xref: /netbsd-src/sys/netinet/in_proto.c (revision 6a493d6bc668897c91594964a732d38505b70cbb)
1 /*	$NetBSD: in_proto.c,v 1.106 2013/06/05 19:01:26 christos 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 1.106 2013/06/05 19:01:26 christos Exp $");
65 
66 #include "opt_mrouting.h"
67 #include "opt_inet.h"
68 #include "opt_ipsec.h"
69 #include "opt_pim.h"
70 #include "opt_gateway.h"
71 
72 #include <sys/param.h>
73 #include <sys/socket.h>
74 #include <sys/protosw.h>
75 #include <sys/domain.h>
76 #include <sys/mbuf.h>
77 
78 #include <net/if.h>
79 #include <net/radix.h>
80 #include <net/route.h>
81 
82 #include <netinet/in.h>
83 #include <netinet/in_systm.h>
84 #include <netinet/ip.h>
85 #include <netinet/ip_var.h>
86 #include <netinet/ip_icmp.h>
87 #include <netinet/in_ifattach.h>
88 #include <netinet/in_pcb.h>
89 #include <netinet/in_proto.h>
90 
91 #ifdef INET6
92 #ifndef INET
93 #include <netinet/in.h>
94 #endif
95 #include <netinet/ip6.h>
96 #endif
97 
98 #include <netinet/igmp_var.h>
99 #ifdef PIM
100 #include <netinet/pim_var.h>
101 #endif
102 #include <netinet/tcp.h>
103 #include <netinet/tcp_fsm.h>
104 #include <netinet/tcp_seq.h>
105 #include <netinet/tcp_timer.h>
106 #include <netinet/tcp_var.h>
107 #include <netinet/tcpip.h>
108 #include <netinet/tcp_debug.h>
109 #include <netinet/udp.h>
110 #include <netinet/udp_var.h>
111 #include <netinet/ip_encap.h>
112 
113 /*
114  * TCP/IP protocol family: IP, ICMP, UDP, TCP.
115  */
116 
117 #ifdef IPSEC
118 #include <netipsec/ipsec.h>
119 #include <netipsec/key.h>
120 #endif	/* IPSEC */
121 
122 #include "carp.h"
123 #if NCARP > 0
124 #include <netinet/ip_carp.h>
125 #endif
126 
127 #include "pfsync.h"
128 #if NPFSYNC > 0
129 #include <net/pfvar.h>
130 #include <net/if_pfsync.h>
131 #endif
132 
133 #include "etherip.h"
134 #if NETHERIP > 0
135 #include <netinet/ip_etherip.h>
136 #endif
137 
138 DOMAIN_DEFINE(inetdomain);	/* forward declare and add to link set */
139 
140 /* Wrappers to acquire kernel_lock. */
141 
142 PR_WRAP_USRREQ(rip_usrreq)
143 PR_WRAP_USRREQ(udp_usrreq)
144 PR_WRAP_USRREQ(tcp_usrreq)
145 
146 #define	rip_usrreq 	rip_usrreq_wrapper
147 #define	udp_usrreq 	udp_usrreq_wrapper
148 #define	tcp_usrreq 	tcp_usrreq_wrapper
149 
150 PR_WRAP_CTLINPUT(rip_ctlinput)
151 PR_WRAP_CTLINPUT(udp_ctlinput)
152 PR_WRAP_CTLINPUT(tcp_ctlinput)
153 
154 #define	rip_ctlinput	rip_ctlinput_wrapper
155 #define	udp_ctlinput	udp_ctlinput_wrapper
156 #define	tcp_ctlinput	tcp_ctlinput_wrapper
157 
158 PR_WRAP_CTLOUTPUT(rip_ctloutput)
159 PR_WRAP_CTLOUTPUT(udp_ctloutput)
160 PR_WRAP_CTLOUTPUT(tcp_ctloutput)
161 
162 #define	rip_ctloutput	rip_ctloutput_wrapper
163 #define	udp_ctloutput	udp_ctloutput_wrapper
164 #define	tcp_ctloutput	tcp_ctloutput_wrapper
