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                          * Detached PCBs can linger in the list if someone
 499                          * holds a reference. (e.g. udp_pcblist)
 500                          */
 501                         if (inp->inp_socket == NULL) {
 502                                 INP_RUNLOCK(inp);
 503                                 continue;
 504                         }
 505
 506                         /*
 507                          * Handle socket delivery policy for any-source
 508                          * and source-specific multicast. [RFC3678]
 509                          */
 510                         imo = inp->inp_moptions;
 511                         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) &&
 512                             imo != NULL) {
 513                                 struct sockaddr_in       group;
 514                                 int                      blocked;
 515
 516                                 bzero(&group, sizeof(struct sockaddr_in));
 517                                 group.sin_len = sizeof(struct sockaddr_in);
 518                                 group.sin_family = AF_INET;
 519                                 group.sin_addr = ip->ip_dst;
 520
 521                                 blocked = imo_multi_filter(imo, ifp,
 522                                         (struct sockaddr *)&group,
 523                                         (struct sockaddr *)&udp_in);
 524                                 if (blocked != MCAST_PASS) {
 525                                         if (blocked == MCAST_NOTGMEMBER)
 526                                                 IPSTAT_INC(ips_notmember);
 527                                         if (blocked == MCAST_NOTSMEMBER ||
 528                                             blocked == MCAST_MUTED)
 529                                                 UDPSTAT_INC(udps_filtermcast);
 530                                         INP_RUNLOCK(inp);
 531                                         continue;
 532                                 }
 533                         }
 534                         if (last != NULL) {
 535                                 struct mbuf *n;
 536
 537                                 n = m_copy(m, 0, M_COPYALL);
 538                                 up = intoudpcb(last);
 539                                 if (up->u_tun_func == NULL) {
 540                                         if (n != NULL)
 541                                                 udp_append(last,
 542                                                     ip, n,
 543                                                     iphlen +
 544                                                     sizeof(struct udphdr),
 545                                                     &udp_in);
 546                                 } else {
 547                                         /*
 548                                          * Engage the tunneling protocol we
 549                                          * will have to leave the info_lock
 550                                          * up, since we are hunting through
 551                                          * multiple UDP's.
 552                                          */
 553
 554                                         (*up->u_tun_func)(n, iphlen, last);
 555                                 }
 556                                 INP_RUNLOCK(last);
 557                         }
 558                         last = inp;
 559                         /*
 560                          * Don't look for additional matches if this one does
 561                          * not have either the SO_REUSEPORT or SO_REUSEADDR
 562                          * socket options set.  This heuristic avoids
 563                          * searching through all pcbs in the common case of a
 564                          * non-shared port.  It assumes that an application
 565                          * will never clear these options after setting them.
 566                          */
 567                         if ((last->inp_socket->so_options &
 568                             (SO_REUSEPORT|SO_REUSEADDR)) == 0)
 569                                 break;
 570                 }
 571
 572                 if (last == NULL) {
 573                         /*
 574                          * No matching pcb found; discard datagram.  (No need
 575                          * to send an ICMP Port Unreachable for a broadcast
 576                          * or multicast datgram.)
 577                          */
 578                         UDPSTAT_INC(udps_noportbcast);
 579                         goto badheadlocked;
 580                 }
 581                 up = intoudpcb(last);
 582                 if (up->u_tun_func == NULL) {
 583                         udp_append(last, ip, m, iphlen + sizeof(struct udphdr),
 584                             &udp_in);
 585                 } else {
 586                         /*
 587                          * Engage the tunneling protocol.
 588                          */
 589                         (*up->u_tun_func)(m, iphlen, last);
 590                 }
 591                 INP_RUNLOCK(last);
 592                 INP_INFO_RUNLOCK(&V_udbinfo);
 593                 return;
 594         }
 595
 596         /*
 597          * Locate pcb for datagram.
 598          */
 599         inp = in_pcblookup_hash(&V_udbinfo, ip->ip_src, uh->uh_sport,
 600             ip->ip_dst, uh->uh_dport, 1, ifp);
 601         if (inp == NULL) {
 602                 if (udp_log_in_vain) {
 603                         char buf[4*sizeof "123"];
 604
 605                         strcpy(buf, inet_ntoa(ip->ip_dst));
 606                         log(LOG_INFO,
 607                             "Connection attempt to UDP %s:%d from %s:%d\n",
 608                             buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src),
 609                             ntohs(uh->uh_sport));
 610                 }
 611                 UDPSTAT_INC(udps_noport);
 612                 if (m->m_flags & (M_BCAST | M_MCAST)) {
 613                         UDPSTAT_INC(udps_noportbcast);
 614                         goto badheadlocked;
 615                 }
 616                 if (V_udp_blackhole)
 617                         goto badheadlocked;
 618                 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
 619                         goto badheadlocked;
 620                 *ip = save_ip;
 621                 ip->ip_len += iphlen;
 622                 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
 623                 INP_INFO_RUNLOCK(&V_udbinfo);
 624                 return;
 625         }
 626
 627         /*
 628          * Check the minimum TTL for socket.
 629          */
 630         INP_RLOCK(inp);
 631         INP_INFO_RUNLOCK(&V_udbinfo);
 632
 633         /*
 634          * Detached PCBs can linger in the hash table if someone holds a
 635          * reference. (e.g. udp_pcblist)
 636          */
 637         if (inp->inp_socket == NULL) {
 638                 INP_RUNLOCK(inp);
 639                 goto badunlocked;
 640         }
 641         if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) {
 642                 INP_RUNLOCK(inp);
 643                 goto badunlocked;
 644         }
 645         up = intoudpcb(inp);
 646         if (up->u_tun_func == NULL) {
 647                 udp_append(inp, ip, m, iphlen + sizeof(struct udphdr), &udp_in);
 648         } else {
 649                 /*
 650                  * Engage the tunneling protocol.
