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 /*
 128  * UDP protocol implementation.
 129  * Per RFC 768, August, 1980.
 130  */
 131
 132 extern struct protosw   inetsw[];
 133 static void             udp6_detach(struct socket *so);
 134
 135 static int
 136 udp6_append(struct inpcb *inp, struct mbuf *n, int off,
 137     struct sockaddr_in6 *fromsa)
 138 {
 139         struct socket *so;
 140         struct mbuf *opts = NULL, *tmp_opts;
 141         struct udpcb *up;
 142
 143         INP_LOCK_ASSERT(inp);
 144
 145         /*
 146          * Engage the tunneling protocol.
 147          */
 148         up = intoudpcb(inp);
 149         if (up->u_tun_func != NULL) {
 150                 in_pcbref(inp);
 151                 INP_RUNLOCK(inp);
 152                 (*up->u_tun_func)(n, off, inp, (struct sockaddr *)&fromsa[0],
 153                     up->u_tun_ctx);
 154                 INP_RLOCK(inp);
 155                 return (in_pcbrele_rlocked(inp));
 156         }
 157 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
 158         /* Check AH/ESP integrity. */
 159         if (IPSEC_ENABLED(ipv6)) {
 160                 if (IPSEC_CHECK_POLICY(ipv6, n, inp) != 0) {
 161                         m_freem(n);
 162                         return (0);
 163                 }
 164         }
 165 #endif /* IPSEC */
 166 #ifdef MAC
 167         if (mac_inpcb_check_deliver(inp, n) != 0) {
 168                 m_freem(n);
 169                 return (0);
 170         }
 171 #endif
 172         opts = NULL;
 173         if (inp->inp_flags & INP_CONTROLOPTS ||
 174             inp->inp_socket->so_options & SO_TIMESTAMP)
 175                 ip6_savecontrol(inp, n, &opts);
 176         if ((inp->inp_vflag & INP_IPV6) && (inp->inp_flags2 & INP_ORIGDSTADDR)) {
 177                 tmp_opts = sbcreatecontrol((caddr_t)&fromsa[1],
 178                         sizeof(struct sockaddr_in6), IPV6_ORIGDSTADDR, IPPROTO_IPV6);
 179                 if (tmp_opts) {
 180                         if (opts) {
 181                                 tmp_opts->m_next = opts;
 182                                 opts = tmp_opts;
 183                         } else
 184                                 opts = tmp_opts;
 185                 }
 186
 187         }
 188         m_adj(n, off + sizeof(struct udphdr));
 189
 190         so = inp->inp_socket;
 191         SOCKBUF_LOCK(&so->so_rcv);
 192         if (sbappendaddr_locked(&so->so_rcv, (struct sockaddr *)&fromsa[0], n,
 193             opts) == 0) {
 194                 SOCKBUF_UNLOCK(&so->so_rcv);
 195                 m_freem(n);
 196                 if (opts)
 197                         m_freem(opts);
 198                 UDPSTAT_INC(udps_fullsock);
 199         } else
 200                 sorwakeup_locked(so);
 201         return (0);
 202 }
 203
 204 int
 205 udp6_input(struct mbuf **mp, int *offp, int proto)
 206 {
 207         struct mbuf *m = *mp;
 208         struct ifnet *ifp;
 209         struct ip6_hdr *ip6;
 210         struct udphdr *uh;
 211         struct inpcb *inp;
 212         struct inpcbinfo *pcbinfo;
 213         struct udpcb *up;
 214         int off = *offp;
 215         int cscov_partial;
 216         int plen, ulen;
 217         struct sockaddr_in6 fromsa[2];
 218         struct m_tag *fwd_tag;
 219         uint16_t uh_sum;
 220         uint8_t nxt;
 221
 222         NET_EPOCH_ASSERT();
 223
 224         ifp = m->m_pkthdr.rcvif;
 225
 226 #ifndef PULLDOWN_TEST
 227         IP6_EXTHDR_CHECK(m, off, sizeof(struct udphdr), IPPROTO_DONE);
 228         ip6 = mtod(m, struct ip6_hdr *);
 229         uh = (struct udphdr *)((caddr_t)ip6 + off);
 230 #else
 231         IP6_EXTHDR_GET(uh, struct udphdr *, m, off, sizeof(*uh));
 232         if (!uh)
 233                 return (IPPROTO_DONE);
 234         ip6 = mtod(m, struct ip6_hdr *);
 235 #endif
 236
 237         UDPSTAT_INC(udps_ipackets);
 238
 239         /*
 240          * Destination port of 0 is illegal, based on RFC768.
 241          */
 242         if (uh->uh_dport == 0)
 243                 goto badunlocked;
 244
 245         plen = ntohs(ip6->ip6_plen) - off + sizeof(*ip6);
 246         ulen = ntohs((u_short)uh->uh_ulen);
 247
 248         nxt = proto;
 249         cscov_partial = (nxt == IPPROTO_UDPLITE) ? 1 : 0;
 250         if (nxt == IPPROTO_UDPLITE) {
 251                 /* Zero means checksum over the complete packet. */
 252                 if (ulen == 0)
 253                         ulen = plen;
 254                 if (ulen == plen)
 255                         cscov_partial = 0;
 256                 if ((ulen < sizeof(struct udphdr)) || (ulen > plen)) {
 257                         /* XXX: What is the right UDPLite MIB counter? */
 258                         goto badunlocked;
 259                 }
 260                 if (uh->uh_sum == 0) {
 261                         /* XXX: What is the right UDPLite MIB counter? */
 262                         goto badunlocked;
 263                 }
 264         } else {
 265                 if ((ulen < sizeof(struct udphdr)) || (plen != ulen)) {
 266                         UDPSTAT_INC(udps_badlen);
 267                         goto badunlocked;
 268                 }
 269                 if (uh->uh_sum == 0) {
 270                         UDPSTAT_INC(udps_nosum);
 271                         goto badunlocked;
 272                 }
 273         }
 274
 275         if ((m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) &&
 276             !cscov_partial) {
 277                 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
 278                         uh_sum = m->m_pkthdr.csum_data;
 279                 else
 280                         uh_sum = in6_cksum_pseudo(ip6, ulen, nxt,
 281                             m->m_pkthdr.csum_data);
 282                 uh_sum ^= 0xffff;
 283         } else
 284                 uh_sum = in6_cksum_partial(m, nxt, off, plen, ulen);
 285
 286         if (uh_sum != 0) {
 287                 UDPSTAT_INC(udps_badsum);
 288                 goto badunlocked;
 289         }
 290
 291         /*
 292          * Construct sockaddr format source address.
