail safe - drop packet */ |
1390 | pool_put(&pf_frent_pl, frent); |
1391 | --pf_nfrents; |
1392 | m_freem(m); |
1393 | return NULL; |
1394 | } |
1395 | |
1396 | static struct mbuf * |
1397 | pf_frag6cache(struct mbuf **m0, struct ip6_hdr *h, struct ip6_frag *fh, |
1398 | struct pf_fragment **frag, int hlen, int mff, int drop, int *nomem) |
1399 | { |
1400 | struct mbuf *m = *m0; |
1401 | u_int16_t plen, off, fr_max; |
1402 | struct pf_frcache *frp, *fra, *cur = NULL; |
1403 | int hosed = 0; |
1404 | |
1405 | VERIFY(*frag == NULL || !BUFFER_FRAGMENTS(*frag)); |
1406 | m = *m0; |
1407 | off = ntohs(fh->ip6f_offlg & IP6F_OFF_MASK); |
1408 | plen = ntohs(h->ip6_plen) - (hlen - sizeof *h); |
1409 | |
1410 | /* |
1411 | * Apple Modification: dimambro@apple.com. The hlen, being passed |
1412 | * into this function Includes all the headers associated with |
1413 | * the packet, and may include routing headers, so to get to |
1414 | * the data payload as stored in the original IPv6 header we need |
1415 | * to subtract al those headers and the IP header. |
1416 | * |
1417 | * The 'max' local variable should also contain the offset from the start |
1418 | * of the reassembled packet to the octet just past the end of the octets |
1419 | * in the current fragment where: |
1420 | * - 'off' is the offset from the start of the reassembled packet to the |
1421 | * first octet in the fragment, |
1422 | * - 'plen' is the length of the "payload data length" Excluding all the |
1423 | * IPv6 headers of the fragment. |
1424 | * - 'hlen' is computed in pf_normalize_ip6() as the offset from the start |
1425 | * of the IPv6 packet to the beginning of the data. |
1426 | */ |
1427 | fr_max = off + plen; |
1428 | |
1429 | DPFPRINTF(("0x%llx plen %u off %u fr_max %u\n" , |
1430 | (uint64_t)VM_KERNEL_ADDRPERM(m), plen, off, fr_max)); |
1431 | |
1432 | /* Create a new range queue for this packet */ |
1433 | if (*frag == NULL) { |
1434 | *frag = pool_get(&pf_cache_pl, PR_NOWAIT); |
1435 | if (*frag == NULL) { |
1436 | pf_flush_fragments(); |
1437 | *frag = pool_get(&pf_cache_pl, PR_NOWAIT); |
1438 | if (*frag == NULL) |
1439 | goto no_mem; |
1440 | } |
1441 | |
1442 | /* Get an entry for the queue */ |
1443 | cur = pool_get(&pf_cent_pl, PR_NOWAIT); |
1444 | if (cur == NULL) { |
1445 | pool_put(&pf_cache_pl, *frag); |
1446 | *frag = NULL; |
1447 | goto no_mem; |
1448 | } |
1449 | pf_ncache++; |
1450 | |
1451 | (*frag)->fr_flags = PFFRAG_NOBUFFER; |
1452 | (*frag)->fr_max = 0; |
1453 | (*frag)->fr_af = AF_INET6; |
1454 | (*frag)->fr_srcx.v6addr = h->ip6_src; |
1455 | (*frag)->fr_dstx.v6addr = h->ip6_dst; |
1456 | (*frag)->fr_p = fh->ip6f_nxt; |
1457 | (*frag)->fr_id6 = fh->ip6f_ident; |
1458 | (*frag)->fr_timeout = pf_time_second(); |
1459 | |
1460 | cur->fr_off = off; |
1461 | cur->fr_end = fr_max; |
1462 | LIST_INIT(&(*frag)->fr_cache); |
1463 | LIST_INSERT_HEAD(&(*frag)->fr_cache, cur, fr_next); |
1464 | |
1465 | RB_INSERT(pf_frag_tree, &pf_cache_tree, *frag); |
1466 | TAILQ_INSERT_HEAD(&pf_cachequeue, *frag, frag_next); |
1467 | |
1468 | DPFPRINTF(("frag6cache[%d]: new %d-%d\n" , ntohl(fh->ip6f_ident), |
1469 | off, fr_max)); |
1470 | |
1471 | goto pass; |
1472 | } |
1473 | |
1474 | /* |
1475 | * Find a fragment after the current one: |
1476 | * - off contains the real shifted offset. |
1477 | */ |
1478 | frp = NULL; |
1479 | LIST_FOREACH(fra, &(*frag)->fr_cache, fr_next) { |
1480 | if (fra->fr_off > off) |
1481 | break; |
1482 | frp = fra; |
1483 | } |
1484 | |
1485 | VERIFY(frp != NULL || fra != NULL); |
1486 | |
1487 | if (frp != NULL) { |
1488 | int precut; |
1489 | |
1490 | precut = frp->fr_end - off; |
1491 | if (precut >= plen) { |
1492 | /* Fragment is entirely a duplicate */ |
1493 | DPFPRINTF(("frag6cache[%u]: dead (%d-%d) %d-%d\n" , |
1494 | ntohl(fh->ip6f_ident), frp->fr_off, frp->fr_end, |
1495 | off, fr_max)); |
1496 | goto drop_fragment; |
1497 | } |
1498 | if (precut == 0) { |
1499 | /* They are adjacent. Fixup cache entry */ |
1500 | DPFPRINTF(("frag6cache[%u]: adjacent (%d-%d) %d-%d\n" , |
1501 | ntohl(fh->ip6f_ident), frp->fr_off, frp->fr_end, |
1502 | off, fr_max)); |
1503 | frp->fr_end = fr_max; |
1504 | } else if (precut > 0) { |
1505 | /* The first part of this payload overlaps with a |
1506 | * fragment that has already been passed. |
1507 | * Need to trim off the first part of the payload. |
1508 | * But to do so easily, we need to create another |
1509 | * mbuf to throw the original header into. |
1510 | */ |
1511 | |
1512 | DPFPRINTF(("frag6cache[%u]: chop %d (%d-%d) %d-%d\n" , |
1513 | ntohl(fh->ip6f_ident), precut, frp->fr_off, |
1514 | frp->fr_end, off, fr_max)); |
1515 | |
1516 | off += precut; |
1517 | fr_max -= precut; |
1518 | /* Update the previous frag to encompass this one */ |
1519 | frp->fr_end = fr_max; |
1520 | |
1521 | if (!drop) { |
1522 | /* XXX Optimization opportunity |
1523 | * This is a very heavy way to trim the payload. |
1524 | * we could do it much faster by diddling mbuf |
1525 | * internals but that would be even less legible |
1526 | * than this mbuf magic. For my next trick, |
1527 | * I'll pull a rabbit out of my laptop. |
1528 | */ |
1529 | *m0 = m_copym(m, 0, hlen, M_NOWAIT); |
1530 | if (*m0 == NULL) |
1531 | goto no_mem; |
1532 | VERIFY((*m0)->m_next == NULL); |
1533 | m_adj(m, precut + hlen); |
1534 | m_cat(*m0, m); |
1535 | m = *m0; |
1536 | if (m->m_flags & M_PKTHDR) { |
1537 | int pktlen = 0; |
1538 | struct mbuf *t; |
1539 | for (t = m; t; t = t->m_next) |
1540 | pktlen += t->m_len; |
1541 | m->m_pkthdr.len = pktlen; |
1542 | } |
1543 | |
1544 | h = mtod(m, struct ip6_hdr *); |
1545 | |
1546 | VERIFY((int)m->m_len == |
1547 | ntohs(h->ip6_plen) - precut); |
1548 | fh->ip6f_offlg &= ~IP6F_OFF_MASK; |
1549 | fh->ip6f_offlg |= |
1550 | htons(ntohs(fh->ip6f_offlg & IP6F_OFF_MASK) |
1551 | + (precut >> 3)); |
1552 | h->ip6_plen = htons(ntohs(h->ip6_plen) - |
1553 | precut); |
1554 | } else { |
1555 | hosed++; |
1556 | } |
1557 | } else { |
1558 | /* There is a gap between fragments */ |
1559 | |
1560 | DPFPRINTF(("frag6cache[%u]: gap %d (%d-%d) %d-%d\n" , |
1561 | ntohl(fh->ip6f_ident), -precut, frp->fr_off, |
1562 | frp->fr_end, off, fr_max)); |
1563 | |
1564 | cur = pool_get(&pf_cent_pl, PR_NOWAIT); |
1565 | if (cur == NULL) |
1566 | goto no_mem; |
1567 | pf_ncache++; |
1568 | |
1569 | cur->fr_off = off; |
1570 | cur->fr_end = fr_max; |
1571 | LIST_INSERT_AFTER(frp, cur, fr_next); |
1572 | } |
1573 | } |
1574 | |
1575 | if (fra != NULL) { |
1576 | int aftercut; |
1577 | int merge = 0; |
1578 | |
1579 | aftercut = fr_max - fra->fr_off; |
1580 | if (aftercut == 0) { |
1581 | /* Adjacent fragments */ |
1582 | DPFPRINTF(("frag6cache[%u]: adjacent %d-%d (%d-%d)\n" , |
1583 | ntohl(fh->ip6f_ident), off, fr_max, fra->fr_off, |
1584 | fra->fr_end)); |
1585 | fra->fr_off = off; |
1586 | merge = 1; |
1587 | } else if (aftercut > 0) { |
1588 | /* Need to chop off the tail of this fragment */ |
1589 | DPFPRINTF(("frag6cache[%u]: chop %d %d-%d (%d-%d)\n" , |
1590 | ntohl(fh->ip6f_ident), aftercut, off, fr_max, |
1591 | fra->fr_off, fra->fr_end)); |
1592 | fra->fr_off = off; |
