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