sys/netinet/udp_usrreq.c

   1 /*-
   2  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
   3  *      The Regents of the University of California.
   4  * Copyright (c) 2008 Robert N. M. Watson
   5  * Copyright (c) 2010-2011 Juniper Networks, Inc.
   6  * All rights reserved.
   7  *
   8  * Portions of this software were developed by Robert N. M. Watson under
   9  * contract to Juniper Networks, Inc.
  10  *
  11  * Redistribution and use in source and binary forms, with or without
  12  * modification, are permitted provided that the following conditions
  13  * are met:
  14  * 1. Redistributions of source code must retain the above copyright
  15  *    notice, this list of conditions and the following disclaimer.
  16  * 2. Redistributions in binary form must reproduce the above copyright
  17  *    notice, this list of conditions and the following disclaimer in the
  18  *    documentation and/or other materials provided with the distribution.
  19  * 4. Neither the name of the University nor the names of its contributors
  20  *    may be used to endorse or promote products derived from this software
  21  *    without specific prior written permission.
  22  *
  23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  33  * SUCH DAMAGE.
  34  *
  35  *      @(#)udp_usrreq.c        8.6 (Berkeley) 5/23/95
  36  */
  37
  38 #include <sys/cdefs.h>
  39 __FBSDID("$FreeBSD$");
  40
  41 #include "opt_ipfw.h"
  42 #include "opt_inet.h"
  43 #include "opt_inet6.h"
  44 #include "opt_ipsec.h"
  45
  46 #include <sys/param.h>
  47 #include <sys/domain.h>
  48 #include <sys/eventhandler.h>
  49 #include <sys/jail.h>
  50 #include <sys/kernel.h>
  51 #include <sys/lock.h>
  52 #include <sys/malloc.h>
  53 #include <sys/mbuf.h>
  54 #include <sys/priv.h>
  55 #include <sys/proc.h>
  56 #include <sys/protosw.h>
  57 #include <sys/signalvar.h>
  58 #include <sys/socket.h>
  59 #include <sys/socketvar.h>
  60 #include <sys/sx.h>
  61 #include <sys/sysctl.h>
  62 #include <sys/syslog.h>
  63 #include <sys/systm.h>
  64
  65 #include <vm/uma.h>
  66
  67 #include <net/if.h>
  68 #include <net/route.h>
  69
  70 #include <netinet/in.h>
  71 #include <netinet/in_pcb.h>
  72 #include <netinet/in_systm.h>
  73 #include <netinet/in_var.h>
  74 #include <netinet/ip.h>
  75 #ifdef INET6
  76 #include <netinet/ip6.h>
  77 #endif
  78 #include <netinet/ip_icmp.h>
  79 #include <netinet/icmp_var.h>
  80 #include <netinet/ip_var.h>
  81 #include <netinet/ip_options.h>
  82 #ifdef INET6
  83 #include <netinet6/ip6_var.h>
  84 #endif
  85 #include <netinet/udp.h>
  86 #include <netinet/udp_var.h>
  87
  88 #ifdef IPSEC
  89 #include <netipsec/ipsec.h>
  90 #include <netipsec/esp.h>
  91 #endif
  92
  93 #include <machine/in_cksum.h>
  94
  95 #include <security/mac/mac_framework.h>
  96
  97 /*
  98  * UDP protocol implementation.
  99  * Per RFC 768, August, 1980.
 100  */
 101
 102 /*
 103  * BSD 4.2 defaulted the udp checksum to be off.  Turning off udp checksums
 104  * removes the only data integrity mechanism for packets and malformed
 105  * packets that would otherwise be discarded due to bad checksums, and may
 106  * cause problems (especially for NFS data blocks).
