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 = NULL;
 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         if ((errno == EHOSTUNREACH || errno == ENETUNREACH ||
 744              errno == EHOSTDOWN) && inp->inp_route.ro_rt) {
 745                 RTFREE(inp->inp_route.ro_rt);
 746                 inp->inp_route.ro_rt = (struct rtentry *)NULL;
 747         }
 748
 749         inp->inp_socket->so_error = errno;
 750         sorwakeup(inp->inp_socket);
 751         sowwakeup(inp->inp_socket);
 752         return (inp);
 753 }
 754
 755 #ifdef INET
 756 static void
 757 udp_common_ctlinput(int cmd, struct sockaddr *sa, void *vip,
 758     struct inpcbinfo *pcbinfo)
 759 {
 760         struct ip *ip = vip;
 761         struct udphdr *uh;
 762         struct in_addr faddr;
 763         struct inpcb *inp;
 764
 765         faddr = ((struct sockaddr_in *)sa)->sin_addr;
 766         if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
 767                 return;
 768
 769         if (PRC_IS_REDIRECT(cmd)) {
 770                 /* signal EHOSTDOWN, as it flushes the cached route */
 771                 in_pcbnotifyall(&V_udbinfo, faddr, EHOSTDOWN, udp_notify);
 772                 return;
 773         }
 774
 775         /*
 776          * Hostdead is ugly because it goes linearly through all PCBs.
 777          *
 778          * XXX: We never get this from ICMP, otherwise it makes an excellent
 779          * DoS attack on machines with many connections.
 780          */
 781         if (cmd == PRC_HOSTDEAD)
 782                 ip = NULL;
 783         else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
 784                 return;
 785         if (ip != NULL) {
 786                 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
 787                 inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport,
 788                     ip->ip_src, uh->uh_sport, INPLOOKUP_RLOCKPCB, NULL);
 789                 if (inp != NULL) {
 790                         INP_RLOCK_ASSERT(inp);
 791                         if (inp->inp_socket != NULL) {
 792                                 udp_notify(inp, inetctlerrmap[cmd]);
 793                         }
 794                         INP_RUNLOCK(inp);
 795                 } else {
 796                         inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport,
 797                                            ip->ip_src, uh->uh_sport,
 798                                            INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
 799                         if (inp != NULL) {
 800                                 struct udpcb *up;
 801
 802                                 up = intoudpcb(inp);
 803                                 if (up->u_icmp_func != NULL) {
 804                                         INP_RUNLOCK(inp);
 805                                         (*up->u_icmp_func)(cmd, sa, vip, up->u_tun_ctx);
 806                                 } else {
 807                                         INP_RUNLOCK(inp);
 808                                 }
 809                         }
 810                 }
 811         } else
 812                 in_pcbnotifyall(pcbinfo, faddr, inetctlerrmap[cmd],
 813                     udp_notify);
 814 }
 815 void
 816 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
 817 {
 818
 819         return (udp_common_ctlinput(cmd, sa, vip, &V_udbinfo));
 820 }
 821
 822 void
 823 udplite_ctlinput(int cmd, struct sockaddr *sa, void *vip)
 824 {
 825
 826         return (udp_common_ctlinput(cmd, sa, vip, &V_ulitecbinfo));
 827 }
 828 #endif /* INET */
 829
 830 static int
 831 udp_pcblist(SYSCTL_HANDLER_ARGS)
 832 {
 833         int error, i, n;
 834         struct inpcb *inp, **inp_list;
 835         inp_gen_t gencnt;
 836         struct xinpgen xig;
 837
 838         /*
 839          * The process of preparing the PCB list is too time-consuming and
 840          * resource-intensive to repeat twice on every request.
 841          */
 842         if (req->oldptr == 0) {
 843                 n = V_udbinfo.ipi_count;
 844                 n += imax(n / 8, 10);
 845                 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
 846                 return (0);
 847         }
 848
 849         if (req->newptr != 0)
 850                 return (EPERM);
 851
 852         /*
 853          * OK, now we're committed to doing something.
 854          */
 855         INP_INFO_RLOCK(&V_udbinfo);
 856         gencnt = V_udbinfo.ipi_gencnt;
 857         n = V_udbinfo.ipi_count;
 858         INP_INFO_RUNLOCK(&V_udbinfo);
 859
 860         error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
 861                 + n * sizeof(struct xinpcb));
 862         if (error != 0)
 863                 return (error);
 864
 865         xig.xig_len = sizeof xig;
 866         xig.xig_count = n;
 867         xig.xig_gen = gencnt;
 868         xig.xig_sogen = so_gencnt;
 869         error = SYSCTL_OUT(req, &xig, sizeof xig);
 870         if (error)
 871                 return (error);
 872
 873         inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
 874         if (inp_list == NULL)
 875                 return (ENOMEM);
 876
 877         INP_INFO_RLOCK(&V_udbinfo);
 878         for (inp = LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n;
 879              inp = LIST_NEXT(inp, inp_list)) {
 880                 INP_WLOCK(inp);
 881                 if (inp->inp_gencnt <= gencnt &&
 882                     cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
 883                         in_pcbref(inp);
 884                         inp_list[i++] = inp;
 885                 }
 886                 INP_WUNLOCK(inp);
 887         }
 888         INP_INFO_RUNLOCK(&V_udbinfo);
 889         n = i;
 890
 891         error = 0;
 892         for (i = 0; i < n; i++) {
 893                 inp = inp_list[i];
 894                 INP_RLOCK(inp);
 895                 if (inp->inp_gencnt <= gencnt) {
 896                         struct xinpcb xi;
 897
 898                         bzero(&xi, sizeof(xi));
 899                         xi.xi_len = sizeof xi;
 900                         /* XXX should avoid extra copy */
 901                         bcopy(inp, &xi.xi_inp, sizeof *inp);
 902                         if (inp->inp_socket)
 903                                 sotoxsocket(inp->inp_socket, &xi.xi_socket);
 904                         xi.xi_inp.inp_gencnt = inp->inp_gencnt;
 905                         INP_RUNLOCK(inp);
 906                         error = SYSCTL_OUT(req, &xi, sizeof xi);
 907                 } else
 908                         INP_RUNLOCK(inp);
 909         }
 910         INP_INFO_WLOCK(&V_udbinfo);
 911         for (i = 0; i < n; i++) {
 912                 inp = inp_list[i];
 913                 INP_RLOCK(inp);
 914                 if (!in_pcbrele_rlocked(inp))
 915                         INP_RUNLOCK(inp);
 916         }
 917         INP_INFO_WUNLOCK(&V_udbinfo);
 918
 919         if (!error) {
 920                 /*
 921                  * Give the user an updated idea of our state.  If the
 922                  * generation differs from what we told her before, she knows
 923                  * that something happened while we were processing this
 924                  * request, and it might be necessary to retry.
