sys/netinet/udp_usrreq.c

   1 /*-
   2  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
   3  *      The Regents of the University of California.
   4  * Copyright (c) 2008 Robert N. M. Watson
   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  * 4. Neither the name of the University 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 REGENTS 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 REGENTS 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  *      @(#)udp_usrreq.c        8.6 (Berkeley) 5/23/95
  37  */
  38
  39 #include <sys/cdefs.h>
  40 __FBSDID("$FreeBSD$");
  41
  42 #include "opt_ipfw.h"
  43 #include "opt_inet.h"
  44 #include "opt_inet6.h"
  45 #include "opt_ipsec.h"
  46 #include "opt_rss.h"
  47
  48 #include <sys/param.h>
  49 #include <sys/domain.h>
  50 #include <sys/eventhandler.h>
  51 #include <sys/jail.h>
  52 #include <sys/kernel.h>
  53 #include <sys/lock.h>
  54 #include <sys/malloc.h>
  55 #include <sys/mbuf.h>
  56 #include <sys/priv.h>
  57 #include <sys/proc.h>
  58 #include <sys/protosw.h>
  59 #include <sys/sdt.h>
  60 #include <sys/signalvar.h>
  61 #include <sys/socket.h>
  62 #include <sys/socketvar.h>
  63 #include <sys/sx.h>
  64 #include <sys/sysctl.h>
  65 #include <sys/syslog.h>
  66 #include <sys/systm.h>
  67
  68 #include <vm/uma.h>
  69
  70 #include <net/if.h>
  71 #include <net/if_var.h>
  72 #include <net/route.h>
  73 #include <net/rss_config.h>
  74
  75 #include <netinet/in.h>
  76 #include <netinet/in_kdtrace.h>
  77 #include <netinet/in_pcb.h>
  78 #include <netinet/in_systm.h>
  79 #include <netinet/in_var.h>
  80 #include <netinet/ip.h>
  81 #ifdef INET6
  82 #include <netinet/ip6.h>
  83 #endif
  84 #include <netinet/ip_icmp.h>
  85 #include <netinet/icmp_var.h>
  86 #include <netinet/ip_var.h>
  87 #include <netinet/ip_options.h>
  88 #ifdef INET6
  89 #include <netinet6/ip6_var.h>
  90 #endif
  91 #include <netinet/udp.h>
  92 #include <netinet/udp_var.h>
  93 #include <netinet/udplite.h>
  94 #include <netinet/in_rss.h>
  95
  96 #ifdef IPSEC
  97 #include <netipsec/ipsec.h>
  98 #include <netipsec/esp.h>
  99 #endif
 100
 101 #include <machine/in_cksum.h>
 102
 103 #include <security/mac/mac_framework.h>
 104
 105 /*
 106  * UDP and UDP-Lite protocols implementation.
 107  * Per RFC 768, August, 1980.
 108  * Per RFC 3828, July, 2004.
 109  */
 110
 111 /*
 112  * BSD 4.2 defaulted the udp checksum to be off.  Turning off udp checksums
 113  * removes the only data integrity mechanism for packets and malformed
 114  * packets that would otherwise be discarded due to bad checksums, and may
 115  * cause problems (especially for NFS data blocks).
 116  */
 117 VNET_DEFINE(int, udp_cksum) = 1;
 118 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_VNET | CTLFLAG_RW,
 119     &VNET_NAME(udp_cksum), 0, "compute udp checksum");
 120
 121 int     udp_log_in_vain = 0;
 122 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW,
 123     &udp_log_in_vain, 0, "Log all incoming UDP packets");
 124
 125 VNET_DEFINE(int, udp_blackhole) = 0;
 126 SYSCTL_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW,
 127     &VNET_NAME(udp_blackhole), 0,
 128     "Do not send port unreachables for refused connects");
 129
 130 u_long  udp_sendspace = 9216;           /* really max datagram size */
 131 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
 132     &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
 133
 134 u_long  udp_recvspace = 40 * (1024 +
 135 #ifdef INET6
 136                                       sizeof(struct sockaddr_in6)
 137 #else
 138                                       sizeof(struct sockaddr_in)
 139 #endif
 140                                       );        /* 40 1K datagrams */
 141
 142 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
 143     &udp_recvspace, 0, "Maximum space for incoming UDP datagrams");
 144
 145 VNET_DEFINE(struct inpcbhead, udb);             /* from udp_var.h */
 146 VNET_DEFINE(struct inpcbinfo, udbinfo);
 147 VNET_DEFINE(struct inpcbhead, ulitecb);
 148 VNET_DEFINE(struct inpcbinfo, ulitecbinfo);
 149 static VNET_DEFINE(uma_zone_t, udpcb_zone);
 150 #define V_udpcb_zone                    VNET(udpcb_zone)
 151
 152 #ifndef UDBHASHSIZE
 153 #define UDBHASHSIZE     128
 154 #endif
 155
 156 VNET_PCPUSTAT_DEFINE(struct udpstat, udpstat);          /* from udp_var.h */
 157 VNET_PCPUSTAT_SYSINIT(udpstat);
 158 SYSCTL_VNET_PCPUSTAT(_net_inet_udp, UDPCTL_STATS, stats, struct udpstat,
 159     udpstat, "UDP statistics (struct udpstat, netinet/udp_var.h)");
 160
 161 #ifdef VIMAGE
 162 VNET_PCPUSTAT_SYSUNINIT(udpstat);
 163 #endif /* VIMAGE */
 164 #ifdef INET
 165 static void     udp_detach(struct socket *so);
 166 static int      udp_output(struct inpcb *, struct mbuf *, struct sockaddr *,
 167                     struct mbuf *, struct thread *);
 168 #endif
 169
 170 #ifdef IPSEC
 171 #ifdef IPSEC_NAT_T
 172 #define UF_ESPINUDP_ALL (UF_ESPINUDP_NON_IKE|UF_ESPINUDP)
 173 #ifdef INET
 174 static struct mbuf *udp4_espdecap(struct inpcb *, struct mbuf *, int);
 175 #endif
 176 #endif /* IPSEC_NAT_T */
 177 #endif /* IPSEC */
 178
 179 static void
 180 udp_zone_change(void *tag)
 181 {
 182
 183         uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets);
 184         uma_zone_set_max(V_udpcb_zone, maxsockets);
 185 }
 186
 187 static int
 188 udp_inpcb_init(void *mem, int size, int flags)
 189 {
 190         struct inpcb *inp;
 191
 192         inp = mem;
 193         INP_LOCK_INIT(inp, "inp", "udpinp");
 194         return (0);
 195 }
 196
 197 static int
 198 udplite_inpcb_init(void *mem, int size, int flags)
 199 {
 200         struct inpcb *inp;
 201
 202         inp = mem;
 203         INP_LOCK_INIT(inp, "inp", "udpliteinp");
 204         return (0);
 205 }
 206
 207 void
 208 udp_init(void)
 209 {
 210
 211         /*
 212          * For now default to 2-tuple UDP hashing - until the fragment
 213          * reassembly code can also update the flowid.
 214          *
 215          * Once we can calculate the flowid that way and re-establish
 216          * a 4-tuple, flip this to 4-tuple.
 217          */
 218         in_pcbinfo_init(&V_udbinfo, "udp", &V_udb, UDBHASHSIZE, UDBHASHSIZE,
 219             "udp_inpcb", udp_inpcb_init, NULL, 0,
 220             IPI_HASHFIELDS_2TUPLE);
 221         V_udpcb_zone = uma_zcreate("udpcb", sizeof(struct udpcb),
 222             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
 223         uma_zone_set_max(V_udpcb_zone, maxsockets);
 224         uma_zone_set_warning(V_udpcb_zone, "kern.ipc.maxsockets limit reached");
 225         EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL,
 226             EVENTHANDLER_PRI_ANY);
 227 }
 228
 229 void
 230 udplite_init(void)
 231 {
 232
 233         in_pcbinfo_init(&V_ulitecbinfo, "udplite", &V_ulitecb, UDBHASHSIZE,
 234             UDBHASHSIZE, "udplite_inpcb", udplite_inpcb_init, NULL,
 235             0, IPI_HASHFIELDS_2TUPLE);
 236 }
 237
 238 /*
 239  * Kernel module interface for updating udpstat.  The argument is an index
 240  * into udpstat treated as an array of u_long.  While this encodes the
 241  * general layout of udpstat into the caller, it doesn't encode its location,
 242  * so that future changes to add, for example, per-CPU stats support won't
 243  * cause binary compatibility problems for kernel modules.
