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         m = m_pullup(m, off + sizeof(struct udphdr));
 227         if (m == NULL) {
 228                 IP6STAT_INC(ip6s_exthdrtoolong);
 229                 *mp = NULL;
 230                 return (IPPROTO_DONE);
 231         }
 232         ip6 = mtod(m, struct ip6_hdr *);
 233         uh = (struct udphdr *)((caddr_t)ip6 + off);
 234
 235         UDPSTAT_INC(udps_ipackets);
 236
 237         /*
 238          * Destination port of 0 is illegal, based on RFC768.
 239          */
 240         if (uh->uh_dport == 0)
 241                 goto badunlocked;
 242
 243         plen = ntohs(ip6->ip6_plen) - off + sizeof(*ip6);
 244         ulen = ntohs((u_short)uh->uh_ulen);
 245
 246         nxt = proto;
 247         cscov_partial = (nxt == IPPROTO_UDPLITE) ? 1 : 0;
 248         if (nxt == IPPROTO_UDPLITE) {
 249                 /* Zero means checksum over the complete packet. */
 250                 if (ulen == 0)
 251                         ulen = plen;
 252                 if (ulen == plen)
 253                         cscov_partial = 0;
 254                 if ((ulen < sizeof(struct udphdr)) || (ulen > plen)) {
 255                         /* XXX: What is the right UDPLite MIB counter? */
 256                         goto badunlocked;
 257                 }
 258                 if (uh->uh_sum == 0) {
 259                         /* XXX: What is the right UDPLite MIB counter? */
 260                         goto badunlocked;
 261                 }
 262         } else {
 263                 if ((ulen < sizeof(struct udphdr)) || (plen != ulen)) {
 264                         UDPSTAT_INC(udps_badlen);
 265                         goto badunlocked;
 266                 }
 267                 if (uh->uh_sum == 0) {
 268                         UDPSTAT_INC(udps_nosum);
 269                         goto badunlocked;
 270                 }
 271         }
 272
 273         if ((m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) &&
 274             !cscov_partial) {
 275                 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
 276                         uh_sum = m->m_pkthdr.csum_data;
 277                 else
 278                         uh_sum = in6_cksum_pseudo(ip6, ulen, nxt,
 279                             m->m_pkthdr.csum_data);
 280                 uh_sum ^= 0xffff;
 281         } else
 282                 uh_sum = in6_cksum_partial(m, nxt, off, plen, ulen);
 283
 284         if (uh_sum != 0) {
 285                 UDPSTAT_INC(udps_badsum);
 286                 goto badunlocked;
 287         }
 288
 289         /*
 290          * Construct sockaddr format source address.
 291          */
 292         init_sin6(&fromsa[0], m, 0);
 293         fromsa[0].sin6_port = uh->uh_sport;
 294         init_sin6(&fromsa[1], m, 1);
 295         fromsa[1].sin6_port = uh->uh_dport;
 296
 297         pcbinfo = udp_get_inpcbinfo(nxt);
 298         if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) {
 299                 struct inpcb *last;
 300                 struct inpcbhead *pcblist;
 301                 struct ip6_moptions *imo;
 302
 303                 /*
 304                  * In the event that laddr should be set to the link-local
 305                  * address (this happens in RIPng), the multicast address
 306                  * specified in the received packet will not match laddr.  To
 307                  * handle this situation, matching is relaxed if the
 308                  * receiving interface is the same as one specified in the
 309                  * socket and if the destination multicast address matches
 310                  * one of the multicast groups specified in the socket.
 311                  */
 312
 313                 /*
 314                  * KAME note: traditionally we dropped udpiphdr from mbuf
 315                  * here.  We need udphdr for IPsec processing so we do that
 316                  * later.
 317                  */
 318                 pcblist = udp_get_pcblist(nxt);
 319                 last = NULL;
 320                 CK_LIST_FOREACH(inp, pcblist, inp_list) {
 321                         if ((inp->inp_vflag & INP_IPV6) == 0)
 322                                 continue;
 323                         if (inp->inp_lport != uh->uh_dport)
 324                                 continue;
 325                         if (inp->inp_fport != 0 &&
 326                             inp->inp_fport != uh->uh_sport)
 327                                 continue;
 328                         if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) {
 329                                 if (!IN6_ARE_ADDR_EQUAL(&inp->in6p_laddr,
 330                                                         &ip6->ip6_dst))
 331                                         continue;
 332                         }
 333                         if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
 334                                 if (!IN6_ARE_ADDR_EQUAL(&inp->in6p_faddr,
 335                                                         &ip6->ip6_src) ||
 336                                     inp->inp_fport != uh->uh_sport)
 337                                         continue;
 338                         }
 339
 340                         /*
 341                          * XXXRW: Because we weren't holding either the inpcb
 342                          * or the hash lock when we checked for a match
 343                          * before, we should probably recheck now that the
 344                          * inpcb lock is (supposed to be) held.