 107  */
 108 static int      udp_cksum = 1;
 109 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_RW, &udp_cksum,
 110     0, "compute udp checksum");
 111
 112 int     udp_log_in_vain = 0;
 113 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW,
 114     &udp_log_in_vain, 0, "Log all incoming UDP packets");
 115
 116 VNET_DEFINE(int, udp_blackhole) = 0;
 117 SYSCTL_VNET_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_RW,
 118     &VNET_NAME(udp_blackhole), 0,
 119     "Do not send port unreachables for refused connects");
 120
 121 u_long  udp_sendspace = 9216;           /* really max datagram size */
 122                                         /* 40 1K datagrams */
 123 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
 124     &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
 125
 126 u_long  udp_recvspace = 40 * (1024 +
 127 #ifdef INET6
 128                                       sizeof(struct sockaddr_in6)
 129 #else
 130                                       sizeof(struct sockaddr_in)
 131 #endif
 132                                       );
 133
 134 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
 135     &udp_recvspace, 0, "Maximum space for incoming UDP datagrams");
 136
 137 VNET_DEFINE(struct inpcbhead, udb);             /* from udp_var.h */
 138 VNET_DEFINE(struct inpcbinfo, udbinfo);
 139 static VNET_DEFINE(uma_zone_t, udpcb_zone);
 140 #define V_udpcb_zone                    VNET(udpcb_zone)
 141
 142 #ifndef UDBHASHSIZE
 143 #define UDBHASHSIZE     128
 144 #endif
 145
 146 VNET_DEFINE(struct udpstat, udpstat);           /* from udp_var.h */
 147 SYSCTL_VNET_STRUCT(_net_inet_udp, UDPCTL_STATS, stats, CTLFLAG_RW,
 148     &VNET_NAME(udpstat), udpstat,
 149     "UDP statistics (struct udpstat, netinet/udp_var.h)");
 150
 151 #ifdef INET
 152 static void     udp_detach(struct socket *so);
 153 static int      udp_output(struct inpcb *, struct mbuf *, struct sockaddr *,
 154                     struct mbuf *, struct thread *);
 155 #endif
 156
 157 #ifdef IPSEC
 158 #ifdef IPSEC_NAT_T
 159 #define UF_ESPINUDP_ALL (UF_ESPINUDP_NON_IKE|UF_ESPINUDP)
 160 #ifdef INET
 161 static struct mbuf *udp4_espdecap(struct inpcb *, struct mbuf *, int);
 162 #endif
 163 #endif /* IPSEC_NAT_T */
 164 #endif /* IPSEC */
 165
 166 static void
 167 udp_zone_change(void *tag)
 168 {
 169
 170         uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets);
 171         uma_zone_set_max(V_udpcb_zone, maxsockets);
 172 }
 173
 174 static int
 175 udp_inpcb_init(void *mem, int size, int flags)
 176 {
 177         struct inpcb *inp;
 178
 179         inp = mem;
 180         INP_LOCK_INIT(inp, "inp", "udpinp");
 181         return (0);
 182 }
 183
 184 void
 185 udp_init(void)
 186 {
 187
 188         in_pcbinfo_init(&V_udbinfo, "udp", &V_udb, UDBHASHSIZE, UDBHASHSIZE,
 189             "udp_inpcb", udp_inpcb_init, NULL, UMA_ZONE_NOFREE,
 190             IPI_HASHFIELDS_2TUPLE);
 191         V_udpcb_zone = uma_zcreate("udpcb", sizeof(struct udpcb),
 192             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
 193         uma_zone_set_max(V_udpcb_zone, maxsockets);
 194         EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL,
 195             EVENTHANDLER_PRI_ANY);
 196 }
 197
 198 /*
 199  * Kernel module interface for updating udpstat.  The argument is an index
 200  * into udpstat treated as an array of u_long.  While this encodes the
 201  * general layout of udpstat into the caller, it doesn't encode its location,
 202  * so that future changes to add, for example, per-CPU stats support won't
 203  * cause binary compatibility problems for kernel modules.