 651                  */
 652
 653                 (*up->u_tun_func)(m, iphlen, inp);
 654         }
 655         INP_RUNLOCK(inp);
 656         return;
 657
 658 badheadlocked:
 659         if (inp)
 660                 INP_RUNLOCK(inp);
 661         INP_INFO_RUNLOCK(&V_udbinfo);
 662 badunlocked:
 663         m_freem(m);
 664 }
 665
 666 /*
 667  * Notify a udp user of an asynchronous error; just wake up so that they can
 668  * collect error status.
 669  */
 670 struct inpcb *
 671 udp_notify(struct inpcb *inp, int errno)
 672 {
 673
 674         /*
 675          * While udp_ctlinput() always calls udp_notify() with a read lock
 676          * when invoking it directly, in_pcbnotifyall() currently uses write
 677          * locks due to sharing code with TCP.  For now, accept either a read
 678          * or a write lock, but a read lock is sufficient.
 679          */
 680         INP_LOCK_ASSERT(inp);
 681
 682         inp->inp_socket->so_error = errno;
 683         sorwakeup(inp->inp_socket);
 684         sowwakeup(inp->inp_socket);
 685         return (inp);
 686 }
 687
 688 void
 689 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
 690 {
 691         struct ip *ip = vip;
 692         struct udphdr *uh;
 693         struct in_addr faddr;
 694         struct inpcb *inp;
 695
 696         faddr = ((struct sockaddr_in *)sa)->sin_addr;
 697         if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
 698                 return;
 699
 700         /*
 701          * Redirects don't need to be handled up here.
 702          */
 703         if (PRC_IS_REDIRECT(cmd))
 704                 return;
 705
 706         /*
 707          * Hostdead is ugly because it goes linearly through all PCBs.
 708          *
 709          * XXX: We never get this from ICMP, otherwise it makes an excellent
 710          * DoS attack on machines with many connections.
 711          */
 712         if (cmd == PRC_HOSTDEAD)
 713                 ip = NULL;
 714         else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
 715                 return;
 716         if (ip != NULL) {
 717                 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
 718                 INP_INFO_RLOCK(&V_udbinfo);
 719                 inp = in_pcblookup_hash(&V_udbinfo, faddr, uh->uh_dport,
 720                     ip->ip_src, uh->uh_sport, 0, NULL);
 721                 if (inp != NULL) {
 722                         INP_RLOCK(inp);
 723                         if (inp->inp_socket != NULL) {
 724                                 udp_notify(inp, inetctlerrmap[cmd]);
 725                         }
 726                         INP_RUNLOCK(inp);
 727                 }
 728                 INP_INFO_RUNLOCK(&V_udbinfo);
 729         } else
 730                 in_pcbnotifyall(&V_udbinfo, faddr, inetctlerrmap[cmd],
 731                     udp_notify);
 732 }
 733
 734 static int
 735 udp_pcblist(SYSCTL_HANDLER_ARGS)
 736 {
 737         int error, i, n;
 738         struct inpcb *inp, **inp_list;
 739         inp_gen_t gencnt;
 740         struct xinpgen xig;
 741
 742         /*
 743          * The process of preparing the PCB list is too time-consuming and
 744          * resource-intensive to repeat twice on every request.
 745          */
 746         if (req->oldptr == 0) {
 747                 n = V_udbinfo.ipi_count;
 748                 n += imax(n / 8, 10);
 749                 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
 750                 return (0);
 751         }
 752
 753         if (req->newptr != 0)
 754                 return (EPERM);
 755
 756         /*
 757          * OK, now we're committed to doing something.
 758          */
 759         INP_INFO_RLOCK(&V_udbinfo);
 760         gencnt = V_udbinfo.ipi_gencnt;
 761         n = V_udbinfo.ipi_count;
 762         INP_INFO_RUNLOCK(&V_udbinfo);
 763
 764         error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
 765                 + n * sizeof(struct xinpcb));
 766         if (error != 0)
 767                 return (error);
 768
 769         xig.xig_len = sizeof xig;
 770         xig.xig_count = n;
 771         xig.xig_gen = gencnt;
 772         xig.xig_sogen = so_gencnt;
 773         error = SYSCTL_OUT(req, &xig, sizeof xig);
 774         if (error)
 775                 return (error);
 776
 777         inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
 778         if (inp_list == 0)
 779                 return (ENOMEM);
 780
 781         INP_INFO_RLOCK(&V_udbinfo);
 782         for (inp = LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n;
 783              inp = LIST_NEXT(inp, inp_list)) {
 784                 INP_WLOCK(inp);
 785                 if (inp->inp_gencnt <= gencnt &&
 786                     cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
 787                         in_pcbref(inp);
 788                         inp_list[i++] = inp;
 789                 }
 790                 INP_WUNLOCK(inp);
 791         }
 792         INP_INFO_RUNLOCK(&V_udbinfo);
 793         n = i;
 794
 795         error = 0;
 796         for (i = 0; i < n; i++) {
 797                 inp = inp_list[i];
 798                 INP_RLOCK(inp);
 799                 if (inp->inp_gencnt <= gencnt) {
 800                         struct xinpcb xi;
 801
 802                         bzero(&xi, sizeof(xi));
 803                         xi.xi_len = sizeof xi;
 804                         /* XXX should avoid extra copy */
 805                         bcopy(inp, &xi.xi_inp, sizeof *inp);
 806                         if (inp->inp_socket)
 807                                 sotoxsocket(inp->inp_socket, &xi.xi_socket);
 808                         xi.xi_inp.inp_gencnt = inp->inp_gencnt;
 809                         INP_RUNLOCK(inp);
 810                         error = SYSCTL_OUT(req, &xi, sizeof xi);
 811                 } else
 812                         INP_RUNLOCK(inp);
 813         }
 814         INP_INFO_WLOCK(&V_udbinfo);
 815         for (i = 0; i < n; i++) {
 816                 inp = inp_list[i];
 817                 INP_WLOCK(inp);
 818                 if (!in_pcbrele(inp))
 819                         INP_WUNLOCK(inp);
 820         }
 821         INP_INFO_WUNLOCK(&V_udbinfo);
 822
 823         if (!error) {
 824                 /*
 825                  * Give the user an updated idea of our state.  If the
 826                  * generation differs from what we told her before, she knows
 827                  * that something happened while we were processing this
 828                  * request, and it might be necessary to retry.
