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_route.h"
  79 #include "opt_rss.h"
  80
  81 #include <sys/param.h>
  82 #include <sys/jail.h>
  83 #include <sys/kernel.h>
  84 #include <sys/lock.h>
  85 #include <sys/mbuf.h>
  86 #include <sys/priv.h>
  87 #include <sys/proc.h>
  88 #include <sys/protosw.h>
  89 #include <sys/sdt.h>
  90 #include <sys/signalvar.h>
  91 #include <sys/socket.h>
  92 #include <sys/socketvar.h>
  93 #include <sys/sx.h>
  94 #include <sys/sysctl.h>
  95 #include <sys/syslog.h>
  96 #include <sys/systm.h>
  97
  98 #include <net/if.h>
  99 #include <net/if_var.h>
 100 #include <net/if_types.h>
 101 #include <net/route.h>
 102 #include <net/rss_config.h>
 103
 104 #include <netinet/in.h>
 105 #include <netinet/in_kdtrace.h>
 106 #include <netinet/in_pcb.h>
 107 #include <netinet/in_systm.h>
 108 #include <netinet/in_var.h>
 109 #include <netinet/ip.h>
 110 #include <netinet/ip6.h>
 111 #include <netinet/icmp6.h>
 112 #include <netinet/ip_var.h>
 113 #include <netinet/udp.h>
 114 #include <netinet/udp_var.h>
 115 #include <netinet/udplite.h>
 116
 117 #include <netinet6/ip6protosw.h>
 118 #include <netinet6/ip6_var.h>
 119 #include <netinet6/in6_fib.h>
 120 #include <netinet6/in6_pcb.h>
 121 #include <netinet6/in6_rss.h>
 122 #include <netinet6/udp6_var.h>
 123 #include <netinet6/scope6_var.h>
 124
 125 #include <netipsec/ipsec_support.h>
 126
 127 #include <security/mac/mac_framework.h>
 128
 129 VNET_DEFINE(int, zero_checksum_port) = 0;
 130 #define V_zero_checksum_port    VNET(zero_checksum_port)
 131 SYSCTL_INT(_net_inet6_udp6, OID_AUTO, rfc6935_port, CTLFLAG_VNET | CTLFLAG_RW,
 132     &VNET_NAME(zero_checksum_port), 0,
 133     "Zero UDP checksum allowed for traffic to/from this port.");
 134 /*
 135  * UDP protocol implementation.
 136  * Per RFC 768, August, 1980.
 137  */
 138
 139 extern struct protosw   inetsw[];
 140 static void             udp6_detach(struct socket *so);
 141
 142 static int
 143 udp6_append(struct inpcb *inp, struct mbuf *n, int off,
 144     struct sockaddr_in6 *fromsa)
 145 {
 146         struct socket *so;
 147         struct mbuf *opts = NULL, *tmp_opts;
 148         struct udpcb *up;
 149
 150         INP_LOCK_ASSERT(inp);
 151
 152         /*
 153          * Engage the tunneling protocol.
 154          */
 155         up = intoudpcb(inp);
 156         if (up->u_tun_func != NULL) {
 157                 in_pcbref(inp);
 158                 INP_RUNLOCK(inp);
 159                 (*up->u_tun_func)(n, off, inp, (struct sockaddr *)&fromsa[0],
 160                     up->u_tun_ctx);
 161                 INP_RLOCK(inp);
 162                 return (in_pcbrele_rlocked(inp));
 163         }
 164 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
 165         /* Check AH/ESP integrity. */
 166         if (IPSEC_ENABLED(ipv6)) {
 167                 if (IPSEC_CHECK_POLICY(ipv6, n, inp) != 0) {
 168                         m_freem(n);
 169                         return (0);
 170                 }
 171         }
 172 #endif /* IPSEC */
 173 #ifdef MAC
 174         if (mac_inpcb_check_deliver(inp, n) != 0) {
 175                 m_freem(n);
 176                 return (0);
 177         }
 178 #endif
 179         opts = NULL;
 180         if (inp->inp_flags & INP_CONTROLOPTS ||
 181             inp->inp_socket->so_options & SO_TIMESTAMP)
 182                 ip6_savecontrol(inp, n, &opts);
 183         if ((inp->inp_vflag & INP_IPV6) && (inp->inp_flags2 & INP_ORIGDSTADDR)) {
 184                 tmp_opts = sbcreatecontrol((caddr_t)&fromsa[1],
 185                         sizeof(struct sockaddr_in6), IPV6_ORIGDSTADDR, IPPROTO_IPV6);
 186                 if (tmp_opts) {
 187                         if (opts) {
 188                                 tmp_opts->m_next = opts;
 189                                 opts = tmp_opts;
 190                         } else
 191                                 opts = tmp_opts;
 192                 }
 193         }
 194         m_adj(n, off + sizeof(struct udphdr));
 195
 196         so = inp->inp_socket;
 197         SOCKBUF_LOCK(&so->so_rcv);
 198         if (sbappendaddr_locked(&so->so_rcv, (struct sockaddr *)&fromsa[0], n,
 199             opts) == 0) {
 200                 SOCKBUF_UNLOCK(&so->so_rcv);
 201                 m_freem(n);
 202                 if (opts)
 203                         m_freem(opts);
 204                 UDPSTAT_INC(udps_fullsock);
 205         } else
 206                 sorwakeup_locked(so);
 207         return (0);
 208 }
 209
 210 int
 211 udp6_input(struct mbuf **mp, int *offp, int proto)
 212 {
 213         struct mbuf *m = *mp;
 214         struct ifnet *ifp;
 215         struct ip6_hdr *ip6;
 216         struct udphdr *uh;
 217         struct inpcb *inp;
 218         struct inpcbinfo *pcbinfo;
 219         struct udpcb *up;
 220         int off = *offp;
 221         int cscov_partial;
 222         int plen, ulen;
 223         struct sockaddr_in6 fromsa[2];
 224         struct m_tag *fwd_tag;
 225         uint16_t uh_sum;
 226         uint8_t nxt;
 227
 228         NET_EPOCH_ASSERT();
 229
 230         ifp = m->m_pkthdr.rcvif;
 231
 232         if (m->m_len < off + sizeof(struct udphdr)) {
 233                 m = m_pullup(m, off + sizeof(struct udphdr));
 234                 if (m == NULL) {
 235                         IP6STAT_INC(ip6s_exthdrtoolong);
 236                         *mp = NULL;
 237                         return (IPPROTO_DONE);
 238                 }
 239         }
 240         ip6 = mtod(m, struct ip6_hdr *);
 241         uh = (struct udphdr *)((caddr_t)ip6 + off);
 242
 243         UDPSTAT_INC(udps_ipackets);
 244
 245         /*
 246          * Destination port of 0 is illegal, based on RFC768.