 345                          */
 346
 347                         /*
 348                          * Handle socket delivery policy for any-source
 349                          * and source-specific multicast. [RFC3678]
 350                          */
 351                         imo = inp->in6p_moptions;
 352                         if (imo && IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) {
 353                                 struct sockaddr_in6      mcaddr;
 354                                 int                      blocked;
 355
 356                                 INP_RLOCK(inp);
 357                                 if (__predict_false(inp->inp_flags2 & INP_FREED)) {
 358                                         INP_RUNLOCK(inp);
 359                                         continue;
 360                                 }
 361
 362                                 bzero(&mcaddr, sizeof(struct sockaddr_in6));
 363                                 mcaddr.sin6_len = sizeof(struct sockaddr_in6);
 364                                 mcaddr.sin6_family = AF_INET6;
 365                                 mcaddr.sin6_addr = ip6->ip6_dst;
 366
 367                                 blocked = im6o_mc_filter(imo, ifp,
 368                                         (struct sockaddr *)&mcaddr,
 369                                         (struct sockaddr *)&fromsa[0]);
 370                                 if (blocked != MCAST_PASS) {
 371                                         if (blocked == MCAST_NOTGMEMBER)
 372                                                 IP6STAT_INC(ip6s_notmember);
 373                                         if (blocked == MCAST_NOTSMEMBER ||
 374                                             blocked == MCAST_MUTED)
 375                                                 UDPSTAT_INC(udps_filtermcast);
 376                                         INP_RUNLOCK(inp); /* XXX */
 377                                         continue;
 378                                 }
 379
 380                                 INP_RUNLOCK(inp);
 381                         }
 382                         if (last != NULL) {
 383                                 struct mbuf *n;
 384
 385                                 if ((n = m_copym(m, 0, M_COPYALL, M_NOWAIT)) !=
 386                                     NULL) {
 387                                         INP_RLOCK(last);
 388                                         if (__predict_true(last->inp_flags2 & INP_FREED) == 0) {
 389                                                 if (nxt == IPPROTO_UDPLITE)
 390                                                         UDPLITE_PROBE(receive, NULL, last,
 391                                                             ip6, last, uh);
 392                                                 else
 393                                                         UDP_PROBE(receive, NULL, last,
 394                                                             ip6, last, uh);
 395                                                 if (udp6_append(last, n, off, fromsa)) {
 396                                                         /* XXX-BZ do we leak m here? */
 397                                                         *mp = NULL;
 398                                                         return (IPPROTO_DONE);
 399                                                 }
 400                                         }
 401                                         INP_RUNLOCK(last);
 402                                 }
 403                         }
 404                         last = inp;
 405                         /*
 406                          * Don't look for additional matches if this one does
 407                          * not have either the SO_REUSEPORT or SO_REUSEADDR
 408                          * socket options set.  This heuristic avoids
 409                          * searching through all pcbs in the common case of a
 410                          * non-shared port.  It assumes that an application
 411                          * will never clear these options after setting them.
 412                          */
 413                         if ((last->inp_socket->so_options &
 414                              (SO_REUSEPORT|SO_REUSEPORT_LB|SO_REUSEADDR)) == 0)
 415                                 break;
 416                 }
 417
 418                 if (last == NULL) {
 419                         /*
 420                          * No matching pcb found; discard datagram.  (No need
 421                          * to send an ICMP Port Unreachable for a broadcast
 422                          * or multicast datgram.)
 423                          */
 424                         UDPSTAT_INC(udps_noport);
 425                         UDPSTAT_INC(udps_noportmcast);
 426                         goto badunlocked;
 427                 }
 428                 INP_RLOCK(last);
 429                 if (__predict_true(last->inp_flags2 & INP_FREED) == 0) {
 430                         if (nxt == IPPROTO_UDPLITE)
 431                                 UDPLITE_PROBE(receive, NULL, last, ip6, last, uh);
 432                         else
 433                                 UDP_PROBE(receive, NULL, last, ip6, last, uh);
 434                         if (udp6_append(last, m, off, fromsa) == 0)
 435                                 INP_RUNLOCK(last);
 436                 } else
 437                         INP_RUNLOCK(last);
 438                 *mp = NULL;
 439                 return (IPPROTO_DONE);
 440         }
 441         /*
 442          * Locate pcb for datagram.
 443          */
 444
 445         /*
 446          * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
 447          */
 448         if ((m->m_flags & M_IP6_NEXTHOP) &&
 449             (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) {
 450                 struct sockaddr_in6 *next_hop6;
 451
 452                 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
 453
 454                 /*
 455                  * Transparently forwarded. Pretend to be the destination.
 456                  * Already got one like this?
 457                  */
 458                 inp = in6_pcblookup_mbuf(pcbinfo, &ip6->ip6_src,
 459                     uh->uh_sport, &ip6->ip6_dst, uh->uh_dport,
 460                     INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif, m);
 461                 if (!inp) {
 462                         /*
 463                          * It's new.  Try to find the ambushing socket.
 464                          * Because we've rewritten the destination address,
 465                          * any hardware-generated hash is ignored.