 293          */
 294         init_sin6(&fromsa[0], m, 0);
 295         fromsa[0].sin6_port = uh->uh_sport;
 296         init_sin6(&fromsa[1], m, 1);
 297         fromsa[1].sin6_port = uh->uh_dport;
 298
 299         pcbinfo = udp_get_inpcbinfo(nxt);
 300         if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) {
 301                 struct inpcb *last;
 302                 struct inpcbhead *pcblist;
 303                 struct ip6_moptions *imo;
 304
 305                 /*
 306                  * In the event that laddr should be set to the link-local
 307                  * address (this happens in RIPng), the multicast address
 308                  * specified in the received packet will not match laddr.  To
 309                  * handle this situation, matching is relaxed if the
 310                  * receiving interface is the same as one specified in the
 311                  * socket and if the destination multicast address matches
 312                  * one of the multicast groups specified in the socket.
 313                  */
 314
 315                 /*
 316                  * KAME note: traditionally we dropped udpiphdr from mbuf
 317                  * here.  We need udphdr for IPsec processing so we do that
 318                  * later.
 319                  */
 320                 pcblist = udp_get_pcblist(nxt);
 321                 last = NULL;
 322                 CK_LIST_FOREACH(inp, pcblist, inp_list) {
 323                         if ((inp->inp_vflag & INP_IPV6) == 0)
 324                                 continue;
 325                         if (inp->inp_lport != uh->uh_dport)
 326                                 continue;
 327                         if (inp->inp_fport != 0 &&
 328                             inp->inp_fport != uh->uh_sport)
 329                                 continue;
 330                         if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) {
 331                                 if (!IN6_ARE_ADDR_EQUAL(&inp->in6p_laddr,
 332                                                         &ip6->ip6_dst))
 333                                         continue;
 334                         }
 335                         if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
 336                                 if (!IN6_ARE_ADDR_EQUAL(&inp->in6p_faddr,
 337                                                         &ip6->ip6_src) ||
 338                                     inp->inp_fport != uh->uh_sport)
 339                                         continue;
 340                         }
 341
 342                         /*
 343                          * XXXRW: Because we weren't holding either the inpcb
 344                          * or the hash lock when we checked for a match
 345                          * before, we should probably recheck now that the
 346                          * inpcb lock is (supposed to be) held.
 347                          */
 348
 349                         /*
 350                          * Handle socket delivery policy for any-source
 351                          * and source-specific multicast. [RFC3678]
 352                          */
 353                         imo = inp->in6p_moptions;
 354                         if (imo && IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) {
 355                                 struct sockaddr_in6      mcaddr;
 356                                 int                      blocked;
 357
 358                                 INP_RLOCK(inp);
 359                                 if (__predict_false(inp->inp_flags2 & INP_FREED)) {
 360                                         INP_RUNLOCK(inp);
 361                                         continue;
 362                                 }
 363
 364                                 bzero(&mcaddr, sizeof(struct sockaddr_in6));
 365                                 mcaddr.sin6_len = sizeof(struct sockaddr_in6);
 366                                 mcaddr.sin6_family = AF_INET6;
 367                                 mcaddr.sin6_addr = ip6->ip6_dst;
 368
 369                                 blocked = im6o_mc_filter(imo, ifp,
 370                                         (struct sockaddr *)&mcaddr,
 371                                         (struct sockaddr *)&fromsa[0]);
 372                                 if (blocked != MCAST_PASS) {
 373                                         if (blocked == MCAST_NOTGMEMBER)
 374                                                 IP6STAT_INC(ip6s_notmember);
 375                                         if (blocked == MCAST_NOTSMEMBER ||
 376                                             blocked == MCAST_MUTED)
 377                                                 UDPSTAT_INC(udps_filtermcast);
 378                                         INP_RUNLOCK(inp); /* XXX */
 379                                         continue;
 380                                 }
 381
 382                                 INP_RUNLOCK(inp);
 383                         }
 384                         if (last != NULL) {
 385                                 struct mbuf *n;
 386
 387                                 if ((n = m_copym(m, 0, M_COPYALL, M_NOWAIT)) !=
 388                                     NULL) {
 389                                         INP_RLOCK(last);
 390                                         if (__predict_true(last->inp_flags2 & INP_FREED) == 0) {
 391                                                 if (nxt == IPPROTO_UDPLITE)
 392                                                         UDPLITE_PROBE(receive, NULL, last,
 393                                                             ip6, last, uh);
 394                                                 else
 395                                                         UDP_PROBE(receive, NULL, last,
 396                                                             ip6, last, uh);
 397                                                 if (udp6_append(last, n, off, fromsa)) {
 398                                                         /* XXX-BZ do we leak m here? */
 399                                                         *mp = NULL;
 400                                                         return (IPPROTO_DONE);
 401                                                 }
 402                                         }
 403                                         INP_RUNLOCK(last);
 404                                 }
 405                         }
 406                         last = inp;
 407                         /*
 408                          * Don't look for additional matches if this one does
 409                          * not have either the SO_REUSEPORT or SO_REUSEADDR
 410                          * socket options set.  This heuristic avoids
 411                          * searching through all pcbs in the common case of a
 412                          * non-shared port.  It assumes that an application
 413                          * will never clear these options after setting them.
 414                          */
 415                         if ((last->inp_socket->so_options &
 416                              (SO_REUSEPORT|SO_REUSEPORT_LB|SO_REUSEADDR)) == 0)
 417                                 break;
 418                 }
 419
 420                 if (last == NULL) {
 421                         /*
 422                          * No matching pcb found; discard datagram.  (No need
 423                          * to send an ICMP Port Unreachable for a broadcast
 424                          * or multicast datgram.)
