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  * All rights reserved.
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  * 1. Redistributions of source code must retain the above copyright
  11  *    notice, this list of conditions and the following disclaimer.
  12  * 2. Redistributions in binary form must reproduce the above copyright
  13  *    notice, this list of conditions and the following disclaimer in the
  14  *    documentation and/or other materials provided with the distribution.
  15  * 4. Neither the name of the University nor the names of its contributors
  16  *    may be used to endorse or promote products derived from this software
  17  *    without specific prior written permission.
  18  *
  19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  29  * SUCH DAMAGE.
  30  *
  31  *      @(#)udp_usrreq.c        8.6 (Berkeley) 5/23/95
  32  */
  33
  34 #include <sys/cdefs.h>
  35 __FBSDID("$FreeBSD$");
  36
  37 #include "opt_ipfw.h"
  38 #include "opt_inet6.h"
  39 #include "opt_ipsec.h"
  40
  41 #include <sys/param.h>
  42 #include <sys/domain.h>
  43 #include <sys/eventhandler.h>
  44 #include <sys/jail.h>
  45 #include <sys/kernel.h>
  46 #include <sys/lock.h>
  47 #include <sys/malloc.h>
  48 #include <sys/mbuf.h>
  49 #include <sys/priv.h>
  50 #include <sys/proc.h>
  51 #include <sys/protosw.h>
  52 #include <sys/signalvar.h>
  53 #include <sys/socket.h>
  54 #include <sys/socketvar.h>
  55 #include <sys/sx.h>
  56 #include <sys/sysctl.h>
  57 #include <sys/syslog.h>
  58 #include <sys/systm.h>
  59
  60 #include <vm/uma.h>
  61
  62 #include <net/if.h>
  63 #include <net/route.h>
  64
  65 #include <netinet/in.h>
  66 #include <netinet/in_pcb.h>
  67 #include <netinet/in_systm.h>
  68 #include <netinet/in_var.h>
  69 #include <netinet/ip.h>
  70 #ifdef INET6
  71 #include <netinet/ip6.h>
  72 #endif
  73 #include <netinet/ip_icmp.h>
  74 #include <netinet/icmp_var.h>
  75 #include <netinet/ip_var.h>
  76 #include <netinet/ip_options.h>
  77 #ifdef INET6
  78 #include <netinet6/ip6_var.h>
  79 #endif
  80 #include <netinet/udp.h>
  81 #include <netinet/udp_var.h>
  82
  83 #ifdef IPSEC
  84 #include <netipsec/ipsec.h>
  85 #include <netipsec/esp.h>
  86 #endif
  87
  88 #include <machine/in_cksum.h>
  89
  90 #include <security/mac/mac_framework.h>
  91
  92 /*
  93  * UDP protocol implementation.
  94  * Per RFC 768, August, 1980.
  95  */
  96
  97 /*
  98  * BSD 4.2 defaulted the udp checksum to be off.  Turning off udp checksums
  99  * removes the only data integrity mechanism for packets and malformed
 100  * packets that would otherwise be discarded due to bad checksums, and may
 101  * cause problems (especially for NFS data blocks).
 102  */
 103 VNET_DEFINE(int, udp_cksum) = 1;
 104 SYSCTL_VNET_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_RW,
 105     &VNET_NAME(udp_cksum), 0, "compute udp checksum");
 106
 107 int     udp_log_in_vain = 0;
 108 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW,
 109     &udp_log_in_vain, 0, "Log all incoming UDP packets");
 110
 111 VNET_DEFINE(int, udp_blackhole) = 0;
 112 SYSCTL_VNET_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_RW,
 113     &VNET_NAME(udp_blackhole), 0,
 114     "Do not send port unreachables for refused connects");
 115
 116 u_long  udp_sendspace = 9216;           /* really max datagram size */
 117                                         /* 40 1K datagrams */
 118 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
 119     &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
 120
 121 u_long  udp_recvspace = 40 * (1024 +
 122 #ifdef INET6
 123                                       sizeof(struct sockaddr_in6)
 124 #else
 125                                       sizeof(struct sockaddr_in)
 126 #endif
 127                                       );
 128
 129 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
 130     &udp_recvspace, 0, "Maximum space for incoming UDP datagrams");
 131
 132 VNET_DEFINE(struct inpcbhead, udb);             /* from udp_var.h */
 133 VNET_DEFINE(struct inpcbinfo, udbinfo);
 134 static VNET_DEFINE(uma_zone_t, udpcb_zone);
 135 #define V_udpcb_zone                    VNET(udpcb_zone)
 136
 137 #ifndef UDBHASHSIZE
 138 #define UDBHASHSIZE     128
 139 #endif
 140
 141 VNET_DEFINE(struct udpstat, udpstat);           /* from udp_var.h */
 142 SYSCTL_VNET_STRUCT(_net_inet_udp, UDPCTL_STATS, stats, CTLFLAG_RW,
 143     &VNET_NAME(udpstat), udpstat,
 144     "UDP statistics (struct udpstat, netinet/udp_var.h)");
 145
 146 static void     udp_detach(struct socket *so);
 147 static int      udp_output(struct inpcb *, struct mbuf *, struct sockaddr *,
 148                     struct mbuf *, struct thread *);
 149 #ifdef IPSEC
 150 #ifdef IPSEC_NAT_T
 151 #define UF_ESPINUDP_ALL (UF_ESPINUDP_NON_IKE|UF_ESPINUDP)
 152 #ifdef INET
 153 static struct mbuf *udp4_espdecap(struct inpcb *, struct mbuf *, int);
 154 #endif
 155 #endif /* IPSEC_NAT_T */
 156 #endif /* IPSEC */
 157
 158 static void
 159 udp_zone_change(void *tag)
 160 {
 161
 162         uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets);
 163         uma_zone_set_max(V_udpcb_zone, maxsockets);
 164 }
 165
 166 static int
 167 udp_inpcb_init(void *mem, int size, int flags)
 168 {
 169         struct inpcb *inp;
 170
 171         inp = mem;
 172         INP_LOCK_INIT(inp, "inp", "udpinp");
 173         return (0);
 174 }
 175
 176 void
 177 udp_init(void)
 178 {
 179
 180
 181         INP_INFO_LOCK_INIT(&V_udbinfo, "udp");
 182         LIST_INIT(&V_udb);
 183 #ifdef VIMAGE
 184         V_udbinfo.ipi_vnet = curvnet;
 185 #endif
 186         V_udbinfo.ipi_listhead = &V_udb;
 187         V_udbinfo.ipi_hashbase = hashinit(UDBHASHSIZE, M_PCB,
 188             &V_udbinfo.ipi_hashmask);
 189         V_udbinfo.ipi_porthashbase = hashinit(UDBHASHSIZE, M_PCB,
 190             &V_udbinfo.ipi_porthashmask);
 191         V_udbinfo.ipi_zone = uma_zcreate("udp_inpcb", sizeof(struct inpcb),
 192             NULL, NULL, udp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
 193         uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets);
 194
 195         V_udpcb_zone = uma_zcreate("udpcb", sizeof(struct udpcb),
 196             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
 197         uma_zone_set_max(V_udpcb_zone, maxsockets);
 198
 199         EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL,
 200             EVENTHANDLER_PRI_ANY);
 201 }
 202
 203 /*
 204  * Kernel module interface for updating udpstat.  The argument is an index
 205  * into udpstat treated as an array of u_long.  While this encodes the
 206  * general layout of udpstat into the caller, it doesn't encode its location,
 207  * so that future changes to add, for example, per-CPU stats support won't
 208  * cause binary compatibility problems for kernel modules.
