1 /*-
2 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. Neither the name of the project nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 *
29 * $KAME: in6_pcb.c,v 1.31 2001/05/21 05:45:10 jinmei Exp $
30 */
31
32 /*-
33 * Copyright (c) 1982, 1986, 1991, 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 * 4. 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_pcb.c 8.2 (Berkeley) 1/4/94
61 */
62
63 #include <sys/cdefs.h>
64 __FBSDID("$FreeBSD: src/sys/netinet6/in6_pcb.c,v 1.106 2008/12/02 21:37:28 bz Exp $");
65
66 #include "opt_inet.h"
67 #include "opt_inet6.h"
68 #include "opt_ipsec.h"
69 #include "opt_mac.h"
70
71 #include <sys/param.h>
72 #include <sys/systm.h>
73 #include <sys/malloc.h>
74 #include <sys/mbuf.h>
75 #include <sys/domain.h>
76 #include <sys/protosw.h>
77 #include <sys/socket.h>
78 #include <sys/socketvar.h>
79 #include <sys/sockio.h>
80 #include <sys/errno.h>
81 #include <sys/time.h>
82 #include <sys/priv.h>
83 #include <sys/proc.h>
84 #include <sys/jail.h>
85 #include <sys/vimage.h>
86
87 #include <vm/uma.h>
88
89 #include <net/if.h>
90 #include <net/if_types.h>
91 #include <net/route.h>
92
93 #include <netinet/in.h>
94 #include <netinet/in_var.h>
95 #include <netinet/in_systm.h>
96 #include <netinet/tcp_var.h>
97 #include <netinet/ip6.h>
98 #include <netinet/ip_var.h>
99 #include <netinet/vinet.h>
100
101 #include <netinet6/ip6_var.h>
102 #include <netinet6/nd6.h>
103 #include <netinet/in_pcb.h>
104 #include <netinet6/in6_pcb.h>
105 #include <netinet6/scope6_var.h>
106 #include <netinet6/vinet6.h>
107
108 #include <security/mac/mac_framework.h>
109
110 struct in6_addr zeroin6_addr;
111
112 int
113 in6_pcbbind(register struct inpcb *inp, struct sockaddr *nam,
114 struct ucred *cred)
115 {
116 INIT_VNET_INET6(inp->inp_vnet);
117 INIT_VNET_INET(inp->inp_vnet);
118 struct socket *so = inp->inp_socket;
119 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)NULL;
120 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
121 u_short lport = 0;
122 int wild = 0, reuseport = (so->so_options & SO_REUSEPORT);
123
124 INP_INFO_WLOCK_ASSERT(pcbinfo);
125 INP_WLOCK_ASSERT(inp);
126
127 if (!V_in6_ifaddr) /* XXX broken! */
128 return (EADDRNOTAVAIL);
129 if (inp->inp_lport || !IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
130 return (EINVAL);
131 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) == 0)
132 wild = INPLOOKUP_WILDCARD;
133 if (nam) {
134 int error;
135
136 sin6 = (struct sockaddr_in6 *)nam;
137 if (nam->sa_len != sizeof(*sin6))
138 return (EINVAL);
139 /*
140 * family check.
141 */
142 if (nam->sa_family != AF_INET6)
143 return (EAFNOSUPPORT);
144
145 if ((error = sa6_embedscope(sin6, V_ip6_use_defzone)) != 0)
146 return(error);
147
148 if (prison_local_ip6(cred, &sin6->sin6_addr,
149 ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0)) != 0)
150 return (EINVAL);
151
152 lport = sin6->sin6_port;
153 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) {
154 /*
155 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
156 * allow compepte duplication of binding if
157 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
158 * and a multicast address is bound on both
159 * new and duplicated sockets.
160 */
161 if (so->so_options & SO_REUSEADDR)
162 reuseport = SO_REUSEADDR|SO_REUSEPORT;
163 } else if (!IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) {
164 struct ifaddr *ia = NULL;
165
166 sin6->sin6_port = 0; /* yech... */
167 if ((ia = ifa_ifwithaddr((struct sockaddr *)sin6)) == 0)
168 return (EADDRNOTAVAIL);
169
170 /*
171 * XXX: bind to an anycast address might accidentally
172 * cause sending a packet with anycast source address.