 425                          */
 426                         UDPSTAT_INC(udps_noport);
 427                         UDPSTAT_INC(udps_noportmcast);
 428                         goto badunlocked;
 429                 }
 430                 INP_RLOCK(last);
 431                 if (__predict_true(last->inp_flags2 & INP_FREED) == 0) {
 432                         if (nxt == IPPROTO_UDPLITE)
 433                                 UDPLITE_PROBE(receive, NULL, last, ip6, last, uh);
 434                         else
 435                                 UDP_PROBE(receive, NULL, last, ip6, last, uh);
 436                         if (udp6_append(last, m, off, fromsa) == 0)
 437                                 INP_RUNLOCK(last);
 438                 } else
 439                         INP_RUNLOCK(last);
 440                 *mp = NULL;
 441                 return (IPPROTO_DONE);
 442         }
 443         /*
 444          * Locate pcb for datagram.
 445          */
 446
 447         /*
 448          * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
 449          */
 450         if ((m->m_flags & M_IP6_NEXTHOP) &&
 451             (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) {
 452                 struct sockaddr_in6 *next_hop6;
 453
 454                 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
 455
 456                 /*
 457                  * Transparently forwarded. Pretend to be the destination.
 458                  * Already got one like this?
 459                  */
 460                 inp = in6_pcblookup_mbuf(pcbinfo, &ip6->ip6_src,
 461                     uh->uh_sport, &ip6->ip6_dst, uh->uh_dport,
 462                     INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif, m);
 463                 if (!inp) {
 464                         /*
 465                          * It's new.  Try to find the ambushing socket.
 466                          * Because we've rewritten the destination address,
 467                          * any hardware-generated hash is ignored.
 468                          */
 469                         inp = in6_pcblookup(pcbinfo, &ip6->ip6_src,
 470                             uh->uh_sport, &next_hop6->sin6_addr,
 471                             next_hop6->sin6_port ? htons(next_hop6->sin6_port) :
 472                             uh->uh_dport, INPLOOKUP_WILDCARD |
 473                             INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif);
 474                 }
 475                 /* Remove the tag from the packet. We don't need it anymore. */
 476                 m_tag_delete(m, fwd_tag);
 477                 m->m_flags &= ~M_IP6_NEXTHOP;
 478         } else
 479                 inp = in6_pcblookup_mbuf(pcbinfo, &ip6->ip6_src,
 480                     uh->uh_sport, &ip6->ip6_dst, uh->uh_dport,
 481                     INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB,
 482                     m->m_pkthdr.rcvif, m);
 483         if (inp == NULL) {
 484                 if (udp_log_in_vain) {
 485                         char ip6bufs[INET6_ADDRSTRLEN];
 486                         char ip6bufd[INET6_ADDRSTRLEN];
 487
 488                         log(LOG_INFO,
 489                             "Connection attempt to UDP [%s]:%d from [%s]:%d\n",
 490                             ip6_sprintf(ip6bufd, &ip6->ip6_dst),
 491                             ntohs(uh->uh_dport),
 492                             ip6_sprintf(ip6bufs, &ip6->ip6_src),
 493                             ntohs(uh->uh_sport));
 494                 }
 495                 if (nxt == IPPROTO_UDPLITE)
 496                         UDPLITE_PROBE(receive, NULL, NULL, ip6, NULL, uh);
 497                 else
 498                         UDP_PROBE(receive, NULL, NULL, ip6, NULL, uh);
 499                 UDPSTAT_INC(udps_noport);
 500                 if (m->m_flags & M_MCAST) {
 501                         printf("UDP6: M_MCAST is set in a unicast packet.\n");
 502                         UDPSTAT_INC(udps_noportmcast);
 503                         goto badunlocked;
 504                 }
 505                 if (V_udp_blackhole)
 506                         goto badunlocked;
 507                 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_NOPORT, 0);
 508                 *mp = NULL;
 509                 return (IPPROTO_DONE);
 510         }
 511         INP_RLOCK_ASSERT(inp);
 512         up = intoudpcb(inp);
 513         if (cscov_partial) {
 514                 if (up->u_rxcslen == 0 || up->u_rxcslen > ulen) {
 515                         INP_RUNLOCK(inp);
 516                         m_freem(m);
 517                         *mp = NULL;
 518                         return (IPPROTO_DONE);
 519                 }
 520         }
 521         if (nxt == IPPROTO_UDPLITE)
 522                 UDPLITE_PROBE(receive, NULL, inp, ip6, inp, uh);
 523         else
 524                 UDP_PROBE(receive, NULL, inp, ip6, inp, uh);
 525         if (udp6_append(inp, m, off, fromsa) == 0)
 526                 INP_RUNLOCK(inp);
 527         *mp = NULL;
 528         return (IPPROTO_DONE);
 529
 530 badunlocked:
 531         if (m)
 532                 m_freem(m);
 533         *mp = NULL;
 534         return (IPPROTO_DONE);
 535 }
 536
 537 static void
 538 udp6_common_ctlinput(int cmd, struct sockaddr *sa, void *d,
 539     struct inpcbinfo *pcbinfo)
 540 {
 541         struct udphdr uh;
 542         struct ip6_hdr *ip6;
 543         struct mbuf *m;
 544         int off = 0;
 545         struct ip6ctlparam *ip6cp = NULL;
 546         const struct sockaddr_in6 *sa6_src = NULL;
 547         void *cmdarg;
 548         struct inpcb *(*notify)(struct inpcb *, int) = udp_notify;
 549         struct udp_portonly {
 550                 u_int16_t uh_sport;
 551                 u_int16_t uh_dport;
 552         } *uhp;
 553
 554         if (sa->sa_family != AF_INET6 ||
 555             sa->sa_len != sizeof(struct sockaddr_in6))
 556                 return;
 557
 558         if ((unsigned)cmd >= PRC_NCMDS)
 559                 return;
 560         if (PRC_IS_REDIRECT(cmd))
 561                 notify = in6_rtchange, d = NULL;
 562         else if (cmd == PRC_HOSTDEAD)
 563                 d = NULL;
 564         else if (inet6ctlerrmap[cmd] == 0)
 565                 return;
 566
 567         /* if the parameter is from icmp6, decode it. */
 568         if (d != NULL) {
 569                 ip6cp = (struct ip6ctlparam *)d;
 570                 m = ip6cp->ip6c_m;
 571                 ip6 = ip6cp->ip6c_ip6;
 572                 off = ip6cp->ip6c_off;
 573                 cmdarg = ip6cp->ip6c_cmdarg;
 574                 sa6_src = ip6cp->ip6c_src;
 575         } else {
 576                 m = NULL;
 577                 ip6 = NULL;
 578                 cmdarg = NULL;
 579                 sa6_src = &sa6_any;
 580         }
 581
 582         if (ip6) {
 583                 /*
 584                  * XXX: We assume that when IPV6 is non NULL,
 585                  * M and OFF are valid.
