]> CyberLeo.Net >> Repos - FreeBSD/releng/9.0.git/blob - sys/netinet6/udp6_usrreq.c
Copy stable/9 to releng/9.0 as part of the FreeBSD 9.0-RELEASE release
[FreeBSD/releng/9.0.git] / sys / netinet6 / udp6_usrreq.c
1 /*-
2  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
3  * Copyright (c) 2010-2011 Juniper Networks, Inc.
4  * All rights reserved.
5  *
6  * Portions of this software were developed by Robert N. M. Watson under
7  * contract to Juniper Networks, Inc.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  * 3. Neither the name of the project nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  *      $KAME: udp6_usrreq.c,v 1.27 2001/05/21 05:45:10 jinmei Exp $
34  *      $KAME: udp6_output.c,v 1.31 2001/05/21 16:39:15 jinmei Exp $
35  */
36
37 /*-
38  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
39  *      The Regents of the University of California.
40  * All rights reserved.
41  *
42  * Redistribution and use in source and binary forms, with or without
43  * modification, are permitted provided that the following conditions
44  * are met:
45  * 1. Redistributions of source code must retain the above copyright
46  *    notice, this list of conditions and the following disclaimer.
47  * 2. Redistributions in binary form must reproduce the above copyright
48  *    notice, this list of conditions and the following disclaimer in the
49  *    documentation and/or other materials provided with the distribution.
50  * 4. Neither the name of the University nor the names of its contributors
51  *    may be used to endorse or promote products derived from this software
52  *    without specific prior written permission.
53  *
54  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64  * SUCH DAMAGE.
65  *
66  *      @(#)udp_usrreq.c        8.6 (Berkeley) 5/23/95
67  */
68
69 #include <sys/cdefs.h>
70 __FBSDID("$FreeBSD$");
71
72 #include "opt_inet.h"
73 #include "opt_inet6.h"
74 #include "opt_ipfw.h"
75 #include "opt_ipsec.h"
76
77 #include <sys/param.h>
78 #include <sys/jail.h>
79 #include <sys/kernel.h>
80 #include <sys/lock.h>
81 #include <sys/mbuf.h>
82 #include <sys/priv.h>
83 #include <sys/proc.h>
84 #include <sys/protosw.h>
85 #include <sys/signalvar.h>
86 #include <sys/socket.h>
87 #include <sys/socketvar.h>
88 #include <sys/sx.h>
89 #include <sys/sysctl.h>
90 #include <sys/syslog.h>
91 #include <sys/systm.h>
92
93 #include <net/if.h>
94 #include <net/if_types.h>
95 #include <net/route.h>
96
97 #include <netinet/in.h>
98 #include <netinet/in_pcb.h>
99 #include <netinet/in_systm.h>
100 #include <netinet/in_var.h>
101 #include <netinet/ip.h>
102 #include <netinet/ip_icmp.h>
103 #include <netinet/ip6.h>
104 #include <netinet/icmp_var.h>
105 #include <netinet/icmp6.h>
106 #include <netinet/ip_var.h>
107 #include <netinet/udp.h>
108 #include <netinet/udp_var.h>
109
110 #include <netinet6/ip6protosw.h>
111 #include <netinet6/ip6_var.h>
112 #include <netinet6/in6_pcb.h>
113 #include <netinet6/udp6_var.h>
114 #include <netinet6/scope6_var.h>
115
116 #ifdef IPSEC
117 #include <netipsec/ipsec.h>
118 #include <netipsec/ipsec6.h>
119 #endif /* IPSEC */
120
121 #include <security/mac/mac_framework.h>
122
123 /*
124  * UDP protocol implementation.
125  * Per RFC 768, August, 1980.
