]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/blob - sys/netinet6/udp6_usrreq.c
OpenZFS: MFV 2.0-rc3-gfc5966
[FreeBSD/FreeBSD.git] / sys / netinet6 / udp6_usrreq.c
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
5  * Copyright (c) 2010-2011 Juniper Networks, Inc.
6  * Copyright (c) 2014 Kevin Lo
7  * All rights reserved.
8  *
9  * Portions of this software were developed by Robert N. M. Watson under
10  * contract to Juniper Networks, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the project nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *      $KAME: udp6_usrreq.c,v 1.27 2001/05/21 05:45:10 jinmei Exp $
37  *      $KAME: udp6_output.c,v 1.31 2001/05/21 16:39:15 jinmei Exp $
38  */
39
40 /*-
41  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
42  *      The Regents of the University of California.
43  * All rights reserved.
44  *
45  * Redistribution and use in source and binary forms, with or without
46  * modification, are permitted provided that the following conditions
47  * are met:
48  * 1. Redistributions of source code must retain the above copyright
49  *    notice, this list of conditions and the following disclaimer.
50  * 2. Redistributions in binary form must reproduce the above copyright
51  *    notice, this list of conditions and the following disclaimer in the
52  *    documentation and/or other materials provided with the distribution.
53  * 3. Neither the name of the University nor the names of its contributors
54  *    may be used to endorse or promote products derived from this software
55  *    without specific prior written permission.
56  *
57  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
58  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
59  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
60  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
61  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
62  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
63  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
64  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
65  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
66  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
67  * SUCH DAMAGE.
68  *
69  *      @(#)udp_usrreq.c        8.6 (Berkeley) 5/23/95
70  */
71
72 #include <sys/cdefs.h>
73 __FBSDID("$FreeBSD$");
74
75 #include "opt_inet.h"
76 #include "opt_inet6.h"
77 #include "opt_ipsec.h"
78 #include "opt_rss.h"
79
80 #include <sys/param.h>
81 #include <sys/jail.h>
82 #include <sys/kernel.h>
83 #include <sys/lock.h>
84 #include <sys/mbuf.h>
85 #include <sys/priv.h>
86 #include <sys/proc.h>
87 #include <sys/protosw.h>
88 #include <sys/sdt.h>
89 #include <sys/signalvar.h>
90 #include <sys/socket.h>
91 #include <sys/socketvar.h>
92 #include <sys/sx.h>
93 #include <sys/sysctl.h>
94 #include <sys/syslog.h>
95 #include <sys/systm.h>
96
97 #include <net/if.h>
98 #include <net/if_var.h>
99 #include <net/if_types.h>
100 #include <net/route.h>
101 #include <net/rss_config.h>
102
103 #include <netinet/in.h>
104 #include <netinet/in_kdtrace.h>
105 #include <netinet/in_pcb.h>
106 #include <netinet/in_systm.h>
107 #include <netinet/in_var.h>
108 #include <netinet/ip.h>
109 #include <netinet/ip6.h>
110 #include <netinet/icmp6.h>
111 #include <netinet/ip_var.h>
112 #include <netinet/udp.h>
113 #include <netinet/udp_var.h>
114 #include <netinet/udplite.h>
115
116 #include <netinet6/ip6protosw.h>
117 #include <netinet6/ip6_var.h>
118 #include <netinet6/in6_pcb.h>
119 #include <netinet6/in6_rss.h>
120 #include <netinet6/udp6_var.h>
121 #include <netinet6/scope6_var.h>
122
123 #include <netipsec/ipsec_support.h>
124
125 #include <security/mac/mac_framework.h>
126
127 VNET_DEFINE(int, zero_checksum_port) = 0;
128 #define V_zero_checksum_port    VNET(zero_checksum_port)
129 SYSCTL_INT(_net_inet6_udp6, OID_AUTO, rfc6935_port, CTLFLAG_VNET | CTLFLAG_RW,
130     &VNET_NAME(zero_checksum_port), 0,
131     "Zero UDP checksum allowed for traffic to/from this port.");
132 /*
133  * UDP protocol implementation.
134  * Per RFC 768, August, 1980.
135  */
136
137 extern struct protosw   inetsw[];
138 static void             udp6_detach(struct socket *so);
139
140 static int
141 udp6_append(struct inpcb *inp, struct mbuf *n, int off,
142     struct sockaddr_in6 *fromsa)
143 {
144         struct socket *so;
145         struct mbuf *opts = NULL, *tmp_opts;
146         struct udpcb *up;
147
148         INP_LOCK_ASSERT(inp);
149
150         /*
151          * Engage the tunneling protocol.
152          */
153         up = intoudpcb(inp);
154         if (up->u_tun_func != NULL) {
155                 in_pcbref(inp);
156                 INP_RUNLOCK(inp);
157                 (*up->u_tun_func)(n, off, inp, (struct sockaddr *)&fromsa[0],
158                     up->u_tun_ctx);
159                 INP_RLOCK(inp);
160                 return (in_pcbrele_rlocked(inp));
161         }
162 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
163         /* Check AH/ESP integrity. */
164         if (IPSEC_ENABLED(ipv6)) {
165                 if (IPSEC_CHECK_POLICY(ipv6, n, inp) != 0) {
166                         m_freem(n);
167                         return (0);
168                 }
169         }
170 #endif /* IPSEC */
171 #ifdef MAC
172         if (mac_inpcb_check_deliver(inp, n) != 0) {
173                 m_freem(n);
174                 return (0);
175         }
176 #endif
177         opts = NULL;
178         if (inp->inp_flags & INP_CONTROLOPTS ||
179             inp->inp_socket->so_options & SO_TIMESTAMP)
180                 ip6_savecontrol(inp, n, &opts);
181         if ((inp->inp_vflag & INP_IPV6) && (inp->inp_flags2 & INP_ORIGDSTADDR)) {
182                 tmp_opts = sbcreatecontrol((caddr_t)&fromsa[1],
183                         sizeof(struct sockaddr_in6), IPV6_ORIGDSTADDR, IPPROTO_IPV6);
184                 if (tmp_opts) {
185                         if (opts) {
186                                 tmp_opts->m_next = opts;
187                                 opts = tmp_opts;
188                         } else
189                                 opts = tmp_opts;
190                 }
191         }
192         m_adj(n, off + sizeof(struct udphdr));
193
194         so = inp->inp_socket;
195         SOCKBUF_LOCK(&so->so_rcv);
196         if (sbappendaddr_locked(&so->so_rcv, (struct sockaddr *)&fromsa[0], n,
197             opts) == 0) {
198                 SOCKBUF_UNLOCK(&so->so_rcv);
199                 m_freem(n);
200                 if (opts)
201                         m_freem(opts);
202                 UDPSTAT_INC(udps_fullsock);
203         } else
204                 sorwakeup_locked(so);
205         return (0);
206 }
207
208 int
209 udp6_input(struct mbuf **mp, int *offp, int proto)
210 {
211         struct mbuf *m = *mp;
212         struct ifnet *ifp;
213         struct ip6_hdr *ip6;
214         struct udphdr *uh;
215         struct inpcb *inp;
216         struct inpcbinfo *pcbinfo;
217         struct udpcb *up;
218         int off = *offp;
219         int cscov_partial;
220         int plen, ulen;
221         struct sockaddr_in6 fromsa[2];
222         struct m_tag *fwd_tag;
223         uint16_t uh_sum;
224         uint8_t nxt;
225
226         NET_EPOCH_ASSERT();
227
228         ifp = m->m_pkthdr.rcvif;
229
230         if (m->m_len < off + sizeof(struct udphdr)) {
231                 m = m_pullup(m, off + sizeof(struct udphdr));
232                 if (m == NULL) {
233                         IP6STAT_INC(ip6s_exthdrtoolong);
234                         *mp = NULL;
235                         return (IPPROTO_DONE);
236                 }
237         }
238         ip6 = mtod(m, struct ip6_hdr *);
239         uh = (struct udphdr *)((caddr_t)ip6 + off);
240
241         UDPSTAT_INC(udps_ipackets);
242
243         /*
244          * Destination port of 0 is illegal, based on RFC768.