 209  */
 210 void
 211 kmod_udpstat_inc(int statnum)
 212 {
 213
 214         (*((u_long *)&V_udpstat + statnum))++;
 215 }
 216
 217 int
 218 udp_newudpcb(struct inpcb *inp)
 219 {
 220         struct udpcb *up;
 221
 222         up = uma_zalloc(V_udpcb_zone, M_NOWAIT | M_ZERO);
 223         if (up == NULL)
 224                 return (ENOBUFS);
 225         inp->inp_ppcb = up;
 226         return (0);
 227 }
 228
 229 void
 230 udp_discardcb(struct udpcb *up)
 231 {
 232
 233         uma_zfree(V_udpcb_zone, up);
 234 }
 235
 236 #ifdef VIMAGE
 237 void
 238 udp_destroy(void)
 239 {
 240
 241         hashdestroy(V_udbinfo.ipi_hashbase, M_PCB,
 242             V_udbinfo.ipi_hashmask);
 243         hashdestroy(V_udbinfo.ipi_porthashbase, M_PCB,
 244             V_udbinfo.ipi_porthashmask);
 245
 246         uma_zdestroy(V_udpcb_zone);
 247         uma_zdestroy(V_udbinfo.ipi_zone);
 248         INP_INFO_LOCK_DESTROY(&V_udbinfo);
 249 }
 250 #endif
 251
 252 /*
 253  * Subroutine of udp_input(), which appends the provided mbuf chain to the
 254  * passed pcb/socket.  The caller must provide a sockaddr_in via udp_in that
 255  * contains the source address.  If the socket ends up being an IPv6 socket,
 256  * udp_append() will convert to a sockaddr_in6 before passing the address
 257  * into the socket code.
 258  */
 259 static void
 260 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off,
 261     struct sockaddr_in *udp_in)
 262 {
 263         struct sockaddr *append_sa;
 264         struct socket *so;
 265         struct mbuf *opts = 0;
 266 #ifdef INET6
 267         struct sockaddr_in6 udp_in6;
 268 #endif
 269 #ifdef IPSEC
 270 #ifdef IPSEC_NAT_T
 271 #ifdef INET
 272         struct udpcb *up;
 273 #endif
 274 #endif
 275 #endif
 276
 277         INP_RLOCK_ASSERT(inp);
 278
 279 #ifdef IPSEC
 280         /* Check AH/ESP integrity. */
 281         if (ipsec4_in_reject(n, inp)) {
 282                 m_freem(n);
 283                 V_ipsec4stat.in_polvio++;
 284                 return;
 285         }
 286 #ifdef IPSEC_NAT_T
 287 #ifdef INET
 288         up = intoudpcb(inp);
 289         KASSERT(up != NULL, ("%s: udpcb NULL", __func__));
 290         if (up->u_flags & UF_ESPINUDP_ALL) {    /* IPSec UDP encaps. */
 291                 n = udp4_espdecap(inp, n, off);
 292                 if (n == NULL)                          /* Consumed. */
 293                         return;
 294         }
 295 #endif /* INET */
 296 #endif /* IPSEC_NAT_T */
 297 #endif /* IPSEC */
 298 #ifdef MAC
 299         if (mac_inpcb_check_deliver(inp, n) != 0) {
 300                 m_freem(n);
 301                 return;
 302         }
 303 #endif
 304         if (inp->inp_flags & INP_CONTROLOPTS ||
 305             inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
 306 #ifdef INET6
 307                 if (inp->inp_vflag & INP_IPV6)
 308                         (void)ip6_savecontrol_v4(inp, n, &opts, NULL);
 309                 else
 310 #endif
 311                         ip_savecontrol(inp, &opts, ip, n);
 312         }
 313 #ifdef INET6
 314         if (inp->inp_vflag & INP_IPV6) {
 315                 bzero(&udp_in6, sizeof(udp_in6));
 316                 udp_in6.sin6_len = sizeof(udp_in6);
 317                 udp_in6.sin6_family = AF_INET6;
 318                 in6_sin_2_v4mapsin6(udp_in, &udp_in6);
 319                 append_sa = (struct sockaddr *)&udp_in6;
 320         } else
 321 #endif
 322                 append_sa = (struct sockaddr *)udp_in;
 323         m_adj(n, off);
 324
 325         so = inp->inp_socket;
 326         SOCKBUF_LOCK(&so->so_rcv);
 327         if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) {
 328                 SOCKBUF_UNLOCK(&so->so_rcv);
 329                 m_freem(n);
 330                 if (opts)
 331                         m_freem(opts);
 332                 UDPSTAT_INC(udps_fullsock);
 333         } else
 334                 sorwakeup_locked(so);
 335 }
 336
 337 void
 338 udp_input(struct mbuf *m, int off)
 339 {
 340         int iphlen = off;
 341         struct ip *ip;
 342         struct udphdr *uh;
 343         struct ifnet *ifp;
 344         struct inpcb *inp;
 345         struct udpcb *up;
 346         int len;
 347         struct ip save_ip;
 348         struct sockaddr_in udp_in;
 349 #ifdef IPFIREWALL_FORWARD
 350         struct m_tag *fwd_tag;
 351 #endif
 352
 353         ifp = m->m_pkthdr.rcvif;
 354         UDPSTAT_INC(udps_ipackets);
 355
 356         /*
 357          * Strip IP options, if any; should skip this, make available to
 358          * user, and use on returned packets, but we don't yet have a way to
 359          * check the checksum with options still present.
 360          */
 361         if (iphlen > sizeof (struct ip)) {
 362                 ip_stripoptions(m, (struct mbuf *)0);
 363                 iphlen = sizeof(struct ip);
 364         }
 365
 366         /*
 367          * Get IP and UDP header together in first mbuf.
 368          */
 369         ip = mtod(m, struct ip *);
 370         if (m->m_len < iphlen + sizeof(struct udphdr)) {
 371                 if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == 0) {
 372                         UDPSTAT_INC(udps_hdrops);
 373                         return;
 374                 }
 375                 ip = mtod(m, struct ip *);
 376         }
 377         uh = (struct udphdr *)((caddr_t)ip + iphlen);
 378
 379         /*
 380          * Destination port of 0 is illegal, based on RFC768.
 381          */
 382         if (uh->uh_dport == 0)
 383                 goto badunlocked;
 384
 385         /*
 386          * Construct sockaddr format source address.  Stuff source address
 387          * and datagram in user buffer.
 388          */
 389         bzero(&udp_in, sizeof(udp_in));
 390         udp_in.sin_len = sizeof(udp_in);
 391         udp_in.sin_family = AF_INET;
 392         udp_in.sin_port = uh->uh_sport;
 393         udp_in.sin_addr = ip->ip_src;
 394
 395         /*
 396          * Make mbuf data length reflect UDP length.  If not enough data to
 397          * reflect UDP length, drop.
 398          */
 399         len = ntohs((u_short)uh->uh_ulen);
 400         if (ip->ip_len != len) {
 401                 if (len > ip->ip_len || len < sizeof(struct udphdr)) {
 402                         UDPSTAT_INC(udps_badlen);
 403                         goto badunlocked;
 404                 }
 405                 m_adj(m, len - ip->ip_len);
 406                 /* ip->ip_len = len; */
 407         }
 408
 409         /*
 410          * Save a copy of the IP header in case we want restore it for
 411          * sending an ICMP error message in response.
 412          */
 413         if (!V_udp_blackhole)
 414                 save_ip = *ip;
 415         else
 416                 memset(&save_ip, 0, sizeof(save_ip));
 417
 418         /*
 419          * Checksum extended UDP header and data.