 466                          */
 467                         inp = in6_pcblookup(pcbinfo, &ip6->ip6_src,
 468                             uh->uh_sport, &next_hop6->sin6_addr,
 469                             next_hop6->sin6_port ? htons(next_hop6->sin6_port) :
 470                             uh->uh_dport, INPLOOKUP_WILDCARD |
 471                             INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif);
 472                 }
 473                 /* Remove the tag from the packet. We don't need it anymore. */
 474                 m_tag_delete(m, fwd_tag);
 475                 m->m_flags &= ~M_IP6_NEXTHOP;
 476         } else
 477                 inp = in6_pcblookup_mbuf(pcbinfo, &ip6->ip6_src,
 478                     uh->uh_sport, &ip6->ip6_dst, uh->uh_dport,
 479                     INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB,
 480                     m->m_pkthdr.rcvif, m);
 481         if (inp == NULL) {
 482                 if (udp_log_in_vain) {
 483                         char ip6bufs[INET6_ADDRSTRLEN];
 484                         char ip6bufd[INET6_ADDRSTRLEN];
 485
 486                         log(LOG_INFO,
 487                             "Connection attempt to UDP [%s]:%d from [%s]:%d\n",
 488                             ip6_sprintf(ip6bufd, &ip6->ip6_dst),
 489                             ntohs(uh->uh_dport),
 490                             ip6_sprintf(ip6bufs, &ip6->ip6_src),
 491                             ntohs(uh->uh_sport));
 492                 }
 493                 if (nxt == IPPROTO_UDPLITE)
 494                         UDPLITE_PROBE(receive, NULL, NULL, ip6, NULL, uh);
 495                 else
 496                         UDP_PROBE(receive, NULL, NULL, ip6, NULL, uh);
 497                 UDPSTAT_INC(udps_noport);
 498                 if (m->m_flags & M_MCAST) {
 499                         printf("UDP6: M_MCAST is set in a unicast packet.\n");
 500                         UDPSTAT_INC(udps_noportmcast);
 501                         goto badunlocked;
 502                 }
 503                 if (V_udp_blackhole)
 504                         goto badunlocked;
 505                 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_NOPORT, 0);
 506                 *mp = NULL;
 507                 return (IPPROTO_DONE);
 508         }
 509         INP_RLOCK_ASSERT(inp);
 510         up = intoudpcb(inp);
 511         if (cscov_partial) {
 512                 if (up->u_rxcslen == 0 || up->u_rxcslen > ulen) {
 513                         INP_RUNLOCK(inp);
 514                         m_freem(m);
 515                         *mp = NULL;
 516                         return (IPPROTO_DONE);
 517                 }
 518         }
 519         if (nxt == IPPROTO_UDPLITE)
 520                 UDPLITE_PROBE(receive, NULL, inp, ip6, inp, uh);
 521         else
 522                 UDP_PROBE(receive, NULL, inp, ip6, inp, uh);
 523         if (udp6_append(inp, m, off, fromsa) == 0)
 524                 INP_RUNLOCK(inp);
 525         *mp = NULL;
 526         return (IPPROTO_DONE);
 527
 528 badunlocked:
 529         if (m)
 530                 m_freem(m);
 531         *mp = NULL;
 532         return (IPPROTO_DONE);
 533 }
 534
 535 static void
 536 udp6_common_ctlinput(int cmd, struct sockaddr *sa, void *d,
 537     struct inpcbinfo *pcbinfo)
 538 {
 539         struct udphdr uh;
 540         struct ip6_hdr *ip6;
 541         struct mbuf *m;
 542         int off = 0;
 543         struct ip6ctlparam *ip6cp = NULL;
 544         const struct sockaddr_in6 *sa6_src = NULL;
 545         void *cmdarg;
 546         struct inpcb *(*notify)(struct inpcb *, int) = udp_notify;
 547         struct udp_portonly {
 548                 u_int16_t uh_sport;
 549                 u_int16_t uh_dport;
 550         } *uhp;
 551
 552         if (sa->sa_family != AF_INET6 ||
 553             sa->sa_len != sizeof(struct sockaddr_in6))
 554                 return;
 555
 556         if ((unsigned)cmd >= PRC_NCMDS)
 557                 return;
 558         if (PRC_IS_REDIRECT(cmd))
 559                 notify = in6_rtchange, d = NULL;
 560         else if (cmd == PRC_HOSTDEAD)
 561                 d = NULL;
 562         else if (inet6ctlerrmap[cmd] == 0)
 563                 return;
 564
 565         /* if the parameter is from icmp6, decode it. */
 566         if (d != NULL) {
 567                 ip6cp = (struct ip6ctlparam *)d;
 568                 m = ip6cp->ip6c_m;
 569                 ip6 = ip6cp->ip6c_ip6;
 570                 off = ip6cp->ip6c_off;
 571                 cmdarg = ip6cp->ip6c_cmdarg;
 572                 sa6_src = ip6cp->ip6c_src;
 573         } else {
 574                 m = NULL;
 575                 ip6 = NULL;
 576                 cmdarg = NULL;
 577                 sa6_src = &sa6_any;
 578         }
 579
 580         if (ip6) {
 581                 /*
 582                  * XXX: We assume that when IPV6 is non NULL,
 583                  * M and OFF are valid.
 584                  */
 585
 586                 /* Check if we can safely examine src and dst ports. */
 587                 if (m->m_pkthdr.len < off + sizeof(*uhp))
 588                         return;
 589
 590                 bzero(&uh, sizeof(uh));
 591                 m_copydata(m, off, sizeof(*uhp), (caddr_t)&uh);
 592
 593                 if (!PRC_IS_REDIRECT(cmd)) {
 594                         /* Check to see if its tunneled */
 595                         struct inpcb *inp;
 596                         inp = in6_pcblookup_mbuf(pcbinfo, &ip6->ip6_dst,
 597                             uh.uh_dport, &ip6->ip6_src, uh.uh_sport,
 598                             INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB,
 599                             m->m_pkthdr.rcvif, m);
 600                         if (inp != NULL) {
 601                                 struct udpcb *up;
 602
 603                                 up = intoudpcb(inp);
 604                                 if (up->u_icmp_func) {
 605                                         /* Yes it is. */
 606                                         INP_RUNLOCK(inp);
 607                                         (*up->u_icmp_func)(cmd, (struct sockaddr *)ip6cp->ip6c_src,
 608                                               d, up->u_tun_ctx);
 609                                         return;
 610                                 } else {
 611                                         /* Can't find it. */
 612                                         INP_RUNLOCK(inp);
 613                                 }
 614                         }
 615                 }
 616                 (void)in6_pcbnotify(pcbinfo, sa, uh.uh_dport,
 617                     (struct sockaddr *)ip6cp->ip6c_src, uh.uh_sport, cmd,
