sys/netinet/udp_usrreq.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-3-Clause
   3  *
   4  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
   5  *      The Regents of the University of California.
   6  * Copyright (c) 2008 Robert N. M. Watson
   7  * Copyright (c) 2010-2011 Juniper Networks, Inc.
   8  * Copyright (c) 2014 Kevin Lo
   9  * All rights reserved.
  10  *
  11  * Portions of this software were developed by Robert N. M. Watson under
  12  * contract to Juniper Networks, Inc.
  13  *
  14  * Redistribution and use in source and binary forms, with or without
  15  * modification, are permitted provided that the following conditions
  16  * are met:
  17  * 1. Redistributions of source code must retain the above copyright
  18  *    notice, this list of conditions and the following disclaimer.
  19  * 2. Redistributions in binary form must reproduce the above copyright
  20  *    notice, this list of conditions and the following disclaimer in the
  21  *    documentation and/or other materials provided with the distribution.
  22  * 3. Neither the name of the University nor the names of its contributors
  23  *    may be used to endorse or promote products derived from this software
  24  *    without specific prior written permission.
  25  *
  26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  36  * SUCH DAMAGE.
  37  *
  38  *      @(#)udp_usrreq.c        8.6 (Berkeley) 5/23/95
  39  */
  40
  41 #include <sys/cdefs.h>
  42 __FBSDID("$FreeBSD$");
  43
  44 #include "opt_inet.h"
  45 #include "opt_inet6.h"
  46 #include "opt_ipsec.h"
  47 #include "opt_rss.h"
  48
  49 #include <sys/param.h>
  50 #include <sys/domain.h>
  51 #include <sys/eventhandler.h>
  52 #include <sys/jail.h>
  53 #include <sys/kernel.h>
  54 #include <sys/lock.h>
  55 #include <sys/malloc.h>
  56 #include <sys/mbuf.h>
  57 #include <sys/priv.h>
  58 #include <sys/proc.h>
  59 #include <sys/protosw.h>
  60 #include <sys/sdt.h>
  61 #include <sys/signalvar.h>
  62 #include <sys/socket.h>
  63 #include <sys/socketvar.h>
  64 #include <sys/sx.h>
  65 #include <sys/sysctl.h>
  66 #include <sys/syslog.h>
  67 #include <sys/systm.h>
  68
  69 #include <vm/uma.h>
  70
  71 #include <net/if.h>
  72 #include <net/if_var.h>
  73 #include <net/route.h>
  74 #include <net/rss_config.h>
  75
  76 #include <netinet/in.h>
  77 #include <netinet/in_kdtrace.h>
  78 #include <netinet/in_pcb.h>
  79 #include <netinet/in_systm.h>
  80 #include <netinet/in_var.h>
  81 #include <netinet/ip.h>
  82 #ifdef INET6
  83 #include <netinet/ip6.h>
  84 #endif
  85 #include <netinet/ip_icmp.h>
  86 #include <netinet/icmp_var.h>
  87 #include <netinet/ip_var.h>
  88 #include <netinet/ip_options.h>
  89 #ifdef INET6
  90 #include <netinet6/ip6_var.h>
  91 #endif
  92 #include <netinet/udp.h>
  93 #include <netinet/udp_var.h>
  94 #include <netinet/udplite.h>
  95 #include <netinet/in_rss.h>
  96
  97 #include <netipsec/ipsec_support.h>
  98
  99 #include <machine/in_cksum.h>
 100
 101 #include <security/mac/mac_framework.h>
 102
 103 /*
 104  * UDP and UDP-Lite protocols implementation.
 105  * Per RFC 768, August, 1980.
 106  * Per RFC 3828, July, 2004.
 107  */
 108
 109 /*
 110  * BSD 4.2 defaulted the udp checksum to be off.  Turning off udp checksums
 111  * removes the only data integrity mechanism for packets and malformed
 112  * packets that would otherwise be discarded due to bad checksums, and may
 113  * cause problems (especially for NFS data blocks).
 114  */
 115 VNET_DEFINE(int, udp_cksum) = 1;
 116 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_VNET | CTLFLAG_RW,
 117     &VNET_NAME(udp_cksum), 0, "compute udp checksum");
 118
 119 int     udp_log_in_vain = 0;
 120 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW,
 121     &udp_log_in_vain, 0, "Log all incoming UDP packets");
 122
 123 VNET_DEFINE(int, udp_blackhole) = 0;
 124 SYSCTL_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW,
 125     &VNET_NAME(udp_blackhole), 0,
 126     "Do not send port unreachables for refused connects");
 127
 128 u_long  udp_sendspace = 9216;           /* really max datagram size */
 129 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
 130     &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
 131
 132 u_long  udp_recvspace = 40 * (1024 +
 133 #ifdef INET6
 134                                       sizeof(struct sockaddr_in6)
 135 #else
 136                                       sizeof(struct sockaddr_in)
 137 #endif
 138                                       );        /* 40 1K datagrams */
 139
 140 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
 141     &udp_recvspace, 0, "Maximum space for incoming UDP datagrams");
 142
 143 VNET_DEFINE(struct inpcbhead, udb);             /* from udp_var.h */
 144 VNET_DEFINE(struct inpcbinfo, udbinfo);
 145 VNET_DEFINE(struct inpcbhead, ulitecb);
 146 VNET_DEFINE(struct inpcbinfo, ulitecbinfo);
 147 static VNET_DEFINE(uma_zone_t, udpcb_zone);
 148 #define V_udpcb_zone                    VNET(udpcb_zone)
 149
 150 #ifndef UDBHASHSIZE
 151 #define UDBHASHSIZE     128
 152 #endif
 153
 154 VNET_PCPUSTAT_DEFINE(struct udpstat, udpstat);          /* from udp_var.h */
 155 VNET_PCPUSTAT_SYSINIT(udpstat);
 156 SYSCTL_VNET_PCPUSTAT(_net_inet_udp, UDPCTL_STATS, stats, struct udpstat,
 157     udpstat, "UDP statistics (struct udpstat, netinet/udp_var.h)");
 158
 159 #ifdef VIMAGE
 160 VNET_PCPUSTAT_SYSUNINIT(udpstat);
 161 #endif /* VIMAGE */
 162 #ifdef INET
 163 static void     udp_detach(struct socket *so);
 164 static int      udp_output(struct inpcb *, struct mbuf *, struct sockaddr *,
 165                     struct mbuf *, struct thread *);
 166 #endif
 167
 168 static void
 169 udp_zone_change(void *tag)
 170 {
 171
 172         uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets);
 173         uma_zone_set_max(V_udpcb_zone, maxsockets);
 174 }
 175
 176 static int
 177 udp_inpcb_init(void *mem, int size, int flags)
 178 {
 179         struct inpcb *inp;
 180
 181         inp = mem;
 182         INP_LOCK_INIT(inp, "inp", "udpinp");
 183         return (0);
 184 }
 185
 186 static int
 187 udplite_inpcb_init(void *mem, int size, int flags)
 188 {
 189         struct inpcb *inp;
 190
 191         inp = mem;
 192         INP_LOCK_INIT(inp, "inp", "udpliteinp");
 193         return (0);
 194 }
 195
 196 void
 197 udp_init(void)
 198 {
 199
 200         /*
 201          * For now default to 2-tuple UDP hashing - until the fragment
 202          * reassembly code can also update the flowid.
 203          *
 204          * Once we can calculate the flowid that way and re-establish
 205          * a 4-tuple, flip this to 4-tuple.
 206          */
 207         in_pcbinfo_init(&V_udbinfo, "udp", &V_udb, UDBHASHSIZE, UDBHASHSIZE,
 208             "udp_inpcb", udp_inpcb_init, IPI_HASHFIELDS_2TUPLE);
 209         V_udpcb_zone = uma_zcreate("udpcb", sizeof(struct udpcb),
 210             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
 211         uma_zone_set_max(V_udpcb_zone, maxsockets);
 212         uma_zone_set_warning(V_udpcb_zone, "kern.ipc.maxsockets limit reached");
 213         EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL,
 214             EVENTHANDLER_PRI_ANY);
 215 }
 216
 217 void
 218 udplite_init(void)
 219 {
 220
 221         in_pcbinfo_init(&V_ulitecbinfo, "udplite", &V_ulitecb, UDBHASHSIZE,
 222             UDBHASHSIZE, "udplite_inpcb", udplite_inpcb_init,
 223             IPI_HASHFIELDS_2TUPLE);
 224 }
 225
 226 /*
 227  * Kernel module interface for updating udpstat.  The argument is an index
 228  * into udpstat treated as an array of u_long.  While this encodes the
 229  * general layout of udpstat into the caller, it doesn't encode its location,
 230  * so that future changes to add, for example, per-CPU stats support won't
 231  * cause binary compatibility problems for kernel modules.
