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         if (m->m_len < off + sizeof(struct udphdr)) {
 227                 m = m_pullup(m, off + sizeof(struct udphdr));
 228                 if (m == NULL) {
 229                         IP6STAT_INC(ip6s_exthdrtoolong);
 230                         *mp = NULL;
 231                         return (IPPROTO_DONE);
 232                 }
 233         }
 234         ip6 = mtod(m, struct ip6_hdr *);
 235         uh = (struct udphdr *)((caddr_t)ip6 + off);
 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 (V_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         m_freem(m);
 532         *mp = NULL;
 533         return (IPPROTO_DONE);
 534 }
 535
 536 static void
 537 udp6_common_ctlinput(int cmd, struct sockaddr *sa, void *d,
 538     struct inpcbinfo *pcbinfo)
 539 {
 540         struct udphdr uh;
 541         struct ip6_hdr *ip6;
 542         struct mbuf *m;
 543         int off = 0;
 544         struct ip6ctlparam *ip6cp = NULL;
 545         const struct sockaddr_in6 *sa6_src = NULL;
 546         void *cmdarg;
 547         struct inpcb *(*notify)(struct inpcb *, int) = udp_notify;
 548         struct udp_portonly {
 549                 u_int16_t uh_sport;
 550                 u_int16_t uh_dport;
 551         } *uhp;
 552
 553         if (sa->sa_family != AF_INET6 ||
 554             sa->sa_len != sizeof(struct sockaddr_in6))
 555                 return;
 556
 557         if ((unsigned)cmd >= PRC_NCMDS)
 558                 return;
 559         if (PRC_IS_REDIRECT(cmd))
 560                 notify = in6_rtchange, d = NULL;
 561         else if (cmd == PRC_HOSTDEAD)
 562                 d = NULL;
 563         else if (inet6ctlerrmap[cmd] == 0)
 564                 return;
 565
 566         /* if the parameter is from icmp6, decode it. */
 567         if (d != NULL) {
 568                 ip6cp = (struct ip6ctlparam *)d;
 569                 m = ip6cp->ip6c_m;
 570                 ip6 = ip6cp->ip6c_ip6;
 571                 off = ip6cp->ip6c_off;
 572                 cmdarg = ip6cp->ip6c_cmdarg;
 573                 sa6_src = ip6cp->ip6c_src;
 574         } else {
 575                 m = NULL;
 576                 ip6 = NULL;
 577                 cmdarg = NULL;
 578                 sa6_src = &sa6_any;
 579         }
 580
 581         if (ip6) {
 582                 /*
 583                  * XXX: We assume that when IPV6 is non NULL,
 584                  * M and OFF are valid.
 585                  */
 586
 587                 /* Check if we can safely examine src and dst ports. */
 588                 if (m->m_pkthdr.len < off + sizeof(*uhp))
 589                         return;
 590
 591                 bzero(&uh, sizeof(uh));
 592                 m_copydata(m, off, sizeof(*uhp), (caddr_t)&uh);
 593
 594                 if (!PRC_IS_REDIRECT(cmd)) {
 595                         /* Check to see if its tunneled */
 596                         struct inpcb *inp;
 597                         inp = in6_pcblookup_mbuf(pcbinfo, &ip6->ip6_dst,
 598                             uh.uh_dport, &ip6->ip6_src, uh.uh_sport,
 599                             INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB,
 600                             m->m_pkthdr.rcvif, m);
 601                         if (inp != NULL) {
 602                                 struct udpcb *up;
 603
 604                                 up = intoudpcb(inp);
 605                                 if (up->u_icmp_func) {
 606                                         /* Yes it is. */
 607                                         INP_RUNLOCK(inp);
 608                                         (*up->u_icmp_func)(cmd, (struct sockaddr *)ip6cp->ip6c_src,
 609                                               d, up->u_tun_ctx);
 610                                         return;
 611                                 } else {
 612                                         /* Can't find it. */
 613                                         INP_RUNLOCK(inp);
 614                                 }
 615                         }
 616                 }
 617                 (void)in6_pcbnotify(pcbinfo, sa, uh.uh_dport,