 829                  */
 830                 INP_INFO_RLOCK(&V_udbinfo);
 831                 xig.xig_gen = V_udbinfo.ipi_gencnt;
 832                 xig.xig_sogen = so_gencnt;
 833                 xig.xig_count = V_udbinfo.ipi_count;
 834                 INP_INFO_RUNLOCK(&V_udbinfo);
 835                 error = SYSCTL_OUT(req, &xig, sizeof xig);
 836         }
 837         free(inp_list, M_TEMP);
 838         return (error);
 839 }
 840
 841 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist,
 842     CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
 843     udp_pcblist, "S,xinpcb", "List of active UDP sockets");
 844
 845 static int
 846 udp_getcred(SYSCTL_HANDLER_ARGS)
 847 {
 848         struct xucred xuc;
 849         struct sockaddr_in addrs[2];
 850         struct inpcb *inp;
 851         int error;
 852
 853         error = priv_check(req->td, PRIV_NETINET_GETCRED);
 854         if (error)
 855                 return (error);
 856         error = SYSCTL_IN(req, addrs, sizeof(addrs));
 857         if (error)
 858                 return (error);
 859         INP_INFO_RLOCK(&V_udbinfo);
 860         inp = in_pcblookup_hash(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
 861                                 addrs[0].sin_addr, addrs[0].sin_port, 1, NULL);
 862         if (inp != NULL) {
 863                 INP_RLOCK(inp);
 864                 INP_INFO_RUNLOCK(&V_udbinfo);
 865                 if (inp->inp_socket == NULL)
 866                         error = ENOENT;
 867                 if (error == 0)
 868                         error = cr_canseeinpcb(req->td->td_ucred, inp);
 869                 if (error == 0)
 870                         cru2x(inp->inp_cred, &xuc);
 871                 INP_RUNLOCK(inp);
 872         } else {
 873                 INP_INFO_RUNLOCK(&V_udbinfo);
 874                 error = ENOENT;
 875         }
 876         if (error == 0)
 877                 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
 878         return (error);
 879 }
 880
 881 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
 882     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
 883     udp_getcred, "S,xucred", "Get the xucred of a UDP connection");
 884
 885 int
 886 udp_ctloutput(struct socket *so, struct sockopt *sopt)
 887 {
 888         int error = 0, optval;
 889         struct inpcb *inp;
 890 #ifdef IPSEC_NAT_T
 891         struct udpcb *up;
 892 #endif
 893
 894         inp = sotoinpcb(so);
 895         KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
 896         INP_WLOCK(inp);
 897         if (sopt->sopt_level != IPPROTO_UDP) {
 898 #ifdef INET6
 899                 if (INP_CHECK_SOCKAF(so, AF_INET6)) {
 900                         INP_WUNLOCK(inp);
 901                         error = ip6_ctloutput(so, sopt);
 902                 } else {
 903 #endif
 904                         INP_WUNLOCK(inp);
 905                         error = ip_ctloutput(so, sopt);
 906 #ifdef INET6
 907                 }
 908 #endif
 909                 return (error);
 910         }
 911
 912         switch (sopt->sopt_dir) {
 913         case SOPT_SET:
 914                 switch (sopt->sopt_name) {
 915                 case UDP_ENCAP:
 916                         INP_WUNLOCK(inp);
 917                         error = sooptcopyin(sopt, &optval, sizeof optval,
 918                                             sizeof optval);
 919                         if (error)
 920                                 break;
 921                         inp = sotoinpcb(so);
 922                         KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
 923                         INP_WLOCK(inp);
 924 #ifdef IPSEC_NAT_T
 925                         up = intoudpcb(inp);
 926                         KASSERT(up != NULL, ("%s: up == NULL", __func__));
 927 #endif
 928                         switch (optval) {
 929                         case 0:
 930                                 /* Clear all UDP encap. */
 931 #ifdef IPSEC_NAT_T
 932                                 up->u_flags &= ~UF_ESPINUDP_ALL;
 933 #endif
 934                                 break;
 935 #ifdef IPSEC_NAT_T
 936                         case UDP_ENCAP_ESPINUDP:
 937                         case UDP_ENCAP_ESPINUDP_NON_IKE:
 938                                 up->u_flags &= ~UF_ESPINUDP_ALL;
 939                                 if (optval == UDP_ENCAP_ESPINUDP)
 940                                         up->u_flags |= UF_ESPINUDP;
 941                                 else if (optval == UDP_ENCAP_ESPINUDP_NON_IKE)
 942                                         up->u_flags |= UF_ESPINUDP_NON_IKE;
 943                                 break;
 944 #endif
 945                         default:
 946                                 error = EINVAL;
 947                                 break;
 948                         }
 949                         INP_WUNLOCK(inp);
 950                         break;
 951                 default:
 952                         INP_WUNLOCK(inp);
 953                         error = ENOPROTOOPT;
 954                         break;
 955                 }
 956                 break;
 957         case SOPT_GET:
 958                 switch (sopt->sopt_name) {
 959 #ifdef IPSEC_NAT_T
 960                 case UDP_ENCAP:
 961                         up = intoudpcb(inp);
 962                         KASSERT(up != NULL, ("%s: up == NULL", __func__));
 963                         optval = up->u_flags & UF_ESPINUDP_ALL;
 964                         INP_WUNLOCK(inp);
 965                         error = sooptcopyout(sopt, &optval, sizeof optval);
 966                         break;
 967 #endif
 968                 default:
 969                         INP_WUNLOCK(inp);
 970                         error = ENOPROTOOPT;
 971                         break;
 972                 }
 973                 break;
 974         }
 975         return (error);
 976 }
 977
 978 static int