1593 | fr_max -= aftercut; |
1594 | |
1595 | merge = 1; |
1596 | |
1597 | if (!drop) { |
1598 | m_adj(m, -aftercut); |
1599 | if (m->m_flags & M_PKTHDR) { |
1600 | int pktlen = 0; |
1601 | struct mbuf *t; |
1602 | for (t = m; t; t = t->m_next) |
1603 | pktlen += t->m_len; |
1604 | m->m_pkthdr.len = pktlen; |
1605 | } |
1606 | h = mtod(m, struct ip6_hdr *); |
1607 | VERIFY((int)m->m_len == |
1608 | ntohs(h->ip6_plen) - aftercut); |
1609 | h->ip6_plen = |
1610 | htons(ntohs(h->ip6_plen) - aftercut); |
1611 | } else { |
1612 | hosed++; |
1613 | } |
1614 | } else if (frp == NULL) { |
1615 | /* There is a gap between fragments */ |
1616 | DPFPRINTF(("frag6cache[%u]: gap %d %d-%d (%d-%d)\n" , |
1617 | ntohl(fh->ip6f_ident), -aftercut, off, fr_max, |
1618 | fra->fr_off, fra->fr_end)); |
1619 | |
1620 | cur = pool_get(&pf_cent_pl, PR_NOWAIT); |
1621 | if (cur == NULL) |
1622 | goto no_mem; |
1623 | pf_ncache++; |
1624 | |
1625 | cur->fr_off = off; |
1626 | cur->fr_end = fr_max; |
1627 | LIST_INSERT_BEFORE(fra, cur, fr_next); |
1628 | } |
1629 | |
1630 | /* Need to glue together two separate fragment descriptors */ |
1631 | if (merge) { |
1632 | if (cur && fra->fr_off <= cur->fr_end) { |
1633 | /* Need to merge in a previous 'cur' */ |
1634 | DPFPRINTF(("frag6cache[%u]: adjacent(merge " |
1635 | "%d-%d) %d-%d (%d-%d)\n" , |
1636 | ntohl(fh->ip6f_ident), cur->fr_off, |
1637 | cur->fr_end, off, fr_max, fra->fr_off, |
1638 | fra->fr_end)); |
1639 | fra->fr_off = cur->fr_off; |
1640 | LIST_REMOVE(cur, fr_next); |
1641 | pool_put(&pf_cent_pl, cur); |
1642 | pf_ncache--; |
1643 | cur = NULL; |
1644 | } else if (frp && fra->fr_off <= frp->fr_end) { |
1645 | /* Need to merge in a modified 'frp' */ |
1646 | VERIFY(cur == NULL); |
1647 | DPFPRINTF(("frag6cache[%u]: adjacent(merge " |
1648 | "%d-%d) %d-%d (%d-%d)\n" , |
1649 | ntohl(fh->ip6f_ident), frp->fr_off, |
1650 | frp->fr_end, off, fr_max, fra->fr_off, |
1651 | fra->fr_end)); |
1652 | fra->fr_off = frp->fr_off; |
1653 | LIST_REMOVE(frp, fr_next); |
1654 | pool_put(&pf_cent_pl, frp); |
1655 | pf_ncache--; |
1656 | frp = NULL; |
1657 | } |
1658 | } |
1659 | } |
1660 | |
1661 | if (hosed) { |
1662 | /* |
1663 | * We must keep tracking the overall fragment even when |
1664 | * we're going to drop it anyway so that we know when to |
1665 | * free the overall descriptor. Thus we drop the frag late. |
1666 | */ |
1667 | goto drop_fragment; |
1668 | } |
1669 | |
1670 | pass: |
1671 | /* Update maximum data size */ |
1672 | if ((*frag)->fr_max < fr_max) |
1673 | (*frag)->fr_max = fr_max; |
1674 | |
1675 | /* This is the last segment */ |
1676 | if (!mff) |
1677 | (*frag)->fr_flags |= PFFRAG_SEENLAST; |
1678 | |
1679 | /* Check if we are completely reassembled */ |
1680 | if (((*frag)->fr_flags & PFFRAG_SEENLAST) && |
1681 | LIST_FIRST(&(*frag)->fr_cache)->fr_off == 0 && |
1682 | LIST_FIRST(&(*frag)->fr_cache)->fr_end == (*frag)->fr_max) { |
1683 | /* Remove from fragment queue */ |
1684 | DPFPRINTF(("frag6cache[%u]: done 0-%d\n" , |
1685 | ntohl(fh->ip6f_ident), (*frag)->fr_max)); |
1686 | pf_free_fragment(*frag); |
1687 | *frag = NULL; |
1688 | } |
1689 | |
1690 | return (m); |
1691 | |
1692 | no_mem: |
1693 | *nomem = 1; |
1694 | |
1695 | /* Still need to pay attention to !IP_MF */ |
1696 | if (!mff && *frag != NULL) |
1697 | (*frag)->fr_flags |= PFFRAG_SEENLAST; |
1698 | |
1699 | m_freem(m); |
1700 | return (NULL); |
1701 | |
1702 | drop_fragment: |
1703 | |
1704 | /* Still need to pay attention to !IP_MF */ |
1705 | if (!mff && *frag != NULL) |
1706 | (*frag)->fr_flags |= PFFRAG_SEENLAST; |
1707 | |
1708 | if (drop) { |
1709 | /* This fragment has been deemed bad. Don't reass */ |
1710 | if (((*frag)->fr_flags & PFFRAG_DROP) == 0) |
1711 | DPFPRINTF(("frag6cache[%u]: dropping overall fragment\n" , |
1712 | ntohl(fh->ip6f_ident))); |
1713 | (*frag)->fr_flags |= PFFRAG_DROP; |
1714 | } |
1715 | |
1716 | m_freem(m); |
1717 | return (NULL); |
1718 | } |
1719 | |
1720 | int |
1721 | pf_normalize_ip(pbuf_t *pbuf, int dir, struct pfi_kif *kif, u_short *reason, |
1722 | struct pf_pdesc *pd) |
1723 | { |
1724 | struct mbuf *m; |
1725 | struct pf_rule *r; |
1726 | struct pf_frent *frent; |
1727 | struct pf_fragment *frag = NULL; |
1728 | struct ip *h = pbuf->pb_data; |
1729 | int mff = (ntohs(h->ip_off) & IP_MF); |
1730 | int hlen = h->ip_hl << 2; |
1731 | u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3; |
1732 | u_int16_t fr_max; |
1733 | int ip_len; |
1734 | int ip_off; |
1735 | int asd = 0; |
1736 | struct pf_ruleset *ruleset = NULL; |
1737 | struct ifnet *ifp = pbuf->pb_ifp; |
1738 | |
1739 | r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr); |
1740 | while (r != NULL) { |
1741 | r->evaluations++; |
1742 | if (pfi_kif_match(r->kif, kif) == r->ifnot) |
1743 | r = r->skip[PF_SKIP_IFP].ptr; |
1744 | else if (r->direction && r->direction != dir) |
1745 | r = r->skip[PF_SKIP_DIR].ptr; |
1746 | else if (r->af && r->af != AF_INET) |
1747 | r = r->skip[PF_SKIP_AF].ptr; |
1748 | else if (r->proto && r->proto != h->ip_p) |
1749 | r = r->skip[PF_SKIP_PROTO].ptr; |
1750 | else if (PF_MISMATCHAW(&r->src.addr, |
1751 | (struct pf_addr *)&h->ip_src.s_addr, AF_INET, |
1752 | r->src.neg, kif)) |
1753 | r = r->skip[PF_SKIP_SRC_ADDR].ptr; |
1754 | else if (PF_MISMATCHAW(&r->dst.addr, |
1755 | (struct pf_addr *)&h->ip_dst.s_addr, AF_INET, |
1756 | r->dst.neg, NULL)) |
1757 | r = r->skip[PF_SKIP_DST_ADDR].ptr; |
1758 | else { |
1759 | if (r->anchor == NULL) |
1760 | break; |
1761 | else |
1762 | pf_step_into_anchor(&asd, &ruleset, |
1763 | PF_RULESET_SCRUB, &r, NULL, NULL); |
1764 | } |
1765 | if (r == NULL && pf_step_out_of_anchor(&asd, &ruleset, |
1766 | PF_RULESET_SCRUB, &r, NULL, NULL)) |
1767 | break; |
1768 | } |
1769 | |
1770 | if (r == NULL || r->action == PF_NOSCRUB) |
1771 | return (PF_PASS); |
1772 | else { |
1773 | r->packets[dir == PF_OUT]++; |
1774 | r->bytes[dir == PF_OUT] += pd->tot_len; |
1775 | } |
1776 | |
1777 | /* Check for illegal packets */ |
1778 | if (hlen < (int)sizeof (struct ip)) |
1779 | goto drop; |
1780 | |
1781 | if (hlen > ntohs(h->ip_len)) |
1782 | goto drop; |
1783 | |
1784 | /* Clear IP_DF if the rule uses the no-df option */ |
1785 | if (r->rule_flag & PFRULE_NODF && h->ip_off & htons(IP_DF)) { |
1786 | u_int16_t ipoff = h->ip_off; |
1787 | |
1788 | h->ip_off &= htons(~IP_DF); |
1789 | h->ip_sum = pf_cksum_fixup(h->ip_sum, ipoff, h->ip_off, 0); |
1790 | } |
1791 | |
1792 | /* We will need other tests here */ |
1793 | if (!fragoff && !mff) |
1794 | goto no_fragment; |
1795 | |
1796 | /* |
1797 | * We're dealing with a fragment now. Don't allow fragments |
1798 | * with IP_DF to enter the cache. If the flag was cleared by |
1799 | * no-df above, fine. Otherwise drop it. |
1800 | */ |
1801 | if (h->ip_off & htons(IP_DF)) { |
1802 | DPFPRINTF(("IP_DF\n" )); |
1803 | goto bad; |
1804 | } |
1805 | |
1806 | ip_len = ntohs(h->ip_len) - hlen; |
1807 | ip_off = (ntohs(h->ip_off) & IP_OFFMASK) << 3; |
1808 | |
1809 | /* All fragments are 8 byte aligned */ |
1810 | if (mff && (ip_len & 0x7)) { |
1811 | DPFPRINTF(("mff and %d\n" , ip_len)); |
1812 | goto bad; |