 244  */
 245 void
 246 kmod_udpstat_inc(int statnum)
 247 {
 248
 249         counter_u64_add(VNET(udpstat)[statnum], 1);
 250 }
 251
 252 int
 253 udp_newudpcb(struct inpcb *inp)
 254 {
 255         struct udpcb *up;
 256
 257         up = uma_zalloc(V_udpcb_zone, M_NOWAIT | M_ZERO);
 258         if (up == NULL)
 259                 return (ENOBUFS);
 260         inp->inp_ppcb = up;
 261         return (0);
 262 }
 263
 264 void
 265 udp_discardcb(struct udpcb *up)
 266 {
 267
 268         uma_zfree(V_udpcb_zone, up);
 269 }
 270
 271 #ifdef VIMAGE
 272 void
 273 udp_destroy(void)
 274 {
 275
 276         in_pcbinfo_destroy(&V_udbinfo);
 277         uma_zdestroy(V_udpcb_zone);
 278 }
 279
 280 void
 281 udplite_destroy(void)
 282 {
 283
 284         in_pcbinfo_destroy(&V_ulitecbinfo);
 285 }
 286 #endif
 287
 288 #ifdef INET
 289 /*
 290  * Subroutine of udp_input(), which appends the provided mbuf chain to the
 291  * passed pcb/socket.  The caller must provide a sockaddr_in via udp_in that
 292  * contains the source address.  If the socket ends up being an IPv6 socket,
 293  * udp_append() will convert to a sockaddr_in6 before passing the address
 294  * into the socket code.
 295  *
 296  * In the normal case udp_append() will return 0, indicating that you
 297  * must unlock the inp. However if a tunneling protocol is in place we increment
 298  * the inpcb refcnt and unlock the inp, on return from the tunneling protocol we
 299  * then decrement the reference count. If the inp_rele returns 1, indicating the
 300  * inp is gone, we return that to the caller to tell them *not* to unlock
 301  * the inp. In the case of multi-cast this will cause the distribution
 302  * to stop (though most tunneling protocols known currently do *not* use
 303  * multicast).
 304  */
 305 static int
 306 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off,
 307     struct sockaddr_in *udp_in)
 308 {
 309         struct sockaddr *append_sa;
 310         struct socket *so;
 311         struct mbuf *opts = 0;
 312 #ifdef INET6
 313         struct sockaddr_in6 udp_in6;
 314 #endif
 315         struct udpcb *up;
 316
 317         INP_LOCK_ASSERT(inp);
 318
 319         /*
 320          * Engage the tunneling protocol.
 321          */
 322         up = intoudpcb(inp);
 323         if (up->u_tun_func != NULL) {
 324                 in_pcbref(inp);
 325                 INP_RUNLOCK(inp);
 326                 (*up->u_tun_func)(n, off, inp, (struct sockaddr *)udp_in,
 327                     up->u_tun_ctx);
 328                 INP_RLOCK(inp);
 329                 return (in_pcbrele_rlocked(inp));
 330         }
 331
 332         off += sizeof(struct udphdr);
 333
 334 #ifdef IPSEC
 335         /* Check AH/ESP integrity. */
 336         if (ipsec4_in_reject(n, inp)) {
 337                 m_freem(n);
 338                 return (0);
 339         }
 340 #ifdef IPSEC_NAT_T
 341         up = intoudpcb(inp);
 342         KASSERT(up != NULL, ("%s: udpcb NULL", __func__));
 343         if (up->u_flags & UF_ESPINUDP_ALL) {    /* IPSec UDP encaps. */
 344                 n = udp4_espdecap(inp, n, off);
 345                 if (n == NULL)                          /* Consumed. */
 346                         return (0);
 347         }
 348 #endif /* IPSEC_NAT_T */
 349 #endif /* IPSEC */
 350 #ifdef MAC
 351         if (mac_inpcb_check_deliver(inp, n) != 0) {
 352                 m_freem(n);
 353                 return (0);
 354         }
 355 #endif /* MAC */
 356         if (inp->inp_flags & INP_CONTROLOPTS ||
 357             inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
 358 #ifdef INET6
 359                 if (inp->inp_vflag & INP_IPV6)
 360                         (void)ip6_savecontrol_v4(inp, n, &opts, NULL);
 361                 else
 362 #endif /* INET6 */
 363                         ip_savecontrol(inp, &opts, ip, n);
 364         }
 365 #ifdef INET6
 366         if (inp->inp_vflag & INP_IPV6) {
 367                 bzero(&udp_in6, sizeof(udp_in6));
 368                 udp_in6.sin6_len = sizeof(udp_in6);
 369                 udp_in6.sin6_family = AF_INET6;
 370                 in6_sin_2_v4mapsin6(udp_in, &udp_in6);
 371                 append_sa = (struct sockaddr *)&udp_in6;
 372         } else
 373 #endif /* INET6 */
 374                 append_sa = (struct sockaddr *)udp_in;
 375         m_adj(n, off);
 376
 377         so = inp->inp_socket;
 378         SOCKBUF_LOCK(&so->so_rcv);
 379         if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) {
 380                 SOCKBUF_UNLOCK(&so->so_rcv);
 381                 m_freem(n);
 382                 if (opts)
 383                         m_freem(opts);
 384                 UDPSTAT_INC(udps_fullsock);
 385         } else
 386                 sorwakeup_locked(so);
 387         return (0);
 388 }
 389
 390 int
 391 udp_input(struct mbuf **mp, int *offp, int proto)
 392 {
 393         struct ip *ip;
 394         struct udphdr *uh;
 395         struct ifnet *ifp;
 396         struct inpcb *inp;
 397         uint16_t len, ip_len;
 398         struct inpcbinfo *pcbinfo;
 399         struct ip save_ip;
 400         struct sockaddr_in udp_in;
 401         struct mbuf *m;
 402         struct m_tag *fwd_tag;
 403         int cscov_partial, iphlen;
 404
 405         m = *mp;
 406         iphlen = *offp;
 407         ifp = m->m_pkthdr.rcvif;
 408         *mp = NULL;
 409         UDPSTAT_INC(udps_ipackets);
 410
 411         /*
 412          * Strip IP options, if any; should skip this, make available to
 413          * user, and use on returned packets, but we don't yet have a way to
 414          * check the checksum with options still present.
 415          */
 416         if (iphlen > sizeof (struct ip)) {
 417                 ip_stripoptions(m);
 418                 iphlen = sizeof(struct ip);
 419         }
 420
 421         /*
 422          * Get IP and UDP header together in first mbuf.
 423          */
 424         ip = mtod(m, struct ip *);
 425         if (m->m_len < iphlen + sizeof(struct udphdr)) {
 426                 if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == NULL) {
 427                         UDPSTAT_INC(udps_hdrops);
 428                         return (IPPROTO_DONE);
 429                 }
 430                 ip = mtod(m, struct ip *);
 431         }
 432         uh = (struct udphdr *)((caddr_t)ip + iphlen);
 433         cscov_partial = (proto == IPPROTO_UDPLITE) ? 1 : 0;
 434
 435         /*
 436          * Destination port of 0 is illegal, based on RFC768.
 437          */
 438         if (uh->uh_dport == 0)
 439                 goto badunlocked;
 440
 441         /*
 442          * Construct sockaddr format source address.  Stuff source address
 443          * and datagram in user buffer.
 444          */
 445         bzero(&udp_in, sizeof(udp_in));
 446         udp_in.sin_len = sizeof(udp_in);
 447         udp_in.sin_family = AF_INET;
 448         udp_in.sin_port = uh->uh_sport;
 449         udp_in.sin_addr = ip->ip_src;
 450
 451         /*
 452          * Make mbuf data length reflect UDP length.  If not enough data to
 453          * reflect UDP length, drop.
 454          */
 455         len = ntohs((u_short)uh->uh_ulen);
 456         ip_len = ntohs(ip->ip_len) - iphlen;
 457         if (proto == IPPROTO_UDPLITE && (len == 0 || len == ip_len)) {
 458                 /* Zero means checksum over the complete packet. */
 459                 if (len == 0)
 460                         len = ip_len;
 461                 cscov_partial = 0;
 462         }
 463         if (ip_len != len) {
 464                 if (len > ip_len || len < sizeof(struct udphdr)) {
 465                         UDPSTAT_INC(udps_badlen);
 466                         goto badunlocked;
 467                 }
 468                 if (proto == IPPROTO_UDP)
 469                         m_adj(m, len - ip_len);
 470         }
 471
 472         /*
 473          * Save a copy of the IP header in case we want restore it for
 474          * sending an ICMP error message in response.
 475          */
 476         if (!V_udp_blackhole)
 477                 save_ip = *ip;
 478         else
 479                 memset(&save_ip, 0, sizeof(save_ip));
 480
 481         /*
 482          * Checksum extended UDP header and data.
 483          */
 484         if (uh->uh_sum) {
 485                 u_short uh_sum;
 486
 487                 if ((m->m_pkthdr.csum_flags & CSUM_DATA_VALID) &&
 488                     !cscov_partial) {
 489                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
 490                                 uh_sum = m->m_pkthdr.csum_data;
 491                         else
 492                                 uh_sum = in_pseudo(ip->ip_src.s_addr,
 493                                     ip->ip_dst.s_addr, htonl((u_short)len +
 494                                     m->m_pkthdr.csum_data + proto));
 495                         uh_sum ^= 0xffff;
 496                 } else {
 497                         char b[9];
 498
 499                         bcopy(((struct ipovly *)ip)->ih_x1, b, 9);
 500                         bzero(((struct ipovly *)ip)->ih_x1, 9);
 501                         ((struct ipovly *)ip)->ih_len = (proto == IPPROTO_UDP) ?