 247          */
 248         if (uh->uh_dport == 0)
 249                 goto badunlocked;
 250
 251         plen = ntohs(ip6->ip6_plen) - off + sizeof(*ip6);
 252         ulen = ntohs((u_short)uh->uh_ulen);
 253
 254         nxt = proto;
 255         cscov_partial = (nxt == IPPROTO_UDPLITE) ? 1 : 0;
 256         if (nxt == IPPROTO_UDPLITE) {
 257                 /* Zero means checksum over the complete packet. */
 258                 if (ulen == 0)
 259                         ulen = plen;
 260                 if (ulen == plen)
 261                         cscov_partial = 0;
 262                 if ((ulen < sizeof(struct udphdr)) || (ulen > plen)) {
 263                         /* XXX: What is the right UDPLite MIB counter? */
 264                         goto badunlocked;
 265                 }
 266                 if (uh->uh_sum == 0) {
 267                         /* XXX: What is the right UDPLite MIB counter? */
 268                         goto badunlocked;
 269                 }
 270         } else {
 271                 if ((ulen < sizeof(struct udphdr)) || (plen != ulen)) {
 272                         UDPSTAT_INC(udps_badlen);
 273                         goto badunlocked;
 274                 }
 275                 if (uh->uh_sum == 0) {
 276                         UDPSTAT_INC(udps_nosum);
 277                         /*
 278                          * dport 0 was rejected earlier so this is OK even if
 279                          * zero_checksum_port is 0 (which is its default value).
 280                          */
 281                         if (ntohs(uh->uh_dport) == V_zero_checksum_port)
 282                                 goto skip_checksum;
 283                         else
 284                                 goto badunlocked;
 285                 }
 286         }
 287
 288         if ((m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) &&
 289             !cscov_partial) {
 290                 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
 291                         uh_sum = m->m_pkthdr.csum_data;
 292                 else
 293                         uh_sum = in6_cksum_pseudo(ip6, ulen, nxt,
 294                             m->m_pkthdr.csum_data);
 295                 uh_sum ^= 0xffff;
 296         } else
 297                 uh_sum = in6_cksum_partial(m, nxt, off, plen, ulen);
 298
 299         if (uh_sum != 0) {
 300                 UDPSTAT_INC(udps_badsum);
 301                 goto badunlocked;
 302         }
 303
 304 skip_checksum:
 305         /*
 306          * Construct sockaddr format source address.
 307          */
 308         init_sin6(&fromsa[0], m, 0);
 309         fromsa[0].sin6_port = uh->uh_sport;
 310         init_sin6(&fromsa[1], m, 1);
 311         fromsa[1].sin6_port = uh->uh_dport;
 312
 313         pcbinfo = udp_get_inpcbinfo(nxt);
 314         if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) {
 315                 struct inpcb *last;
 316                 struct inpcbhead *pcblist;
 317                 struct ip6_moptions *imo;
 318
 319                 /*
 320                  * In the event that laddr should be set to the link-local
 321                  * address (this happens in RIPng), the multicast address
 322                  * specified in the received packet will not match laddr.  To
 323                  * handle this situation, matching is relaxed if the
 324                  * receiving interface is the same as one specified in the
 325                  * socket and if the destination multicast address matches
 326                  * one of the multicast groups specified in the socket.
 327                  */
 328
 329                 /*
 330                  * KAME note: traditionally we dropped udpiphdr from mbuf
 331                  * here.  We need udphdr for IPsec processing so we do that
 332                  * later.
 333                  */
 334                 pcblist = udp_get_pcblist(nxt);
 335                 last = NULL;
 336                 CK_LIST_FOREACH(inp, pcblist, inp_list) {
 337                         if ((inp->inp_vflag & INP_IPV6) == 0)
 338                                 continue;
 339                         if (inp->inp_lport != uh->uh_dport)
 340                                 continue;
 341                         if (inp->inp_fport != 0 &&
 342                             inp->inp_fport != uh->uh_sport)
 343                                 continue;
 344                         if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) {
 345                                 if (!IN6_ARE_ADDR_EQUAL(&inp->in6p_laddr,
 346                                                         &ip6->ip6_dst))
 347                                         continue;
 348                         }
 349                         if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
 350                                 if (!IN6_ARE_ADDR_EQUAL(&inp->in6p_faddr,
 351                                                         &ip6->ip6_src) ||
 352                                     inp->inp_fport != uh->uh_sport)
 353                                         continue;
 354                         }
 355
 356                         /*
 357                          * XXXRW: Because we weren't holding either the inpcb
 358                          * or the hash lock when we checked for a match
 359                          * before, we should probably recheck now that the
 360                          * inpcb lock is (supposed to be) held.