126  */
127
128 extern struct protosw   inetsw[];
129 static void             udp6_detach(struct socket *so);
130
131 static void
132 udp6_append(struct inpcb *inp, struct mbuf *n, int off,
133     struct sockaddr_in6 *fromsa)
134 {
135         struct socket *so;
136         struct mbuf *opts;
137
138         INP_LOCK_ASSERT(inp);
139
140 #ifdef IPSEC
141         /* Check AH/ESP integrity. */
142         if (ipsec6_in_reject(n, inp)) {
143                 m_freem(n);
144                 V_ipsec6stat.in_polvio++;
145                 return;
146         }
147 #endif /* IPSEC */
148 #ifdef MAC
149         if (mac_inpcb_check_deliver(inp, n) != 0) {
150                 m_freem(n);
151                 return;
152         }
153 #endif
154         opts = NULL;
155         if (inp->inp_flags & INP_CONTROLOPTS ||
156             inp->inp_socket->so_options & SO_TIMESTAMP)
157                 ip6_savecontrol(inp, n, &opts);
158         m_adj(n, off + sizeof(struct udphdr));
159
160         so = inp->inp_socket;
161         SOCKBUF_LOCK(&so->so_rcv);
162         if (sbappendaddr_locked(&so->so_rcv, (struct sockaddr *)fromsa, n,
163             opts) == 0) {
164                 SOCKBUF_UNLOCK(&so->so_rcv);
165                 m_freem(n);
166                 if (opts)
167                         m_freem(opts);
168                 UDPSTAT_INC(udps_fullsock);
169         } else
170                 sorwakeup_locked(so);
171 }
172
173 int
174 udp6_input(struct mbuf **mp, int *offp, int proto)
175 {
176         struct mbuf *m = *mp;
177         struct ifnet *ifp;
178         struct ip6_hdr *ip6;
179         struct udphdr *uh;
180         struct inpcb *inp;
181         struct udpcb *up;
182         int off = *offp;
183         int plen, ulen;
184         struct sockaddr_in6 fromsa;
185 #ifdef IPFIREWALL_FORWARD
186         struct m_tag *fwd_tag;
187 #endif
188
189         ifp = m->m_pkthdr.rcvif;
190         ip6 = mtod(m, struct ip6_hdr *);
191
192         if (faithprefix_p != NULL && (*faithprefix_p)(&ip6->ip6_dst)) {
193                 /* XXX send icmp6 host/port unreach? */
194                 m_freem(m);
195                 return (IPPROTO_DONE);
196         }
197
198 #ifndef PULLDOWN_TEST
199         IP6_EXTHDR_CHECK(m, off, sizeof(struct udphdr), IPPROTO_DONE);
200         ip6 = mtod(m, struct ip6_hdr *);
201         uh = (struct udphdr *)((caddr_t)ip6 + off);
202 #else
203         IP6_EXTHDR_GET(uh, struct udphdr *, m, off, sizeof(*uh));
204         if (!uh)
205                 return (IPPROTO_DONE);
206 #endif
207
208         UDPSTAT_INC(udps_ipackets);
209
210         /*
211          * Destination port of 0 is illegal, based on RFC768.
212          */
213         if (uh->uh_dport == 0)
214                 goto badunlocked;
215
216         plen = ntohs(ip6->ip6_plen) - off + sizeof(*ip6);
217         ulen = ntohs((u_short)uh->uh_ulen);
218
219         if (plen != ulen) {
220                 UDPSTAT_INC(udps_badlen);
221                 goto badunlocked;
222         }
223
224         /*
225          * Checksum extended UDP header and data.
226          */
227         if (uh->uh_sum == 0) {
228                 UDPSTAT_INC(udps_nosum);
229                 goto badunlocked;
230         }
231         if (in6_cksum(m, IPPROTO_UDP, off, ulen) != 0) {
232                 UDPSTAT_INC(udps_badsum);
233                 goto badunlocked;
234         }
235
236         /*
237          * Construct sockaddr format source address.
238          */
239         init_sin6(&fromsa, m);
240         fromsa.sin6_port = uh->uh_sport;
241
242         if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) {
243                 struct inpcb *last;
244                 struct ip6_moptions *imo;
245
246                 INP_INFO_RLOCK(&V_udbinfo);
247                 /*
248                  * In the event that laddr should be set to the link-local
249                  * address (this happens in RIPng), the multicast address
250                  * specified in the received packet will not match laddr.  To
251                  * handle this situation, matching is relaxed if the
252                  * receiving interface is the same as one specified in the
253                  * socket and if the destination multicast address matches
254                  * one of the multicast groups specified in the socket.
255                  */
256
257                 /*
258                  * KAME note: traditionally we dropped udpiphdr from mbuf
259                  * here.  We need udphdr for IPsec processing so we do that
260                  * later.
261                  */
262                 last = NULL;
263                 LIST_FOREACH(inp, &V_udb, inp_list) {
264                         if ((inp->inp_vflag & INP_IPV6) == 0)
265                                 continue;
266                         if (inp->inp_lport != uh->uh_dport)
267                                 continue;
268                         if (inp->inp_fport != 0 &&
269                             inp->inp_fport != uh->uh_sport)
270                                 continue;
271                         if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) {
272                                 if (!IN6_ARE_ADDR_EQUAL(&inp->in6p_laddr,
273                                                         &ip6->ip6_dst))
274                                         continue;
275                         }
276                         if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
277                                 if (!IN6_ARE_ADDR_EQUAL(&inp->in6p_faddr,
278                                                         &ip6->ip6_src) ||
279                                     inp->inp_fport != uh->uh_sport)
280                                         continue;
281                         }
282
283                         /*
284                          * XXXRW: Because we weren't holding either the inpcb
285                          * or the hash lock when we checked for a match 
286                          * before, we should probably recheck now that the 
287                          * inpcb lock is (supposed to be) held.