245          */
246         if (uh->uh_dport == 0)
247                 goto badunlocked;
248
249         plen = ntohs(ip6->ip6_plen) - off + sizeof(*ip6);
250         ulen = ntohs((u_short)uh->uh_ulen);
251
252         nxt = proto;
253         cscov_partial = (nxt == IPPROTO_UDPLITE) ? 1 : 0;
254         if (nxt == IPPROTO_UDPLITE) {
255                 /* Zero means checksum over the complete packet. */
256                 if (ulen == 0)
257                         ulen = plen;
258                 if (ulen == plen)
259                         cscov_partial = 0;
260                 if ((ulen < sizeof(struct udphdr)) || (ulen > plen)) {
261                         /* XXX: What is the right UDPLite MIB counter? */
262                         goto badunlocked;
263                 }
264                 if (uh->uh_sum == 0) {
265                         /* XXX: What is the right UDPLite MIB counter? */
266                         goto badunlocked;
267                 }
268         } else {
269                 if ((ulen < sizeof(struct udphdr)) || (plen != ulen)) {
270                         UDPSTAT_INC(udps_badlen);
271                         goto badunlocked;
272                 }
273                 if (uh->uh_sum == 0) {
274                         UDPSTAT_INC(udps_nosum);
275                         /*
276                          * dport 0 was rejected earlier so this is OK even if
277                          * zero_checksum_port is 0 (which is its default value).
278                          */
279                         if (ntohs(uh->uh_dport) == V_zero_checksum_port)
280                                 goto skip_checksum;
281                         else
282                                 goto badunlocked;
283                 }
284         }
285
286         if ((m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) &&
287             !cscov_partial) {
288                 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
289                         uh_sum = m->m_pkthdr.csum_data;
290                 else
291                         uh_sum = in6_cksum_pseudo(ip6, ulen, nxt,
292                             m->m_pkthdr.csum_data);
293                 uh_sum ^= 0xffff;
294         } else
295                 uh_sum = in6_cksum_partial(m, nxt, off, plen, ulen);
296
297         if (uh_sum != 0) {
298                 UDPSTAT_INC(udps_badsum);
299                 goto badunlocked;
300         }
301
302 skip_checksum:
303         /*
304          * Construct sockaddr format source address.
305          */
306         init_sin6(&fromsa[0], m, 0);
307         fromsa[0].sin6_port = uh->uh_sport;
308         init_sin6(&fromsa[1], m, 1);
309         fromsa[1].sin6_port = uh->uh_dport;
310
311         pcbinfo = udp_get_inpcbinfo(nxt);
312         if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) {
313                 struct inpcb *last;
314                 struct inpcbhead *pcblist;
315                 struct ip6_moptions *imo;
316
317                 /*
318                  * In the event that laddr should be set to the link-local
319                  * address (this happens in RIPng), the multicast address
320                  * specified in the received packet will not match laddr.  To
321                  * handle this situation, matching is relaxed if the
322                  * receiving interface is the same as one specified in the
323                  * socket and if the destination multicast address matches
324                  * one of the multicast groups specified in the socket.
325                  */
326
327                 /*
328                  * KAME note: traditionally we dropped udpiphdr from mbuf
329                  * here.  We need udphdr for IPsec processing so we do that
330                  * later.
331                  */
332                 pcblist = udp_get_pcblist(nxt);
333                 last = NULL;
334                 CK_LIST_FOREACH(inp, pcblist, inp_list) {
335                         if ((inp->inp_vflag & INP_IPV6) == 0)
336                                 continue;
337                         if (inp->inp_lport != uh->uh_dport)
338                                 continue;
339                         if (inp->inp_fport != 0 &&
340                             inp->inp_fport != uh->uh_sport)
341                                 continue;
342                         if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) {
343                                 if (!IN6_ARE_ADDR_EQUAL(&inp->in6p_laddr,
344                                                         &ip6->ip6_dst))
345                                         continue;
346                         }
347                         if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
348                                 if (!IN6_ARE_ADDR_EQUAL(&inp->in6p_faddr,
349                                                         &ip6->ip6_src) ||
350                                     inp->inp_fport != uh->uh_sport)
351                                         continue;
352                         }
353
354                         /*
355                          * XXXRW: Because we weren't holding either the inpcb
356                          * or the hash lock when we checked for a match 
357                          * before, we should probably recheck now that the 
358                          * inpcb lock is (supposed to be) held.