 361                          */
 362
 363                         /*
 364                          * Handle socket delivery policy for any-source
 365                          * and source-specific multicast. [RFC3678]
 366                          */
 367                         imo = inp->in6p_moptions;
 368                         if (imo && IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) {
 369                                 struct sockaddr_in6      mcaddr;
 370                                 int                      blocked;
 371
 372                                 INP_RLOCK(inp);
 373                                 if (__predict_false(inp->inp_flags2 & INP_FREED)) {
 374                                         INP_RUNLOCK(inp);
 375                                         continue;
 376                                 }
 377
 378                                 bzero(&mcaddr, sizeof(struct sockaddr_in6));
 379                                 mcaddr.sin6_len = sizeof(struct sockaddr_in6);
 380                                 mcaddr.sin6_family = AF_INET6;
 381                                 mcaddr.sin6_addr = ip6->ip6_dst;
 382
 383                                 blocked = im6o_mc_filter(imo, ifp,
 384                                         (struct sockaddr *)&mcaddr,
 385                                         (struct sockaddr *)&fromsa[0]);
 386                                 if (blocked != MCAST_PASS) {
 387                                         if (blocked == MCAST_NOTGMEMBER)
 388                                                 IP6STAT_INC(ip6s_notmember);
 389                                         if (blocked == MCAST_NOTSMEMBER ||
 390                                             blocked == MCAST_MUTED)
 391                                                 UDPSTAT_INC(udps_filtermcast);
 392                                         INP_RUNLOCK(inp); /* XXX */
 393                                         continue;
 394                                 }
 395
 396                                 INP_RUNLOCK(inp);
 397                         }
 398                         if (last != NULL) {
 399                                 struct mbuf *n;
 400
 401                                 if ((n = m_copym(m, 0, M_COPYALL, M_NOWAIT)) !=
 402                                     NULL) {
 403                                         INP_RLOCK(last);
 404                                         if (__predict_true(last->inp_flags2 & INP_FREED) == 0) {
 405                                                 if (nxt == IPPROTO_UDPLITE)
 406                                                         UDPLITE_PROBE(receive, NULL, last,
 407                                                             ip6, last, uh);
 408                                                 else
 409                                                         UDP_PROBE(receive, NULL, last,
 410                                                             ip6, last, uh);
 411                                                 if (udp6_append(last, n, off, fromsa)) {
 412                                                         /* XXX-BZ do we leak m here? */
 413                                                         *mp = NULL;
 414                                                         return (IPPROTO_DONE);
 415                                                 }
 416                                         }
 417                                         INP_RUNLOCK(last);
 418                                 }
 419                         }
 420                         last = inp;
 421                         /*
 422                          * Don't look for additional matches if this one does
 423                          * not have either the SO_REUSEPORT or SO_REUSEADDR
 424                          * socket options set.  This heuristic avoids
 425                          * searching through all pcbs in the common case of a
 426                          * non-shared port.  It assumes that an application
 427                          * will never clear these options after setting them.
 428                          */
 429                         if ((last->inp_socket->so_options &
 430                              (SO_REUSEPORT|SO_REUSEPORT_LB|SO_REUSEADDR)) == 0)
 431                                 break;
 432                 }
 433
 434                 if (last == NULL) {
 435                         /*
 436                          * No matching pcb found; discard datagram.  (No need
 437                          * to send an ICMP Port Unreachable for a broadcast
 438                          * or multicast datgram.)
 439                          */
 440                         UDPSTAT_INC(udps_noport);
 441                         UDPSTAT_INC(udps_noportmcast);
 442                         goto badunlocked;
 443                 }
 444                 INP_RLOCK(last);
 445                 if (__predict_true(last->inp_flags2 & INP_FREED) == 0) {
 446                         if (nxt == IPPROTO_UDPLITE)
 447                                 UDPLITE_PROBE(receive, NULL, last, ip6, last, uh);
 448                         else
 449                                 UDP_PROBE(receive, NULL, last, ip6, last, uh);
 450                         if (udp6_append(last, m, off, fromsa) == 0)
 451                                 INP_RUNLOCK(last);
 452                 } else
 453                         INP_RUNLOCK(last);
 454                 *mp = NULL;
 455                 return (IPPROTO_DONE);
 456         }
 457         /*
 458          * Locate pcb for datagram.
 459          */
 460
 461         /*
 462          * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
 463          */
 464         if ((m->m_flags & M_IP6_NEXTHOP) &&
 465             (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) {
 466                 struct sockaddr_in6 *next_hop6;
 467
 468                 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
 469
 470                 /*
 471                  * Transparently forwarded. Pretend to be the destination.
 472                  * Already got one like this?
 473                  */
 474                 inp = in6_pcblookup_mbuf(pcbinfo, &ip6->ip6_src,
 475                     uh->uh_sport, &ip6->ip6_dst, uh->uh_dport,
 476                     INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif, m);
 477                 if (!inp) {
 478                         /*
 479                          * It's new.  Try to find the ambushing socket.
 480                          * Because we've rewritten the destination address,
 481                          * any hardware-generated hash is ignored.
 482                          */
 483                         inp = in6_pcblookup(pcbinfo, &ip6->ip6_src,
 484                             uh->uh_sport, &next_hop6->sin6_addr,
 485                             next_hop6->sin6_port ? htons(next_hop6->sin6_port) :
 486                             uh->uh_dport, INPLOOKUP_WILDCARD |
 487                             INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif);
 488                 }
 489                 /* Remove the tag from the packet. We don't need it anymore. */
 490                 m_tag_delete(m, fwd_tag);
 491                 m->m_flags &= ~M_IP6_NEXTHOP;
 492         } else
 493                 inp = in6_pcblookup_mbuf(pcbinfo, &ip6->ip6_src,
 494                     uh->uh_sport, &ip6->ip6_dst, uh->uh_dport,
 495                     INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB,
 496                     m->m_pkthdr.rcvif, m);
 497         if (inp == NULL) {
 498                 if (V_udp_log_in_vain) {
 499                         char ip6bufs[INET6_ADDRSTRLEN];
 500                         char ip6bufd[INET6_ADDRSTRLEN];
 501
 502                         log(LOG_INFO,
 503                             "Connection attempt to UDP [%s]:%d from [%s]:%d\n",
 504                             ip6_sprintf(ip6bufd, &ip6->ip6_dst),
 505                             ntohs(uh->uh_dport),
 506                             ip6_sprintf(ip6bufs, &ip6->ip6_src),
 507                             ntohs(uh->uh_sport));
 508                 }
 509                 if (nxt == IPPROTO_UDPLITE)
 510                         UDPLITE_PROBE(receive, NULL, NULL, ip6, NULL, uh);
 511                 else
 512                         UDP_PROBE(receive, NULL, NULL, ip6, NULL, uh);