 979 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
 980     struct mbuf *control, struct thread *td)
 981 {
 982         struct udpiphdr *ui;
 983         int len = m->m_pkthdr.len;
 984         struct in_addr faddr, laddr;
 985         struct cmsghdr *cm;
 986         struct sockaddr_in *sin, src;
 987         int error = 0;
 988         int ipflags;
 989         u_short fport, lport;
 990         int unlock_udbinfo;
 991         u_char tos;
 992
 993         /*
 994          * udp_output() may need to temporarily bind or connect the current
 995          * inpcb.  As such, we don't know up front whether we will need the
 996          * pcbinfo lock or not.  Do any work to decide what is needed up
 997          * front before acquiring any locks.
 998          */
 999         if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
1000                 if (control)
1001                         m_freem(control);
1002                 m_freem(m);
1003                 return (EMSGSIZE);
1004         }
1005
1006         src.sin_family = 0;
1007         tos = inp->inp_ip_tos;
1008         if (control != NULL) {
1009                 /*
1010                  * XXX: Currently, we assume all the optional information is
1011                  * stored in a single mbuf.
1012                  */
1013                 if (control->m_next) {
1014                         m_freem(control);
1015                         m_freem(m);
1016                         return (EINVAL);
1017                 }
1018                 for (; control->m_len > 0;
1019                     control->m_data += CMSG_ALIGN(cm->cmsg_len),
1020                     control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
1021                         cm = mtod(control, struct cmsghdr *);
1022                         if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0
1023                             || cm->cmsg_len > control->m_len) {
1024                                 error = EINVAL;
1025                                 break;
1026                         }
1027                         if (cm->cmsg_level != IPPROTO_IP)
1028                                 continue;
1029
1030                         switch (cm->cmsg_type) {
1031                         case IP_SENDSRCADDR:
1032                                 if (cm->cmsg_len !=
1033                                     CMSG_LEN(sizeof(struct in_addr))) {
1034                                         error = EINVAL;
1035                                         break;
1036                                 }
1037                                 bzero(&src, sizeof(src));
1038                                 src.sin_family = AF_INET;
1039                                 src.sin_len = sizeof(src);
1040                                 src.sin_port = inp->inp_lport;
1041                                 src.sin_addr =
1042                                     *(struct in_addr *)CMSG_DATA(cm);
1043                                 break;
1044
1045                         case IP_TOS:
1046                                 if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) {
1047                                         error = EINVAL;
1048                                         break;
1049                                 }
1050                                 tos = *(u_char *)CMSG_DATA(cm);
1051                                 break;
1052
1053                         default:
1054                                 error = ENOPROTOOPT;
1055                                 break;
1056                         }
1057                         if (error)
1058                                 break;
1059                 }
1060                 m_freem(control);
1061         }
1062         if (error) {
1063                 m_freem(m);
1064                 return (error);
1065         }
1066
1067         /*
1068          * Depending on whether or not the application has bound or connected
1069          * the socket, we may have to do varying levels of work.  The optimal
1070          * case is for a connected UDP socket, as a global lock isn't
1071          * required at all.
1072          *
1073          * In order to decide which we need, we require stability of the
1074          * inpcb binding, which we ensure by acquiring a read lock on the
1075          * inpcb.  This doesn't strictly follow the lock order, so we play
1076          * the trylock and retry game; note that we may end up with more
1077          * conservative locks than required the second time around, so later
1078          * assertions have to accept that.  Further analysis of the number of
1079          * misses under contention is required.
1080          */
1081         sin = (struct sockaddr_in *)addr;
1082         INP_RLOCK(inp);
1083         if (sin != NULL &&
1084             (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
1085                 INP_RUNLOCK(inp);
1086                 INP_INFO_WLOCK(&V_udbinfo);
1087                 INP_WLOCK(inp);
1088                 unlock_udbinfo = 2;
1089         } else if ((sin != NULL && (
1090             (sin->sin_addr.s_addr == INADDR_ANY) ||
1091             (sin->sin_addr.s_addr == INADDR_BROADCAST) ||
1092             (inp->inp_laddr.s_addr == INADDR_ANY) ||
1093             (inp->inp_lport == 0))) ||
1094             (src.sin_family == AF_INET)) {
1095                 if (!INP_INFO_TRY_RLOCK(&V_udbinfo)) {
1096                         INP_RUNLOCK(inp);
1097                         INP_INFO_RLOCK(&V_udbinfo);
1098                         INP_RLOCK(inp);
1099                 }
1100                 unlock_udbinfo = 1;
1101         } else
1102                 unlock_udbinfo = 0;
1103
1104         /*
1105          * If the IP_SENDSRCADDR control message was specified, override the
1106          * source address for this datagram.  Its use is invalidated if the
1107          * address thus specified is incomplete or clobbers other inpcbs.