 502                             uh->uh_ulen : htons(ip_len);
 503                         uh_sum = in_cksum(m, len + sizeof (struct ip));
 504                         bcopy(b, ((struct ipovly *)ip)->ih_x1, 9);
 505                 }
 506                 if (uh_sum) {
 507                         UDPSTAT_INC(udps_badsum);
 508                         m_freem(m);
 509                         return (IPPROTO_DONE);
 510                 }
 511         } else {
 512                 if (proto == IPPROTO_UDP) {
 513                         UDPSTAT_INC(udps_nosum);
 514                 } else {
 515                         /* UDPLite requires a checksum */
 516                         /* XXX: What is the right UDPLite MIB counter here? */
 517                         m_freem(m);
 518                         return (IPPROTO_DONE);
 519                 }
 520         }
 521
 522         pcbinfo = udp_get_inpcbinfo(proto);
 523         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
 524             in_broadcast(ip->ip_dst, ifp)) {
 525                 struct inpcb *last;
 526                 struct inpcbhead *pcblist;
 527                 struct ip_moptions *imo;
 528
 529                 INP_INFO_RLOCK(pcbinfo);
 530                 pcblist = udp_get_pcblist(proto);
 531                 last = NULL;
 532                 LIST_FOREACH(inp, pcblist, inp_list) {
 533                         if (inp->inp_lport != uh->uh_dport)
 534                                 continue;
 535 #ifdef INET6
 536                         if ((inp->inp_vflag & INP_IPV4) == 0)
 537                                 continue;
 538 #endif
 539                         if (inp->inp_laddr.s_addr != INADDR_ANY &&
 540                             inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
 541                                 continue;
 542                         if (inp->inp_faddr.s_addr != INADDR_ANY &&
 543                             inp->inp_faddr.s_addr != ip->ip_src.s_addr)
 544                                 continue;
 545                         if (inp->inp_fport != 0 &&
 546                             inp->inp_fport != uh->uh_sport)
 547                                 continue;
 548
 549                         INP_RLOCK(inp);
 550
 551                         /*
 552                          * XXXRW: Because we weren't holding either the inpcb
 553                          * or the hash lock when we checked for a match
 554                          * before, we should probably recheck now that the
 555                          * inpcb lock is held.
 556                          */
 557
 558                         /*
 559                          * Handle socket delivery policy for any-source
 560                          * and source-specific multicast. [RFC3678]
 561                          */
 562                         imo = inp->inp_moptions;
 563                         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
 564                                 struct sockaddr_in       group;
 565                                 int                      blocked;
 566                                 if (imo == NULL) {
 567                                         INP_RUNLOCK(inp);
 568                                         continue;
 569                                 }
 570                                 bzero(&group, sizeof(struct sockaddr_in));
 571                                 group.sin_len = sizeof(struct sockaddr_in);
 572                                 group.sin_family = AF_INET;
 573                                 group.sin_addr = ip->ip_dst;
 574
 575                                 blocked = imo_multi_filter(imo, ifp,
 576                                         (struct sockaddr *)&group,
 577                                         (struct sockaddr *)&udp_in);
 578                                 if (blocked != MCAST_PASS) {
 579                                         if (blocked == MCAST_NOTGMEMBER)
 580                                                 IPSTAT_INC(ips_notmember);
 581                                         if (blocked == MCAST_NOTSMEMBER ||
 582                                             blocked == MCAST_MUTED)
 583                                                 UDPSTAT_INC(udps_filtermcast);
 584                                         INP_RUNLOCK(inp);
 585                                         continue;
 586                                 }
 587                         }
 588                         if (last != NULL) {
 589                                 struct mbuf *n;
 590
 591                                 if ((n = m_copy(m, 0, M_COPYALL)) != NULL) {
 592                                         UDP_PROBE(receive, NULL, last, ip,
 593                                             last, uh);
 594                                         if (udp_append(last, ip, n, iphlen,
 595                                                 &udp_in)) {
 596                                                 goto inp_lost;
 597                                         }
 598                                 }
 599                                 INP_RUNLOCK(last);
 600                         }
 601                         last = inp;
 602                         /*
 603                          * Don't look for additional matches if this one does
 604                          * not have either the SO_REUSEPORT or SO_REUSEADDR
 605                          * socket options set.  This heuristic avoids
 606                          * searching through all pcbs in the common case of a
 607                          * non-shared port.  It assumes that an application
 608                          * will never clear these options after setting them.
 609                          */
 610                         if ((last->inp_socket->so_options &
 611                             (SO_REUSEPORT|SO_REUSEADDR)) == 0)
 612                                 break;
 613                 }
 614
 615                 if (last == NULL) {
 616                         /*
 617                          * No matching pcb found; discard datagram.  (No need
 618                          * to send an ICMP Port Unreachable for a broadcast
 619                          * or multicast datgram.)
 620                          */
 621                         UDPSTAT_INC(udps_noportbcast);
 622                         if (inp)
 623                                 INP_RUNLOCK(inp);
 624                         INP_INFO_RUNLOCK(pcbinfo);
 625                         goto badunlocked;
 626                 }
 627                 UDP_PROBE(receive, NULL, last, ip, last, uh);
 628                 if (udp_append(last, ip, m, iphlen, &udp_in) == 0)
 629                         INP_RUNLOCK(last);
 630         inp_lost:
 631                 INP_INFO_RUNLOCK(pcbinfo);
 632                 return (IPPROTO_DONE);
 633         }
 634
 635         /*
 636          * Locate pcb for datagram.
 637          */
 638
 639         /*
 640          * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
 641          */
 642         if ((m->m_flags & M_IP_NEXTHOP) &&
 643             (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) {
 644                 struct sockaddr_in *next_hop;
 645
 646                 next_hop = (struct sockaddr_in *)(fwd_tag + 1);
 647
 648                 /*
 649                  * Transparently forwarded. Pretend to be the destination.
 650                  * Already got one like this?
 651                  */
 652                 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport,
 653                     ip->ip_dst, uh->uh_dport, INPLOOKUP_RLOCKPCB, ifp, m);
 654                 if (!inp) {
 655                         /*
 656                          * It's new.  Try to find the ambushing socket.
 657                          * Because we've rewritten the destination address,
 658                          * any hardware-generated hash is ignored.
 659                          */
 660                         inp = in_pcblookup(pcbinfo, ip->ip_src,
 661                             uh->uh_sport, next_hop->sin_addr,
 662                             next_hop->sin_port ? htons(next_hop->sin_port) :
 663                             uh->uh_dport, INPLOOKUP_WILDCARD |
 664                             INPLOOKUP_RLOCKPCB, ifp);
 665                 }
 666                 /* Remove the tag from the packet. We don't need it anymore. */
 667                 m_tag_delete(m, fwd_tag);
 668                 m->m_flags &= ~M_IP_NEXTHOP;
 669         } else
 670                 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport,
 671                     ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD |
 672                     INPLOOKUP_RLOCKPCB, ifp, m);
 673         if (inp == NULL) {
 674                 if (udp_log_in_vain) {
 675                         char buf[4*sizeof "123"];
 676
 677                         strcpy(buf, inet_ntoa(ip->ip_dst));
 678                         log(LOG_INFO,
 679                             "Connection attempt to UDP %s:%d from %s:%d\n",
 680                             buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src),
 681                             ntohs(uh->uh_sport));
 682                 }
 683                 UDPSTAT_INC(udps_noport);
 684                 if (m->m_flags & (M_BCAST | M_MCAST)) {
 685                         UDPSTAT_INC(udps_noportbcast);
 686                         goto badunlocked;
 687                 }
 688                 if (V_udp_blackhole)
 689                         goto badunlocked;
 690                 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
 691                         goto badunlocked;
 692                 *ip = save_ip;
 693                 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
 694                 return (IPPROTO_DONE);
 695         }
 696
 697         /*
 698          * Check the minimum TTL for socket.
 699          */
 700         INP_RLOCK_ASSERT(inp);
 701         if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) {
 702                 INP_RUNLOCK(inp);
 703                 m_freem(m);
 704                 return (IPPROTO_DONE);
 705         }
 706         if (cscov_partial) {
 707                 struct udpcb *up;
 708
 709                 up = intoudpcb(inp);
 710                 if (up->u_rxcslen == 0 || up->u_rxcslen > len) {
 711                         INP_RUNLOCK(inp);
 712                         m_freem(m);
 713                         return (IPPROTO_DONE);
 714                 }
 715         }
 716
 717         UDP_PROBE(receive, NULL, inp, ip, inp, uh);
 718         if (udp_append(inp, ip, m, iphlen, &udp_in) == 0)
 719                 INP_RUNLOCK(inp);
 720         return (IPPROTO_DONE);
 721
 722 badunlocked:
 723         m_freem(m);
 724         return (IPPROTO_DONE);
 725 }
 726 #endif /* INET */
 727
 728 /*
 729  * Notify a udp user of an asynchronous error; just wake up so that they can
 730  * collect error status.