288                          */
289
290                         /*
291                          * Handle socket delivery policy for any-source
292                          * and source-specific multicast. [RFC3678]
293                          */
294                         imo = inp->in6p_moptions;
295                         if (imo && IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) {
296                                 struct sockaddr_in6      mcaddr;
297                                 int                      blocked;
298
299                                 INP_RLOCK(inp);
300
301                                 bzero(&mcaddr, sizeof(struct sockaddr_in6));
302                                 mcaddr.sin6_len = sizeof(struct sockaddr_in6);
303                                 mcaddr.sin6_family = AF_INET6;
304                                 mcaddr.sin6_addr = ip6->ip6_dst;
305
306                                 blocked = im6o_mc_filter(imo, ifp,
307                                         (struct sockaddr *)&mcaddr,
308                                         (struct sockaddr *)&fromsa);
309                                 if (blocked != MCAST_PASS) {
310                                         if (blocked == MCAST_NOTGMEMBER)
311                                                 IP6STAT_INC(ip6s_notmember);
312                                         if (blocked == MCAST_NOTSMEMBER ||
313                                             blocked == MCAST_MUTED)
314                                                 UDPSTAT_INC(udps_filtermcast);
315                                         INP_RUNLOCK(inp); /* XXX */
316                                         continue;
317                                 }
318
319                                 INP_RUNLOCK(inp);
320                         }
321                         if (last != NULL) {
322                                 struct mbuf *n;
323
324                                 if ((n = m_copy(m, 0, M_COPYALL)) != NULL) {
325                                         INP_RLOCK(last);
326                                         up = intoudpcb(last);
327                                         if (up->u_tun_func == NULL) {
328                                                 udp6_append(last, n, off, &fromsa);
329                                         } else {
330                                                 /*
331                                                  * Engage the tunneling
332                                                  * protocol we will have to
333                                                  * leave the info_lock up,
334                                                  * since we are hunting
335                                                  * through multiple UDP's.
336                                                  * 
337                                                  */
338                                                 (*up->u_tun_func)(n, off, last);
339                                         }
340                                         INP_RUNLOCK(last);
341                                 }
342                         }
343                         last = inp;
344                         /*
345                          * Don't look for additional matches if this one does
346                          * not have either the SO_REUSEPORT or SO_REUSEADDR
347                          * socket options set.  This heuristic avoids
348                          * searching through all pcbs in the common case of a
349                          * non-shared port.  It assumes that an application
350                          * will never clear these options after setting them.
351                          */
352                         if ((last->inp_socket->so_options &
353                              (SO_REUSEPORT|SO_REUSEADDR)) == 0)
354                                 break;
355                 }
356
357                 if (last == NULL) {
358                         /*
359                          * No matching pcb found; discard datagram.  (No need
360                          * to send an ICMP Port Unreachable for a broadcast
361                          * or multicast datgram.)
362                          */
363                         UDPSTAT_INC(udps_noport);
364                         UDPSTAT_INC(udps_noportmcast);
365                         goto badheadlocked;
366                 }
367                 INP_RLOCK(last);
368                 INP_INFO_RUNLOCK(&V_udbinfo);
369                 up = intoudpcb(last);
370                 if (up->u_tun_func == NULL) {
371                         udp6_append(last, m, off, &fromsa);
372                 } else {
373                         /*
374                          * Engage the tunneling protocol.
375                          */
376                         (*up->u_tun_func)(m, off, last);
377                 }
378                 INP_RUNLOCK(last);
379                 return (IPPROTO_DONE);
380         }
381         /*
382          * Locate pcb for datagram.
383          */
384 #ifdef IPFIREWALL_FORWARD
385         /*
386          * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
387          */
388         fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
389         if (fwd_tag != NULL) {
390                 struct sockaddr_in6 *next_hop6;
391
392                 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
393
394                 /*
395                  * Transparently forwarded. Pretend to be the destination.
396                  * Already got one like this?
397                  */
398                 inp = in6_pcblookup_mbuf(&V_udbinfo,
399                     &ip6->ip6_src, uh->uh_sport, &ip6->ip6_dst, uh->uh_dport,
400                     INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif, m);
401                 if (!inp) {
402                         /*
403                          * It's new.  Try to find the ambushing socket.
404                          * Because we've rewritten the destination address,
405                          * any hardware-generated hash is ignored.
406                          */
407                         inp = in6_pcblookup(&V_udbinfo, &ip6->ip6_src,
408                             uh->uh_sport, &next_hop6->sin6_addr,
409                             next_hop6->sin6_port ? htons(next_hop6->sin6_port) :
410                             uh->uh_dport, INPLOOKUP_WILDCARD |
411                             INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif);
412                 }
413                 /* Remove the tag from the packet. We don't need it anymore. */
414                 m_tag_delete(m, fwd_tag);
415         } else
416 #endif /* IPFIREWALL_FORWARD */
417                 inp = in6_pcblookup_mbuf(&V_udbinfo, &ip6->ip6_src,
418                     uh->uh_sport, &ip6->ip6_dst, uh->uh_dport,
419                     INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB,
420                     m->m_pkthdr.rcvif, m);
421         if (inp == NULL) {
422                 if (udp_log_in_vain) {
423                         char ip6bufs[INET6_ADDRSTRLEN];
424                         char ip6bufd[INET6_ADDRSTRLEN];
425
426                         log(LOG_INFO,
427                             "Connection attempt to UDP [%s]:%d from [%s]:%d\n",
428                             ip6_sprintf(ip6bufd, &ip6->ip6_dst),
429                             ntohs(uh->uh_dport),
430                             ip6_sprintf(ip6bufs, &ip6->ip6_src),
431                             ntohs(uh->uh_sport));
432                 }
433                 UDPSTAT_INC(udps_noport);
434                 if (m->m_flags & M_MCAST) {
435                         printf("UDP6: M_MCAST is set in a unicast packet.\n");
436                         UDPSTAT_INC(udps_noportmcast);
437                         goto badunlocked;
438                 }
439                 if (V_udp_blackhole)
440                         goto badunlocked;
441                 if (badport_bandlim(BANDLIM_ICMP6_UNREACH) < 0)
442                         goto badunlocked;
443                 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_NOPORT, 0);
444                 return (IPPROTO_DONE);
445         }
446         INP_RLOCK_ASSERT(inp);
447         up = intoudpcb(inp);
448         if (up->u_tun_func == NULL) {
449                 udp6_append(inp, m, off, &fromsa);
450         } else {
451                 /*
452                  * Engage the tunneling protocol.