165 
166 #if defined(IPSEC)
167 PR_WRAP_CTLINPUT(ah4_ctlinput)
168 
169 #define	ah4_ctlinput	ah4_ctlinput_wrapper
170 PR_WRAP_CTLINPUT(esp4_ctlinput)
171 
172 #define	esp4_ctlinput	esp4_ctlinput_wrapper
173 #endif
174 
175 const struct protosw inetsw[] = {
176 {	.pr_domain = &inetdomain,
177 	.pr_init = ip_init,
178 	.pr_output = ip_output,
179 	.pr_fasttimo = ip_fasttimo,
180 	.pr_slowtimo = ip_slowtimo,
181 	.pr_drain = ip_drainstub,
182 },
183 {	.pr_type = SOCK_DGRAM,
184 	.pr_domain = &inetdomain,
185 	.pr_protocol = IPPROTO_UDP,
186 	.pr_flags = PR_ATOMIC|PR_ADDR|PR_PURGEIF,
187 	.pr_input = udp_input,
188 	.pr_ctlinput = udp_ctlinput,
189 	.pr_ctloutput = udp_ctloutput,
190 	.pr_usrreq = udp_usrreq,
191 	.pr_init = udp_init,
192 },
193 {	.pr_type = SOCK_STREAM,
194 	.pr_domain = &inetdomain,
195 	.pr_protocol = IPPROTO_TCP,
196 	.pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_LISTEN|PR_ABRTACPTDIS|PR_PURGEIF,
197 	.pr_input = tcp_input,
198 	.pr_ctlinput = tcp_ctlinput,
199 	.pr_ctloutput = tcp_ctloutput,
200 	.pr_usrreq = tcp_usrreq,
201 	.pr_init = tcp_init,
202 	.pr_fasttimo = tcp_fasttimo,
203 	.pr_slowtimo = tcp_slowtimo,
204 	.pr_drain = tcp_drainstub,
205 },
206 {	.pr_type = SOCK_RAW,
207 	.pr_domain = &inetdomain,
208 	.pr_protocol = IPPROTO_RAW,
209 	.pr_flags = PR_ATOMIC|PR_ADDR|PR_PURGEIF,
210 	.pr_input = rip_input,
211 	.pr_output = rip_output,
212 	.pr_ctlinput = rip_ctlinput,
213 	.pr_ctloutput = rip_ctloutput,
214 	.pr_usrreq = rip_usrreq,
215 },
216 {	.pr_type = SOCK_RAW,
217 	.pr_domain = &inetdomain,
218 	.pr_protocol = IPPROTO_ICMP,
219 	.pr_flags = PR_ATOMIC|PR_ADDR|PR_LASTHDR,
220 	.pr_input = icmp_input,
221 	.pr_output = rip_output,
222 	.pr_ctlinput = rip_ctlinput,
223 	.pr_ctloutput = rip_ctloutput,
224 	.pr_usrreq = rip_usrreq,
225 	.pr_init = icmp_init,
226 },
227 #ifdef GATEWAY
228 {	.pr_domain = &inetdomain,
229 	.pr_protocol = IPPROTO_IP,
230 	.pr_slowtimo = ipflow_slowtimo,
231 	.pr_init = ipflow_poolinit,
232 },
233 #endif /* GATEWAY */
234 #ifdef IPSEC
235 {	.pr_type = SOCK_RAW,
236 	.pr_domain = &inetdomain,
237 	.pr_protocol = IPPROTO_AH,
238 	.pr_flags = PR_ATOMIC|PR_ADDR,
239 	.pr_input = ipsec4_common_input,
240 	.pr_ctlinput = ah4_ctlinput,
241 },
242 {	.pr_type = SOCK_RAW,
243 	.pr_domain = &inetdomain,
244 	.pr_protocol = IPPROTO_ESP,
245 	.pr_flags = PR_ATOMIC|PR_ADDR,
246 	.pr_input = ipsec4_common_input,
247 	.pr_ctlinput = esp4_ctlinput,
248 },
249 {	.pr_type = SOCK_RAW,
250 	.pr_domain = &inetdomain,
251 	.pr_protocol = IPPROTO_IPCOMP,
252 	.pr_flags = PR_ATOMIC|PR_ADDR,
253 	.pr_input = ipsec4_common_input,
254 },
255 #endif /* IPSEC */
256 {	.pr_type = SOCK_RAW,
257 	.pr_domain = &inetdomain,