 731  */
 732 struct inpcb *
 733 udp_notify(struct inpcb *inp, int errno)
 734 {
 735
 736         /*
 737          * While udp_ctlinput() always calls udp_notify() with a read lock
 738          * when invoking it directly, in_pcbnotifyall() currently uses write
 739          * locks due to sharing code with TCP.  For now, accept either a read
 740          * or a write lock, but a read lock is sufficient.
 741          */
 742         INP_LOCK_ASSERT(inp);
 743
 744         inp->inp_socket->so_error = errno;
 745         sorwakeup(inp->inp_socket);
 746         sowwakeup(inp->inp_socket);
 747         return (inp);
 748 }
 749
 750 #ifdef INET
 751 static void
 752 udp_common_ctlinput(int cmd, struct sockaddr *sa, void *vip,
 753     struct inpcbinfo *pcbinfo)
 754 {
 755         struct ip *ip = vip;
 756         struct udphdr *uh;
 757         struct in_addr faddr;
 758         struct inpcb *inp;
 759
 760         faddr = ((struct sockaddr_in *)sa)->sin_addr;
 761         if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
 762                 return;
 763
 764         /*
 765          * Redirects don't need to be handled up here.
 766          */
 767         if (PRC_IS_REDIRECT(cmd))
 768                 return;
 769
 770         /*
 771          * Hostdead is ugly because it goes linearly through all PCBs.
 772          *
 773          * XXX: We never get this from ICMP, otherwise it makes an excellent
 774          * DoS attack on machines with many connections.
 775          */
 776         if (cmd == PRC_HOSTDEAD)
 777                 ip = NULL;
 778         else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
 779                 return;
 780         if (ip != NULL) {
 781                 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
 782                 inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport,
 783                     ip->ip_src, uh->uh_sport, INPLOOKUP_RLOCKPCB, NULL);
 784                 if (inp != NULL) {
 785                         INP_RLOCK_ASSERT(inp);
 786                         if (inp->inp_socket != NULL) {
 787                                 udp_notify(inp, inetctlerrmap[cmd]);
 788                         }
 789                         INP_RUNLOCK(inp);
 790                 }
 791         } else
 792                 in_pcbnotifyall(pcbinfo, faddr, inetctlerrmap[cmd],
 793                     udp_notify);
 794 }
 795 void
 796 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
 797 {
 798
 799         return (udp_common_ctlinput(cmd, sa, vip, &V_udbinfo));
 800 }
 801
 802 void
 803 udplite_ctlinput(int cmd, struct sockaddr *sa, void *vip)
 804 {
 805
 806         return (udp_common_ctlinput(cmd, sa, vip, &V_ulitecbinfo));
 807 }
 808 #endif /* INET */
 809
 810 static int
 811 udp_pcblist(SYSCTL_HANDLER_ARGS)
 812 {
 813         int error, i, n;
 814         struct inpcb *inp, **inp_list;
 815         inp_gen_t gencnt;
 816         struct xinpgen xig;
 817
 818         /*
 819          * The process of preparing the PCB list is too time-consuming and
 820          * resource-intensive to repeat twice on every request.
 821          */
 822         if (req->oldptr == 0) {
 823                 n = V_udbinfo.ipi_count;
 824                 n += imax(n / 8, 10);
 825                 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
 826                 return (0);
 827         }
 828
 829         if (req->newptr != 0)
 830                 return (EPERM);
 831
 832         /*
 833          * OK, now we're committed to doing something.
 834          */
 835         INP_INFO_RLOCK(&V_udbinfo);
 836         gencnt = V_udbinfo.ipi_gencnt;
 837         n = V_udbinfo.ipi_count;
 838         INP_INFO_RUNLOCK(&V_udbinfo);
 839
 840         error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
 841                 + n * sizeof(struct xinpcb));
 842         if (error != 0)
 843                 return (error);
 844
 845         xig.xig_len = sizeof xig;
 846         xig.xig_count = n;
 847         xig.xig_gen = gencnt;
 848         xig.xig_sogen = so_gencnt;
 849         error = SYSCTL_OUT(req, &xig, sizeof xig);
 850         if (error)
 851                 return (error);
 852
 853         inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
 854         if (inp_list == 0)
 855                 return (ENOMEM);
 856
 857         INP_INFO_RLOCK(&V_udbinfo);
 858         for (inp = LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n;
 859              inp = LIST_NEXT(inp, inp_list)) {
 860                 INP_WLOCK(inp);
 861                 if (inp->inp_gencnt <= gencnt &&
 862                     cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
 863                         in_pcbref(inp);
 864                         inp_list[i++] = inp;
 865                 }
 866                 INP_WUNLOCK(inp);
 867         }
 868         INP_INFO_RUNLOCK(&V_udbinfo);
 869         n = i;
 870
 871         error = 0;
 872         for (i = 0; i < n; i++) {
 873                 inp = inp_list[i];
 874                 INP_RLOCK(inp);
 875                 if (inp->inp_gencnt <= gencnt) {
 876                         struct xinpcb xi;
 877
 878                         bzero(&xi, sizeof(xi));
 879                         xi.xi_len = sizeof xi;
 880                         /* XXX should avoid extra copy */
 881                         bcopy(inp, &xi.xi_inp, sizeof *inp);
 882                         if (inp->inp_socket)
 883                                 sotoxsocket(inp->inp_socket, &xi.xi_socket);
 884                         xi.xi_inp.inp_gencnt = inp->inp_gencnt;
 885                         INP_RUNLOCK(inp);
 886                         error = SYSCTL_OUT(req, &xi, sizeof xi);
 887                 } else
 888                         INP_RUNLOCK(inp);
 889         }
 890         INP_INFO_WLOCK(&V_udbinfo);
 891         for (i = 0; i < n; i++) {
 892                 inp = inp_list[i];
 893                 INP_RLOCK(inp);
 894                 if (!in_pcbrele_rlocked(inp))
 895                         INP_RUNLOCK(inp);
 896         }
 897         INP_INFO_WUNLOCK(&V_udbinfo);
 898
 899         if (!error) {
 900                 /*
 901                  * Give the user an updated idea of our state.  If the
 902                  * generation differs from what we told her before, she knows
 903                  * that something happened while we were processing this
 904                  * request, and it might be necessary to retry.