 204  */
 205 void
 206 kmod_udpstat_inc(int statnum)
 207 {
 208
 209         (*((u_long *)&V_udpstat + statnum))++;
 210 }
 211
 212 int
 213 udp_newudpcb(struct inpcb *inp)
 214 {
 215         struct udpcb *up;
 216
 217         up = uma_zalloc(V_udpcb_zone, M_NOWAIT | M_ZERO);
 218         if (up == NULL)
 219                 return (ENOBUFS);
 220         inp->inp_ppcb = up;
 221         return (0);
 222 }
 223
 224 void
 225 udp_discardcb(struct udpcb *up)
 226 {
 227
 228         uma_zfree(V_udpcb_zone, up);
 229 }
 230
 231 #ifdef VIMAGE
 232 void
 233 udp_destroy(void)
 234 {
 235
 236         in_pcbinfo_destroy(&V_udbinfo);
 237         uma_zdestroy(V_udpcb_zone);
 238 }
 239 #endif
 240
 241 #ifdef INET
 242 /*
 243  * Subroutine of udp_input(), which appends the provided mbuf chain to the
 244  * passed pcb/socket.  The caller must provide a sockaddr_in via udp_in that
 245  * contains the source address.  If the socket ends up being an IPv6 socket,
 246  * udp_append() will convert to a sockaddr_in6 before passing the address
 247  * into the socket code.
 248  */
 249 static void
 250 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off,
 251     struct sockaddr_in *udp_in)
 252 {
 253         struct sockaddr *append_sa;
 254         struct socket *so;
 255         struct mbuf *opts = 0;
 256 #ifdef INET6
 257         struct sockaddr_in6 udp_in6;
 258 #endif
 259         struct udpcb *up;
 260
 261         INP_LOCK_ASSERT(inp);
 262
 263         /*
 264          * Engage the tunneling protocol.
 265          */
 266         up = intoudpcb(inp);
 267         if (up->u_tun_func != NULL) {
 268                 (*up->u_tun_func)(n, off, inp);
 269                 return;
 270         }
 271
 272         if (n == NULL)
 273                 return;
 274
 275         off += sizeof(struct udphdr);
 276
 277 #ifdef IPSEC
 278         /* Check AH/ESP integrity. */
 279         if (ipsec4_in_reject(n, inp)) {
 280                 m_freem(n);
 281                 V_ipsec4stat.in_polvio++;
 282                 return;
 283         }
 284 #ifdef IPSEC_NAT_T
 285         up = intoudpcb(inp);
 286         KASSERT(up != NULL, ("%s: udpcb NULL", __func__));
 287         if (up->u_flags & UF_ESPINUDP_ALL) {    /* IPSec UDP encaps. */
 288                 n = udp4_espdecap(inp, n, off);
 289                 if (n == NULL)                          /* Consumed. */
 290                         return;
 291         }
 292 #endif /* IPSEC_NAT_T */
 293 #endif /* IPSEC */
 294 #ifdef MAC
 295         if (mac_inpcb_check_deliver(inp, n) != 0) {
 296                 m_freem(n);
 297                 return;
 298         }
 299 #endif /* MAC */
 300         if (inp->inp_flags & INP_CONTROLOPTS ||
 301             inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
 302 #ifdef INET6
 303                 if (inp->inp_vflag & INP_IPV6)
 304                         (void)ip6_savecontrol_v4(inp, n, &opts, NULL);
 305                 else
 306 #endif /* INET6 */
 307                         ip_savecontrol(inp, &opts, ip, n);
 308         }
 309 #ifdef INET6
 310         if (inp->inp_vflag & INP_IPV6) {
 311                 bzero(&udp_in6, sizeof(udp_in6));
 312                 udp_in6.sin6_len = sizeof(udp_in6);
 313                 udp_in6.sin6_family = AF_INET6;
 314                 in6_sin_2_v4mapsin6(udp_in, &udp_in6);
 315                 append_sa = (struct sockaddr *)&udp_in6;
 316         } else
 317 #endif /* INET6 */