 618                     (struct sockaddr *)ip6cp->ip6c_src, uh.uh_sport, cmd,
 619                     cmdarg, notify);
 620         } else
 621                 (void)in6_pcbnotify(pcbinfo, sa, 0,
 622                     (const struct sockaddr *)sa6_src, 0, cmd, cmdarg, notify);
 623 }
 624
 625 void
 626 udp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
 627 {
 628
 629         return (udp6_common_ctlinput(cmd, sa, d, &V_udbinfo));
 630 }
 631
 632 void
 633 udplite6_ctlinput(int cmd, struct sockaddr *sa, void *d)
 634 {
 635
 636         return (udp6_common_ctlinput(cmd, sa, d, &V_ulitecbinfo));
 637 }
 638
 639 static int
 640 udp6_getcred(SYSCTL_HANDLER_ARGS)
 641 {
 642         struct xucred xuc;
 643         struct sockaddr_in6 addrs[2];
 644         struct epoch_tracker et;
 645         struct inpcb *inp;
 646         int error;
 647
 648         error = priv_check(req->td, PRIV_NETINET_GETCRED);
 649         if (error)
 650                 return (error);
 651
 652         if (req->newlen != sizeof(addrs))
 653                 return (EINVAL);
 654         if (req->oldlen != sizeof(struct xucred))
 655                 return (EINVAL);
 656         error = SYSCTL_IN(req, addrs, sizeof(addrs));
 657         if (error)
 658                 return (error);
 659         if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
 660             (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
 661                 return (error);
 662         }
 663         NET_EPOCH_ENTER(et);
 664         inp = in6_pcblookup(&V_udbinfo, &addrs[1].sin6_addr,
 665             addrs[1].sin6_port, &addrs[0].sin6_addr, addrs[0].sin6_port,
 666             INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
 667         NET_EPOCH_EXIT(et);
 668         if (inp != NULL) {
 669                 INP_RLOCK_ASSERT(inp);
 670                 if (inp->inp_socket == NULL)
 671                         error = ENOENT;
 672                 if (error == 0)
 673                         error = cr_canseesocket(req->td->td_ucred,
 674                             inp->inp_socket);
 675                 if (error == 0)
 676                         cru2x(inp->inp_cred, &xuc);
 677                 INP_RUNLOCK(inp);
 678         } else
 679                 error = ENOENT;
 680         if (error == 0)
 681                 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
 682         return (error);
 683 }
 684
 685 SYSCTL_PROC(_net_inet6_udp6, OID_AUTO, getcred, CTLTYPE_OPAQUE|CTLFLAG_RW, 0,
 686     0, udp6_getcred, "S,xucred", "Get the xucred of a UDP6 connection");
 687
 688 static int
 689 udp6_output(struct socket *so, int flags_arg, struct mbuf *m,
 690     struct sockaddr *addr6, struct mbuf *control, struct thread *td)
 691 {
 692         struct inpcb *inp;
 693         struct ip6_hdr *ip6;
 694         struct udphdr *udp6;
 695         struct in6_addr *laddr, *faddr, in6a;
 696         struct ip6_pktopts *optp, opt;
 697         struct sockaddr_in6 *sin6, tmp;
 698         struct epoch_tracker et;
 699         int cscov_partial, error, flags, hlen, scope_ambiguous;
 700         u_int32_t ulen, plen;
 701         uint16_t cscov;
 702         u_short fport;
 703         uint8_t nxt;
 704
 705         /* addr6 has been validated in udp6_send(). */
 706         sin6 = (struct sockaddr_in6 *)addr6;
 707
 708         /*
 709          * In contrast to to IPv4 we do not validate the max. packet length
 710          * here due to IPv6 Jumbograms (RFC2675).
 711          */
 712
 713         scope_ambiguous = 0;
 714         if (sin6) {
 715                 /* Protect *addr6 from overwrites. */
 716                 tmp = *sin6;
 717                 sin6 = &tmp;
 718
 719                 /*
 720                  * Application should provide a proper zone ID or the use of
 721                  * default zone IDs should be enabled.  Unfortunately, some
 722                  * applications do not behave as it should, so we need a
 723                  * workaround.  Even if an appropriate ID is not determined,
 724                  * we'll see if we can determine the outgoing interface.  If we
 725                  * can, determine the zone ID based on the interface below.