359                          */
360
361                         /*
362                          * Handle socket delivery policy for any-source
363                          * and source-specific multicast. [RFC3678]
364                          */
365                         imo = inp->in6p_moptions;
366                         if (imo && IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) {
367                                 struct sockaddr_in6      mcaddr;
368                                 int                      blocked;
369
370                                 INP_RLOCK(inp);
371                                 if (__predict_false(inp->inp_flags2 & INP_FREED)) {
372                                         INP_RUNLOCK(inp);
373                                         continue;
374                                 }
375
376                                 bzero(&mcaddr, sizeof(struct sockaddr_in6));
377                                 mcaddr.sin6_len = sizeof(struct sockaddr_in6);
378                                 mcaddr.sin6_family = AF_INET6;
379                                 mcaddr.sin6_addr = ip6->ip6_dst;
380
381                                 blocked = im6o_mc_filter(imo, ifp,
382                                         (struct sockaddr *)&mcaddr,
383                                         (struct sockaddr *)&fromsa[0]);
384                                 if (blocked != MCAST_PASS) {
385                                         if (blocked == MCAST_NOTGMEMBER)
386                                                 IP6STAT_INC(ip6s_notmember);
387                                         if (blocked == MCAST_NOTSMEMBER ||
388                                             blocked == MCAST_MUTED)
389                                                 UDPSTAT_INC(udps_filtermcast);
390                                         INP_RUNLOCK(inp); /* XXX */
391                                         continue;
392                                 }
393
394                                 INP_RUNLOCK(inp);
395                         }
396                         if (last != NULL) {
397                                 struct mbuf *n;
398
399                                 if ((n = m_copym(m, 0, M_COPYALL, M_NOWAIT)) !=
400                                     NULL) {
401                                         INP_RLOCK(last);
402                                         if (__predict_true(last->inp_flags2 & INP_FREED) == 0) {
403                                                 if (nxt == IPPROTO_UDPLITE)
404                                                         UDPLITE_PROBE(receive, NULL, last,
405                                                             ip6, last, uh);
406                                                 else
407                                                         UDP_PROBE(receive, NULL, last,
408                                                             ip6, last, uh);
409                                                 if (udp6_append(last, n, off, fromsa)) {
410                                                         /* XXX-BZ do we leak m here? */
411                                                         *mp = NULL;
412                                                         return (IPPROTO_DONE);
413                                                 }
414                                         }
415                                         INP_RUNLOCK(last);
416                                 }
417                         }
418                         last = inp;
419                         /*
420                          * Don't look for additional matches if this one does
421                          * not have either the SO_REUSEPORT or SO_REUSEADDR
422                          * socket options set.  This heuristic avoids
423                          * searching through all pcbs in the common case of a
424                          * non-shared port.  It assumes that an application
425                          * will never clear these options after setting them.
426                          */
427                         if ((last->inp_socket->so_options &
428                              (SO_REUSEPORT|SO_REUSEPORT_LB|SO_REUSEADDR)) == 0)
429                                 break;
430                 }
431
432                 if (last == NULL) {
433                         /*
434                          * No matching pcb found; discard datagram.  (No need
435                          * to send an ICMP Port Unreachable for a broadcast
436                          * or multicast datgram.)
437                          */
438                         UDPSTAT_INC(udps_noport);
439                         UDPSTAT_INC(udps_noportmcast);
440                         goto badunlocked;
441                 }
442                 INP_RLOCK(last);
443                 if (__predict_true(last->inp_flags2 & INP_FREED) == 0) {
444                         if (nxt == IPPROTO_UDPLITE)
445                                 UDPLITE_PROBE(receive, NULL, last, ip6, last, uh);
446                         else
447                                 UDP_PROBE(receive, NULL, last, ip6, last, uh);
448                         if (udp6_append(last, m, off, fromsa) == 0)
449                                 INP_RUNLOCK(last);
450                 } else
451                         INP_RUNLOCK(last);
452                 *mp = NULL;
453                 return (IPPROTO_DONE);
454         }
455         /*
456          * Locate pcb for datagram.
457          */
458
459         /*
460          * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
461          */
462         if ((m->m_flags & M_IP6_NEXTHOP) &&
463             (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) {
464                 struct sockaddr_in6 *next_hop6;
465
466                 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
467
468                 /*
469                  * Transparently forwarded. Pretend to be the destination.
470                  * Already got one like this?
471                  */
472                 inp = in6_pcblookup_mbuf(pcbinfo, &ip6->ip6_src,
473                     uh->uh_sport, &ip6->ip6_dst, uh->uh_dport,
474                     INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif, m);
475                 if (!inp) {
476                         /*
477                          * It's new.  Try to find the ambushing socket.
478                          * Because we've rewritten the destination address,
479                          * any hardware-generated hash is ignored.
480                          */
481                         inp = in6_pcblookup(pcbinfo, &ip6->ip6_src,
482                             uh->uh_sport, &next_hop6->sin6_addr,
483                             next_hop6->sin6_port ? htons(next_hop6->sin6_port) :
484                             uh->uh_dport, INPLOOKUP_WILDCARD |
485                             INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif);
486                 }
487                 /* Remove the tag from the packet. We don't need it anymore. */
488                 m_tag_delete(m, fwd_tag);
489                 m->m_flags &= ~M_IP6_NEXTHOP;
490         } else
491                 inp = in6_pcblookup_mbuf(pcbinfo, &ip6->ip6_src,
492                     uh->uh_sport, &ip6->ip6_dst, uh->uh_dport,
493                     INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB,
494                     m->m_pkthdr.rcvif, m);
495         if (inp == NULL) {
496                 if (V_udp_log_in_vain) {
497                         char ip6bufs[INET6_ADDRSTRLEN];
498                         char ip6bufd[INET6_ADDRSTRLEN];
499
500                         log(LOG_INFO,
501                             "Connection attempt to UDP [%s]:%d from [%s]:%d\n",
502                             ip6_sprintf(ip6bufd, &ip6->ip6_dst),
503                             ntohs(uh->uh_dport),
504                             ip6_sprintf(ip6bufs, &ip6->ip6_src),
505                             ntohs(uh->uh_sport));
506                 }
507                 if (nxt == IPPROTO_UDPLITE)
508                         UDPLITE_PROBE(receive, NULL, NULL, ip6, NULL, uh);
509                 else
510                         UDP_PROBE(receive, NULL, NULL, ip6, NULL, uh);