 905                  */
 906                 INP_INFO_RLOCK(&V_udbinfo);
 907                 xig.xig_gen = V_udbinfo.ipi_gencnt;
 908                 xig.xig_sogen = so_gencnt;
 909                 xig.xig_count = V_udbinfo.ipi_count;
 910                 INP_INFO_RUNLOCK(&V_udbinfo);
 911                 error = SYSCTL_OUT(req, &xig, sizeof xig);
 912         }
 913         free(inp_list, M_TEMP);
 914         return (error);
 915 }
 916
 917 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist,
 918     CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
 919     udp_pcblist, "S,xinpcb", "List of active UDP sockets");
 920
 921 #ifdef INET
 922 static int
 923 udp_getcred(SYSCTL_HANDLER_ARGS)
 924 {
 925         struct xucred xuc;
 926         struct sockaddr_in addrs[2];
 927         struct inpcb *inp;
 928         int error;
 929
 930         error = priv_check(req->td, PRIV_NETINET_GETCRED);
 931         if (error)
 932                 return (error);
 933         error = SYSCTL_IN(req, addrs, sizeof(addrs));
 934         if (error)
 935                 return (error);
 936         inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
 937             addrs[0].sin_addr, addrs[0].sin_port,
 938             INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
 939         if (inp != NULL) {
 940                 INP_RLOCK_ASSERT(inp);
 941                 if (inp->inp_socket == NULL)
 942                         error = ENOENT;
 943                 if (error == 0)
 944                         error = cr_canseeinpcb(req->td->td_ucred, inp);
 945                 if (error == 0)
 946                         cru2x(inp->inp_cred, &xuc);
 947                 INP_RUNLOCK(inp);
 948         } else
 949                 error = ENOENT;
 950         if (error == 0)
 951                 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
 952         return (error);
 953 }
 954
 955 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
 956     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
 957     udp_getcred, "S,xucred", "Get the xucred of a UDP connection");
 958 #endif /* INET */
 959
 960 int
 961 udp_ctloutput(struct socket *so, struct sockopt *sopt)
 962 {
 963         struct inpcb *inp;
 964         struct udpcb *up;
 965         int isudplite, error, optval;
 966
 967         error = 0;
 968         isudplite = (so->so_proto->pr_protocol == IPPROTO_UDPLITE) ? 1 : 0;
 969         inp = sotoinpcb(so);
 970         KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
 971         INP_WLOCK(inp);
 972         if (sopt->sopt_level != so->so_proto->pr_protocol) {
 973 #ifdef INET6
 974                 if (INP_CHECK_SOCKAF(so, AF_INET6)) {
 975                         INP_WUNLOCK(inp);
 976                         error = ip6_ctloutput(so, sopt);
 977                 }
 978 #endif
 979 #if defined(INET) && defined(INET6)
 980                 else
 981 #endif
 982 #ifdef INET
 983                 {
 984                         INP_WUNLOCK(inp);
 985                         error = ip_ctloutput(so, sopt);
 986                 }
 987 #endif
 988                 return (error);
 989         }
 990
 991         switch (sopt->sopt_dir) {
 992         case SOPT_SET:
 993                 switch (sopt->sopt_name) {
 994                 case UDP_ENCAP:
 995                         INP_WUNLOCK(inp);
 996                         error = sooptcopyin(sopt, &optval, sizeof optval,
 997                                             sizeof optval);
 998                         if (error)
 999                                 break;
1000                         inp = sotoinpcb(so);
1001                         KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
1002                         INP_WLOCK(inp);
1003 #ifdef IPSEC_NAT_T
1004                         up = intoudpcb(inp);
1005                         KASSERT(up != NULL, ("%s: up == NULL", __func__));
1006 #endif
1007                         switch (optval) {
1008                         case 0:
1009                                 /* Clear all UDP encap. */
1010 #ifdef IPSEC_NAT_T
1011                                 up->u_flags &= ~UF_ESPINUDP_ALL;
1012 #endif
1013                                 break;
1014 #ifdef IPSEC_NAT_T
1015                         case UDP_ENCAP_ESPINUDP:
1016                         case UDP_ENCAP_ESPINUDP_NON_IKE:
1017                                 up->u_flags &= ~UF_ESPINUDP_ALL;
1018                                 if (optval == UDP_ENCAP_ESPINUDP)
1019                                         up->u_flags |= UF_ESPINUDP;
1020                                 else if (optval == UDP_ENCAP_ESPINUDP_NON_IKE)
1021                                         up->u_flags |= UF_ESPINUDP_NON_IKE;
1022                                 break;
1023 #endif
1024                         default:
1025                                 error = EINVAL;
1026                                 break;
1027                         }
1028                         INP_WUNLOCK(inp);
1029                         break;
1030                 case UDPLITE_SEND_CSCOV:
1031                 case UDPLITE_RECV_CSCOV:
1032                         if (!isudplite) {
1033                                 INP_WUNLOCK(inp);
1034                                 error = ENOPROTOOPT;
1035                                 break;
1036                         }
1037                         INP_WUNLOCK(inp);
1038                         error = sooptcopyin(sopt, &optval, sizeof(optval),
1039                             sizeof(optval));
1040                         if (error != 0)
1041                                 break;
1042                         inp = sotoinpcb(so);
1043                         KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
1044                         INP_WLOCK(inp);
1045                         up = intoudpcb(inp);
1046                         KASSERT(up != NULL, ("%s: up == NULL", __func__));
1047                         if ((optval != 0 && optval < 8) || (optval > 65535)) {
1048                                 INP_WUNLOCK(inp);
1049                                 error = EINVAL;
1050                                 break;
1051                         }
1052                         if (sopt->sopt_name == UDPLITE_SEND_CSCOV)
1053                                 up->u_txcslen = optval;
1054                         else
1055                                 up->u_rxcslen = optval;
1056                         INP_WUNLOCK(inp);
1057                         break;
1058                 default:
1059                         INP_WUNLOCK(inp);
1060                         error = ENOPROTOOPT;
1061                         break;
1062                 }
1063                 break;
1064         case SOPT_GET:
1065                 switch (sopt->sopt_name) {
1066 #ifdef IPSEC_NAT_T
1067                 case UDP_ENCAP:
1068                         up = intoudpcb(inp);
1069                         KASSERT(up != NULL, ("%s: up == NULL", __func__));
1070                         optval = up->u_flags & UF_ESPINUDP_ALL;
1071                         INP_WUNLOCK(inp);
1072                         error = sooptcopyout(sopt, &optval, sizeof optval);
1073                         break;
1074 #endif
1075                 case UDPLITE_SEND_CSCOV:
1076                 case UDPLITE_RECV_CSCOV:
1077                         if (!isudplite) {
1078                                 INP_WUNLOCK(inp);
1079                                 error = ENOPROTOOPT;
1080                                 break;
1081                         }
1082                         up = intoudpcb(inp);
1083                         KASSERT(up != NULL, ("%s: up == NULL", __func__));
1084                         if (sopt->sopt_name == UDPLITE_SEND_CSCOV)
1085                                 optval = up->u_txcslen;
1086                         else
1087                                 optval = up->u_rxcslen;
1088                         INP_WUNLOCK(inp);
1089                         error = sooptcopyout(sopt, &optval, sizeof(optval));
1090                         break;
1091                 default:
1092                         INP_WUNLOCK(inp);
1093                         error = ENOPROTOOPT;
1094                         break;
1095                 }
1096                 break;
1097         }
1098         return (error);
1099 }
1100
1101 #ifdef INET
1102 #define UH_WLOCKED      2
1103 #define UH_RLOCKED      1
1104 #define UH_UNLOCKED     0
1105 static int
1106 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
1107     struct mbuf *control, struct thread *td)
1108 {
1109         struct udpiphdr *ui;
1110         int len = m->m_pkthdr.len;
1111         struct in_addr faddr, laddr;
1112         struct cmsghdr *cm;
1113         struct inpcbinfo *pcbinfo;
1114         struct sockaddr_in *sin, src;
1115         int cscov_partial = 0;
1116         int error = 0;
1117         int ipflags;
1118         u_short fport, lport;
1119         int unlock_udbinfo;
1120         u_char tos;
1121         uint8_t pr;
1122         uint16_t cscov = 0;
1123         uint32_t flowid = 0;
1124         uint8_t flowtype = M_HASHTYPE_NONE;
1125
1126         /*
1127          * udp_output() may need to temporarily bind or connect the current
1128          * inpcb.  As such, we don't know up front whether we will need the
1129          * pcbinfo lock or not.  Do any work to decide what is needed up
1130          * front before acquiring any locks.
1131          */
1132         if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
1133                 if (control)
1134                         m_freem(control);
1135                 m_freem(m);
1136                 return (EMSGSIZE);
1137         }
1138
1139         src.sin_family = 0;
1140         INP_RLOCK(inp);
1141         tos = inp->inp_ip_tos;
1142         if (control != NULL) {
1143                 /*
1144                  * XXX: Currently, we assume all the optional information is
1145                  * stored in a single mbuf.
1146                  */
1147                 if (control->m_next) {
1148                         INP_RUNLOCK(inp);
1149                         m_freem(control);
1150                         m_freem(m);
1151                         return (EINVAL);
1152                 }
1153                 for (; control->m_len > 0;
1154                     control->m_data += CMSG_ALIGN(cm->cmsg_len),
1155                     control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
1156                         cm = mtod(control, struct cmsghdr *);
1157                         if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0
1158                             || cm->cmsg_len > control->m_len) {
1159                                 error = EINVAL;
1160                                 break;
1161                         }
1162                         if (cm->cmsg_level != IPPROTO_IP)
1163                                 continue;
1164
1165                         switch (cm->cmsg_type) {
1166                         case IP_SENDSRCADDR:
1167                                 if (cm->cmsg_len !=
1168                                     CMSG_LEN(sizeof(struct in_addr))) {
1169                                         error = EINVAL;
1170                                         break;
1171                                 }
1172                                 bzero(&src, sizeof(src));
1173                                 src.sin_family = AF_INET;
1174                                 src.sin_len = sizeof(src);
1175                                 src.sin_port = inp->inp_lport;
1176                                 src.sin_addr =
1177                                     *(struct in_addr *)CMSG_DATA(cm);
1178                                 break;
1179
1180                         case IP_TOS:
1181                                 if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) {
1182                                         error = EINVAL;
1183                                         break;
1184                                 }
1185                                 tos = *(u_char *)CMSG_DATA(cm);
1186                                 break;
1187
1188                         case IP_FLOWID:
1189                                 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1190                                         error = EINVAL;
1191                                         break;
1192                                 }
1193                                 flowid = *(uint32_t *) CMSG_DATA(cm);
1194                                 break;
1195
1196                         case IP_FLOWTYPE:
1197                                 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1198                                         error = EINVAL;
1199                                         break;
1200                                 }
1201                                 flowtype = *(uint32_t *) CMSG_DATA(cm);
1202                                 break;
1203
1204 #ifdef  RSS
1205                         case IP_RSSBUCKETID:
1206                                 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1207                                         error = EINVAL;
1208                                         break;
1209                                 }
1210                                 /* This is just a placeholder for now */
1211                                 break;
1212 #endif  /* RSS */
1213                         default:
1214                                 error = ENOPROTOOPT;
1215                                 break;
1216                         }
1217                         if (error)
1218                                 break;
1219                 }
1220                 m_freem(control);
1221         }
1222         if (error) {
1223                 INP_RUNLOCK(inp);
1224                 m_freem(m);
1225                 return (error);
1226         }
1227
1228         /*
1229          * Depending on whether or not the application has bound or connected
1230          * the socket, we may have to do varying levels of work.  The optimal
1231          * case is for a connected UDP socket, as a global lock isn't
1232          * required at all.