 232  */
 233 void
 234 kmod_udpstat_inc(int statnum)
 235 {
 236
 237         counter_u64_add(VNET(udpstat)[statnum], 1);
 238 }
 239
 240 int
 241 udp_newudpcb(struct inpcb *inp)
 242 {
 243         struct udpcb *up;
 244
 245         up = uma_zalloc(V_udpcb_zone, M_NOWAIT | M_ZERO);
 246         if (up == NULL)
 247                 return (ENOBUFS);
 248         inp->inp_ppcb = up;
 249         return (0);
 250 }
 251
 252 void
 253 udp_discardcb(struct udpcb *up)
 254 {
 255
 256         uma_zfree(V_udpcb_zone, up);
 257 }
 258
 259 #ifdef VIMAGE
 260 static void
 261 udp_destroy(void *unused __unused)
 262 {
 263
 264         in_pcbinfo_destroy(&V_udbinfo);
 265         uma_zdestroy(V_udpcb_zone);
 266 }
 267 VNET_SYSUNINIT(udp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udp_destroy, NULL);
 268
 269 static void
 270 udplite_destroy(void *unused __unused)
 271 {
 272
 273         in_pcbinfo_destroy(&V_ulitecbinfo);
 274 }
 275 VNET_SYSUNINIT(udplite, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udplite_destroy,
 276     NULL);
 277 #endif
 278
 279 #ifdef INET
 280 /*
 281  * Subroutine of udp_input(), which appends the provided mbuf chain to the
 282  * passed pcb/socket.  The caller must provide a sockaddr_in via udp_in that
 283  * contains the source address.  If the socket ends up being an IPv6 socket,
 284  * udp_append() will convert to a sockaddr_in6 before passing the address
 285  * into the socket code.
 286  *
 287  * In the normal case udp_append() will return 0, indicating that you
 288  * must unlock the inp. However if a tunneling protocol is in place we increment
 289  * the inpcb refcnt and unlock the inp, on return from the tunneling protocol we
 290  * then decrement the reference count. If the inp_rele returns 1, indicating the
 291  * inp is gone, we return that to the caller to tell them *not* to unlock
 292  * the inp. In the case of multi-cast this will cause the distribution
 293  * to stop (though most tunneling protocols known currently do *not* use
 294  * multicast).
 295  */
 296 static int
 297 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off,
 298     struct sockaddr_in *udp_in)
 299 {
 300         struct sockaddr *append_sa;
 301         struct socket *so;
 302         struct mbuf *tmpopts, *opts = NULL;
 303 #ifdef INET6
 304         struct sockaddr_in6 udp_in6;
 305 #endif
 306         struct udpcb *up;
 307
 308         INP_LOCK_ASSERT(inp);
 309
 310         /*
 311          * Engage the tunneling protocol.
 312          */
 313         up = intoudpcb(inp);
 314         if (up->u_tun_func != NULL) {
 315                 in_pcbref(inp);
 316                 INP_RUNLOCK(inp);
 317                 (*up->u_tun_func)(n, off, inp, (struct sockaddr *)&udp_in[0],
 318                     up->u_tun_ctx);
 319                 INP_RLOCK(inp);
 320                 return (in_pcbrele_rlocked(inp));
 321         }
 322
 323         off += sizeof(struct udphdr);
 324
 325 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
 326         /* Check AH/ESP integrity. */
 327         if (IPSEC_ENABLED(ipv4) &&
 328             IPSEC_CHECK_POLICY(ipv4, n, inp) != 0) {
 329                 m_freem(n);
 330                 return (0);
 331         }
 332         if (up->u_flags & UF_ESPINUDP) {/* IPSec UDP encaps. */
 333                 if (IPSEC_ENABLED(ipv4) &&
 334                     UDPENCAP_INPUT(n, off, AF_INET) != 0)
 335                         return (0);     /* Consumed. */
 336         }
 337 #endif /* IPSEC */
 338 #ifdef MAC
 339         if (mac_inpcb_check_deliver(inp, n) != 0) {
 340                 m_freem(n);
 341                 return (0);
 342         }
 343 #endif /* MAC */
 344         if (inp->inp_flags & INP_CONTROLOPTS ||
 345             inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
 346 #ifdef INET6
 347                 if (inp->inp_vflag & INP_IPV6)
 348                         (void)ip6_savecontrol_v4(inp, n, &opts, NULL);
 349                 else
 350 #endif /* INET6 */
 351                         ip_savecontrol(inp, &opts, ip, n);
 352         }
 353         if ((inp->inp_vflag & INP_IPV4) && (inp->inp_flags2 & INP_ORIGDSTADDR)) {
 354                 tmpopts = sbcreatecontrol((caddr_t)&udp_in[1],
 355                         sizeof(struct sockaddr_in), IP_ORIGDSTADDR, IPPROTO_IP);
 356                 if (tmpopts) {
 357                         if (opts) {
 358                                 tmpopts->m_next = opts;
 359                                 opts = tmpopts;
 360                         } else
 361                                 opts = tmpopts;
 362                 }
 363         }
 364 #ifdef INET6
 365         if (inp->inp_vflag & INP_IPV6) {
 366                 bzero(&udp_in6, sizeof(udp_in6));
 367                 udp_in6.sin6_len = sizeof(udp_in6);
 368                 udp_in6.sin6_family = AF_INET6;
 369                 in6_sin_2_v4mapsin6(&udp_in[0], &udp_in6);
 370                 append_sa = (struct sockaddr *)&udp_in6;
 371         } else
 372 #endif /* INET6 */
 373                 append_sa = (struct sockaddr *)&udp_in[0];
 374         m_adj(n, off);
 375
 376         so = inp->inp_socket;
 377         SOCKBUF_LOCK(&so->so_rcv);
 378         if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) {
 379                 SOCKBUF_UNLOCK(&so->so_rcv);
 380                 m_freem(n);
 381                 if (opts)
 382                         m_freem(opts);
 383                 UDPSTAT_INC(udps_fullsock);
 384         } else
 385                 sorwakeup_locked(so);
 386         return (0);
 387 }
 388
 389 int
 390 udp_input(struct mbuf **mp, int *offp, int proto)
 391 {
 392         struct ip *ip;
 393         struct udphdr *uh;
 394         struct ifnet *ifp;
 395         struct inpcb *inp;
 396         uint16_t len, ip_len;
 397         struct inpcbinfo *pcbinfo;
 398         struct ip save_ip;
 399         struct sockaddr_in udp_in[2];
 400         struct mbuf *m;
 401         struct m_tag *fwd_tag;
 402         int cscov_partial, iphlen;
 403
 404         m = *mp;
 405         iphlen = *offp;
 406         ifp = m->m_pkthdr.rcvif;
 407         *mp = NULL;
 408         UDPSTAT_INC(udps_ipackets);
 409
 410         /*
 411          * Strip IP options, if any; should skip this, make available to
 412          * user, and use on returned packets, but we don't yet have a way to
 413          * check the checksum with options still present.
 414          */
 415         if (iphlen > sizeof (struct ip)) {
 416                 ip_stripoptions(m);
 417                 iphlen = sizeof(struct ip);
 418         }
 419
 420         /*
 421          * Get IP and UDP header together in first mbuf.
 422          */
 423         ip = mtod(m, struct ip *);
 424         if (m->m_len < iphlen + sizeof(struct udphdr)) {
 425                 if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == NULL) {
 426                         UDPSTAT_INC(udps_hdrops);
 427                         return (IPPROTO_DONE);
 428                 }
 429                 ip = mtod(m, struct ip *);
 430         }
 431         uh = (struct udphdr *)((caddr_t)ip + iphlen);
 432         cscov_partial = (proto == IPPROTO_UDPLITE) ? 1 : 0;
 433
 434         /*
 435          * Destination port of 0 is illegal, based on RFC768.
 436          */
 437         if (uh->uh_dport == 0)
 438                 goto badunlocked;
 439
 440         /*
 441          * Construct sockaddr format source address.  Stuff source address
 442          * and datagram in user buffer.