511                 UDPSTAT_INC(udps_noport);
512                 if (m->m_flags & M_MCAST) {
513                         printf("UDP6: M_MCAST is set in a unicast packet.\n");
514                         UDPSTAT_INC(udps_noportmcast);
515                         goto badunlocked;
516                 }
517                 if (V_udp_blackhole)
518                         goto badunlocked;
519                 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_NOPORT, 0);
520                 *mp = NULL;
521                 return (IPPROTO_DONE);
522         }
523         INP_RLOCK_ASSERT(inp);
524         up = intoudpcb(inp);
525         if (cscov_partial) {
526                 if (up->u_rxcslen == 0 || up->u_rxcslen > ulen) {
527                         INP_RUNLOCK(inp);
528                         m_freem(m);
529                         *mp = NULL;
530                         return (IPPROTO_DONE);
531                 }
532         }
533         if (nxt == IPPROTO_UDPLITE)
534                 UDPLITE_PROBE(receive, NULL, inp, ip6, inp, uh);
535         else
536                 UDP_PROBE(receive, NULL, inp, ip6, inp, uh);
537         if (udp6_append(inp, m, off, fromsa) == 0)
538                 INP_RUNLOCK(inp);
539         *mp = NULL;
540         return (IPPROTO_DONE);
541
542 badunlocked:
543         m_freem(m);
544         *mp = NULL;
545         return (IPPROTO_DONE);
546 }
547
548 static void
549 udp6_common_ctlinput(int cmd, struct sockaddr *sa, void *d,
550     struct inpcbinfo *pcbinfo)
551 {
552         struct udphdr uh;
553         struct ip6_hdr *ip6;
554         struct mbuf *m;
555         int off = 0;
556         struct ip6ctlparam *ip6cp = NULL;
557         const struct sockaddr_in6 *sa6_src = NULL;
558         void *cmdarg;
559         struct inpcb *(*notify)(struct inpcb *, int) = udp_notify;
560         struct udp_portonly {
561                 u_int16_t uh_sport;
562                 u_int16_t uh_dport;
563         } *uhp;
564
565         if (sa->sa_family != AF_INET6 ||
566             sa->sa_len != sizeof(struct sockaddr_in6))
567                 return;
568
569         if ((unsigned)cmd >= PRC_NCMDS)
570                 return;
571         if (PRC_IS_REDIRECT(cmd))
572                 notify = in6_rtchange, d = NULL;
573         else if (cmd == PRC_HOSTDEAD)
574                 d = NULL;
575         else if (inet6ctlerrmap[cmd] == 0)
576                 return;
577
578         /* if the parameter is from icmp6, decode it. */
579         if (d != NULL) {
580                 ip6cp = (struct ip6ctlparam *)d;
581                 m = ip6cp->ip6c_m;
582                 ip6 = ip6cp->ip6c_ip6;
583                 off = ip6cp->ip6c_off;
584                 cmdarg = ip6cp->ip6c_cmdarg;
585                 sa6_src = ip6cp->ip6c_src;
586         } else {
587                 m = NULL;
588                 ip6 = NULL;
589                 cmdarg = NULL;
590                 sa6_src = &sa6_any;
591         }
592
593         if (ip6) {
594                 /*
595                  * XXX: We assume that when IPV6 is non NULL,
596                  * M and OFF are valid.
597                  */
598
599                 /* Check if we can safely examine src and dst ports. */
600                 if (m->m_pkthdr.len < off + sizeof(*uhp))
601                         return;
602
603                 bzero(&uh, sizeof(uh));
604                 m_copydata(m, off, sizeof(*uhp), (caddr_t)&uh);
605
606                 if (!PRC_IS_REDIRECT(cmd)) {
607                         /* Check to see if its tunneled */
608                         struct inpcb *inp;
609                         inp = in6_pcblookup_mbuf(pcbinfo, &ip6->ip6_dst,
610                             uh.uh_dport, &ip6->ip6_src, uh.uh_sport,
611                             INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB,
612                             m->m_pkthdr.rcvif, m);
613                         if (inp != NULL) {
614                                 struct udpcb *up;
615                                 
616                                 up = intoudpcb(inp);
617                                 if (up->u_icmp_func) {
618                                         /* Yes it is. */
619                                         INP_RUNLOCK(inp);
620                                         (*up->u_icmp_func)(cmd, (struct sockaddr *)ip6cp->ip6c_src,
621                                               d, up->u_tun_ctx);
622                                         return;
623                                 } else {
624                                         /* Can't find it. */
625                                         INP_RUNLOCK(inp);
626                                 }
627                         }
628                 }
629                 (void)in6_pcbnotify(pcbinfo, sa, uh.uh_dport,
630                     (struct sockaddr *)ip6cp->ip6c_src, uh.uh_sport, cmd,
631                     cmdarg, notify);
632         } else
633                 (void)in6_pcbnotify(pcbinfo, sa, 0,
634                     (const struct sockaddr *)sa6_src, 0, cmd, cmdarg, notify);
635 }
636
637 void
638 udp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
639 {
640
641         return (udp6_common_ctlinput(cmd, sa, d, &V_udbinfo));
642 }
643
644 void
645 udplite6_ctlinput(int cmd, struct sockaddr *sa, void *d)
646 {
647
648         return (udp6_common_ctlinput(cmd, sa, d, &V_ulitecbinfo));
649 }
650
651 static int
652 udp6_getcred(SYSCTL_HANDLER_ARGS)
653 {
654         struct xucred xuc;
655         struct sockaddr_in6 addrs[2];
656         struct epoch_tracker et;
657         struct inpcb *inp;
658         int error;
659
660         error = priv_check(req->td, PRIV_NETINET_GETCRED);
661         if (error)
662                 return (error);
663
664         if (req->newlen != sizeof(addrs))
665                 return (EINVAL);
666         if (req->oldlen != sizeof(struct xucred))
667                 return (EINVAL);
668         error = SYSCTL_IN(req, addrs, sizeof(addrs));
669         if (error)
670                 return (error);
671         if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
672             (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
673                 return (error);
674         }
675         NET_EPOCH_ENTER(et);
676         inp = in6_pcblookup(&V_udbinfo, &addrs[1].sin6_addr,
677             addrs[1].sin6_port, &addrs[0].sin6_addr, addrs[0].sin6_port,
678             INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
679         NET_EPOCH_EXIT(et);
680         if (inp != NULL) {
681                 INP_RLOCK_ASSERT(inp);
682                 if (inp->inp_socket == NULL)
683                         error = ENOENT;
684                 if (error == 0)
685                         error = cr_canseesocket(req->td->td_ucred,
686                             inp->inp_socket);
687                 if (error == 0)
688                         cru2x(inp->inp_cred, &xuc);
689                 INP_RUNLOCK(inp);
690         } else
691                 error = ENOENT;
692         if (error == 0)
693                 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
694         return (error);
695 }
696
697 SYSCTL_PROC(_net_inet6_udp6, OID_AUTO, getcred,
698     CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_MPSAFE,
699     0, 0, udp6_getcred, "S,xucred",
700     "Get the xucred of a UDP6 connection");
701
702 static int
703 udp6_output(struct socket *so, int flags_arg, struct mbuf *m,
704     struct sockaddr *addr6, struct mbuf *control, struct thread *td)
705 {
706         struct inpcb *inp;
707         struct ip6_hdr *ip6;
708         struct udphdr *udp6;
709         struct in6_addr *laddr, *faddr, in6a;
710         struct ip6_pktopts *optp, opt;
711         struct sockaddr_in6 *sin6, tmp;
712         struct epoch_tracker et;
713         int cscov_partial, error, flags, hlen, scope_ambiguous;
714         u_int32_t ulen, plen;
715         uint16_t cscov;
716         u_short fport;
717         uint8_t nxt;
718
719         /* addr6 has been validated in udp6_send(). */
720         sin6 = (struct sockaddr_in6 *)addr6;
721
722         /*
723          * In contrast to to IPv4 we do not validate the max. packet length
724          * here due to IPv6 Jumbograms (RFC2675).