 726                  */
 727                 if (sin6->sin6_scope_id == 0 && !V_ip6_use_defzone)
 728                         scope_ambiguous = 1;
 729                 if ((error = sa6_embedscope(sin6, V_ip6_use_defzone)) != 0) {
 730                         if (control)
 731                                 m_freem(control);
 732                         m_freem(m);
 733                         return (error);
 734                 }
 735         }
 736
 737         inp = sotoinpcb(so);
 738         KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
 739         /*
 740          * In the following cases we want a write lock on the inp for either
 741          * local operations or for possible route cache updates in the IPv6
 742          * output path:
 743          * - on connected sockets (sin6 is NULL) for route cache updates,
 744          * - when we are not bound to an address and source port (it is
 745          *   in6_pcbsetport() which will require the write lock).
 746          *
 747          * We check the inp fields before actually locking the inp, so
 748          * here exists a race, and we may WLOCK the inp and end with already
 749          * bound one by other thread. This is fine.
 750          */
 751         if (sin6 == NULL || (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) &&
 752             inp->inp_lport == 0))
 753                 INP_WLOCK(inp);
 754         else
 755                 INP_RLOCK(inp);
 756
 757         nxt = (inp->inp_socket->so_proto->pr_protocol == IPPROTO_UDP) ?
 758             IPPROTO_UDP : IPPROTO_UDPLITE;
 759
 760 #ifdef INET
 761         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
 762                 int hasv4addr;
 763
 764                 if (sin6 == NULL)
 765                         hasv4addr = (inp->inp_vflag & INP_IPV4);
 766                 else
 767                         hasv4addr = IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)
 768                             ? 1 : 0;
 769                 if (hasv4addr) {
 770                         struct pr_usrreqs *pru;
 771
 772                         /*
 773                          * XXXRW: We release UDP-layer locks before calling
 774                          * udp_send() in order to avoid recursion.  However,
 775                          * this does mean there is a short window where inp's
 776                          * fields are unstable.  Could this lead to a
 777                          * potential race in which the factors causing us to
 778                          * select the UDPv4 output routine are invalidated?
 779                          */
 780                         INP_UNLOCK(inp);
 781                         if (sin6)
 782                                 in6_sin6_2_sin_in_sock((struct sockaddr *)sin6);
 783                         pru = inetsw[ip_protox[nxt]].pr_usrreqs;
 784                         /* addr will just be freed in sendit(). */
 785                         return ((*pru->pru_send)(so, flags_arg, m,
 786                             (struct sockaddr *)sin6, control, td));
 787                 }
 788         } else
 789 #endif
 790         if (sin6 && IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
 791                 /*
 792                  * Given this is either an IPv6-only socket or no INET is
 793                  * supported we will fail the send if the given destination
 794                  * address is a v4mapped address.
 795                  *
 796                  * XXXGL: do we leak m and control?
 797                  */
 798                 INP_UNLOCK(inp);
 799                 return (EINVAL);
 800         }
 801
 802         if (control) {
 803                 if ((error = ip6_setpktopts(control, &opt,
 804                     inp->in6p_outputopts, td->td_ucred, nxt)) != 0) {
 805                         INP_UNLOCK(inp);
 806                         ip6_clearpktopts(&opt, -1);
 807                         if (control)
 808                                 m_freem(control);
 809                         m_freem(m);
 810                         return (error);
 811                 }
 812                 optp = &opt;
 813         } else
 814                 optp = inp->in6p_outputopts;
 815
 816         NET_EPOCH_ENTER(et);
 817         if (sin6) {
 818
 819                 /*
 820                  * Since we saw no essential reason for calling in_pcbconnect,
 821                  * we get rid of such kind of logic, and call in6_selectsrc
 822                  * and in6_pcbsetport in order to fill in the local address
 823                  * and the local port.
 824                  */
 825                 if (sin6->sin6_port == 0) {
 826                         error = EADDRNOTAVAIL;
 827                         goto release;
 828                 }
 829
 830                 if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
 831                         /* how about ::ffff:0.0.0.0 case? */
 832                         error = EISCONN;
 833                         goto release;
 834                 }
 835
 836                 /*
 837                  * Given we handle the v4mapped case in the INET block above
 838                  * assert here that it must not happen anymore.