1813 | } |
1814 | |
1815 | /* Respect maximum length */ |
1816 | if (fragoff + ip_len > IP_MAXPACKET) { |
1817 | DPFPRINTF(("max packet %d\n" , fragoff + ip_len)); |
1818 | goto bad; |
1819 | } |
1820 | fr_max = fragoff + ip_len; |
1821 | |
1822 | if ((r->rule_flag & (PFRULE_FRAGCROP|PFRULE_FRAGDROP)) == 0) { |
1823 | /* Fully buffer all of the fragments */ |
1824 | |
1825 | frag = pf_find_fragment_by_ipv4_header(h, &pf_frag_tree); |
1826 | /* Check if we saw the last fragment already */ |
1827 | if (frag != NULL && (frag->fr_flags & PFFRAG_SEENLAST) && |
1828 | fr_max > frag->fr_max) |
1829 | goto bad; |
1830 | |
1831 | if ((m = pbuf_to_mbuf(pbuf, TRUE)) == NULL) { |
1832 | REASON_SET(reason, PFRES_MEMORY); |
1833 | return (PF_DROP); |
1834 | } |
1835 | |
1836 | VERIFY(!pbuf_is_valid(pbuf)); |
1837 | |
1838 | /* Restore iph pointer after pbuf_to_mbuf() */ |
1839 | h = mtod(m, struct ip *); |
1840 | |
1841 | /* Get an entry for the fragment queue */ |
1842 | frent = pool_get(&pf_frent_pl, PR_NOWAIT); |
1843 | if (frent == NULL) { |
1844 | REASON_SET(reason, PFRES_MEMORY); |
1845 | m_freem(m); |
1846 | return (PF_DROP); |
1847 | } |
1848 | pf_nfrents++; |
1849 | frent->fr_ip = h; |
1850 | frent->fr_m = m; |
1851 | |
1852 | /* Might return a completely reassembled mbuf, or NULL */ |
1853 | DPFPRINTF(("reass IPv4 frag %d @ %d-%d\n" , ntohs(h->ip_id), |
1854 | fragoff, fr_max)); |
1855 | m = pf_reassemble(m, &frag, frent, mff); |
1856 | |
1857 | if (m == NULL) |
1858 | return (PF_DROP); |
1859 | |
1860 | VERIFY(m->m_flags & M_PKTHDR); |
1861 | pbuf_init_mbuf(pbuf, m, ifp); |
1862 | |
1863 | /* use mtag from concatenated mbuf chain */ |
1864 | pd->pf_mtag = pf_find_mtag_pbuf(pbuf); |
1865 | #if 0 |
1866 | // SCW: This check is superfluous |
1867 | #if DIAGNOSTIC |
1868 | if (pd->pf_mtag == NULL) { |
1869 | printf("%s: pf_find_mtag returned NULL(1)\n" , __func__); |
1870 | if ((pd->pf_mtag = pf_get_mtag(m)) == NULL) { |
1871 | m_freem(m); |
1872 | m = NULL; |
1873 | goto no_mem; |
1874 | } |
1875 | } |
1876 | #endif |
1877 | #endif |
1878 | |
1879 | h = mtod(m, struct ip *); |
1880 | |
1881 | if (frag != NULL && (frag->fr_flags & PFFRAG_DROP)) |
1882 | goto drop; |
1883 | } else { |
1884 | /* non-buffering fragment cache (drops or masks overlaps) */ |
1885 | int nomem = 0; |
1886 | |
1887 | if (dir == PF_OUT && (pd->pf_mtag->pftag_flags & PF_TAG_FRAGCACHE)) { |
1888 | /* |
1889 | * Already passed the fragment cache in the |
1890 | * input direction. If we continued, it would |
1891 | * appear to be a dup and would be dropped. |
1892 | */ |
1893 | goto fragment_pass; |
1894 | } |
1895 | |
1896 | frag = pf_find_fragment_by_ipv4_header(h, &pf_cache_tree); |
1897 | |
1898 | /* Check if we saw the last fragment already */ |
1899 | if (frag != NULL && (frag->fr_flags & PFFRAG_SEENLAST) && |
1900 | fr_max > frag->fr_max) { |
1901 | if (r->rule_flag & PFRULE_FRAGDROP) |
1902 | frag->fr_flags |= PFFRAG_DROP; |
1903 | goto bad; |
1904 | } |
1905 | |
1906 | if ((m = pbuf_to_mbuf(pbuf, TRUE)) == NULL) { |
1907 | REASON_SET(reason, PFRES_MEMORY); |
1908 | goto bad; |
1909 | } |
1910 | |
1911 | VERIFY(!pbuf_is_valid(pbuf)); |
1912 | |
1913 | /* Restore iph pointer after pbuf_to_mbuf() */ |
1914 | h = mtod(m, struct ip *); |
1915 | |
1916 | m = pf_fragcache(&m, h, &frag, mff, |
1917 | (r->rule_flag & PFRULE_FRAGDROP) ? 1 : 0, &nomem); |
1918 | if (m == NULL) { |
1919 | // Note: pf_fragcache() has already m_freem'd the mbuf |
1920 | if (nomem) |
1921 | goto no_mem; |
1922 | goto drop; |
1923 | } |
1924 | |
1925 | VERIFY(m->m_flags & M_PKTHDR); |
1926 | pbuf_init_mbuf(pbuf, m, ifp); |
1927 | |
1928 | /* use mtag from copied and trimmed mbuf chain */ |
1929 | pd->pf_mtag = pf_find_mtag_pbuf(pbuf); |
1930 | #if 0 |
1931 | // SCW: This check is superfluous |
1932 | #if DIAGNOSTIC |
1933 | if (pd->pf_mtag == NULL) { |
1934 | printf("%s: pf_find_mtag returned NULL(2)\n" , __func__); |
1935 | if ((pd->pf_mtag = pf_get_mtag(m)) == NULL) { |
1936 | m_freem(m); |
1937 | m = NULL; |
1938 | goto no_mem; |
1939 | } |
1940 | } |
1941 | #endif |
1942 | #endif |
1943 | if (dir == PF_IN) |
1944 | pd->pf_mtag->pftag_flags |= PF_TAG_FRAGCACHE; |
1945 | |
1946 | if (frag != NULL && (frag->fr_flags & PFFRAG_DROP)) |
1947 | goto drop; |
1948 | |
1949 | goto fragment_pass; |
1950 | } |
1951 | |
1952 | no_fragment: |
1953 | /* At this point, only IP_DF is allowed in ip_off */ |
1954 | if (h->ip_off & ~htons(IP_DF)) { |
1955 | u_int16_t ipoff = h->ip_off; |
1956 | |
1957 | h->ip_off &= htons(IP_DF); |
1958 | h->ip_sum = pf_cksum_fixup(h->ip_sum, ipoff, h->ip_off, 0); |
1959 | } |
1960 | |
1961 | /* Enforce a minimum ttl, may cause endless packet loops */ |
1962 | if (r->min_ttl && h->ip_ttl < r->min_ttl) { |
1963 | u_int16_t ip_ttl = h->ip_ttl; |
1964 | |
1965 | h->ip_ttl = r->min_ttl; |
1966 | h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_ttl, h->ip_ttl, 0); |
1967 | } |
1968 | if (r->rule_flag & PFRULE_RANDOMID) { |
1969 | u_int16_t oip_id = h->ip_id; |
1970 | |
1971 | if (rfc6864 && IP_OFF_IS_ATOMIC(ntohs(h->ip_off))) { |
1972 | h->ip_id = 0; |
1973 | } else { |
1974 | h->ip_id = ip_randomid(); |
1975 | } |
1976 | h->ip_sum = pf_cksum_fixup(h->ip_sum, oip_id, h->ip_id, 0); |
1977 | } |
1978 | if ((r->rule_flag & (PFRULE_FRAGCROP|PFRULE_FRAGDROP)) == 0) |
1979 | pd->flags |= PFDESC_IP_REAS; |
1980 | |
1981 | return (PF_PASS); |
1982 | |
1983 | fragment_pass: |
1984 | /* Enforce a minimum ttl, may cause endless packet loops */ |
1985 | if (r->min_ttl && h->ip_ttl < r->min_ttl) { |
1986 | u_int16_t ip_ttl = h->ip_ttl; |
1987 | |
1988 | h->ip_ttl = r->min_ttl; |
1989 | h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_ttl, h->ip_ttl, 0); |
1990 | } |
1991 | if ((r->rule_flag & (PFRULE_FRAGCROP|PFRULE_FRAGDROP)) == 0) |
1992 | pd->flags |= PFDESC_IP_REAS; |
1993 | return (PF_PASS); |
1994 | |
1995 | no_mem: |
1996 | REASON_SET(reason, PFRES_MEMORY); |
1997 | if (r != NULL && r->log && pbuf_is_valid(pbuf)) |
1998 | PFLOG_PACKET(kif, h, pbuf, AF_INET, dir, *reason, r, |
1999 | NULL, NULL, pd); |
2000 | return (PF_DROP); |
2001 | |
2002 | drop: |
2003 | REASON_SET(reason, PFRES_NORM); |
2004 | if (r != NULL && r->log && pbuf_is_valid(pbuf)) |
2005 | PFLOG_PACKET(kif, h, pbuf, AF_INET, dir, *reason, r, |
2006 | NULL, NULL, pd); |
2007 | return (PF_DROP); |
2008 | |
2009 | bad: |
2010 | DPFPRINTF(("dropping bad IPv4 fragment\n" )); |
2011 | |
2012 | /* Free associated fragments */ |
2013 | if (frag != NULL) |
2014 | pf_free_fragment(frag); |
2015 | |
2016 | REASON_SET(reason, PFRES_FRAG); |
2017 | if (r != NULL && r->log && pbuf_is_valid(pbuf)) |
2018 | PFLOG_PACKET(kif, h, pbuf, AF_INET, dir, *reason, r, NULL, NULL, pd); |
2019 | |
2020 | return (PF_DROP); |
2021 | } |
2022 | |
2023 | #if INET6 |
2024 | int |
2025 | pf_normalize_ip6(pbuf_t *pbuf, int dir, struct pfi_kif *kif, |
2026 | u_short *reason, struct pf_pdesc *pd) |
2027 | { |
2028 | struct mbuf *m; |
2029 | struct pf_rule *r; |
2030 | struct ip6_hdr *h = pbuf->pb_data; |
2031 | int off; |
2032 | struct ip6_ext ext; |
2033 | /* adi XXX */ |
2034 | #if 0 |
2035 | struct ip6_opt opt; |