1108          */
1109         laddr = inp->inp_laddr;
1110         lport = inp->inp_lport;
1111         if (src.sin_family == AF_INET) {
1112                 INP_INFO_LOCK_ASSERT(&V_udbinfo);
1113                 if ((lport == 0) ||
1114                     (laddr.s_addr == INADDR_ANY &&
1115                      src.sin_addr.s_addr == INADDR_ANY)) {
1116                         error = EINVAL;
1117                         goto release;
1118                 }
1119                 error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
1120                     &laddr.s_addr, &lport, td->td_ucred);
1121                 if (error)
1122                         goto release;
1123         }
1124
1125         /*
1126          * If a UDP socket has been connected, then a local address/port will
1127          * have been selected and bound.
1128          *
1129          * If a UDP socket has not been connected to, then an explicit
1130          * destination address must be used, in which case a local
1131          * address/port may not have been selected and bound.
1132          */
1133         if (sin != NULL) {
1134                 INP_LOCK_ASSERT(inp);
1135                 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1136                         error = EISCONN;
1137                         goto release;
1138                 }
1139
1140                 /*
1141                  * Jail may rewrite the destination address, so let it do
1142                  * that before we use it.
1143                  */
1144                 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1145                 if (error)
1146                         goto release;
1147
1148                 /*
1149                  * If a local address or port hasn't yet been selected, or if
1150                  * the destination address needs to be rewritten due to using
1151                  * a special INADDR_ constant, invoke in_pcbconnect_setup()
1152                  * to do the heavy lifting.  Once a port is selected, we
1153                  * commit the binding back to the socket; we also commit the
1154                  * binding of the address if in jail.
1155                  *
1156                  * If we already have a valid binding and we're not
1157                  * requesting a destination address rewrite, use a fast path.
1158                  */
1159                 if (inp->inp_laddr.s_addr == INADDR_ANY ||
1160                     inp->inp_lport == 0 ||
1161                     sin->sin_addr.s_addr == INADDR_ANY ||
1162                     sin->sin_addr.s_addr == INADDR_BROADCAST) {
1163                         INP_INFO_LOCK_ASSERT(&V_udbinfo);
1164                         error = in_pcbconnect_setup(inp, addr, &laddr.s_addr,
1165                             &lport, &faddr.s_addr, &fport, NULL,
1166                             td->td_ucred);
1167                         if (error)
1168                                 goto release;
1169
1170                         /*
1171                          * XXXRW: Why not commit the port if the address is
1172                          * !INADDR_ANY?
1173                          */
1174                         /* Commit the local port if newly assigned. */
1175                         if (inp->inp_laddr.s_addr == INADDR_ANY &&
1176                             inp->inp_lport == 0) {
1177                                 INP_INFO_WLOCK_ASSERT(&V_udbinfo);
1178                                 INP_WLOCK_ASSERT(inp);
1179                                 /*
1180                                  * Remember addr if jailed, to prevent
1181                                  * rebinding.
1182                                  */
1183                                 if (prison_flag(td->td_ucred, PR_IP4))
1184                                         inp->inp_laddr = laddr;
1185                                 inp->inp_lport = lport;
1186                                 if (in_pcbinshash(inp) != 0) {
1187                                         inp->inp_lport = 0;
1188                                         error = EAGAIN;
1189                                         goto release;
1190                                 }
1191                                 inp->inp_flags |= INP_ANONPORT;
1192                         }
1193                 } else {
1194                         faddr = sin->sin_addr;
1195                         fport = sin->sin_port;
1196                 }
1197         } else {
1198                 INP_LOCK_ASSERT(inp);
1199                 faddr = inp->inp_faddr;
1200                 fport = inp->inp_fport;
1201                 if (faddr.s_addr == INADDR_ANY) {
1202                         error = ENOTCONN;
1203                         goto release;
1204                 }
1205         }
1206
1207         /*
1208          * Calculate data length and get a mbuf for UDP, IP, and possible
1209          * link-layer headers.  Immediate slide the data pointer back forward
1210          * since we won't use that space at this layer.
1211          */
1212         M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_DONTWAIT);
1213         if (m == NULL) {
1214                 error = ENOBUFS;
1215                 goto release;
1216         }
1217         m->m_data += max_linkhdr;
1218         m->m_len -= max_linkhdr;
1219         m->m_pkthdr.len -= max_linkhdr;
1220
1221         /*
1222          * Fill in mbuf with extended UDP header and addresses and length put
1223          * into network format.
1224          */
1225         ui = mtod(m, struct udpiphdr *);
1226         bzero(ui->ui_x1, sizeof(ui->ui_x1));    /* XXX still needed? */
1227         ui->ui_pr = IPPROTO_UDP;
1228         ui->ui_src = laddr;
1229         ui->ui_dst = faddr;
1230         ui->ui_sport = lport;
1231         ui->ui_dport = fport;
1232         ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
1233
1234         /*
1235          * Set the Don't Fragment bit in the IP header.