453                  */
454
455                 (*up->u_tun_func)(m, off, inp);
456         }
457         INP_RUNLOCK(inp);
458         return (IPPROTO_DONE);
459
460 badheadlocked:
461         INP_INFO_RUNLOCK(&V_udbinfo);
462 badunlocked:
463         if (m)
464                 m_freem(m);
465         return (IPPROTO_DONE);
466 }
467
468 void
469 udp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
470 {
471         struct udphdr uh;
472         struct ip6_hdr *ip6;
473         struct mbuf *m;
474         int off = 0;
475         struct ip6ctlparam *ip6cp = NULL;
476         const struct sockaddr_in6 *sa6_src = NULL;
477         void *cmdarg;
478         struct inpcb *(*notify)(struct inpcb *, int) = udp_notify;
479         struct udp_portonly {
480                 u_int16_t uh_sport;
481                 u_int16_t uh_dport;
482         } *uhp;
483
484         if (sa->sa_family != AF_INET6 ||
485             sa->sa_len != sizeof(struct sockaddr_in6))
486                 return;
487
488         if ((unsigned)cmd >= PRC_NCMDS)
489                 return;
490         if (PRC_IS_REDIRECT(cmd))
491                 notify = in6_rtchange, d = NULL;
492         else if (cmd == PRC_HOSTDEAD)
493                 d = NULL;
494         else if (inet6ctlerrmap[cmd] == 0)
495                 return;
496
497         /* if the parameter is from icmp6, decode it. */
498         if (d != NULL) {
499                 ip6cp = (struct ip6ctlparam *)d;
500                 m = ip6cp->ip6c_m;
501                 ip6 = ip6cp->ip6c_ip6;
502                 off = ip6cp->ip6c_off;
503                 cmdarg = ip6cp->ip6c_cmdarg;
504                 sa6_src = ip6cp->ip6c_src;
505         } else {
506                 m = NULL;
507                 ip6 = NULL;
508                 cmdarg = NULL;
509                 sa6_src = &sa6_any;
510         }
511
512         if (ip6) {
513                 /*
514                  * XXX: We assume that when IPV6 is non NULL,
515                  * M and OFF are valid.
516                  */
517
518                 /* Check if we can safely examine src and dst ports. */
519                 if (m->m_pkthdr.len < off + sizeof(*uhp))
520                         return;
521
522                 bzero(&uh, sizeof(uh));
523                 m_copydata(m, off, sizeof(*uhp), (caddr_t)&uh);
524
525                 (void) in6_pcbnotify(&V_udbinfo, sa, uh.uh_dport,
526                     (struct sockaddr *)ip6cp->ip6c_src, uh.uh_sport, cmd,
527                     cmdarg, notify);
528         } else
529                 (void) in6_pcbnotify(&V_udbinfo, sa, 0,
530                     (const struct sockaddr *)sa6_src, 0, cmd, cmdarg, notify);
531 }
532
533 static int
534 udp6_getcred(SYSCTL_HANDLER_ARGS)
535 {
536         struct xucred xuc;
537         struct sockaddr_in6 addrs[2];
538         struct inpcb *inp;
539         int error;
540
541         error = priv_check(req->td, PRIV_NETINET_GETCRED);
542         if (error)
543                 return (error);
544
545         if (req->newlen != sizeof(addrs))
546                 return (EINVAL);
547         if (req->oldlen != sizeof(struct xucred))
548                 return (EINVAL);
549         error = SYSCTL_IN(req, addrs, sizeof(addrs));
550         if (error)
551                 return (error);
552         if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
553             (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
554                 return (error);
555         }
556         inp = in6_pcblookup(&V_udbinfo, &addrs[1].sin6_addr,
557             addrs[1].sin6_port, &addrs[0].sin6_addr, addrs[0].sin6_port,
558             INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
559         if (inp != NULL) {
560                 INP_RLOCK_ASSERT(inp);
561                 if (inp->inp_socket == NULL)
562                         error = ENOENT;
563                 if (error == 0)
564                         error = cr_canseesocket(req->td->td_ucred,
565                             inp->inp_socket);
566                 if (error == 0)
567                         cru2x(inp->inp_cred, &xuc);
568                 INP_RUNLOCK(inp);
569         } else
570                 error = ENOENT;
571         if (error == 0)
572                 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
573         return (error);
574 }
575
576 SYSCTL_PROC(_net_inet6_udp6, OID_AUTO, getcred, CTLTYPE_OPAQUE|CTLFLAG_RW, 0,
577     0, udp6_getcred, "S,xucred", "Get the xucred of a UDP6 connection");
578
579 static int
580 udp6_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr6,
581     struct mbuf *control, struct thread *td)
582 {
583         u_int32_t ulen = m->m_pkthdr.len;
584         u_int32_t plen = sizeof(struct udphdr) + ulen;
585         struct ip6_hdr *ip6;
586         struct udphdr *udp6;
587         struct in6_addr *laddr, *faddr, in6a;
588         struct sockaddr_in6 *sin6 = NULL;
589         struct ifnet *oifp = NULL;
590         int scope_ambiguous = 0;
591         u_short fport;
592         int error = 0;
593         struct ip6_pktopts *optp, opt;
594         int af = AF_INET6, hlen = sizeof(struct ip6_hdr);
595         int flags;
596         struct sockaddr_in6 tmp;
597
598         INP_WLOCK_ASSERT(inp);
599         INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
600
601         if (addr6) {
602                 /* addr6 has been validated in udp6_send(). */
603                 sin6 = (struct sockaddr_in6 *)addr6;
604
605                 /* protect *sin6 from overwrites */
606                 tmp = *sin6;
607                 sin6 = &tmp;
608
609                 /*
610                  * Application should provide a proper zone ID or the use of
611                  * default zone IDs should be enabled.  Unfortunately, some
612                  * applications do not behave as it should, so we need a
613                  * workaround.  Even if an appropriate ID is not determined,
614                  * we'll see if we can determine the outgoing interface.  If we
615                  * can, determine the zone ID based on the interface below.