 443          */
 444         bzero(&udp_in[0], sizeof(struct sockaddr_in) * 2);
 445         udp_in[0].sin_len = sizeof(struct sockaddr_in);
 446         udp_in[0].sin_family = AF_INET;
 447         udp_in[0].sin_port = uh->uh_sport;
 448         udp_in[0].sin_addr = ip->ip_src;
 449         udp_in[1].sin_len = sizeof(struct sockaddr_in);
 450         udp_in[1].sin_family = AF_INET;
 451         udp_in[1].sin_port = uh->uh_dport;
 452         udp_in[1].sin_addr = ip->ip_dst;
 453
 454         /*
 455          * Make mbuf data length reflect UDP length.  If not enough data to
 456          * reflect UDP length, drop.
 457          */
 458         len = ntohs((u_short)uh->uh_ulen);
 459         ip_len = ntohs(ip->ip_len) - iphlen;
 460         if (proto == IPPROTO_UDPLITE && (len == 0 || len == ip_len)) {
 461                 /* Zero means checksum over the complete packet. */
 462                 if (len == 0)
 463                         len = ip_len;
 464                 cscov_partial = 0;
 465         }
 466         if (ip_len != len) {
 467                 if (len > ip_len || len < sizeof(struct udphdr)) {
 468                         UDPSTAT_INC(udps_badlen);
 469                         goto badunlocked;
 470                 }
 471                 if (proto == IPPROTO_UDP)
 472                         m_adj(m, len - ip_len);
 473         }
 474
 475         /*
 476          * Save a copy of the IP header in case we want restore it for
 477          * sending an ICMP error message in response.
 478          */
 479         if (!V_udp_blackhole)
 480                 save_ip = *ip;
 481         else
 482                 memset(&save_ip, 0, sizeof(save_ip));
 483
 484         /*
 485          * Checksum extended UDP header and data.
 486          */
 487         if (uh->uh_sum) {
 488                 u_short uh_sum;
 489
 490                 if ((m->m_pkthdr.csum_flags & CSUM_DATA_VALID) &&
 491                     !cscov_partial) {
 492                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
 493                                 uh_sum = m->m_pkthdr.csum_data;
 494                         else
 495                                 uh_sum = in_pseudo(ip->ip_src.s_addr,
 496                                     ip->ip_dst.s_addr, htonl((u_short)len +
 497                                     m->m_pkthdr.csum_data + proto));
 498                         uh_sum ^= 0xffff;
 499                 } else {
 500                         char b[9];
 501
 502                         bcopy(((struct ipovly *)ip)->ih_x1, b, 9);
 503                         bzero(((struct ipovly *)ip)->ih_x1, 9);
 504                         ((struct ipovly *)ip)->ih_len = (proto == IPPROTO_UDP) ?
 505                             uh->uh_ulen : htons(ip_len);
 506                         uh_sum = in_cksum(m, len + sizeof (struct ip));
 507                         bcopy(b, ((struct ipovly *)ip)->ih_x1, 9);
 508                 }
 509                 if (uh_sum) {
 510                         UDPSTAT_INC(udps_badsum);
 511                         m_freem(m);
 512                         return (IPPROTO_DONE);
 513                 }
 514         } else {
 515                 if (proto == IPPROTO_UDP) {
 516                         UDPSTAT_INC(udps_nosum);
 517                 } else {
 518                         /* UDPLite requires a checksum */
 519                         /* XXX: What is the right UDPLite MIB counter here? */
 520                         m_freem(m);
 521                         return (IPPROTO_DONE);
 522                 }
 523         }
 524
 525         pcbinfo = udp_get_inpcbinfo(proto);
 526         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
 527             in_broadcast(ip->ip_dst, ifp)) {
 528                 struct inpcb *last;
 529                 struct inpcbhead *pcblist;
 530                 struct ip_moptions *imo;
 531
 532                 INP_INFO_RLOCK(pcbinfo);
 533                 pcblist = udp_get_pcblist(proto);
 534                 last = NULL;
 535                 LIST_FOREACH(inp, pcblist, inp_list) {
 536                         if (inp->inp_lport != uh->uh_dport)
 537                                 continue;
 538 #ifdef INET6
 539                         if ((inp->inp_vflag & INP_IPV4) == 0)
 540                                 continue;
 541 #endif
 542                         if (inp->inp_laddr.s_addr != INADDR_ANY &&
 543                             inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
 544                                 continue;
 545                         if (inp->inp_faddr.s_addr != INADDR_ANY &&
 546                             inp->inp_faddr.s_addr != ip->ip_src.s_addr)
 547                                 continue;
 548                         if (inp->inp_fport != 0 &&
 549                             inp->inp_fport != uh->uh_sport)
 550                                 continue;
 551
 552                         INP_RLOCK(inp);
 553
 554                         /*
 555                          * XXXRW: Because we weren't holding either the inpcb
 556                          * or the hash lock when we checked for a match
 557                          * before, we should probably recheck now that the
 558                          * inpcb lock is held.
 559                          */
 560
 561                         /*
 562                          * Handle socket delivery policy for any-source
 563                          * and source-specific multicast. [RFC3678]
 564                          */
 565                         imo = inp->inp_moptions;
 566                         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
 567                                 struct sockaddr_in       group;
 568                                 int                      blocked;
 569                                 if (imo == NULL) {
 570                                         INP_RUNLOCK(inp);
 571                                         continue;
 572                                 }
 573                                 bzero(&group, sizeof(struct sockaddr_in));
 574                                 group.sin_len = sizeof(struct sockaddr_in);
 575                                 group.sin_family = AF_INET;
 576                                 group.sin_addr = ip->ip_dst;
 577
 578                                 blocked = imo_multi_filter(imo, ifp,
 579                                         (struct sockaddr *)&group,
 580                                         (struct sockaddr *)&udp_in[0]);
 581                                 if (blocked != MCAST_PASS) {
 582                                         if (blocked == MCAST_NOTGMEMBER)
 583                                                 IPSTAT_INC(ips_notmember);
 584                                         if (blocked == MCAST_NOTSMEMBER ||
 585                                             blocked == MCAST_MUTED)
 586                                                 UDPSTAT_INC(udps_filtermcast);
 587                                         INP_RUNLOCK(inp);
 588                                         continue;
 589                                 }
 590                         }
 591                         if (last != NULL) {
 592                                 struct mbuf *n;
 593
 594                                 if ((n = m_copym(m, 0, M_COPYALL, M_NOWAIT)) !=
 595                                     NULL) {
 596                                         UDP_PROBE(receive, NULL, last, ip,
 597                                             last, uh);
 598                                         if (udp_append(last, ip, n, iphlen,
 599                                                 udp_in)) {
 600                                                 goto inp_lost;
 601                                         }
 602                                 }
 603                                 INP_RUNLOCK(last);
 604                         }
 605                         last = inp;
 606                         /*
 607                          * Don't look for additional matches if this one does
 608                          * not have either the SO_REUSEPORT or SO_REUSEADDR
 609                          * socket options set.  This heuristic avoids
 610                          * searching through all pcbs in the common case of a
 611                          * non-shared port.  It assumes that an application
 612                          * will never clear these options after setting them.
 613                          */
 614                         if ((last->inp_socket->so_options &
 615                             (SO_REUSEPORT|SO_REUSEADDR)) == 0)
 616                                 break;
 617                 }
 618
 619                 if (last == NULL) {
 620                         /*
 621                          * No matching pcb found; discard datagram.  (No need
 622                          * to send an ICMP Port Unreachable for a broadcast
 623                          * or multicast datgram.)
 624                          */
 625                         UDPSTAT_INC(udps_noportbcast);
 626                         if (inp)
 627                                 INP_RUNLOCK(inp);
 628                         INP_INFO_RUNLOCK(pcbinfo);
 629                         goto badunlocked;
 630                 }
 631                 UDP_PROBE(receive, NULL, last, ip, last, uh);
 632                 if (udp_append(last, ip, m, iphlen, udp_in) == 0)
 633                         INP_RUNLOCK(last);
 634         inp_lost:
 635                 INP_INFO_RUNLOCK(pcbinfo);
 636                 return (IPPROTO_DONE);
 637         }
 638
 639         /*
 640          * Locate pcb for datagram.
 641          */
 642
 643         /*
 644          * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
 645          */
 646         if ((m->m_flags & M_IP_NEXTHOP) &&
 647             (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) {
 648                 struct sockaddr_in *next_hop;
 649
 650                 next_hop = (struct sockaddr_in *)(fwd_tag + 1);
 651
 652                 /*
 653                  * Transparently forwarded. Pretend to be the destination.
 654                  * Already got one like this?