 420          */
 421         if (uh->uh_sum) {
 422                 u_short uh_sum;
 423
 424                 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
 425                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
 426                                 uh_sum = m->m_pkthdr.csum_data;
 427                         else
 428                                 uh_sum = in_pseudo(ip->ip_src.s_addr,
 429                                     ip->ip_dst.s_addr, htonl((u_short)len +
 430                                     m->m_pkthdr.csum_data + IPPROTO_UDP));
 431                         uh_sum ^= 0xffff;
 432                 } else {
 433                         char b[9];
 434
 435                         bcopy(((struct ipovly *)ip)->ih_x1, b, 9);
 436                         bzero(((struct ipovly *)ip)->ih_x1, 9);
 437                         ((struct ipovly *)ip)->ih_len = uh->uh_ulen;
 438                         uh_sum = in_cksum(m, len + sizeof (struct ip));
 439                         bcopy(b, ((struct ipovly *)ip)->ih_x1, 9);
 440                 }
 441                 if (uh_sum) {
 442                         UDPSTAT_INC(udps_badsum);
 443                         m_freem(m);
 444                         return;
 445                 }
 446         } else
 447                 UDPSTAT_INC(udps_nosum);
 448
 449 #ifdef IPFIREWALL_FORWARD
 450         /*
 451          * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
 452          */
 453         fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 454         if (fwd_tag != NULL) {
 455                 struct sockaddr_in *next_hop;
 456
 457                 /*
 458                  * Do the hack.
 459                  */
 460                 next_hop = (struct sockaddr_in *)(fwd_tag + 1);
 461                 ip->ip_dst = next_hop->sin_addr;
 462                 uh->uh_dport = ntohs(next_hop->sin_port);
 463
 464                 /*
 465                  * Remove the tag from the packet.  We don't need it anymore.
 466                  */
 467                 m_tag_delete(m, fwd_tag);
 468         }
 469 #endif
 470
 471         INP_INFO_RLOCK(&V_udbinfo);
 472         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
 473             in_broadcast(ip->ip_dst, ifp)) {
 474                 struct inpcb *last;
 475                 struct ip_moptions *imo;
 476
 477                 last = NULL;
 478                 LIST_FOREACH(inp, &V_udb, inp_list) {
 479                         if (inp->inp_lport != uh->uh_dport)
 480                                 continue;
 481 #ifdef INET6
 482                         if ((inp->inp_vflag & INP_IPV4) == 0)
 483                                 continue;
 484 #endif
 485                         if (inp->inp_laddr.s_addr != INADDR_ANY &&
 486                             inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
 487                                 continue;
 488                         if (inp->inp_faddr.s_addr != INADDR_ANY &&
 489                             inp->inp_faddr.s_addr != ip->ip_src.s_addr)
 490                                 continue;
 491                         if (inp->inp_fport != 0 &&
 492                             inp->inp_fport != uh->uh_sport)
 493                                 continue;
 494
 495                         INP_RLOCK(inp);
 496
 497                         /*
 498                          * Handle socket delivery policy for any-source
 499                          * and source-specific multicast. [RFC3678]
 500                          */
 501                         imo = inp->inp_moptions;
 502                         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) &&
 503                             imo != NULL) {
 504                                 struct sockaddr_in       group;
 505                                 int                      blocked;
 506
 507                                 bzero(&group, sizeof(struct sockaddr_in));
 508                                 group.sin_len = sizeof(struct sockaddr_in);
 509                                 group.sin_family = AF_INET;
 510                                 group.sin_addr = ip->ip_dst;
 511
 512                                 blocked = imo_multi_filter(imo, ifp,
 513                                         (struct sockaddr *)&group,
 514                                         (struct sockaddr *)&udp_in);
 515                                 if (blocked != MCAST_PASS) {
 516                                         if (blocked == MCAST_NOTGMEMBER)
 517                                                 IPSTAT_INC(ips_notmember);
 518                                         if (blocked == MCAST_NOTSMEMBER ||
 519                                             blocked == MCAST_MUTED)
 520                                                 UDPSTAT_INC(udps_filtermcast);
 521                                         INP_RUNLOCK(inp);
 522                                         continue;
 523                                 }
 524                         }
 525                         if (last != NULL) {
 526                                 struct mbuf *n;
 527
 528                                 n = m_copy(m, 0, M_COPYALL);
 529                                 up = intoudpcb(last);
 530                                 if (up->u_tun_func == NULL) {
 531                                         if (n != NULL)
 532                                                 udp_append(last,
 533                                                     ip, n,
 534                                                     iphlen +
 535                                                     sizeof(struct udphdr),
 536                                                     &udp_in);
 537                                 } else {
 538                                         /*
 539                                          * Engage the tunneling protocol we
 540                                          * will have to leave the info_lock
 541                                          * up, since we are hunting through
 542                                          * multiple UDP's.
 543                                          */
 544
 545                                         (*up->u_tun_func)(n, iphlen, last);
 546                                 }
 547                                 INP_RUNLOCK(last);
 548                         }
 549                         last = inp;
 550                         /*
 551                          * Don't look for additional matches if this one does
 552                          * not have either the SO_REUSEPORT or SO_REUSEADDR
 553                          * socket options set.  This heuristic avoids
 554                          * searching through all pcbs in the common case of a
 555                          * non-shared port.  It assumes that an application
 556                          * will never clear these options after setting them.
 557                          */
 558                         if ((last->inp_socket->so_options &
 559                             (SO_REUSEPORT|SO_REUSEADDR)) == 0)
 560                                 break;
 561                 }
 562
 563                 if (last == NULL) {
 564                         /*
 565                          * No matching pcb found; discard datagram.  (No need
 566                          * to send an ICMP Port Unreachable for a broadcast
 567                          * or multicast datgram.)
 568                          */
 569                         UDPSTAT_INC(udps_noportbcast);
 570                         goto badheadlocked;
 571                 }
 572                 up = intoudpcb(last);
 573                 if (up->u_tun_func == NULL) {
 574                         udp_append(last, ip, m, iphlen + sizeof(struct udphdr),
 575                             &udp_in);
 576                 } else {
 577                         /*
 578                          * Engage the tunneling protocol.
 579                          */
 580                         (*up->u_tun_func)(m, iphlen, last);
 581                 }
 582                 INP_RUNLOCK(last);
 583                 INP_INFO_RUNLOCK(&V_udbinfo);
 584                 return;
 585         }
 586
 587         /*
 588          * Locate pcb for datagram.
 589          */
 590         inp = in_pcblookup_hash(&V_udbinfo, ip->ip_src, uh->uh_sport,
 591             ip->ip_dst, uh->uh_dport, 1, ifp);
 592         if (inp == NULL) {
 593                 if (udp_log_in_vain) {
 594                         char buf[4*sizeof "123"];
 595
 596                         strcpy(buf, inet_ntoa(ip->ip_dst));
 597                         log(LOG_INFO,
 598                             "Connection attempt to UDP %s:%d from %s:%d\n",
 599                             buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src),
 600                             ntohs(uh->uh_sport));
 601                 }
 602                 UDPSTAT_INC(udps_noport);
 603                 if (m->m_flags & (M_BCAST | M_MCAST)) {
 604                         UDPSTAT_INC(udps_noportbcast);
 605                         goto badheadlocked;
 606                 }
 607                 if (V_udp_blackhole)
 608                         goto badheadlocked;
 609                 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
 610                         goto badheadlocked;
 611                 *ip = save_ip;
 612                 ip->ip_len += iphlen;
 613                 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
 614                 INP_INFO_RUNLOCK(&V_udbinfo);
 615                 return;
 616         }
 617
 618         /*
 619          * Check the minimum TTL for socket.