258 	.pr_protocol = IPPROTO_IPV4,
259 	.pr_flags = PR_ATOMIC|PR_ADDR|PR_LASTHDR,
260 	.pr_input = encap4_input,
261 	.pr_output = rip_output,
262 	.pr_ctlinput = rip_ctlinput,
263 	.pr_ctloutput = rip_ctloutput,
264 	.pr_usrreq = rip_usrreq,
265 	.pr_init = encap_init,
266 },
267 #ifdef INET6
268 {	.pr_type = SOCK_RAW,
269 	.pr_domain = &inetdomain,
270 	.pr_protocol = IPPROTO_IPV6,
271 	.pr_flags = PR_ATOMIC|PR_ADDR|PR_LASTHDR,
272 	.pr_input = encap4_input,
273 	.pr_output = rip_output,
274 	.pr_ctlinput = rip_ctlinput,
275 	.pr_ctloutput = rip_ctloutput,
276 	.pr_usrreq = rip_usrreq,
277 	.pr_init = encap_init,
278 },
279 #endif /* INET6 */
280 #if NETHERIP > 0
281 {	.pr_type = SOCK_RAW,
282 	.pr_domain = &inetdomain,
283 	.pr_protocol = IPPROTO_ETHERIP,
284 	.pr_flags = PR_ATOMIC|PR_ADDR|PR_LASTHDR,
285 	.pr_input = ip_etherip_input,
286 	.pr_output = rip_output,
287 	.pr_ctlinput = rip_ctlinput,
288 	.pr_ctloutput = rip_ctloutput,
289 	.pr_usrreq = rip_usrreq,
290 },
291 #endif /* NETHERIP > 0 */
292 #if NCARP > 0
293 {	.pr_type = SOCK_RAW,
294 	.pr_domain = &inetdomain,
295 	.pr_protocol = IPPROTO_CARP,
296 	.pr_flags = PR_ATOMIC|PR_ADDR,
297 	.pr_input = carp_proto_input,
298 	.pr_output = rip_output,
299 	.pr_ctloutput = rip_ctloutput,
300 	.pr_usrreq = rip_usrreq,
301 	.pr_init = carp_init,
302 },
303 #endif /* NCARP > 0 */
304 #if NPFSYNC > 0
305 {	.pr_type = SOCK_RAW,
306 	.pr_domain = &inetdomain,
307 	.pr_protocol = IPPROTO_PFSYNC,
308 	.pr_flags	 = PR_ATOMIC|PR_ADDR,
309 	.pr_input	 = pfsync_input,
310 	.pr_output	 = rip_output,
311 	.pr_ctloutput = rip_ctloutput,
312 	.pr_usrreq	 = rip_usrreq,
313 },
314 #endif /* NPFSYNC > 0 */
315 {	.pr_type = SOCK_RAW,
316 	.pr_domain = &inetdomain,
317 	.pr_protocol = IPPROTO_IGMP,
318 	.pr_flags = PR_ATOMIC|PR_ADDR|PR_LASTHDR,
319 	.pr_input = igmp_input,
320 	.pr_output = rip_output,
321 	.pr_ctloutput = rip_ctloutput,
322 	.pr_ctlinput = rip_ctlinput,
323 	.pr_usrreq = rip_usrreq,
324 	.pr_fasttimo = igmp_fasttimo,
325 	.pr_slowtimo = igmp_slowtimo,
326 	.pr_init = igmp_init,
327 },
328 #ifdef PIM
329 {	.pr_type = SOCK_RAW,
330 	.pr_domain = &inetdomain,
331 	.pr_protocol = IPPROTO_PIM,
332 	.pr_flags = PR_ATOMIC|PR_ADDR|PR_LASTHDR,
333 	.pr_input = pim_input,
334 	.pr_output = rip_output,
335 	.pr_ctloutput = rip_ctloutput,
336 	.pr_ctlinput = rip_ctlinput,
337 	.pr_usrreq = rip_usrreq,
338 },
339 #endif /* PIM */
340 /* raw wildcard */
341 {	.pr_type = SOCK_RAW,
342 	.pr_domain = &inetdomain,
343 	.pr_flags = PR_ATOMIC|PR_ADDR|PR_LASTHDR,
344 	.pr_input = rip_input,
345 	.pr_output = rip_output,
346 	.pr_ctloutput = rip_ctloutput,
347 	.pr_ctlinput = rip_ctlinput,
348 	.pr_usrreq = rip_usrreq,
349 	.pr_init = rip_init,
350 },
351 };