 655                  */
 656                 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport,
 657                     ip->ip_dst, uh->uh_dport, INPLOOKUP_RLOCKPCB, ifp, m);
 658                 if (!inp) {
 659                         /*
 660                          * It's new.  Try to find the ambushing socket.
 661                          * Because we've rewritten the destination address,
 662                          * any hardware-generated hash is ignored.
 663                          */
 664                         inp = in_pcblookup(pcbinfo, ip->ip_src,
 665                             uh->uh_sport, next_hop->sin_addr,
 666                             next_hop->sin_port ? htons(next_hop->sin_port) :
 667                             uh->uh_dport, INPLOOKUP_WILDCARD |
 668                             INPLOOKUP_RLOCKPCB, ifp);
 669                 }
 670                 /* Remove the tag from the packet. We don't need it anymore. */
 671                 m_tag_delete(m, fwd_tag);
 672                 m->m_flags &= ~M_IP_NEXTHOP;
 673         } else
 674                 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport,
 675                     ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD |
 676                     INPLOOKUP_RLOCKPCB, ifp, m);
 677         if (inp == NULL) {
 678                 if (udp_log_in_vain) {
 679                         char src[INET_ADDRSTRLEN];
 680                         char dst[INET_ADDRSTRLEN];
 681
 682                         log(LOG_INFO,
 683                             "Connection attempt to UDP %s:%d from %s:%d\n",
 684                             inet_ntoa_r(ip->ip_dst, dst), ntohs(uh->uh_dport),
 685                             inet_ntoa_r(ip->ip_src, src), ntohs(uh->uh_sport));
 686                 }
 687                 UDPSTAT_INC(udps_noport);
 688                 if (m->m_flags & (M_BCAST | M_MCAST)) {
 689                         UDPSTAT_INC(udps_noportbcast);
 690                         goto badunlocked;
 691                 }
 692                 if (V_udp_blackhole)
 693                         goto badunlocked;
 694                 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
 695                         goto badunlocked;
 696                 *ip = save_ip;
 697                 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
 698                 return (IPPROTO_DONE);
 699         }
 700
 701         /*
 702          * Check the minimum TTL for socket.
 703          */
 704         INP_RLOCK_ASSERT(inp);
 705         if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) {
 706                 INP_RUNLOCK(inp);
 707                 m_freem(m);
 708                 return (IPPROTO_DONE);
 709         }
 710         if (cscov_partial) {
 711                 struct udpcb *up;
 712
 713                 up = intoudpcb(inp);
 714                 if (up->u_rxcslen == 0 || up->u_rxcslen > len) {
 715                         INP_RUNLOCK(inp);
 716                         m_freem(m);
 717                         return (IPPROTO_DONE);
 718                 }
 719         }
 720
 721         UDP_PROBE(receive, NULL, inp, ip, inp, uh);
 722         if (udp_append(inp, ip, m, iphlen, udp_in) == 0)
 723                 INP_RUNLOCK(inp);
 724         return (IPPROTO_DONE);
 725
 726 badunlocked:
 727         m_freem(m);
 728         return (IPPROTO_DONE);
 729 }
 730 #endif /* INET */
 731
 732 /*
 733  * Notify a udp user of an asynchronous error; just wake up so that they can
 734  * collect error status.
 735  */
 736 struct inpcb *
 737 udp_notify(struct inpcb *inp, int errno)
 738 {
 739
 740         /*
 741          * While udp_ctlinput() always calls udp_notify() with a read lock
 742          * when invoking it directly, in_pcbnotifyall() currently uses write
 743          * locks due to sharing code with TCP.  For now, accept either a read
 744          * or a write lock, but a read lock is sufficient.
 745          */
 746         INP_LOCK_ASSERT(inp);
 747         if ((errno == EHOSTUNREACH || errno == ENETUNREACH ||
 748              errno == EHOSTDOWN) && inp->inp_route.ro_rt) {
 749                 RTFREE(inp->inp_route.ro_rt);
 750                 inp->inp_route.ro_rt = (struct rtentry *)NULL;
 751         }
 752
 753         inp->inp_socket->so_error = errno;
 754         sorwakeup(inp->inp_socket);
 755         sowwakeup(inp->inp_socket);
 756         return (inp);
 757 }
 758
 759 #ifdef INET
 760 static void
 761 udp_common_ctlinput(int cmd, struct sockaddr *sa, void *vip,
 762     struct inpcbinfo *pcbinfo)
 763 {
 764         struct ip *ip = vip;
 765         struct udphdr *uh;
 766         struct in_addr faddr;
 767         struct inpcb *inp;
 768
 769         faddr = ((struct sockaddr_in *)sa)->sin_addr;
 770         if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
 771                 return;
 772
 773         if (PRC_IS_REDIRECT(cmd)) {
 774                 /* signal EHOSTDOWN, as it flushes the cached route */
 775                 in_pcbnotifyall(&V_udbinfo, faddr, EHOSTDOWN, udp_notify);
 776                 return;
 777         }
 778
 779         /*
 780          * Hostdead is ugly because it goes linearly through all PCBs.
 781          *
 782          * XXX: We never get this from ICMP, otherwise it makes an excellent
 783          * DoS attack on machines with many connections.
 784          */
 785         if (cmd == PRC_HOSTDEAD)
 786                 ip = NULL;
 787         else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
 788                 return;
 789         if (ip != NULL) {
 790                 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
 791                 inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport,
 792                     ip->ip_src, uh->uh_sport, INPLOOKUP_RLOCKPCB, NULL);
 793                 if (inp != NULL) {
 794                         INP_RLOCK_ASSERT(inp);
 795                         if (inp->inp_socket != NULL) {
 796                                 udp_notify(inp, inetctlerrmap[cmd]);
 797                         }
 798                         INP_RUNLOCK(inp);
 799                 } else {
 800                         inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport,
 801                                            ip->ip_src, uh->uh_sport,
 802                                            INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
 803                         if (inp != NULL) {
 804                                 struct udpcb *up;
 805
 806                                 up = intoudpcb(inp);
 807                                 if (up->u_icmp_func != NULL) {
 808                                         INP_RUNLOCK(inp);
 809                                         (*up->u_icmp_func)(cmd, sa, vip, up->u_tun_ctx);
 810                                 } else {
 811                                         INP_RUNLOCK(inp);
 812                                 }
 813                         }
 814                 }
 815         } else
 816                 in_pcbnotifyall(pcbinfo, faddr, inetctlerrmap[cmd],
 817                     udp_notify);
 818 }
 819 void
 820 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
 821 {
 822
 823         return (udp_common_ctlinput(cmd, sa, vip, &V_udbinfo));
 824 }
 825
 826 void
 827 udplite_ctlinput(int cmd, struct sockaddr *sa, void *vip)
 828 {
 829
 830         return (udp_common_ctlinput(cmd, sa, vip, &V_ulitecbinfo));
 831 }
 832 #endif /* INET */
 833
 834 static int
 835 udp_pcblist(SYSCTL_HANDLER_ARGS)
 836 {
 837         int error, i, n;
 838         struct inpcb *inp, **inp_list;
 839         inp_gen_t gencnt;
 840         struct xinpgen xig;
 841
 842         /*
 843          * The process of preparing the PCB list is too time-consuming and
 844          * resource-intensive to repeat twice on every request.
 845          */
 846         if (req->oldptr == 0) {
 847                 n = V_udbinfo.ipi_count;
 848                 n += imax(n / 8, 10);
 849                 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
 850                 return (0);
 851         }
 852
 853         if (req->newptr != 0)
 854                 return (EPERM);
 855
 856         /*
 857          * OK, now we're committed to doing something.