 620          */
 621         INP_RLOCK(inp);
 622         INP_INFO_RUNLOCK(&V_udbinfo);
 623         if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) {
 624                 INP_RUNLOCK(inp);
 625                 goto badunlocked;
 626         }
 627         up = intoudpcb(inp);
 628         if (up->u_tun_func == NULL) {
 629                 udp_append(inp, ip, m, iphlen + sizeof(struct udphdr), &udp_in);
 630         } else {
 631                 /*
 632                  * Engage the tunneling protocol.
 633                  */
 634
 635                 (*up->u_tun_func)(m, iphlen, inp);
 636         }
 637         INP_RUNLOCK(inp);
 638         return;
 639
 640 badheadlocked:
 641         if (inp)
 642                 INP_RUNLOCK(inp);
 643         INP_INFO_RUNLOCK(&V_udbinfo);
 644 badunlocked:
 645         m_freem(m);
 646 }
 647
 648 /*
 649  * Notify a udp user of an asynchronous error; just wake up so that they can
 650  * collect error status.
 651  */
 652 struct inpcb *
 653 udp_notify(struct inpcb *inp, int errno)
 654 {
 655
 656         /*
 657          * While udp_ctlinput() always calls udp_notify() with a read lock
 658          * when invoking it directly, in_pcbnotifyall() currently uses write
 659          * locks due to sharing code with TCP.  For now, accept either a read
 660          * or a write lock, but a read lock is sufficient.
 661          */
 662         INP_LOCK_ASSERT(inp);
 663
 664         inp->inp_socket->so_error = errno;
 665         sorwakeup(inp->inp_socket);
 666         sowwakeup(inp->inp_socket);
 667         return (inp);
 668 }
 669
 670 void
 671 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
 672 {
 673         struct ip *ip = vip;
 674         struct udphdr *uh;
 675         struct in_addr faddr;
 676         struct inpcb *inp;
 677
 678         faddr = ((struct sockaddr_in *)sa)->sin_addr;
 679         if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
 680                 return;
 681
 682         /*
 683          * Redirects don't need to be handled up here.
 684          */
 685         if (PRC_IS_REDIRECT(cmd))
 686                 return;
 687
 688         /*
 689          * Hostdead is ugly because it goes linearly through all PCBs.
 690          *
 691          * XXX: We never get this from ICMP, otherwise it makes an excellent
 692          * DoS attack on machines with many connections.
 693          */
 694         if (cmd == PRC_HOSTDEAD)
 695                 ip = NULL;
 696         else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
 697                 return;
 698         if (ip != NULL) {
 699                 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
 700                 INP_INFO_RLOCK(&V_udbinfo);
 701                 inp = in_pcblookup_hash(&V_udbinfo, faddr, uh->uh_dport,
 702                     ip->ip_src, uh->uh_sport, 0, NULL);
 703                 if (inp != NULL) {
 704                         INP_RLOCK(inp);
 705                         if (inp->inp_socket != NULL) {
 706                                 udp_notify(inp, inetctlerrmap[cmd]);
 707                         }
 708                         INP_RUNLOCK(inp);
 709                 }
 710                 INP_INFO_RUNLOCK(&V_udbinfo);
 711         } else
 712                 in_pcbnotifyall(&V_udbinfo, faddr, inetctlerrmap[cmd],
 713                     udp_notify);
 714 }
 715
 716 static int
 717 udp_pcblist(SYSCTL_HANDLER_ARGS)
 718 {
 719         int error, i, n;
 720         struct inpcb *inp, **inp_list;
 721         inp_gen_t gencnt;
 722         struct xinpgen xig;
 723
 724         /*
 725          * The process of preparing the PCB list is too time-consuming and
 726          * resource-intensive to repeat twice on every request.
 727          */
 728         if (req->oldptr == 0) {
 729                 n = V_udbinfo.ipi_count;
 730                 n += imax(n / 8, 10);
 731                 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
 732                 return (0);
 733         }
 734
 735         if (req->newptr != 0)
 736                 return (EPERM);
 737
 738         /*
 739          * OK, now we're committed to doing something.
 740          */
 741         INP_INFO_RLOCK(&V_udbinfo);
 742         gencnt = V_udbinfo.ipi_gencnt;
 743         n = V_udbinfo.ipi_count;
 744         INP_INFO_RUNLOCK(&V_udbinfo);
 745
 746         error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
 747                 + n * sizeof(struct xinpcb));
 748         if (error != 0)
 749                 return (error);
 750
 751         xig.xig_len = sizeof xig;
 752         xig.xig_count = n;
 753         xig.xig_gen = gencnt;
 754         xig.xig_sogen = so_gencnt;
 755         error = SYSCTL_OUT(req, &xig, sizeof xig);
 756         if (error)
 757                 return (error);
 758
 759         inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
 760         if (inp_list == 0)
 761                 return (ENOMEM);
 762
 763         INP_INFO_RLOCK(&V_udbinfo);
 764         for (inp = LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n;
 765              inp = LIST_NEXT(inp, inp_list)) {
 766                 INP_WLOCK(inp);
 767                 if (inp->inp_gencnt <= gencnt &&
 768                     cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
 769                         in_pcbref(inp);
 770                         inp_list[i++] = inp;
 771                 }
 772                 INP_WUNLOCK(inp);
 773         }
 774         INP_INFO_RUNLOCK(&V_udbinfo);
 775         n = i;
 776
 777         error = 0;
 778         for (i = 0; i < n; i++) {
 779                 inp = inp_list[i];
 780                 INP_RLOCK(inp);
 781                 if (inp->inp_gencnt <= gencnt) {
 782                         struct xinpcb xi;
 783
 784                         bzero(&xi, sizeof(xi));
 785                         xi.xi_len = sizeof xi;
 786                         /* XXX should avoid extra copy */
 787                         bcopy(inp, &xi.xi_inp, sizeof *inp);
 788                         if (inp->inp_socket)
 789                                 sotoxsocket(inp->inp_socket, &xi.xi_socket);
 790                         xi.xi_inp.inp_gencnt = inp->inp_gencnt;
 791                         INP_RUNLOCK(inp);
 792                         error = SYSCTL_OUT(req, &xi, sizeof xi);
 793                 } else
 794                         INP_RUNLOCK(inp);
 795         }
 796         INP_INFO_WLOCK(&V_udbinfo);
 797         for (i = 0; i < n; i++) {
 798                 inp = inp_list[i];
 799                 INP_WLOCK(inp);
 800                 if (!in_pcbrele(inp))
 801                         INP_WUNLOCK(inp);
 802         }
 803         INP_INFO_WUNLOCK(&V_udbinfo);
 804
 805         if (!error) {
 806                 /*
 807                  * Give the user an updated idea of our state.  If the
 808                  * generation differs from what we told her before, she knows
 809                  * that something happened while we were processing this
 810                  * request, and it might be necessary to retry.