725          */
726
727         scope_ambiguous = 0;
728         if (sin6) {
729                 /* Protect *addr6 from overwrites. */
730                 tmp = *sin6;
731                 sin6 = &tmp;
732
733                 /*
734                  * Application should provide a proper zone ID or the use of
735                  * default zone IDs should be enabled.  Unfortunately, some
736                  * applications do not behave as it should, so we need a
737                  * workaround.  Even if an appropriate ID is not determined,
738                  * we'll see if we can determine the outgoing interface.  If we
739                  * can, determine the zone ID based on the interface below.
740                  */
741                 if (sin6->sin6_scope_id == 0 && !V_ip6_use_defzone)
742                         scope_ambiguous = 1;
743                 if ((error = sa6_embedscope(sin6, V_ip6_use_defzone)) != 0) {
744                         if (control)
745                                 m_freem(control);
746                         m_freem(m);
747                         return (error);
748                 }
749         }
750
751         inp = sotoinpcb(so);
752         KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
753         /*
754          * In the following cases we want a write lock on the inp for either
755          * local operations or for possible route cache updates in the IPv6
756          * output path:
757          * - on connected sockets (sin6 is NULL) for route cache updates,
758          * - when we are not bound to an address and source port (it is
759          *   in6_pcbsetport() which will require the write lock).
760          *
761          * We check the inp fields before actually locking the inp, so
762          * here exists a race, and we may WLOCK the inp and end with already
763          * bound one by other thread. This is fine.
764          */
765         if (sin6 == NULL || (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) &&
766             inp->inp_lport == 0))
767                 INP_WLOCK(inp);
768         else
769                 INP_RLOCK(inp);
770
771         nxt = (inp->inp_socket->so_proto->pr_protocol == IPPROTO_UDP) ?
772             IPPROTO_UDP : IPPROTO_UDPLITE;
773
774 #ifdef INET
775         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
776                 int hasv4addr;
777
778                 if (sin6 == NULL)
779                         hasv4addr = (inp->inp_vflag & INP_IPV4);
780                 else
781                         hasv4addr = IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)
782                             ? 1 : 0;
783                 if (hasv4addr) {
784                         struct pr_usrreqs *pru;
785
786                         /*
787                          * XXXRW: We release UDP-layer locks before calling
788                          * udp_send() in order to avoid recursion.  However,
789                          * this does mean there is a short window where inp's
790                          * fields are unstable.  Could this lead to a
791                          * potential race in which the factors causing us to
792                          * select the UDPv4 output routine are invalidated?
793                          */
794                         INP_UNLOCK(inp);
795                         if (sin6)
796                                 in6_sin6_2_sin_in_sock((struct sockaddr *)sin6);
797                         pru = inetsw[ip_protox[nxt]].pr_usrreqs;
798                         /* addr will just be freed in sendit(). */
799                         return ((*pru->pru_send)(so, flags_arg | PRUS_IPV6, m,
800                             (struct sockaddr *)sin6, control, td));
801                 }
802         } else
803 #endif
804         if (sin6 && IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
805                 /*
806                  * Given this is either an IPv6-only socket or no INET is
807                  * supported we will fail the send if the given destination
808                  * address is a v4mapped address.
809                  *
810                  * XXXGL: do we leak m and control?
811                  */
812                 INP_UNLOCK(inp);
813                 return (EINVAL);
814         }
815
816         if (control) {
817                 if ((error = ip6_setpktopts(control, &opt,
818                     inp->in6p_outputopts, td->td_ucred, nxt)) != 0) {
819                         INP_UNLOCK(inp);
820                         ip6_clearpktopts(&opt, -1);
821                         if (control)
822                                 m_freem(control);
823                         m_freem(m);
824                         return (error);
825                 }
826                 optp = &opt;
827         } else
828                 optp = inp->in6p_outputopts;
829
830         NET_EPOCH_ENTER(et);
831         if (sin6) {
832                 /*
833                  * Since we saw no essential reason for calling in_pcbconnect,
834                  * we get rid of such kind of logic, and call in6_selectsrc
835                  * and in6_pcbsetport in order to fill in the local address
836                  * and the local port.
837                  */
838                 if (sin6->sin6_port == 0) {
839                         error = EADDRNOTAVAIL;
840                         goto release;
841                 }
842
843                 if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
844                         /* how about ::ffff:0.0.0.0 case? */
845                         error = EISCONN;
846                         goto release;
847                 }
848
849                 /*
850                  * Given we handle the v4mapped case in the INET block above
851                  * assert here that it must not happen anymore.