 586                  */
 587
 588                 /* Check if we can safely examine src and dst ports. */
 589                 if (m->m_pkthdr.len < off + sizeof(*uhp))
 590                         return;
 591
 592                 bzero(&uh, sizeof(uh));
 593                 m_copydata(m, off, sizeof(*uhp), (caddr_t)&uh);
 594
 595                 if (!PRC_IS_REDIRECT(cmd)) {
 596                         /* Check to see if its tunneled */
 597                         struct inpcb *inp;
 598                         inp = in6_pcblookup_mbuf(pcbinfo, &ip6->ip6_dst,
 599                             uh.uh_dport, &ip6->ip6_src, uh.uh_sport,
 600                             INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB,
 601                             m->m_pkthdr.rcvif, m);
 602                         if (inp != NULL) {
 603                                 struct udpcb *up;
 604
 605                                 up = intoudpcb(inp);
 606                                 if (up->u_icmp_func) {
 607                                         /* Yes it is. */
 608                                         INP_RUNLOCK(inp);
 609                                         (*up->u_icmp_func)(cmd, (struct sockaddr *)ip6cp->ip6c_src,
 610                                               d, up->u_tun_ctx);
 611                                         return;
 612                                 } else {
 613                                         /* Can't find it. */
 614                                         INP_RUNLOCK(inp);
 615                                 }
 616                         }
 617                 }
 618                 (void)in6_pcbnotify(pcbinfo, sa, uh.uh_dport,
 619                     (struct sockaddr *)ip6cp->ip6c_src, uh.uh_sport, cmd,
 620                     cmdarg, notify);
 621         } else
 622                 (void)in6_pcbnotify(pcbinfo, sa, 0,
 623                     (const struct sockaddr *)sa6_src, 0, cmd, cmdarg, notify);
 624 }
 625
 626 void
 627 udp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
 628 {
 629
 630         return (udp6_common_ctlinput(cmd, sa, d, &V_udbinfo));
 631 }
 632
 633 void
 634 udplite6_ctlinput(int cmd, struct sockaddr *sa, void *d)
 635 {
 636
 637         return (udp6_common_ctlinput(cmd, sa, d, &V_ulitecbinfo));
 638 }
 639
 640 static int
 641 udp6_getcred(SYSCTL_HANDLER_ARGS)
 642 {
 643         struct xucred xuc;
 644         struct sockaddr_in6 addrs[2];
 645         struct epoch_tracker et;
 646         struct inpcb *inp;
 647         int error;
 648
 649         error = priv_check(req->td, PRIV_NETINET_GETCRED);
 650         if (error)
 651                 return (error);
 652
 653         if (req->newlen != sizeof(addrs))
 654                 return (EINVAL);
 655         if (req->oldlen != sizeof(struct xucred))
 656                 return (EINVAL);
 657         error = SYSCTL_IN(req, addrs, sizeof(addrs));
 658         if (error)
 659                 return (error);
 660         if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
 661             (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
 662                 return (error);
 663         }
 664         NET_EPOCH_ENTER(et);
 665         inp = in6_pcblookup(&V_udbinfo, &addrs[1].sin6_addr,
 666             addrs[1].sin6_port, &addrs[0].sin6_addr, addrs[0].sin6_port,
 667             INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
 668         NET_EPOCH_EXIT(et);
 669         if (inp != NULL) {
 670                 INP_RLOCK_ASSERT(inp);
 671                 if (inp->inp_socket == NULL)
 672                         error = ENOENT;
 673                 if (error == 0)
 674                         error = cr_canseesocket(req->td->td_ucred,
 675                             inp->inp_socket);
 676                 if (error == 0)
 677                         cru2x(inp->inp_cred, &xuc);
 678                 INP_RUNLOCK(inp);
 679         } else
 680                 error = ENOENT;
 681         if (error == 0)
 682                 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
 683         return (error);
 684 }
 685
 686 SYSCTL_PROC(_net_inet6_udp6, OID_AUTO, getcred, CTLTYPE_OPAQUE|CTLFLAG_RW, 0,
 687     0, udp6_getcred, "S,xucred", "Get the xucred of a UDP6 connection");
 688
 689 static int
 690 udp6_output(struct socket *so, int flags_arg, struct mbuf *m,
 691     struct sockaddr *addr6, struct mbuf *control, struct thread *td)
 692 {
 693         struct inpcb *inp;
 694         struct ip6_hdr *ip6;
 695         struct udphdr *udp6;
 696         struct in6_addr *laddr, *faddr, in6a;
 697         struct ip6_pktopts *optp, opt;
 698         struct sockaddr_in6 *sin6, tmp;
 699         struct epoch_tracker et;
 700         int cscov_partial, error, flags, hlen, scope_ambiguous;
 701         u_int32_t ulen, plen;
 702         uint16_t cscov;
 703         u_short fport;
 704         uint8_t nxt;
 705
 706         /* addr6 has been validated in udp6_send(). */
 707         sin6 = (struct sockaddr_in6 *)addr6;
 708
 709         /*
 710          * In contrast to to IPv4 we do not validate the max. packet length
 711          * here due to IPv6 Jumbograms (RFC2675).
 712          */
 713
 714         scope_ambiguous = 0;
 715         if (sin6) {
 716                 /* Protect *addr6 from overwrites. */
 717                 tmp = *sin6;
 718                 sin6 = &tmp;
 719
 720                 /*
 721                  * Application should provide a proper zone ID or the use of
 722                  * default zone IDs should be enabled.  Unfortunately, some
 723                  * applications do not behave as it should, so we need a
 724                  * workaround.  Even if an appropriate ID is not determined,
 725                  * we'll see if we can determine the outgoing interface.  If we
 726                  * can, determine the zone ID based on the interface below.