616                  */
617                 if (sin6->sin6_scope_id == 0 && !V_ip6_use_defzone)
618                         scope_ambiguous = 1;
619                 if ((error = sa6_embedscope(sin6, V_ip6_use_defzone)) != 0)
620                         return (error);
621         }
622
623         if (control) {
624                 if ((error = ip6_setpktopts(control, &opt,
625                     inp->in6p_outputopts, td->td_ucred, IPPROTO_UDP)) != 0)
626                         goto release;
627                 optp = &opt;
628         } else
629                 optp = inp->in6p_outputopts;
630
631         if (sin6) {
632                 faddr = &sin6->sin6_addr;
633
634                 /*
635                  * IPv4 version of udp_output calls in_pcbconnect in this case,
636                  * which needs splnet and affects performance.
637                  * Since we saw no essential reason for calling in_pcbconnect,
638                  * we get rid of such kind of logic, and call in6_selectsrc
639                  * and in6_pcbsetport in order to fill in the local address
640                  * and the local port.
641                  */
642                 if (sin6->sin6_port == 0) {
643                         error = EADDRNOTAVAIL;
644                         goto release;
645                 }
646
647                 if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
648                         /* how about ::ffff:0.0.0.0 case? */
649                         error = EISCONN;
650                         goto release;
651                 }
652
653                 fport = sin6->sin6_port; /* allow 0 port */
654
655                 if (IN6_IS_ADDR_V4MAPPED(faddr)) {
656                         if ((inp->inp_flags & IN6P_IPV6_V6ONLY)) {
657                                 /*
658                                  * I believe we should explicitly discard the
659                                  * packet when mapped addresses are disabled,
660                                  * rather than send the packet as an IPv6 one.
661                                  * If we chose the latter approach, the packet
662                                  * might be sent out on the wire based on the
663                                  * default route, the situation which we'd
664                                  * probably want to avoid.
665                                  * (20010421 jinmei@kame.net)
666                                  */
667                                 error = EINVAL;
668                                 goto release;
669                         }
670                         if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) &&
671                             !IN6_IS_ADDR_V4MAPPED(&inp->in6p_laddr)) {
672                                 /*
673                                  * when remote addr is an IPv4-mapped address,
674                                  * local addr should not be an IPv6 address,
675                                  * since you cannot determine how to map IPv6
676                                  * source address to IPv4.
677                                  */
678                                 error = EINVAL;
679                                 goto release;
680                         }
681
682                         af = AF_INET;
683                 }
684
685                 if (!IN6_IS_ADDR_V4MAPPED(faddr)) {
686                         error = in6_selectsrc(sin6, optp, inp, NULL,
687                             td->td_ucred, &oifp, &in6a);
688                         if (error)
689                                 goto release;
690                         if (oifp && scope_ambiguous &&
691                             (error = in6_setscope(&sin6->sin6_addr,
692                             oifp, NULL))) {
693                                 goto release;
694                         }
695                         laddr = &in6a;
696                 } else
697                         laddr = &inp->in6p_laddr;       /* XXX */
698                 if (laddr == NULL) {
699                         if (error == 0)
700                                 error = EADDRNOTAVAIL;
701                         goto release;
702                 }
703                 if (inp->inp_lport == 0 &&
704                     (error = in6_pcbsetport(laddr, inp, td->td_ucred)) != 0) {
705                         /* Undo an address bind that may have occurred. */
706                         inp->in6p_laddr = in6addr_any;
707                         goto release;
708                 }
709         } else {
710                 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
711                         error = ENOTCONN;
712                         goto release;
713                 }
714                 if (IN6_IS_ADDR_V4MAPPED(&inp->in6p_faddr)) {
715                         if ((inp->inp_flags & IN6P_IPV6_V6ONLY)) {
716                                 /*
717                                  * XXX: this case would happen when the
718                                  * application sets the V6ONLY flag after
719                                  * connecting the foreign address.