852                  */
853                 KASSERT(!IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr),
854                     ("%s: sin6(%p)->sin6_addr is v4mapped which we "
855                     "should have handled.", __func__, sin6));
856
857                 /* This only requires read-locking. */
858                 error = in6_selectsrc_socket(sin6, optp, inp,
859                     td->td_ucred, scope_ambiguous, &in6a, NULL);
860                 if (error)
861                         goto release;
862                 laddr = &in6a;
863
864                 if (inp->inp_lport == 0) {
865                         struct inpcbinfo *pcbinfo;
866
867                         INP_WLOCK_ASSERT(inp);
868
869                         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
870                         INP_HASH_WLOCK(pcbinfo);
871                         error = in6_pcbsetport(laddr, inp, td->td_ucred);
872                         INP_HASH_WUNLOCK(pcbinfo);
873                         if (error != 0) {
874                                 /* Undo an address bind that may have occurred. */
875                                 inp->in6p_laddr = in6addr_any;
876                                 goto release;
877                         }
878                 }
879                 faddr = &sin6->sin6_addr;
880                 fport = sin6->sin6_port; /* allow 0 port */
881
882         } else {
883                 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
884                         error = ENOTCONN;
885                         goto release;
886                 }
887                 laddr = &inp->in6p_laddr;
888                 faddr = &inp->in6p_faddr;
889                 fport = inp->inp_fport;
890         }
891
892         ulen = m->m_pkthdr.len;
893         plen = sizeof(struct udphdr) + ulen;
894         hlen = sizeof(struct ip6_hdr);
895
896         /*
897          * Calculate data length and get a mbuf
898          * for UDP and IP6 headers.
899          */
900         M_PREPEND(m, hlen + sizeof(struct udphdr), M_NOWAIT);
901         if (m == NULL) {
902                 error = ENOBUFS;
903                 goto release;
904         }
905
906         /*
907          * Stuff checksum and output datagram.
908          */
909         cscov = cscov_partial = 0;
910         udp6 = (struct udphdr *)(mtod(m, caddr_t) + hlen);
911         udp6->uh_sport = inp->inp_lport; /* lport is always set in the PCB */
912         udp6->uh_dport = fport;
913         if (nxt == IPPROTO_UDPLITE) {
914                 struct udpcb *up;
915
916                 up = intoudpcb(inp);
917                 cscov = up->u_txcslen;
918                 if (cscov >= plen)
919                         cscov = 0;
920                 udp6->uh_ulen = htons(cscov);
921                 /*
922                  * For UDP-Lite, checksum coverage length of zero means
923                  * the entire UDPLite packet is covered by the checksum.
924                  */
925                 cscov_partial = (cscov == 0) ? 0 : 1;
926         } else if (plen <= 0xffff)
927                 udp6->uh_ulen = htons((u_short)plen);
928         else
929                 udp6->uh_ulen = 0;
930         udp6->uh_sum = 0;
931
932         ip6 = mtod(m, struct ip6_hdr *);
933         ip6->ip6_flow   = inp->inp_flow & IPV6_FLOWINFO_MASK;
934         ip6->ip6_vfc    &= ~IPV6_VERSION_MASK;
935         ip6->ip6_vfc    |= IPV6_VERSION;
936         ip6->ip6_plen   = htons((u_short)plen);
937         ip6->ip6_nxt    = nxt;
938         ip6->ip6_hlim   = in6_selecthlim(inp, NULL);
939         ip6->ip6_src    = *laddr;
940         ip6->ip6_dst    = *faddr;
941
942 #ifdef MAC
943         mac_inpcb_create_mbuf(inp, m);
944 #endif
945
946         if (cscov_partial) {
947                 if ((udp6->uh_sum = in6_cksum_partial(m, nxt,
948                     sizeof(struct ip6_hdr), plen, cscov)) == 0)
949                         udp6->uh_sum = 0xffff;
950         } else {
951                 udp6->uh_sum = in6_cksum_pseudo(ip6, plen, nxt, 0);
952                 m->m_pkthdr.csum_flags = CSUM_UDP_IPV6;
953                 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
954         }
955
956         flags = 0;
957 #ifdef  RSS
958         {
959                 uint32_t hash_val, hash_type;
960                 uint8_t pr;
961
962                 pr = inp->inp_socket->so_proto->pr_protocol;
963                 /*
964                  * Calculate an appropriate RSS hash for UDP and
965                  * UDP Lite.
966                  *
967                  * The called function will take care of figuring out
968                  * whether a 2-tuple or 4-tuple hash is required based
969                  * on the currently configured scheme.
970                  *
971                  * Later later on connected socket values should be
972                  * cached in the inpcb and reused, rather than constantly
973                  * re-calculating it.
974                  *
975                  * UDP Lite is a different protocol number and will
976                  * likely end up being hashed as a 2-tuple until
977                  * RSS / NICs grow UDP Lite protocol awareness.
978                  */
979                 if (rss_proto_software_hash_v6(faddr, laddr, fport,
980                     inp->inp_lport, pr, &hash_val, &hash_type) == 0) {
981                         m->m_pkthdr.flowid = hash_val;
982                         M_HASHTYPE_SET(m, hash_type);
983                 }
984
985                 /*
986                  * Don't override with the inp cached flowid.
987                  *
988                  * Until the whole UDP path is vetted, it may actually
989                  * be incorrect.
990                  */
991                 flags |= IP_NODEFAULTFLOWID;
992         }
993 #endif
994
995         UDPSTAT_INC(udps_opackets);
996         if (nxt == IPPROTO_UDPLITE)
997                 UDPLITE_PROBE(send, NULL, inp, ip6, inp, udp6);
998         else
999                 UDP_PROBE(send, NULL, inp, ip6, inp, udp6);
1000         error = ip6_output(m, optp,
1001             INP_WLOCKED(inp) ? &inp->inp_route6 : NULL, flags,
1002             inp->in6p_moptions, NULL, inp);
1003         INP_UNLOCK(inp);
1004         NET_EPOCH_EXIT(et);
1005
1006         if (control) {
1007                 ip6_clearpktopts(&opt, -1);
1008                 m_freem(control);
1009         }
1010         return (error);
1011
1012 release:
1013         INP_UNLOCK(inp);
1014         NET_EPOCH_EXIT(et);
1015         if (control) {
1016                 ip6_clearpktopts(&opt, -1);
1017                 m_freem(control);
1018         }
1019         m_freem(m);
1020
1021         return (error);
1022 }
1023
1024 static void
1025 udp6_abort(struct socket *so)
1026 {
1027         struct inpcb *inp;
1028         struct inpcbinfo *pcbinfo;
1029
1030         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1031         inp = sotoinpcb(so);
1032         KASSERT(inp != NULL, ("udp6_abort: inp == NULL"));
1033
1034         INP_WLOCK(inp);
1035 #ifdef INET
1036         if (inp->inp_vflag & INP_IPV4) {
1037                 struct pr_usrreqs *pru;
1038                 uint8_t nxt;
1039
1040                 nxt = (inp->inp_socket->so_proto->pr_protocol == IPPROTO_UDP) ?
