xref: /netbsd-src/sys/netinet/in_proto.c (revision 10ad5ffa714ce1a679dcc9dd8159648df2d67b5a)
1 /*	$NetBSD: in_proto.c,v 1.97 2009/02/28 18:31:12 pooka 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 1.97 2009/02/28 18:31:12 pooka 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 "etherip.h"
148 #if NETHERIP > 0
149 #include <netinet/ip_etherip.h>
150 #endif
151 
152 DOMAIN_DEFINE(inetdomain);	/* forward declare and add to link set */
153 
154 /* Wrappers to acquire kernel_lock. */
155 
156 PR_WRAP_USRREQ(rip_usrreq)
157 PR_WRAP_USRREQ(udp_usrreq)
158 PR_WRAP_USRREQ(tcp_usrreq)
159 
160 #define	rip_usrreq 	rip_usrreq_wrapper
161 #define	udp_usrreq 	udp_usrreq_wrapper
162 #define	tcp_usrreq 	tcp_usrreq_wrapper
163 
164 PR_WRAP_CTLINPUT(rip_ctlinput)
165 PR_WRAP_CTLINPUT(udp_ctlinput)
166 PR_WRAP_CTLINPUT(tcp_ctlinput)
167 
168 #define	rip_ctlinput	rip_ctlinput_wrapper
169 #define	udp_ctlinput	udp_ctlinput_wrapper
170 #define	tcp_ctlinput	tcp_ctlinput_wrapper
171 
172 PR_WRAP_CTLOUTPUT(rip_ctloutput)
173 PR_WRAP_CTLOUTPUT(udp_ctloutput)
174 PR_WRAP_CTLOUTPUT(tcp_ctloutput)
175 
176 #define	rip_ctloutput	rip_ctloutput_wrapper
177 #define	udp_ctloutput	udp_ctloutput_wrapper
178 #define	tcp_ctloutput	tcp_ctloutput_wrapper
179 
180 #if defined(IPSEC) || defined(FAST_IPSEC)
181 PR_WRAP_CTLINPUT(ah4_ctlinput)
182 
183 #define	ah4_ctlinput	ah4_ctlinput_wrapper
184 #endif
185 
186 #if defined(IPSEC_ESP) || defined(FAST_IPSEC)
187 PR_WRAP_CTLINPUT(esp4_ctlinput)
188 
189 #define	esp4_ctlinput	esp4_ctlinput_wrapper
190 #endif
191 
192 #ifdef TPIP
193 PR_WRAP_CTLOUTPUT(tp_ctloutput)
194 
195 #define	tp_ctloutput	tp_ctloutput_wrapper
196 
197 PR_WRAP_CTLINPUT(tpip_ctlinput)
198 
199 #define	tpip_ctlinput	tpip_ctlinput_wrapper
200 #endif
201 
202 #ifdef EON
203 PR_WRAP_CTLINPUT(eonctlinput)
204 
205 #define	eonctlinput	eonctlinput_wrapper
206 #endif
207 
208 const struct protosw inetsw[] = {
209 {	.pr_domain = &inetdomain,
210 	.pr_init = ip_init,
211 	.pr_output = ip_output,
212 	.pr_slowtimo = ip_slowtimo,
213 	.pr_drain = ip_drain,
214 },
215 {	.pr_type = SOCK_DGRAM,
216 	.pr_domain = &inetdomain,
217 	.pr_protocol = IPPROTO_UDP,
218 	.pr_flags = PR_ATOMIC|PR_ADDR|PR_PURGEIF,
219 	.pr_input = udp_input,
220 	.pr_ctlinput = udp_ctlinput,
221 	.pr_ctloutput = udp_ctloutput,
222 	.pr_usrreq = udp_usrreq,
223 	.pr_init = udp_init,
224 },
225 {	.pr_type = SOCK_STREAM,
226 	.pr_domain = &inetdomain,
227 	.pr_protocol = IPPROTO_TCP,
228 	.pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_LISTEN|PR_ABRTACPTDIS|PR_PURGEIF,
229 	.pr_input = tcp_input,
230 	.pr_ctlinput = tcp_ctlinput,
231 	.pr_ctloutput = tcp_ctloutput,
232 	.pr_usrreq = tcp_usrreq,
233 	.pr_init = tcp_init,
234 	.pr_slowtimo = tcp_slowtimo,
235 	.pr_drain = tcp_drain,
236 },
237 {	.pr_type = SOCK_RAW,