2036 | struct ip6_opt_jumbo jumbo; |
2037 | int optend; |
2038 | int ooff; |
2039 | #endif |
2040 | struct ip6_frag frag; |
2041 | u_int32_t jumbolen = 0, plen; |
2042 | u_int16_t fragoff = 0; |
2043 | u_int8_t proto; |
2044 | int terminal; |
2045 | struct pf_frent *frent; |
2046 | struct pf_fragment *pff = NULL; |
2047 | int mff = 0, rh_cnt = 0; |
2048 | u_int16_t fr_max; |
2049 | int asd = 0; |
2050 | struct pf_ruleset *ruleset = NULL; |
2051 | struct ifnet *ifp = pbuf->pb_ifp; |
2052 | |
2053 | r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr); |
2054 | while (r != NULL) { |
2055 | r->evaluations++; |
2056 | if (pfi_kif_match(r->kif, kif) == r->ifnot) |
2057 | r = r->skip[PF_SKIP_IFP].ptr; |
2058 | else if (r->direction && r->direction != dir) |
2059 | r = r->skip[PF_SKIP_DIR].ptr; |
2060 | else if (r->af && r->af != AF_INET6) |
2061 | r = r->skip[PF_SKIP_AF].ptr; |
2062 | #if 0 /* header chain! */ |
2063 | else if (r->proto && r->proto != h->ip6_nxt) |
2064 | r = r->skip[PF_SKIP_PROTO].ptr; |
2065 | #endif |
2066 | else if (PF_MISMATCHAW(&r->src.addr, |
2067 | (struct pf_addr *)(uintptr_t)&h->ip6_src, AF_INET6, |
2068 | r->src.neg, kif)) |
2069 | r = r->skip[PF_SKIP_SRC_ADDR].ptr; |
2070 | else if (PF_MISMATCHAW(&r->dst.addr, |
2071 | (struct pf_addr *)(uintptr_t)&h->ip6_dst, AF_INET6, |
2072 | r->dst.neg, NULL)) |
2073 | r = r->skip[PF_SKIP_DST_ADDR].ptr; |
2074 | else { |
2075 | if (r->anchor == NULL) |
2076 | break; |
2077 | else |
2078 | pf_step_into_anchor(&asd, &ruleset, |
2079 | PF_RULESET_SCRUB, &r, NULL, NULL); |
2080 | } |
2081 | if (r == NULL && pf_step_out_of_anchor(&asd, &ruleset, |
2082 | PF_RULESET_SCRUB, &r, NULL, NULL)) |
2083 | break; |
2084 | } |
2085 | |
2086 | if (r == NULL || r->action == PF_NOSCRUB) |
2087 | return (PF_PASS); |
2088 | else { |
2089 | r->packets[dir == PF_OUT]++; |
2090 | r->bytes[dir == PF_OUT] += pd->tot_len; |
2091 | } |
2092 | |
2093 | /* Check for illegal packets */ |
2094 | if ((uint32_t)(sizeof (struct ip6_hdr) + IPV6_MAXPACKET) < |
2095 | pbuf->pb_packet_len) |
2096 | goto drop; |
2097 | |
2098 | off = sizeof (struct ip6_hdr); |
2099 | proto = h->ip6_nxt; |
2100 | terminal = 0; |
2101 | do { |
2102 | pd->proto = proto; |
2103 | switch (proto) { |
2104 | case IPPROTO_FRAGMENT: |
2105 | goto fragment; |
2106 | case IPPROTO_AH: |
2107 | case IPPROTO_ROUTING: |
2108 | case IPPROTO_DSTOPTS: |
2109 | if (!pf_pull_hdr(pbuf, off, &ext, sizeof (ext), NULL, |
2110 | NULL, AF_INET6)) |
2111 | goto shortpkt; |
2112 | /* |
2113 | * <jhw@apple.com> |
2114 | * Multiple routing headers not allowed. |
2115 | * Routing header type zero considered harmful. |
2116 | */ |
2117 | if (proto == IPPROTO_ROUTING) { |
2118 | const struct ip6_rthdr *rh = |
2119 | (const struct ip6_rthdr *)&ext; |
2120 | if (rh_cnt++) |
2121 | goto drop; |
2122 | if (rh->ip6r_type == IPV6_RTHDR_TYPE_0) |
2123 | goto drop; |
2124 | } |
2125 | else |
2126 | if (proto == IPPROTO_AH) |
2127 | off += (ext.ip6e_len + 2) * 4; |
2128 | else |
2129 | off += (ext.ip6e_len + 1) * 8; |
2130 | proto = ext.ip6e_nxt; |
2131 | break; |
2132 | case IPPROTO_HOPOPTS: |
2133 | /* adi XXX */ |
2134 | #if 0 |
2135 | if (!pf_pull_hdr(m, off, &ext, sizeof (ext), NULL, |
2136 | NULL, AF_INET6)) |
2137 | goto shortpkt; |
2138 | optend = off + (ext.ip6e_len + 1) * 8; |
2139 | ooff = off + sizeof (ext); |
2140 | do { |
2141 | if (!pf_pull_hdr(m, ooff, &opt.ip6o_type, |
2142 | sizeof (opt.ip6o_type), NULL, NULL, |
2143 | AF_INET6)) |
2144 | goto shortpkt; |
2145 | if (opt.ip6o_type == IP6OPT_PAD1) { |
2146 | ooff++; |
2147 | continue; |
2148 | } |
2149 | if (!pf_pull_hdr(m, ooff, &opt, sizeof (opt), |
2150 | NULL, NULL, AF_INET6)) |
2151 | goto shortpkt; |
2152 | if (ooff + sizeof (opt) + opt.ip6o_len > optend) |
2153 | goto drop; |
2154 | switch (opt.ip6o_type) { |
2155 | case IP6OPT_JUMBO: |
2156 | if (h->ip6_plen != 0) |
2157 | goto drop; |
2158 | if (!pf_pull_hdr(m, ooff, &jumbo, |
2159 | sizeof (jumbo), NULL, NULL, |
2160 | AF_INET6)) |
2161 | goto shortpkt; |
2162 | memcpy(&jumbolen, jumbo.ip6oj_jumbo_len, |
2163 | sizeof (jumbolen)); |
2164 | jumbolen = ntohl(jumbolen); |
2165 | if (jumbolen <= IPV6_MAXPACKET) |
2166 | goto drop; |
2167 | if (sizeof (struct ip6_hdr) + |
2168 | jumbolen != m->m_pkthdr.len) |
2169 | goto drop; |
2170 | break; |
2171 | default: |
2172 | break; |
2173 | } |
2174 | ooff += sizeof (opt) + opt.ip6o_len; |
2175 | } while (ooff < optend); |
2176 | |
2177 | off = optend; |
2178 | proto = ext.ip6e_nxt; |
2179 | break; |
2180 | #endif |
2181 | default: |
2182 | terminal = 1; |
2183 | break; |
2184 | } |
2185 | } while (!terminal); |
2186 | |
2187 | /* jumbo payload option must be present, or plen > 0 */ |
2188 | if (ntohs(h->ip6_plen) == 0) |
2189 | plen = jumbolen; |
2190 | else |
2191 | plen = ntohs(h->ip6_plen); |
2192 | if (plen == 0) |
2193 | goto drop; |
2194 | if ((uint32_t)(sizeof (struct ip6_hdr) + plen) > pbuf->pb_packet_len) |
2195 | goto shortpkt; |
2196 | |
2197 | /* Enforce a minimum ttl, may cause endless packet loops */ |
2198 | if (r->min_ttl && h->ip6_hlim < r->min_ttl) |
2199 | h->ip6_hlim = r->min_ttl; |
2200 | |
2201 | return (PF_PASS); |
2202 | |
2203 | fragment: |
2204 | if (ntohs(h->ip6_plen) == 0 || jumbolen) |
2205 | goto drop; |
2206 | plen = ntohs(h->ip6_plen); |
2207 | |
2208 | if (!pf_pull_hdr(pbuf, off, &frag, sizeof (frag), NULL, NULL, AF_INET6)) |
2209 | goto shortpkt; |
2210 | fragoff = ntohs(frag.ip6f_offlg & IP6F_OFF_MASK); |
2211 | pd->proto = frag.ip6f_nxt; |
2212 | mff = ntohs(frag.ip6f_offlg & IP6F_MORE_FRAG); |
2213 | off += sizeof frag; |
2214 | if (fragoff + (plen - off) > IPV6_MAXPACKET) |
2215 | goto badfrag; |
2216 | |
2217 | fr_max = fragoff + plen - (off - sizeof(struct ip6_hdr)); |
2218 | // XXX SCW: mbuf-specific |
2219 | // DPFPRINTF(("0x%llx IPv6 frag plen %u mff %d off %u fragoff %u " |
2220 | // "fr_max %u\n", (uint64_t)VM_KERNEL_ADDRPERM(m), plen, mff, off, |
2221 | // fragoff, fr_max)); |
2222 | |
2223 | if ((r->rule_flag & (PFRULE_FRAGCROP|PFRULE_FRAGDROP)) == 0) { |
2224 | /* Fully buffer all of the fragments */ |
2225 | pd->flags |= PFDESC_IP_REAS; |
2226 | |
2227 | pff = pf_find_fragment_by_ipv6_header(h, &frag, |
2228 | &pf_frag_tree); |
2229 | |
2230 | /* Check if we saw the last fragment already */ |
2231 | if (pff != NULL && (pff->fr_flags & PFFRAG_SEENLAST) && |
2232 | fr_max > pff->fr_max) |
2233 | goto badfrag; |
2234 | |
2235 | if ((m = pbuf_to_mbuf(pbuf, TRUE)) == NULL) { |
2236 | REASON_SET(reason, PFRES_MEMORY); |
2237 | return (PF_DROP); |
2238 | } |
2239 | |
2240 | /* Restore iph pointer after pbuf_to_mbuf() */ |
2241 | h = mtod(m, struct ip6_hdr *); |
2242 | |
2243 | /* Get an entry for the fragment queue */ |
2244 | frent = pool_get(&pf_frent_pl, PR_NOWAIT); |
2245 | if (frent == NULL) { |
2246 | REASON_SET(reason, PFRES_MEMORY); |
2247 | return (PF_DROP); |
2248 | } |
2249 | |
2250 | pf_nfrents++; |
2251 | frent->fr_ip6 = h; |
2252 | frent->fr_m = m; |
2253 | frent->fr_ip6f_opt = frag; |
2254 | frent->fr_ip6f_hlen = off; |
2255 | |
2256 | /* Might return a completely reassembled mbuf, or NULL */ |