 318                 append_sa = (struct sockaddr *)udp_in;
 319         m_adj(n, off);
 320
 321         so = inp->inp_socket;
 322         SOCKBUF_LOCK(&so->so_rcv);
 323         if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) {
 324                 SOCKBUF_UNLOCK(&so->so_rcv);
 325                 m_freem(n);
 326                 if (opts)
 327                         m_freem(opts);
 328                 UDPSTAT_INC(udps_fullsock);
 329         } else
 330                 sorwakeup_locked(so);
 331 }
 332
 333 void
 334 udp_input(struct mbuf *m, int off)
 335 {
 336         int iphlen = off;
 337         struct ip *ip;
 338         struct udphdr *uh;
 339         struct ifnet *ifp;
 340         struct inpcb *inp;
 341         int len;
 342         struct ip save_ip;
 343         struct sockaddr_in udp_in;
 344 #ifdef IPFIREWALL_FORWARD
 345         struct m_tag *fwd_tag;
 346 #endif
 347
 348         ifp = m->m_pkthdr.rcvif;
 349         UDPSTAT_INC(udps_ipackets);
 350
 351         /*
 352          * Strip IP options, if any; should skip this, make available to
 353          * user, and use on returned packets, but we don't yet have a way to
 354          * check the checksum with options still present.
 355          */
 356         if (iphlen > sizeof (struct ip)) {
 357                 ip_stripoptions(m, (struct mbuf *)0);
 358                 iphlen = sizeof(struct ip);
 359         }
 360
 361         /*
 362          * Get IP and UDP header together in first mbuf.
 363          */
 364         ip = mtod(m, struct ip *);
 365         if (m->m_len < iphlen + sizeof(struct udphdr)) {
 366                 if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == 0) {
 367                         UDPSTAT_INC(udps_hdrops);
 368                         return;
 369                 }
 370                 ip = mtod(m, struct ip *);
 371         }
 372         uh = (struct udphdr *)((caddr_t)ip + iphlen);
 373
 374         /*
 375          * Destination port of 0 is illegal, based on RFC768.
 376          */
 377         if (uh->uh_dport == 0)
 378                 goto badunlocked;
 379
 380         /*
 381          * Construct sockaddr format source address.  Stuff source address
 382          * and datagram in user buffer.
 383          */
 384         bzero(&udp_in, sizeof(udp_in));
 385         udp_in.sin_len = sizeof(udp_in);
 386         udp_in.sin_family = AF_INET;
 387         udp_in.sin_port = uh->uh_sport;
 388         udp_in.sin_addr = ip->ip_src;
 389
 390         /*
 391          * Make mbuf data length reflect UDP length.  If not enough data to
 392          * reflect UDP length, drop.
 393          */
 394         len = ntohs((u_short)uh->uh_ulen);
 395         if (ip->ip_len != len) {
 396                 if (len > ip->ip_len || len < sizeof(struct udphdr)) {
 397                         UDPSTAT_INC(udps_badlen);
 398                         goto badunlocked;
 399                 }
 400                 m_adj(m, len - ip->ip_len);
 401                 /* ip->ip_len = len; */
 402         }
 403
 404         /*
 405          * Save a copy of the IP header in case we want restore it for
 406          * sending an ICMP error message in response.
 407          */
 408         if (!V_udp_blackhole)
 409                 save_ip = *ip;
 410         else
 411                 memset(&save_ip, 0, sizeof(save_ip));
 412
 413         /*
 414          * Checksum extended UDP header and data.