238 	.pr_domain = &inetdomain,
239 	.pr_protocol = IPPROTO_RAW,
240 	.pr_flags = PR_ATOMIC|PR_ADDR|PR_PURGEIF,
241 	.pr_input = rip_input,
242 	.pr_output = rip_output,
243 	.pr_ctlinput = rip_ctlinput,
244 	.pr_ctloutput = rip_ctloutput,
245 	.pr_usrreq = rip_usrreq,
246 },
247 {	.pr_type = SOCK_RAW,
248 	.pr_domain = &inetdomain,
249 	.pr_protocol = IPPROTO_ICMP,
250 	.pr_flags = PR_ATOMIC|PR_ADDR|PR_LASTHDR,
251 	.pr_input = icmp_input,
252 	.pr_output = rip_output,
253 	.pr_ctlinput = rip_ctlinput,
254 	.pr_ctloutput = rip_ctloutput,
255 	.pr_usrreq = rip_usrreq,
256 	.pr_init = icmp_init,
257 },
258 #ifdef GATEWAY
259 {	.pr_domain = &inetdomain,
260 	.pr_protocol = IPPROTO_IP,
261 	.pr_slowtimo = ipflow_slowtimo,
262 	.pr_init = ipflow_poolinit,
263 },
264 #endif /* GATEWAY */
265 #ifdef IPSEC
266 {	.pr_type = SOCK_RAW,
267 	.pr_domain = &inetdomain,
268 	.pr_protocol = IPPROTO_AH,
269 	.pr_flags = PR_ATOMIC|PR_ADDR,
270 	.pr_input = ah4_input,
271 	.pr_ctlinput = ah4_ctlinput,
272 	.pr_init = ah4_init,
273 },
274 #ifdef IPSEC_ESP
275 {	.pr_type = SOCK_RAW,
276 	.pr_domain = &inetdomain,
277 	.pr_protocol = IPPROTO_ESP,
278 	.pr_flags = PR_ATOMIC|PR_ADDR,
279 	.pr_input = esp4_input,
280 	.pr_ctlinput = esp4_ctlinput,
281 	.pr_init = esp4_init,
282 },
283 #endif /* IPSEC_ESP */
284 {	.pr_type = SOCK_RAW,
285 	.pr_domain = &inetdomain,
286 	.pr_protocol = IPPROTO_IPCOMP,
287 	.pr_flags = PR_ATOMIC|PR_ADDR,
288 	.pr_input = ipcomp4_input,
289 	.pr_init = ipcomp4_init,
290 },
291 #endif /* IPSEC */
292 #ifdef FAST_IPSEC
293 {	.pr_type = SOCK_RAW,
294 	.pr_domain = &inetdomain,
295 	.pr_protocol = IPPROTO_AH,
296 	.pr_flags = PR_ATOMIC|PR_ADDR,
297 	.pr_input = ipsec4_common_input,
298 	.pr_ctlinput = ah4_ctlinput,
299 },
300 {	.pr_type = SOCK_RAW,
301 	.pr_domain = &inetdomain,
302 	.pr_protocol = IPPROTO_ESP,
303 	.pr_flags = PR_ATOMIC|PR_ADDR,
304 	.pr_input = ipsec4_common_input,
305 	.pr_ctlinput = esp4_ctlinput,
306 },
307 {	.pr_type = SOCK_RAW,
308 	.pr_domain = &inetdomain,
309 	.pr_protocol = IPPROTO_IPCOMP,
310 	.pr_flags = PR_ATOMIC|PR_ADDR,
311 	.pr_input = ipsec4_common_input,
312 },
313 #endif /* FAST_IPSEC */
314 {	.pr_type = SOCK_RAW,
315 	.pr_domain = &inetdomain,
316 	.pr_protocol = IPPROTO_IPV4,
317 	.pr_flags = PR_ATOMIC|PR_ADDR|PR_LASTHDR,
318 	.pr_input = encap4_input,
319 	.pr_output = rip_output,
320 	.pr_ctlinput = rip_ctlinput,
321 	.pr_ctloutput = rip_ctloutput,
322 	.pr_usrreq = rip_usrreq,
323 	.pr_init = encap_init,
324 },
325 #ifdef INET6
326 {	.pr_type = SOCK_RAW,
327 	.pr_domain = &inetdomain,
328 	.pr_protocol = IPPROTO_IPV6,
329 	.pr_flags = PR_ATOMIC|PR_ADDR|PR_LASTHDR,
330 	.pr_input = encap4_input,
331 	.pr_output = rip_output,
332 	.pr_ctlinput = rip_ctlinput,
333 	.pr_ctloutput = rip_ctloutput,
334 	.pr_usrreq = rip_usrreq,