1236          */
1237         if (inp->inp_flags & INP_DONTFRAG) {
1238                 struct ip *ip;
1239
1240                 ip = (struct ip *)&ui->ui_i;
1241                 ip->ip_off |= IP_DF;
1242         }
1243
1244         ipflags = 0;
1245         if (inp->inp_socket->so_options & SO_DONTROUTE)
1246                 ipflags |= IP_ROUTETOIF;
1247         if (inp->inp_socket->so_options & SO_BROADCAST)
1248                 ipflags |= IP_ALLOWBROADCAST;
1249         if (inp->inp_flags & INP_ONESBCAST)
1250                 ipflags |= IP_SENDONES;
1251
1252 #ifdef MAC
1253         mac_inpcb_create_mbuf(inp, m);
1254 #endif
1255
1256         /*
1257          * Set up checksum and output datagram.
1258          */
1259         if (V_udp_cksum) {
1260                 if (inp->inp_flags & INP_ONESBCAST)
1261                         faddr.s_addr = INADDR_BROADCAST;
1262                 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
1263                     htons((u_short)len + sizeof(struct udphdr) + IPPROTO_UDP));
1264                 m->m_pkthdr.csum_flags = CSUM_UDP;
1265                 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1266         } else
1267                 ui->ui_sum = 0;
1268         ((struct ip *)ui)->ip_len = sizeof (struct udpiphdr) + len;
1269         ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl;    /* XXX */
1270         ((struct ip *)ui)->ip_tos = tos;                /* XXX */
1271         UDPSTAT_INC(udps_opackets);
1272
1273         if (unlock_udbinfo == 2)
1274                 INP_INFO_WUNLOCK(&V_udbinfo);
1275         else if (unlock_udbinfo == 1)
1276                 INP_INFO_RUNLOCK(&V_udbinfo);
1277         error = ip_output(m, inp->inp_options, NULL, ipflags,
1278             inp->inp_moptions, inp);
1279         if (unlock_udbinfo == 2)
1280                 INP_WUNLOCK(inp);
1281         else
1282                 INP_RUNLOCK(inp);
1283         return (error);
1284
1285 release:
1286         if (unlock_udbinfo == 2) {
1287                 INP_WUNLOCK(inp);
1288                 INP_INFO_WUNLOCK(&V_udbinfo);
1289         } else if (unlock_udbinfo == 1) {
1290                 INP_RUNLOCK(inp);
1291                 INP_INFO_RUNLOCK(&V_udbinfo);
1292         } else
1293                 INP_RUNLOCK(inp);
1294         m_freem(m);
1295         return (error);
1296 }
1297
1298
1299 #if defined(IPSEC) && defined(IPSEC_NAT_T)
1300 #ifdef INET
1301 /*
1302  * Potentially decap ESP in UDP frame.  Check for an ESP header
1303  * and optional marker; if present, strip the UDP header and
1304  * push the result through IPSec.
1305  *
1306  * Returns mbuf to be processed (potentially re-allocated) or
1307  * NULL if consumed and/or processed.
1308  */
1309 static struct mbuf *
1310 udp4_espdecap(struct inpcb *inp, struct mbuf *m, int off)
1311 {
1312         size_t minlen, payload, skip, iphlen;
1313         caddr_t data;
1314         struct udpcb *up;
1315         struct m_tag *tag;
1316         struct udphdr *udphdr;
1317         struct ip *ip;
1318
1319         INP_RLOCK_ASSERT(inp);
1320
1321         /*
1322          * Pull up data so the longest case is contiguous:
1323          *    IP/UDP hdr + non ESP marker + ESP hdr.
1324          */
1325         minlen = off + sizeof(uint64_t) + sizeof(struct esp);
1326         if (minlen > m->m_pkthdr.len)
1327                 minlen = m->m_pkthdr.len;
1328         if ((m = m_pullup(m, minlen)) == NULL) {
1329                 V_ipsec4stat.in_inval++;
1330                 return (NULL);          /* Bypass caller processing. */
1331         }
1332         data = mtod(m, caddr_t);        /* Points to ip header. */
1333         payload = m->m_len - off;       /* Size of payload. */
1334
1335         if (payload == 1 && data[off] == '\xff')
1336                 return (m);             /* NB: keepalive packet, no decap. */
1337
1338         up = intoudpcb(inp);
1339         KASSERT(up != NULL, ("%s: udpcb NULL", __func__));
1340         KASSERT((up->u_flags & UF_ESPINUDP_ALL) != 0,
1341             ("u_flags 0x%x", up->u_flags));
1342
1343         /*
1344          * Check that the payload is large enough to hold an
1345          * ESP header and compute the amount of data to remove.
1346          *
1347          * NB: the caller has already done a pullup for us.
1348          * XXX can we assume alignment and eliminate bcopys?
1349          */
1350         if (up->u_flags & UF_ESPINUDP_NON_IKE) {
1351                 /*
1352                  * draft-ietf-ipsec-nat-t-ike-0[01].txt and
1353                  * draft-ietf-ipsec-udp-encaps-(00/)01.txt, ignoring
1354                  * possible AH mode non-IKE marker+non-ESP marker
1355                  * from draft-ietf-ipsec-udp-encaps-00.txt.