 925                  */
 926                 INP_INFO_RLOCK(&V_udbinfo);
 927                 xig.xig_gen = V_udbinfo.ipi_gencnt;
 928                 xig.xig_sogen = so_gencnt;
 929                 xig.xig_count = V_udbinfo.ipi_count;
 930                 INP_INFO_RUNLOCK(&V_udbinfo);
 931                 error = SYSCTL_OUT(req, &xig, sizeof xig);
 932         }
 933         free(inp_list, M_TEMP);
 934         return (error);
 935 }
 936
 937 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist,
 938     CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
 939     udp_pcblist, "S,xinpcb", "List of active UDP sockets");
 940
 941 #ifdef INET
 942 static int
 943 udp_getcred(SYSCTL_HANDLER_ARGS)
 944 {
 945         struct xucred xuc;
 946         struct sockaddr_in addrs[2];
 947         struct inpcb *inp;
 948         int error;
 949
 950         error = priv_check(req->td, PRIV_NETINET_GETCRED);
 951         if (error)
 952                 return (error);
 953         error = SYSCTL_IN(req, addrs, sizeof(addrs));
 954         if (error)
 955                 return (error);
 956         inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
 957             addrs[0].sin_addr, addrs[0].sin_port,
 958             INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
 959         if (inp != NULL) {
 960                 INP_RLOCK_ASSERT(inp);
 961                 if (inp->inp_socket == NULL)
 962                         error = ENOENT;
 963                 if (error == 0)
 964                         error = cr_canseeinpcb(req->td->td_ucred, inp);
 965                 if (error == 0)
 966                         cru2x(inp->inp_cred, &xuc);
 967                 INP_RUNLOCK(inp);
 968         } else
 969                 error = ENOENT;
 970         if (error == 0)
 971                 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
 972         return (error);
 973 }
 974
 975 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
 976     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
 977     udp_getcred, "S,xucred", "Get the xucred of a UDP connection");
 978 #endif /* INET */
 979
 980 int
 981 udp_ctloutput(struct socket *so, struct sockopt *sopt)
 982 {
 983         struct inpcb *inp;
 984         struct udpcb *up;
 985         int isudplite, error, optval;
 986
 987         error = 0;
 988         isudplite = (so->so_proto->pr_protocol == IPPROTO_UDPLITE) ? 1 : 0;
 989         inp = sotoinpcb(so);
 990         KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
 991         INP_WLOCK(inp);
 992         if (sopt->sopt_level != so->so_proto->pr_protocol) {
 993 #ifdef INET6
 994                 if (INP_CHECK_SOCKAF(so, AF_INET6)) {
 995                         INP_WUNLOCK(inp);
 996                         error = ip6_ctloutput(so, sopt);
 997                 }
 998 #endif
 999 #if defined(INET) && defined(INET6)
1000                 else
1001 #endif
1002 #ifdef INET
1003                 {
1004                         INP_WUNLOCK(inp);
1005                         error = ip_ctloutput(so, sopt);
1006                 }
1007 #endif
1008                 return (error);
1009         }
1010
1011         switch (sopt->sopt_dir) {
1012         case SOPT_SET:
1013                 switch (sopt->sopt_name) {
1014                 case UDP_ENCAP:
1015                         INP_WUNLOCK(inp);
1016                         error = sooptcopyin(sopt, &optval, sizeof optval,
1017                                             sizeof optval);
1018                         if (error)
1019                                 break;
1020                         inp = sotoinpcb(so);
1021                         KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
1022                         INP_WLOCK(inp);
1023 #ifdef IPSEC_NAT_T
1024                         up = intoudpcb(inp);
1025                         KASSERT(up != NULL, ("%s: up == NULL", __func__));
1026 #endif
1027                         switch (optval) {
1028                         case 0:
1029                                 /* Clear all UDP encap. */
1030 #ifdef IPSEC_NAT_T
1031                                 up->u_flags &= ~UF_ESPINUDP_ALL;
1032 #endif
1033                                 break;
1034 #ifdef IPSEC_NAT_T
1035                         case UDP_ENCAP_ESPINUDP:
1036                         case UDP_ENCAP_ESPINUDP_NON_IKE:
1037                                 up->u_flags &= ~UF_ESPINUDP_ALL;
1038                                 if (optval == UDP_ENCAP_ESPINUDP)
1039                                         up->u_flags |= UF_ESPINUDP;
1040                                 else if (optval == UDP_ENCAP_ESPINUDP_NON_IKE)
1041                                         up->u_flags |= UF_ESPINUDP_NON_IKE;
1042                                 break;
1043 #endif
1044                         default:
1045                                 error = EINVAL;
1046                                 break;
1047                         }
1048                         INP_WUNLOCK(inp);
1049                         break;
1050                 case UDPLITE_SEND_CSCOV:
1051                 case UDPLITE_RECV_CSCOV:
1052                         if (!isudplite) {