 513                 UDPSTAT_INC(udps_noport);
 514                 if (m->m_flags & M_MCAST) {
 515                         printf("UDP6: M_MCAST is set in a unicast packet.\n");
 516                         UDPSTAT_INC(udps_noportmcast);
 517                         goto badunlocked;
 518                 }
 519                 if (V_udp_blackhole)
 520                         goto badunlocked;
 521                 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_NOPORT, 0);
 522                 *mp = NULL;
 523                 return (IPPROTO_DONE);
 524         }
 525         INP_RLOCK_ASSERT(inp);
 526         up = intoudpcb(inp);
 527         if (cscov_partial) {
 528                 if (up->u_rxcslen == 0 || up->u_rxcslen > ulen) {
 529                         INP_RUNLOCK(inp);
 530                         m_freem(m);
 531                         *mp = NULL;
 532                         return (IPPROTO_DONE);
 533                 }
 534         }
 535         if (nxt == IPPROTO_UDPLITE)
 536                 UDPLITE_PROBE(receive, NULL, inp, ip6, inp, uh);
 537         else
 538                 UDP_PROBE(receive, NULL, inp, ip6, inp, uh);
 539         if (udp6_append(inp, m, off, fromsa) == 0)
 540                 INP_RUNLOCK(inp);
 541         *mp = NULL;
 542         return (IPPROTO_DONE);
 543
 544 badunlocked:
 545         m_freem(m);
 546         *mp = NULL;
 547         return (IPPROTO_DONE);
 548 }
 549
 550 static void
 551 udp6_common_ctlinput(int cmd, struct sockaddr *sa, void *d,
 552     struct inpcbinfo *pcbinfo)
 553 {
 554         struct udphdr uh;
 555         struct ip6_hdr *ip6;
 556         struct mbuf *m;
 557         int off = 0;
 558         struct ip6ctlparam *ip6cp = NULL;
 559         const struct sockaddr_in6 *sa6_src = NULL;
 560         void *cmdarg;
 561         struct inpcb *(*notify)(struct inpcb *, int) = udp_notify;
 562         struct udp_portonly {
 563                 u_int16_t uh_sport;
 564                 u_int16_t uh_dport;
 565         } *uhp;
 566
 567         if (sa->sa_family != AF_INET6 ||
 568             sa->sa_len != sizeof(struct sockaddr_in6))
 569                 return;
 570
 571         if ((unsigned)cmd >= PRC_NCMDS)
 572                 return;
 573         if (PRC_IS_REDIRECT(cmd))
 574                 notify = in6_rtchange, d = NULL;
 575         else if (cmd == PRC_HOSTDEAD)
 576                 d = NULL;
 577         else if (inet6ctlerrmap[cmd] == 0)
 578                 return;
 579
 580         /* if the parameter is from icmp6, decode it. */
 581         if (d != NULL) {
 582                 ip6cp = (struct ip6ctlparam *)d;
 583                 m = ip6cp->ip6c_m;
 584                 ip6 = ip6cp->ip6c_ip6;
 585                 off = ip6cp->ip6c_off;
 586                 cmdarg = ip6cp->ip6c_cmdarg;
 587                 sa6_src = ip6cp->ip6c_src;
 588         } else {
 589                 m = NULL;
 590                 ip6 = NULL;
 591                 cmdarg = NULL;
 592                 sa6_src = &sa6_any;
 593         }
 594
 595         if (ip6) {
 596                 /*
 597                  * XXX: We assume that when IPV6 is non NULL,
 598                  * M and OFF are valid.
 599                  */
 600
 601                 /* Check if we can safely examine src and dst ports. */
 602                 if (m->m_pkthdr.len < off + sizeof(*uhp))
 603                         return;
 604
 605                 bzero(&uh, sizeof(uh));
 606                 m_copydata(m, off, sizeof(*uhp), (caddr_t)&uh);
 607
 608                 if (!PRC_IS_REDIRECT(cmd)) {
 609                         /* Check to see if its tunneled */
 610                         struct inpcb *inp;
 611                         inp = in6_pcblookup_mbuf(pcbinfo, &ip6->ip6_dst,
 612                             uh.uh_dport, &ip6->ip6_src, uh.uh_sport,
 613                             INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB,
 614                             m->m_pkthdr.rcvif, m);
 615                         if (inp != NULL) {
 616                                 struct udpcb *up;
 617
 618                                 up = intoudpcb(inp);
 619                                 if (up->u_icmp_func) {
 620                                         /* Yes it is. */
 621                                         INP_RUNLOCK(inp);
 622                                         (*up->u_icmp_func)(cmd, (struct sockaddr *)ip6cp->ip6c_src,
 623                                               d, up->u_tun_ctx);
 624                                         return;
 625                                 } else {
 626                                         /* Can't find it. */
 627                                         INP_RUNLOCK(inp);
 628                                 }
 629                         }
 630                 }
 631                 (void)in6_pcbnotify(pcbinfo, sa, uh.uh_dport,
 632                     (struct sockaddr *)ip6cp->ip6c_src, uh.uh_sport, cmd,
 633                     cmdarg, notify);
 634         } else
 635                 (void)in6_pcbnotify(pcbinfo, sa, 0,
 636                     (const struct sockaddr *)sa6_src, 0, cmd, cmdarg, notify);
 637 }
 638
 639 void
 640 udp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
 641 {
 642
 643         return (udp6_common_ctlinput(cmd, sa, d, &V_udbinfo));
 644 }
 645
 646 void
 647 udplite6_ctlinput(int cmd, struct sockaddr *sa, void *d)
 648 {
 649
 650         return (udp6_common_ctlinput(cmd, sa, d, &V_ulitecbinfo));
 651 }
 652
 653 static int
 654 udp6_getcred(SYSCTL_HANDLER_ARGS)
 655 {
 656         struct xucred xuc;
 657         struct sockaddr_in6 addrs[2];
 658         struct epoch_tracker et;
 659         struct inpcb *inp;
 660         int error;
 661
 662         error = priv_check(req->td, PRIV_NETINET_GETCRED);
 663         if (error)
 664                 return (error);
 665
 666         if (req->newlen != sizeof(addrs))
 667                 return (EINVAL);
 668         if (req->oldlen != sizeof(struct xucred))
 669                 return (EINVAL);
 670         error = SYSCTL_IN(req, addrs, sizeof(addrs));
 671         if (error)
 672                 return (error);
 673         if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
 674             (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
 675                 return (error);
 676         }
 677         NET_EPOCH_ENTER(et);
 678         inp = in6_pcblookup(&V_udbinfo, &addrs[1].sin6_addr,
 679             addrs[1].sin6_port, &addrs[0].sin6_addr, addrs[0].sin6_port,
 680             INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
 681         NET_EPOCH_EXIT(et);
 682         if (inp != NULL) {
 683                 INP_RLOCK_ASSERT(inp);
 684                 if (inp->inp_socket == NULL)
 685                         error = ENOENT;
 686                 if (error == 0)
 687                         error = cr_canseesocket(req->td->td_ucred,
 688                             inp->inp_socket);
 689                 if (error == 0)
 690                         cru2x(inp->inp_cred, &xuc);
 691                 INP_RUNLOCK(inp);
 692         } else
 693                 error = ENOENT;
 694         if (error == 0)
 695                 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
 696         return (error);
 697 }
 698
 699 SYSCTL_PROC(_net_inet6_udp6, OID_AUTO, getcred,
 700     CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_MPSAFE,
 701     0, 0, udp6_getcred, "S,xucred",
 702     "Get the xucred of a UDP6 connection");
 703
 704 static int
 705 udp6_output(struct socket *so, int flags_arg, struct mbuf *m,
 706     struct sockaddr *addr6, struct mbuf *control, struct thread *td)
 707 {
 708         struct inpcb *inp;
 709         struct ip6_hdr *ip6;
 710         struct udphdr *udp6;
 711         struct in6_addr *laddr, *faddr, in6a;
 712         struct ip6_pktopts *optp, opt;
 713         struct sockaddr_in6 *sin6, tmp;
 714         struct epoch_tracker et;
 715         int cscov_partial, error, flags, hlen, scope_ambiguous;
 716         u_int32_t ulen, plen;
 717         uint16_t cscov;
 718         u_short fport;
 719         uint8_t nxt;
 720
 721         /* addr6 has been validated in udp6_send(). */
 722         sin6 = (struct sockaddr_in6 *)addr6;
 723
 724         /*
 725          * In contrast to to IPv4 we do not validate the max. packet length
 726          * here due to IPv6 Jumbograms (RFC2675).