 839                  */
 840                 KASSERT(!IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr),
 841                     ("%s: sin6(%p)->sin6_addr is v4mapped which we "
 842                     "should have handled.", __func__, sin6));
 843
 844                 /* This only requires read-locking. */
 845                 error = in6_selectsrc_socket(sin6, optp, inp,
 846                     td->td_ucred, scope_ambiguous, &in6a, NULL);
 847                 if (error)
 848                         goto release;
 849                 laddr = &in6a;
 850
 851                 if (inp->inp_lport == 0) {
 852                         struct inpcbinfo *pcbinfo;
 853
 854                         INP_WLOCK_ASSERT(inp);
 855
 856                         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
 857                         INP_HASH_WLOCK(pcbinfo);
 858                         error = in6_pcbsetport(laddr, inp, td->td_ucred);
 859                         INP_HASH_WUNLOCK(pcbinfo);
 860                         if (error != 0) {
 861                                 /* Undo an address bind that may have occurred. */
 862                                 inp->in6p_laddr = in6addr_any;
 863                                 goto release;
 864                         }
 865                 }
 866                 faddr = &sin6->sin6_addr;
 867                 fport = sin6->sin6_port; /* allow 0 port */
 868
 869         } else {
 870                 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
 871                         error = ENOTCONN;
 872                         goto release;
 873                 }
 874                 laddr = &inp->in6p_laddr;
 875                 faddr = &inp->in6p_faddr;
 876                 fport = inp->inp_fport;
 877         }
 878
 879         ulen = m->m_pkthdr.len;
 880         plen = sizeof(struct udphdr) + ulen;
 881         hlen = sizeof(struct ip6_hdr);
 882
 883         /*
 884          * Calculate data length and get a mbuf
 885          * for UDP and IP6 headers.
 886          */
 887         M_PREPEND(m, hlen + sizeof(struct udphdr), M_NOWAIT);
 888         if (m == NULL) {
 889                 error = ENOBUFS;
 890                 goto release;
 891         }
 892
 893         /*
 894          * Stuff checksum and output datagram.
 895          */
 896         cscov = cscov_partial = 0;
 897         udp6 = (struct udphdr *)(mtod(m, caddr_t) + hlen);
 898         udp6->uh_sport = inp->inp_lport; /* lport is always set in the PCB */
 899         udp6->uh_dport = fport;
 900         if (nxt == IPPROTO_UDPLITE) {
 901                 struct udpcb *up;
 902
 903                 up = intoudpcb(inp);
 904                 cscov = up->u_txcslen;
 905                 if (cscov >= plen)
 906                         cscov = 0;
 907                 udp6->uh_ulen = htons(cscov);
 908                 /*
 909                  * For UDP-Lite, checksum coverage length of zero means
 910                  * the entire UDPLite packet is covered by the checksum.
 911                  */
 912                 cscov_partial = (cscov == 0) ? 0 : 1;
 913         } else if (plen <= 0xffff)
 914                 udp6->uh_ulen = htons((u_short)plen);
 915         else
 916                 udp6->uh_ulen = 0;
 917         udp6->uh_sum = 0;
 918
 919         ip6 = mtod(m, struct ip6_hdr *);
 920         ip6->ip6_flow   = inp->inp_flow & IPV6_FLOWINFO_MASK;
 921         ip6->ip6_vfc    &= ~IPV6_VERSION_MASK;
 922         ip6->ip6_vfc    |= IPV6_VERSION;
 923         ip6->ip6_plen   = htons((u_short)plen);
 924         ip6->ip6_nxt    = nxt;
 925         ip6->ip6_hlim   = in6_selecthlim(inp, NULL);
 926         ip6->ip6_src    = *laddr;
 927         ip6->ip6_dst    = *faddr;
 928
 929 #ifdef MAC
 930         mac_inpcb_create_mbuf(inp, m);
 931 #endif
 932
 933         if (cscov_partial) {
 934                 if ((udp6->uh_sum = in6_cksum_partial(m, nxt,
 935                     sizeof(struct ip6_hdr), plen, cscov)) == 0)
 936                         udp6->uh_sum = 0xffff;
 937         } else {
 938                 udp6->uh_sum = in6_cksum_pseudo(ip6, plen, nxt, 0);
 939                 m->m_pkthdr.csum_flags = CSUM_UDP_IPV6;
 940                 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
 941         }
 942
 943         flags = 0;
 944 #ifdef  RSS
 945         {
 946                 uint32_t hash_val, hash_type;
 947                 uint8_t pr;
 948
 949                 pr = inp->inp_socket->so_proto->pr_protocol;
 950                 /*
 951                  * Calculate an appropriate RSS hash for UDP and
 952                  * UDP Lite.