 858          */
 859         INP_INFO_RLOCK(&V_udbinfo);
 860         gencnt = V_udbinfo.ipi_gencnt;
 861         n = V_udbinfo.ipi_count;
 862         INP_INFO_RUNLOCK(&V_udbinfo);
 863
 864         error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
 865                 + n * sizeof(struct xinpcb));
 866         if (error != 0)
 867                 return (error);
 868
 869         xig.xig_len = sizeof xig;
 870         xig.xig_count = n;
 871         xig.xig_gen = gencnt;
 872         xig.xig_sogen = so_gencnt;
 873         error = SYSCTL_OUT(req, &xig, sizeof xig);
 874         if (error)
 875                 return (error);
 876
 877         inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
 878         if (inp_list == NULL)
 879                 return (ENOMEM);
 880
 881         INP_INFO_RLOCK(&V_udbinfo);
 882         for (inp = LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n;
 883              inp = LIST_NEXT(inp, inp_list)) {
 884                 INP_WLOCK(inp);
 885                 if (inp->inp_gencnt <= gencnt &&
 886                     cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
 887                         in_pcbref(inp);
 888                         inp_list[i++] = inp;
 889                 }
 890                 INP_WUNLOCK(inp);
 891         }
 892         INP_INFO_RUNLOCK(&V_udbinfo);
 893         n = i;
 894
 895         error = 0;
 896         for (i = 0; i < n; i++) {
 897                 inp = inp_list[i];
 898                 INP_RLOCK(inp);
 899                 if (inp->inp_gencnt <= gencnt) {
 900                         struct xinpcb xi;
 901
 902                         in_pcbtoxinpcb(inp, &xi);
 903                         INP_RUNLOCK(inp);
 904                         error = SYSCTL_OUT(req, &xi, sizeof xi);
 905                 } else
 906                         INP_RUNLOCK(inp);
 907         }
 908         INP_INFO_WLOCK(&V_udbinfo);
 909         for (i = 0; i < n; i++) {
 910                 inp = inp_list[i];
 911                 INP_RLOCK(inp);
 912                 if (!in_pcbrele_rlocked(inp))
 913                         INP_RUNLOCK(inp);
 914         }
 915         INP_INFO_WUNLOCK(&V_udbinfo);
 916
 917         if (!error) {
 918                 /*
 919                  * Give the user an updated idea of our state.  If the
 920                  * generation differs from what we told her before, she knows
 921                  * that something happened while we were processing this
 922                  * request, and it might be necessary to retry.
 923                  */
 924                 INP_INFO_RLOCK(&V_udbinfo);
 925                 xig.xig_gen = V_udbinfo.ipi_gencnt;
 926                 xig.xig_sogen = so_gencnt;
 927                 xig.xig_count = V_udbinfo.ipi_count;
 928                 INP_INFO_RUNLOCK(&V_udbinfo);
 929                 error = SYSCTL_OUT(req, &xig, sizeof xig);
 930         }
 931         free(inp_list, M_TEMP);
 932         return (error);
 933 }
 934
 935 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist,
 936     CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
 937     udp_pcblist, "S,xinpcb", "List of active UDP sockets");
 938
 939 #ifdef INET
 940 static int
 941 udp_getcred(SYSCTL_HANDLER_ARGS)
 942 {
 943         struct xucred xuc;
 944         struct sockaddr_in addrs[2];
 945         struct inpcb *inp;
 946         int error;
 947
 948         error = priv_check(req->td, PRIV_NETINET_GETCRED);
 949         if (error)
 950                 return (error);
 951         error = SYSCTL_IN(req, addrs, sizeof(addrs));
 952         if (error)
 953                 return (error);
 954         inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
 955             addrs[0].sin_addr, addrs[0].sin_port,
 956             INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
 957         if (inp != NULL) {
 958                 INP_RLOCK_ASSERT(inp);
 959                 if (inp->inp_socket == NULL)
 960                         error = ENOENT;
 961                 if (error == 0)
 962                         error = cr_canseeinpcb(req->td->td_ucred, inp);
 963                 if (error == 0)
 964                         cru2x(inp->inp_cred, &xuc);
 965                 INP_RUNLOCK(inp);
 966         } else
 967                 error = ENOENT;
 968         if (error == 0)
 969                 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
 970         return (error);
 971 }
 972
 973 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
 974     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
 975     udp_getcred, "S,xucred", "Get the xucred of a UDP connection");
 976 #endif /* INET */
 977
 978 int
 979 udp_ctloutput(struct socket *so, struct sockopt *sopt)
 980 {
 981         struct inpcb *inp;
 982         struct udpcb *up;
 983         int isudplite, error, optval;
 984
 985         error = 0;
 986         isudplite = (so->so_proto->pr_protocol == IPPROTO_UDPLITE) ? 1 : 0;
 987         inp = sotoinpcb(so);
 988         KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
 989         INP_WLOCK(inp);
 990         if (sopt->sopt_level != so->so_proto->pr_protocol) {
 991 #ifdef INET6
 992                 if (INP_CHECK_SOCKAF(so, AF_INET6)) {
 993                         INP_WUNLOCK(inp);
 994                         error = ip6_ctloutput(so, sopt);
 995                 }
 996 #endif
 997 #if defined(INET) && defined(INET6)
 998                 else
 999 #endif
1000 #ifdef INET
1001                 {
1002                         INP_WUNLOCK(inp);
1003                         error = ip_ctloutput(so, sopt);
1004                 }
1005 #endif
1006                 return (error);
1007         }
1008
1009         switch (sopt->sopt_dir) {
1010         case SOPT_SET:
1011                 switch (sopt->sopt_name) {
1012 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1013 #ifdef INET
1014                 case UDP_ENCAP:
1015                         if (!IPSEC_ENABLED(ipv4)) {
1016                                 INP_WUNLOCK(inp);
1017                                 return (ENOPROTOOPT);
1018                         }
1019                         error = UDPENCAP_PCBCTL(inp, sopt);
1020                         break;
1021 #endif /* INET */
1022 #endif /* IPSEC */
1023                 case UDPLITE_SEND_CSCOV:
1024                 case UDPLITE_RECV_CSCOV:
1025                         if (!isudplite) {
1026                                 INP_WUNLOCK(inp);
1027                                 error = ENOPROTOOPT;
1028                                 break;
1029                         }
1030                         INP_WUNLOCK(inp);
1031                         error = sooptcopyin(sopt, &optval, sizeof(optval),
1032                             sizeof(optval));
1033                         if (error != 0)
1034                                 break;
1035                         inp = sotoinpcb(so);
1036                         KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
1037                         INP_WLOCK(inp);
1038                         up = intoudpcb(inp);
1039                         KASSERT(up != NULL, ("%s: up == NULL", __func__));
1040                         if ((optval != 0 && optval < 8) || (optval > 65535)) {
1041                                 INP_WUNLOCK(inp);
1042                                 error = EINVAL;
1043                                 break;
1044                         }
1045                         if (sopt->sopt_name == UDPLITE_SEND_CSCOV)
1046                                 up->u_txcslen = optval;
1047                         else
1048                                 up->u_rxcslen = optval;
1049                         INP_WUNLOCK(inp);
1050                         break;
1051                 default:
1052                         INP_WUNLOCK(inp);
1053                         error = ENOPROTOOPT;
1054                         break;
1055                 }
1056                 break;
1057         case SOPT_GET:
1058                 switch (sopt->sopt_name) {
1059 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1060 #ifdef INET
1061                 case UDP_ENCAP:
1062                         if (!IPSEC_ENABLED(ipv4)) {
1063                                 INP_WUNLOCK(inp);
1064                                 return (ENOPROTOOPT);
1065                         }
1066                         error = UDPENCAP_PCBCTL(inp, sopt);
1067                         break;
1068 #endif /* INET */
1069 #endif /* IPSEC */
1070                 case UDPLITE_SEND_CSCOV:
1071                 case UDPLITE_RECV_CSCOV:
1072                         if (!isudplite) {
1073                                 INP_WUNLOCK(inp);
1074                                 error = ENOPROTOOPT;
1075                                 break;
1076                         }
1077                         up = intoudpcb(inp);
1078                         KASSERT(up != NULL, ("%s: up == NULL", __func__));
1079                         if (sopt->sopt_name == UDPLITE_SEND_CSCOV)
1080                                 optval = up->u_txcslen;
1081                         else
1082                                 optval = up->u_rxcslen;
1083                         INP_WUNLOCK(inp);
1084                         error = sooptcopyout(sopt, &optval, sizeof(optval));
1085                         break;
1086                 default:
1087                         INP_WUNLOCK(inp);
1088                         error = ENOPROTOOPT;
1089                         break;
1090                 }
1091                 break;
1092         }
1093         return (error);
1094 }
1095
1096 #ifdef INET
1097 #define UH_WLOCKED      2
1098 #define UH_RLOCKED      1
1099 #define UH_UNLOCKED     0
1100 static int
1101 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
1102     struct mbuf *control, struct thread *td)
1103 {
1104         struct udpiphdr *ui;
1105         int len = m->m_pkthdr.len;
1106         struct in_addr faddr, laddr;
1107         struct cmsghdr *cm;
1108         struct inpcbinfo *pcbinfo;
1109         struct sockaddr_in *sin, src;
1110         int cscov_partial = 0;
1111         int error = 0;
1112         int ipflags;
1113         u_short fport, lport;
1114         int unlock_udbinfo, unlock_inp;
1115         u_char tos;
1116         uint8_t pr;
1117         uint16_t cscov = 0;
1118         uint32_t flowid = 0;
1119         uint8_t flowtype = M_HASHTYPE_NONE;
1120
1121         /*
1122          * udp_output() may need to temporarily bind or connect the current
1123          * inpcb.  As such, we don't know up front whether we will need the
1124          * pcbinfo lock or not.  Do any work to decide what is needed up
1125          * front before acquiring any locks.