1053                                 INP_WUNLOCK(inp);
1054                                 error = ENOPROTOOPT;
1055                                 break;
1056                         }
1057                         INP_WUNLOCK(inp);
1058                         error = sooptcopyin(sopt, &optval, sizeof(optval),
1059                             sizeof(optval));
1060                         if (error != 0)
1061                                 break;
1062                         inp = sotoinpcb(so);
1063                         KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
1064                         INP_WLOCK(inp);
1065                         up = intoudpcb(inp);
1066                         KASSERT(up != NULL, ("%s: up == NULL", __func__));
1067                         if ((optval != 0 && optval < 8) || (optval > 65535)) {
1068                                 INP_WUNLOCK(inp);
1069                                 error = EINVAL;
1070                                 break;
1071                         }
1072                         if (sopt->sopt_name == UDPLITE_SEND_CSCOV)
1073                                 up->u_txcslen = optval;
1074                         else
1075                                 up->u_rxcslen = optval;
1076                         INP_WUNLOCK(inp);
1077                         break;
1078                 default:
1079                         INP_WUNLOCK(inp);
1080                         error = ENOPROTOOPT;
1081                         break;
1082                 }
1083                 break;
1084         case SOPT_GET:
1085                 switch (sopt->sopt_name) {
1086 #ifdef IPSEC_NAT_T
1087                 case UDP_ENCAP:
1088                         up = intoudpcb(inp);
1089                         KASSERT(up != NULL, ("%s: up == NULL", __func__));
1090                         optval = up->u_flags & UF_ESPINUDP_ALL;
1091                         INP_WUNLOCK(inp);
1092                         error = sooptcopyout(sopt, &optval, sizeof optval);
1093                         break;
1094 #endif
1095                 case UDPLITE_SEND_CSCOV:
1096                 case UDPLITE_RECV_CSCOV:
1097                         if (!isudplite) {
1098                                 INP_WUNLOCK(inp);
1099                                 error = ENOPROTOOPT;
1100                                 break;
1101                         }
1102                         up = intoudpcb(inp);
1103                         KASSERT(up != NULL, ("%s: up == NULL", __func__));
1104                         if (sopt->sopt_name == UDPLITE_SEND_CSCOV)
1105                                 optval = up->u_txcslen;
1106                         else
1107                                 optval = up->u_rxcslen;
1108                         INP_WUNLOCK(inp);
1109                         error = sooptcopyout(sopt, &optval, sizeof(optval));
1110                         break;
1111                 default:
1112                         INP_WUNLOCK(inp);
1113                         error = ENOPROTOOPT;
1114                         break;
1115                 }
1116                 break;
1117         }
1118         return (error);
1119 }
1120
1121 #ifdef INET
1122 #define UH_WLOCKED      2
1123 #define UH_RLOCKED      1
1124 #define UH_UNLOCKED     0
1125 static int
1126 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
1127     struct mbuf *control, struct thread *td)
1128 {
1129         struct udpiphdr *ui;
1130         int len = m->m_pkthdr.len;
1131         struct in_addr faddr, laddr;
1132         struct cmsghdr *cm;
1133         struct inpcbinfo *pcbinfo;
1134         struct sockaddr_in *sin, src;
1135         int cscov_partial = 0;
1136         int error = 0;
1137         int ipflags;
1138         u_short fport, lport;
1139         int unlock_udbinfo, unlock_inp;
1140         u_char tos;
1141         uint8_t pr;
1142         uint16_t cscov = 0;
1143         uint32_t flowid = 0;
1144         uint8_t flowtype = M_HASHTYPE_NONE;
1145
1146         /*
1147          * udp_output() may need to temporarily bind or connect the current
1148          * inpcb.  As such, we don't know up front whether we will need the
1149          * pcbinfo lock or not.  Do any work to decide what is needed up
1150          * front before acquiring any locks.
1151          */
1152         if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
1153                 if (control)
1154                         m_freem(control);
1155                 m_freem(m);
1156                 return (EMSGSIZE);
1157         }
1158
1159         src.sin_family = 0;
1160         sin = (struct sockaddr_in *)addr;
1161         if (sin == NULL ||
1162             (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
1163                 INP_WLOCK(inp);
1164                 unlock_inp = UH_WLOCKED;
1165         } else {
1166                 INP_RLOCK(inp);
1167                 unlock_inp = UH_RLOCKED;
1168         }
1169         tos = inp->inp_ip_tos;
1170         if (control != NULL) {
1171                 /*
1172                  * XXX: Currently, we assume all the optional information is
1173                  * stored in a single mbuf.