335 	.pr_init = encap_init,
336 },
337 #endif /* INET6 */
338 #if NETHERIP > 0
339 {	.pr_type = SOCK_RAW,
340 	.pr_domain = &inetdomain,
341 	.pr_protocol = IPPROTO_ETHERIP,
342 	.pr_flags = PR_ATOMIC|PR_ADDR|PR_LASTHDR,
343 	.pr_input = ip_etherip_input,
344 	.pr_output = rip_output,
345 	.pr_ctlinput = rip_ctlinput,
346 	.pr_ctloutput = rip_ctloutput,
347 	.pr_usrreq = rip_usrreq,
348 },
349 #endif /* NETHERIP > 0 */
350 #if NCARP > 0
351 {	.pr_type = SOCK_RAW,
352 	.pr_domain = &inetdomain,
353 	.pr_protocol = IPPROTO_CARP,
354 	.pr_flags = PR_ATOMIC|PR_ADDR,
355 	.pr_input = carp_proto_input,
356 	.pr_output = rip_output,
357 	.pr_ctloutput = rip_ctloutput,
358 	.pr_usrreq = rip_usrreq,
359 },
360 #endif /* NCARP > 0 */
361 {	.pr_type = SOCK_RAW,
362 	.pr_domain = &inetdomain,
363 	.pr_protocol = IPPROTO_IGMP,
364 	.pr_flags = PR_ATOMIC|PR_ADDR|PR_LASTHDR,
365 	.pr_input = igmp_input,
366 	.pr_output = rip_output,
367 	.pr_ctloutput = rip_ctloutput,
368 	.pr_ctlinput = rip_ctlinput,
369 	.pr_usrreq = rip_usrreq,
370 	.pr_fasttimo = igmp_fasttimo,
371 	.pr_slowtimo = igmp_slowtimo,
372 	.pr_init = igmp_init,
373 },
374 #ifdef PIM
375 {	.pr_type = SOCK_RAW,
376 	.pr_domain = &inetdomain,
377 	.pr_protocol = IPPROTO_PIM,
378 	.pr_flags = PR_ATOMIC|PR_ADDR|PR_LASTHDR,
379 	.pr_input = pim_input,
380 	.pr_output = rip_output,
381 	.pr_ctloutput = rip_ctloutput,
382 	.pr_ctlinput = rip_ctlinput,
383 	.pr_usrreq = rip_usrreq,
384 },
385 #endif /* PIM */
386 #ifdef TPIP
387 {	.pr_type = SOCK_SEQPACKET,
388 	.pr_domain = &inetdomain,
389 	.pr_protocol = IPPROTO_TP,
390 	.pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_LISTEN|PR_LASTHDR|PR_ABRTACPTDIS,
391 	.pr_input = tpip_input,
392 	.pr_ctloutput = tp_ctloutput,
393 	.pr_ctlinput = tpip_ctlinput,
394 	.pr_usrreq = tp_usrreq,
395 	.pr_init = tp_init,
396 	.pr_slowtimo = tp_slowtimo,
397 	.pr_drain = tp_drain,
398 },
399 #endif /* TPIP */
400 #ifdef ISO
401 /* EON (ISO CLNL over IP) */
402 #ifdef EON
403 {	.pr_type = SOCK_RAW,
404 	.pr_domain = &inetdomain,
405 	.pr_protocol = IPPROTO_EON,
406 	.pr_flags = PR_LASTHDR,
407 	.pr_input = eoninput,
408 	.pr_ctlinput = eonctlinput,
409 	.pr_init = eonprotoinit,
410 },
411 #else
412 {	.pr_type = SOCK_RAW,
413 	.pr_domain = &inetdomain,
414 	.pr_protocol = IPPROTO_EON,
415 	.pr_flags = PR_ATOMIC|PR_ADDR|PR_LASTHDR,
416 	.pr_input = encap4_input,
417 	.pr_output = rip_output,
418 	.pr_ctloutput = rip_ctloutput,
419 	.pr_ctlinput = rip_ctlinput,
420 	.pr_usrreq = rip_usrreq,
421 	.pr_init = encap_init,
422 },
423 #endif /* EON */
424 #endif /* ISO */
425 /* raw wildcard */
426 {	.pr_type = SOCK_RAW,
427 	.pr_domain = &inetdomain,
428 	.pr_flags = PR_ATOMIC|PR_ADDR|PR_LASTHDR,
429 	.pr_input = rip_input,
430 	.pr_output = rip_output,
431 	.pr_ctloutput = rip_ctloutput,