 811                  */
 812                 INP_INFO_RLOCK(&V_udbinfo);
 813                 xig.xig_gen = V_udbinfo.ipi_gencnt;
 814                 xig.xig_sogen = so_gencnt;
 815                 xig.xig_count = V_udbinfo.ipi_count;
 816                 INP_INFO_RUNLOCK(&V_udbinfo);
 817                 error = SYSCTL_OUT(req, &xig, sizeof xig);
 818         }
 819         free(inp_list, M_TEMP);
 820         return (error);
 821 }
 822
 823 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist,
 824     CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
 825     udp_pcblist, "S,xinpcb", "List of active UDP sockets");
 826
 827 static int
 828 udp_getcred(SYSCTL_HANDLER_ARGS)
 829 {
 830         struct xucred xuc;
 831         struct sockaddr_in addrs[2];
 832         struct inpcb *inp;
 833         int error;
 834
 835         error = priv_check(req->td, PRIV_NETINET_GETCRED);
 836         if (error)
 837                 return (error);
 838         error = SYSCTL_IN(req, addrs, sizeof(addrs));
 839         if (error)
 840                 return (error);
 841         INP_INFO_RLOCK(&V_udbinfo);
 842         inp = in_pcblookup_hash(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
 843                                 addrs[0].sin_addr, addrs[0].sin_port, 1, NULL);
 844         if (inp != NULL) {
 845                 INP_RLOCK(inp);
 846                 INP_INFO_RUNLOCK(&V_udbinfo);
 847                 if (inp->inp_socket == NULL)
 848                         error = ENOENT;
 849                 if (error == 0)
 850                         error = cr_canseeinpcb(req->td->td_ucred, inp);
 851                 if (error == 0)
 852                         cru2x(inp->inp_cred, &xuc);
 853                 INP_RUNLOCK(inp);
 854         } else {
 855                 INP_INFO_RUNLOCK(&V_udbinfo);
 856                 error = ENOENT;
 857         }
 858         if (error == 0)
 859                 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
 860         return (error);
 861 }
 862
 863 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
 864     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
 865     udp_getcred, "S,xucred", "Get the xucred of a UDP connection");
 866
 867 int
 868 udp_ctloutput(struct socket *so, struct sockopt *sopt)
 869 {
 870         int error = 0, optval;
 871         struct inpcb *inp;
 872 #ifdef IPSEC_NAT_T
 873         struct udpcb *up;
 874 #endif
 875
 876         inp = sotoinpcb(so);
 877         KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
 878         INP_WLOCK(inp);
 879         if (sopt->sopt_level != IPPROTO_UDP) {
 880 #ifdef INET6
 881                 if (INP_CHECK_SOCKAF(so, AF_INET6)) {
 882                         INP_WUNLOCK(inp);
 883                         error = ip6_ctloutput(so, sopt);
 884                 } else {
 885 #endif
 886                         INP_WUNLOCK(inp);
 887                         error = ip_ctloutput(so, sopt);
 888 #ifdef INET6
 889                 }
 890 #endif
 891                 return (error);
 892         }
 893
 894         switch (sopt->sopt_dir) {
 895         case SOPT_SET:
 896                 switch (sopt->sopt_name) {
 897                 case UDP_ENCAP:
 898                         INP_WUNLOCK(inp);
 899                         error = sooptcopyin(sopt, &optval, sizeof optval,
 900                                             sizeof optval);
 901                         if (error)
 902                                 break;
 903                         inp = sotoinpcb(so);
 904                         KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
 905                         INP_WLOCK(inp);
 906 #ifdef IPSEC_NAT_T
 907                         up = intoudpcb(inp);
 908                         KASSERT(up != NULL, ("%s: up == NULL", __func__));
 909 #endif
 910                         switch (optval) {
 911                         case 0:
 912                                 /* Clear all UDP encap. */
 913 #ifdef IPSEC_NAT_T
 914                                 up->u_flags &= ~UF_ESPINUDP_ALL;
 915 #endif
 916                                 break;
 917 #ifdef IPSEC_NAT_T
 918                         case UDP_ENCAP_ESPINUDP:
 919                         case UDP_ENCAP_ESPINUDP_NON_IKE:
 920                                 up->u_flags &= ~UF_ESPINUDP_ALL;
 921                                 if (optval == UDP_ENCAP_ESPINUDP)
 922                                         up->u_flags |= UF_ESPINUDP;
 923                                 else if (optval == UDP_ENCAP_ESPINUDP_NON_IKE)
 924                                         up->u_flags |= UF_ESPINUDP_NON_IKE;
 925                                 break;
 926 #endif
 927                         default:
 928                                 error = EINVAL;
 929                                 break;
 930                         }
 931                         INP_WUNLOCK(inp);
 932                         break;
 933                 default:
 934                         INP_WUNLOCK(inp);
 935                         error = ENOPROTOOPT;
 936                         break;
 937                 }
 938                 break;
 939         case SOPT_GET:
 940                 switch (sopt->sopt_name) {
 941 #ifdef IPSEC_NAT_T
 942                 case UDP_ENCAP:
 943                         up = intoudpcb(inp);
 944                         KASSERT(up != NULL, ("%s: up == NULL", __func__));
 945                         optval = up->u_flags & UF_ESPINUDP_ALL;
 946                         INP_WUNLOCK(inp);
 947                         error = sooptcopyout(sopt, &optval, sizeof optval);
 948                         break;
 949 #endif
 950                 default:
 951                         INP_WUNLOCK(inp);
 952                         error = ENOPROTOOPT;
 953                         break;
 954                 }
 955                 break;
 956         }
 957         return (error);
 958 }
 959
 960 static int
 961 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
 962     struct mbuf *control, struct thread *td)
 963 {
 964         struct udpiphdr *ui;
 965         int len = m->m_pkthdr.len;
 966         struct in_addr faddr, laddr;
 967         struct cmsghdr *cm;
 968         struct sockaddr_in *sin, src;
 969         int error = 0;
 970         int ipflags;
 971         u_short fport, lport;
 972         int unlock_udbinfo;
 973
 974         /*
 975          * udp_output() may need to temporarily bind or connect the current
 976          * inpcb.  As such, we don't know up front whether we will need the
 977          * pcbinfo lock or not.  Do any work to decide what is needed up
 978          * front before acquiring any locks.
 979          */
 980         if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
 981                 if (control)
 982                         m_freem(control);
 983                 m_freem(m);
 984                 return (EMSGSIZE);
 985         }
 986
 987         src.sin_family = 0;
 988         if (control != NULL) {
 989                 /*
 990                  * XXX: Currently, we assume all the optional information is
 991                  * stored in a single mbuf.
 992                  */
 993                 if (control->m_next) {
 994                         m_freem(control);
 995                         m_freem(m);
 996                         return (EINVAL);
 997                 }
 998                 for (; control->m_len > 0;
 999                     control->m_data += CMSG_ALIGN(cm->cmsg_len),
1000                     control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
1001                         cm = mtod(control, struct cmsghdr *);
1002                         if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0
1003                             || cm->cmsg_len > control->m_len) {
1004                                 error = EINVAL;
1005                                 break;
1006                         }
1007                         if (cm->cmsg_level != IPPROTO_IP)
1008                                 continue;
1009
1010                         switch (cm->cmsg_type) {
1011                         case IP_SENDSRCADDR:
1012                                 if (cm->cmsg_len !=
1013                                     CMSG_LEN(sizeof(struct in_addr))) {
1014                                         error = EINVAL;
1015                                         break;
1016                                 }
1017                                 bzero(&src, sizeof(src));
1018                                 src.sin_family = AF_INET;
1019                                 src.sin_len = sizeof(src);
1020                                 src.sin_port = inp->inp_lport;
1021                                 src.sin_addr =
1022                                     *(struct in_addr *)CMSG_DATA(cm);
1023                                 break;
1024
1025                         default:
1026                                 error = ENOPROTOOPT;
1027                                 break;
1028                         }
1029                         if (error)
1030                                 break;
1031                 }
1032                 m_freem(control);
1033         }
1034         if (error) {
1035                 m_freem(m);
1036                 return (error);
1037         }
1038
1039         /*
1040          * Depending on whether or not the application has bound or connected
1041          * the socket, we may have to do varying levels of work.  The optimal
1042          * case is for a connected UDP socket, as a global lock isn't
1043          * required at all.