 727          */
 728
 729         scope_ambiguous = 0;
 730         if (sin6) {
 731                 /* Protect *addr6 from overwrites. */
 732                 tmp = *sin6;
 733                 sin6 = &tmp;
 734
 735                 /*
 736                  * Application should provide a proper zone ID or the use of
 737                  * default zone IDs should be enabled.  Unfortunately, some
 738                  * applications do not behave as it should, so we need a
 739                  * workaround.  Even if an appropriate ID is not determined,
 740                  * we'll see if we can determine the outgoing interface.  If we
 741                  * can, determine the zone ID based on the interface below.
 742                  */
 743                 if (sin6->sin6_scope_id == 0 && !V_ip6_use_defzone)
 744                         scope_ambiguous = 1;
 745                 if ((error = sa6_embedscope(sin6, V_ip6_use_defzone)) != 0) {
 746                         if (control)
 747                                 m_freem(control);
 748                         m_freem(m);
 749                         return (error);
 750                 }
 751         }
 752
 753         inp = sotoinpcb(so);
 754         KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
 755         /*
 756          * In the following cases we want a write lock on the inp for either
 757          * local operations or for possible route cache updates in the IPv6
 758          * output path:
 759          * - on connected sockets (sin6 is NULL) for route cache updates,
 760          * - when we are not bound to an address and source port (it is
 761          *   in6_pcbsetport() which will require the write lock).
 762          *
 763          * We check the inp fields before actually locking the inp, so
 764          * here exists a race, and we may WLOCK the inp and end with already
 765          * bound one by other thread. This is fine.
 766          */
 767         if (sin6 == NULL || (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) &&
 768             inp->inp_lport == 0))
 769                 INP_WLOCK(inp);
 770         else
 771                 INP_RLOCK(inp);
 772
 773         nxt = (inp->inp_socket->so_proto->pr_protocol == IPPROTO_UDP) ?
 774             IPPROTO_UDP : IPPROTO_UDPLITE;
 775
 776 #ifdef INET
 777         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
 778                 int hasv4addr;
 779
 780                 if (sin6 == NULL)
 781                         hasv4addr = (inp->inp_vflag & INP_IPV4);
 782                 else
 783                         hasv4addr = IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)
 784                             ? 1 : 0;
 785                 if (hasv4addr) {
 786                         struct pr_usrreqs *pru;
 787
 788                         /*
 789                          * XXXRW: We release UDP-layer locks before calling
 790                          * udp_send() in order to avoid recursion.  However,
 791                          * this does mean there is a short window where inp's
 792                          * fields are unstable.  Could this lead to a
 793                          * potential race in which the factors causing us to
 794                          * select the UDPv4 output routine are invalidated?
 795                          */
 796                         INP_UNLOCK(inp);
 797                         if (sin6)
 798                                 in6_sin6_2_sin_in_sock((struct sockaddr *)sin6);
 799                         pru = inetsw[ip_protox[nxt]].pr_usrreqs;
 800                         /* addr will just be freed in sendit(). */
 801                         return ((*pru->pru_send)(so, flags_arg | PRUS_IPV6, m,
 802                             (struct sockaddr *)sin6, control, td));
 803                 }
 804         } else
 805 #endif
 806         if (sin6 && IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
 807                 /*
 808                  * Given this is either an IPv6-only socket or no INET is
 809                  * supported we will fail the send if the given destination
 810                  * address is a v4mapped address.
 811                  *
 812                  * XXXGL: do we leak m and control?
 813                  */
 814                 INP_UNLOCK(inp);
 815                 return (EINVAL);
 816         }
 817
 818         if (control) {
 819                 if ((error = ip6_setpktopts(control, &opt,
 820                     inp->in6p_outputopts, td->td_ucred, nxt)) != 0) {
 821                         INP_UNLOCK(inp);
 822                         ip6_clearpktopts(&opt, -1);
 823                         if (control)
 824                                 m_freem(control);
 825                         m_freem(m);
 826                         return (error);
 827                 }
 828                 optp = &opt;
 829         } else
 830                 optp = inp->in6p_outputopts;
 831
 832         NET_EPOCH_ENTER(et);
 833         if (sin6) {
 834                 /*
 835                  * Since we saw no essential reason for calling in_pcbconnect,
 836                  * we get rid of such kind of logic, and call in6_selectsrc
 837                  * and in6_pcbsetport in order to fill in the local address
 838                  * and the local port.
 839                  */
 840                 if (sin6->sin6_port == 0) {
 841                         error = EADDRNOTAVAIL;
 842                         goto release;
 843                 }
 844
 845                 if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
 846                         /* how about ::ffff:0.0.0.0 case? */
 847                         error = EISCONN;
 848                         goto release;
 849                 }
 850
 851                 /*
 852                  * Given we handle the v4mapped case in the INET block above
 853                  * assert here that it must not happen anymore.