173 * We should allow to bind to a deprecated address, since
174 * the application dares to use it.
175 */
176 if (ia &&
177 ((struct in6_ifaddr *)ia)->ia6_flags &
178 (IN6_IFF_ANYCAST|IN6_IFF_NOTREADY|IN6_IFF_DETACHED)) {
179 return (EADDRNOTAVAIL);
180 }
181 }
182 if (lport) {
183 struct inpcb *t;
184
185 /* GROSS */
186 if (ntohs(lport) <= V_ipport_reservedhigh &&
187 ntohs(lport) >= V_ipport_reservedlow &&
188 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT,
189 0))
190 return (EACCES);
191 if (!IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr) &&
192 priv_check_cred(inp->inp_cred,
193 PRIV_NETINET_REUSEPORT, 0) != 0) {
194 t = in6_pcblookup_local(pcbinfo,
195 &sin6->sin6_addr, lport,
196 INPLOOKUP_WILDCARD, cred);
197 if (t &&
198 ((t->inp_vflag & INP_TIMEWAIT) == 0) &&
199 (so->so_type != SOCK_STREAM ||
200 IN6_IS_ADDR_UNSPECIFIED(&t->in6p_faddr)) &&
201 (!IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr) ||
202 !IN6_IS_ADDR_UNSPECIFIED(&t->in6p_laddr) ||
203 (t->inp_socket->so_options & SO_REUSEPORT)
204 == 0) && (inp->inp_cred->cr_uid !=
205 t->inp_cred->cr_uid))
206 return (EADDRINUSE);
207 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0 &&
208 IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) {
209 struct sockaddr_in sin;
210
211 in6_sin6_2_sin(&sin, sin6);
212 t = in_pcblookup_local(pcbinfo,
213 sin.sin_addr, lport,
214 INPLOOKUP_WILDCARD, cred);
215 if (t &&
216 ((t->inp_vflag &
217 INP_TIMEWAIT) == 0) &&
218 (so->so_type != SOCK_STREAM ||
219 ntohl(t->inp_faddr.s_addr) ==
220 INADDR_ANY) &&
221 (inp->inp_cred->cr_uid !=
222 t->inp_cred->cr_uid))
223 return (EADDRINUSE);
224 }
225 }
226 if (prison_local_ip6(cred, &sin6->sin6_addr,
227 ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0)) != 0)
228 return (EADDRNOTAVAIL);
229 t = in6_pcblookup_local(pcbinfo, &sin6->sin6_addr,
230 lport, wild, cred);
231 if (t && (reuseport & ((t->inp_vflag & INP_TIMEWAIT) ?
232 intotw(t)->tw_so_options :
233 t->inp_socket->so_options)) == 0)
234 return (EADDRINUSE);
235 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0 &&
236 IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) {
237 struct sockaddr_in sin;
238
239 in6_sin6_2_sin(&sin, sin6);
240 t = in_pcblookup_local(pcbinfo, sin.sin_addr,
241 lport, wild, cred);
242 if (t && t->inp_vflag & INP_TIMEWAIT) {
243 if ((reuseport &
244 intotw(t)->tw_so_options) == 0 &&
245 (ntohl(t->inp_laddr.s_addr) !=
246 INADDR_ANY || ((inp->inp_vflag &
247 INP_IPV6PROTO) ==
248 (t->inp_vflag & INP_IPV6PROTO))))
249 return (EADDRINUSE);
250 }
251 else if (t &&
252 (reuseport & t->inp_socket->so_options)
253 == 0 && (ntohl(t->inp_laddr.s_addr) !=
254 INADDR_ANY || INP_SOCKAF(so) ==
255 INP_SOCKAF(t->inp_socket)))
256 return (EADDRINUSE);
257 }
258 }
259 inp->in6p_laddr = sin6->sin6_addr;
260 }
261 if (prison_local_ip6(cred, &inp->in6p_laddr,
262 ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0)) != 0)
263 return (EINVAL);
264 if (lport == 0) {
265 int e;
266 if ((e = in6_pcbsetport(&inp->in6p_laddr, inp, cred)) != 0)
267 return (e);
268 } else {
269 inp->inp_lport = lport;
270 if (in_pcbinshash(inp) != 0) {
271 inp->in6p_laddr = in6addr_any;
272 inp->inp_lport = 0;
273 return (EAGAIN);
274 }
275 }
276 return (0);
277 }
278
279 /*
280 * Transform old in6_pcbconnect() into an inner subroutine for new
281 * in6_pcbconnect(): Do some validity-checking on the remote
282 * address (in mbuf 'nam') and then determine local host address
283 * (i.e., which interface) to use to access that remote host.