 727                  */
 728                 if (sin6->sin6_scope_id == 0 && !V_ip6_use_defzone)
 729                         scope_ambiguous = 1;
 730                 if ((error = sa6_embedscope(sin6, V_ip6_use_defzone)) != 0) {
 731                         if (control)
 732                                 m_freem(control);
 733                         m_freem(m);
 734                         return (error);
 735                 }
 736         }
 737
 738         inp = sotoinpcb(so);
 739         KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
 740         /*
 741          * In the following cases we want a write lock on the inp for either
 742          * local operations or for possible route cache updates in the IPv6
 743          * output path:
 744          * - on connected sockets (sin6 is NULL) for route cache updates,
 745          * - when we are not bound to an address and source port (it is
 746          *   in6_pcbsetport() which will require the write lock).
 747          *
 748          * We check the inp fields before actually locking the inp, so
 749          * here exists a race, and we may WLOCK the inp and end with already
 750          * bound one by other thread. This is fine.
 751          */
 752         if (sin6 == NULL || (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) &&
 753             inp->inp_lport == 0))
 754                 INP_WLOCK(inp);
 755         else
 756                 INP_RLOCK(inp);
 757
 758         nxt = (inp->inp_socket->so_proto->pr_protocol == IPPROTO_UDP) ?
 759             IPPROTO_UDP : IPPROTO_UDPLITE;
 760
 761 #ifdef INET
 762         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
 763                 int hasv4addr;
 764
 765                 if (sin6 == NULL)
 766                         hasv4addr = (inp->inp_vflag & INP_IPV4);
 767                 else
 768                         hasv4addr = IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)
 769                             ? 1 : 0;
 770                 if (hasv4addr) {
 771                         struct pr_usrreqs *pru;
 772
 773                         /*
 774                          * XXXRW: We release UDP-layer locks before calling
 775                          * udp_send() in order to avoid recursion.  However,
 776                          * this does mean there is a short window where inp's
 777                          * fields are unstable.  Could this lead to a
 778                          * potential race in which the factors causing us to
 779                          * select the UDPv4 output routine are invalidated?
 780                          */
 781                         INP_UNLOCK(inp);
 782                         if (sin6)
 783                                 in6_sin6_2_sin_in_sock((struct sockaddr *)sin6);
 784                         pru = inetsw[ip_protox[nxt]].pr_usrreqs;
 785                         /* addr will just be freed in sendit(). */
 786                         return ((*pru->pru_send)(so, flags_arg, m,
 787                             (struct sockaddr *)sin6, control, td));
 788                 }
 789         } else
 790 #endif
 791         if (sin6 && IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
 792                 /*
 793                  * Given this is either an IPv6-only socket or no INET is
 794                  * supported we will fail the send if the given destination
 795                  * address is a v4mapped address.
 796                  *
 797                  * XXXGL: do we leak m and control?
 798                  */
 799                 INP_UNLOCK(inp);
 800                 return (EINVAL);
 801         }
 802
 803         if (control) {
 804                 if ((error = ip6_setpktopts(control, &opt,
 805                     inp->in6p_outputopts, td->td_ucred, nxt)) != 0) {
 806                         INP_UNLOCK(inp);
 807                         ip6_clearpktopts(&opt, -1);
 808                         if (control)
 809                                 m_freem(control);
 810                         m_freem(m);
 811                         return (error);
 812                 }
 813                 optp = &opt;
 814         } else
 815                 optp = inp->in6p_outputopts;
 816
 817         NET_EPOCH_ENTER(et);
 818         if (sin6) {
 819
 820                 /*
 821                  * Since we saw no essential reason for calling in_pcbconnect,
 822                  * we get rid of such kind of logic, and call in6_selectsrc
 823                  * and in6_pcbsetport in order to fill in the local address
 824                  * and the local port.
 825                  */
 826                 if (sin6->sin6_port == 0) {
 827                         error = EADDRNOTAVAIL;
 828                         goto release;
 829                 }
 830
 831                 if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
 832                         /* how about ::ffff:0.0.0.0 case? */
 833                         error = EISCONN;
 834                         goto release;
 835                 }
 836
 837                 /*
 838                  * Given we handle the v4mapped case in the INET block above
 839                  * assert here that it must not happen anymore.
 840                  */
 841                 KASSERT(!IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr),
 842                     ("%s: sin6(%p)->sin6_addr is v4mapped which we "
 843                     "should have handled.", __func__, sin6));
 844
 845                 /* This only requires read-locking. */
 846                 error = in6_selectsrc_socket(sin6, optp, inp,
 847                     td->td_ucred, scope_ambiguous, &in6a, NULL);
 848                 if (error)
 849                         goto release;
 850                 laddr = &in6a;
 851
 852                 if (inp->inp_lport == 0) {
 853                         struct inpcbinfo *pcbinfo;
 854
 855                         INP_WLOCK_ASSERT(inp);
 856
 857                         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
 858                         INP_HASH_WLOCK(pcbinfo);
 859                         error = in6_pcbsetport(laddr, inp, td->td_ucred);
 860                         INP_HASH_WUNLOCK(pcbinfo);
 861                         if (error != 0) {
 862                                 /* Undo an address bind that may have occurred. */
 863                                 inp->in6p_laddr = in6addr_any;
 864                                 goto release;
 865                         }
 866                 }
 867                 faddr = &sin6->sin6_addr;
 868                 fport = sin6->sin6_port; /* allow 0 port */
 869
 870         } else {
 871                 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
 872                         error = ENOTCONN;
 873                         goto release;
 874                 }
 875                 laddr = &inp->in6p_laddr;
 876                 faddr = &inp->in6p_faddr;
 877                 fport = inp->inp_fport;
 878         }
 879
 880         ulen = m->m_pkthdr.len;
 881         plen = sizeof(struct udphdr) + ulen;
 882         hlen = sizeof(struct ip6_hdr);
 883
 884         /*
 885          * Calculate data length and get a mbuf
 886          * for UDP and IP6 headers.