1356                  */
1357                 uint64_t marker;
1358
1359                 if (payload <= sizeof(uint64_t) + sizeof(struct esp))
1360                         return (m);     /* NB: no decap. */
1361                 bcopy(data + off, &marker, sizeof(uint64_t));
1362                 if (marker != 0)        /* Non-IKE marker. */
1363                         return (m);     /* NB: no decap. */
1364                 skip = sizeof(uint64_t) + sizeof(struct udphdr);
1365         } else {
1366                 uint32_t spi;
1367
1368                 if (payload <= sizeof(struct esp)) {
1369                         V_ipsec4stat.in_inval++;
1370                         m_freem(m);
1371                         return (NULL);  /* Discard. */
1372                 }
1373                 bcopy(data + off, &spi, sizeof(uint32_t));
1374                 if (spi == 0)           /* Non-ESP marker. */
1375                         return (m);     /* NB: no decap. */
1376                 skip = sizeof(struct udphdr);
1377         }
1378
1379         /*
1380          * Setup a PACKET_TAG_IPSEC_NAT_T_PORT tag to remember
1381          * the UDP ports. This is required if we want to select
1382          * the right SPD for multiple hosts behind same NAT.
1383          *
1384          * NB: ports are maintained in network byte order everywhere
1385          *     in the NAT-T code.
1386          */
1387         tag = m_tag_get(PACKET_TAG_IPSEC_NAT_T_PORTS,
1388                 2 * sizeof(uint16_t), M_NOWAIT);
1389         if (tag == NULL) {
1390                 V_ipsec4stat.in_nomem++;
1391                 m_freem(m);
1392                 return (NULL);          /* Discard. */
1393         }
1394         iphlen = off - sizeof(struct udphdr);
1395         udphdr = (struct udphdr *)(data + iphlen);
1396         ((uint16_t *)(tag + 1))[0] = udphdr->uh_sport;
1397         ((uint16_t *)(tag + 1))[1] = udphdr->uh_dport;
1398         m_tag_prepend(m, tag);
1399
1400         /*
1401          * Remove the UDP header (and possibly the non ESP marker)
1402          * IP header length is iphlen
1403          * Before:
1404          *   <--- off --->
1405          *   +----+------+-----+
1406          *   | IP |  UDP | ESP |
1407          *   +----+------+-----+
1408          *        <-skip->
1409          * After:
1410          *          +----+-----+
1411          *          | IP | ESP |
1412          *          +----+-----+
1413          *   <-skip->
1414          */
1415         ovbcopy(data, data + skip, iphlen);
1416         m_adj(m, skip);
1417
1418         ip = mtod(m, struct ip *);
1419         ip->ip_len -= skip;
1420         ip->ip_p = IPPROTO_ESP;
1421
1422         /*
1423          * We cannot yet update the cksums so clear any
1424          * h/w cksum flags as they are no longer valid.
1425          */
1426         if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID)
1427                 m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID|CSUM_PSEUDO_HDR);
1428
1429         (void) ipsec4_common_input(m, iphlen, ip->ip_p);
1430         return (NULL);                  /* NB: consumed, bypass processing. */
1431 }
1432 #endif /* INET */
1433 #endif /* defined(IPSEC) && defined(IPSEC_NAT_T) */
1434
1435 static void
1436 udp_abort(struct socket *so)
1437 {
1438         struct inpcb *inp;
1439
1440         inp = sotoinpcb(so);
1441         KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
1442         INP_INFO_WLOCK(&V_udbinfo);
1443         INP_WLOCK(inp);
1444         if (inp->inp_faddr.s_addr != INADDR_ANY) {
1445                 in_pcbdisconnect(inp);
1446                 inp->inp_laddr.s_addr = INADDR_ANY;
1447                 soisdisconnected(so);
1448         }
1449         INP_WUNLOCK(inp);
1450         INP_INFO_WUNLOCK(&V_udbinfo);
1451 }
1452
1453 static int
1454 udp_attach(struct socket *so, int proto, struct thread *td)
1455 {
1456         struct inpcb *inp;
1457         int error;
1458
1459         inp = sotoinpcb(so);
1460         KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
1461         error = soreserve(so, udp_sendspace, udp_recvspace);
1462         if (error)
1463                 return (error);
1464         INP_INFO_WLOCK(&V_udbinfo);
1465         error = in_pcballoc(so, &V_udbinfo);
1466         if (error) {
1467                 INP_INFO_WUNLOCK(&V_udbinfo);
1468                 return (error);
1469         }
1470
1471         inp = sotoinpcb(so);
1472         inp->inp_vflag |= INP_IPV4;
1473         inp->inp_ip_ttl = V_ip_defttl;
1474
1475         error = udp_newudpcb(inp);
1476         if (error) {
1477                 in_pcbdetach(inp);
1478                 in_pcbfree(inp);
1479                 INP_INFO_WUNLOCK(&V_udbinfo);
1480                 return (error);
1481         }
1482
1483         INP_WUNLOCK(inp);
1484         INP_INFO_WUNLOCK(&V_udbinfo);
1485         return (0);
1486 }
1487
1488 int
1489 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f)
1490 {
1491         struct inpcb *inp;
1492         struct udpcb *up;
1493
1494         KASSERT(so->so_type == SOCK_DGRAM,
1495             ("udp_set_kernel_tunneling: !dgram"));
1496         inp = sotoinpcb(so);
1497         KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL"));
1498         INP_WLOCK(inp);
1499         up = intoudpcb(inp);