284 *
285 * This preserves definition of in6_pcbconnect(), while supporting a
286 * slightly different version for T/TCP. (This is more than
287 * a bit of a kludge, but cleaning up the internal interfaces would
288 * have forced minor changes in every protocol).
289 */
290 int
291 in6_pcbladdr(register struct inpcb *inp, struct sockaddr *nam,
292 struct in6_addr **plocal_addr6)
293 {
294 INIT_VNET_INET6(inp->inp_vnet);
295 register struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
296 int error = 0;
297 struct ifnet *ifp = NULL;
298 int scope_ambiguous = 0;
299
300 INP_INFO_WLOCK_ASSERT(inp->inp_pcbinfo);
301 INP_WLOCK_ASSERT(inp);
302
303 if (nam->sa_len != sizeof (*sin6))
304 return (EINVAL);
305 if (sin6->sin6_family != AF_INET6)
306 return (EAFNOSUPPORT);
307 if (sin6->sin6_port == 0)
308 return (EADDRNOTAVAIL);
309
310 if (sin6->sin6_scope_id == 0 && !V_ip6_use_defzone)
311 scope_ambiguous = 1;
312 if ((error = sa6_embedscope(sin6, V_ip6_use_defzone)) != 0)
313 return(error);
314
315 if (V_in6_ifaddr) {
316 /*
317 * If the destination address is UNSPECIFIED addr,
318 * use the loopback addr, e.g ::1.
319 */
320 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
321 sin6->sin6_addr = in6addr_loopback;
322 }
323 if (prison_remote_ip6(inp->inp_cred, &sin6->sin6_addr) != 0)
324 return (EADDRNOTAVAIL);
325
326 /*
327 * XXX: in6_selectsrc might replace the bound local address
328 * with the address specified by setsockopt(IPV6_PKTINFO).
329 * Is it the intended behavior?
330 */
331 *plocal_addr6 = in6_selectsrc(sin6, inp->in6p_outputopts,
332 inp, NULL,
333 inp->inp_cred,
334 &ifp, &error);
335 if (ifp && scope_ambiguous &&
336 (error = in6_setscope(&sin6->sin6_addr, ifp, NULL)) != 0) {
337 return(error);
338 }
339
340 if (*plocal_addr6 == NULL) {
341 if (error == 0)
342 error = EADDRNOTAVAIL;
343 return (error);
344 }
345 /*
346 * Don't do pcblookup call here; return interface in
347 * plocal_addr6
348 * and exit to caller, that will do the lookup.
349 */
350
351 return (0);
352 }
353
354 /*
355 * Outer subroutine:
356 * Connect from a socket to a specified address.
357 * Both address and port must be specified in argument sin.
358 * If don't have a local address for this socket yet,
359 * then pick one.
360 */
361 int
362 in6_pcbconnect(register struct inpcb *inp, struct sockaddr *nam,
363 struct ucred *cred)
364 {
365 struct in6_addr *addr6;
366 register struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
367 int error;
368
369 INP_INFO_WLOCK_ASSERT(inp->inp_pcbinfo);
370 INP_WLOCK_ASSERT(inp);
371
372 /*
373 * Call inner routine, to assign local interface address.