720                                  * Such applications should be fixed,
721                                  * so we bark here.
722                                  */
723                                 log(LOG_INFO, "udp6_output: IPV6_V6ONLY "
724                                     "option was set for a connected socket\n");
725                                 error = EINVAL;
726                                 goto release;
727                         } else
728                                 af = AF_INET;
729                 }
730                 laddr = &inp->in6p_laddr;
731                 faddr = &inp->in6p_faddr;
732                 fport = inp->inp_fport;
733         }
734
735         if (af == AF_INET)
736                 hlen = sizeof(struct ip);
737
738         /*
739          * Calculate data length and get a mbuf
740          * for UDP and IP6 headers.
741          */
742         M_PREPEND(m, hlen + sizeof(struct udphdr), M_DONTWAIT);
743         if (m == 0) {
744                 error = ENOBUFS;
745                 goto release;
746         }
747
748         /*
749          * Stuff checksum and output datagram.
750          */
751         udp6 = (struct udphdr *)(mtod(m, caddr_t) + hlen);
752         udp6->uh_sport = inp->inp_lport; /* lport is always set in the PCB */
753         udp6->uh_dport = fport;
754         if (plen <= 0xffff)
755                 udp6->uh_ulen = htons((u_short)plen);
756         else
757                 udp6->uh_ulen = 0;
758         udp6->uh_sum = 0;
759
760         switch (af) {
761         case AF_INET6:
762                 ip6 = mtod(m, struct ip6_hdr *);
763                 ip6->ip6_flow   = inp->inp_flow & IPV6_FLOWINFO_MASK;
764                 ip6->ip6_vfc    &= ~IPV6_VERSION_MASK;
765                 ip6->ip6_vfc    |= IPV6_VERSION;
766 #if 0                           /* ip6_plen will be filled in ip6_output. */
767                 ip6->ip6_plen   = htons((u_short)plen);
768 #endif
769                 ip6->ip6_nxt    = IPPROTO_UDP;
770                 ip6->ip6_hlim   = in6_selecthlim(inp, NULL);
771                 ip6->ip6_src    = *laddr;
772                 ip6->ip6_dst    = *faddr;
773
774                 if ((udp6->uh_sum = in6_cksum(m, IPPROTO_UDP,
775                                 sizeof(struct ip6_hdr), plen)) == 0) {
776                         udp6->uh_sum = 0xffff;
777                 }
778
779                 flags = 0;
780
781                 UDPSTAT_INC(udps_opackets);
782                 error = ip6_output(m, optp, NULL, flags, inp->in6p_moptions,
783                     NULL, inp);
784                 break;
785         case AF_INET:
786                 error = EAFNOSUPPORT;
787                 goto release;
788         }
789         goto releaseopt;
790
791 release:
792         m_freem(m);
793
794 releaseopt:
795         if (control) {
796                 ip6_clearpktopts(&opt, -1);
797                 m_freem(control);
798         }
799         return (error);
800 }
801
802 static void
803 udp6_abort(struct socket *so)
804 {
805         struct inpcb *inp;
806
807         inp = sotoinpcb(so);
808         KASSERT(inp != NULL, ("udp6_abort: inp == NULL"));
809
810 #ifdef INET
811         if (inp->inp_vflag & INP_IPV4) {
812                 struct pr_usrreqs *pru;
813
814                 pru = inetsw[ip_protox[IPPROTO_UDP]].pr_usrreqs;
815                 (*pru->pru_abort)(so);
816                 return;
817         }
818 #endif
819
820         INP_WLOCK(inp);
821         if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
822                 INP_HASH_WLOCK(&V_udbinfo);
823                 in6_pcbdisconnect(inp);
824                 inp->in6p_laddr = in6addr_any;
825                 INP_HASH_WUNLOCK(&V_udbinfo);
826                 soisdisconnected(so);
827         }
828         INP_WUNLOCK(inp);
829 }
830
831 static int
832 udp6_attach(struct socket *so, int proto, struct thread *td)
833 {
834         struct inpcb *inp;
835         int error;
836
837         inp = sotoinpcb(so);
838         KASSERT(inp == NULL, ("udp6_attach: inp != NULL"));
839
840         if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
841                 error = soreserve(so, udp_sendspace, udp_recvspace);
842                 if (error)
843                         return (error);
844         }
845         INP_INFO_WLOCK(&V_udbinfo);
846         error = in_pcballoc(so, &V_udbinfo);
847         if (error) {
848                 INP_INFO_WUNLOCK(&V_udbinfo);
849                 return (error);
850         }
851         inp = (struct inpcb *)so->so_pcb;
852         inp->inp_vflag |= INP_IPV6;
853         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
854                 inp->inp_vflag |= INP_IPV4;
855         inp->in6p_hops = -1;    /* use kernel default */
856         inp->in6p_cksum = -1;   /* just to be sure */
857         /*
858          * XXX: ugly!!
859          * IPv4 TTL initialization is necessary for an IPv6 socket as well,
860          * because the socket may be bound to an IPv6 wildcard address,
861          * which may match an IPv4-mapped IPv6 address.