1174                  */
1175                 if (control->m_next) {
1176                         if (unlock_inp == UH_WLOCKED)
1177                                 INP_WUNLOCK(inp);
1178                         else
1179                                 INP_RUNLOCK(inp);
1180                         m_freem(control);
1181                         m_freem(m);
1182                         return (EINVAL);
1183                 }
1184                 for (; control->m_len > 0;
1185                     control->m_data += CMSG_ALIGN(cm->cmsg_len),
1186                     control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
1187                         cm = mtod(control, struct cmsghdr *);
1188                         if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0
1189                             || cm->cmsg_len > control->m_len) {
1190                                 error = EINVAL;
1191                                 break;
1192                         }
1193                         if (cm->cmsg_level != IPPROTO_IP)
1194                                 continue;
1195
1196                         switch (cm->cmsg_type) {
1197                         case IP_SENDSRCADDR:
1198                                 if (cm->cmsg_len !=
1199                                     CMSG_LEN(sizeof(struct in_addr))) {
1200                                         error = EINVAL;
1201                                         break;
1202                                 }
1203                                 bzero(&src, sizeof(src));
1204                                 src.sin_family = AF_INET;
1205                                 src.sin_len = sizeof(src);
1206                                 src.sin_port = inp->inp_lport;
1207                                 src.sin_addr =
1208                                     *(struct in_addr *)CMSG_DATA(cm);
1209                                 break;
1210
1211                         case IP_TOS:
1212                                 if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) {
1213                                         error = EINVAL;
1214                                         break;
1215                                 }
1216                                 tos = *(u_char *)CMSG_DATA(cm);
1217                                 break;
1218
1219                         case IP_FLOWID:
1220                                 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1221                                         error = EINVAL;
1222                                         break;
1223                                 }
1224                                 flowid = *(uint32_t *) CMSG_DATA(cm);
1225                                 break;
1226
1227                         case IP_FLOWTYPE:
1228                                 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1229                                         error = EINVAL;
1230                                         break;
1231                                 }
1232                                 flowtype = *(uint32_t *) CMSG_DATA(cm);
1233                                 break;
1234
1235 #ifdef  RSS
1236                         case IP_RSSBUCKETID:
1237                                 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1238                                         error = EINVAL;
1239                                         break;
1240                                 }
1241                                 /* This is just a placeholder for now */
1242                                 break;
1243 #endif  /* RSS */
1244                         default:
1245                                 error = ENOPROTOOPT;
1246                                 break;
1247                         }
1248                         if (error)
1249                                 break;
1250                 }
1251                 m_freem(control);
1252         }
1253         if (error) {
1254                 if (unlock_inp == UH_WLOCKED)
1255                         INP_WUNLOCK(inp);
1256                 else
1257                         INP_RUNLOCK(inp);
1258                 m_freem(m);
1259                 return (error);
1260         }
1261
1262         /*
1263          * Depending on whether or not the application has bound or connected
1264          * the socket, we may have to do varying levels of work.  The optimal
1265          * case is for a connected UDP socket, as a global lock isn't
1266          * required at all.
1267          *
1268          * In order to decide which we need, we require stability of the
1269          * inpcb binding, which we ensure by acquiring a read lock on the
1270          * inpcb.  This doesn't strictly follow the lock order, so we play
1271          * the trylock and retry game; note that we may end up with more
1272          * conservative locks than required the second time around, so later
1273          * assertions have to accept that.  Further analysis of the number of
1274          * misses under contention is required.
1275          *
1276          * XXXRW: Check that hash locking update here is correct.
1277          */
1278         pr = inp->inp_socket->so_proto->pr_protocol;
1279         pcbinfo = udp_get_inpcbinfo(pr);
1280         sin = (struct sockaddr_in *)addr;
1281         if (sin != NULL &&
1282             (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
1283                 INP_HASH_WLOCK(pcbinfo);
1284                 unlock_udbinfo = UH_WLOCKED;
1285         } else if ((sin != NULL && (
1286             (sin->sin_addr.s_addr == INADDR_ANY) ||
1287             (sin->sin_addr.s_addr == INADDR_BROADCAST) ||
1288             (inp->inp_laddr.s_addr == INADDR_ANY) ||
1289             (inp->inp_lport == 0))) ||
1290             (src.sin_family == AF_INET)) {
1291                 INP_HASH_RLOCK(pcbinfo);
1292                 unlock_udbinfo = UH_RLOCKED;
1293         } else
1294                 unlock_udbinfo = UH_UNLOCKED;
1295
1296         /*
1297          * If the IP_SENDSRCADDR control message was specified, override the
1298          * source address for this datagram.  Its use is invalidated if the
1299          * address thus specified is incomplete or clobbers other inpcbs.
1300          */
1301         laddr = inp->inp_laddr;
1302         lport = inp->inp_lport;
1303         if (src.sin_family == AF_INET) {
1304                 INP_HASH_LOCK_ASSERT(pcbinfo);
1305                 if ((lport == 0) ||
1306                     (laddr.s_addr == INADDR_ANY &&
1307                      src.sin_addr.s_addr == INADDR_ANY)) {
1308                         error = EINVAL;
1309                         goto release;
1310                 }
1311                 error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
1312                     &laddr.s_addr, &lport, td->td_ucred);
1313                 if (error)
1314                         goto release;
1315         }
1316
1317         /*
1318          * If a UDP socket has been connected, then a local address/port will
1319          * have been selected and bound.
1320          *
1321          * If a UDP socket has not been connected to, then an explicit
1322          * destination address must be used, in which case a local
1323          * address/port may not have been selected and bound.
1324          */
1325         if (sin != NULL) {
1326                 INP_LOCK_ASSERT(inp);
1327                 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1328                         error = EISCONN;
1329                         goto release;
1330                 }
1331
1332                 /*
1333                  * Jail may rewrite the destination address, so let it do
1334                  * that before we use it.
1335                  */
1336                 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1337                 if (error)
1338                         goto release;
1339
1340                 /*
1341                  * If a local address or port hasn't yet been selected, or if
1342                  * the destination address needs to be rewritten due to using
1343                  * a special INADDR_ constant, invoke in_pcbconnect_setup()
1344                  * to do the heavy lifting.  Once a port is selected, we
1345                  * commit the binding back to the socket; we also commit the
1346                  * binding of the address if in jail.
1347                  *
1348                  * If we already have a valid binding and we're not
1349                  * requesting a destination address rewrite, use a fast path.
1350                  */
1351                 if (inp->inp_laddr.s_addr == INADDR_ANY ||
1352                     inp->inp_lport == 0 ||
1353                     sin->sin_addr.s_addr == INADDR_ANY ||
1354                     sin->sin_addr.s_addr == INADDR_BROADCAST) {
1355                         INP_HASH_LOCK_ASSERT(pcbinfo);
1356                         error = in_pcbconnect_setup(inp, addr, &laddr.s_addr,
1357                             &lport, &faddr.s_addr, &fport, NULL,
1358                             td->td_ucred);
1359                         if (error)
1360                                 goto release;
1361
1362                         /*
1363                          * XXXRW: Why not commit the port if the address is
1364                          * !INADDR_ANY?