2257 | DPFPRINTF(("reass IPv6 frag %d @ %d-%d\n" , |
2258 | ntohl(frag.ip6f_ident), fragoff, fr_max)); |
2259 | m = pf_reassemble6(&m, &pff, frent, mff); |
2260 | |
2261 | if (m == NULL) |
2262 | return (PF_DROP); |
2263 | |
2264 | pbuf_init_mbuf(pbuf, m, ifp); |
2265 | h = pbuf->pb_data; |
2266 | |
2267 | if (pff != NULL && (pff->fr_flags & PFFRAG_DROP)) |
2268 | goto drop; |
2269 | } |
2270 | else if (dir == PF_IN || !(pd->pf_mtag->pftag_flags & PF_TAG_FRAGCACHE)) { |
2271 | /* non-buffering fragment cache (overlaps: see RFC 5722) */ |
2272 | int nomem = 0; |
2273 | |
2274 | pff = pf_find_fragment_by_ipv6_header(h, &frag, |
2275 | &pf_cache_tree); |
2276 | |
2277 | /* Check if we saw the last fragment already */ |
2278 | if (pff != NULL && (pff->fr_flags & PFFRAG_SEENLAST) && |
2279 | fr_max > pff->fr_max) { |
2280 | if (r->rule_flag & PFRULE_FRAGDROP) |
2281 | pff->fr_flags |= PFFRAG_DROP; |
2282 | goto badfrag; |
2283 | } |
2284 | |
2285 | if ((m = pbuf_to_mbuf(pbuf, TRUE)) == NULL) { |
2286 | goto no_mem; |
2287 | } |
2288 | |
2289 | /* Restore iph pointer after pbuf_to_mbuf() */ |
2290 | h = mtod(m, struct ip6_hdr *); |
2291 | |
2292 | m = pf_frag6cache(&m, h, &frag, &pff, off, mff, |
2293 | (r->rule_flag & PFRULE_FRAGDROP) ? 1 : 0, &nomem); |
2294 | if (m == NULL) { |
2295 | // Note: pf_frag6cache() has already m_freem'd the mbuf |
2296 | if (nomem) |
2297 | goto no_mem; |
2298 | goto drop; |
2299 | } |
2300 | |
2301 | pbuf_init_mbuf(pbuf, m, ifp); |
2302 | pd->pf_mtag = pf_find_mtag_pbuf(pbuf); |
2303 | h = pbuf->pb_data; |
2304 | |
2305 | if (dir == PF_IN) |
2306 | pd->pf_mtag->pftag_flags |= PF_TAG_FRAGCACHE; |
2307 | |
2308 | if (pff != NULL && (pff->fr_flags & PFFRAG_DROP)) |
2309 | goto drop; |
2310 | } |
2311 | |
2312 | /* Enforce a minimum ttl, may cause endless packet loops */ |
2313 | if (r->min_ttl && h->ip6_hlim < r->min_ttl) |
2314 | h->ip6_hlim = r->min_ttl; |
2315 | return (PF_PASS); |
2316 | |
2317 | no_mem: |
2318 | REASON_SET(reason, PFRES_MEMORY); |
2319 | goto dropout; |
2320 | |
2321 | shortpkt: |
2322 | REASON_SET(reason, PFRES_SHORT); |
2323 | goto dropout; |
2324 | |
2325 | drop: |
2326 | REASON_SET(reason, PFRES_NORM); |
2327 | goto dropout; |
2328 | |
2329 | badfrag: |
2330 | DPFPRINTF(("dropping bad IPv6 fragment\n" )); |
2331 | REASON_SET(reason, PFRES_FRAG); |
2332 | goto dropout; |
2333 | |
2334 | dropout: |
2335 | if (pff != NULL) |
2336 | pf_free_fragment(pff); |
2337 | if (r != NULL && r->log && pbuf_is_valid(pbuf)) |
2338 | PFLOG_PACKET(kif, h, pbuf, AF_INET6, dir, *reason, r, NULL, NULL, pd); |
2339 | return (PF_DROP); |
2340 | } |
2341 | #endif /* INET6 */ |
2342 | |
2343 | int |
2344 | pf_normalize_tcp(int dir, struct pfi_kif *kif, pbuf_t *pbuf, int ipoff, |
2345 | int off, void *h, struct pf_pdesc *pd) |
2346 | { |
2347 | #pragma unused(ipoff, h) |
2348 | struct pf_rule *r, *rm = NULL; |
2349 | struct tcphdr *th = pd->hdr.tcp; |
2350 | int rewrite = 0; |
2351 | int asd = 0; |
2352 | u_short reason; |
2353 | u_int8_t flags; |
2354 | sa_family_t af = pd->af; |
2355 | struct pf_ruleset *ruleset = NULL; |
2356 | union pf_state_xport sxport, dxport; |
2357 | |
2358 | sxport.port = th->th_sport; |
2359 | dxport.port = th->th_dport; |
2360 | |
2361 | r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr); |
2362 | while (r != NULL) { |
2363 | r->evaluations++; |
2364 | if (pfi_kif_match(r->kif, kif) == r->ifnot) |
2365 | r = r->skip[PF_SKIP_IFP].ptr; |
2366 | else if (r->direction && r->direction != dir) |
2367 | r = r->skip[PF_SKIP_DIR].ptr; |
2368 | else if (r->af && r->af != af) |
2369 | r = r->skip[PF_SKIP_AF].ptr; |
2370 | else if (r->proto && r->proto != pd->proto) |
2371 | r = r->skip[PF_SKIP_PROTO].ptr; |
2372 | else if (PF_MISMATCHAW(&r->src.addr, pd->src, af, |
2373 | r->src.neg, kif)) |
2374 | r = r->skip[PF_SKIP_SRC_ADDR].ptr; |
2375 | else if (r->src.xport.range.op && |
2376 | !pf_match_xport(r->src.xport.range.op, r->proto_variant, |
2377 | &r->src.xport, &sxport)) |
2378 | r = r->skip[PF_SKIP_SRC_PORT].ptr; |
2379 | else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af, |
2380 | r->dst.neg, NULL)) |
2381 | r = r->skip[PF_SKIP_DST_ADDR].ptr; |
2382 | else if (r->dst.xport.range.op && |
2383 | !pf_match_xport(r->dst.xport.range.op, r->proto_variant, |
2384 | &r->dst.xport, &dxport)) |
2385 | r = r->skip[PF_SKIP_DST_PORT].ptr; |
2386 | else if (r->os_fingerprint != PF_OSFP_ANY && |
2387 | !pf_osfp_match(pf_osfp_fingerprint(pd, pbuf, off, th), |
2388 | r->os_fingerprint)) |
2389 | r = TAILQ_NEXT(r, entries); |
2390 | else { |
2391 | if (r->anchor == NULL) { |
2392 | rm = r; |
2393 | break; |
2394 | } else { |
2395 | pf_step_into_anchor(&asd, &ruleset, |
2396 | PF_RULESET_SCRUB, &r, NULL, NULL); |
2397 | } |
2398 | } |
2399 | if (r == NULL && pf_step_out_of_anchor(&asd, &ruleset, |
2400 | PF_RULESET_SCRUB, &r, NULL, NULL)) |
2401 | break; |
2402 | } |
2403 | |
2404 | if (rm == NULL || rm->action == PF_NOSCRUB) |
2405 | return (PF_PASS); |
2406 | else { |
2407 | r->packets[dir == PF_OUT]++; |
2408 | r->bytes[dir == PF_OUT] += pd->tot_len; |
2409 | } |
2410 | |
2411 | if (rm->rule_flag & PFRULE_REASSEMBLE_TCP) |
2412 | pd->flags |= PFDESC_TCP_NORM; |
2413 | |
2414 | flags = th->th_flags; |
2415 | if (flags & TH_SYN) { |
2416 | /* Illegal packet */ |
2417 | if (flags & TH_RST) |
2418 | goto tcp_drop; |
2419 | |
2420 | if (flags & TH_FIN) |
2421 | flags &= ~TH_FIN; |
2422 | } else { |
2423 | /* Illegal packet */ |
2424 | if (!(flags & (TH_ACK|TH_RST))) |
2425 | goto tcp_drop; |
2426 | } |
2427 | |
2428 | if (!(flags & TH_ACK)) { |
2429 | /* These flags are only valid if ACK is set */ |
2430 | if ((flags & TH_FIN) || (flags & TH_PUSH) || (flags & TH_URG)) |
2431 | goto tcp_drop; |
2432 | } |
2433 | |
2434 | /* Check for illegal header length */ |
2435 | if (th->th_off < (sizeof (struct tcphdr) >> 2)) |
2436 | goto tcp_drop; |
2437 | |
2438 | /* If flags changed, or reserved data set, then adjust */ |
2439 | if (flags != th->th_flags || th->th_x2 != 0) { |
2440 | u_int16_t ov, nv; |
2441 | |
2442 | ov = *(u_int16_t *)(&th->th_ack + 1); |
2443 | th->th_flags = flags; |
2444 | th->th_x2 = 0; |
2445 | nv = *(u_int16_t *)(&th->th_ack + 1); |
2446 | |
2447 | th->th_sum = pf_cksum_fixup(th->th_sum, ov, nv, 0); |
2448 | rewrite = 1; |
2449 | } |
2450 | |
2451 | /* Remove urgent pointer, if TH_URG is not set */ |
2452 | if (!(flags & TH_URG) && th->th_urp) { |
2453 | th->th_sum = pf_cksum_fixup(th->th_sum, th->th_urp, 0, 0); |
2454 | th->th_urp = 0; |
2455 | rewrite = 1; |
2456 | } |
2457 | |
2458 | /* copy back packet headers if we sanitized */ |
2459 | /* Process options */ |
2460 | if (r->max_mss) { |
2461 | int rv = pf_normalize_tcpopt(r, dir, kif, pd, pbuf, th, off, |
2462 | &rewrite); |
2463 | if (rv == PF_DROP) |
2464 | return rv; |
2465 | pbuf = pd->mp; |
2466 | } |
2467 | |
2468 | if (rewrite) { |
2469 | if (pf_lazy_makewritable(pd, pbuf, |
2470 | off + sizeof (*th)) == NULL) { |
2471 | REASON_SET(&reason, PFRES_MEMORY); |
2472 | if (r->log) |
2473 | PFLOG_PACKET(kif, h, pbuf, AF_INET, dir, reason, |
2474 | r, 0, 0, pd); |
2475 | return PF_DROP; |