 618                     cmdarg, notify);
 619         } else
 620                 (void)in6_pcbnotify(pcbinfo, sa, 0,
 621                     (const struct sockaddr *)sa6_src, 0, cmd, cmdarg, notify);
 622 }
 623
 624 void
 625 udp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
 626 {
 627
 628         return (udp6_common_ctlinput(cmd, sa, d, &V_udbinfo));
 629 }
 630
 631 void
 632 udplite6_ctlinput(int cmd, struct sockaddr *sa, void *d)
 633 {
 634
 635         return (udp6_common_ctlinput(cmd, sa, d, &V_ulitecbinfo));
 636 }
 637
 638 static int
 639 udp6_getcred(SYSCTL_HANDLER_ARGS)
 640 {
 641         struct xucred xuc;
 642         struct sockaddr_in6 addrs[2];
 643         struct epoch_tracker et;
 644         struct inpcb *inp;
 645         int error;
 646
 647         error = priv_check(req->td, PRIV_NETINET_GETCRED);
 648         if (error)
 649                 return (error);
 650
 651         if (req->newlen != sizeof(addrs))
 652                 return (EINVAL);
 653         if (req->oldlen != sizeof(struct xucred))
 654                 return (EINVAL);
 655         error = SYSCTL_IN(req, addrs, sizeof(addrs));
 656         if (error)
 657                 return (error);
 658         if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
 659             (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
 660                 return (error);
 661         }
 662         NET_EPOCH_ENTER(et);
 663         inp = in6_pcblookup(&V_udbinfo, &addrs[1].sin6_addr,
 664             addrs[1].sin6_port, &addrs[0].sin6_addr, addrs[0].sin6_port,
 665             INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
 666         NET_EPOCH_EXIT(et);
 667         if (inp != NULL) {
 668                 INP_RLOCK_ASSERT(inp);
 669                 if (inp->inp_socket == NULL)
 670                         error = ENOENT;
 671                 if (error == 0)
 672                         error = cr_canseesocket(req->td->td_ucred,
 673                             inp->inp_socket);
 674                 if (error == 0)
 675                         cru2x(inp->inp_cred, &xuc);
 676                 INP_RUNLOCK(inp);
 677         } else
 678                 error = ENOENT;
 679         if (error == 0)
 680                 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
 681         return (error);
 682 }
 683
 684 SYSCTL_PROC(_net_inet6_udp6, OID_AUTO, getcred, CTLTYPE_OPAQUE|CTLFLAG_RW, 0,
 685     0, udp6_getcred, "S,xucred", "Get the xucred of a UDP6 connection");
 686
 687 static int
 688 udp6_output(struct socket *so, int flags_arg, struct mbuf *m,
 689     struct sockaddr *addr6, struct mbuf *control, struct thread *td)
 690 {
 691         struct inpcb *inp;
 692         struct ip6_hdr *ip6;
 693         struct udphdr *udp6;
 694         struct in6_addr *laddr, *faddr, in6a;
 695         struct ip6_pktopts *optp, opt;
 696         struct sockaddr_in6 *sin6, tmp;
 697         struct epoch_tracker et;
 698         int cscov_partial, error, flags, hlen, scope_ambiguous;
 699         u_int32_t ulen, plen;
 700         uint16_t cscov;
 701         u_short fport;
 702         uint8_t nxt;
 703
 704         /* addr6 has been validated in udp6_send(). */
 705         sin6 = (struct sockaddr_in6 *)addr6;
 706
 707         /*
 708          * In contrast to to IPv4 we do not validate the max. packet length
 709          * here due to IPv6 Jumbograms (RFC2675).
 710          */
 711
 712         scope_ambiguous = 0;
 713         if (sin6) {
 714                 /* Protect *addr6 from overwrites. */
 715                 tmp = *sin6;
 716                 sin6 = &tmp;
 717
 718                 /*
 719                  * Application should provide a proper zone ID or the use of
 720                  * default zone IDs should be enabled.  Unfortunately, some
 721                  * applications do not behave as it should, so we need a
 722                  * workaround.  Even if an appropriate ID is not determined,
 723                  * we'll see if we can determine the outgoing interface.  If we
 724                  * can, determine the zone ID based on the interface below.
 725                  */
 726                 if (sin6->sin6_scope_id == 0 && !V_ip6_use_defzone)
 727                         scope_ambiguous = 1;
 728                 if ((error = sa6_embedscope(sin6, V_ip6_use_defzone)) != 0) {
 729                         if (control)
 730                                 m_freem(control);
 731                         m_freem(m);
 732                         return (error);
 733                 }
 734         }
 735
 736         inp = sotoinpcb(so);
 737         KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
 738         /*
 739          * In the following cases we want a write lock on the inp for either
 740          * local operations or for possible route cache updates in the IPv6
 741          * output path:
 742          * - on connected sockets (sin6 is NULL) for route cache updates,
 743          * - when we are not bound to an address and source port (it is
 744          *   in6_pcbsetport() which will require the write lock).
 745          *
 746          * We check the inp fields before actually locking the inp, so
 747          * here exists a race, and we may WLOCK the inp and end with already
 748          * bound one by other thread. This is fine.
 749          */
 750         if (sin6 == NULL || (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) &&
 751             inp->inp_lport == 0))
 752                 INP_WLOCK(inp);
 753         else
 754                 INP_RLOCK(inp);
 755
 756         nxt = (inp->inp_socket->so_proto->pr_protocol == IPPROTO_UDP) ?
 757             IPPROTO_UDP : IPPROTO_UDPLITE;
 758
 759 #ifdef INET
 760         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
 761                 int hasv4addr;
 762
 763                 if (sin6 == NULL)
 764                         hasv4addr = (inp->inp_vflag & INP_IPV4);
 765                 else
 766                         hasv4addr = IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)
 767                             ? 1 : 0;
 768                 if (hasv4addr) {
 769                         struct pr_usrreqs *pru;
 770
 771                         /*
 772                          * XXXRW: We release UDP-layer locks before calling
 773                          * udp_send() in order to avoid recursion.  However,
 774                          * this does mean there is a short window where inp's
 775                          * fields are unstable.  Could this lead to a
 776                          * potential race in which the factors causing us to
 777                          * select the UDPv4 output routine are invalidated?