 854                  */
 855                 KASSERT(!IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr),
 856                     ("%s: sin6(%p)->sin6_addr is v4mapped which we "
 857                     "should have handled.", __func__, sin6));
 858
 859                 /* This only requires read-locking. */
 860                 error = in6_selectsrc_socket(sin6, optp, inp,
 861                     td->td_ucred, scope_ambiguous, &in6a, NULL);
 862                 if (error)
 863                         goto release;
 864                 laddr = &in6a;
 865
 866                 if (inp->inp_lport == 0) {
 867                         struct inpcbinfo *pcbinfo;
 868
 869                         INP_WLOCK_ASSERT(inp);
 870
 871                         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
 872                         INP_HASH_WLOCK(pcbinfo);
 873                         error = in6_pcbsetport(laddr, inp, td->td_ucred);
 874                         INP_HASH_WUNLOCK(pcbinfo);
 875                         if (error != 0) {
 876                                 /* Undo an address bind that may have occurred. */
 877                                 inp->in6p_laddr = in6addr_any;
 878                                 goto release;
 879                         }
 880                 }
 881                 faddr = &sin6->sin6_addr;
 882                 fport = sin6->sin6_port; /* allow 0 port */
 883
 884         } else {
 885                 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
 886                         error = ENOTCONN;
 887                         goto release;
 888                 }
 889                 laddr = &inp->in6p_laddr;
 890                 faddr = &inp->in6p_faddr;
 891                 fport = inp->inp_fport;
 892         }
 893
 894         ulen = m->m_pkthdr.len;
 895         plen = sizeof(struct udphdr) + ulen;
 896         hlen = sizeof(struct ip6_hdr);
 897
 898         /*
 899          * Calculate data length and get a mbuf
 900          * for UDP and IP6 headers.
 901          */
 902         M_PREPEND(m, hlen + sizeof(struct udphdr), M_NOWAIT);
 903         if (m == NULL) {
 904                 error = ENOBUFS;
 905                 goto release;
 906         }
 907
 908         /*
 909          * Stuff checksum and output datagram.
 910          */
 911         cscov = cscov_partial = 0;
 912         udp6 = (struct udphdr *)(mtod(m, caddr_t) + hlen);
 913         udp6->uh_sport = inp->inp_lport; /* lport is always set in the PCB */
 914         udp6->uh_dport = fport;
 915         if (nxt == IPPROTO_UDPLITE) {
 916                 struct udpcb *up;
 917
 918                 up = intoudpcb(inp);
 919                 cscov = up->u_txcslen;
 920                 if (cscov >= plen)
 921                         cscov = 0;
 922                 udp6->uh_ulen = htons(cscov);
 923                 /*
 924                  * For UDP-Lite, checksum coverage length of zero means
 925                  * the entire UDPLite packet is covered by the checksum.
 926                  */
 927                 cscov_partial = (cscov == 0) ? 0 : 1;
 928         } else if (plen <= 0xffff)
 929                 udp6->uh_ulen = htons((u_short)plen);
 930         else
 931                 udp6->uh_ulen = 0;
 932         udp6->uh_sum = 0;
 933
 934         ip6 = mtod(m, struct ip6_hdr *);
 935         ip6->ip6_flow   = inp->inp_flow & IPV6_FLOWINFO_MASK;
 936         ip6->ip6_vfc    &= ~IPV6_VERSION_MASK;
 937         ip6->ip6_vfc    |= IPV6_VERSION;
 938         ip6->ip6_plen   = htons((u_short)plen);
 939         ip6->ip6_nxt    = nxt;
 940         ip6->ip6_hlim   = in6_selecthlim(inp, NULL);
 941         ip6->ip6_src    = *laddr;
 942         ip6->ip6_dst    = *faddr;
 943
 944 #ifdef MAC
 945         mac_inpcb_create_mbuf(inp, m);
 946 #endif
 947
 948         if (cscov_partial) {
 949                 if ((udp6->uh_sum = in6_cksum_partial(m, nxt,
 950                     sizeof(struct ip6_hdr), plen, cscov)) == 0)
 951                         udp6->uh_sum = 0xffff;
 952         } else {
 953                 udp6->uh_sum = in6_cksum_pseudo(ip6, plen, nxt, 0);
 954                 m->m_pkthdr.csum_flags = CSUM_UDP_IPV6;
 955                 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
 956         }
 957
 958         flags = 0;
 959 #if defined(ROUTE_MPATH) || defined(RSS)
 960         if (CALC_FLOWID_OUTBOUND_SENDTO) {
 961                 uint32_t hash_type, hash_val;
 962                 uint8_t pr;
 963
 964                 pr = inp->inp_socket->so_proto->pr_protocol;
 965
 966                 hash_val = fib6_calc_packet_hash(laddr, faddr,
 967                     inp->inp_lport, fport, pr, &hash_type);
 968                 m->m_pkthdr.flowid = hash_val;
 969                 M_HASHTYPE_SET(m, hash_type);
 970         }
 971         /* do not use inp flowid */
 972         flags |= IP_NODEFAULTFLOWID;
 973 #endif
 974
 975         UDPSTAT_INC(udps_opackets);
 976         if (nxt == IPPROTO_UDPLITE)
 977                 UDPLITE_PROBE(send, NULL, inp, ip6, inp, udp6);
 978         else
 979                 UDP_PROBE(send, NULL, inp, ip6, inp, udp6);
 980         error = ip6_output(m, optp,
 981             INP_WLOCKED(inp) ? &inp->inp_route6 : NULL, flags,
 982             inp->in6p_moptions, NULL, inp);
 983         INP_UNLOCK(inp);
 984         NET_EPOCH_EXIT(et);
 985
 986         if (control) {
 987                 ip6_clearpktopts(&opt, -1);
 988                 m_freem(control);
 989         }
 990         return (error);
 991
 992 release:
 993         INP_UNLOCK(inp);
 994         NET_EPOCH_EXIT(et);
 995         if (control) {
 996                 ip6_clearpktopts(&opt, -1);
 997                 m_freem(control);
 998         }
 999         m_freem(m);
1000
1001         return (error);
1002 }
1003
1004 static void
1005 udp6_abort(struct socket *so)
1006 {
1007         struct inpcb *inp;
1008         struct inpcbinfo *pcbinfo;
1009
1010         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1011         inp = sotoinpcb(so);
1012         KASSERT(inp != NULL, ("udp6_abort: inp == NULL"));
1013
1014         INP_WLOCK(inp);
1015 #ifdef INET
1016         if (inp->inp_vflag & INP_IPV4) {
1017                 struct pr_usrreqs *pru;
1018                 uint8_t nxt;
1019
1020                 nxt = (inp->inp_socket->so_proto->pr_protocol == IPPROTO_UDP) ?