374 * in6_pcbladdr() may automatically fill in sin6_scope_id.
375 */
376 if ((error = in6_pcbladdr(inp, nam, &addr6)) != 0)
377 return (error);
378
379 if (in6_pcblookup_hash(inp->inp_pcbinfo, &sin6->sin6_addr,
380 sin6->sin6_port,
381 IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)
382 ? addr6 : &inp->in6p_laddr,
383 inp->inp_lport, 0, NULL) != NULL) {
384 return (EADDRINUSE);
385 }
386 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) {
387 if (inp->inp_lport == 0) {
388 error = in6_pcbbind(inp, (struct sockaddr *)0, cred);
389 if (error)
390 return (error);
391 }
392 inp->in6p_laddr = *addr6;
393 }
394 inp->in6p_faddr = sin6->sin6_addr;
395 inp->inp_fport = sin6->sin6_port;
396 /* update flowinfo - draft-itojun-ipv6-flowlabel-api-00 */
397 inp->in6p_flowinfo &= ~IPV6_FLOWLABEL_MASK;
398 if (inp->in6p_flags & IN6P_AUTOFLOWLABEL)
399 inp->in6p_flowinfo |=
400 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
401
402 in_pcbrehash(inp);
403
404 return (0);
405 }
406
407 void
408 in6_pcbdisconnect(struct inpcb *inp)
409 {
410
411 INP_INFO_WLOCK_ASSERT(inp->inp_pcbinfo);
412 INP_WLOCK_ASSERT(inp);
413
414 bzero((caddr_t)&inp->in6p_faddr, sizeof(inp->in6p_faddr));
415 inp->inp_fport = 0;
416 /* clear flowinfo - draft-itojun-ipv6-flowlabel-api-00 */
417 inp->in6p_flowinfo &= ~IPV6_FLOWLABEL_MASK;
418 in_pcbrehash(inp);
419 }
420
421 struct sockaddr *
422 in6_sockaddr(in_port_t port, struct in6_addr *addr_p)
423 {
424 struct sockaddr_in6 *sin6;
425
426 sin6 = malloc(sizeof *sin6, M_SONAME, M_WAITOK);
427 bzero(sin6, sizeof *sin6);
428 sin6->sin6_family = AF_INET6;
429 sin6->sin6_len = sizeof(*sin6);
430 sin6->sin6_port = port;
431 sin6->sin6_addr = *addr_p;
432 (void)sa6_recoverscope(sin6); /* XXX: should catch errors */
433
434 return (struct sockaddr *)sin6;
435 }
436
437 struct sockaddr *
438 in6_v4mapsin6_sockaddr(in_port_t port, struct in_addr *addr_p)
439 {
440 struct sockaddr_in sin;
441 struct sockaddr_in6 *sin6_p;
442
443 bzero(&sin, sizeof sin);
444 sin.sin_family = AF_INET;
445 sin.sin_len = sizeof(sin);
446 sin.sin_port = port;
447 sin.sin_addr = *addr_p;
448
449 sin6_p = malloc(sizeof *sin6_p, M_SONAME,
450 M_WAITOK);
451 in6_sin_2_v4mapsin6(&sin, sin6_p);
452
453 return (struct sockaddr *)sin6_p;
454 }
455
456 int
457 in6_getsockaddr(struct socket *so, struct sockaddr **nam)
458 {
459 register struct inpcb *inp;
460 struct in6_addr addr;
461 in_port_t port;
462
463 inp = sotoinpcb(so);
464 KASSERT(inp != NULL, ("in6_getsockaddr: inp == NULL"));
465
466 INP_RLOCK(inp);
467 port = inp->inp_lport;
468 addr = inp->in6p_laddr;
469 INP_RUNLOCK(inp);
470
471 *nam = in6_sockaddr(port, &addr);
472 return 0;
473 }
474
475 int
476 in6_getpeeraddr(struct socket *so, struct sockaddr **nam)
477 {
478 struct inpcb *inp;
479 struct in6_addr addr;
480 in_port_t port;
481
482 inp = sotoinpcb(so);