862          */
863         inp->inp_ip_ttl = V_ip_defttl;
864
865         error = udp_newudpcb(inp);
866         if (error) {
867                 in_pcbdetach(inp);
868                 in_pcbfree(inp);
869                 INP_INFO_WUNLOCK(&V_udbinfo);
870                 return (error);
871         }
872         INP_WUNLOCK(inp);
873         INP_INFO_WUNLOCK(&V_udbinfo);
874         return (0);
875 }
876
877 static int
878 udp6_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
879 {
880         struct inpcb *inp;
881         int error;
882
883         inp = sotoinpcb(so);
884         KASSERT(inp != NULL, ("udp6_bind: inp == NULL"));
885
886         INP_WLOCK(inp);
887         INP_HASH_WLOCK(&V_udbinfo);
888         inp->inp_vflag &= ~INP_IPV4;
889         inp->inp_vflag |= INP_IPV6;
890         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
891                 struct sockaddr_in6 *sin6_p;
892
893                 sin6_p = (struct sockaddr_in6 *)nam;
894
895                 if (IN6_IS_ADDR_UNSPECIFIED(&sin6_p->sin6_addr))
896                         inp->inp_vflag |= INP_IPV4;
897 #ifdef INET
898                 else if (IN6_IS_ADDR_V4MAPPED(&sin6_p->sin6_addr)) {
899                         struct sockaddr_in sin;
900
901                         in6_sin6_2_sin(&sin, sin6_p);
902                         inp->inp_vflag |= INP_IPV4;
903                         inp->inp_vflag &= ~INP_IPV6;
904                         error = in_pcbbind(inp, (struct sockaddr *)&sin,
905                             td->td_ucred);
906                         goto out;
907                 }
908 #endif
909         }
910
911         error = in6_pcbbind(inp, nam, td->td_ucred);
912 #ifdef INET
913 out:
914 #endif
915         INP_HASH_WUNLOCK(&V_udbinfo);
916         INP_WUNLOCK(inp);
917         return (error);
918 }
919
920 static void
921 udp6_close(struct socket *so)
922 {
923         struct inpcb *inp;
924
925         inp = sotoinpcb(so);
926         KASSERT(inp != NULL, ("udp6_close: inp == NULL"));
927
928 #ifdef INET
929         if (inp->inp_vflag & INP_IPV4) {
930                 struct pr_usrreqs *pru;
931
932                 pru = inetsw[ip_protox[IPPROTO_UDP]].pr_usrreqs;
933                 (*pru->pru_disconnect)(so);
934                 return;
935         }
936 #endif
937         INP_WLOCK(inp);
938         if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
939                 INP_HASH_WLOCK(&V_udbinfo);
940                 in6_pcbdisconnect(inp);
941                 inp->in6p_laddr = in6addr_any;
942                 INP_HASH_WUNLOCK(&V_udbinfo);
943                 soisdisconnected(so);
944         }
945         INP_WUNLOCK(inp);
946 }
947
948 static int
949 udp6_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
950 {
951         struct inpcb *inp;
952         struct sockaddr_in6 *sin6;
953         int error;
954
955         inp = sotoinpcb(so);
956         sin6 = (struct sockaddr_in6 *)nam;
957         KASSERT(inp != NULL, ("udp6_connect: inp == NULL"));
958
959         /*
960          * XXXRW: Need to clarify locking of v4/v6 flags.
961          */
962         INP_WLOCK(inp);
963 #ifdef INET
964         if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
965                 struct sockaddr_in sin;
966
967                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
968                         error = EINVAL;
969                         goto out;
970                 }
971                 if (inp->inp_faddr.s_addr != INADDR_ANY) {
972                         error = EISCONN;
973                         goto out;
974                 }
975                 in6_sin6_2_sin(&sin, sin6);
976                 inp->inp_vflag |= INP_IPV4;
977                 inp->inp_vflag &= ~INP_IPV6;
978                 error = prison_remote_ip4(td->td_ucred, &sin.sin_addr);
979                 if (error != 0)
980                         goto out;
981                 INP_HASH_WLOCK(&V_udbinfo);
982                 error = in_pcbconnect(inp, (struct sockaddr *)&sin,
983                     td->td_ucred);
984                 INP_HASH_WUNLOCK(&V_udbinfo);
985                 if (error == 0)
986                         soisconnected(so);
987                 goto out;
988         }
989 #endif
990         if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
991                 error = EISCONN;
992                 goto out;
993         }
994         inp->inp_vflag &= ~INP_IPV4;
995         inp->inp_vflag |= INP_IPV6;
996         error = prison_remote_ip6(td->td_ucred, &sin6->sin6_addr);
997         if (error != 0)
998                 goto out;
999         INP_HASH_WLOCK(&V_udbinfo);
1000         error = in6_pcbconnect(inp, nam, td->td_ucred);
1001         INP_HASH_WUNLOCK(&V_udbinfo);
1002         if (error == 0)
1003                 soisconnected(so);
1004 out:
1005         INP_WUNLOCK(inp);
1006         return (error);
1007 }
1008
1009 static void
1010 udp6_detach(struct socket *so)
1011 {
1012         struct inpcb *inp;
1013         struct udpcb *up;
1014
1015         inp = sotoinpcb(so);