 415          */
 416         if (uh->uh_sum) {
 417                 u_short uh_sum;
 418
 419                 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
 420                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
 421                                 uh_sum = m->m_pkthdr.csum_data;
 422                         else
 423                                 uh_sum = in_pseudo(ip->ip_src.s_addr,
 424                                     ip->ip_dst.s_addr, htonl((u_short)len +
 425                                     m->m_pkthdr.csum_data + IPPROTO_UDP));
 426                         uh_sum ^= 0xffff;
 427                 } else {
 428                         char b[9];
 429
 430                         bcopy(((struct ipovly *)ip)->ih_x1, b, 9);
 431                         bzero(((struct ipovly *)ip)->ih_x1, 9);
 432                         ((struct ipovly *)ip)->ih_len = uh->uh_ulen;
 433                         uh_sum = in_cksum(m, len + sizeof (struct ip));
 434                         bcopy(b, ((struct ipovly *)ip)->ih_x1, 9);
 435                 }
 436                 if (uh_sum) {
 437                         UDPSTAT_INC(udps_badsum);
 438                         m_freem(m);
 439                         return;
 440                 }
 441         } else
 442                 UDPSTAT_INC(udps_nosum);
 443
 444 #ifdef IPFIREWALL_FORWARD
 445         /*
 446          * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
 447          */
 448         fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
 449         if (fwd_tag != NULL) {
 450                 struct sockaddr_in *next_hop;
 451
 452                 /*
 453                  * Do the hack.
 454                  */
 455                 next_hop = (struct sockaddr_in *)(fwd_tag + 1);
 456                 ip->ip_dst = next_hop->sin_addr;
 457                 uh->uh_dport = ntohs(next_hop->sin_port);
 458
 459                 /*
 460                  * Remove the tag from the packet.  We don't need it anymore.
 461                  */
 462                 m_tag_delete(m, fwd_tag);
 463         }
 464 #endif
 465
 466         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
 467             in_broadcast(ip->ip_dst, ifp)) {
 468                 struct inpcb *last;
 469                 struct ip_moptions *imo;
 470
 471                 INP_INFO_RLOCK(&V_udbinfo);
 472                 last = NULL;
 473                 LIST_FOREACH(inp, &V_udb, inp_list) {
 474                         if (inp->inp_lport != uh->uh_dport)
 475                                 continue;
 476 #ifdef INET6
 477                         if ((inp->inp_vflag & INP_IPV4) == 0)
 478                                 continue;
 479 #endif
 480                         if (inp->inp_laddr.s_addr != INADDR_ANY &&
 481                             inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
 482                                 continue;
 483                         if (inp->inp_faddr.s_addr != INADDR_ANY &&
 484                             inp->inp_faddr.s_addr != ip->ip_src.s_addr)
 485                                 continue;
 486                         if (inp->inp_fport != 0 &&
 487                             inp->inp_fport != uh->uh_sport)
 488                                 continue;
 489
 490                         INP_RLOCK(inp);
 491
 492                         /*
 493                          * XXXRW: Because we weren't holding either the inpcb
 494                          * or the hash lock when we checked for a match
 495                          * before, we should probably recheck now that the
 496                          * inpcb lock is held.