 778                          */
 779                         INP_UNLOCK(inp);
 780                         if (sin6)
 781                                 in6_sin6_2_sin_in_sock((struct sockaddr *)sin6);
 782                         pru = inetsw[ip_protox[nxt]].pr_usrreqs;
 783                         /* addr will just be freed in sendit(). */
 784                         return ((*pru->pru_send)(so, flags_arg, m,
 785                             (struct sockaddr *)sin6, control, td));
 786                 }
 787         } else
 788 #endif
 789         if (sin6 && IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
 790                 /*
 791                  * Given this is either an IPv6-only socket or no INET is
 792                  * supported we will fail the send if the given destination
 793                  * address is a v4mapped address.
 794                  *
 795                  * XXXGL: do we leak m and control?
 796                  */
 797                 INP_UNLOCK(inp);
 798                 return (EINVAL);
 799         }
 800
 801         if (control) {
 802                 if ((error = ip6_setpktopts(control, &opt,
 803                     inp->in6p_outputopts, td->td_ucred, nxt)) != 0) {
 804                         INP_UNLOCK(inp);
 805                         ip6_clearpktopts(&opt, -1);
 806                         if (control)
 807                                 m_freem(control);
 808                         m_freem(m);
 809                         return (error);
 810                 }
 811                 optp = &opt;
 812         } else
 813                 optp = inp->in6p_outputopts;
 814
 815         NET_EPOCH_ENTER(et);
 816         if (sin6) {
 817
 818                 /*
 819                  * Since we saw no essential reason for calling in_pcbconnect,
 820                  * we get rid of such kind of logic, and call in6_selectsrc
 821                  * and in6_pcbsetport in order to fill in the local address
 822                  * and the local port.
 823                  */
 824                 if (sin6->sin6_port == 0) {
 825                         error = EADDRNOTAVAIL;
 826                         goto release;
 827                 }
 828
 829                 if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
 830                         /* how about ::ffff:0.0.0.0 case? */
 831                         error = EISCONN;
 832                         goto release;
 833                 }
 834
 835                 /*
 836                  * Given we handle the v4mapped case in the INET block above
 837                  * assert here that it must not happen anymore.
 838                  */
 839                 KASSERT(!IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr),
 840                     ("%s: sin6(%p)->sin6_addr is v4mapped which we "
 841                     "should have handled.", __func__, sin6));
 842
 843                 /* This only requires read-locking. */
 844                 error = in6_selectsrc_socket(sin6, optp, inp,
 845                     td->td_ucred, scope_ambiguous, &in6a, NULL);
 846                 if (error)
 847                         goto release;
 848                 laddr = &in6a;
 849
 850                 if (inp->inp_lport == 0) {
 851                         struct inpcbinfo *pcbinfo;
 852
 853                         INP_WLOCK_ASSERT(inp);
 854
 855                         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
 856                         INP_HASH_WLOCK(pcbinfo);
 857                         error = in6_pcbsetport(laddr, inp, td->td_ucred);
 858                         INP_HASH_WUNLOCK(pcbinfo);
 859                         if (error != 0) {
 860                                 /* Undo an address bind that may have occurred. */
 861                                 inp->in6p_laddr = in6addr_any;
 862                                 goto release;
 863                         }
 864                 }
 865                 faddr = &sin6->sin6_addr;
 866                 fport = sin6->sin6_port; /* allow 0 port */
 867
 868         } else {
 869                 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
 870                         error = ENOTCONN;
 871                         goto release;
 872                 }
 873                 laddr = &inp->in6p_laddr;
 874                 faddr = &inp->in6p_faddr;
 875                 fport = inp->inp_fport;
 876         }
 877
 878         ulen = m->m_pkthdr.len;
 879         plen = sizeof(struct udphdr) + ulen;
 880         hlen = sizeof(struct ip6_hdr);
 881
 882         /*
 883          * Calculate data length and get a mbuf
 884          * for UDP and IP6 headers.
 885          */
 886         M_PREPEND(m, hlen + sizeof(struct udphdr), M_NOWAIT);
 887         if (m == NULL) {
 888                 error = ENOBUFS;
 889                 goto release;
 890         }
 891
 892         /*
 893          * Stuff checksum and output datagram.
 894          */
 895         cscov = cscov_partial = 0;
 896         udp6 = (struct udphdr *)(mtod(m, caddr_t) + hlen);
 897         udp6->uh_sport = inp->inp_lport; /* lport is always set in the PCB */
 898         udp6->uh_dport = fport;
 899         if (nxt == IPPROTO_UDPLITE) {
 900                 struct udpcb *up;
 901
 902                 up = intoudpcb(inp);
 903                 cscov = up->u_txcslen;
 904                 if (cscov >= plen)
 905                         cscov = 0;
 906                 udp6->uh_ulen = htons(cscov);
 907                 /*
 908                  * For UDP-Lite, checksum coverage length of zero means
 909                  * the entire UDPLite packet is covered by the checksum.