483 KASSERT(inp != NULL, ("in6_getpeeraddr: inp == NULL"));
484
485 INP_RLOCK(inp);
486 port = inp->inp_fport;
487 addr = inp->in6p_faddr;
488 INP_RUNLOCK(inp);
489
490 *nam = in6_sockaddr(port, &addr);
491 return 0;
492 }
493
494 int
495 in6_mapped_sockaddr(struct socket *so, struct sockaddr **nam)
496 {
497 struct inpcb *inp;
498 int error;
499
500 inp = sotoinpcb(so);
501 KASSERT(inp != NULL, ("in6_mapped_sockaddr: inp == NULL"));
502
503 if ((inp->inp_vflag & (INP_IPV4 | INP_IPV6)) == INP_IPV4) {
504 error = in_getsockaddr(so, nam);
505 if (error == 0)
506 in6_sin_2_v4mapsin6_in_sock(nam);
507 } else {
508 /* scope issues will be handled in in6_getsockaddr(). */
509 error = in6_getsockaddr(so, nam);
510 }
511
512 return error;
513 }
514
515 int
516 in6_mapped_peeraddr(struct socket *so, struct sockaddr **nam)
517 {
518 struct inpcb *inp;
519 int error;
520
521 inp = sotoinpcb(so);
522 KASSERT(inp != NULL, ("in6_mapped_peeraddr: inp == NULL"));
523
524 if ((inp->inp_vflag & (INP_IPV4 | INP_IPV6)) == INP_IPV4) {
525 error = in_getpeeraddr(so, nam);
526 if (error == 0)
527 in6_sin_2_v4mapsin6_in_sock(nam);
528 } else
529 /* scope issues will be handled in in6_getpeeraddr(). */
530 error = in6_getpeeraddr(so, nam);
531
532 return error;
533 }
534
535 /*
536 * Pass some notification to all connections of a protocol
537 * associated with address dst. The local address and/or port numbers
538 * may be specified to limit the search. The "usual action" will be
539 * taken, depending on the ctlinput cmd. The caller must filter any
540 * cmds that are uninteresting (e.g., no error in the map).
541 * Call the protocol specific routine (if any) to report
542 * any errors for each matching socket.
543 */
544 void
545 in6_pcbnotify(struct inpcbinfo *pcbinfo, struct sockaddr *dst,
546 u_int fport_arg, const struct sockaddr *src, u_int lport_arg,
547 int cmd, void *cmdarg,
548 struct inpcb *(*notify)(struct inpcb *, int))
549 {
550 struct inpcb *inp, *inp_temp;
551 struct sockaddr_in6 sa6_src, *sa6_dst;
552 u_short fport = fport_arg, lport = lport_arg;
553 u_int32_t flowinfo;
554 int errno;
555
556 if ((unsigned)cmd >= PRC_NCMDS || dst->sa_family != AF_INET6)
557 return;
558
559 sa6_dst = (struct sockaddr_in6 *)dst;
560 if (IN6_IS_ADDR_UNSPECIFIED(&sa6_dst->sin6_addr))
561 return;
562
563 /*
564 * note that src can be NULL when we get notify by local fragmentation.
565 */
566 sa6_src = (src == NULL) ? sa6_any : *(const struct sockaddr_in6 *)src;
567 flowinfo = sa6_src.sin6_flowinfo;
568
569 /*
570 * Redirects go to all references to the destination,
571 * and use in6_rtchange to invalidate the route cache.
572 * Dead host indications: also use in6_rtchange to invalidate
573 * the cache, and deliver the error to all the sockets.
574 * Otherwise, if we have knowledge of the local port and address,
575 * deliver only to that socket.