 887          */
 888         M_PREPEND(m, hlen + sizeof(struct udphdr), M_NOWAIT);
 889         if (m == NULL) {
 890                 error = ENOBUFS;
 891                 goto release;
 892         }
 893
 894         /*
 895          * Stuff checksum and output datagram.
 896          */
 897         cscov = cscov_partial = 0;
 898         udp6 = (struct udphdr *)(mtod(m, caddr_t) + hlen);
 899         udp6->uh_sport = inp->inp_lport; /* lport is always set in the PCB */
 900         udp6->uh_dport = fport;
 901         if (nxt == IPPROTO_UDPLITE) {
 902                 struct udpcb *up;
 903
 904                 up = intoudpcb(inp);
 905                 cscov = up->u_txcslen;
 906                 if (cscov >= plen)
 907                         cscov = 0;
 908                 udp6->uh_ulen = htons(cscov);
 909                 /*
 910                  * For UDP-Lite, checksum coverage length of zero means
 911                  * the entire UDPLite packet is covered by the checksum.
 912                  */
 913                 cscov_partial = (cscov == 0) ? 0 : 1;
 914         } else if (plen <= 0xffff)
 915                 udp6->uh_ulen = htons((u_short)plen);
 916         else
 917                 udp6->uh_ulen = 0;
 918         udp6->uh_sum = 0;
 919
 920         ip6 = mtod(m, struct ip6_hdr *);
 921         ip6->ip6_flow   = inp->inp_flow & IPV6_FLOWINFO_MASK;
 922         ip6->ip6_vfc    &= ~IPV6_VERSION_MASK;
 923         ip6->ip6_vfc    |= IPV6_VERSION;
 924         ip6->ip6_plen   = htons((u_short)plen);
 925         ip6->ip6_nxt    = nxt;
 926         ip6->ip6_hlim   = in6_selecthlim(inp, NULL);
 927         ip6->ip6_src    = *laddr;
 928         ip6->ip6_dst    = *faddr;
 929
 930 #ifdef MAC
 931         mac_inpcb_create_mbuf(inp, m);
 932 #endif
 933
 934         if (cscov_partial) {
 935                 if ((udp6->uh_sum = in6_cksum_partial(m, nxt,
 936                     sizeof(struct ip6_hdr), plen, cscov)) == 0)
 937                         udp6->uh_sum = 0xffff;
 938         } else {
 939                 udp6->uh_sum = in6_cksum_pseudo(ip6, plen, nxt, 0);
 940                 m->m_pkthdr.csum_flags = CSUM_UDP_IPV6;
 941                 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
 942         }
 943
 944         flags = 0;
 945 #ifdef  RSS
 946         {
 947                 uint32_t hash_val, hash_type;
 948                 uint8_t pr;
 949
 950                 pr = inp->inp_socket->so_proto->pr_protocol;
 951                 /*
 952                  * Calculate an appropriate RSS hash for UDP and
 953                  * UDP Lite.
 954                  *
 955                  * The called function will take care of figuring out
 956                  * whether a 2-tuple or 4-tuple hash is required based
 957                  * on the currently configured scheme.
 958                  *
 959                  * Later later on connected socket values should be
 960                  * cached in the inpcb and reused, rather than constantly
 961                  * re-calculating it.
 962                  *
 963                  * UDP Lite is a different protocol number and will
 964                  * likely end up being hashed as a 2-tuple until
 965                  * RSS / NICs grow UDP Lite protocol awareness.
 966                  */
 967                 if (rss_proto_software_hash_v6(faddr, laddr, fport,
 968                     inp->inp_lport, pr, &hash_val, &hash_type) == 0) {
 969                         m->m_pkthdr.flowid = hash_val;
 970                         M_HASHTYPE_SET(m, hash_type);
 971                 }
 972
 973                 /*
 974                  * Don't override with the inp cached flowid.
 975                  *
 976                  * Until the whole UDP path is vetted, it may actually
 977                  * be incorrect.
 978                  */
 979                 flags |= IP_NODEFAULTFLOWID;
 980         }
 981 #endif
 982
 983         UDPSTAT_INC(udps_opackets);
 984         if (nxt == IPPROTO_UDPLITE)
 985                 UDPLITE_PROBE(send, NULL, inp, ip6, inp, udp6);
 986         else
 987                 UDP_PROBE(send, NULL, inp, ip6, inp, udp6);
 988         error = ip6_output(m, optp,
 989             INP_WLOCKED(inp) ? &inp->inp_route6 : NULL, flags,
 990             inp->in6p_moptions, NULL, inp);
 991         INP_UNLOCK(inp);
 992         NET_EPOCH_EXIT(et);
 993
 994         if (control) {
 995                 ip6_clearpktopts(&opt, -1);
 996                 m_freem(control);
 997         }
 998         return (error);
 999
1000 release:
1001         INP_UNLOCK(inp);
1002         NET_EPOCH_EXIT(et);
1003         if (control) {
1004                 ip6_clearpktopts(&opt, -1);
1005                 m_freem(control);
1006         }
1007         m_freem(m);
1008
1009         return (error);
1010 }
1011
1012 static void
1013 udp6_abort(struct socket *so)
1014 {
1015         struct inpcb *inp;
1016         struct inpcbinfo *pcbinfo;
1017
1018         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1019         inp = sotoinpcb(so);
1020         KASSERT(inp != NULL, ("udp6_abort: inp == NULL"));
1021
1022         INP_WLOCK(inp);
1023 #ifdef INET
1024         if (inp->inp_vflag & INP_IPV4) {
1025                 struct pr_usrreqs *pru;
1026                 uint8_t nxt;
1027
1028                 nxt = (inp->inp_socket->so_proto->pr_protocol == IPPROTO_UDP) ?