1044          *
1045          * In order to decide which we need, we require stability of the
1046          * inpcb binding, which we ensure by acquiring a read lock on the
1047          * inpcb.  This doesn't strictly follow the lock order, so we play
1048          * the trylock and retry game; note that we may end up with more
1049          * conservative locks than required the second time around, so later
1050          * assertions have to accept that.  Further analysis of the number of
1051          * misses under contention is required.
1052          */
1053         sin = (struct sockaddr_in *)addr;
1054         INP_RLOCK(inp);
1055         if (sin != NULL &&
1056             (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
1057                 INP_RUNLOCK(inp);
1058                 INP_INFO_WLOCK(&V_udbinfo);
1059                 INP_WLOCK(inp);
1060                 unlock_udbinfo = 2;
1061         } else if ((sin != NULL && (
1062             (sin->sin_addr.s_addr == INADDR_ANY) ||
1063             (sin->sin_addr.s_addr == INADDR_BROADCAST) ||
1064             (inp->inp_laddr.s_addr == INADDR_ANY) ||
1065             (inp->inp_lport == 0))) ||
1066             (src.sin_family == AF_INET)) {
1067                 if (!INP_INFO_TRY_RLOCK(&V_udbinfo)) {
1068                         INP_RUNLOCK(inp);
1069                         INP_INFO_RLOCK(&V_udbinfo);
1070                         INP_RLOCK(inp);
1071                 }
1072                 unlock_udbinfo = 1;
1073         } else
1074                 unlock_udbinfo = 0;
1075
1076         /*
1077          * If the IP_SENDSRCADDR control message was specified, override the
1078          * source address for this datagram.  Its use is invalidated if the
1079          * address thus specified is incomplete or clobbers other inpcbs.
1080          */
1081         laddr = inp->inp_laddr;
1082         lport = inp->inp_lport;
1083         if (src.sin_family == AF_INET) {
1084                 INP_INFO_LOCK_ASSERT(&V_udbinfo);
1085                 if ((lport == 0) ||
1086                     (laddr.s_addr == INADDR_ANY &&
1087                      src.sin_addr.s_addr == INADDR_ANY)) {
1088                         error = EINVAL;
1089                         goto release;
1090                 }
1091                 error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
1092                     &laddr.s_addr, &lport, td->td_ucred);
1093                 if (error)
1094                         goto release;
1095         }
1096
1097         /*
1098          * If a UDP socket has been connected, then a local address/port will
1099          * have been selected and bound.
1100          *
1101          * If a UDP socket has not been connected to, then an explicit
1102          * destination address must be used, in which case a local
1103          * address/port may not have been selected and bound.
1104          */
1105         if (sin != NULL) {
1106                 INP_LOCK_ASSERT(inp);
1107                 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1108                         error = EISCONN;
1109                         goto release;
1110                 }
1111
1112                 /*
1113                  * Jail may rewrite the destination address, so let it do
1114                  * that before we use it.
1115                  */
1116                 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1117                 if (error)
1118                         goto release;
1119
1120                 /*
1121                  * If a local address or port hasn't yet been selected, or if
1122                  * the destination address needs to be rewritten due to using
1123                  * a special INADDR_ constant, invoke in_pcbconnect_setup()
1124                  * to do the heavy lifting.  Once a port is selected, we
1125                  * commit the binding back to the socket; we also commit the
1126                  * binding of the address if in jail.
1127                  *
1128                  * If we already have a valid binding and we're not
1129                  * requesting a destination address rewrite, use a fast path.
1130                  */
1131                 if (inp->inp_laddr.s_addr == INADDR_ANY ||
1132                     inp->inp_lport == 0 ||
1133                     sin->sin_addr.s_addr == INADDR_ANY ||
1134                     sin->sin_addr.s_addr == INADDR_BROADCAST) {
1135                         INP_INFO_LOCK_ASSERT(&V_udbinfo);
1136                         error = in_pcbconnect_setup(inp, addr, &laddr.s_addr,
1137                             &lport, &faddr.s_addr, &fport, NULL,
1138                             td->td_ucred);
1139                         if (error)
1140                                 goto release;
1141
1142                         /*
1143                          * XXXRW: Why not commit the port if the address is
1144                          * !INADDR_ANY?
1145                          */
1146                         /* Commit the local port if newly assigned. */
1147                         if (inp->inp_laddr.s_addr == INADDR_ANY &&
1148                             inp->inp_lport == 0) {
1149                                 INP_INFO_WLOCK_ASSERT(&V_udbinfo);
1150                                 INP_WLOCK_ASSERT(inp);
1151                                 /*
1152                                  * Remember addr if jailed, to prevent
1153                                  * rebinding.
1154                                  */
1155                                 if (prison_flag(td->td_ucred, PR_IP4))
1156                                         inp->inp_laddr = laddr;
1157                                 inp->inp_lport = lport;
1158                                 if (in_pcbinshash(inp) != 0) {
1159                                         inp->inp_lport = 0;
1160                                         error = EAGAIN;
1161                                         goto release;
1162                                 }
1163                                 inp->inp_flags |= INP_ANONPORT;
1164                         }
1165                 } else {
1166                         faddr = sin->sin_addr;
1167                         fport = sin->sin_port;
1168                 }
1169         } else {
1170                 INP_LOCK_ASSERT(inp);
1171                 faddr = inp->inp_faddr;
1172                 fport = inp->inp_fport;
1173                 if (faddr.s_addr == INADDR_ANY) {
1174                         error = ENOTCONN;
1175                         goto release;
1176                 }
1177         }
1178
1179         /*
1180          * Calculate data length and get a mbuf for UDP, IP, and possible
1181          * link-layer headers.  Immediate slide the data pointer back forward
1182          * since we won't use that space at this layer.
1183          */
1184         M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_DONTWAIT);
1185         if (m == NULL) {
1186                 error = ENOBUFS;
1187                 goto release;
1188         }
1189         m->m_data += max_linkhdr;
1190         m->m_len -= max_linkhdr;
1191         m->m_pkthdr.len -= max_linkhdr;
1192
1193         /*
1194          * Fill in mbuf with extended UDP header and addresses and length put
1195          * into network format.
1196          */
1197         ui = mtod(m, struct udpiphdr *);
1198         bzero(ui->ui_x1, sizeof(ui->ui_x1));    /* XXX still needed? */
1199         ui->ui_pr = IPPROTO_UDP;
1200         ui->ui_src = laddr;
1201         ui->ui_dst = faddr;
1202         ui->ui_sport = lport;
1203         ui->ui_dport = fport;
1204         ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
1205
1206         /*
1207          * Set the Don't Fragment bit in the IP header.
1208          */
1209         if (inp->inp_flags & INP_DONTFRAG) {
1210                 struct ip *ip;
1211
1212                 ip = (struct ip *)&ui->ui_i;
1213                 ip->ip_off |= IP_DF;
1214         }
1215
1216         ipflags = 0;
1217         if (inp->inp_socket->so_options & SO_DONTROUTE)
1218                 ipflags |= IP_ROUTETOIF;
1219         if (inp->inp_socket->so_options & SO_BROADCAST)
1220                 ipflags |= IP_ALLOWBROADCAST;
1221         if (inp->inp_flags & INP_ONESBCAST)
1222                 ipflags |= IP_SENDONES;
1223
1224 #ifdef MAC
1225         mac_inpcb_create_mbuf(inp, m);
1226 #endif
1227
1228         /*
1229          * Set up checksum and output datagram.