 953                  *
 954                  * The called function will take care of figuring out
 955                  * whether a 2-tuple or 4-tuple hash is required based
 956                  * on the currently configured scheme.
 957                  *
 958                  * Later later on connected socket values should be
 959                  * cached in the inpcb and reused, rather than constantly
 960                  * re-calculating it.
 961                  *
 962                  * UDP Lite is a different protocol number and will
 963                  * likely end up being hashed as a 2-tuple until
 964                  * RSS / NICs grow UDP Lite protocol awareness.
 965                  */
 966                 if (rss_proto_software_hash_v6(faddr, laddr, fport,
 967                     inp->inp_lport, pr, &hash_val, &hash_type) == 0) {
 968                         m->m_pkthdr.flowid = hash_val;
 969                         M_HASHTYPE_SET(m, hash_type);
 970                 }
 971
 972                 /*
 973                  * Don't override with the inp cached flowid.
 974                  *
 975                  * Until the whole UDP path is vetted, it may actually
 976                  * be incorrect.
 977                  */
 978                 flags |= IP_NODEFAULTFLOWID;
 979         }
 980 #endif
 981
 982         UDPSTAT_INC(udps_opackets);
 983         if (nxt == IPPROTO_UDPLITE)
 984                 UDPLITE_PROBE(send, NULL, inp, ip6, inp, udp6);
 985         else
 986                 UDP_PROBE(send, NULL, inp, ip6, inp, udp6);
 987         error = ip6_output(m, optp,
 988             INP_WLOCKED(inp) ? &inp->inp_route6 : NULL, flags,
 989             inp->in6p_moptions, NULL, inp);
 990         INP_UNLOCK(inp);
 991         NET_EPOCH_EXIT(et);
 992
 993         if (control) {
 994                 ip6_clearpktopts(&opt, -1);
 995                 m_freem(control);
 996         }
 997         return (error);
 998
 999 release:
1000         INP_UNLOCK(inp);
1001         NET_EPOCH_EXIT(et);
1002         if (control) {
1003                 ip6_clearpktopts(&opt, -1);
1004                 m_freem(control);
1005         }
1006         m_freem(m);
1007
1008         return (error);
1009 }
1010
1011 static void
1012 udp6_abort(struct socket *so)
1013 {
1014         struct inpcb *inp;
1015         struct inpcbinfo *pcbinfo;
1016
1017         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1018         inp = sotoinpcb(so);
1019         KASSERT(inp != NULL, ("udp6_abort: inp == NULL"));
1020
1021         INP_WLOCK(inp);
1022 #ifdef INET
1023         if (inp->inp_vflag & INP_IPV4) {
1024                 struct pr_usrreqs *pru;
1025                 uint8_t nxt;
1026
1027                 nxt = (inp->inp_socket->so_proto->pr_protocol == IPPROTO_UDP) ?