576 */
577 if (PRC_IS_REDIRECT(cmd) || cmd == PRC_HOSTDEAD) {
578 fport = 0;
579 lport = 0;
580 bzero((caddr_t)&sa6_src.sin6_addr, sizeof(sa6_src.sin6_addr));
581
582 if (cmd != PRC_HOSTDEAD)
583 notify = in6_rtchange;
584 }
585 errno = inet6ctlerrmap[cmd];
586 INP_INFO_WLOCK(pcbinfo);
587 LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) {
588 INP_WLOCK(inp);
589 if ((inp->inp_vflag & INP_IPV6) == 0) {
590 INP_WUNLOCK(inp);
591 continue;
592 }
593
594 /*
595 * If the error designates a new path MTU for a destination
596 * and the application (associated with this socket) wanted to
597 * know the value, notify. Note that we notify for all
598 * disconnected sockets if the corresponding application
599 * wanted. This is because some UDP applications keep sending
600 * sockets disconnected.
601 * XXX: should we avoid to notify the value to TCP sockets?
602 */
603 if (cmd == PRC_MSGSIZE && (inp->inp_flags & IN6P_MTU) != 0 &&
604 (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr) ||
605 IN6_ARE_ADDR_EQUAL(&inp->in6p_faddr, &sa6_dst->sin6_addr))) {
606 ip6_notify_pmtu(inp, (struct sockaddr_in6 *)dst,
607 (u_int32_t *)cmdarg);
608 }
609
610 /*
611 * Detect if we should notify the error. If no source and
612 * destination ports are specifed, but non-zero flowinfo and
613 * local address match, notify the error. This is the case
614 * when the error is delivered with an encrypted buffer
615 * by ESP. Otherwise, just compare addresses and ports
616 * as usual.
617 */
618 if (lport == 0 && fport == 0 && flowinfo &&
619 inp->inp_socket != NULL &&
620 flowinfo == (inp->in6p_flowinfo & IPV6_FLOWLABEL_MASK) &&
621 IN6_ARE_ADDR_EQUAL(&inp->in6p_laddr, &sa6_src.sin6_addr))
622 goto do_notify;
623 else if (!IN6_ARE_ADDR_EQUAL(&inp->in6p_faddr,
624 &sa6_dst->sin6_addr) ||
625 inp->inp_socket == 0 ||
626 (lport && inp->inp_lport != lport) ||
627 (!IN6_IS_ADDR_UNSPECIFIED(&sa6_src.sin6_addr) &&
628 !IN6_ARE_ADDR_EQUAL(&inp->in6p_laddr,
629 &sa6_src.sin6_addr)) ||
630 (fport && inp->inp_fport != fport)) {
631 INP_WUNLOCK(inp);
632 continue;
633 }
634
635 do_notify:
636 if (notify) {
637 if ((*notify)(inp, errno))
638 INP_WUNLOCK(inp);
639 } else
640 INP_WUNLOCK(inp);
641 }
642 INP_INFO_WUNLOCK(pcbinfo);
643 }
644
645 /*
646 * Lookup a PCB based on the local address and port.
647 */
648 struct inpcb *
649 in6_pcblookup_local(struct inpcbinfo *pcbinfo, struct in6_addr *laddr,
650 u_short lport, int wild_okay, struct ucred *cred)
651 {
652 register struct inpcb *inp;
653 int matchwild = 3, wildcard;
654
655 INP_INFO_WLOCK_ASSERT(pcbinfo);
656
657 if (!wild_okay) {
658 struct inpcbhead *head;
659 /*
660 * Look for an unconnected (wildcard foreign addr) PCB that
661 * matches the local address and port we're looking for.
662 */
663 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
664 0, pcbinfo->ipi_hashmask)];
665 LIST_FOREACH(inp, head, inp_hash) {
666 /* XXX inp locking */
667 if ((inp->inp_vflag & INP_IPV6) == 0)
668 continue;
669 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr) &&
670 IN6_ARE_ADDR_EQUAL(&inp->in6p_laddr, laddr) &&
671 inp->inp_lport == lport) {
672 /* Found. */
673 if (cred == NULL ||
674 inp->inp_cred->cr_prison == cred->cr_prison)
675 return (inp);
676 }
677 }
678 /*
679 * Not found.