1365                          */
1366                         /* Commit the local port if newly assigned. */
1367                         if (inp->inp_laddr.s_addr == INADDR_ANY &&
1368                             inp->inp_lport == 0) {
1369                                 INP_WLOCK_ASSERT(inp);
1370                                 INP_HASH_WLOCK_ASSERT(pcbinfo);
1371                                 /*
1372                                  * Remember addr if jailed, to prevent
1373                                  * rebinding.
1374                                  */
1375                                 if (prison_flag(td->td_ucred, PR_IP4))
1376                                         inp->inp_laddr = laddr;
1377                                 inp->inp_lport = lport;
1378                                 if (in_pcbinshash(inp) != 0) {
1379                                         inp->inp_lport = 0;
1380                                         error = EAGAIN;
1381                                         goto release;
1382                                 }
1383                                 inp->inp_flags |= INP_ANONPORT;
1384                         }
1385                 } else {
1386                         faddr = sin->sin_addr;
1387                         fport = sin->sin_port;
1388                 }
1389         } else {
1390                 INP_LOCK_ASSERT(inp);
1391                 faddr = inp->inp_faddr;
1392                 fport = inp->inp_fport;
1393                 if (faddr.s_addr == INADDR_ANY) {
1394                         error = ENOTCONN;
1395                         goto release;
1396                 }
1397         }
1398
1399         /*
1400          * Calculate data length and get a mbuf for UDP, IP, and possible
1401          * link-layer headers.  Immediate slide the data pointer back forward
1402          * since we won't use that space at this layer.
1403          */
1404         M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT);
1405         if (m == NULL) {
1406                 error = ENOBUFS;
1407                 goto release;
1408         }
1409         m->m_data += max_linkhdr;
1410         m->m_len -= max_linkhdr;
1411         m->m_pkthdr.len -= max_linkhdr;
1412
1413         /*
1414          * Fill in mbuf with extended UDP header and addresses and length put
1415          * into network format.
1416          */
1417         ui = mtod(m, struct udpiphdr *);
1418         bzero(ui->ui_x1, sizeof(ui->ui_x1));    /* XXX still needed? */
1419         ui->ui_pr = pr;
1420         ui->ui_src = laddr;
1421         ui->ui_dst = faddr;
1422         ui->ui_sport = lport;
1423         ui->ui_dport = fport;
1424         ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
1425         if (pr == IPPROTO_UDPLITE) {
1426                 struct udpcb *up;
1427                 uint16_t plen;
1428
1429                 up = intoudpcb(inp);
1430                 cscov = up->u_txcslen;
1431                 plen = (u_short)len + sizeof(struct udphdr);
1432                 if (cscov >= plen)
1433                         cscov = 0;
1434                 ui->ui_len = htons(plen);
1435                 ui->ui_ulen = htons(cscov);
1436                 /*
1437                  * For UDP-Lite, checksum coverage length of zero means
1438                  * the entire UDPLite packet is covered by the checksum.
1439                  */
1440                 cscov_partial = (cscov == 0) ? 0 : 1;
1441         } else
1442                 ui->ui_v = IPVERSION << 4;
1443
1444         /*
1445          * Set the Don't Fragment bit in the IP header.
1446          */
1447         if (inp->inp_flags & INP_DONTFRAG) {
1448                 struct ip *ip;
1449
1450                 ip = (struct ip *)&ui->ui_i;
1451                 ip->ip_off |= htons(IP_DF);
1452         }
1453
1454         ipflags = 0;
1455         if (inp->inp_socket->so_options & SO_DONTROUTE)
1456                 ipflags |= IP_ROUTETOIF;
1457         if (inp->inp_socket->so_options & SO_BROADCAST)
1458                 ipflags |= IP_ALLOWBROADCAST;
1459         if (inp->inp_flags & INP_ONESBCAST)
1460                 ipflags |= IP_SENDONES;
1461
1462 #ifdef MAC
1463         mac_inpcb_create_mbuf(inp, m);
1464 #endif
1465
1466         /*
1467          * Set up checksum and output datagram.
1468          */
1469         ui->ui_sum = 0;
1470         if (pr == IPPROTO_UDPLITE) {
1471                 if (inp->inp_flags & INP_ONESBCAST)
1472                         faddr.s_addr = INADDR_BROADCAST;
1473                 if (cscov_partial) {
1474                         if ((ui->ui_sum = in_cksum(m, sizeof(struct ip) + cscov)) == 0)
1475                                 ui->ui_sum = 0xffff;
1476                 } else {
1477                         if ((ui->ui_sum = in_cksum(m, sizeof(struct udpiphdr) + len)) == 0)
1478                                 ui->ui_sum = 0xffff;
1479                 }
1480         } else if (V_udp_cksum) {
1481                 if (inp->inp_flags & INP_ONESBCAST)
1482                         faddr.s_addr = INADDR_BROADCAST;
1483                 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
1484                     htons((u_short)len + sizeof(struct udphdr) + pr));
1485                 m->m_pkthdr.csum_flags = CSUM_UDP;
1486                 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1487         }
1488         ((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len);
1489         ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl;    /* XXX */
1490         ((struct ip *)ui)->ip_tos = tos;                /* XXX */
1491         UDPSTAT_INC(udps_opackets);
1492
1493         /*
1494          * Setup flowid / RSS information for outbound socket.
1495          *
1496          * Once the UDP code decides to set a flowid some other way,
1497          * this allows the flowid to be overridden by userland.