1021                     IPPROTO_UDP : IPPROTO_UDPLITE;
1022                 INP_WUNLOCK(inp);
1023                 pru = inetsw[ip_protox[nxt]].pr_usrreqs;
1024                 (*pru->pru_abort)(so);
1025                 return;
1026         }
1027 #endif
1028
1029         if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
1030                 INP_HASH_WLOCK(pcbinfo);
1031                 in6_pcbdisconnect(inp);
1032                 inp->in6p_laddr = in6addr_any;
1033                 INP_HASH_WUNLOCK(pcbinfo);
1034                 soisdisconnected(so);
1035         }
1036         INP_WUNLOCK(inp);
1037 }
1038
1039 static int
1040 udp6_attach(struct socket *so, int proto, struct thread *td)
1041 {
1042         struct inpcb *inp;
1043         struct inpcbinfo *pcbinfo;
1044         int error;
1045
1046         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1047         inp = sotoinpcb(so);
1048         KASSERT(inp == NULL, ("udp6_attach: inp != NULL"));
1049
1050         if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
1051                 error = soreserve(so, udp_sendspace, udp_recvspace);
1052                 if (error)
1053                         return (error);
1054         }
1055         INP_INFO_WLOCK(pcbinfo);
1056         error = in_pcballoc(so, pcbinfo);
1057         if (error) {
1058                 INP_INFO_WUNLOCK(pcbinfo);
1059                 return (error);
1060         }
1061         inp = (struct inpcb *)so->so_pcb;
1062         inp->inp_vflag |= INP_IPV6;
1063         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
1064                 inp->inp_vflag |= INP_IPV4;
1065         inp->in6p_hops = -1;    /* use kernel default */
1066         inp->in6p_cksum = -1;   /* just to be sure */
1067         /*
1068          * XXX: ugly!!
1069          * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1070          * because the socket may be bound to an IPv6 wildcard address,
1071          * which may match an IPv4-mapped IPv6 address.
1072          */
1073         inp->inp_ip_ttl = V_ip_defttl;
1074
1075         error = udp_newudpcb(inp);
1076         if (error) {
1077                 in_pcbdetach(inp);
1078                 in_pcbfree(inp);
1079                 INP_INFO_WUNLOCK(pcbinfo);
1080                 return (error);
1081         }
1082         INP_WUNLOCK(inp);
1083         INP_INFO_WUNLOCK(pcbinfo);
1084         return (0);
1085 }
1086
1087 static int
1088 udp6_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1089 {
1090         struct inpcb *inp;
1091         struct inpcbinfo *pcbinfo;
1092         int error;
1093         u_char vflagsav;
1094
1095         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1096         inp = sotoinpcb(so);
1097         KASSERT(inp != NULL, ("udp6_bind: inp == NULL"));
1098
1099         INP_WLOCK(inp);
1100         INP_HASH_WLOCK(pcbinfo);
1101         vflagsav = inp->inp_vflag;
1102         inp->inp_vflag &= ~INP_IPV4;
1103         inp->inp_vflag |= INP_IPV6;
1104         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
1105                 struct sockaddr_in6 *sin6_p;
1106
1107                 sin6_p = (struct sockaddr_in6 *)nam;
1108
1109                 if (IN6_IS_ADDR_UNSPECIFIED(&sin6_p->sin6_addr))
1110                         inp->inp_vflag |= INP_IPV4;
1111 #ifdef INET
1112                 else if (IN6_IS_ADDR_V4MAPPED(&sin6_p->sin6_addr)) {
1113                         struct sockaddr_in sin;
1114
1115                         in6_sin6_2_sin(&sin, sin6_p);
1116                         inp->inp_vflag |= INP_IPV4;
1117                         inp->inp_vflag &= ~INP_IPV6;
1118                         error = in_pcbbind(inp, (struct sockaddr *)&sin,
1119                             td->td_ucred);
1120                         goto out;
1121                 }
1122 #endif
1123         }
1124
1125         error = in6_pcbbind(inp, nam, td->td_ucred);
1126 #ifdef INET
1127 out:
1128 #endif
1129         if (error != 0)
1130                 inp->inp_vflag = vflagsav;
1131         INP_HASH_WUNLOCK(pcbinfo);
1132         INP_WUNLOCK(inp);
1133         return (error);
1134 }
1135
1136 static void
1137 udp6_close(struct socket *so)
1138 {
1139         struct inpcb *inp;
1140         struct inpcbinfo *pcbinfo;
1141
1142         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1143         inp = sotoinpcb(so);
1144         KASSERT(inp != NULL, ("udp6_close: inp == NULL"));
1145
1146         INP_WLOCK(inp);
1147 #ifdef INET
1148         if (inp->inp_vflag & INP_IPV4) {
1149                 struct pr_usrreqs *pru;
1150                 uint8_t nxt;
1151
1152                 nxt = (inp->inp_socket->so_proto->pr_protocol == IPPROTO_UDP) ?
1153                     IPPROTO_UDP : IPPROTO_UDPLITE;
1154                 INP_WUNLOCK(inp);
1155                 pru = inetsw[ip_protox[nxt]].pr_usrreqs;
1156                 (*pru->pru_disconnect)(so);
1157                 return;
1158         }
1159 #endif
1160         if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
1161                 INP_HASH_WLOCK(pcbinfo);
1162                 in6_pcbdisconnect(inp);
1163                 inp->in6p_laddr = in6addr_any;
1164                 INP_HASH_WUNLOCK(pcbinfo);
1165                 soisdisconnected(so);
1166         }
1167         INP_WUNLOCK(inp);
1168 }
1169
1170 static int
1171 udp6_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1172 {
1173 #ifdef INET
1174         struct epoch_tracker et;
1175 #endif
1176         struct inpcb *inp;
1177         struct inpcbinfo *pcbinfo;
1178         struct sockaddr_in6 *sin6;
1179         int error;
1180         u_char vflagsav;
1181
1182         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1183         inp = sotoinpcb(so);
1184         sin6 = (struct sockaddr_in6 *)nam;
1185         KASSERT(inp != NULL, ("udp6_connect: inp == NULL"));
1186
1187         /*
1188          * XXXRW: Need to clarify locking of v4/v6 flags.