680 */
681 return (NULL);
682 } else {
683 struct inpcbporthead *porthash;
684 struct inpcbport *phd;
685 struct inpcb *match = NULL;
686 /*
687 * Best fit PCB lookup.
688 *
689 * First see if this local port is in use by looking on the
690 * port hash list.
691 */
692 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
693 pcbinfo->ipi_porthashmask)];
694 LIST_FOREACH(phd, porthash, phd_hash) {
695 if (phd->phd_port == lport)
696 break;
697 }
698 if (phd != NULL) {
699 /*
700 * Port is in use by one or more PCBs. Look for best
701 * fit.
702 */
703 LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
704 wildcard = 0;
705 if (cred != NULL &&
706 inp->inp_cred->cr_prison != cred->cr_prison)
707 continue;
708 /* XXX inp locking */
709 if ((inp->inp_vflag & INP_IPV6) == 0)
710 continue;
711 if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr))
712 wildcard++;
713 if (!IN6_IS_ADDR_UNSPECIFIED(
714 &inp->in6p_laddr)) {
715 if (IN6_IS_ADDR_UNSPECIFIED(laddr))
716 wildcard++;
717 else if (!IN6_ARE_ADDR_EQUAL(
718 &inp->in6p_laddr, laddr))
719 continue;
720 } else {
721 if (!IN6_IS_ADDR_UNSPECIFIED(laddr))
722 wildcard++;
723 }
724 if (wildcard < matchwild) {
725 match = inp;
726 matchwild = wildcard;
727 if (matchwild == 0)
728 break;
729 }
730 }
731 }
732 return (match);
733 }
734 }
735
736 void
737 in6_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
738 {
739 struct in6pcb *in6p;
740 struct ip6_moptions *im6o;
741 struct in6_multi_mship *imm, *nimm;
742
743 INP_INFO_RLOCK(pcbinfo);
744 LIST_FOREACH(in6p, pcbinfo->ipi_listhead, inp_list) {
745 INP_WLOCK(in6p);
746 im6o = in6p->in6p_moptions;
747 if ((in6p->inp_vflag & INP_IPV6) &&
748 im6o) {
749 /*
750 * Unselect the outgoing interface if it is being
751 * detached.
752 */
753 if (im6o->im6o_multicast_ifp == ifp)
754 im6o->im6o_multicast_ifp = NULL;
755
756 /*
757 * Drop multicast group membership if we joined
758 * through the interface being detached.
759 * XXX controversial - is it really legal for kernel
760 * to force this?
761 */
762 for (imm = im6o->im6o_memberships.lh_first;
763 imm != NULL; imm = nimm) {
764 nimm = imm->i6mm_chain.le_next;
765 if (imm->i6mm_maddr->in6m_ifp == ifp) {
766 LIST_REMOVE(imm, i6mm_chain);
767 in6_delmulti(imm->i6mm_maddr);
768 free(imm, M_IP6MADDR);
769 }
770 }
771 }
772 INP_WUNLOCK(in6p);
773 }
774 INP_INFO_RUNLOCK(pcbinfo);
775 }
776
777 /*
778 * Check for alternatives when higher level complains
779 * about service problems. For now, invalidate cached
780 * routing information. If the route was created dynamically
781 * (by a redirect), time to try a default gateway again.
782 */
783 void
784 in6_losing(struct inpcb *in6p)
785 {
786
787 /*
788 * We don't store route pointers in the routing table anymore
789 */
790 return;
791 }
792
793 /*
794 * After a routing change, flush old routing
795 * and allocate a (hopefully) better one.
796 */
797 struct inpcb *
798 in6_rtchange(struct inpcb *inp, int errno)
799 {
800 /*
801 * We don't store route pointers in the routing table anymore
802 */
803 return inp;
804 }
805
806 /*
807 * Lookup PCB in hash list.