1230          */
1231         if (V_udp_cksum) {
1232                 if (inp->inp_flags & INP_ONESBCAST)
1233                         faddr.s_addr = INADDR_BROADCAST;
1234                 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
1235                     htons((u_short)len + sizeof(struct udphdr) + IPPROTO_UDP));
1236                 m->m_pkthdr.csum_flags = CSUM_UDP;
1237                 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1238         } else
1239                 ui->ui_sum = 0;
1240         ((struct ip *)ui)->ip_len = sizeof (struct udpiphdr) + len;
1241         ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl;    /* XXX */
1242         ((struct ip *)ui)->ip_tos = inp->inp_ip_tos;    /* XXX */
1243         UDPSTAT_INC(udps_opackets);
1244
1245         if (unlock_udbinfo == 2)
1246                 INP_INFO_WUNLOCK(&V_udbinfo);
1247         else if (unlock_udbinfo == 1)
1248                 INP_INFO_RUNLOCK(&V_udbinfo);
1249         error = ip_output(m, inp->inp_options, NULL, ipflags,
1250             inp->inp_moptions, inp);
1251         if (unlock_udbinfo == 2)
1252                 INP_WUNLOCK(inp);
1253         else
1254                 INP_RUNLOCK(inp);
1255         return (error);
1256
1257 release:
1258         if (unlock_udbinfo == 2) {
1259                 INP_WUNLOCK(inp);
1260                 INP_INFO_WUNLOCK(&V_udbinfo);
1261         } else if (unlock_udbinfo == 1) {
1262                 INP_RUNLOCK(inp);
1263                 INP_INFO_RUNLOCK(&V_udbinfo);
1264         } else
1265                 INP_RUNLOCK(inp);
1266         m_freem(m);
1267         return (error);
1268 }
1269
1270
1271 #if defined(IPSEC) && defined(IPSEC_NAT_T)
1272 #ifdef INET
1273 /*
1274  * Potentially decap ESP in UDP frame.  Check for an ESP header
1275  * and optional marker; if present, strip the UDP header and
1276  * push the result through IPSec.
1277  *
1278  * Returns mbuf to be processed (potentially re-allocated) or
1279  * NULL if consumed and/or processed.
1280  */
1281 static struct mbuf *
1282 udp4_espdecap(struct inpcb *inp, struct mbuf *m, int off)
1283 {
1284         size_t minlen, payload, skip, iphlen;
1285         caddr_t data;
1286         struct udpcb *up;
1287         struct m_tag *tag;
1288         struct udphdr *udphdr;
1289         struct ip *ip;
1290
1291         INP_RLOCK_ASSERT(inp);
1292
1293         /*
1294          * Pull up data so the longest case is contiguous:
1295          *    IP/UDP hdr + non ESP marker + ESP hdr.
1296          */
1297         minlen = off + sizeof(uint64_t) + sizeof(struct esp);
1298         if (minlen > m->m_pkthdr.len)
1299                 minlen = m->m_pkthdr.len;
1300         if ((m = m_pullup(m, minlen)) == NULL) {
1301                 V_ipsec4stat.in_inval++;
1302                 return (NULL);          /* Bypass caller processing. */
1303         }
1304         data = mtod(m, caddr_t);        /* Points to ip header. */
1305         payload = m->m_len - off;       /* Size of payload. */
1306
1307         if (payload == 1 && data[off] == '\xff')
1308                 return (m);             /* NB: keepalive packet, no decap. */
1309
1310         up = intoudpcb(inp);
1311         KASSERT(up != NULL, ("%s: udpcb NULL", __func__));
1312         KASSERT((up->u_flags & UF_ESPINUDP_ALL) != 0,
1313             ("u_flags 0x%x", up->u_flags));
1314
1315         /*
1316          * Check that the payload is large enough to hold an
1317          * ESP header and compute the amount of data to remove.
1318          *
1319          * NB: the caller has already done a pullup for us.
1320          * XXX can we assume alignment and eliminate bcopys?
1321          */
1322         if (up->u_flags & UF_ESPINUDP_NON_IKE) {
1323                 /*
1324                  * draft-ietf-ipsec-nat-t-ike-0[01].txt and
1325                  * draft-ietf-ipsec-udp-encaps-(00/)01.txt, ignoring
1326                  * possible AH mode non-IKE marker+non-ESP marker
1327                  * from draft-ietf-ipsec-udp-encaps-00.txt.
1328                  */
1329                 uint64_t marker;
1330
1331                 if (payload <= sizeof(uint64_t) + sizeof(struct esp))
1332                         return (m);     /* NB: no decap. */
1333                 bcopy(data + off, &marker, sizeof(uint64_t));
1334                 if (marker != 0)        /* Non-IKE marker. */
1335                         return (m);     /* NB: no decap. */
1336                 skip = sizeof(uint64_t) + sizeof(struct udphdr);
1337         } else {
1338                 uint32_t spi;
1339
1340                 if (payload <= sizeof(struct esp)) {
1341                         V_ipsec4stat.in_inval++;
1342                         m_freem(m);
1343                         return (NULL);  /* Discard. */
1344                 }
1345                 bcopy(data + off, &spi, sizeof(uint32_t));
1346                 if (spi == 0)           /* Non-ESP marker. */
1347                         return (m);     /* NB: no decap. */
1348                 skip = sizeof(struct udphdr);
1349         }
1350
1351         /*
1352          * Setup a PACKET_TAG_IPSEC_NAT_T_PORT tag to remember
1353          * the UDP ports. This is required if we want to select
1354          * the right SPD for multiple hosts behind same NAT.
1355          *
1356          * NB: ports are maintained in network byte order everywhere
1357          *     in the NAT-T code.
1358          */
1359         tag = m_tag_get(PACKET_TAG_IPSEC_NAT_T_PORTS,
1360                 2 * sizeof(uint16_t), M_NOWAIT);
1361         if (tag == NULL) {
1362                 V_ipsec4stat.in_nomem++;
1363                 m_freem(m);
1364                 return (NULL);          /* Discard. */
1365         }
1366         iphlen = off - sizeof(struct udphdr);
1367         udphdr = (struct udphdr *)(data + iphlen);
1368         ((uint16_t *)(tag + 1))[0] = udphdr->uh_sport;
1369         ((uint16_t *)(tag + 1))[1] = udphdr->uh_dport;
1370         m_tag_prepend(m, tag);
1371
1372         /*
1373          * Remove the UDP header (and possibly the non ESP marker)
1374          * IP header length is iphlen
1375          * Before:
1376          *   <--- off --->
1377          *   +----+------+-----+
1378          *   | IP |  UDP | ESP |
1379          *   +----+------+-----+
1380          *        <-skip->
1381          * After:
1382          *          +----+-----+
1383          *          | IP | ESP |
1384          *          +----+-----+
1385          *   <-skip->
1386          */
1387         ovbcopy(data, data + skip, iphlen);
1388         m_adj(m, skip);
1389
1390         ip = mtod(m, struct ip *);
1391         ip->ip_len -= skip;
1392         ip->ip_p = IPPROTO_ESP;
1393
1394         /*
1395          * We cannot yet update the cksums so clear any
1396          * h/w cksum flags as they are no longer valid.