1029                     IPPROTO_UDP : IPPROTO_UDPLITE;
1030                 INP_WUNLOCK(inp);
1031                 pru = inetsw[ip_protox[nxt]].pr_usrreqs;
1032                 (*pru->pru_abort)(so);
1033                 return;
1034         }
1035 #endif
1036
1037         if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
1038                 INP_HASH_WLOCK(pcbinfo);
1039                 in6_pcbdisconnect(inp);
1040                 inp->in6p_laddr = in6addr_any;
1041                 INP_HASH_WUNLOCK(pcbinfo);
1042                 soisdisconnected(so);
1043         }
1044         INP_WUNLOCK(inp);
1045 }
1046
1047 static int
1048 udp6_attach(struct socket *so, int proto, struct thread *td)
1049 {
1050         struct inpcb *inp;
1051         struct inpcbinfo *pcbinfo;
1052         int error;
1053
1054         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1055         inp = sotoinpcb(so);
1056         KASSERT(inp == NULL, ("udp6_attach: inp != NULL"));
1057
1058         if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
1059                 error = soreserve(so, udp_sendspace, udp_recvspace);
1060                 if (error)
1061                         return (error);
1062         }
1063         INP_INFO_WLOCK(pcbinfo);
1064         error = in_pcballoc(so, pcbinfo);
1065         if (error) {
1066                 INP_INFO_WUNLOCK(pcbinfo);
1067                 return (error);
1068         }
1069         inp = (struct inpcb *)so->so_pcb;
1070         inp->inp_vflag |= INP_IPV6;
1071         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
1072                 inp->inp_vflag |= INP_IPV4;
1073         inp->in6p_hops = -1;    /* use kernel default */
1074         inp->in6p_cksum = -1;   /* just to be sure */
1075         /*
1076          * XXX: ugly!!
1077          * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1078          * because the socket may be bound to an IPv6 wildcard address,
1079          * which may match an IPv4-mapped IPv6 address.
1080          */
1081         inp->inp_ip_ttl = V_ip_defttl;
1082
1083         error = udp_newudpcb(inp);
1084         if (error) {
1085                 in_pcbdetach(inp);
1086                 in_pcbfree(inp);
1087                 INP_INFO_WUNLOCK(pcbinfo);
1088                 return (error);
1089         }
1090         INP_WUNLOCK(inp);
1091         INP_INFO_WUNLOCK(pcbinfo);
1092         return (0);
1093 }
1094
1095 static int
1096 udp6_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1097 {
1098         struct inpcb *inp;
1099         struct inpcbinfo *pcbinfo;
1100         int error;
1101         u_char vflagsav;
1102
1103         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1104         inp = sotoinpcb(so);
1105         KASSERT(inp != NULL, ("udp6_bind: inp == NULL"));
1106
1107         INP_WLOCK(inp);
1108         INP_HASH_WLOCK(pcbinfo);
1109         vflagsav = inp->inp_vflag;
1110         inp->inp_vflag &= ~INP_IPV4;
1111         inp->inp_vflag |= INP_IPV6;
1112         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
1113                 struct sockaddr_in6 *sin6_p;
1114
1115                 sin6_p = (struct sockaddr_in6 *)nam;
1116
1117                 if (IN6_IS_ADDR_UNSPECIFIED(&sin6_p->sin6_addr))
1118                         inp->inp_vflag |= INP_IPV4;
1119 #ifdef INET
1120                 else if (IN6_IS_ADDR_V4MAPPED(&sin6_p->sin6_addr)) {
1121                         struct sockaddr_in sin;
1122
1123                         in6_sin6_2_sin(&sin, sin6_p);
1124                         inp->inp_vflag |= INP_IPV4;
1125                         inp->inp_vflag &= ~INP_IPV6;
1126                         error = in_pcbbind(inp, (struct sockaddr *)&sin,
1127                             td->td_ucred);
1128                         goto out;
1129                 }
1130 #endif
1131         }
1132
1133         error = in6_pcbbind(inp, nam, td->td_ucred);
1134 #ifdef INET
1135 out:
1136 #endif
1137         if (error != 0)
1138                 inp->inp_vflag = vflagsav;
1139         INP_HASH_WUNLOCK(pcbinfo);
1140         INP_WUNLOCK(inp);
1141         return (error);
1142 }
1143
1144 static void
1145 udp6_close(struct socket *so)
1146 {
1147         struct inpcb *inp;
1148         struct inpcbinfo *pcbinfo;
1149
1150         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1151         inp = sotoinpcb(so);
1152         KASSERT(inp != NULL, ("udp6_close: inp == NULL"));
1153
1154         INP_WLOCK(inp);
1155 #ifdef INET
1156         if (inp->inp_vflag & INP_IPV4) {
1157                 struct pr_usrreqs *pru;
1158                 uint8_t nxt;
1159
1160                 nxt = (inp->inp_socket->so_proto->pr_protocol == IPPROTO_UDP) ?
1161                     IPPROTO_UDP : IPPROTO_UDPLITE;
1162                 INP_WUNLOCK(inp);
1163                 pru = inetsw[ip_protox[nxt]].pr_usrreqs;
1164                 (*pru->pru_disconnect)(so);
1165                 return;
1166         }
1167 #endif
1168         if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
1169                 INP_HASH_WLOCK(pcbinfo);
1170                 in6_pcbdisconnect(inp);
1171                 inp->in6p_laddr = in6addr_any;
1172                 INP_HASH_WUNLOCK(pcbinfo);
1173                 soisdisconnected(so);
1174         }
1175         INP_WUNLOCK(inp);
1176 }
1177
1178 static int
1179 udp6_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1180 {
1181         struct inpcb *inp;
1182         struct inpcbinfo *pcbinfo;
1183         struct sockaddr_in6 *sin6;
1184         int error;
1185         u_char vflagsav;
1186
1187         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1188         inp = sotoinpcb(so);
1189         sin6 = (struct sockaddr_in6 *)nam;
1190         KASSERT(inp != NULL, ("udp6_connect: inp == NULL"));
1191
1192         /*
1193          * XXXRW: Need to clarify locking of v4/v6 flags.