1498          */
1499         if (flowtype != M_HASHTYPE_NONE) {
1500                 m->m_pkthdr.flowid = flowid;
1501                 M_HASHTYPE_SET(m, flowtype);
1502 #ifdef  RSS
1503         } else {
1504                 uint32_t hash_val, hash_type;
1505                 /*
1506                  * Calculate an appropriate RSS hash for UDP and
1507                  * UDP Lite.
1508                  *
1509                  * The called function will take care of figuring out
1510                  * whether a 2-tuple or 4-tuple hash is required based
1511                  * on the currently configured scheme.
1512                  *
1513                  * Later later on connected socket values should be
1514                  * cached in the inpcb and reused, rather than constantly
1515                  * re-calculating it.
1516                  *
1517                  * UDP Lite is a different protocol number and will
1518                  * likely end up being hashed as a 2-tuple until
1519                  * RSS / NICs grow UDP Lite protocol awareness.
1520                  */
1521                 if (rss_proto_software_hash_v4(faddr, laddr, fport, lport,
1522                     pr, &hash_val, &hash_type) == 0) {
1523                         m->m_pkthdr.flowid = hash_val;
1524                         M_HASHTYPE_SET(m, hash_type);
1525                 }
1526 #endif
1527         }
1528
1529 #ifdef  RSS
1530         /*
1531          * Don't override with the inp cached flowid value.
1532          *
1533          * Depending upon the kind of send being done, the inp
1534          * flowid/flowtype values may actually not be appropriate
1535          * for this particular socket send.
1536          *
1537          * We should either leave the flowid at zero (which is what is
1538          * currently done) or set it to some software generated
1539          * hash value based on the packet contents.
1540          */
1541         ipflags |= IP_NODEFAULTFLOWID;
1542 #endif  /* RSS */
1543
1544         if (unlock_udbinfo == UH_WLOCKED)
1545                 INP_HASH_WUNLOCK(pcbinfo);
1546         else if (unlock_udbinfo == UH_RLOCKED)
1547                 INP_HASH_RUNLOCK(pcbinfo);
1548         UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u);
1549         error = ip_output(m, inp->inp_options,
1550             (unlock_inp == UH_WLOCKED ? &inp->inp_route : NULL), ipflags,
1551             inp->inp_moptions, inp);
1552         if (unlock_inp == UH_WLOCKED)
1553                 INP_WUNLOCK(inp);
1554         else
1555                 INP_RUNLOCK(inp);
1556         return (error);
1557
1558 release:
1559         if (unlock_udbinfo == UH_WLOCKED) {
1560                 INP_HASH_WUNLOCK(pcbinfo);
1561                 INP_WUNLOCK(inp);
1562         } else if (unlock_udbinfo == UH_RLOCKED) {
1563                 INP_HASH_RUNLOCK(pcbinfo);
1564                 INP_RUNLOCK(inp);
1565         } else
1566                 INP_RUNLOCK(inp);
1567         m_freem(m);
1568         return (error);
1569 }
1570
1571
1572 #if defined(IPSEC) && defined(IPSEC_NAT_T)
1573 /*
1574  * Potentially decap ESP in UDP frame.  Check for an ESP header
1575  * and optional marker; if present, strip the UDP header and
1576  * push the result through IPSec.
1577  *
1578  * Returns mbuf to be processed (potentially re-allocated) or
1579  * NULL if consumed and/or processed.
1580  */
1581 static struct mbuf *
1582 udp4_espdecap(struct inpcb *inp, struct mbuf *m, int off)
1583 {
1584         size_t minlen, payload, skip, iphlen;
1585         caddr_t data;
1586         struct udpcb *up;
1587         struct m_tag *tag;
1588         struct udphdr *udphdr;
1589         struct ip *ip;
1590
1591         INP_RLOCK_ASSERT(inp);
1592
1593         /*
1594          * Pull up data so the longest case is contiguous:
1595          *    IP/UDP hdr + non ESP marker + ESP hdr.
1596          */
1597         minlen = off + sizeof(uint64_t) + sizeof(struct esp);
1598         if (minlen > m->m_pkthdr.len)
1599                 minlen = m->m_pkthdr.len;
1600         if ((m = m_pullup(m, minlen)) == NULL) {
1601                 IPSECSTAT_INC(ips_in_inval);
1602                 return (NULL);          /* Bypass caller processing. */
1603         }
1604         data = mtod(m, caddr_t);        /* Points to ip header. */
1605         payload = m->m_len - off;       /* Size of payload. */
1606
1607         if (payload == 1 && data[off] == '\xff')
1608                 return (m);             /* NB: keepalive packet, no decap. */
1609
1610         up = intoudpcb(inp);
1611         KASSERT(up != NULL, ("%s: udpcb NULL", __func__));
1612         KASSERT((up->u_flags & UF_ESPINUDP_ALL) != 0,
1613             ("u_flags 0x%x", up->u_flags));
1614
1615         /*
1616          * Check that the payload is large enough to hold an
1617          * ESP header and compute the amount of data to remove.
1618          *
1619          * NB: the caller has already done a pullup for us.
1620          * XXX can we assume alignment and eliminate bcopys?
1621          */
1622         if (up->u_flags & UF_ESPINUDP_NON_IKE) {
1623                 /*
1624                  * draft-ietf-ipsec-nat-t-ike-0[01].txt and
1625                  * draft-ietf-ipsec-udp-encaps-(00/)01.txt, ignoring
1626                  * possible AH mode non-IKE marker+non-ESP marker
1627                  * from draft-ietf-ipsec-udp-encaps-00.txt.