1041                     IPPROTO_UDP : IPPROTO_UDPLITE;
1042                 INP_WUNLOCK(inp);
1043                 pru = inetsw[ip_protox[nxt]].pr_usrreqs;
1044                 (*pru->pru_abort)(so);
1045                 return;
1046         }
1047 #endif
1048
1049         if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
1050                 INP_HASH_WLOCK(pcbinfo);
1051                 in6_pcbdisconnect(inp);
1052                 inp->in6p_laddr = in6addr_any;
1053                 INP_HASH_WUNLOCK(pcbinfo);
1054                 soisdisconnected(so);
1055         }
1056         INP_WUNLOCK(inp);
1057 }
1058
1059 static int
1060 udp6_attach(struct socket *so, int proto, struct thread *td)
1061 {
1062         struct inpcb *inp;
1063         struct inpcbinfo *pcbinfo;
1064         int error;
1065
1066         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1067         inp = sotoinpcb(so);
1068         KASSERT(inp == NULL, ("udp6_attach: inp != NULL"));
1069
1070         if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
1071                 error = soreserve(so, udp_sendspace, udp_recvspace);
1072                 if (error)
1073                         return (error);
1074         }
1075         INP_INFO_WLOCK(pcbinfo);
1076         error = in_pcballoc(so, pcbinfo);
1077         if (error) {
1078                 INP_INFO_WUNLOCK(pcbinfo);
1079                 return (error);
1080         }
1081         inp = (struct inpcb *)so->so_pcb;
1082         inp->inp_vflag |= INP_IPV6;
1083         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
1084                 inp->inp_vflag |= INP_IPV4;
1085         inp->in6p_hops = -1;    /* use kernel default */
1086         inp->in6p_cksum = -1;   /* just to be sure */
1087         /*
1088          * XXX: ugly!!
1089          * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1090          * because the socket may be bound to an IPv6 wildcard address,
1091          * which may match an IPv4-mapped IPv6 address.
1092          */
1093         inp->inp_ip_ttl = V_ip_defttl;
1094
1095         error = udp_newudpcb(inp);
1096         if (error) {
1097                 in_pcbdetach(inp);
1098                 in_pcbfree(inp);
1099                 INP_INFO_WUNLOCK(pcbinfo);
1100                 return (error);
1101         }
1102         INP_WUNLOCK(inp);
1103         INP_INFO_WUNLOCK(pcbinfo);
1104         return (0);
1105 }
1106
1107 static int
1108 udp6_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1109 {
1110         struct inpcb *inp;
1111         struct inpcbinfo *pcbinfo;
1112         int error;
1113         u_char vflagsav;
1114
1115         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1116         inp = sotoinpcb(so);
1117         KASSERT(inp != NULL, ("udp6_bind: inp == NULL"));
1118
1119         INP_WLOCK(inp);
1120         INP_HASH_WLOCK(pcbinfo);
1121         vflagsav = inp->inp_vflag;
1122         inp->inp_vflag &= ~INP_IPV4;
1123         inp->inp_vflag |= INP_IPV6;
1124         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
1125                 struct sockaddr_in6 *sin6_p;
1126
1127                 sin6_p = (struct sockaddr_in6 *)nam;
1128
1129                 if (IN6_IS_ADDR_UNSPECIFIED(&sin6_p->sin6_addr))
1130                         inp->inp_vflag |= INP_IPV4;
1131 #ifdef INET
1132                 else if (IN6_IS_ADDR_V4MAPPED(&sin6_p->sin6_addr)) {
1133                         struct sockaddr_in sin;
1134
1135                         in6_sin6_2_sin(&sin, sin6_p);
1136                         inp->inp_vflag |= INP_IPV4;
1137                         inp->inp_vflag &= ~INP_IPV6;
1138                         error = in_pcbbind(inp, (struct sockaddr *)&sin,
1139                             td->td_ucred);
1140                         goto out;
1141                 }
1142 #endif
1143         }
1144
1145         error = in6_pcbbind(inp, nam, td->td_ucred);
1146 #ifdef INET
1147 out:
1148 #endif
1149         if (error != 0)
1150                 inp->inp_vflag = vflagsav;
1151         INP_HASH_WUNLOCK(pcbinfo);
1152         INP_WUNLOCK(inp);
1153         return (error);
1154 }
1155
1156 static void
1157 udp6_close(struct socket *so)
1158 {
1159         struct inpcb *inp;
1160         struct inpcbinfo *pcbinfo;
1161
1162         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1163         inp = sotoinpcb(so);
1164         KASSERT(inp != NULL, ("udp6_close: inp == NULL"));
1165
1166         INP_WLOCK(inp);
1167 #ifdef INET
1168         if (inp->inp_vflag & INP_IPV4) {
1169                 struct pr_usrreqs *pru;
1170                 uint8_t nxt;
1171
1172                 nxt = (inp->inp_socket->so_proto->pr_protocol == IPPROTO_UDP) ?
1173                     IPPROTO_UDP : IPPROTO_UDPLITE;
1174                 INP_WUNLOCK(inp);
1175                 pru = inetsw[ip_protox[nxt]].pr_usrreqs;
1176                 (*pru->pru_disconnect)(so);
1177                 return;
1178         }
1179 #endif
1180         if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
1181                 INP_HASH_WLOCK(pcbinfo);
1182                 in6_pcbdisconnect(inp);
1183                 inp->in6p_laddr = in6addr_any;
1184                 INP_HASH_WUNLOCK(pcbinfo);
1185                 soisdisconnected(so);
1186         }
1187         INP_WUNLOCK(inp);
1188 }
1189
1190 static int
1191 udp6_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1192 {
1193 #ifdef INET
1194         struct epoch_tracker et;
1195 #endif
1196         struct inpcb *inp;
1197         struct inpcbinfo *pcbinfo;
1198         struct sockaddr_in6 *sin6;
1199         int error;
1200         u_char vflagsav;
1201
1202         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1203         inp = sotoinpcb(so);
1204         sin6 = (struct sockaddr_in6 *)nam;
1205         KASSERT(inp != NULL, ("udp6_connect: inp == NULL"));
1206
1207         /*
1208          * XXXRW: Need to clarify locking of v4/v6 flags.