2476 | } |
2477 | |
2478 | pbuf_copy_back(pbuf, off, sizeof (*th), th); |
2479 | } |
2480 | |
2481 | return (PF_PASS); |
2482 | |
2483 | tcp_drop: |
2484 | REASON_SET(&reason, PFRES_NORM); |
2485 | if (rm != NULL && r->log) |
2486 | PFLOG_PACKET(kif, h, pbuf, AF_INET, dir, reason, r, NULL, NULL, pd); |
2487 | return (PF_DROP); |
2488 | } |
2489 | |
2490 | int |
2491 | pf_normalize_tcp_init(pbuf_t *pbuf, int off, struct pf_pdesc *pd, |
2492 | struct tcphdr *th, struct pf_state_peer *src, struct pf_state_peer *dst) |
2493 | { |
2494 | #pragma unused(dst) |
2495 | u_int32_t tsval, tsecr; |
2496 | u_int8_t hdr[60]; |
2497 | u_int8_t *opt; |
2498 | |
2499 | VERIFY(src->scrub == NULL); |
2500 | |
2501 | src->scrub = pool_get(&pf_state_scrub_pl, PR_NOWAIT); |
2502 | if (src->scrub == NULL) |
2503 | return (1); |
2504 | bzero(src->scrub, sizeof (*src->scrub)); |
2505 | |
2506 | switch (pd->af) { |
2507 | #if INET |
2508 | case AF_INET: { |
2509 | struct ip *h = pbuf->pb_data; |
2510 | src->scrub->pfss_ttl = h->ip_ttl; |
2511 | break; |
2512 | } |
2513 | #endif /* INET */ |
2514 | #if INET6 |
2515 | case AF_INET6: { |
2516 | struct ip6_hdr *h = pbuf->pb_data; |
2517 | src->scrub->pfss_ttl = h->ip6_hlim; |
2518 | break; |
2519 | } |
2520 | #endif /* INET6 */ |
2521 | } |
2522 | |
2523 | |
2524 | /* |
2525 | * All normalizations below are only begun if we see the start of |
2526 | * the connections. They must all set an enabled bit in pfss_flags |
2527 | */ |
2528 | if ((th->th_flags & TH_SYN) == 0) |
2529 | return (0); |
2530 | |
2531 | |
2532 | if (th->th_off > (sizeof (struct tcphdr) >> 2) && src->scrub && |
2533 | pf_pull_hdr(pbuf, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) { |
2534 | /* Diddle with TCP options */ |
2535 | int hlen; |
2536 | opt = hdr + sizeof (struct tcphdr); |
2537 | hlen = (th->th_off << 2) - sizeof (struct tcphdr); |
2538 | while (hlen >= TCPOLEN_TIMESTAMP) { |
2539 | switch (*opt) { |
2540 | case TCPOPT_EOL: /* FALLTHROUGH */ |
2541 | case TCPOPT_NOP: |
2542 | opt++; |
2543 | hlen--; |
2544 | break; |
2545 | case TCPOPT_TIMESTAMP: |
2546 | if (opt[1] >= TCPOLEN_TIMESTAMP) { |
2547 | src->scrub->pfss_flags |= |
2548 | PFSS_TIMESTAMP; |
2549 | src->scrub->pfss_ts_mod = |
2550 | htonl(random()); |
2551 | |
2552 | /* note PFSS_PAWS not set yet */ |
2553 | memcpy(&tsval, &opt[2], |
2554 | sizeof (u_int32_t)); |
2555 | memcpy(&tsecr, &opt[6], |
2556 | sizeof (u_int32_t)); |
2557 | src->scrub->pfss_tsval0 = ntohl(tsval); |
2558 | src->scrub->pfss_tsval = ntohl(tsval); |
2559 | src->scrub->pfss_tsecr = ntohl(tsecr); |
2560 | getmicrouptime(&src->scrub->pfss_last); |
2561 | } |
2562 | /* FALLTHROUGH */ |
2563 | default: |
2564 | hlen -= MAX(opt[1], 2); |
2565 | opt += MAX(opt[1], 2); |
2566 | break; |
2567 | } |
2568 | } |
2569 | } |
2570 | |
2571 | return (0); |
2572 | } |
2573 | |
2574 | void |
2575 | pf_normalize_tcp_cleanup(struct pf_state *state) |
2576 | { |
2577 | if (state->src.scrub) |
2578 | pool_put(&pf_state_scrub_pl, state->src.scrub); |
2579 | if (state->dst.scrub) |
2580 | pool_put(&pf_state_scrub_pl, state->dst.scrub); |
2581 | |
2582 | /* Someday... flush the TCP segment reassembly descriptors. */ |
2583 | } |
2584 | |
2585 | int |
2586 | pf_normalize_tcp_stateful(pbuf_t *pbuf, int off, struct pf_pdesc *pd, |
2587 | u_short *reason, struct tcphdr *th, struct pf_state *state, |
2588 | struct pf_state_peer *src, struct pf_state_peer *dst, int *writeback) |
2589 | { |
2590 | struct timeval uptime; |
2591 | u_int32_t tsval = 0, tsecr = 0; |
2592 | u_int tsval_from_last; |
2593 | u_int8_t hdr[60]; |
2594 | u_int8_t *opt; |
2595 | int copyback = 0; |
2596 | int got_ts = 0; |
2597 | |
2598 | VERIFY(src->scrub || dst->scrub); |
2599 | |
2600 | /* |
2601 | * Enforce the minimum TTL seen for this connection. Negate a common |
2602 | * technique to evade an intrusion detection system and confuse |
2603 | * firewall state code. |
2604 | */ |
2605 | switch (pd->af) { |
2606 | #if INET |
2607 | case AF_INET: { |
2608 | if (src->scrub) { |
2609 | struct ip *h = pbuf->pb_data; |
2610 | if (h->ip_ttl > src->scrub->pfss_ttl) |
2611 | src->scrub->pfss_ttl = h->ip_ttl; |
2612 | h->ip_ttl = src->scrub->pfss_ttl; |
2613 | } |
2614 | break; |
2615 | } |
2616 | #endif /* INET */ |
2617 | #if INET6 |
2618 | case AF_INET6: { |
2619 | if (src->scrub) { |
2620 | struct ip6_hdr *h = pbuf->pb_data; |
2621 | if (h->ip6_hlim > src->scrub->pfss_ttl) |
2622 | src->scrub->pfss_ttl = h->ip6_hlim; |
2623 | h->ip6_hlim = src->scrub->pfss_ttl; |
2624 | } |
2625 | break; |
2626 | } |
2627 | #endif /* INET6 */ |
2628 | } |
2629 | |
2630 | if (th->th_off > (sizeof (struct tcphdr) >> 2) && |
2631 | ((src->scrub && (src->scrub->pfss_flags & PFSS_TIMESTAMP)) || |
2632 | (dst->scrub && (dst->scrub->pfss_flags & PFSS_TIMESTAMP))) && |
2633 | pf_pull_hdr(pbuf, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) { |
2634 | /* Diddle with TCP options */ |
2635 | int hlen; |
2636 | opt = hdr + sizeof (struct tcphdr); |
2637 | hlen = (th->th_off << 2) - sizeof (struct tcphdr); |
2638 | while (hlen >= TCPOLEN_TIMESTAMP) { |
2639 | switch (*opt) { |
2640 | case TCPOPT_EOL: /* FALLTHROUGH */ |
2641 | case TCPOPT_NOP: |
2642 | opt++; |
2643 | hlen--; |
2644 | break; |
2645 | case TCPOPT_TIMESTAMP: |
2646 | /* |
2647 | * Modulate the timestamps. Can be used for |
2648 | * NAT detection, OS uptime determination or |
2649 | * reboot detection. |
2650 | */ |
2651 | |
2652 | if (got_ts) { |
2653 | /* Huh? Multiple timestamps!? */ |
2654 | if (pf_status.debug >= PF_DEBUG_MISC) { |
2655 | DPFPRINTF(("multiple TS??" )); |
2656 | pf_print_state(state); |
2657 | printf("\n" ); |
2658 | } |
2659 | REASON_SET(reason, PFRES_TS); |
2660 | return (PF_DROP); |
2661 | } |
2662 | if (opt[1] >= TCPOLEN_TIMESTAMP) { |
2663 | memcpy(&tsval, &opt[2], |
2664 | sizeof (u_int32_t)); |
2665 | if (tsval && src->scrub && |
2666 | (src->scrub->pfss_flags & |
2667 | PFSS_TIMESTAMP)) { |
2668 | tsval = ntohl(tsval); |
2669 | pf_change_a(&opt[2], |
2670 | &th->th_sum, |
2671 | htonl(tsval + |
2672 | src->scrub->pfss_ts_mod), |
2673 | 0); |
2674 | copyback = 1; |
2675 | } |
2676 | |
2677 | /* Modulate TS reply iff valid (!0) */ |
2678 | memcpy(&tsecr, &opt[6], |
2679 | sizeof (u_int32_t)); |
2680 | if (tsecr && dst->scrub && |
2681 | (dst->scrub->pfss_flags & |
2682 | PFSS_TIMESTAMP)) { |
2683 | tsecr = ntohl(tsecr) |
2684 | - dst->scrub->pfss_ts_mod; |
2685 | pf_change_a(&opt[6], |
2686 | &th->th_sum, htonl(tsecr), |
2687 | 0); |
2688 | copyback = 1; |
2689 | } |
2690 | got_ts = 1; |
2691 | } |
2692 | /* FALLTHROUGH */ |
2693 | default: |
2694 | hlen -= MAX(opt[1], 2); |
2695 | opt += MAX(opt[1], 2); |
2696 | break; |
2697 | } |
2698 | } |
2699 | if (copyback) { |
2700 | /* Copyback the options, caller copys back header */ |
2701 | int optoff = off + sizeof (*th); |
2702 | int optlen = (th->th_off << 2) - sizeof (*th); |
2703 | if (pf_lazy_makewritable(pd, pbuf, optoff + optlen) == |
2704 | NULL) { |
2705 | REASON_SET(reason, PFRES_MEMORY); |
2706 | return PF_DROP; |
2707 | } |
2708 | *writeback = optoff + optlen; |
2709 | pbuf_copy_back(pbuf, optoff, optlen, hdr + sizeof(*th)); |