1028                     IPPROTO_UDP : IPPROTO_UDPLITE;
1029                 INP_WUNLOCK(inp);
1030                 pru = inetsw[ip_protox[nxt]].pr_usrreqs;
1031                 (*pru->pru_abort)(so);
1032                 return;
1033         }
1034 #endif
1035
1036         if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
1037                 INP_HASH_WLOCK(pcbinfo);
1038                 in6_pcbdisconnect(inp);
1039                 inp->in6p_laddr = in6addr_any;
1040                 INP_HASH_WUNLOCK(pcbinfo);
1041                 soisdisconnected(so);
1042         }
1043         INP_WUNLOCK(inp);
1044 }
1045
1046 static int
1047 udp6_attach(struct socket *so, int proto, struct thread *td)
1048 {
1049         struct inpcb *inp;
1050         struct inpcbinfo *pcbinfo;
1051         int error;
1052
1053         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1054         inp = sotoinpcb(so);
1055         KASSERT(inp == NULL, ("udp6_attach: inp != NULL"));
1056
1057         if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
1058                 error = soreserve(so, udp_sendspace, udp_recvspace);
1059                 if (error)
1060                         return (error);
1061         }
1062         INP_INFO_WLOCK(pcbinfo);
1063         error = in_pcballoc(so, pcbinfo);
1064         if (error) {
1065                 INP_INFO_WUNLOCK(pcbinfo);
1066                 return (error);
1067         }
1068         inp = (struct inpcb *)so->so_pcb;
1069         inp->inp_vflag |= INP_IPV6;
1070         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
1071                 inp->inp_vflag |= INP_IPV4;
1072         inp->in6p_hops = -1;    /* use kernel default */
1073         inp->in6p_cksum = -1;   /* just to be sure */
1074         /*
1075          * XXX: ugly!!
1076          * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1077          * because the socket may be bound to an IPv6 wildcard address,
1078          * which may match an IPv4-mapped IPv6 address.
1079          */
1080         inp->inp_ip_ttl = V_ip_defttl;
1081
1082         error = udp_newudpcb(inp);
1083         if (error) {
1084                 in_pcbdetach(inp);
1085                 in_pcbfree(inp);
1086                 INP_INFO_WUNLOCK(pcbinfo);
1087                 return (error);
1088         }
1089         INP_WUNLOCK(inp);
1090         INP_INFO_WUNLOCK(pcbinfo);
1091         return (0);
1092 }
1093
1094 static int
1095 udp6_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1096 {
1097         struct inpcb *inp;
1098         struct inpcbinfo *pcbinfo;
1099         int error;
1100         u_char vflagsav;
1101
1102         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1103         inp = sotoinpcb(so);
1104         KASSERT(inp != NULL, ("udp6_bind: inp == NULL"));
1105
1106         INP_WLOCK(inp);
1107         INP_HASH_WLOCK(pcbinfo);
1108         vflagsav = inp->inp_vflag;
1109         inp->inp_vflag &= ~INP_IPV4;
1110         inp->inp_vflag |= INP_IPV6;
1111         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
1112                 struct sockaddr_in6 *sin6_p;
1113
1114                 sin6_p = (struct sockaddr_in6 *)nam;
1115
1116                 if (IN6_IS_ADDR_UNSPECIFIED(&sin6_p->sin6_addr))
1117                         inp->inp_vflag |= INP_IPV4;
1118 #ifdef INET
1119                 else if (IN6_IS_ADDR_V4MAPPED(&sin6_p->sin6_addr)) {
1120                         struct sockaddr_in sin;
1121
1122                         in6_sin6_2_sin(&sin, sin6_p);
1123                         inp->inp_vflag |= INP_IPV4;
1124                         inp->inp_vflag &= ~INP_IPV6;
1125                         error = in_pcbbind(inp, (struct sockaddr *)&sin,
1126                             td->td_ucred);
1127                         goto out;
1128                 }
1129 #endif
1130         }
1131
1132         error = in6_pcbbind(inp, nam, td->td_ucred);
1133 #ifdef INET
1134 out:
1135 #endif
1136         if (error != 0)
1137                 inp->inp_vflag = vflagsav;
1138         INP_HASH_WUNLOCK(pcbinfo);
1139         INP_WUNLOCK(inp);
1140         return (error);
1141 }
1142
1143 static void
1144 udp6_close(struct socket *so)
1145 {
1146         struct inpcb *inp;
1147         struct inpcbinfo *pcbinfo;
1148
1149         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1150         inp = sotoinpcb(so);
1151         KASSERT(inp != NULL, ("udp6_close: inp == NULL"));
1152
1153         INP_WLOCK(inp);
1154 #ifdef INET
1155         if (inp->inp_vflag & INP_IPV4) {
1156                 struct pr_usrreqs *pru;
1157                 uint8_t nxt;
1158
1159                 nxt = (inp->inp_socket->so_proto->pr_protocol == IPPROTO_UDP) ?