1126          */
1127         if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
1128                 if (control)
1129                         m_freem(control);
1130                 m_freem(m);
1131                 return (EMSGSIZE);
1132         }
1133
1134         src.sin_family = 0;
1135         sin = (struct sockaddr_in *)addr;
1136         if (sin == NULL ||
1137             (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
1138                 INP_WLOCK(inp);
1139                 unlock_inp = UH_WLOCKED;
1140         } else {
1141                 INP_RLOCK(inp);
1142                 unlock_inp = UH_RLOCKED;
1143         }
1144         tos = inp->inp_ip_tos;
1145         if (control != NULL) {
1146                 /*
1147                  * XXX: Currently, we assume all the optional information is
1148                  * stored in a single mbuf.
1149                  */
1150                 if (control->m_next) {
1151                         if (unlock_inp == UH_WLOCKED)
1152                                 INP_WUNLOCK(inp);
1153                         else
1154                                 INP_RUNLOCK(inp);
1155                         m_freem(control);
1156                         m_freem(m);
1157                         return (EINVAL);
1158                 }
1159                 for (; control->m_len > 0;
1160                     control->m_data += CMSG_ALIGN(cm->cmsg_len),
1161                     control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
1162                         cm = mtod(control, struct cmsghdr *);
1163                         if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0
1164                             || cm->cmsg_len > control->m_len) {
1165                                 error = EINVAL;
1166                                 break;
1167                         }
1168                         if (cm->cmsg_level != IPPROTO_IP)
1169                                 continue;
1170
1171                         switch (cm->cmsg_type) {
1172                         case IP_SENDSRCADDR:
1173                                 if (cm->cmsg_len !=
1174                                     CMSG_LEN(sizeof(struct in_addr))) {
1175                                         error = EINVAL;
1176                                         break;
1177                                 }
1178                                 bzero(&src, sizeof(src));
1179                                 src.sin_family = AF_INET;
1180                                 src.sin_len = sizeof(src);
1181                                 src.sin_port = inp->inp_lport;
1182                                 src.sin_addr =
1183                                     *(struct in_addr *)CMSG_DATA(cm);
1184                                 break;
1185
1186                         case IP_TOS:
1187                                 if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) {
1188                                         error = EINVAL;
1189                                         break;
1190                                 }
1191                                 tos = *(u_char *)CMSG_DATA(cm);
1192                                 break;
1193
1194                         case IP_FLOWID:
1195                                 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1196                                         error = EINVAL;
1197                                         break;
1198                                 }
1199                                 flowid = *(uint32_t *) CMSG_DATA(cm);
1200                                 break;
1201
1202                         case IP_FLOWTYPE:
1203                                 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1204                                         error = EINVAL;
1205                                         break;
1206                                 }
1207                                 flowtype = *(uint32_t *) CMSG_DATA(cm);
1208                                 break;
1209
1210 #ifdef  RSS
1211                         case IP_RSSBUCKETID:
1212                                 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1213                                         error = EINVAL;
1214                                         break;
1215                                 }
1216                                 /* This is just a placeholder for now */
1217                                 break;
1218 #endif  /* RSS */
1219                         default:
1220                                 error = ENOPROTOOPT;
1221                                 break;
1222                         }
1223                         if (error)
1224                                 break;
1225                 }
1226                 m_freem(control);
1227         }
1228         if (error) {
1229                 if (unlock_inp == UH_WLOCKED)
1230                         INP_WUNLOCK(inp);
1231                 else
1232                         INP_RUNLOCK(inp);
1233                 m_freem(m);
1234                 return (error);
1235         }
1236
1237         /*
1238          * Depending on whether or not the application has bound or connected
1239          * the socket, we may have to do varying levels of work.  The optimal
1240          * case is for a connected UDP socket, as a global lock isn't
1241          * required at all.
1242          *
1243          * In order to decide which we need, we require stability of the
1244          * inpcb binding, which we ensure by acquiring a read lock on the
1245          * inpcb.  This doesn't strictly follow the lock order, so we play
1246          * the trylock and retry game; note that we may end up with more
1247          * conservative locks than required the second time around, so later
1248          * assertions have to accept that.  Further analysis of the number of
1249          * misses under contention is required.
1250          *
1251          * XXXRW: Check that hash locking update here is correct.
1252          */
1253         pr = inp->inp_socket->so_proto->pr_protocol;
1254         pcbinfo = udp_get_inpcbinfo(pr);
1255         sin = (struct sockaddr_in *)addr;
1256         if (sin != NULL &&
1257             (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
1258                 INP_HASH_WLOCK(pcbinfo);
1259                 unlock_udbinfo = UH_WLOCKED;
1260         } else if ((sin != NULL && (
1261             (sin->sin_addr.s_addr == INADDR_ANY) ||
1262             (sin->sin_addr.s_addr == INADDR_BROADCAST) ||
1263             (inp->inp_laddr.s_addr == INADDR_ANY) ||
1264             (inp->inp_lport == 0))) ||
1265             (src.sin_family == AF_INET)) {
1266                 INP_HASH_RLOCK(pcbinfo);
1267                 unlock_udbinfo = UH_RLOCKED;
1268         } else
1269                 unlock_udbinfo = UH_UNLOCKED;
1270
1271         /*
1272          * If the IP_SENDSRCADDR control message was specified, override the
1273          * source address for this datagram.  Its use is invalidated if the
1274          * address thus specified is incomplete or clobbers other inpcbs.
1275          */
1276         laddr = inp->inp_laddr;
1277         lport = inp->inp_lport;
1278         if (src.sin_family == AF_INET) {
1279                 INP_HASH_LOCK_ASSERT(pcbinfo);
1280                 if ((lport == 0) ||
1281                     (laddr.s_addr == INADDR_ANY &&
1282                      src.sin_addr.s_addr == INADDR_ANY)) {
1283                         error = EINVAL;
1284                         goto release;
1285                 }
1286                 error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
1287                     &laddr.s_addr, &lport, td->td_ucred);
1288                 if (error)
1289                         goto release;
1290         }
1291
1292         /*
1293          * If a UDP socket has been connected, then a local address/port will
1294          * have been selected and bound.
1295          *
1296          * If a UDP socket has not been connected to, then an explicit
1297          * destination address must be used, in which case a local
1298          * address/port may not have been selected and bound.
1299          */
1300         if (sin != NULL) {
1301                 INP_LOCK_ASSERT(inp);
1302                 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1303                         error = EISCONN;
1304                         goto release;
1305                 }
1306
1307                 /*
1308                  * Jail may rewrite the destination address, so let it do
1309                  * that before we use it.
1310                  */
1311                 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1312                 if (error)
1313                         goto release;
1314
1315                 /*
1316                  * If a local address or port hasn't yet been selected, or if
1317                  * the destination address needs to be rewritten due to using
1318                  * a special INADDR_ constant, invoke in_pcbconnect_setup()
1319                  * to do the heavy lifting.  Once a port is selected, we
1320                  * commit the binding back to the socket; we also commit the
1321                  * binding of the address if in jail.
1322                  *
1323                  * If we already have a valid binding and we're not
1324                  * requesting a destination address rewrite, use a fast path.
1325                  */
1326                 if (inp->inp_laddr.s_addr == INADDR_ANY ||
1327                     inp->inp_lport == 0 ||
1328                     sin->sin_addr.s_addr == INADDR_ANY ||
1329                     sin->sin_addr.s_addr == INADDR_BROADCAST) {
1330                         INP_HASH_LOCK_ASSERT(pcbinfo);
1331                         error = in_pcbconnect_setup(inp, addr, &laddr.s_addr,
1332                             &lport, &faddr.s_addr, &fport, NULL,
1333                             td->td_ucred);
1334                         if (error)
1335                                 goto release;
1336
1337                         /*
1338                          * XXXRW: Why not commit the port if the address is
1339                          * !INADDR_ANY?