2710 | } |
2711 | } |
2712 | |
2713 | |
2714 | /* |
2715 | * Must invalidate PAWS checks on connections idle for too long. |
2716 | * The fastest allowed timestamp clock is 1ms. That turns out to |
2717 | * be about 24 days before it wraps. XXX Right now our lowerbound |
2718 | * TS echo check only works for the first 12 days of a connection |
2719 | * when the TS has exhausted half its 32bit space |
2720 | */ |
2721 | #define TS_MAX_IDLE (24*24*60*60) |
2722 | #define TS_MAX_CONN (12*24*60*60) /* XXX remove when better tsecr check */ |
2723 | |
2724 | getmicrouptime(&uptime); |
2725 | if (src->scrub && (src->scrub->pfss_flags & PFSS_PAWS) && |
2726 | (uptime.tv_sec - src->scrub->pfss_last.tv_sec > TS_MAX_IDLE || |
2727 | pf_time_second() - state->creation > TS_MAX_CONN)) { |
2728 | if (pf_status.debug >= PF_DEBUG_MISC) { |
2729 | DPFPRINTF(("src idled out of PAWS\n" )); |
2730 | pf_print_state(state); |
2731 | printf("\n" ); |
2732 | } |
2733 | src->scrub->pfss_flags = (src->scrub->pfss_flags & ~PFSS_PAWS) |
2734 | | PFSS_PAWS_IDLED; |
2735 | } |
2736 | if (dst->scrub && (dst->scrub->pfss_flags & PFSS_PAWS) && |
2737 | uptime.tv_sec - dst->scrub->pfss_last.tv_sec > TS_MAX_IDLE) { |
2738 | if (pf_status.debug >= PF_DEBUG_MISC) { |
2739 | DPFPRINTF(("dst idled out of PAWS\n" )); |
2740 | pf_print_state(state); |
2741 | printf("\n" ); |
2742 | } |
2743 | dst->scrub->pfss_flags = (dst->scrub->pfss_flags & ~PFSS_PAWS) |
2744 | | PFSS_PAWS_IDLED; |
2745 | } |
2746 | |
2747 | if (got_ts && src->scrub && dst->scrub && |
2748 | (src->scrub->pfss_flags & PFSS_PAWS) && |
2749 | (dst->scrub->pfss_flags & PFSS_PAWS)) { |
2750 | /* |
2751 | * Validate that the timestamps are "in-window". |
2752 | * RFC1323 describes TCP Timestamp options that allow |
2753 | * measurement of RTT (round trip time) and PAWS |
2754 | * (protection against wrapped sequence numbers). PAWS |
2755 | * gives us a set of rules for rejecting packets on |
2756 | * long fat pipes (packets that were somehow delayed |
2757 | * in transit longer than the time it took to send the |
2758 | * full TCP sequence space of 4Gb). We can use these |
2759 | * rules and infer a few others that will let us treat |
2760 | * the 32bit timestamp and the 32bit echoed timestamp |
2761 | * as sequence numbers to prevent a blind attacker from |
2762 | * inserting packets into a connection. |
2763 | * |
2764 | * RFC1323 tells us: |
2765 | * - The timestamp on this packet must be greater than |
2766 | * or equal to the last value echoed by the other |
2767 | * endpoint. The RFC says those will be discarded |
2768 | * since it is a dup that has already been acked. |
2769 | * This gives us a lowerbound on the timestamp. |
2770 | * timestamp >= other last echoed timestamp |
2771 | * - The timestamp will be less than or equal to |
2772 | * the last timestamp plus the time between the |
2773 | * last packet and now. The RFC defines the max |
2774 | * clock rate as 1ms. We will allow clocks to be |
2775 | * up to 10% fast and will allow a total difference |
2776 | * or 30 seconds due to a route change. And this |
2777 | * gives us an upperbound on the timestamp. |
2778 | * timestamp <= last timestamp + max ticks |
2779 | * We have to be careful here. Windows will send an |
2780 | * initial timestamp of zero and then initialize it |
2781 | * to a random value after the 3whs; presumably to |
2782 | * avoid a DoS by having to call an expensive RNG |
2783 | * during a SYN flood. Proof MS has at least one |
2784 | * good security geek. |
2785 | * |
2786 | * - The TCP timestamp option must also echo the other |
2787 | * endpoints timestamp. The timestamp echoed is the |
2788 | * one carried on the earliest unacknowledged segment |
2789 | * on the left edge of the sequence window. The RFC |
2790 | * states that the host will reject any echoed |
2791 | * timestamps that were larger than any ever sent. |
2792 | * This gives us an upperbound on the TS echo. |
2793 | * tescr <= largest_tsval |
2794 | * - The lowerbound on the TS echo is a little more |
2795 | * tricky to determine. The other endpoint's echoed |
2796 | * values will not decrease. But there may be |
2797 | * network conditions that re-order packets and |
2798 | * cause our view of them to decrease. For now the |
2799 | * only lowerbound we can safely determine is that |
2800 | * the TS echo will never be less than the original |
2801 | * TS. XXX There is probably a better lowerbound. |
2802 | * Remove TS_MAX_CONN with better lowerbound check. |
2803 | * tescr >= other original TS |
2804 | * |
2805 | * It is also important to note that the fastest |
2806 | * timestamp clock of 1ms will wrap its 32bit space in |
2807 | * 24 days. So we just disable TS checking after 24 |
2808 | * days of idle time. We actually must use a 12d |
2809 | * connection limit until we can come up with a better |
2810 | * lowerbound to the TS echo check. |
2811 | */ |
2812 | struct timeval delta_ts; |
2813 | int ts_fudge; |
2814 | |
2815 | |
2816 | /* |
2817 | * PFTM_TS_DIFF is how many seconds of leeway to allow |
2818 | * a host's timestamp. This can happen if the previous |
2819 | * packet got delayed in transit for much longer than |
2820 | * this packet. |
2821 | */ |
2822 | if ((ts_fudge = state->rule.ptr->timeout[PFTM_TS_DIFF]) == 0) |
2823 | ts_fudge = pf_default_rule.timeout[PFTM_TS_DIFF]; |
2824 | |
2825 | |
2826 | /* Calculate max ticks since the last timestamp */ |
2827 | #define TS_MAXFREQ 1100 /* RFC max TS freq of 1Khz + 10% skew */ |
2828 | #define TS_MICROSECS 1000000 /* microseconds per second */ |
2829 | timersub(&uptime, &src->scrub->pfss_last, &delta_ts); |
2830 | tsval_from_last = (delta_ts.tv_sec + ts_fudge) * TS_MAXFREQ; |
2831 | tsval_from_last += delta_ts.tv_usec / (TS_MICROSECS/TS_MAXFREQ); |
2832 | |
2833 | |
2834 | if ((src->state >= TCPS_ESTABLISHED && |
2835 | dst->state >= TCPS_ESTABLISHED) && |
2836 | (SEQ_LT(tsval, dst->scrub->pfss_tsecr) || |
2837 | SEQ_GT(tsval, src->scrub->pfss_tsval + tsval_from_last) || |
2838 | (tsecr && (SEQ_GT(tsecr, dst->scrub->pfss_tsval) || |
2839 | SEQ_LT(tsecr, dst->scrub->pfss_tsval0))))) { |
2840 | /* |
2841 | * Bad RFC1323 implementation or an insertion attack. |
2842 | * |
2843 | * - Solaris 2.6 and 2.7 are known to send another ACK |
2844 | * after the FIN,FIN|ACK,ACK closing that carries |
2845 | * an old timestamp. |
2846 | */ |
2847 | |
2848 | DPFPRINTF(("Timestamp failed %c%c%c%c\n" , |
2849 | SEQ_LT(tsval, dst->scrub->pfss_tsecr) ? '0' : ' ', |
2850 | SEQ_GT(tsval, src->scrub->pfss_tsval + |
2851 | tsval_from_last) ? '1' : ' ', |
2852 | SEQ_GT(tsecr, dst->scrub->pfss_tsval) ? '2' : ' ', |
2853 | SEQ_LT(tsecr, dst->scrub->pfss_tsval0)? '3' : ' ')); |
2854 | DPFPRINTF((" tsval: %u tsecr: %u +ticks: %u " |
2855 | "idle: %lus %ums\n" , |
2856 | tsval, tsecr, tsval_from_last, delta_ts.tv_sec, |
2857 | delta_ts.tv_usec / 1000)); |
2858 | DPFPRINTF((" src->tsval: %u tsecr: %u\n" , |
2859 | src->scrub->pfss_tsval, src->scrub->pfss_tsecr)); |
2860 | DPFPRINTF((" dst->tsval: %u tsecr: %u tsval0: %u\n" , |
2861 | dst->scrub->pfss_tsval, dst->scrub->pfss_tsecr, |