 910                  */
 911                 cscov_partial = (cscov == 0) ? 0 : 1;
 912         } else if (plen <= 0xffff)
 913                 udp6->uh_ulen = htons((u_short)plen);
 914         else
 915                 udp6->uh_ulen = 0;
 916         udp6->uh_sum = 0;
 917
 918         ip6 = mtod(m, struct ip6_hdr *);
 919         ip6->ip6_flow   = inp->inp_flow & IPV6_FLOWINFO_MASK;
 920         ip6->ip6_vfc    &= ~IPV6_VERSION_MASK;
 921         ip6->ip6_vfc    |= IPV6_VERSION;
 922         ip6->ip6_plen   = htons((u_short)plen);
 923         ip6->ip6_nxt    = nxt;
 924         ip6->ip6_hlim   = in6_selecthlim(inp, NULL);
 925         ip6->ip6_src    = *laddr;
 926         ip6->ip6_dst    = *faddr;
 927
 928 #ifdef MAC
 929         mac_inpcb_create_mbuf(inp, m);
 930 #endif
 931
 932         if (cscov_partial) {
 933                 if ((udp6->uh_sum = in6_cksum_partial(m, nxt,
 934                     sizeof(struct ip6_hdr), plen, cscov)) == 0)
 935                         udp6->uh_sum = 0xffff;
 936         } else {
 937                 udp6->uh_sum = in6_cksum_pseudo(ip6, plen, nxt, 0);
 938                 m->m_pkthdr.csum_flags = CSUM_UDP_IPV6;
 939                 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
 940         }
 941
 942         flags = 0;
 943 #ifdef  RSS
 944         {
 945                 uint32_t hash_val, hash_type;
 946                 uint8_t pr;
 947
 948                 pr = inp->inp_socket->so_proto->pr_protocol;
 949                 /*
 950                  * Calculate an appropriate RSS hash for UDP and
 951                  * UDP Lite.
 952                  *
 953                  * The called function will take care of figuring out
 954                  * whether a 2-tuple or 4-tuple hash is required based
 955                  * on the currently configured scheme.
 956                  *
 957                  * Later later on connected socket values should be
 958                  * cached in the inpcb and reused, rather than constantly
 959                  * re-calculating it.
 960                  *
 961                  * UDP Lite is a different protocol number and will
 962                  * likely end up being hashed as a 2-tuple until
 963                  * RSS / NICs grow UDP Lite protocol awareness.
 964                  */
 965                 if (rss_proto_software_hash_v6(faddr, laddr, fport,
 966                     inp->inp_lport, pr, &hash_val, &hash_type) == 0) {
 967                         m->m_pkthdr.flowid = hash_val;
 968                         M_HASHTYPE_SET(m, hash_type);
 969                 }
 970
 971                 /*
 972                  * Don't override with the inp cached flowid.
 973                  *
 974                  * Until the whole UDP path is vetted, it may actually
 975                  * be incorrect.
 976                  */
 977                 flags |= IP_NODEFAULTFLOWID;
 978         }
 979 #endif
 980
 981         UDPSTAT_INC(udps_opackets);
 982         if (nxt == IPPROTO_UDPLITE)
 983                 UDPLITE_PROBE(send, NULL, inp, ip6, inp, udp6);
 984         else
 985                 UDP_PROBE(send, NULL, inp, ip6, inp, udp6);
 986         error = ip6_output(m, optp,
 987             INP_WLOCKED(inp) ? &inp->inp_route6 : NULL, flags,
 988             inp->in6p_moptions, NULL, inp);
 989         INP_UNLOCK(inp);
 990         NET_EPOCH_EXIT(et);
 991
 992         if (control) {
 993                 ip6_clearpktopts(&opt, -1);
 994                 m_freem(control);
 995         }
 996         return (error);
 997
 998 release:
 999         INP_UNLOCK(inp);
1000         NET_EPOCH_EXIT(et);
1001         if (control) {
1002                 ip6_clearpktopts(&opt, -1);
1003                 m_freem(control);
1004         }
1005         m_freem(m);
1006
1007         return (error);
1008 }
1009
1010 static void
1011 udp6_abort(struct socket *so)
1012 {
1013         struct inpcb *inp;
1014         struct inpcbinfo *pcbinfo;
1015
1016         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1017         inp = sotoinpcb(so);
1018         KASSERT(inp != NULL, ("udp6_abort: inp == NULL"));
1019
1020         INP_WLOCK(inp);
1021 #ifdef INET
1022         if (inp->inp_vflag & INP_IPV4) {
1023                 struct pr_usrreqs *pru;
1024                 uint8_t nxt;
1025
1026                 nxt = (inp->inp_socket->so_proto->pr_protocol == IPPROTO_UDP) ?
1027                     IPPROTO_UDP : IPPROTO_UDPLITE;
1028                 INP_WUNLOCK(inp);
1029                 pru = inetsw[ip_protox[nxt]].pr_usrreqs;
1030                 (*pru->pru_abort)(so);
1031                 return;
1032         }
1033 #endif
1034
1035         if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
1036                 INP_HASH_WLOCK(pcbinfo);
1037                 in6_pcbdisconnect(inp);
1038                 inp->in6p_laddr = in6addr_any;
1039                 INP_HASH_WUNLOCK(pcbinfo);
1040                 soisdisconnected(so);
1041         }
1042         INP_WUNLOCK(inp);
1043 }
1044
1045 static int
1046 udp6_attach(struct socket *so, int proto, struct thread *td)
1047 {
1048         struct inpcb *inp;
1049         struct inpcbinfo *pcbinfo;
1050         int error;
1051
1052         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1053         inp = sotoinpcb(so);
1054         KASSERT(inp == NULL, ("udp6_attach: inp != NULL"));
1055
1056         if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
1057                 error = soreserve(so, udp_sendspace, udp_recvspace);
1058                 if (error)
1059                         return (error);
1060         }
1061         INP_INFO_WLOCK(pcbinfo);
1062         error = in_pcballoc(so, pcbinfo);
1063         if (error) {
1064                 INP_INFO_WUNLOCK(pcbinfo);
1065                 return (error);
1066         }
1067         inp = (struct inpcb *)so->so_pcb;
1068         inp->inp_vflag |= INP_IPV6;
1069         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
1070                 inp->inp_vflag |= INP_IPV4;
1071         inp->in6p_hops = -1;    /* use kernel default */
1072         inp->in6p_cksum = -1;   /* just to be sure */
1073         /*
1074          * XXX: ugly!!