1340                          */
1341                         /* Commit the local port if newly assigned. */
1342                         if (inp->inp_laddr.s_addr == INADDR_ANY &&
1343                             inp->inp_lport == 0) {
1344                                 INP_WLOCK_ASSERT(inp);
1345                                 INP_HASH_WLOCK_ASSERT(pcbinfo);
1346                                 /*
1347                                  * Remember addr if jailed, to prevent
1348                                  * rebinding.
1349                                  */
1350                                 if (prison_flag(td->td_ucred, PR_IP4))
1351                                         inp->inp_laddr = laddr;
1352                                 inp->inp_lport = lport;
1353                                 if (in_pcbinshash(inp) != 0) {
1354                                         inp->inp_lport = 0;
1355                                         error = EAGAIN;
1356                                         goto release;
1357                                 }
1358                                 inp->inp_flags |= INP_ANONPORT;
1359                         }
1360                 } else {
1361                         faddr = sin->sin_addr;
1362                         fport = sin->sin_port;
1363                 }
1364         } else {
1365                 INP_LOCK_ASSERT(inp);
1366                 faddr = inp->inp_faddr;
1367                 fport = inp->inp_fport;
1368                 if (faddr.s_addr == INADDR_ANY) {
1369                         error = ENOTCONN;
1370                         goto release;
1371                 }
1372         }
1373
1374         /*
1375          * Calculate data length and get a mbuf for UDP, IP, and possible
1376          * link-layer headers.  Immediate slide the data pointer back forward
1377          * since we won't use that space at this layer.
1378          */
1379         M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT);
1380         if (m == NULL) {
1381                 error = ENOBUFS;
1382                 goto release;
1383         }
1384         m->m_data += max_linkhdr;
1385         m->m_len -= max_linkhdr;
1386         m->m_pkthdr.len -= max_linkhdr;
1387
1388         /*
1389          * Fill in mbuf with extended UDP header and addresses and length put
1390          * into network format.
1391          */
1392         ui = mtod(m, struct udpiphdr *);
1393         bzero(ui->ui_x1, sizeof(ui->ui_x1));    /* XXX still needed? */
1394         ui->ui_pr = pr;
1395         ui->ui_src = laddr;
1396         ui->ui_dst = faddr;
1397         ui->ui_sport = lport;
1398         ui->ui_dport = fport;
1399         ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
1400         if (pr == IPPROTO_UDPLITE) {
1401                 struct udpcb *up;
1402                 uint16_t plen;
1403
1404                 up = intoudpcb(inp);
1405                 cscov = up->u_txcslen;
1406                 plen = (u_short)len + sizeof(struct udphdr);
1407                 if (cscov >= plen)
1408                         cscov = 0;
1409                 ui->ui_len = htons(plen);
1410                 ui->ui_ulen = htons(cscov);
1411                 /*
1412                  * For UDP-Lite, checksum coverage length of zero means
1413                  * the entire UDPLite packet is covered by the checksum.
1414                  */
1415                 cscov_partial = (cscov == 0) ? 0 : 1;
1416         } else
1417                 ui->ui_v = IPVERSION << 4;
1418
1419         /*
1420          * Set the Don't Fragment bit in the IP header.
1421          */
1422         if (inp->inp_flags & INP_DONTFRAG) {
1423                 struct ip *ip;
1424
1425                 ip = (struct ip *)&ui->ui_i;
1426                 ip->ip_off |= htons(IP_DF);
1427         }
1428
1429         ipflags = 0;
1430         if (inp->inp_socket->so_options & SO_DONTROUTE)
1431                 ipflags |= IP_ROUTETOIF;
1432         if (inp->inp_socket->so_options & SO_BROADCAST)
1433                 ipflags |= IP_ALLOWBROADCAST;
1434         if (inp->inp_flags & INP_ONESBCAST)
1435                 ipflags |= IP_SENDONES;
1436
1437 #ifdef MAC
1438         mac_inpcb_create_mbuf(inp, m);
1439 #endif
1440
1441         /*
1442          * Set up checksum and output datagram.
1443          */
1444         ui->ui_sum = 0;
1445         if (pr == IPPROTO_UDPLITE) {
1446                 if (inp->inp_flags & INP_ONESBCAST)
1447                         faddr.s_addr = INADDR_BROADCAST;
1448                 if (cscov_partial) {
1449                         if ((ui->ui_sum = in_cksum(m, sizeof(struct ip) + cscov)) == 0)
1450                                 ui->ui_sum = 0xffff;
1451                 } else {
1452                         if ((ui->ui_sum = in_cksum(m, sizeof(struct udpiphdr) + len)) == 0)
1453                                 ui->ui_sum = 0xffff;
1454                 }
1455         } else if (V_udp_cksum) {
1456                 if (inp->inp_flags & INP_ONESBCAST)
1457                         faddr.s_addr = INADDR_BROADCAST;
1458                 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
1459                     htons((u_short)len + sizeof(struct udphdr) + pr));
1460                 m->m_pkthdr.csum_flags = CSUM_UDP;
1461                 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1462         }
1463         ((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len);
1464         ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl;    /* XXX */
1465         ((struct ip *)ui)->ip_tos = tos;                /* XXX */
1466         UDPSTAT_INC(udps_opackets);
1467
1468         /*
1469          * Setup flowid / RSS information for outbound socket.
1470          *
1471          * Once the UDP code decides to set a flowid some other way,
1472          * this allows the flowid to be overridden by userland.
1473          */
1474         if (flowtype != M_HASHTYPE_NONE) {
1475                 m->m_pkthdr.flowid = flowid;
1476                 M_HASHTYPE_SET(m, flowtype);
1477 #ifdef  RSS
1478         } else {
1479                 uint32_t hash_val, hash_type;
1480                 /*
1481                  * Calculate an appropriate RSS hash for UDP and
1482                  * UDP Lite.
1483                  *
1484                  * The called function will take care of figuring out
1485                  * whether a 2-tuple or 4-tuple hash is required based
1486                  * on the currently configured scheme.
1487                  *
1488                  * Later later on connected socket values should be
1489                  * cached in the inpcb and reused, rather than constantly
1490                  * re-calculating it.
1491                  *
1492                  * UDP Lite is a different protocol number and will
1493                  * likely end up being hashed as a 2-tuple until
1494                  * RSS / NICs grow UDP Lite protocol awareness.
1495                  */
1496                 if (rss_proto_software_hash_v4(faddr, laddr, fport, lport,
1497                     pr, &hash_val, &hash_type) == 0) {
1498                         m->m_pkthdr.flowid = hash_val;
1499                         M_HASHTYPE_SET(m, hash_type);
1500                 }
1501 #endif
1502         }
1503
1504 #ifdef  RSS
1505         /*
1506          * Don't override with the inp cached flowid value.
1507          *
1508          * Depending upon the kind of send being done, the inp
1509          * flowid/flowtype values may actually not be appropriate
1510          * for this particular socket send.
1511          *
1512          * We should either leave the flowid at zero (which is what is
1513          * currently done) or set it to some software generated
1514          * hash value based on the packet contents.