1189          */
1190         INP_WLOCK(inp);
1191 #ifdef INET
1192         if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
1193                 struct sockaddr_in sin;
1194
1195                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
1196                         error = EINVAL;
1197                         goto out;
1198                 }
1199                 if ((inp->inp_vflag & INP_IPV4) == 0) {
1200                         error = EAFNOSUPPORT;
1201                         goto out;
1202                 }
1203                 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1204                         error = EISCONN;
1205                         goto out;
1206                 }
1207                 in6_sin6_2_sin(&sin, sin6);
1208                 error = prison_remote_ip4(td->td_ucred, &sin.sin_addr);
1209                 if (error != 0)
1210                         goto out;
1211                 vflagsav = inp->inp_vflag;
1212                 inp->inp_vflag |= INP_IPV4;
1213                 inp->inp_vflag &= ~INP_IPV6;
1214                 NET_EPOCH_ENTER(et);
1215                 INP_HASH_WLOCK(pcbinfo);
1216                 error = in_pcbconnect(inp, (struct sockaddr *)&sin,
1217                     td->td_ucred);
1218                 INP_HASH_WUNLOCK(pcbinfo);
1219                 NET_EPOCH_EXIT(et);
1220                 /*
1221                  * If connect succeeds, mark socket as connected. If
1222                  * connect fails and socket is unbound, reset inp_vflag
1223                  * field.
1224                  */
1225                 if (error == 0)
1226                         soisconnected(so);
1227                 else if (inp->inp_laddr.s_addr == INADDR_ANY &&
1228                     inp->inp_lport == 0)
1229                         inp->inp_vflag = vflagsav;
1230                 goto out;
1231         } else {
1232                 if ((inp->inp_vflag & INP_IPV6) == 0) {
1233                         error = EAFNOSUPPORT;
1234                         goto out;
1235                 }
1236         }
1237 #endif
1238         if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
1239                 error = EISCONN;
1240                 goto out;
1241         }
1242         error = prison_remote_ip6(td->td_ucred, &sin6->sin6_addr);
1243         if (error != 0)
1244                 goto out;
1245         vflagsav = inp->inp_vflag;
1246         inp->inp_vflag &= ~INP_IPV4;
1247         inp->inp_vflag |= INP_IPV6;
1248         INP_HASH_WLOCK(pcbinfo);
1249         error = in6_pcbconnect(inp, nam, td->td_ucred);
1250         INP_HASH_WUNLOCK(pcbinfo);
1251         /*
1252          * If connect succeeds, mark socket as connected. If
1253          * connect fails and socket is unbound, reset inp_vflag
1254          * field.
1255          */
1256         if (error == 0)
1257                 soisconnected(so);
1258         else if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) &&
1259             inp->inp_lport == 0)
1260                 inp->inp_vflag = vflagsav;
1261 out:
1262         INP_WUNLOCK(inp);
1263         return (error);
1264 }
1265
1266 static void
1267 udp6_detach(struct socket *so)
1268 {
1269         struct inpcb *inp;
1270         struct inpcbinfo *pcbinfo;
1271         struct udpcb *up;
1272
1273         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1274         inp = sotoinpcb(so);
1275         KASSERT(inp != NULL, ("udp6_detach: inp == NULL"));
1276
1277         INP_INFO_WLOCK(pcbinfo);
1278         INP_WLOCK(inp);
1279         up = intoudpcb(inp);
1280         KASSERT(up != NULL, ("%s: up == NULL", __func__));
1281         in_pcbdetach(inp);
1282         in_pcbfree(inp);
1283         INP_INFO_WUNLOCK(pcbinfo);
1284         udp_discardcb(up);
1285 }
1286
1287 static int
1288 udp6_disconnect(struct socket *so)
1289 {
1290         struct inpcb *inp;
1291         struct inpcbinfo *pcbinfo;
1292
1293         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1294         inp = sotoinpcb(so);
1295         KASSERT(inp != NULL, ("udp6_disconnect: inp == NULL"));
1296
1297         INP_WLOCK(inp);
1298 #ifdef INET
1299         if (inp->inp_vflag & INP_IPV4) {
1300                 struct pr_usrreqs *pru;
1301                 uint8_t nxt;
1302
1303                 nxt = (inp->inp_socket->so_proto->pr_protocol == IPPROTO_UDP) ?
1304                     IPPROTO_UDP : IPPROTO_UDPLITE;
1305                 INP_WUNLOCK(inp);
1306                 pru = inetsw[ip_protox[nxt]].pr_usrreqs;
1307                 (void)(*pru->pru_disconnect)(so);
1308                 return (0);
1309         }
1310 #endif
1311
1312         if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
1313                 INP_WUNLOCK(inp);
1314                 return (ENOTCONN);
1315         }
1316
1317         INP_HASH_WLOCK(pcbinfo);
1318         in6_pcbdisconnect(inp);
1319         inp->in6p_laddr = in6addr_any;
1320         INP_HASH_WUNLOCK(pcbinfo);
1321         SOCK_LOCK(so);
1322         so->so_state &= ~SS_ISCONNECTED;                /* XXX */
1323         SOCK_UNLOCK(so);
1324         INP_WUNLOCK(inp);
1325         return (0);
1326 }
1327
1328 static int
1329 udp6_send(struct socket *so, int flags, struct mbuf *m,
1330     struct sockaddr *addr, struct mbuf *control, struct thread *td)
1331 {
1332         int error;
1333
1334         if (addr) {
1335                 if (addr->sa_len != sizeof(struct sockaddr_in6)) {
1336                         error = EINVAL;
1337                         goto bad;
1338                 }
1339                 if (addr->sa_family != AF_INET6) {
1340                         error = EAFNOSUPPORT;
1341                         goto bad;
1342                 }
1343         }
1344
1345         return (udp6_output(so, flags, m, addr, control, td));
1346
1347 bad:
1348         if (control)
1349                 m_freem(control);
1350         m_freem(m);
1351         return (error);
1352 }
1353
1354 struct pr_usrreqs udp6_usrreqs = {
1355         .pru_abort =            udp6_abort,
1356         .pru_attach =           udp6_attach,
1357         .pru_bind =             udp6_bind,
1358         .pru_connect =          udp6_connect,
1359         .pru_control =          in6_control,
1360         .pru_detach =           udp6_detach,
1361         .pru_disconnect =       udp6_disconnect,
1362         .pru_peeraddr =         in6_mapped_peeraddr,
1363         .pru_send =             udp6_send,
1364         .pru_shutdown =         udp_shutdown,
1365         .pru_sockaddr =         in6_mapped_sockaddr,
1366         .pru_soreceive =        soreceive_dgram,
1367         .pru_sosend =           sosend_dgram,
1368         .pru_sosetlabel =       in_pcbsosetlabel,
1369         .pru_close =            udp6_close
1370 };