 497                          */
 498
 499                         /*
 500                          * Handle socket delivery policy for any-source
 501                          * and source-specific multicast. [RFC3678]
 502                          */
 503                         imo = inp->inp_moptions;
 504                         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
 505                                 struct sockaddr_in       group;
 506                                 int                      blocked;
 507                                 if (imo == NULL) {
 508                                         INP_RUNLOCK(inp);
 509                                         continue;
 510                                 }
 511                                 bzero(&group, sizeof(struct sockaddr_in));
 512                                 group.sin_len = sizeof(struct sockaddr_in);
 513                                 group.sin_family = AF_INET;
 514                                 group.sin_addr = ip->ip_dst;
 515
 516                                 blocked = imo_multi_filter(imo, ifp,
 517                                         (struct sockaddr *)&group,
 518                                         (struct sockaddr *)&udp_in);
 519                                 if (blocked != MCAST_PASS) {
 520                                         if (blocked == MCAST_NOTGMEMBER)
 521                                                 IPSTAT_INC(ips_notmember);
 522                                         if (blocked == MCAST_NOTSMEMBER ||
 523                                             blocked == MCAST_MUTED)
 524                                                 UDPSTAT_INC(udps_filtermcast);
 525                                         INP_RUNLOCK(inp);
 526                                         continue;
 527                                 }
 528                         }
 529                         if (last != NULL) {
 530                                 struct mbuf *n;
 531
 532                                 n = m_copy(m, 0, M_COPYALL);
 533                                 udp_append(last, ip, n, iphlen, &udp_in);
 534                                 INP_RUNLOCK(last);
 535                         }
 536                         last = inp;
 537                         /*
 538                          * Don't look for additional matches if this one does
 539                          * not have either the SO_REUSEPORT or SO_REUSEADDR
 540                          * socket options set.  This heuristic avoids
 541                          * searching through all pcbs in the common case of a
 542                          * non-shared port.  It assumes that an application
 543                          * will never clear these options after setting them.
 544                          */
 545                         if ((last->inp_socket->so_options &
 546                             (SO_REUSEPORT|SO_REUSEADDR)) == 0)
 547                                 break;
 548                 }
 549
 550                 if (last == NULL) {
 551                         /*
 552                          * No matching pcb found; discard datagram.  (No need
 553                          * to send an ICMP Port Unreachable for a broadcast
 554                          * or multicast datgram.)
 555                          */
 556                         UDPSTAT_INC(udps_noportbcast);
 557                         if (inp)
 558                                 INP_RUNLOCK(inp);
 559                         INP_INFO_RUNLOCK(&V_udbinfo);
 560                         goto badunlocked;
 561                 }
 562                 udp_append(last, ip, m, iphlen, &udp_in);
 563                 INP_RUNLOCK(last);
 564                 INP_INFO_RUNLOCK(&V_udbinfo);
 565                 return;
 566         }
 567
 568         /*
 569          * Locate pcb for datagram.
 570          */
 571         inp = in_pcblookup_mbuf(&V_udbinfo, ip->ip_src, uh->uh_sport,
 572             ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB,
 573             ifp, m);
 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 badunlocked;
 588                 }
 589                 if (V_udp_blackhole)
 590                         goto badunlocked;
 591                 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
 592                         goto badunlocked;
 593                 *ip = save_ip;
 594                 ip->ip_len += iphlen;
 595                 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
 596                 return;
 597         }
 598
 599         /*
 600          * Check the minimum TTL for socket.
 601          */
 602         INP_RLOCK_ASSERT(inp);
 603         if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) {
 604                 INP_RUNLOCK(inp);
 605                 m_freem(m);
 606                 return;
 607         }
 608         udp_append(inp, ip, m, iphlen, &udp_in);
 609         INP_RUNLOCK(inp);
 610         return;
 611
 612 badunlocked:
 613         m_freem(m);
 614 }
 615 #endif /* INET */
 616
 617 /*
 618  * Notify a udp user of an asynchronous error; just wake up so that they can
 619  * collect error status.
 620  */
 621 struct inpcb *
 622 udp_notify(struct inpcb *inp, int errno)
 623 {
 624
 625         /*
 626          * While udp_ctlinput() always calls udp_notify() with a read lock
 627          * when invoking it directly, in_pcbnotifyall() currently uses write
 628          * locks due to sharing code with TCP.  For now, accept either a read
 629          * or a write lock, but a read lock is sufficient.
 630          */
 631         INP_LOCK_ASSERT(inp);
 632
 633         inp->inp_socket->so_error = errno;
 634         sorwakeup(inp->inp_socket);
 635         sowwakeup(inp->inp_socket);
 636         return (inp);
 637 }
 638
 639 #ifdef INET
 640 void
 641 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
 642 {
 643         struct ip *ip = vip;
 644         struct udphdr *uh;
 645         struct in_addr faddr;
 646         struct inpcb *inp;
 647
 648         faddr = ((struct sockaddr_in *)sa)->sin_addr;
 649         if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
 650                 return;
 651
 652         /*
 653          * Redirects don't need to be handled up here.
