in_proto.c revision 1.101 1 /* $NetBSD: in_proto.c,v 1.101 2011/05/03 17:44:31 dyoung Exp $ */
2
3 /*
4 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of the project nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * SUCH DAMAGE.
30 */
31
32 /*
33 * Copyright (c) 1982, 1986, 1993
34 * The Regents of the University of California. 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 * 3. Neither the name of the University nor the names of its contributors
45 * may be used to endorse or promote products derived from this software
46 * without specific prior written permission.
47 *
48 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
49 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
50 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
51 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
52 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
53 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
54 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
55 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
56 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
57 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
58 * SUCH DAMAGE.
59 *
60 * @(#)in_proto.c 8.2 (Berkeley) 2/9/95
61 */
62
63 #include <sys/cdefs.h>
64 __KERNEL_RCSID(0, "$NetBSD: in_proto.c,v 1.101 2011/05/03 17:44:31 dyoung Exp $");
65
66 #include "opt_mrouting.h"
67 #include "opt_eon.h" /* ISO CLNL over IP */
68 #include "opt_iso.h" /* ISO TP tunneled over IP */
69 #include "opt_inet.h"
70 #include "opt_ipsec.h"
71 #include "opt_pim.h"
72 #include "opt_gateway.h"
73
74 #include <sys/param.h>
75 #include <sys/socket.h>
76 #include <sys/protosw.h>
77 #include <sys/domain.h>
78 #include <sys/mbuf.h>
79
80 #include <net/if.h>
81 #include <net/radix.h>
82 #include <net/route.h>
83
84 #include <netinet/in.h>
85 #include <netinet/in_systm.h>
86 #include <netinet/ip.h>
87 #include <netinet/ip_var.h>
88 #include <netinet/ip_icmp.h>
89 #include <netinet/in_ifattach.h>
90 #include <netinet/in_pcb.h>
91 #include <netinet/in_proto.h>
92
93 #ifdef INET6
94 #ifndef INET
95 #include <netinet/in.h>
96 #endif
97 #include <netinet/ip6.h>
98 #endif
99
100 #include <netinet/igmp_var.h>
101 #ifdef PIM
102 #include <netinet/pim_var.h>
103 #endif
104 #include <netinet/tcp.h>
105 #include <netinet/tcp_fsm.h>
106 #include <netinet/tcp_seq.h>
107 #include <netinet/tcp_timer.h>
108 #include <netinet/tcp_var.h>
109 #include <netinet/tcpip.h>
110 #include <netinet/tcp_debug.h>
111 #include <netinet/udp.h>
112 #include <netinet/udp_var.h>
113 #include <netinet/ip_encap.h>
114
115 /*
116 * TCP/IP protocol family: IP, ICMP, UDP, TCP.
117 */
118
119 #ifdef IPSEC
120 #include <netinet6/ipsec.h>
121 #include <netinet6/ah.h>
122 #ifdef IPSEC_ESP
123 #include <netinet6/esp.h>
124 #endif
125 #include <netinet6/ipcomp.h>
126 #endif /* IPSEC */
127
128 #ifdef FAST_IPSEC