2862 | dst->scrub->pfss_tsval0)); |
2863 | if (pf_status.debug >= PF_DEBUG_MISC) { |
2864 | pf_print_state(state); |
2865 | pf_print_flags(th->th_flags); |
2866 | printf("\n" ); |
2867 | } |
2868 | REASON_SET(reason, PFRES_TS); |
2869 | return (PF_DROP); |
2870 | } |
2871 | |
2872 | /* XXX I'd really like to require tsecr but it's optional */ |
2873 | |
2874 | } else if (!got_ts && (th->th_flags & TH_RST) == 0 && |
2875 | ((src->state == TCPS_ESTABLISHED && dst->state == TCPS_ESTABLISHED) |
2876 | || pd->p_len > 0 || (th->th_flags & TH_SYN)) && |
2877 | src->scrub && dst->scrub && |
2878 | (src->scrub->pfss_flags & PFSS_PAWS) && |
2879 | (dst->scrub->pfss_flags & PFSS_PAWS)) { |
2880 | /* |
2881 | * Didn't send a timestamp. Timestamps aren't really useful |
2882 | * when: |
2883 | * - connection opening or closing (often not even sent). |
2884 | * but we must not let an attacker to put a FIN on a |
2885 | * data packet to sneak it through our ESTABLISHED check. |
2886 | * - on a TCP reset. RFC suggests not even looking at TS. |
2887 | * - on an empty ACK. The TS will not be echoed so it will |
2888 | * probably not help keep the RTT calculation in sync and |
2889 | * there isn't as much danger when the sequence numbers |
2890 | * got wrapped. So some stacks don't include TS on empty |
2891 | * ACKs :-( |
2892 | * |
2893 | * To minimize the disruption to mostly RFC1323 conformant |
2894 | * stacks, we will only require timestamps on data packets. |
2895 | * |
2896 | * And what do ya know, we cannot require timestamps on data |
2897 | * packets. There appear to be devices that do legitimate |
2898 | * TCP connection hijacking. There are HTTP devices that allow |
2899 | * a 3whs (with timestamps) and then buffer the HTTP request. |
2900 | * If the intermediate device has the HTTP response cache, it |
2901 | * will spoof the response but not bother timestamping its |
2902 | * packets. So we can look for the presence of a timestamp in |
2903 | * the first data packet and if there, require it in all future |
2904 | * packets. |
2905 | */ |
2906 | |
2907 | if (pd->p_len > 0 && (src->scrub->pfss_flags & PFSS_DATA_TS)) { |
2908 | /* |
2909 | * Hey! Someone tried to sneak a packet in. Or the |
2910 | * stack changed its RFC1323 behavior?!?! |
2911 | */ |
2912 | if (pf_status.debug >= PF_DEBUG_MISC) { |
2913 | DPFPRINTF(("Did not receive expected RFC1323 " |
2914 | "timestamp\n" )); |
2915 | pf_print_state(state); |
2916 | pf_print_flags(th->th_flags); |
2917 | printf("\n" ); |
2918 | } |
2919 | REASON_SET(reason, PFRES_TS); |
2920 | return (PF_DROP); |
2921 | } |
2922 | } |
2923 | |
2924 | |
2925 | /* |
2926 | * We will note if a host sends his data packets with or without |
2927 | * timestamps. And require all data packets to contain a timestamp |
2928 | * if the first does. PAWS implicitly requires that all data packets be |
2929 | * timestamped. But I think there are middle-man devices that hijack |
2930 | * TCP streams immediately after the 3whs and don't timestamp their |
2931 | * packets (seen in a WWW accelerator or cache). |
2932 | */ |
2933 | if (pd->p_len > 0 && src->scrub && (src->scrub->pfss_flags & |
2934 | (PFSS_TIMESTAMP|PFSS_DATA_TS|PFSS_DATA_NOTS)) == PFSS_TIMESTAMP) { |
2935 | if (got_ts) |
2936 | src->scrub->pfss_flags |= PFSS_DATA_TS; |
2937 | else { |
2938 | src->scrub->pfss_flags |= PFSS_DATA_NOTS; |
2939 | if (pf_status.debug >= PF_DEBUG_MISC && dst->scrub && |
2940 | (dst->scrub->pfss_flags & PFSS_TIMESTAMP)) { |
2941 | /* Don't warn if other host rejected RFC1323 */ |
2942 | DPFPRINTF(("Broken RFC1323 stack did not " |
2943 | "timestamp data packet. Disabled PAWS " |
2944 | "security.\n" )); |
2945 | pf_print_state(state); |
2946 | pf_print_flags(th->th_flags); |
2947 | printf("\n" ); |
2948 | } |
2949 | } |
2950 | } |
2951 | |
2952 | |
2953 | /* |
2954 | * Update PAWS values |
2955 | */ |
2956 | if (got_ts && src->scrub && PFSS_TIMESTAMP == (src->scrub->pfss_flags & |
2957 | (PFSS_PAWS_IDLED|PFSS_TIMESTAMP))) { |
2958 | getmicrouptime(&src->scrub->pfss_last); |
2959 | if (SEQ_GEQ(tsval, src->scrub->pfss_tsval) || |
2960 | (src->scrub->pfss_flags & PFSS_PAWS) == 0) |
2961 | src->scrub->pfss_tsval = tsval; |
2962 | |
2963 | if (tsecr) { |
2964 | if (SEQ_GEQ(tsecr, src->scrub->pfss_tsecr) || |
2965 | (src->scrub->pfss_flags & PFSS_PAWS) == 0) |
2966 | src->scrub->pfss_tsecr = tsecr; |
2967 | |
2968 | if ((src->scrub->pfss_flags & PFSS_PAWS) == 0 && |
2969 | (SEQ_LT(tsval, src->scrub->pfss_tsval0) || |
2970 | src->scrub->pfss_tsval0 == 0)) { |
2971 | /* tsval0 MUST be the lowest timestamp */ |
2972 | src->scrub->pfss_tsval0 = tsval; |
2973 | } |
2974 | |
2975 | /* Only fully initialized after a TS gets echoed */ |
2976 | if ((src->scrub->pfss_flags & PFSS_PAWS) == 0) |
2977 | src->scrub->pfss_flags |= PFSS_PAWS; |
2978 | } |
2979 | } |
2980 | |
2981 | /* I have a dream.... TCP segment reassembly.... */ |
2982 | return (0); |
2983 | } |
2984 | |
2985 | static int |
2986 | pf_normalize_tcpopt(struct pf_rule *r, int dir, struct pfi_kif *kif, |
2987 | struct pf_pdesc *pd, pbuf_t *pbuf, struct tcphdr *th, int off, |
2988 | int *rewrptr) |
2989 | { |
2990 | #pragma unused(dir, kif) |
2991 | sa_family_t af = pd->af; |
2992 | u_int16_t *mss; |
2993 | int thoff; |
2994 | int opt, cnt, optlen = 0; |
2995 | int rewrite = 0; |
2996 | u_char opts[MAX_TCPOPTLEN]; |
2997 | u_char *optp = opts; |
2998 | |
2999 | thoff = th->th_off << 2; |
3000 | cnt = thoff - sizeof (struct tcphdr); |
3001 | |
3002 | if (cnt > 0 && !pf_pull_hdr(pbuf, off + sizeof (*th), opts, cnt, |
3003 | NULL, NULL, af)) |
3004 | return PF_DROP; |
3005 | |
3006 | for (; cnt > 0; cnt -= optlen, optp += optlen) { |
3007 | opt = optp[0]; |
3008 | if (opt == TCPOPT_EOL) |
3009 | break; |
3010 | if (opt == TCPOPT_NOP) |
3011 | optlen = 1; |
3012 | else { |
3013 | if (cnt < 2) |
3014 | break; |
3015 | optlen = optp[1]; |
3016 | if (optlen < 2 || optlen > cnt) |
3017 | break; |
3018 | } |
3019 | switch (opt) { |
3020 | case TCPOPT_MAXSEG: |
3021 | mss = (u_int16_t *)(void *)(optp + 2); |
3022 | if ((ntohs(*mss)) > r->max_mss) { |
3023 | /* |
3024 | * <jhw@apple.com> |
3025 | * Only do the TCP checksum fixup if delayed |
3026 | * checksum calculation will not be performed. |
3027 | */ |
3028 | if (pbuf->pb_ifp || |
3029 | !(*pbuf->pb_csum_flags & CSUM_TCP)) |
3030 | th->th_sum = pf_cksum_fixup(th->th_sum, |
3031 | *mss, htons(r->max_mss), 0); |
3032 | *mss = htons(r->max_mss); |
3033 | rewrite = 1; |
3034 | } |
3035 | break; |
3036 | default: |
3037 | break; |
3038 | } |
3039 | } |
3040 | |
3041 | if (rewrite) { |
3042 | u_short reason; |
3043 | |
3044 | VERIFY(pbuf == pd->mp); |
3045 | |
3046 | if (pf_lazy_makewritable(pd, pd->mp, |
3047 | off + sizeof (*th) + thoff) == NULL) { |
3048 | REASON_SET(&reason, PFRES_MEMORY); |
3049 | if (r->log) |
3050 | PFLOG_PACKET(kif, h, pbuf, AF_INET, dir, reason, |
3051 | r, 0, 0, pd); |
3052 | return PF_DROP; |
3053 | } |
3054 | |
3055 | *rewrptr = 1; |
3056 | pbuf_copy_back(pd->mp, off + sizeof (*th), thoff - sizeof (*th), opts); |
3057 | } |
3058 | |
3059 | return PF_PASS; |
3060 | } |
3061 | |