1194          */
1195         INP_WLOCK(inp);
1196 #ifdef INET
1197         if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
1198                 struct sockaddr_in sin;
1199
1200                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
1201                         error = EINVAL;
1202                         goto out;
1203                 }
1204                 if ((inp->inp_vflag & INP_IPV4) == 0) {
1205                         error = EAFNOSUPPORT;
1206                         goto out;
1207                 }
1208                 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1209                         error = EISCONN;
1210                         goto out;
1211                 }
1212                 in6_sin6_2_sin(&sin, sin6);
1213                 error = prison_remote_ip4(td->td_ucred, &sin.sin_addr);
1214                 if (error != 0)
1215                         goto out;
1216                 vflagsav = inp->inp_vflag;
1217                 inp->inp_vflag |= INP_IPV4;
1218                 inp->inp_vflag &= ~INP_IPV6;
1219                 INP_HASH_WLOCK(pcbinfo);
1220                 error = in_pcbconnect(inp, (struct sockaddr *)&sin,
1221                     td->td_ucred);
1222                 INP_HASH_WUNLOCK(pcbinfo);
1223                 /*
1224                  * If connect succeeds, mark socket as connected. If
1225                  * connect fails and socket is unbound, reset inp_vflag
1226                  * field.
1227                  */
1228                 if (error == 0)
1229                         soisconnected(so);
1230                 else if (inp->inp_laddr.s_addr == INADDR_ANY &&
1231                     inp->inp_lport == 0)
1232                         inp->inp_vflag = vflagsav;
1233                 goto out;
1234         } else {
1235                 if ((inp->inp_vflag & INP_IPV6) == 0) {
1236                         error = EAFNOSUPPORT;
1237                         goto out;
1238                 }
1239         }
1240 #endif
1241         if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
1242                 error = EISCONN;
1243                 goto out;
1244         }
1245         error = prison_remote_ip6(td->td_ucred, &sin6->sin6_addr);
1246         if (error != 0)
1247                 goto out;
1248         vflagsav = inp->inp_vflag;
1249         inp->inp_vflag &= ~INP_IPV4;
1250         inp->inp_vflag |= INP_IPV6;
1251         INP_HASH_WLOCK(pcbinfo);
1252         error = in6_pcbconnect(inp, nam, td->td_ucred);
1253         INP_HASH_WUNLOCK(pcbinfo);
1254         /*
1255          * If connect succeeds, mark socket as connected. If
1256          * connect fails and socket is unbound, reset inp_vflag
1257          * field.
1258          */
1259         if (error == 0)
1260                 soisconnected(so);
1261         else if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) &&
1262             inp->inp_lport == 0)
1263                 inp->inp_vflag = vflagsav;
1264 out:
1265         INP_WUNLOCK(inp);
1266         return (error);
1267 }
1268
1269 static void
1270 udp6_detach(struct socket *so)
1271 {
1272         struct inpcb *inp;
1273         struct inpcbinfo *pcbinfo;
1274         struct udpcb *up;
1275
1276         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1277         inp = sotoinpcb(so);
1278         KASSERT(inp != NULL, ("udp6_detach: inp == NULL"));
1279
1280         INP_INFO_WLOCK(pcbinfo);
1281         INP_WLOCK(inp);
1282         up = intoudpcb(inp);
1283         KASSERT(up != NULL, ("%s: up == NULL", __func__));
1284         in_pcbdetach(inp);
1285         in_pcbfree(inp);
1286         INP_INFO_WUNLOCK(pcbinfo);
1287         udp_discardcb(up);
1288 }
1289
1290 static int
1291 udp6_disconnect(struct socket *so)
1292 {
1293         struct inpcb *inp;
1294         struct inpcbinfo *pcbinfo;
1295
1296         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1297         inp = sotoinpcb(so);
1298         KASSERT(inp != NULL, ("udp6_disconnect: inp == NULL"));
1299
1300         INP_WLOCK(inp);
1301 #ifdef INET
1302         if (inp->inp_vflag & INP_IPV4) {
1303                 struct pr_usrreqs *pru;
1304                 uint8_t nxt;
1305
1306                 nxt = (inp->inp_socket->so_proto->pr_protocol == IPPROTO_UDP) ?
1307                     IPPROTO_UDP : IPPROTO_UDPLITE;
1308                 INP_WUNLOCK(inp);
1309                 pru = inetsw[ip_protox[nxt]].pr_usrreqs;
1310                 (void)(*pru->pru_disconnect)(so);
1311                 return (0);
1312         }
1313 #endif
1314
1315         if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
1316                 INP_WUNLOCK(inp);
1317                 return (ENOTCONN);
1318         }
1319
1320         INP_HASH_WLOCK(pcbinfo);
1321         in6_pcbdisconnect(inp);
1322         inp->in6p_laddr = in6addr_any;
1323         INP_HASH_WUNLOCK(pcbinfo);
1324         SOCK_LOCK(so);
1325         so->so_state &= ~SS_ISCONNECTED;                /* XXX */
1326         SOCK_UNLOCK(so);
1327         INP_WUNLOCK(inp);
1328         return (0);
1329 }
1330
1331 static int
1332 udp6_send(struct socket *so, int flags, struct mbuf *m,
1333     struct sockaddr *addr, struct mbuf *control, struct thread *td)
1334 {
1335         int error;
1336
1337         if (addr) {
1338                 if (addr->sa_len != sizeof(struct sockaddr_in6)) {
1339                         error = EINVAL;
1340                         goto bad;
1341                 }
1342                 if (addr->sa_family != AF_INET6) {
1343                         error = EAFNOSUPPORT;
1344                         goto bad;
1345                 }
1346         }
1347
1348         return (udp6_output(so, flags, m, addr, control, td));
1349
1350 bad:
1351         if (control)
1352                 m_freem(control);
1353         m_freem(m);
1354         return (error);
1355 }
1356
1357 struct pr_usrreqs udp6_usrreqs = {
1358         .pru_abort =            udp6_abort,
1359         .pru_attach =           udp6_attach,
1360         .pru_bind =             udp6_bind,
1361         .pru_connect =          udp6_connect,
1362         .pru_control =          in6_control,
1363         .pru_detach =           udp6_detach,
1364         .pru_disconnect =       udp6_disconnect,
1365         .pru_peeraddr =         in6_mapped_peeraddr,
1366         .pru_send =             udp6_send,
1367         .pru_shutdown =         udp_shutdown,
1368         .pru_sockaddr =         in6_mapped_sockaddr,
1369         .pru_soreceive =        soreceive_dgram,
1370         .pru_sosend =           sosend_dgram,
1371         .pru_sosetlabel =       in_pcbsosetlabel,
1372         .pru_close =            udp6_close
1373 };