1628                  */
1629                 uint64_t marker;
1630
1631                 if (payload <= sizeof(uint64_t) + sizeof(struct esp))
1632                         return (m);     /* NB: no decap. */
1633                 bcopy(data + off, &marker, sizeof(uint64_t));
1634                 if (marker != 0)        /* Non-IKE marker. */
1635                         return (m);     /* NB: no decap. */
1636                 skip = sizeof(uint64_t) + sizeof(struct udphdr);
1637         } else {
1638                 uint32_t spi;
1639
1640                 if (payload <= sizeof(struct esp)) {
1641                         IPSECSTAT_INC(ips_in_inval);
1642                         m_freem(m);
1643                         return (NULL);  /* Discard. */
1644                 }
1645                 bcopy(data + off, &spi, sizeof(uint32_t));
1646                 if (spi == 0)           /* Non-ESP marker. */
1647                         return (m);     /* NB: no decap. */
1648                 skip = sizeof(struct udphdr);
1649         }
1650
1651         /*
1652          * Setup a PACKET_TAG_IPSEC_NAT_T_PORT tag to remember
1653          * the UDP ports. This is required if we want to select
1654          * the right SPD for multiple hosts behind same NAT.
1655          *
1656          * NB: ports are maintained in network byte order everywhere
1657          *     in the NAT-T code.
1658          */
1659         tag = m_tag_get(PACKET_TAG_IPSEC_NAT_T_PORTS,
1660                 2 * sizeof(uint16_t), M_NOWAIT);
1661         if (tag == NULL) {
1662                 IPSECSTAT_INC(ips_in_nomem);
1663                 m_freem(m);
1664                 return (NULL);          /* Discard. */
1665         }
1666         iphlen = off - sizeof(struct udphdr);
1667         udphdr = (struct udphdr *)(data + iphlen);
1668         ((uint16_t *)(tag + 1))[0] = udphdr->uh_sport;
1669         ((uint16_t *)(tag + 1))[1] = udphdr->uh_dport;
1670         m_tag_prepend(m, tag);
1671
1672         /*
1673          * Remove the UDP header (and possibly the non ESP marker)
1674          * IP header length is iphlen
1675          * Before:
1676          *   <--- off --->
1677          *   +----+------+-----+
1678          *   | IP |  UDP | ESP |
1679          *   +----+------+-----+
1680          *        <-skip->
1681          * After:
1682          *          +----+-----+
1683          *          | IP | ESP |
1684          *          +----+-----+
1685          *   <-skip->
1686          */
1687         ovbcopy(data, data + skip, iphlen);
1688         m_adj(m, skip);
1689
1690         ip = mtod(m, struct ip *);
1691         ip->ip_len = htons(ntohs(ip->ip_len) - skip);
1692         ip->ip_p = IPPROTO_ESP;
1693
1694         /*
1695          * We cannot yet update the cksums so clear any
1696          * h/w cksum flags as they are no longer valid.
1697          */
1698         if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID)
1699                 m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID|CSUM_PSEUDO_HDR);
1700
1701         (void) ipsec_common_input(m, iphlen, offsetof(struct ip, ip_p),
1702                                 AF_INET, ip->ip_p);
1703         return (NULL);                  /* NB: consumed, bypass processing. */
1704 }
1705 #endif /* defined(IPSEC) && defined(IPSEC_NAT_T) */
1706
1707 static void
1708 udp_abort(struct socket *so)
1709 {
1710         struct inpcb *inp;
1711         struct inpcbinfo *pcbinfo;
1712
1713         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1714         inp = sotoinpcb(so);
1715         KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
1716         INP_WLOCK(inp);
1717         if (inp->inp_faddr.s_addr != INADDR_ANY) {
1718                 INP_HASH_WLOCK(pcbinfo);
1719                 in_pcbdisconnect(inp);
1720                 inp->inp_laddr.s_addr = INADDR_ANY;
1721                 INP_HASH_WUNLOCK(pcbinfo);
1722                 soisdisconnected(so);
1723         }
1724         INP_WUNLOCK(inp);
1725 }
1726
1727 static int
1728 udp_attach(struct socket *so, int proto, struct thread *td)
1729 {
1730         struct inpcb *inp;
1731         struct inpcbinfo *pcbinfo;
1732         int error;
1733
1734         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1735         inp = sotoinpcb(so);
1736         KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
1737         error = soreserve(so, udp_sendspace, udp_recvspace);
1738         if (error)
1739                 return (error);
1740         INP_INFO_WLOCK(pcbinfo);
1741         error = in_pcballoc(so, pcbinfo);
1742         if (error) {
1743                 INP_INFO_WUNLOCK(pcbinfo);
1744                 return (error);
1745         }
1746
1747         inp = sotoinpcb(so);
1748         inp->inp_vflag |= INP_IPV4;
1749         inp->inp_ip_ttl = V_ip_defttl;
1750
1751         error = udp_newudpcb(inp);
1752         if (error) {
1753                 in_pcbdetach(inp);
1754                 in_pcbfree(inp);
1755                 INP_INFO_WUNLOCK(pcbinfo);
1756                 return (error);
1757         }
1758
1759         INP_WUNLOCK(inp);
1760         INP_INFO_WUNLOCK(pcbinfo);
1761         return (0);
1762 }
1763 #endif /* INET */
1764
1765 int
1766 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f, udp_tun_icmp_t i, void *ctx)
1767 {
1768         struct inpcb *inp;
1769         struct udpcb *up;
1770
1771         KASSERT(so->so_type == SOCK_DGRAM,
1772             ("udp_set_kernel_tunneling: !dgram"));
1773         inp = sotoinpcb(so);
1774         KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL"));
1775         INP_WLOCK(inp);
1776         up = intoudpcb(inp);
1777         if ((up->u_tun_func != NULL) ||
1778             (up->u_icmp_func != NULL)) {