 654          */
 655         if (PRC_IS_REDIRECT(cmd))
 656                 return;
 657
 658         /*
 659          * Hostdead is ugly because it goes linearly through all PCBs.
 660          *
 661          * XXX: We never get this from ICMP, otherwise it makes an excellent
 662          * DoS attack on machines with many connections.
 663          */
 664         if (cmd == PRC_HOSTDEAD)
 665                 ip = NULL;
 666         else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
 667                 return;
 668         if (ip != NULL) {
 669                 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
 670                 inp = in_pcblookup(&V_udbinfo, faddr, uh->uh_dport,
 671                     ip->ip_src, uh->uh_sport, INPLOOKUP_RLOCKPCB, NULL);
 672                 if (inp != NULL) {
 673                         INP_RLOCK_ASSERT(inp);
 674                         if (inp->inp_socket != NULL) {
 675                                 udp_notify(inp, inetctlerrmap[cmd]);
 676                         }
 677                         INP_RUNLOCK(inp);
 678                 }
 679         } else
 680                 in_pcbnotifyall(&V_udbinfo, faddr, inetctlerrmap[cmd],
 681                     udp_notify);
 682 }
 683 #endif /* INET */
 684
 685 static int
 686 udp_pcblist(SYSCTL_HANDLER_ARGS)
 687 {
 688         int error, i, n;
 689         struct inpcb *inp, **inp_list;
 690         inp_gen_t gencnt;
 691         struct xinpgen xig;
 692
 693         /*
 694          * The process of preparing the PCB list is too time-consuming and
 695          * resource-intensive to repeat twice on every request.
 696          */
 697         if (req->oldptr == 0) {
 698                 n = V_udbinfo.ipi_count;
 699                 n += imax(n / 8, 10);
 700                 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
 701                 return (0);
 702         }
 703
 704         if (req->newptr != 0)
 705                 return (EPERM);
 706
 707         /*
 708          * OK, now we're committed to doing something.
 709          */
 710         INP_INFO_RLOCK(&V_udbinfo);
 711         gencnt = V_udbinfo.ipi_gencnt;
 712         n = V_udbinfo.ipi_count;
 713         INP_INFO_RUNLOCK(&V_udbinfo);
 714
 715         error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
 716                 + n * sizeof(struct xinpcb));
 717         if (error != 0)
 718                 return (error);
 719
 720         xig.xig_len = sizeof xig;
 721         xig.xig_count = n;
 722         xig.xig_gen = gencnt;
 723         xig.xig_sogen = so_gencnt;
 724         error = SYSCTL_OUT(req, &xig, sizeof xig);
 725         if (error)
 726                 return (error);
 727
 728         inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
 729         if (inp_list == 0)
 730                 return (ENOMEM);
 731
 732         INP_INFO_RLOCK(&V_udbinfo);
 733         for (inp = LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n;
 734              inp = LIST_NEXT(inp, inp_list)) {
 735                 INP_WLOCK(inp);
 736                 if (inp->inp_gencnt <= gencnt &&
 737                     cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
 738                         in_pcbref(inp);
 739                         inp_list[i++] = inp;
 740                 }
 741                 INP_WUNLOCK(inp);
 742         }
 743         INP_INFO_RUNLOCK(&V_udbinfo);
 744         n = i;
 745
 746         error = 0;
 747         for (i = 0; i < n; i++) {
 748                 inp = inp_list[i];
 749                 INP_RLOCK(inp);
 750                 if (inp->inp_gencnt <= gencnt) {
 751                         struct xinpcb xi;
 752
 753                         bzero(&xi, sizeof(xi));
 754                         xi.xi_len = sizeof xi;
 755                         /* XXX should avoid extra copy */