1515          */
1516         ipflags |= IP_NODEFAULTFLOWID;
1517 #endif  /* RSS */
1518
1519         if (unlock_udbinfo == UH_WLOCKED)
1520                 INP_HASH_WUNLOCK(pcbinfo);
1521         else if (unlock_udbinfo == UH_RLOCKED)
1522                 INP_HASH_RUNLOCK(pcbinfo);
1523         UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u);
1524         error = ip_output(m, inp->inp_options,
1525             (unlock_inp == UH_WLOCKED ? &inp->inp_route : NULL), ipflags,
1526             inp->inp_moptions, inp);
1527         if (unlock_inp == UH_WLOCKED)
1528                 INP_WUNLOCK(inp);
1529         else
1530                 INP_RUNLOCK(inp);
1531         return (error);
1532
1533 release:
1534         if (unlock_udbinfo == UH_WLOCKED) {
1535                 KASSERT(unlock_inp == UH_WLOCKED,
1536                     ("%s: excl udbinfo lock, shared inp lock", __func__));
1537                 INP_HASH_WUNLOCK(pcbinfo);
1538                 INP_WUNLOCK(inp);
1539         } else if (unlock_udbinfo == UH_RLOCKED) {
1540                 KASSERT(unlock_inp == UH_RLOCKED,
1541                     ("%s: shared udbinfo lock, excl inp lock", __func__));
1542                 INP_HASH_RUNLOCK(pcbinfo);
1543                 INP_RUNLOCK(inp);
1544         } else if (unlock_inp == UH_WLOCKED)
1545                 INP_WUNLOCK(inp);
1546         else
1547                 INP_RUNLOCK(inp);
1548         m_freem(m);
1549         return (error);
1550 }
1551
1552 static void
1553 udp_abort(struct socket *so)
1554 {
1555         struct inpcb *inp;
1556         struct inpcbinfo *pcbinfo;
1557
1558         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1559         inp = sotoinpcb(so);
1560         KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
1561         INP_WLOCK(inp);
1562         if (inp->inp_faddr.s_addr != INADDR_ANY) {
1563                 INP_HASH_WLOCK(pcbinfo);
1564                 in_pcbdisconnect(inp);
1565                 inp->inp_laddr.s_addr = INADDR_ANY;
1566                 INP_HASH_WUNLOCK(pcbinfo);
1567                 soisdisconnected(so);
1568         }
1569         INP_WUNLOCK(inp);
1570 }
1571
1572 static int
1573 udp_attach(struct socket *so, int proto, struct thread *td)
1574 {
1575         struct inpcb *inp;
1576         struct inpcbinfo *pcbinfo;
1577         int error;
1578
1579         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1580         inp = sotoinpcb(so);
1581         KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
1582         error = soreserve(so, udp_sendspace, udp_recvspace);
1583         if (error)
1584                 return (error);
1585         INP_INFO_WLOCK(pcbinfo);
1586         error = in_pcballoc(so, pcbinfo);
1587         if (error) {
1588                 INP_INFO_WUNLOCK(pcbinfo);
1589                 return (error);
1590         }
1591
1592         inp = sotoinpcb(so);
1593         inp->inp_vflag |= INP_IPV4;
1594         inp->inp_ip_ttl = V_ip_defttl;
1595
1596         error = udp_newudpcb(inp);
1597         if (error) {
1598                 in_pcbdetach(inp);
1599                 in_pcbfree(inp);
1600                 INP_INFO_WUNLOCK(pcbinfo);
1601                 return (error);
1602         }
1603
1604         INP_WUNLOCK(inp);
1605         INP_INFO_WUNLOCK(pcbinfo);
1606         return (0);
1607 }
1608 #endif /* INET */
1609
1610 int
1611 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f, udp_tun_icmp_t i, void *ctx)
1612 {
1613         struct inpcb *inp;
1614         struct udpcb *up;
1615
1616         KASSERT(so->so_type == SOCK_DGRAM,
1617             ("udp_set_kernel_tunneling: !dgram"));
1618         inp = sotoinpcb(so);
1619         KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL"));
1620         INP_WLOCK(inp);
1621         up = intoudpcb(inp);
1622         if ((up->u_tun_func != NULL) ||
1623             (up->u_icmp_func != NULL)) {
1624                 INP_WUNLOCK(inp);
1625                 return (EBUSY);
1626         }
1627         up->u_tun_func = f;
1628         up->u_icmp_func = i;
1629         up->u_tun_ctx = ctx;
1630         INP_WUNLOCK(inp);
1631         return (0);
1632 }
1633
1634 #ifdef INET
1635 static int
1636 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1637 {
1638         struct inpcb *inp;
1639         struct inpcbinfo *pcbinfo;
1640         int error;
1641
1642         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1643         inp = sotoinpcb(so);
1644         KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1645         INP_WLOCK(inp);
1646         INP_HASH_WLOCK(pcbinfo);
1647         error = in_pcbbind(inp, nam, td->td_ucred);
1648         INP_HASH_WUNLOCK(pcbinfo);
1649         INP_WUNLOCK(inp);
1650         return (error);
1651 }
1652
1653 static void
1654 udp_close(struct socket *so)
1655 {
1656         struct inpcb *inp;
1657         struct inpcbinfo *pcbinfo;
1658
1659         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1660         inp = sotoinpcb(so);
1661         KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1662         INP_WLOCK(inp);
1663         if (inp->inp_faddr.s_addr != INADDR_ANY) {
1664                 INP_HASH_WLOCK(pcbinfo);
1665                 in_pcbdisconnect(inp);
1666                 inp->inp_laddr.s_addr = INADDR_ANY;
1667                 INP_HASH_WUNLOCK(pcbinfo);
1668                 soisdisconnected(so);
1669         }
1670         INP_WUNLOCK(inp);
1671 }
1672
1673 static int
1674 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1675 {
1676         struct inpcb *inp;
1677         struct inpcbinfo *pcbinfo;
1678         struct sockaddr_in *sin;
1679         int error;
1680
1681         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1682         inp = sotoinpcb(so);
1683         KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1684         INP_WLOCK(inp);
1685         if (inp->inp_faddr.s_addr != INADDR_ANY) {
1686                 INP_WUNLOCK(inp);
1687                 return (EISCONN);
1688         }
1689         sin = (struct sockaddr_in *)nam;
1690         error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1691         if (error != 0) {
1692                 INP_WUNLOCK(inp);
1693                 return (error);
1694         }
1695         INP_HASH_WLOCK(pcbinfo);
1696         error = in_pcbconnect(inp, nam, td->td_ucred);
1697         INP_HASH_WUNLOCK(pcbinfo);
1698         if (error == 0)
1699                 soisconnected(so);
1700         INP_WUNLOCK(inp);
1701         return (error);
1702 }
1703
1704 static void
1705 udp_detach(struct socket *so)
1706 {
1707         struct inpcb *inp;
1708         struct inpcbinfo *pcbinfo;
1709         struct udpcb *up;
1710
1711         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1712         inp = sotoinpcb(so);
1713         KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1714         KASSERT(inp->inp_faddr.s_addr == INADDR_ANY,
1715             ("udp_detach: not disconnected"));
1716         INP_INFO_WLOCK(pcbinfo);
1717         INP_WLOCK(inp);
1718         up = intoudpcb(inp);
1719         KASSERT(up != NULL, ("%s: up == NULL", __func__));
1720         inp->inp_ppcb = NULL;
1721         in_pcbdetach(inp);
1722         in_pcbfree(inp);
1723         INP_INFO_WUNLOCK(pcbinfo);
1724         udp_discardcb(up);
1725 }
1726
1727 static int
1728 udp_disconnect(struct socket *so)
1729 {
1730         struct inpcb *inp;
1731         struct inpcbinfo *pcbinfo;
1732
1733         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1734         inp = sotoinpcb(so);
1735         KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1736         INP_WLOCK(inp);
1737         if (inp->inp_faddr.s_addr == INADDR_ANY) {
1738                 INP_WUNLOCK(inp);
1739                 return (ENOTCONN);
1740         }
1741         INP_HASH_WLOCK(pcbinfo);
1742         in_pcbdisconnect(inp);
1743         inp->inp_laddr.s_addr = INADDR_ANY;
1744         INP_HASH_WUNLOCK(pcbinfo);
1745         SOCK_LOCK(so);
1746         so->so_state &= ~SS_ISCONNECTED;                /* XXX */
1747         SOCK_UNLOCK(so);
1748         INP_WUNLOCK(inp);
1749         return (0);
1750 }
1751
1752 static int
1753 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1754     struct mbuf *control, struct thread *td)
1755 {
1756         struct inpcb *inp;
1757
1758         inp = sotoinpcb(so);
1759         KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1760         return (udp_output(inp, m, addr, control, td));
1761 }
1762 #endif /* INET */
1763
1764 int
1765 udp_shutdown(struct socket *so)
1766 {
1767         struct inpcb *inp;
1768
1769         inp = sotoinpcb(so);
1770         KASSERT(inp != NULL, ("udp_shutdown: inp == NULL"));
1771         INP_WLOCK(inp);
1772         socantsendmore(so);
1773         INP_WUNLOCK(inp);
1774         return (0);
1775 }
1776
1777 #ifdef INET
1778 struct pr_usrreqs udp_usrreqs = {
1779         .pru_abort =            udp_abort,
1780         .pru_attach =           udp_attach,
1781         .pru_bind =             udp_bind,
1782         .pru_connect =          udp_connect,
1783         .pru_control =          in_control,
1784         .pru_detach =           udp_detach,
1785         .pru_disconnect =       udp_disconnect,
1786         .pru_peeraddr =         in_getpeeraddr,
1787         .pru_send =             udp_send,
1788         .pru_soreceive =        soreceive_dgram,
1789         .pru_sosend =           sosend_dgram,
1790         .pru_shutdown =         udp_shutdown,
1791         .pru_sockaddr =         in_getsockaddr,
1792         .pru_sosetlabel =       in_pcbsosetlabel,
1793         .pru_close =            udp_close,
1794 };
1795 #endif /* INET */