1233          *
1234          * In order to decide which we need, we require stability of the
1235          * inpcb binding, which we ensure by acquiring a read lock on the
1236          * inpcb.  This doesn't strictly follow the lock order, so we play
1237          * the trylock and retry game; note that we may end up with more
1238          * conservative locks than required the second time around, so later
1239          * assertions have to accept that.  Further analysis of the number of
1240          * misses under contention is required.
1241          *
1242          * XXXRW: Check that hash locking update here is correct.
1243          */
1244         pr = inp->inp_socket->so_proto->pr_protocol;
1245         pcbinfo = udp_get_inpcbinfo(pr);
1246         sin = (struct sockaddr_in *)addr;
1247         if (sin != NULL &&
1248             (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
1249                 INP_RUNLOCK(inp);
1250                 INP_WLOCK(inp);
1251                 INP_HASH_WLOCK(pcbinfo);
1252                 unlock_udbinfo = UH_WLOCKED;
1253         } else if ((sin != NULL && (
1254             (sin->sin_addr.s_addr == INADDR_ANY) ||
1255             (sin->sin_addr.s_addr == INADDR_BROADCAST) ||
1256             (inp->inp_laddr.s_addr == INADDR_ANY) ||
1257             (inp->inp_lport == 0))) ||
1258             (src.sin_family == AF_INET)) {
1259                 INP_HASH_RLOCK(pcbinfo);
1260                 unlock_udbinfo = UH_RLOCKED;
1261         } else
1262                 unlock_udbinfo = UH_UNLOCKED;
1263
1264         /*
1265          * If the IP_SENDSRCADDR control message was specified, override the
1266          * source address for this datagram.  Its use is invalidated if the
1267          * address thus specified is incomplete or clobbers other inpcbs.
1268          */
1269         laddr = inp->inp_laddr;
1270         lport = inp->inp_lport;
1271         if (src.sin_family == AF_INET) {
1272                 INP_HASH_LOCK_ASSERT(pcbinfo);
1273                 if ((lport == 0) ||
1274                     (laddr.s_addr == INADDR_ANY &&
1275                      src.sin_addr.s_addr == INADDR_ANY)) {
1276                         error = EINVAL;
1277                         goto release;
1278                 }
1279                 error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
1280                     &laddr.s_addr, &lport, td->td_ucred);
1281                 if (error)
1282                         goto release;
1283         }
1284
1285         /*
1286          * If a UDP socket has been connected, then a local address/port will
1287          * have been selected and bound.
1288          *
1289          * If a UDP socket has not been connected to, then an explicit
1290          * destination address must be used, in which case a local
1291          * address/port may not have been selected and bound.
1292          */
1293         if (sin != NULL) {
1294                 INP_LOCK_ASSERT(inp);
1295                 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1296                         error = EISCONN;
1297                         goto release;
1298                 }
1299
1300                 /*
1301                  * Jail may rewrite the destination address, so let it do
1302                  * that before we use it.
1303                  */
1304                 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1305                 if (error)
1306                         goto release;
1307
1308                 /*
1309                  * If a local address or port hasn't yet been selected, or if
1310                  * the destination address needs to be rewritten due to using
1311                  * a special INADDR_ constant, invoke in_pcbconnect_setup()
1312                  * to do the heavy lifting.  Once a port is selected, we
1313                  * commit the binding back to the socket; we also commit the
1314                  * binding of the address if in jail.
1315                  *
1316                  * If we already have a valid binding and we're not
1317                  * requesting a destination address rewrite, use a fast path.
1318                  */
1319                 if (inp->inp_laddr.s_addr == INADDR_ANY ||
1320                     inp->inp_lport == 0 ||
1321                     sin->sin_addr.s_addr == INADDR_ANY ||
1322                     sin->sin_addr.s_addr == INADDR_BROADCAST) {
1323                         INP_HASH_LOCK_ASSERT(pcbinfo);
1324                         error = in_pcbconnect_setup(inp, addr, &laddr.s_addr,
1325                             &lport, &faddr.s_addr, &fport, NULL,
1326                             td->td_ucred);
1327                         if (error)
1328                                 goto release;
1329
1330                         /*
1331                          * XXXRW: Why not commit the port if the address is
1332                          * !INADDR_ANY?
1333                          */
1334                         /* Commit the local port if newly assigned. */
1335                         if (inp->inp_laddr.s_addr == INADDR_ANY &&
1336                             inp->inp_lport == 0) {
1337                                 INP_WLOCK_ASSERT(inp);
1338                                 INP_HASH_WLOCK_ASSERT(pcbinfo);
1339                                 /*
1340                                  * Remember addr if jailed, to prevent
1341                                  * rebinding.
1342                                  */
1343                                 if (prison_flag(td->td_ucred, PR_IP4))
1344                                         inp->inp_laddr = laddr;
1345                                 inp->inp_lport = lport;
1346                                 if (in_pcbinshash(inp) != 0) {
1347                                         inp->inp_lport = 0;
1348                                         error = EAGAIN;
1349                                         goto release;
1350                                 }
1351                                 inp->inp_flags |= INP_ANONPORT;
1352                         }
1353                 } else {
1354                         faddr = sin->sin_addr;
1355                         fport = sin->sin_port;
1356                 }
1357         } else {
1358                 INP_LOCK_ASSERT(inp);
1359                 faddr = inp->inp_faddr;
1360                 fport = inp->inp_fport;
1361                 if (faddr.s_addr == INADDR_ANY) {
1362                         error = ENOTCONN;
1363                         goto release;
1364                 }
1365         }
1366
1367         /*
1368          * Calculate data length and get a mbuf for UDP, IP, and possible
1369          * link-layer headers.  Immediate slide the data pointer back forward
1370          * since we won't use that space at this layer.
1371          */
1372         M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT);
1373         if (m == NULL) {
1374                 error = ENOBUFS;
1375                 goto release;
1376         }
1377         m->m_data += max_linkhdr;
1378         m->m_len -= max_linkhdr;
1379         m->m_pkthdr.len -= max_linkhdr;
1380
1381         /*
1382          * Fill in mbuf with extended UDP header and addresses and length put
1383          * into network format.
1384          */
1385         ui = mtod(m, struct udpiphdr *);
1386         bzero(ui->ui_x1, sizeof(ui->ui_x1));    /* XXX still needed? */
1387         ui->ui_pr = pr;
1388         ui->ui_src = laddr;
1389         ui->ui_dst = faddr;
1390         ui->ui_sport = lport;
1391         ui->ui_dport = fport;
1392         ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
1393         if (pr == IPPROTO_UDPLITE) {
1394                 struct udpcb *up;
1395                 uint16_t plen;
1396
1397                 up = intoudpcb(inp);
1398                 cscov = up->u_txcslen;
1399                 plen = (u_short)len + sizeof(struct udphdr);
1400                 if (cscov >= plen)
1401                         cscov = 0;
1402                 ui->ui_len = htons(plen);
1403                 ui->ui_ulen = htons(cscov);
1404                 /*
1405                  * For UDP-Lite, checksum coverage length of zero means
1406                  * the entire UDPLite packet is covered by the checksum.
1407                  */
1408                 cscov_partial = (cscov == 0) ? 0 : 1;
1409         } else
1410                 ui->ui_v = IPVERSION << 4;
1411
1412         /*
1413          * Set the Don't Fragment bit in the IP header.
1414          */
1415         if (inp->inp_flags & INP_DONTFRAG) {
1416                 struct ip *ip;
1417
1418                 ip = (struct ip *)&ui->ui_i;
1419                 ip->ip_off |= htons(IP_DF);
1420         }
1421
1422         ipflags = 0;
1423         if (inp->inp_socket->so_options & SO_DONTROUTE)
1424                 ipflags |= IP_ROUTETOIF;
1425         if (inp->inp_socket->so_options & SO_BROADCAST)
1426                 ipflags |= IP_ALLOWBROADCAST;
1427         if (inp->inp_flags & INP_ONESBCAST)
1428                 ipflags |= IP_SENDONES;
1429
1430 #ifdef MAC
1431         mac_inpcb_create_mbuf(inp, m);
1432 #endif
1433
1434         /*
1435          * Set up checksum and output datagram.