1779                 INP_WUNLOCK(inp);
1780                 return (EBUSY);
1781         }
1782         up->u_tun_func = f;
1783         up->u_icmp_func = i;
1784         up->u_tun_ctx = ctx;
1785         INP_WUNLOCK(inp);
1786         return (0);
1787 }
1788
1789 #ifdef INET
1790 static int
1791 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1792 {
1793         struct inpcb *inp;
1794         struct inpcbinfo *pcbinfo;
1795         int error;
1796
1797         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1798         inp = sotoinpcb(so);
1799         KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1800         INP_WLOCK(inp);
1801         INP_HASH_WLOCK(pcbinfo);
1802         error = in_pcbbind(inp, nam, td->td_ucred);
1803         INP_HASH_WUNLOCK(pcbinfo);
1804         INP_WUNLOCK(inp);
1805         return (error);
1806 }
1807
1808 static void
1809 udp_close(struct socket *so)
1810 {
1811         struct inpcb *inp;
1812         struct inpcbinfo *pcbinfo;
1813
1814         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1815         inp = sotoinpcb(so);
1816         KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1817         INP_WLOCK(inp);
1818         if (inp->inp_faddr.s_addr != INADDR_ANY) {
1819                 INP_HASH_WLOCK(pcbinfo);
1820                 in_pcbdisconnect(inp);
1821                 inp->inp_laddr.s_addr = INADDR_ANY;
1822                 INP_HASH_WUNLOCK(pcbinfo);
1823                 soisdisconnected(so);
1824         }
1825         INP_WUNLOCK(inp);
1826 }
1827
1828 static int
1829 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1830 {
1831         struct inpcb *inp;
1832         struct inpcbinfo *pcbinfo;
1833         struct sockaddr_in *sin;
1834         int error;
1835
1836         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1837         inp = sotoinpcb(so);
1838         KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1839         INP_WLOCK(inp);
1840         if (inp->inp_faddr.s_addr != INADDR_ANY) {
1841                 INP_WUNLOCK(inp);
1842                 return (EISCONN);
1843         }
1844         sin = (struct sockaddr_in *)nam;
1845         error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1846         if (error != 0) {
1847                 INP_WUNLOCK(inp);
1848                 return (error);
1849         }
1850         INP_HASH_WLOCK(pcbinfo);
1851         error = in_pcbconnect(inp, nam, td->td_ucred);
1852         INP_HASH_WUNLOCK(pcbinfo);
1853         if (error == 0)
1854                 soisconnected(so);
1855         INP_WUNLOCK(inp);
1856         return (error);
1857 }
1858
1859 static void
1860 udp_detach(struct socket *so)
1861 {
1862         struct inpcb *inp;
1863         struct inpcbinfo *pcbinfo;
1864         struct udpcb *up;
1865
1866         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1867         inp = sotoinpcb(so);
1868         KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1869         KASSERT(inp->inp_faddr.s_addr == INADDR_ANY,
1870             ("udp_detach: not disconnected"));
1871         INP_INFO_WLOCK(pcbinfo);
1872         INP_WLOCK(inp);
1873         up = intoudpcb(inp);
1874         KASSERT(up != NULL, ("%s: up == NULL", __func__));
1875         inp->inp_ppcb = NULL;
1876         in_pcbdetach(inp);
1877         in_pcbfree(inp);
1878         INP_INFO_WUNLOCK(pcbinfo);
1879         udp_discardcb(up);
1880 }
1881
1882 static int
1883 udp_disconnect(struct socket *so)
1884 {
1885         struct inpcb *inp;
1886         struct inpcbinfo *pcbinfo;
1887
1888         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1889         inp = sotoinpcb(so);
1890         KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1891         INP_WLOCK(inp);
1892         if (inp->inp_faddr.s_addr == INADDR_ANY) {
1893                 INP_WUNLOCK(inp);
1894                 return (ENOTCONN);
1895         }
1896         INP_HASH_WLOCK(pcbinfo);
1897         in_pcbdisconnect(inp);
1898         inp->inp_laddr.s_addr = INADDR_ANY;
1899         INP_HASH_WUNLOCK(pcbinfo);
1900         SOCK_LOCK(so);
1901         so->so_state &= ~SS_ISCONNECTED;                /* XXX */
1902         SOCK_UNLOCK(so);
1903         INP_WUNLOCK(inp);
1904         return (0);
1905 }
1906
1907 static int
1908 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1909     struct mbuf *control, struct thread *td)
1910 {
1911         struct inpcb *inp;
1912
1913         inp = sotoinpcb(so);
1914         KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1915         return (udp_output(inp, m, addr, control, td));
1916 }
1917 #endif /* INET */
1918
1919 int
1920 udp_shutdown(struct socket *so)
1921 {
1922         struct inpcb *inp;
1923
1924         inp = sotoinpcb(so);
1925         KASSERT(inp != NULL, ("udp_shutdown: inp == NULL"));
1926         INP_WLOCK(inp);
1927         socantsendmore(so);
1928         INP_WUNLOCK(inp);
1929         return (0);
1930 }
1931
1932 #ifdef INET
1933 struct pr_usrreqs udp_usrreqs = {
1934         .pru_abort =            udp_abort,
1935         .pru_attach =           udp_attach,
1936         .pru_bind =             udp_bind,
1937         .pru_connect =          udp_connect,
1938         .pru_control =          in_control,
1939         .pru_detach =           udp_detach,
1940         .pru_disconnect =       udp_disconnect,
1941         .pru_peeraddr =         in_getpeeraddr,
1942         .pru_send =             udp_send,
1943         .pru_soreceive =        soreceive_dgram,
1944         .pru_sosend =           sosend_dgram,
1945         .pru_shutdown =         udp_shutdown,
1946         .pru_sockaddr =         in_getsockaddr,
1947         .pru_sosetlabel =       in_pcbsosetlabel,
1948         .pru_close =            udp_close,
1949 };
1950 #endif /* INET */