129 #include <netipsec/ipsec.h>
130 #include <netipsec/key.h>
131 #endif /* FAST_IPSEC */
132
133 #ifdef TPIP
134 #include <netiso/tp_param.h>
135 #include <netiso/tp_var.h>
136 #endif /* TPIP */
137
138 #ifdef EON
139 #include <netiso/eonvar.h>
140 #endif /* EON */
141
142 #include "carp.h"
143 #if NCARP > 0
144 #include <netinet/ip_carp.h>
145 #endif
146
147 #include "pfsync.h"
148 #if NPFSYNC > 0
149 #include <net/pfvar.h>
150 #include <net/if_pfsync.h>
151 #endif
152
153 #include "etherip.h"
154 #if NETHERIP > 0
155 #include <netinet/ip_etherip.h>
156 #endif
157
158 DOMAIN_DEFINE(inetdomain); /* forward declare and add to link set */
159
160 /* Wrappers to acquire kernel_lock. */
161
162 PR_WRAP_USRREQ(rip_usrreq)
163 PR_WRAP_USRREQ(udp_usrreq)
164 PR_WRAP_USRREQ(tcp_usrreq)
165
166 #define rip_usrreq rip_usrreq_wrapper
167 #define udp_usrreq udp_usrreq_wrapper
168 #define tcp_usrreq tcp_usrreq_wrapper
169
170 PR_WRAP_CTLINPUT(rip_ctlinput)
171 PR_WRAP_CTLINPUT(udp_ctlinput)
172 PR_WRAP_CTLINPUT(tcp_ctlinput)
173
174 #define rip_ctlinput rip_ctlinput_wrapper
175 #define udp_ctlinput udp_ctlinput_wrapper
176 #define tcp_ctlinput tcp_ctlinput_wrapper
177
178 PR_WRAP_CTLOUTPUT(rip_ctloutput)
179 PR_WRAP_CTLOUTPUT(udp_ctloutput)
180 PR_WRAP_CTLOUTPUT(tcp_ctloutput)
181
182 #define rip_ctloutput rip_ctloutput_wrapper
183 #define udp_ctloutput udp_ctloutput_wrapper
184 #define tcp_ctloutput tcp_ctloutput_wrapper
185
186 #if defined(IPSEC) || defined(FAST_IPSEC)
187 PR_WRAP_CTLINPUT(ah4_ctlinput)
188
189 #define ah4_ctlinput ah4_ctlinput_wrapper
190 #endif
191
192 #if defined(IPSEC_ESP) || defined(FAST_IPSEC)
193 PR_WRAP_CTLINPUT(esp4_ctlinput)
194
195 #define esp4_ctlinput esp4_ctlinput_wrapper
196 #endif
197
198 #ifdef TPIP
199 PR_WRAP_CTLOUTPUT(tp_ctloutput)
200
201 #define tp_ctloutput tp_ctloutput_wrapper
202
203 PR_WRAP_CTLINPUT(tpip_ctlinput)
204
205 #define tpip_ctlinput tpip_ctlinput_wrapper
206 #endif
207
208 #ifdef EON
209 PR_WRAP_CTLINPUT(eonctlinput)
210
211 #define eonctlinput eonctlinput_wrapper
212 #endif
213
214 const struct protosw inetsw[] = {
215 { .pr_domain = &inetdomain,
216 .pr_init = ip_init,
217 .pr_output = ip_output,
218 .pr_fasttimo = ip_fasttimo,
219 .pr_slowtimo = ip_slowtimo,
220 .pr_drain = ip_drainstub,
221 },
222 { .pr_type = SOCK_DGRAM,
223 .pr_domain = &inetdomain,
224 .pr_protocol = IPPROTO_UDP,
225 .pr_flags = PR_ATOMIC|PR_ADDR|PR_PURGEIF,
226 .pr_input = udp_input,
227 .pr_ctlinput = udp_ctlinput,
228 .pr_ctloutput = udp_ctloutput,
229 .pr_usrreq = udp_usrreq,
230 .pr_init = udp_init,
231 },
232 { .pr_type = SOCK_STREAM,
233 .pr_domain = &inetdomain,
234 .pr_protocol = IPPROTO_TCP,
235 .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_LISTEN|PR_ABRTACPTDIS|PR_PURGEIF,
236 .pr_input = tcp_input,
237 .pr_ctlinput = tcp_ctlinput,
238 .pr_ctloutput = tcp_ctloutput,
239 .pr_usrreq = tcp_usrreq,
240 .pr_init = tcp_init,