1016         KASSERT(inp != NULL, ("udp6_detach: inp == NULL"));
1017
1018         INP_INFO_WLOCK(&V_udbinfo);
1019         INP_WLOCK(inp);
1020         up = intoudpcb(inp);
1021         KASSERT(up != NULL, ("%s: up == NULL", __func__));
1022         in_pcbdetach(inp);
1023         in_pcbfree(inp);
1024         INP_INFO_WUNLOCK(&V_udbinfo);
1025         udp_discardcb(up);
1026 }
1027
1028 static int
1029 udp6_disconnect(struct socket *so)
1030 {
1031         struct inpcb *inp;
1032         int error;
1033
1034         inp = sotoinpcb(so);
1035         KASSERT(inp != NULL, ("udp6_disconnect: inp == NULL"));
1036
1037 #ifdef INET
1038         if (inp->inp_vflag & INP_IPV4) {
1039                 struct pr_usrreqs *pru;
1040
1041                 pru = inetsw[ip_protox[IPPROTO_UDP]].pr_usrreqs;
1042                 (void)(*pru->pru_disconnect)(so);
1043                 return (0);
1044         }
1045 #endif
1046
1047         INP_WLOCK(inp);
1048
1049         if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
1050                 error = ENOTCONN;
1051                 goto out;
1052         }
1053
1054         INP_HASH_WLOCK(&V_udbinfo);
1055         in6_pcbdisconnect(inp);
1056         inp->in6p_laddr = in6addr_any;
1057         INP_HASH_WUNLOCK(&V_udbinfo);
1058         SOCK_LOCK(so);
1059         so->so_state &= ~SS_ISCONNECTED;                /* XXX */
1060         SOCK_UNLOCK(so);
1061 out:
1062         INP_WUNLOCK(inp);
1063         return (0);
1064 }
1065
1066 static int
1067 udp6_send(struct socket *so, int flags, struct mbuf *m,
1068     struct sockaddr *addr, struct mbuf *control, struct thread *td)
1069 {
1070         struct inpcb *inp;
1071         int error = 0;
1072
1073         inp = sotoinpcb(so);
1074         KASSERT(inp != NULL, ("udp6_send: inp == NULL"));
1075
1076         INP_WLOCK(inp);
1077         if (addr) {
1078                 if (addr->sa_len != sizeof(struct sockaddr_in6)) {
1079                         error = EINVAL;
1080                         goto bad;
1081                 }
1082                 if (addr->sa_family != AF_INET6) {
1083                         error = EAFNOSUPPORT;
1084                         goto bad;
1085                 }
1086         }
1087
1088 #ifdef INET
1089         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
1090                 int hasv4addr;
1091                 struct sockaddr_in6 *sin6 = 0;
1092
1093                 if (addr == 0)
1094                         hasv4addr = (inp->inp_vflag & INP_IPV4);
1095                 else {
1096                         sin6 = (struct sockaddr_in6 *)addr;
1097                         hasv4addr = IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)
1098                             ? 1 : 0;
1099                 }
1100                 if (hasv4addr) {
1101                         struct pr_usrreqs *pru;
1102
1103                         /*
1104                          * XXXRW: We release UDP-layer locks before calling
1105                          * udp_send() in order to avoid recursion.  However,
1106                          * this does mean there is a short window where inp's
1107                          * fields are unstable.  Could this lead to a
1108                          * potential race in which the factors causing us to
1109                          * select the UDPv4 output routine are invalidated?
1110                          */
1111                         INP_WUNLOCK(inp);
1112                         if (sin6)
1113                                 in6_sin6_2_sin_in_sock(addr);
1114                         pru = inetsw[ip_protox[IPPROTO_UDP]].pr_usrreqs;
1115                         /* addr will just be freed in sendit(). */
1116                         return ((*pru->pru_send)(so, flags, m, addr, control,
1117                             td));
1118                 }
1119         }
1120 #endif
1121 #ifdef MAC
1122         mac_inpcb_create_mbuf(inp, m);
1123 #endif
1124         INP_HASH_WLOCK(&V_udbinfo);
1125         error = udp6_output(inp, m, addr, control, td);
1126         INP_HASH_WUNLOCK(&V_udbinfo);
1127 #ifdef INET
1128 #endif  
1129         INP_WUNLOCK(inp);
1130         return (error);
1131
1132 bad:
1133         INP_WUNLOCK(inp);
1134         m_freem(m);
1135         return (error);
1136 }
1137
1138 struct pr_usrreqs udp6_usrreqs = {
1139         .pru_abort =            udp6_abort,
1140         .pru_attach =           udp6_attach,
1141         .pru_bind =             udp6_bind,
1142         .pru_connect =          udp6_connect,
1143         .pru_control =          in6_control,
1144         .pru_detach =           udp6_detach,
1145         .pru_disconnect =       udp6_disconnect,
1146         .pru_peeraddr =         in6_mapped_peeraddr,
1147         .pru_send =             udp6_send,
1148         .pru_shutdown =         udp_shutdown,
1149         .pru_sockaddr =         in6_mapped_sockaddr,
1150         .pru_soreceive =        soreceive_dgram,
1151         .pru_sosend =           sosend_dgram,
1152         .pru_sosetlabel =       in_pcbsosetlabel,
1153         .pru_close =            udp6_close
1154 };