1209          */
1210         INP_WLOCK(inp);
1211 #ifdef INET
1212         if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
1213                 struct sockaddr_in sin;
1214
1215                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
1216                         error = EINVAL;
1217                         goto out;
1218                 }
1219                 if ((inp->inp_vflag & INP_IPV4) == 0) {
1220                         error = EAFNOSUPPORT;
1221                         goto out;
1222                 }
1223                 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1224                         error = EISCONN;
1225                         goto out;
1226                 }
1227                 in6_sin6_2_sin(&sin, sin6);
1228                 error = prison_remote_ip4(td->td_ucred, &sin.sin_addr);
1229                 if (error != 0)
1230                         goto out;
1231                 vflagsav = inp->inp_vflag;
1232                 inp->inp_vflag |= INP_IPV4;
1233                 inp->inp_vflag &= ~INP_IPV6;
1234                 NET_EPOCH_ENTER(et);
1235                 INP_HASH_WLOCK(pcbinfo);
1236                 error = in_pcbconnect(inp, (struct sockaddr *)&sin,
1237                     td->td_ucred);
1238                 INP_HASH_WUNLOCK(pcbinfo);
1239                 NET_EPOCH_EXIT(et);
1240                 /*
1241                  * If connect succeeds, mark socket as connected. If
1242                  * connect fails and socket is unbound, reset inp_vflag
1243                  * field.
1244                  */
1245                 if (error == 0)
1246                         soisconnected(so);
1247                 else if (inp->inp_laddr.s_addr == INADDR_ANY &&
1248                     inp->inp_lport == 0)
1249                         inp->inp_vflag = vflagsav;
1250                 goto out;
1251         } else {
1252                 if ((inp->inp_vflag & INP_IPV6) == 0) {
1253                         error = EAFNOSUPPORT;
1254                         goto out;
1255                 }
1256         }
1257 #endif
1258         if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
1259                 error = EISCONN;
1260                 goto out;
1261         }
1262         error = prison_remote_ip6(td->td_ucred, &sin6->sin6_addr);
1263         if (error != 0)
1264                 goto out;
1265         vflagsav = inp->inp_vflag;
1266         inp->inp_vflag &= ~INP_IPV4;
1267         inp->inp_vflag |= INP_IPV6;
1268         INP_HASH_WLOCK(pcbinfo);
1269         error = in6_pcbconnect(inp, nam, td->td_ucred);
1270         INP_HASH_WUNLOCK(pcbinfo);
1271         /*
1272          * If connect succeeds, mark socket as connected. If
1273          * connect fails and socket is unbound, reset inp_vflag
1274          * field.
1275          */
1276         if (error == 0)
1277                 soisconnected(so);
1278         else if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) &&
1279             inp->inp_lport == 0)
1280                 inp->inp_vflag = vflagsav;
1281 out:
1282         INP_WUNLOCK(inp);
1283         return (error);
1284 }
1285
1286 static void
1287 udp6_detach(struct socket *so)
1288 {
1289         struct inpcb *inp;
1290         struct inpcbinfo *pcbinfo;
1291         struct udpcb *up;
1292
1293         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1294         inp = sotoinpcb(so);
1295         KASSERT(inp != NULL, ("udp6_detach: inp == NULL"));
1296
1297         INP_INFO_WLOCK(pcbinfo);
1298         INP_WLOCK(inp);
1299         up = intoudpcb(inp);
1300         KASSERT(up != NULL, ("%s: up == NULL", __func__));
1301         in_pcbdetach(inp);
1302         in_pcbfree(inp);
1303         INP_INFO_WUNLOCK(pcbinfo);
1304         udp_discardcb(up);
1305 }
1306
1307 static int
1308 udp6_disconnect(struct socket *so)
1309 {
1310         struct inpcb *inp;
1311         struct inpcbinfo *pcbinfo;
1312
1313         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1314         inp = sotoinpcb(so);
1315         KASSERT(inp != NULL, ("udp6_disconnect: inp == NULL"));
1316
1317         INP_WLOCK(inp);
1318 #ifdef INET
1319         if (inp->inp_vflag & INP_IPV4) {
1320                 struct pr_usrreqs *pru;
1321                 uint8_t nxt;
1322
1323                 nxt = (inp->inp_socket->so_proto->pr_protocol == IPPROTO_UDP) ?
1324                     IPPROTO_UDP : IPPROTO_UDPLITE;
1325                 INP_WUNLOCK(inp);
1326                 pru = inetsw[ip_protox[nxt]].pr_usrreqs;
1327                 (void)(*pru->pru_disconnect)(so);
1328                 return (0);
1329         }
1330 #endif
1331
1332         if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
1333                 INP_WUNLOCK(inp);
1334                 return (ENOTCONN);
1335         }
1336
1337         INP_HASH_WLOCK(pcbinfo);
1338         in6_pcbdisconnect(inp);
1339         inp->in6p_laddr = in6addr_any;
1340         INP_HASH_WUNLOCK(pcbinfo);
1341         SOCK_LOCK(so);
1342         so->so_state &= ~SS_ISCONNECTED;                /* XXX */
1343         SOCK_UNLOCK(so);
1344         INP_WUNLOCK(inp);
1345         return (0);
1346 }
1347
1348 static int
1349 udp6_send(struct socket *so, int flags, struct mbuf *m,
1350     struct sockaddr *addr, struct mbuf *control, struct thread *td)
1351 {
1352         int error;
1353
1354         if (addr) {
1355                 if (addr->sa_len != sizeof(struct sockaddr_in6)) {
1356                         error = EINVAL;
1357                         goto bad;
1358                 }
1359                 if (addr->sa_family != AF_INET6) {
1360                         error = EAFNOSUPPORT;
1361                         goto bad;
1362                 }
1363         }
1364
1365         return (udp6_output(so, flags, m, addr, control, td));
1366
1367 bad:
1368         if (control)
1369                 m_freem(control);
1370         m_freem(m);
1371         return (error);
1372 }
1373
1374 struct pr_usrreqs udp6_usrreqs = {
1375         .pru_abort =            udp6_abort,
1376         .pru_attach =           udp6_attach,
1377         .pru_bind =             udp6_bind,
1378         .pru_connect =          udp6_connect,
1379         .pru_control =          in6_control,
1380         .pru_detach =           udp6_detach,
1381         .pru_disconnect =       udp6_disconnect,
1382         .pru_peeraddr =         in6_mapped_peeraddr,
1383         .pru_send =             udp6_send,
1384         .pru_shutdown =         udp_shutdown,
1385         .pru_sockaddr =         in6_mapped_sockaddr,
1386         .pru_soreceive =        soreceive_dgram,
1387         .pru_sosend =           sosend_dgram,
1388         .pru_sosetlabel =       in_pcbsosetlabel,
1389         .pru_close =            udp6_close
1390 };