241 .pr_fasttimo = tcp_fasttimo,
242 .pr_slowtimo = tcp_slowtimo,
243 .pr_drain = tcp_drainstub,
244 },
245 { .pr_type = SOCK_RAW,
246 .pr_domain = &inetdomain,
247 .pr_protocol = IPPROTO_RAW,
248 .pr_flags = PR_ATOMIC|PR_ADDR|PR_PURGEIF,
249 .pr_input = rip_input,
250 .pr_output = rip_output,
251 .pr_ctlinput = rip_ctlinput,
252 .pr_ctloutput = rip_ctloutput,
253 .pr_usrreq = rip_usrreq,
254 },
255 { .pr_type = SOCK_RAW,
256 .pr_domain = &inetdomain,
257 .pr_protocol = IPPROTO_ICMP,
258 .pr_flags = PR_ATOMIC|PR_ADDR|PR_LASTHDR,
259 .pr_input = icmp_input,
260 .pr_output = rip_output,
261 .pr_ctlinput = rip_ctlinput,
262 .pr_ctloutput = rip_ctloutput,
263 .pr_usrreq = rip_usrreq,
264 .pr_init = icmp_init,
265 },
266 #ifdef GATEWAY
267 { .pr_domain = &inetdomain,
268 .pr_protocol = IPPROTO_IP,
269 .pr_slowtimo = ipflow_slowtimo,
270 .pr_init = ipflow_poolinit,
271 },
272 #endif /* GATEWAY */
273 #ifdef IPSEC
274 { .pr_type = SOCK_RAW,
275 .pr_domain = &inetdomain,
276 .pr_protocol = IPPROTO_AH,
277 .pr_flags = PR_ATOMIC|PR_ADDR,
278 .pr_input = ah4_input,
279 .pr_ctlinput = ah4_ctlinput,
280 .pr_init = ah4_init,
281 },
282 #ifdef IPSEC_ESP
283 { .pr_type = SOCK_RAW,
284 .pr_domain = &inetdomain,
285 .pr_protocol = IPPROTO_ESP,
286 .pr_flags = PR_ATOMIC|PR_ADDR,
287 .pr_input = esp4_input,
288 .pr_ctlinput = esp4_ctlinput,
289 .pr_init = esp4_init,
290 },
291 #endif /* IPSEC_ESP */
292 { .pr_type = SOCK_RAW,
293 .pr_domain = &inetdomain,
294 .pr_protocol = IPPROTO_IPCOMP,
295 .pr_flags = PR_ATOMIC|PR_ADDR,
296 .pr_input = ipcomp4_input,
297 .pr_init = ipcomp4_init,
298 },
299 #endif /* IPSEC */
300 #ifdef FAST_IPSEC
301 { .pr_type = SOCK_RAW,
302 .pr_domain = &inetdomain,
303 .pr_protocol = IPPROTO_AH,
304 .pr_flags = PR_ATOMIC|PR_ADDR,
305 .pr_input = ipsec4_common_input,
306 .pr_ctlinput = ah4_ctlinput,
307 },
308 { .pr_type = SOCK_RAW,
309 .pr_domain = &inetdomain,
310 .pr_protocol = IPPROTO_ESP,
311 .pr_flags = PR_ATOMIC|PR_ADDR,
312 .pr_input = ipsec4_common_input,
313 .pr_ctlinput = esp4_ctlinput,
314 },
315 { .pr_type = SOCK_RAW,
316 .pr_domain = &inetdomain,
317 .pr_protocol = IPPROTO_IPCOMP,
318 .pr_flags = PR_ATOMIC|PR_ADDR,
319 .pr_input = ipsec4_common_input,
320 },
321 #endif /* FAST_IPSEC */
322 { .pr_type = SOCK_RAW,
323 .pr_domain = &inetdomain,
324 .pr_protocol = IPPROTO_IPV4,
325 .pr_flags = PR_ATOMIC|PR_ADDR|PR_LASTHDR,
326 .pr_input = encap4_input,
327 .pr_output = rip_output,
328 .pr_ctlinput = rip_ctlinput,
329 .pr_ctloutput = rip_ctloutput,
330 .pr_usrreq = rip_usrreq,
331 .pr_init = encap_init,
332 },
333 #ifdef INET6
334 { .pr_type = SOCK_RAW,
335 .pr_domain = &inetdomain,
336 .pr_protocol = IPPROTO_IPV6,
337 .pr_flags = PR_ATOMIC|PR_ADDR|PR_LASTHDR,
338 .pr_input = encap4_input,
339 .pr_output = rip_output,
340 .pr_ctlinput = rip_ctlinput,
341 .pr_ctloutput = rip_ctloutput,
342 .pr_usrreq = rip_usrreq,