1160                     IPPROTO_UDP : IPPROTO_UDPLITE;
1161                 INP_WUNLOCK(inp);
1162                 pru = inetsw[ip_protox[nxt]].pr_usrreqs;
1163                 (*pru->pru_disconnect)(so);
1164                 return;
1165         }
1166 #endif
1167         if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
1168                 INP_HASH_WLOCK(pcbinfo);
1169                 in6_pcbdisconnect(inp);
1170                 inp->in6p_laddr = in6addr_any;
1171                 INP_HASH_WUNLOCK(pcbinfo);
1172                 soisdisconnected(so);
1173         }
1174         INP_WUNLOCK(inp);
1175 }
1176
1177 static int
1178 udp6_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1179 {
1180 #ifdef INET
1181         struct epoch_tracker et;
1182 #endif
1183         struct inpcb *inp;
1184         struct inpcbinfo *pcbinfo;
1185         struct sockaddr_in6 *sin6;
1186         int error;
1187         u_char vflagsav;
1188
1189         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1190         inp = sotoinpcb(so);
1191         sin6 = (struct sockaddr_in6 *)nam;
1192         KASSERT(inp != NULL, ("udp6_connect: inp == NULL"));
1193
1194         /*
1195          * XXXRW: Need to clarify locking of v4/v6 flags.
1196          */
1197         INP_WLOCK(inp);
1198 #ifdef INET
1199         if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
1200                 struct sockaddr_in sin;
1201
1202                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
1203                         error = EINVAL;
1204                         goto out;
1205                 }
1206                 if ((inp->inp_vflag & INP_IPV4) == 0) {
1207                         error = EAFNOSUPPORT;
1208                         goto out;
1209                 }
1210                 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1211                         error = EISCONN;
1212                         goto out;
1213                 }
1214                 in6_sin6_2_sin(&sin, sin6);
1215                 error = prison_remote_ip4(td->td_ucred, &sin.sin_addr);
1216                 if (error != 0)
1217                         goto out;
1218                 vflagsav = inp->inp_vflag;
1219                 inp->inp_vflag |= INP_IPV4;
1220                 inp->inp_vflag &= ~INP_IPV6;
1221                 NET_EPOCH_ENTER(et);
1222                 INP_HASH_WLOCK(pcbinfo);
1223                 error = in_pcbconnect(inp, (struct sockaddr *)&sin,
1224                     td->td_ucred);
1225                 INP_HASH_WUNLOCK(pcbinfo);
1226                 NET_EPOCH_EXIT(et);
1227                 /*
1228                  * If connect succeeds, mark socket as connected. If
1229                  * connect fails and socket is unbound, reset inp_vflag
1230                  * field.
1231                  */
1232                 if (error == 0)
1233                         soisconnected(so);
1234                 else if (inp->inp_laddr.s_addr == INADDR_ANY &&
1235                     inp->inp_lport == 0)
1236                         inp->inp_vflag = vflagsav;
1237                 goto out;
1238         } else {
1239                 if ((inp->inp_vflag & INP_IPV6) == 0) {
1240                         error = EAFNOSUPPORT;
1241                         goto out;
1242                 }
1243         }
1244 #endif
1245         if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
1246                 error = EISCONN;
1247                 goto out;
1248         }
1249         error = prison_remote_ip6(td->td_ucred, &sin6->sin6_addr);
1250         if (error != 0)
1251                 goto out;
1252         vflagsav = inp->inp_vflag;
1253         inp->inp_vflag &= ~INP_IPV4;
1254         inp->inp_vflag |= INP_IPV6;
1255         INP_HASH_WLOCK(pcbinfo);
1256         error = in6_pcbconnect(inp, nam, td->td_ucred);
1257         INP_HASH_WUNLOCK(pcbinfo);
1258         /*
1259          * If connect succeeds, mark socket as connected. If
1260          * connect fails and socket is unbound, reset inp_vflag
1261          * field.