1436          */
1437         ui->ui_sum = 0;
1438         if (pr == IPPROTO_UDPLITE) {
1439                 if (inp->inp_flags & INP_ONESBCAST)
1440                         faddr.s_addr = INADDR_BROADCAST;
1441                 if (cscov_partial) {
1442                         if ((ui->ui_sum = in_cksum(m, sizeof(struct ip) + cscov)) == 0)
1443                                 ui->ui_sum = 0xffff;
1444                 } else {
1445                         if ((ui->ui_sum = in_cksum(m, sizeof(struct udpiphdr) + len)) == 0)
1446                                 ui->ui_sum = 0xffff;
1447                 }
1448         } else if (V_udp_cksum) {
1449                 if (inp->inp_flags & INP_ONESBCAST)
1450                         faddr.s_addr = INADDR_BROADCAST;
1451                 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
1452                     htons((u_short)len + sizeof(struct udphdr) + pr));
1453                 m->m_pkthdr.csum_flags = CSUM_UDP;
1454                 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1455         }
1456         ((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len);
1457         ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl;    /* XXX */
1458         ((struct ip *)ui)->ip_tos = tos;                /* XXX */
1459         UDPSTAT_INC(udps_opackets);
1460
1461         /*
1462          * Setup flowid / RSS information for outbound socket.
1463          *
1464          * Once the UDP code decides to set a flowid some other way,
1465          * this allows the flowid to be overridden by userland.
1466          */
1467         if (flowtype != M_HASHTYPE_NONE) {
1468                 m->m_pkthdr.flowid = flowid;
1469                 M_HASHTYPE_SET(m, flowtype);
1470 #ifdef  RSS
1471         } else {
1472                 uint32_t hash_val, hash_type;
1473                 /*
1474                  * Calculate an appropriate RSS hash for UDP and
1475                  * UDP Lite.
1476                  *
1477                  * The called function will take care of figuring out
1478                  * whether a 2-tuple or 4-tuple hash is required based
1479                  * on the currently configured scheme.
1480                  *
1481                  * Later later on connected socket values should be
1482                  * cached in the inpcb and reused, rather than constantly
1483                  * re-calculating it.
1484                  *
1485                  * UDP Lite is a different protocol number and will
1486                  * likely end up being hashed as a 2-tuple until
1487                  * RSS / NICs grow UDP Lite protocol awareness.
1488                  */
1489                 if (rss_proto_software_hash_v4(faddr, laddr, fport, lport,
1490                     pr, &hash_val, &hash_type) == 0) {
1491                         m->m_pkthdr.flowid = hash_val;
1492                         M_HASHTYPE_SET(m, hash_type);
1493                 }
1494 #endif
1495         }
1496
1497 #ifdef  RSS
1498         /*
1499          * Don't override with the inp cached flowid value.
1500          *
1501          * Depending upon the kind of send being done, the inp
1502          * flowid/flowtype values may actually not be appropriate
1503          * for this particular socket send.
1504          *
1505          * We should either leave the flowid at zero (which is what is
1506          * currently done) or set it to some software generated
1507          * hash value based on the packet contents.
1508          */
1509         ipflags |= IP_NODEFAULTFLOWID;
1510 #endif  /* RSS */
1511
1512         if (unlock_udbinfo == UH_WLOCKED)
1513                 INP_HASH_WUNLOCK(pcbinfo);
1514         else if (unlock_udbinfo == UH_RLOCKED)
1515                 INP_HASH_RUNLOCK(pcbinfo);
1516         UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u);
1517         error = ip_output(m, inp->inp_options, NULL, ipflags,
1518             inp->inp_moptions, inp);
1519         if (unlock_udbinfo == UH_WLOCKED)
1520                 INP_WUNLOCK(inp);
1521         else
1522                 INP_RUNLOCK(inp);
1523         return (error);
1524
1525 release:
1526         if (unlock_udbinfo == UH_WLOCKED) {
1527                 INP_HASH_WUNLOCK(pcbinfo);
1528                 INP_WUNLOCK(inp);
1529         } else if (unlock_udbinfo == UH_RLOCKED) {
1530                 INP_HASH_RUNLOCK(pcbinfo);
1531                 INP_RUNLOCK(inp);
1532         } else
1533                 INP_RUNLOCK(inp);
1534         m_freem(m);
1535         return (error);
1536 }
1537
1538
1539 #if defined(IPSEC) && defined(IPSEC_NAT_T)
1540 /*
1541  * Potentially decap ESP in UDP frame.  Check for an ESP header
1542  * and optional marker; if present, strip the UDP header and
1543  * push the result through IPSec.
1544  *
1545  * Returns mbuf to be processed (potentially re-allocated) or
1546  * NULL if consumed and/or processed.
1547  */
1548 static struct mbuf *
1549 udp4_espdecap(struct inpcb *inp, struct mbuf *m, int off)
1550 {
1551         size_t minlen, payload, skip, iphlen;
1552         caddr_t data;
1553         struct udpcb *up;
1554         struct m_tag *tag;
1555         struct udphdr *udphdr;
1556         struct ip *ip;
1557
1558         INP_RLOCK_ASSERT(inp);
1559
1560         /*
1561          * Pull up data so the longest case is contiguous:
1562          *    IP/UDP hdr + non ESP marker + ESP hdr.
1563          */
1564         minlen = off + sizeof(uint64_t) + sizeof(struct esp);
1565         if (minlen > m->m_pkthdr.len)
1566                 minlen = m->m_pkthdr.len;
1567         if ((m = m_pullup(m, minlen)) == NULL) {
1568                 IPSECSTAT_INC(ips_in_inval);
1569                 return (NULL);          /* Bypass caller processing. */
1570         }
1571         data = mtod(m, caddr_t);        /* Points to ip header. */
1572         payload = m->m_len - off;       /* Size of payload. */
1573
1574         if (payload == 1 && data[off] == '\xff')
1575                 return (m);             /* NB: keepalive packet, no decap. */
1576
1577         up = intoudpcb(inp);
1578         KASSERT(up != NULL, ("%s: udpcb NULL", __func__));
1579         KASSERT((up->u_flags & UF_ESPINUDP_ALL) != 0,
1580             ("u_flags 0x%x", up->u_flags));
1581
1582         /*
1583          * Check that the payload is large enough to hold an
1584          * ESP header and compute the amount of data to remove.
1585          *
1586          * NB: the caller has already done a pullup for us.
1587          * XXX can we assume alignment and eliminate bcopys?
1588          */
1589         if (up->u_flags & UF_ESPINUDP_NON_IKE) {
1590                 /*
1591                  * draft-ietf-ipsec-nat-t-ike-0[01].txt and
1592                  * draft-ietf-ipsec-udp-encaps-(00/)01.txt, ignoring
1593                  * possible AH mode non-IKE marker+non-ESP marker
1594                  * from draft-ietf-ipsec-udp-encaps-00.txt.
1595                  */
1596                 uint64_t marker;
1597
1598                 if (payload <= sizeof(uint64_t) + sizeof(struct esp))
1599                         return (m);     /* NB: no decap. */
1600                 bcopy(data + off, &marker, sizeof(uint64_t));
1601                 if (marker != 0)        /* Non-IKE marker. */
1602                         return (m);     /* NB: no decap. */
1603                 skip = sizeof(uint64_t) + sizeof(struct udphdr);
1604         } else {
1605                 uint32_t spi;
1606
1607                 if (payload <= sizeof(struct esp)) {
1608                         IPSECSTAT_INC(ips_in_inval);
1609                         m_freem(m);
1610                         return (NULL);  /* Discard. */
1611                 }
1612                 bcopy(data + off, &spi, sizeof(uint32_t));
1613                 if (spi == 0)           /* Non-ESP marker. */
1614                         return (m);     /* NB: no decap. */
1615                 skip = sizeof(struct udphdr);
1616         }
1617
1618         /*
1619          * Setup a PACKET_TAG_IPSEC_NAT_T_PORT tag to remember
1620          * the UDP ports. This is required if we want to select
1621          * the right SPD for multiple hosts behind same NAT.
1622          *
1623          * NB: ports are maintained in network byte order everywhere
1624          *     in the NAT-T code.
1625          */
1626         tag = m_tag_get(PACKET_TAG_IPSEC_NAT_T_PORTS,
1627                 2 * sizeof(uint16_t), M_NOWAIT);
1628         if (tag == NULL) {
1629                 IPSECSTAT_INC(ips_in_nomem);
1630                 m_freem(m);
1631                 return (NULL);          /* Discard. */
1632         }
1633         iphlen = off - sizeof(struct udphdr);
1634         udphdr = (struct udphdr *)(data + iphlen);
1635         ((uint16_t *)(tag + 1))[0] = udphdr->uh_sport;
1636         ((uint16_t *)(tag + 1))[1] = udphdr->uh_dport;
1637         m_tag_prepend(m, tag);
1638
1639         /*
1640          * Remove the UDP header (and possibly the non ESP marker)
1641          * IP header length is iphlen
1642          * Before:
1643          *   <--- off --->
1644          *   +----+------+-----+
1645          *   | IP |  UDP | ESP |
1646          *   +----+------+-----+
1647          *        <-skip->
1648          * After:
1649          *          +----+-----+
1650          *          | IP | ESP |
1651          *          +----+-----+
1652          *   <-skip->
1653          */
1654         ovbcopy(data, data + skip, iphlen);
1655         m_adj(m, skip);
1656
1657         ip = mtod(m, struct ip *);
1658         ip->ip_len = htons(ntohs(ip->ip_len) - skip);
1659         ip->ip_p = IPPROTO_ESP;
1660
1661         /*
1662          * We cannot yet update the cksums so clear any
1663          * h/w cksum flags as they are no longer valid.