432 	.pr_ctlinput = rip_ctlinput,
433 	.pr_usrreq = rip_usrreq,
434 	.pr_init = rip_init,
435 },
436 };
437 
438 extern struct ifqueue ipintrq;
439 
440 const struct sockaddr_in in_any = {
441 	  .sin_len = sizeof(struct sockaddr_in)
442 	, .sin_family = AF_INET
443 	, .sin_port = 0
444 	, .sin_addr = {.s_addr = 0 /* INADDR_ANY */}
445 };
446 
447 struct domain inetdomain = {
448 	.dom_family = PF_INET, .dom_name = "internet", .dom_init = NULL,
449 	.dom_externalize = NULL, .dom_dispose = NULL,
450 	.dom_protosw = inetsw,
451 	.dom_protoswNPROTOSW = &inetsw[__arraycount(inetsw)],
452 	.dom_rtattach = rn_inithead,
453 	.dom_rtoffset = 32,
454 	.dom_maxrtkey = sizeof(struct ip_pack4),
455 #ifdef IPSELSRC
456 	.dom_ifattach = in_domifattach,
457 	.dom_ifdetach = in_domifdetach,
458 #else
459 	.dom_ifattach = NULL,
460 	.dom_ifdetach = NULL,
461 #endif
462 	.dom_ifqueues = { &ipintrq, NULL },
463 	.dom_link = { NULL },
464 	.dom_mowner = MOWNER_INIT("",""),
465 	.dom_sa_cmpofs = offsetof(struct sockaddr_in, sin_addr),
466 	.dom_sa_cmplen = sizeof(struct in_addr),
467 	.dom_sa_any = (const struct sockaddr *)&in_any,
468 	.dom_sockaddr_const_addr = sockaddr_in_const_addr,
469 	.dom_sockaddr_addr = sockaddr_in_addr,
470 	.dom_rtcache = LIST_HEAD_INITIALIZER(inetdomain.dom_rtcache)
471 };
472 
473 u_char	ip_protox[IPPROTO_MAX];
474 
475 int icmperrppslim = 100;			/* 100pps */
476 
477 static void
478 sockaddr_in_addrlen(const struct sockaddr *sa, socklen_t *slenp)
479 {
480 	socklen_t slen;
481 
482 	if (slenp == NULL)
483 		return;
484 
485 	slen = sockaddr_getlen(sa);
486 	*slenp = (socklen_t)MIN(sizeof(struct in_addr),
487 	    slen - MIN(slen, offsetof(struct sockaddr_in, sin_addr)));
488 }
489 
490 const void *
491 sockaddr_in_const_addr(const struct sockaddr *sa, socklen_t *slenp)
492 {
493 	const struct sockaddr_in *sin;
494 
495 	sockaddr_in_addrlen(sa, slenp);
496 	sin = (const struct sockaddr_in *)sa;
497 	return &sin->sin_addr;
498 }
499 
500 void *
501 sockaddr_in_addr(struct sockaddr *sa, socklen_t *slenp)
502 {
503 	struct sockaddr_in *sin;
504 
505 	sockaddr_in_addrlen(sa, slenp);
506 	sin = (struct sockaddr_in *)sa;
507 	return &sin->sin_addr;
508 }
509 
510 int
511 sockaddr_in_cmp(const struct sockaddr *sa1, const struct sockaddr *sa2)
512 {
513 	uint_fast8_t len;
514 	const uint_fast8_t addrofs = offsetof(struct sockaddr_in, sin_addr),
515 			   addrend = addrofs + sizeof(struct in_addr);
516 	int rc;
517 	const struct sockaddr_in *sin1, *sin2;
518 
519 	sin1 = satocsin(sa1);
520 	sin2 = satocsin(sa2);
521 
522 	len = MIN(addrend, MIN(sin1->sin_len, sin2->sin_len));
523 
524 	if (len > addrofs &&
525 	     (rc = memcmp(&sin1->sin_addr, &sin2->sin_addr,
526 	                  len - addrofs)) != 0)
527 		return rc;
528 
529 	return sin1->sin_len - sin2->sin_len;
530 }
531