1262          */
1263         if (error == 0)
1264                 soisconnected(so);
1265         else if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) &&
1266             inp->inp_lport == 0)
1267                 inp->inp_vflag = vflagsav;
1268 out:
1269         INP_WUNLOCK(inp);
1270         return (error);
1271 }
1272
1273 static void
1274 udp6_detach(struct socket *so)
1275 {
1276         struct inpcb *inp;
1277         struct inpcbinfo *pcbinfo;
1278         struct udpcb *up;
1279
1280         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1281         inp = sotoinpcb(so);
1282         KASSERT(inp != NULL, ("udp6_detach: inp == NULL"));
1283
1284         INP_INFO_WLOCK(pcbinfo);
1285         INP_WLOCK(inp);
1286         up = intoudpcb(inp);
1287         KASSERT(up != NULL, ("%s: up == NULL", __func__));
1288         in_pcbdetach(inp);
1289         in_pcbfree(inp);
1290         INP_INFO_WUNLOCK(pcbinfo);
1291         udp_discardcb(up);
1292 }
1293
1294 static int
1295 udp6_disconnect(struct socket *so)
1296 {
1297         struct inpcb *inp;
1298         struct inpcbinfo *pcbinfo;
1299
1300         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1301         inp = sotoinpcb(so);
1302         KASSERT(inp != NULL, ("udp6_disconnect: inp == NULL"));
1303
1304         INP_WLOCK(inp);
1305 #ifdef INET
1306         if (inp->inp_vflag & INP_IPV4) {
1307                 struct pr_usrreqs *pru;
1308                 uint8_t nxt;
1309
1310                 nxt = (inp->inp_socket->so_proto->pr_protocol == IPPROTO_UDP) ?
1311                     IPPROTO_UDP : IPPROTO_UDPLITE;
1312                 INP_WUNLOCK(inp);
1313                 pru = inetsw[ip_protox[nxt]].pr_usrreqs;
1314                 (void)(*pru->pru_disconnect)(so);
1315                 return (0);
1316         }
1317 #endif
1318
1319         if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
1320                 INP_WUNLOCK(inp);
1321                 return (ENOTCONN);
1322         }
1323
1324         INP_HASH_WLOCK(pcbinfo);
1325         in6_pcbdisconnect(inp);
1326         inp->in6p_laddr = in6addr_any;
1327         INP_HASH_WUNLOCK(pcbinfo);
1328         SOCK_LOCK(so);
1329         so->so_state &= ~SS_ISCONNECTED;                /* XXX */
1330         SOCK_UNLOCK(so);
1331         INP_WUNLOCK(inp);
1332         return (0);
1333 }
1334
1335 static int
1336 udp6_send(struct socket *so, int flags, struct mbuf *m,
1337     struct sockaddr *addr, struct mbuf *control, struct thread *td)
1338 {
1339         int error;
1340
1341         if (addr) {
1342                 if (addr->sa_len != sizeof(struct sockaddr_in6)) {
1343                         error = EINVAL;
1344                         goto bad;
1345                 }
1346                 if (addr->sa_family != AF_INET6) {
1347                         error = EAFNOSUPPORT;
1348                         goto bad;
1349                 }
1350         }
1351
1352         return (udp6_output(so, flags, m, addr, control, td));
1353
1354 bad:
1355         if (control)
1356                 m_freem(control);
1357         m_freem(m);
1358         return (error);
1359 }
1360
1361 struct pr_usrreqs udp6_usrreqs = {
1362         .pru_abort =            udp6_abort,
1363         .pru_attach =           udp6_attach,
1364         .pru_bind =             udp6_bind,
1365         .pru_connect =          udp6_connect,
1366         .pru_control =          in6_control,
1367         .pru_detach =           udp6_detach,
1368         .pru_disconnect =       udp6_disconnect,
1369         .pru_peeraddr =         in6_mapped_peeraddr,
1370         .pru_send =             udp6_send,
1371         .pru_shutdown =         udp_shutdown,
1372         .pru_sockaddr =         in6_mapped_sockaddr,
1373         .pru_soreceive =        soreceive_dgram,
1374         .pru_sosend =           sosend_dgram,
1375         .pru_sosetlabel =       in_pcbsosetlabel,
1376         .pru_close =            udp6_close
1377 };