343 .pr_init = encap_init,
344 },
345 #endif /* INET6 */
346 #if NETHERIP > 0
347 { .pr_type = SOCK_RAW,
348 .pr_domain = &inetdomain,
349 .pr_protocol = IPPROTO_ETHERIP,
350 .pr_flags = PR_ATOMIC|PR_ADDR|PR_LASTHDR,
351 .pr_input = ip_etherip_input,
352 .pr_output = rip_output,
353 .pr_ctlinput = rip_ctlinput,
354 .pr_ctloutput = rip_ctloutput,
355 .pr_usrreq = rip_usrreq,
356 },
357 #endif /* NETHERIP > 0 */
358 #if NCARP > 0
359 { .pr_type = SOCK_RAW,
360 .pr_domain = &inetdomain,
361 .pr_protocol = IPPROTO_CARP,
362 .pr_flags = PR_ATOMIC|PR_ADDR,
363 .pr_input = carp_proto_input,
364 .pr_output = rip_output,
365 .pr_ctloutput = rip_ctloutput,
366 .pr_usrreq = rip_usrreq,
367 .pr_init = carp_init,
368 },
369 #endif /* NCARP > 0 */
370 #if NPFSYNC > 0
371 { .pr_type = SOCK_RAW,
372 .pr_domain = &inetdomain,
373 .pr_protocol = IPPROTO_PFSYNC,
374 .pr_flags = PR_ATOMIC|PR_ADDR,
375 .pr_input = pfsync_input,
376 .pr_output = rip_output,
377 .pr_ctloutput = rip_ctloutput,
378 .pr_usrreq = rip_usrreq,
379 },
380 #endif /* NPFSYNC > 0 */
381 { .pr_type = SOCK_RAW,
382 .pr_domain = &inetdomain,
383 .pr_protocol = IPPROTO_IGMP,
384 .pr_flags = PR_ATOMIC|PR_ADDR|PR_LASTHDR,
385 .pr_input = igmp_input,
386 .pr_output = rip_output,
387 .pr_ctloutput = rip_ctloutput,
388 .pr_ctlinput = rip_ctlinput,
389 .pr_usrreq = rip_usrreq,
390 .pr_fasttimo = igmp_fasttimo,
391 .pr_slowtimo = igmp_slowtimo,
392 .pr_init = igmp_init,
393 },
394 #ifdef PIM
395 { .pr_type = SOCK_RAW,
396 .pr_domain = &inetdomain,
397 .pr_protocol = IPPROTO_PIM,
398 .pr_flags = PR_ATOMIC|PR_ADDR|PR_LASTHDR,
399 .pr_input = pim_input,
400 .pr_output = rip_output,
401 .pr_ctloutput = rip_ctloutput,
402 .pr_ctlinput = rip_ctlinput,
403 .pr_usrreq = rip_usrreq,
404 },
405 #endif /* PIM */
406 #ifdef TPIP
407 { .pr_type = SOCK_SEQPACKET,
408 .pr_domain = &inetdomain,
409 .pr_protocol = IPPROTO_TP,
410 .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_LISTEN|PR_LASTHDR|PR_ABRTACPTDIS,
411 .pr_input = tpip_input,
412 .pr_ctloutput = tp_ctloutput,
413 .pr_ctlinput = tpip_ctlinput,
414 .pr_usrreq = tp_usrreq,
415 .pr_init = tp_init,
416 .pr_slowtimo = tp_slowtimo,
417 .pr_drain = tp_drain,
418 },
419 #endif /* TPIP */
420 #ifdef ISO
421 /* EON (ISO CLNL over IP) */
422 #ifdef EON
423 { .pr_type = SOCK_RAW,
424 .pr_domain = &inetdomain,
425 .pr_protocol = IPPROTO_EON,
426 .pr_flags = PR_LASTHDR,
427 .pr_input = eoninput,
428 .pr_ctlinput = eonctlinput,
429 .pr_init = eonprotoinit,
430 },
431 #else
432 { .pr_type = SOCK_RAW,
433 .pr_domain = &inetdomain,
434 .pr_protocol = IPPROTO_EON,
435 .pr_flags = PR_ATOMIC|PR_ADDR|PR_LASTHDR,
436 .pr_input = encap4_input,
437 .pr_output = rip_output,
438 .pr_ctloutput = rip_ctloutput,
439 .pr_ctlinput = rip_ctlinput,
440 .pr_usrreq = rip_usrreq,
441 .pr_init = encap_init,
442 },
443 #endif /* EON */
444 #endif /* ISO */
445 /* raw wildcard */
446 { .pr_type = SOCK_RAW,