1500         if (up->u_tun_func != NULL) {
1501                 INP_WUNLOCK(inp);
1502                 return (EBUSY);
1503         }
1504         up->u_tun_func = f;
1505         INP_WUNLOCK(inp);
1506         return (0);
1507 }
1508
1509 static int
1510 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1511 {
1512         struct inpcb *inp;
1513         int error;
1514
1515         inp = sotoinpcb(so);
1516         KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1517         INP_INFO_WLOCK(&V_udbinfo);
1518         INP_WLOCK(inp);
1519         error = in_pcbbind(inp, nam, td->td_ucred);
1520         INP_WUNLOCK(inp);
1521         INP_INFO_WUNLOCK(&V_udbinfo);
1522         return (error);
1523 }
1524
1525 static void
1526 udp_close(struct socket *so)
1527 {
1528         struct inpcb *inp;
1529
1530         inp = sotoinpcb(so);
1531         KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1532         INP_INFO_WLOCK(&V_udbinfo);
1533         INP_WLOCK(inp);
1534         if (inp->inp_faddr.s_addr != INADDR_ANY) {
1535                 in_pcbdisconnect(inp);
1536                 inp->inp_laddr.s_addr = INADDR_ANY;
1537                 soisdisconnected(so);
1538         }
1539         INP_WUNLOCK(inp);
1540         INP_INFO_WUNLOCK(&V_udbinfo);
1541 }
1542
1543 static int
1544 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1545 {
1546         struct inpcb *inp;
1547         int error;
1548         struct sockaddr_in *sin;
1549
1550         inp = sotoinpcb(so);
1551         KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1552         INP_INFO_WLOCK(&V_udbinfo);
1553         INP_WLOCK(inp);
1554         if (inp->inp_faddr.s_addr != INADDR_ANY) {
1555                 INP_WUNLOCK(inp);
1556                 INP_INFO_WUNLOCK(&V_udbinfo);
1557                 return (EISCONN);
1558         }
1559         sin = (struct sockaddr_in *)nam;
1560         error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1561         if (error != 0) {
1562                 INP_WUNLOCK(inp);
1563                 INP_INFO_WUNLOCK(&V_udbinfo);
1564                 return (error);
1565         }
1566         error = in_pcbconnect(inp, nam, td->td_ucred);
1567         if (error == 0)
1568                 soisconnected(so);
1569         INP_WUNLOCK(inp);
1570         INP_INFO_WUNLOCK(&V_udbinfo);
1571         return (error);
1572 }
1573
1574 static void
1575 udp_detach(struct socket *so)
1576 {
1577         struct inpcb *inp;
1578         struct udpcb *up;
1579
1580         inp = sotoinpcb(so);
1581         KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1582         KASSERT(inp->inp_faddr.s_addr == INADDR_ANY,
1583             ("udp_detach: not disconnected"));
1584         INP_INFO_WLOCK(&V_udbinfo);
1585         INP_WLOCK(inp);
1586         up = intoudpcb(inp);
1587         KASSERT(up != NULL, ("%s: up == NULL", __func__));
1588         inp->inp_ppcb = NULL;
1589         in_pcbdetach(inp);
1590         in_pcbfree(inp);
1591         INP_INFO_WUNLOCK(&V_udbinfo);
1592         udp_discardcb(up);
1593 }
1594
1595 static int
1596 udp_disconnect(struct socket *so)
1597 {
1598         struct inpcb *inp;
1599
1600         inp = sotoinpcb(so);
1601         KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1602         INP_INFO_WLOCK(&V_udbinfo);
1603         INP_WLOCK(inp);
1604         if (inp->inp_faddr.s_addr == INADDR_ANY) {
1605                 INP_WUNLOCK(inp);
1606                 INP_INFO_WUNLOCK(&V_udbinfo);
1607                 return (ENOTCONN);
1608         }
1609
1610         in_pcbdisconnect(inp);
1611         inp->inp_laddr.s_addr = INADDR_ANY;
1612         SOCK_LOCK(so);
1613         so->so_state &= ~SS_ISCONNECTED;                /* XXX */
1614         SOCK_UNLOCK(so);
1615         INP_WUNLOCK(inp);
1616         INP_INFO_WUNLOCK(&V_udbinfo);
1617         return (0);
1618 }
1619
1620 static int
1621 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1622     struct mbuf *control, struct thread *td)
1623 {
1624         struct inpcb *inp;
1625
1626         inp = sotoinpcb(so);
1627         KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1628         return (udp_output(inp, m, addr, control, td));
1629 }
1630
1631 int
1632 udp_shutdown(struct socket *so)
1633 {
1634         struct inpcb *inp;
1635
1636         inp = sotoinpcb(so);
1637         KASSERT(inp != NULL, ("udp_shutdown: inp == NULL"));
1638         INP_WLOCK(inp);
1639         socantsendmore(so);
1640         INP_WUNLOCK(inp);
1641         return (0);
1642 }
1643
1644 struct pr_usrreqs udp_usrreqs = {
1645         .pru_abort =            udp_abort,
1646         .pru_attach =           udp_attach,
1647         .pru_bind =             udp_bind,
1648         .pru_connect =          udp_connect,
1649         .pru_control =          in_control,
1650         .pru_detach =           udp_detach,
1651         .pru_disconnect =       udp_disconnect,
1652         .pru_peeraddr =         in_getpeeraddr,
1653         .pru_send =             udp_send,
1654         .pru_soreceive =        soreceive_dgram,
1655         .pru_sosend =           sosend_dgram,
1656         .pru_shutdown =         udp_shutdown,
1657         .pru_sockaddr =         in_getsockaddr,
1658         .pru_sosetlabel =       in_pcbsosetlabel,
1659         .pru_close =            udp_close,
1660 };