1664          */
1665         if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID)
1666                 m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID|CSUM_PSEUDO_HDR);
1667
1668         (void) ipsec_common_input(m, iphlen, offsetof(struct ip, ip_p),
1669                                 AF_INET, ip->ip_p);
1670         return (NULL);                  /* NB: consumed, bypass processing. */
1671 }
1672 #endif /* defined(IPSEC) && defined(IPSEC_NAT_T) */
1673
1674 static void
1675 udp_abort(struct socket *so)
1676 {
1677         struct inpcb *inp;
1678         struct inpcbinfo *pcbinfo;
1679
1680         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1681         inp = sotoinpcb(so);
1682         KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
1683         INP_WLOCK(inp);
1684         if (inp->inp_faddr.s_addr != INADDR_ANY) {
1685                 INP_HASH_WLOCK(pcbinfo);
1686                 in_pcbdisconnect(inp);
1687                 inp->inp_laddr.s_addr = INADDR_ANY;
1688                 INP_HASH_WUNLOCK(pcbinfo);
1689                 soisdisconnected(so);
1690         }
1691         INP_WUNLOCK(inp);
1692 }
1693
1694 static int
1695 udp_attach(struct socket *so, int proto, struct thread *td)
1696 {
1697         struct inpcb *inp;
1698         struct inpcbinfo *pcbinfo;
1699         int error;
1700
1701         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1702         inp = sotoinpcb(so);
1703         KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
1704         error = soreserve(so, udp_sendspace, udp_recvspace);
1705         if (error)
1706                 return (error);
1707         INP_INFO_WLOCK(pcbinfo);
1708         error = in_pcballoc(so, pcbinfo);
1709         if (error) {
1710                 INP_INFO_WUNLOCK(pcbinfo);
1711                 return (error);
1712         }
1713
1714         inp = sotoinpcb(so);
1715         inp->inp_vflag |= INP_IPV4;
1716         inp->inp_ip_ttl = V_ip_defttl;
1717
1718         error = udp_newudpcb(inp);
1719         if (error) {
1720                 in_pcbdetach(inp);
1721                 in_pcbfree(inp);
1722                 INP_INFO_WUNLOCK(pcbinfo);
1723                 return (error);
1724         }
1725
1726         INP_WUNLOCK(inp);
1727         INP_INFO_WUNLOCK(pcbinfo);
1728         return (0);
1729 }
1730 #endif /* INET */
1731
1732 int
1733 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f, void *ctx)
1734 {
1735         struct inpcb *inp;
1736         struct udpcb *up;
1737
1738         KASSERT(so->so_type == SOCK_DGRAM,
1739             ("udp_set_kernel_tunneling: !dgram"));
1740         inp = sotoinpcb(so);
1741         KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL"));
1742         INP_WLOCK(inp);
1743         up = intoudpcb(inp);
1744         if (up->u_tun_func != NULL) {
1745                 INP_WUNLOCK(inp);
1746                 return (EBUSY);
1747         }
1748         up->u_tun_func = f;
1749         up->u_tun_ctx = ctx;
1750         INP_WUNLOCK(inp);
1751         return (0);
1752 }
1753
1754 #ifdef INET
1755 static int
1756 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1757 {
1758         struct inpcb *inp;
1759         struct inpcbinfo *pcbinfo;
1760         int error;
1761
1762         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1763         inp = sotoinpcb(so);
1764         KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1765         INP_WLOCK(inp);
1766         INP_HASH_WLOCK(pcbinfo);
1767         error = in_pcbbind(inp, nam, td->td_ucred);
1768         INP_HASH_WUNLOCK(pcbinfo);
1769         INP_WUNLOCK(inp);
1770         return (error);
1771 }
1772
1773 static void
1774 udp_close(struct socket *so)
1775 {
1776         struct inpcb *inp;
1777         struct inpcbinfo *pcbinfo;
1778
1779         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1780         inp = sotoinpcb(so);
1781         KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1782         INP_WLOCK(inp);
1783         if (inp->inp_faddr.s_addr != INADDR_ANY) {
1784                 INP_HASH_WLOCK(pcbinfo);
1785                 in_pcbdisconnect(inp);
1786                 inp->inp_laddr.s_addr = INADDR_ANY;
1787                 INP_HASH_WUNLOCK(pcbinfo);
1788                 soisdisconnected(so);
1789         }
1790         INP_WUNLOCK(inp);
1791 }
1792
1793 static int
1794 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1795 {
1796         struct inpcb *inp;
1797         struct inpcbinfo *pcbinfo;
1798         struct sockaddr_in *sin;
1799         int error;
1800
1801         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1802         inp = sotoinpcb(so);
1803         KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1804         INP_WLOCK(inp);
1805         if (inp->inp_faddr.s_addr != INADDR_ANY) {
1806                 INP_WUNLOCK(inp);
1807                 return (EISCONN);
1808         }
1809         sin = (struct sockaddr_in *)nam;
1810         error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1811         if (error != 0) {
1812                 INP_WUNLOCK(inp);
1813                 return (error);
1814         }
1815         INP_HASH_WLOCK(pcbinfo);
1816         error = in_pcbconnect(inp, nam, td->td_ucred);
1817         INP_HASH_WUNLOCK(pcbinfo);
1818         if (error == 0)
1819                 soisconnected(so);
1820         INP_WUNLOCK(inp);
1821         return (error);
1822 }
1823
1824 static void
1825 udp_detach(struct socket *so)
1826 {
1827         struct inpcb *inp;
1828         struct inpcbinfo *pcbinfo;
1829         struct udpcb *up;
1830
1831         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1832         inp = sotoinpcb(so);
1833         KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1834         KASSERT(inp->inp_faddr.s_addr == INADDR_ANY,
1835             ("udp_detach: not disconnected"));
1836         INP_INFO_WLOCK(pcbinfo);
1837         INP_WLOCK(inp);
1838         up = intoudpcb(inp);
1839         KASSERT(up != NULL, ("%s: up == NULL", __func__));
1840         inp->inp_ppcb = NULL;
1841         in_pcbdetach(inp);
1842         in_pcbfree(inp);
1843         INP_INFO_WUNLOCK(pcbinfo);
1844         udp_discardcb(up);
1845 }
1846
1847 static int
1848 udp_disconnect(struct socket *so)
1849 {
1850         struct inpcb *inp;
1851         struct inpcbinfo *pcbinfo;
1852
1853         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1854         inp = sotoinpcb(so);
1855         KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1856         INP_WLOCK(inp);
1857         if (inp->inp_faddr.s_addr == INADDR_ANY) {
1858                 INP_WUNLOCK(inp);
1859                 return (ENOTCONN);
1860         }
1861         INP_HASH_WLOCK(pcbinfo);
1862         in_pcbdisconnect(inp);
1863         inp->inp_laddr.s_addr = INADDR_ANY;
1864         INP_HASH_WUNLOCK(pcbinfo);
1865         SOCK_LOCK(so);
1866         so->so_state &= ~SS_ISCONNECTED;                /* XXX */
1867         SOCK_UNLOCK(so);
1868         INP_WUNLOCK(inp);
1869         return (0);
1870 }
1871
1872 static int
1873 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1874     struct mbuf *control, struct thread *td)
1875 {
1876         struct inpcb *inp;
1877
1878         inp = sotoinpcb(so);
1879         KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1880         return (udp_output(inp, m, addr, control, td));
1881 }
1882 #endif /* INET */
1883
1884 int
1885 udp_shutdown(struct socket *so)
1886 {
1887         struct inpcb *inp;
1888
1889         inp = sotoinpcb(so);
1890         KASSERT(inp != NULL, ("udp_shutdown: inp == NULL"));
1891         INP_WLOCK(inp);
1892         socantsendmore(so);
1893         INP_WUNLOCK(inp);
1894         return (0);
1895 }
1896
1897 #ifdef INET
1898 struct pr_usrreqs udp_usrreqs = {
1899         .pru_abort =            udp_abort,
1900         .pru_attach =           udp_attach,
1901         .pru_bind =             udp_bind,
1902         .pru_connect =          udp_connect,
1903         .pru_control =          in_control,
1904         .pru_detach =           udp_detach,
1905         .pru_disconnect =       udp_disconnect,
1906         .pru_peeraddr =         in_getpeeraddr,
1907         .pru_send =             udp_send,
1908         .pru_soreceive =        soreceive_dgram,
1909         .pru_sosend =           sosend_dgram,
1910         .pru_shutdown =         udp_shutdown,
1911         .pru_sockaddr =         in_getsockaddr,
1912         .pru_sosetlabel =       in_pcbsosetlabel,
1913         .pru_close =            udp_close,
1914 };
1915 #endif /* INET */