447 .pr_domain = &inetdomain,
448 .pr_flags = PR_ATOMIC|PR_ADDR|PR_LASTHDR,
449 .pr_input = rip_input,
450 .pr_output = rip_output,
451 .pr_ctloutput = rip_ctloutput,
452 .pr_ctlinput = rip_ctlinput,
453 .pr_usrreq = rip_usrreq,
454 .pr_init = rip_init,
455 },
456 };
457
458 extern struct ifqueue ipintrq;
459
460 const struct sockaddr_in in_any = {
461 .sin_len = sizeof(struct sockaddr_in)
462 , .sin_family = AF_INET
463 , .sin_port = 0
464 , .sin_addr = {.s_addr = 0 /* INADDR_ANY */}
465 };
466
467 struct domain inetdomain = {
468 .dom_family = PF_INET, .dom_name = "internet", .dom_init = NULL,
469 .dom_externalize = NULL, .dom_dispose = NULL,
470 .dom_protosw = inetsw,
471 .dom_protoswNPROTOSW = &inetsw[__arraycount(inetsw)],
472 .dom_rtattach = rt_inithead,
473 .dom_rtoffset = 32,
474 .dom_maxrtkey = sizeof(struct ip_pack4),
475 #ifdef IPSELSRC
476 .dom_ifattach = in_domifattach,
477 .dom_ifdetach = in_domifdetach,
478 #else
479 .dom_ifattach = NULL,
480 .dom_ifdetach = NULL,
481 #endif
482 .dom_ifqueues = { &ipintrq, NULL },
483 .dom_link = { NULL },
484 .dom_mowner = MOWNER_INIT("",""),
485 .dom_sa_cmpofs = offsetof(struct sockaddr_in, sin_addr),
486 .dom_sa_cmplen = sizeof(struct in_addr),
487 .dom_sa_any = (const struct sockaddr *)&in_any,
488 .dom_sockaddr_const_addr = sockaddr_in_const_addr,
489 .dom_sockaddr_addr = sockaddr_in_addr,
490 .dom_rtcache = LIST_HEAD_INITIALIZER(inetdomain.dom_rtcache)
491 };
492
493 u_char ip_protox[IPPROTO_MAX];
494
495 int icmperrppslim = 100; /* 100pps */
496
497 static void
498 sockaddr_in_addrlen(const struct sockaddr *sa, socklen_t *slenp)
499 {
500 socklen_t slen;
501
502 if (slenp == NULL)
503 return;
504
505 slen = sockaddr_getlen(sa);
506 *slenp = (socklen_t)MIN(sizeof(struct in_addr),
507 slen - MIN(slen, offsetof(struct sockaddr_in, sin_addr)));
508 }
509
510 const void *
511 sockaddr_in_const_addr(const struct sockaddr *sa, socklen_t *slenp)
512 {
513 const struct sockaddr_in *sin;
514
515 sockaddr_in_addrlen(sa, slenp);
516 sin = (const struct sockaddr_in *)sa;
517 return &sin->sin_addr;
518 }
519
520 void *
521 sockaddr_in_addr(struct sockaddr *sa, socklen_t *slenp)
522 {
523 struct sockaddr_in *sin;
524
525 sockaddr_in_addrlen(sa, slenp);
526 sin = (struct sockaddr_in *)sa;
527 return &sin->sin_addr;
528 }
529
530 int
531 sockaddr_in_cmp(const struct sockaddr *sa1, const struct sockaddr *sa2)
532 {
533 uint_fast8_t len;
534 const uint_fast8_t addrofs = offsetof(struct sockaddr_in, sin_addr),
535 addrend = addrofs + sizeof(struct in_addr);
536 int rc;
537 const struct sockaddr_in *sin1, *sin2;
538
539 sin1 = satocsin(sa1);
540 sin2 = satocsin(sa2);
541
542 len = MIN(addrend, MIN(sin1->sin_len, sin2->sin_len));
543
544 if (len > addrofs &&
545 (rc = memcmp(&sin1->sin_addr, &sin2->sin_addr,
546 len - addrofs)) != 0)
547 return rc;
548
549 return sin1->sin_len - sin2->sin_len;
550 }
551