352 
353 extern struct ifqueue ipintrq;
354 
355 const struct sockaddr_in in_any = {
356 	  .sin_len = sizeof(struct sockaddr_in)
357 	, .sin_family = AF_INET
358 	, .sin_port = 0
359 	, .sin_addr = {.s_addr = 0 /* INADDR_ANY */}
360 };
361 
362 struct domain inetdomain = {
363 	.dom_family = PF_INET, .dom_name = "internet", .dom_init = NULL,
364 	.dom_externalize = NULL, .dom_dispose = NULL,
365 	.dom_protosw = inetsw,
366 	.dom_protoswNPROTOSW = &inetsw[__arraycount(inetsw)],
367 	.dom_rtattach = rt_inithead,
368 	.dom_rtoffset = 32,
369 	.dom_maxrtkey = sizeof(struct ip_pack4),
370 #ifdef IPSELSRC
371 	.dom_ifattach = in_domifattach,
372 	.dom_ifdetach = in_domifdetach,
373 #else
374 	.dom_ifattach = NULL,
375 	.dom_ifdetach = NULL,
376 #endif
377 	.dom_ifqueues = { &ipintrq, NULL },
378 	.dom_link = { NULL },
379 	.dom_mowner = MOWNER_INIT("",""),
380 	.dom_sa_cmpofs = offsetof(struct sockaddr_in, sin_addr),
381 	.dom_sa_cmplen = sizeof(struct in_addr),
382 	.dom_sa_any = (const struct sockaddr *)&in_any,
383 	.dom_sockaddr_const_addr = sockaddr_in_const_addr,
384 	.dom_sockaddr_addr = sockaddr_in_addr,
385 	.dom_rtcache = LIST_HEAD_INITIALIZER(inetdomain.dom_rtcache)
386 };
387 
388 u_char	ip_protox[IPPROTO_MAX];
389 
390 int icmperrppslim = 100;			/* 100pps */
391 
392 static void
393 sockaddr_in_addrlen(const struct sockaddr *sa, socklen_t *slenp)
394 {
395 	socklen_t slen;
396 
397 	if (slenp == NULL)
398 		return;
399 
400 	slen = sockaddr_getlen(sa);
401 	*slenp = (socklen_t)MIN(sizeof(struct in_addr),
402 	    slen - MIN(slen, offsetof(struct sockaddr_in, sin_addr)));
403 }
404 
405 const void *
406 sockaddr_in_const_addr(const struct sockaddr *sa, socklen_t *slenp)
407 {
408 	const struct sockaddr_in *sin;
409 
410 	sockaddr_in_addrlen(sa, slenp);
411 	sin = (const struct sockaddr_in *)sa;
412 	return &sin->sin_addr;
413 }
414 
415 void *
416 sockaddr_in_addr(struct sockaddr *sa, socklen_t *slenp)
417 {
418 	struct sockaddr_in *sin;
419 
420 	sockaddr_in_addrlen(sa, slenp);
421 	sin = (struct sockaddr_in *)sa;
422 	return &sin->sin_addr;
423 }
424 
425 int
426 sockaddr_in_cmp(const struct sockaddr *sa1, const struct sockaddr *sa2)
427 {
428 	uint_fast8_t len;
429 	const uint_fast8_t addrofs = offsetof(struct sockaddr_in, sin_addr),
430 			   addrend = addrofs + sizeof(struct in_addr);
431 	int rc;
432 	const struct sockaddr_in *sin1, *sin2;
433 
434 	sin1 = satocsin(sa1);
435 	sin2 = satocsin(sa2);
436 
437 	len = MIN(addrend, MIN(sin1->sin_len, sin2->sin_len));
438 
439 	if (len > addrofs &&
440 	     (rc = memcmp(&sin1->sin_addr, &sin2->sin_addr,
441 	                  len - addrofs)) != 0)
442 		return rc;
443 
444 	return sin1->sin_len - sin2->sin_len;
445 }
446