1075          * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1076          * because the socket may be bound to an IPv6 wildcard address,
1077          * which may match an IPv4-mapped IPv6 address.
1078          */
1079         inp->inp_ip_ttl = V_ip_defttl;
1080
1081         error = udp_newudpcb(inp);
1082         if (error) {
1083                 in_pcbdetach(inp);
1084                 in_pcbfree(inp);
1085                 INP_INFO_WUNLOCK(pcbinfo);
1086                 return (error);
1087         }
1088         INP_WUNLOCK(inp);
1089         INP_INFO_WUNLOCK(pcbinfo);
1090         return (0);
1091 }
1092
1093 static int
1094 udp6_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1095 {
1096         struct inpcb *inp;
1097         struct inpcbinfo *pcbinfo;
1098         int error;
1099         u_char vflagsav;
1100
1101         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1102         inp = sotoinpcb(so);
1103         KASSERT(inp != NULL, ("udp6_bind: inp == NULL"));
1104
1105         INP_WLOCK(inp);
1106         INP_HASH_WLOCK(pcbinfo);
1107         vflagsav = inp->inp_vflag;
1108         inp->inp_vflag &= ~INP_IPV4;
1109         inp->inp_vflag |= INP_IPV6;
1110         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
1111                 struct sockaddr_in6 *sin6_p;
1112
1113                 sin6_p = (struct sockaddr_in6 *)nam;
1114
1115                 if (IN6_IS_ADDR_UNSPECIFIED(&sin6_p->sin6_addr))
1116                         inp->inp_vflag |= INP_IPV4;
1117 #ifdef INET
1118                 else if (IN6_IS_ADDR_V4MAPPED(&sin6_p->sin6_addr)) {
1119                         struct sockaddr_in sin;
1120
1121                         in6_sin6_2_sin(&sin, sin6_p);
1122                         inp->inp_vflag |= INP_IPV4;
1123                         inp->inp_vflag &= ~INP_IPV6;
1124                         error = in_pcbbind(inp, (struct sockaddr *)&sin,
1125                             td->td_ucred);
1126                         goto out;
1127                 }
1128 #endif
1129         }
1130
1131         error = in6_pcbbind(inp, nam, td->td_ucred);
1132 #ifdef INET
1133 out:
1134 #endif
1135         if (error != 0)
1136                 inp->inp_vflag = vflagsav;
1137         INP_HASH_WUNLOCK(pcbinfo);
1138         INP_WUNLOCK(inp);
1139         return (error);
1140 }
1141
1142 static void
1143 udp6_close(struct socket *so)
1144 {
1145         struct inpcb *inp;
1146         struct inpcbinfo *pcbinfo;
1147
1148         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1149         inp = sotoinpcb(so);
1150         KASSERT(inp != NULL, ("udp6_close: inp == NULL"));
1151
1152         INP_WLOCK(inp);
1153 #ifdef INET
1154         if (inp->inp_vflag & INP_IPV4) {
1155                 struct pr_usrreqs *pru;
1156                 uint8_t nxt;
1157
1158                 nxt = (inp->inp_socket->so_proto->pr_protocol == IPPROTO_UDP) ?
1159                     IPPROTO_UDP : IPPROTO_UDPLITE;
1160                 INP_WUNLOCK(inp);
1161                 pru = inetsw[ip_protox[nxt]].pr_usrreqs;
1162                 (*pru->pru_disconnect)(so);
1163                 return;
1164         }
1165 #endif
1166         if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
1167                 INP_HASH_WLOCK(pcbinfo);
1168                 in6_pcbdisconnect(inp);
1169                 inp->in6p_laddr = in6addr_any;
1170                 INP_HASH_WUNLOCK(pcbinfo);
1171                 soisdisconnected(so);
1172         }
1173         INP_WUNLOCK(inp);
1174 }
1175
1176 static int
1177 udp6_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1178 {
1179         struct inpcb *inp;
1180         struct inpcbinfo *pcbinfo;
1181         struct sockaddr_in6 *sin6;
1182         int error;
1183         u_char vflagsav;
1184
1185         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1186         inp = sotoinpcb(so);
1187         sin6 = (struct sockaddr_in6 *)nam;
1188         KASSERT(inp != NULL, ("udp6_connect: inp == NULL"));
1189
1190         /*
1191          * XXXRW: Need to clarify locking of v4/v6 flags.
1192          */
1193         INP_WLOCK(inp);
1194 #ifdef INET
1195         if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
1196                 struct sockaddr_in sin;
1197
1198                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
1199                         error = EINVAL;
1200                         goto out;
1201                 }
1202                 if ((inp->inp_vflag & INP_IPV4) == 0) {
1203                         error = EAFNOSUPPORT;
1204                         goto out;
1205                 }
1206                 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1207                         error = EISCONN;
1208                         goto out;
1209                 }
1210                 in6_sin6_2_sin(&sin, sin6);
1211                 error = prison_remote_ip4(td->td_ucred, &sin.sin_addr);
1212                 if (error != 0)
1213                         goto out;
1214                 vflagsav = inp->inp_vflag;
1215                 inp->inp_vflag |= INP_IPV4;
1216                 inp->inp_vflag &= ~INP_IPV6;
1217                 INP_HASH_WLOCK(pcbinfo);
1218                 error = in_pcbconnect(inp, (struct sockaddr *)&sin,
1219                     td->td_ucred);
1220                 INP_HASH_WUNLOCK(pcbinfo);
1221                 /*
1222                  * If connect succeeds, mark socket as connected. If
1223                  * connect fails and socket is unbound, reset inp_vflag
1224                  * field.