1397          */
1398         if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID)
1399                 m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID|CSUM_PSEUDO_HDR);
1400
1401         (void) ipsec4_common_input(m, iphlen, ip->ip_p);
1402         return (NULL);                  /* NB: consumed, bypass processing. */
1403 }
1404 #endif /* INET */
1405 #endif /* defined(IPSEC) && defined(IPSEC_NAT_T) */
1406
1407 static void
1408 udp_abort(struct socket *so)
1409 {
1410         struct inpcb *inp;
1411
1412         inp = sotoinpcb(so);
1413         KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
1414         INP_INFO_WLOCK(&V_udbinfo);
1415         INP_WLOCK(inp);
1416         if (inp->inp_faddr.s_addr != INADDR_ANY) {
1417                 in_pcbdisconnect(inp);
1418                 inp->inp_laddr.s_addr = INADDR_ANY;
1419                 soisdisconnected(so);
1420         }
1421         INP_WUNLOCK(inp);
1422         INP_INFO_WUNLOCK(&V_udbinfo);
1423 }
1424
1425 static int
1426 udp_attach(struct socket *so, int proto, struct thread *td)
1427 {
1428         struct inpcb *inp;
1429         int error;
1430
1431         inp = sotoinpcb(so);
1432         KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
1433         error = soreserve(so, udp_sendspace, udp_recvspace);
1434         if (error)
1435                 return (error);
1436         INP_INFO_WLOCK(&V_udbinfo);
1437         error = in_pcballoc(so, &V_udbinfo);
1438         if (error) {
1439                 INP_INFO_WUNLOCK(&V_udbinfo);
1440                 return (error);
1441         }
1442
1443         inp = sotoinpcb(so);
1444         inp->inp_vflag |= INP_IPV4;
1445         inp->inp_ip_ttl = V_ip_defttl;
1446
1447         error = udp_newudpcb(inp);
1448         if (error) {
1449                 in_pcbdetach(inp);
1450                 in_pcbfree(inp);
1451                 INP_INFO_WUNLOCK(&V_udbinfo);
1452                 return (error);
1453         }
1454
1455         INP_WUNLOCK(inp);
1456         INP_INFO_WUNLOCK(&V_udbinfo);
1457         return (0);
1458 }
1459
1460 int
1461 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f)
1462 {
1463         struct inpcb *inp;
1464         struct udpcb *up;
1465
1466         KASSERT(so->so_type == SOCK_DGRAM,
1467             ("udp_set_kernel_tunneling: !dgram"));
1468         inp = sotoinpcb(so);
1469         KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL"));
1470         INP_WLOCK(inp);
1471         up = intoudpcb(inp);
1472         if (up->u_tun_func != NULL) {
1473                 INP_WUNLOCK(inp);
1474                 return (EBUSY);
1475         }
1476         up->u_tun_func = f;
1477         INP_WUNLOCK(inp);
1478         return (0);
1479 }
1480
1481 static int
1482 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1483 {
1484         struct inpcb *inp;
1485         int error;
1486
1487         inp = sotoinpcb(so);
1488         KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1489         INP_INFO_WLOCK(&V_udbinfo);
1490         INP_WLOCK(inp);
1491         error = in_pcbbind(inp, nam, td->td_ucred);
1492         INP_WUNLOCK(inp);
1493         INP_INFO_WUNLOCK(&V_udbinfo);
1494         return (error);
1495 }
1496
1497 static void
1498 udp_close(struct socket *so)
1499 {
1500         struct inpcb *inp;
1501
1502         inp = sotoinpcb(so);
1503         KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1504         INP_INFO_WLOCK(&V_udbinfo);
1505         INP_WLOCK(inp);
1506         if (inp->inp_faddr.s_addr != INADDR_ANY) {
1507                 in_pcbdisconnect(inp);
1508                 inp->inp_laddr.s_addr = INADDR_ANY;
1509                 soisdisconnected(so);
1510         }
1511         INP_WUNLOCK(inp);
1512         INP_INFO_WUNLOCK(&V_udbinfo);
1513 }
1514
1515 static int
1516 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1517 {
1518         struct inpcb *inp;
1519         int error;
1520         struct sockaddr_in *sin;
1521
1522         inp = sotoinpcb(so);
1523         KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1524         INP_INFO_WLOCK(&V_udbinfo);
1525         INP_WLOCK(inp);
1526         if (inp->inp_faddr.s_addr != INADDR_ANY) {
1527                 INP_WUNLOCK(inp);
1528                 INP_INFO_WUNLOCK(&V_udbinfo);
1529                 return (EISCONN);
1530         }
1531         sin = (struct sockaddr_in *)nam;
1532         error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1533         if (error != 0) {
1534                 INP_WUNLOCK(inp);
1535                 INP_INFO_WUNLOCK(&V_udbinfo);
1536                 return (error);
1537         }
1538         error = in_pcbconnect(inp, nam, td->td_ucred);
1539         if (error == 0)
1540                 soisconnected(so);
1541         INP_WUNLOCK(inp);
1542         INP_INFO_WUNLOCK(&V_udbinfo);
1543         return (error);
1544 }
1545
1546 static void
1547 udp_detach(struct socket *so)
1548 {
1549         struct inpcb *inp;
1550         struct udpcb *up;
1551
1552         inp = sotoinpcb(so);
1553         KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1554         KASSERT(inp->inp_faddr.s_addr == INADDR_ANY,
1555             ("udp_detach: not disconnected"));
1556         INP_INFO_WLOCK(&V_udbinfo);
1557         INP_WLOCK(inp);
1558         up = intoudpcb(inp);
1559         KASSERT(up != NULL, ("%s: up == NULL", __func__));
1560         inp->inp_ppcb = NULL;
1561         in_pcbdetach(inp);
1562         in_pcbfree(inp);
1563         INP_INFO_WUNLOCK(&V_udbinfo);
1564         udp_discardcb(up);
1565 }
1566
1567 static int
1568 udp_disconnect(struct socket *so)
1569 {
1570         struct inpcb *inp;
1571
1572         inp = sotoinpcb(so);
1573         KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1574         INP_INFO_WLOCK(&V_udbinfo);
1575         INP_WLOCK(inp);
1576         if (inp->inp_faddr.s_addr == INADDR_ANY) {
1577                 INP_WUNLOCK(inp);
1578                 INP_INFO_WUNLOCK(&V_udbinfo);
1579                 return (ENOTCONN);
1580         }
1581
1582         in_pcbdisconnect(inp);
1583         inp->inp_laddr.s_addr = INADDR_ANY;
1584         SOCK_LOCK(so);
1585         so->so_state &= ~SS_ISCONNECTED;                /* XXX */
1586         SOCK_UNLOCK(so);
1587         INP_WUNLOCK(inp);
1588         INP_INFO_WUNLOCK(&V_udbinfo);
1589         return (0);
1590 }
1591
1592 static int
1593 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1594     struct mbuf *control, struct thread *td)
1595 {
1596         struct inpcb *inp;
1597
1598         inp = sotoinpcb(so);
1599         KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1600         return (udp_output(inp, m, addr, control, td));
1601 }
1602
1603 int
1604 udp_shutdown(struct socket *so)
1605 {
1606         struct inpcb *inp;
1607
1608         inp = sotoinpcb(so);
1609         KASSERT(inp != NULL, ("udp_shutdown: inp == NULL"));
1610         INP_WLOCK(inp);
1611         socantsendmore(so);
1612         INP_WUNLOCK(inp);
1613         return (0);
1614 }
1615
1616 struct pr_usrreqs udp_usrreqs = {
1617         .pru_abort =            udp_abort,
1618         .pru_attach =           udp_attach,
1619         .pru_bind =             udp_bind,
1620         .pru_connect =          udp_connect,
1621         .pru_control =          in_control,
1622         .pru_detach =           udp_detach,
1623         .pru_disconnect =       udp_disconnect,
1624         .pru_peeraddr =         in_getpeeraddr,
1625         .pru_send =             udp_send,
1626         .pru_soreceive =        soreceive_dgram,
1627         .pru_sosend =           sosend_dgram,
1628         .pru_shutdown =         udp_shutdown,
1629         .pru_sockaddr =         in_getsockaddr,
1630         .pru_sosetlabel =       in_pcbsosetlabel,
1631         .pru_close =            udp_close,
1632 };