808 */
809 struct inpcb *
810 in6_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in6_addr *faddr,
811 u_int fport_arg, struct in6_addr *laddr, u_int lport_arg, int wildcard,
812 struct ifnet *ifp)
813 {
814 struct inpcbhead *head;
815 struct inpcb *inp, *tmpinp;
816 u_short fport = fport_arg, lport = lport_arg;
817 int faith;
818
819 INP_INFO_LOCK_ASSERT(pcbinfo);
820
821 if (faithprefix_p != NULL)
822 faith = (*faithprefix_p)(laddr);
823 else
824 faith = 0;
825
826 /*
827 * First look for an exact match.
828 */
829 tmpinp = NULL;
830 head = &pcbinfo->ipi_hashbase[
831 INP_PCBHASH(faddr->s6_addr32[3] /* XXX */, lport, fport,
832 pcbinfo->ipi_hashmask)];
833 LIST_FOREACH(inp, head, inp_hash) {
834 /* XXX inp locking */
835 if ((inp->inp_vflag & INP_IPV6) == 0)
836 continue;
837 if (IN6_ARE_ADDR_EQUAL(&inp->in6p_faddr, faddr) &&
838 IN6_ARE_ADDR_EQUAL(&inp->in6p_laddr, laddr) &&
839 inp->inp_fport == fport &&
840 inp->inp_lport == lport) {
841 /*
842 * XXX We should be able to directly return
843 * the inp here, without any checks.
844 * Well unless both bound with SO_REUSEPORT?
845 */
846 if (jailed(inp->inp_cred))
847 return (inp);
848 if (tmpinp == NULL)
849 tmpinp = inp;
850 }
851 }
852 if (tmpinp != NULL)
853 return (tmpinp);
854
855 /*
856 * Then look for a wildcard match, if requested.
857 */
858 if (wildcard == INPLOOKUP_WILDCARD) {
859 struct inpcb *local_wild = NULL, *local_exact = NULL;
860 struct inpcb *jail_wild = NULL;
861 int injail;
862
863 /*
864 * Order of socket selection - we always prefer jails.
865 * 1. jailed, non-wild.
866 * 2. jailed, wild.
867 * 3. non-jailed, non-wild.
868 * 4. non-jailed, wild.
869 */
870 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
871 0, pcbinfo->ipi_hashmask)];
872 LIST_FOREACH(inp, head, inp_hash) {
873 /* XXX inp locking */
874 if ((inp->inp_vflag & INP_IPV6) == 0)
875 continue;
876
877 if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr) ||
878 inp->inp_lport != lport) {
879 continue;
880 }
881
882 /* XXX inp locking */
883 if (faith && (inp->inp_flags & INP_FAITH) == 0)
884 continue;
885
886 injail = jailed(inp->inp_cred);
887 if (injail) {
888 if (!prison_check_ip6(inp->inp_cred, laddr))
889 continue;
890 } else {
891 if (local_exact != NULL)
892 continue;
893 }
894
895 if (IN6_ARE_ADDR_EQUAL(&inp->in6p_laddr, laddr)) {
896 if (injail)
897 return (inp);
898 else
899 local_exact = inp;
900 } else if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) {
901 if (injail)
902 jail_wild = inp;
903 else
904 local_wild = inp;
905 }
906 } /* LIST_FOREACH */
907
908 if (jail_wild != NULL)
909 return (jail_wild);
910 if (local_exact != NULL)
911 return (local_exact);
912 if (local_wild != NULL)
913 return (local_wild);
914 } /* if (wildcard == INPLOOKUP_WILDCARD) */
915
916 /*
917 * Not found.
918 */
919 return (NULL);
920 }
921
922 void
923 init_sin6(struct sockaddr_in6 *sin6, struct mbuf *m)
924 {
925 struct ip6_hdr *ip;
926
927 ip = mtod(m, struct ip6_hdr *);
928 bzero(sin6, sizeof(*sin6));
929 sin6->sin6_len = sizeof(*sin6);
930 sin6->sin6_family = AF_INET6;
931 sin6->sin6_addr = ip->ip6_src;
932
933 (void)sa6_recoverscope(sin6); /* XXX: should catch errors... */
934
935 return;
936 }
937
|