1225                  */
1226                 if (error == 0)
1227                         soisconnected(so);
1228                 else if (inp->inp_laddr.s_addr == INADDR_ANY &&
1229                     inp->inp_lport == 0)
1230                         inp->inp_vflag = vflagsav;
1231                 goto out;
1232         } else {
1233                 if ((inp->inp_vflag & INP_IPV6) == 0) {
1234                         error = EAFNOSUPPORT;
1235                         goto out;
1236                 }
1237         }
1238 #endif
1239         if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
1240                 error = EISCONN;
1241                 goto out;
1242         }
1243         error = prison_remote_ip6(td->td_ucred, &sin6->sin6_addr);
1244         if (error != 0)
1245                 goto out;
1246         vflagsav = inp->inp_vflag;
1247         inp->inp_vflag &= ~INP_IPV4;
1248         inp->inp_vflag |= INP_IPV6;
1249         INP_HASH_WLOCK(pcbinfo);
1250         error = in6_pcbconnect(inp, nam, td->td_ucred);
1251         INP_HASH_WUNLOCK(pcbinfo);
1252         /*
1253          * If connect succeeds, mark socket as connected. If
1254          * connect fails and socket is unbound, reset inp_vflag
1255          * field.
1256          */
1257         if (error == 0)
1258                 soisconnected(so);
1259         else if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) &&
1260             inp->inp_lport == 0)
1261                 inp->inp_vflag = vflagsav;
1262 out:
1263         INP_WUNLOCK(inp);
1264         return (error);
1265 }
1266
1267 static void
1268 udp6_detach(struct socket *so)
1269 {
1270         struct inpcb *inp;
1271         struct inpcbinfo *pcbinfo;
1272         struct udpcb *up;
1273
1274         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1275         inp = sotoinpcb(so);
1276         KASSERT(inp != NULL, ("udp6_detach: inp == NULL"));
1277
1278         INP_INFO_WLOCK(pcbinfo);
1279         INP_WLOCK(inp);
1280         up = intoudpcb(inp);
1281         KASSERT(up != NULL, ("%s: up == NULL", __func__));
1282         in_pcbdetach(inp);
1283         in_pcbfree(inp);
1284         INP_INFO_WUNLOCK(pcbinfo);
1285         udp_discardcb(up);
1286 }
1287
1288 static int
1289 udp6_disconnect(struct socket *so)
1290 {
1291         struct inpcb *inp;
1292         struct inpcbinfo *pcbinfo;
1293
1294         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1295         inp = sotoinpcb(so);
1296         KASSERT(inp != NULL, ("udp6_disconnect: inp == NULL"));
1297
1298         INP_WLOCK(inp);
1299 #ifdef INET
1300         if (inp->inp_vflag & INP_IPV4) {
1301                 struct pr_usrreqs *pru;
1302                 uint8_t nxt;
1303
1304                 nxt = (inp->inp_socket->so_proto->pr_protocol == IPPROTO_UDP) ?
1305                     IPPROTO_UDP : IPPROTO_UDPLITE;
1306                 INP_WUNLOCK(inp);
1307                 pru = inetsw[ip_protox[nxt]].pr_usrreqs;
1308                 (void)(*pru->pru_disconnect)(so);
1309                 return (0);
1310         }
1311 #endif
1312
1313         if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
1314                 INP_WUNLOCK(inp);
1315                 return (ENOTCONN);
1316         }
1317
1318         INP_HASH_WLOCK(pcbinfo);
1319         in6_pcbdisconnect(inp);
1320         inp->in6p_laddr = in6addr_any;
1321         INP_HASH_WUNLOCK(pcbinfo);
1322         SOCK_LOCK(so);
1323         so->so_state &= ~SS_ISCONNECTED;                /* XXX */
1324         SOCK_UNLOCK(so);
1325         INP_WUNLOCK(inp);
1326         return (0);
1327 }
1328
1329 static int
1330 udp6_send(struct socket *so, int flags, struct mbuf *m,
1331     struct sockaddr *addr, struct mbuf *control, struct thread *td)
1332 {
1333         int error;
1334
1335         if (addr) {
1336                 if (addr->sa_len != sizeof(struct sockaddr_in6)) {
1337                         error = EINVAL;
1338                         goto bad;
1339                 }
1340                 if (addr->sa_family != AF_INET6) {
1341                         error = EAFNOSUPPORT;
1342                         goto bad;
1343                 }
1344         }
1345
1346         return (udp6_output(so, flags, m, addr, control, td));
1347
1348 bad:
1349         if (control)
1350                 m_freem(control);
1351         m_freem(m);
1352         return (error);
1353 }
1354
1355 struct pr_usrreqs udp6_usrreqs = {
1356         .pru_abort =            udp6_abort,
1357         .pru_attach =           udp6_attach,
1358         .pru_bind =             udp6_bind,
1359         .pru_connect =          udp6_connect,
1360         .pru_control =          in6_control,
1361         .pru_detach =           udp6_detach,
1362         .pru_disconnect =       udp6_disconnect,
1363         .pru_peeraddr =         in6_mapped_peeraddr,
1364         .pru_send =             udp6_send,
1365         .pru_shutdown =         udp_shutdown,
1366         .pru_sockaddr =         in6_mapped_sockaddr,
1367         .pru_soreceive =        soreceive_dgram,
1368         .pru_sosend =           sosend_dgram,
1369         .pru_sosetlabel =       in_pcbsosetlabel,
1370         .pru_close =            udp6_close
1371 };