 756                         bcopy(inp, &xi.xi_inp, sizeof *inp);
 757                         if (inp->inp_socket)
 758                                 sotoxsocket(inp->inp_socket, &xi.xi_socket);
 759                         xi.xi_inp.inp_gencnt = inp->inp_gencnt;
 760                         INP_RUNLOCK(inp);
 761                         error = SYSCTL_OUT(req, &xi, sizeof xi);
 762                 } else
 763                         INP_RUNLOCK(inp);
 764         }
 765         INP_INFO_WLOCK(&V_udbinfo);
 766         for (i = 0; i < n; i++) {
 767                 inp = inp_list[i];
 768                 INP_RLOCK(inp);
 769                 if (!in_pcbrele_rlocked(inp))
 770                         INP_RUNLOCK(inp);
 771         }
 772         INP_INFO_WUNLOCK(&V_udbinfo);
 773
 774         if (!error) {
 775                 /*
 776                  * Give the user an updated idea of our state.  If the
 777                  * generation differs from what we told her before, she knows
 778                  * that something happened while we were processing this
 779                  * request, and it might be necessary to retry.
 780                  */
 781                 INP_INFO_RLOCK(&V_udbinfo);
 782                 xig.xig_gen = V_udbinfo.ipi_gencnt;
 783                 xig.xig_sogen = so_gencnt;
 784                 xig.xig_count = V_udbinfo.ipi_count;
 785                 INP_INFO_RUNLOCK(&V_udbinfo);
 786                 error = SYSCTL_OUT(req, &xig, sizeof xig);
 787         }
 788         free(inp_list, M_TEMP);
 789         return (error);
 790 }
 791
 792 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist,
 793     CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
 794     udp_pcblist, "S,xinpcb", "List of active UDP sockets");
 795
 796 #ifdef INET
 797 static int
 798 udp_getcred(SYSCTL_HANDLER_ARGS)
 799 {
 800         struct xucred xuc;
 801         struct sockaddr_in addrs[2];
 802         struct inpcb *inp;
 803         int error;
 804
 805         error = priv_check(req->td, PRIV_NETINET_GETCRED);
 806         if (error)
 807                 return (error);
 808         error = SYSCTL_IN(req, addrs, sizeof(addrs));
 809         if (error)
 810                 return (error);
 811         inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
 812             addrs[0].sin_addr, addrs[0].sin_port,
 813             INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
 814         if (inp != NULL) {
 815                 INP_RLOCK_ASSERT(inp);
 816                 if (inp->inp_socket == NULL)
 817                         error = ENOENT;
 818                 if (error == 0)
 819                         error = cr_canseeinpcb(req->td->td_ucred, inp);
 820                 if (error == 0)
 821                         cru2x(inp->inp_cred, &xuc);
 822                 INP_RUNLOCK(inp);
 823         } else
 824                 error = ENOENT;
 825         if (error == 0)
 826                 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
 827         return (error);
 828 }
 829
 830 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
 831     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
 832     udp_getcred, "S,xucred", "Get the xucred of a UDP connection");
 833 #endif /* INET */
 834
 835 int
 836 udp_ctloutput(struct socket *so, struct sockopt *sopt)
 837 {
 838         int error = 0, optval;
 839         struct inpcb *inp;
 840 #ifdef IPSEC_NAT_T
 841         struct udpcb *up;
 842 #endif
 843
 844         inp = sotoinpcb(so);
 845         KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
 846         INP_WLOCK(inp);
 847         if (sopt->sopt_level != IPPROTO_UDP) {
 848 #ifdef INET6
 849                 if (INP_CHECK_SOCKAF(so, AF_INET6)) {
 850                         INP_WUNLOCK(inp);
 851                         error = ip6_ctloutput(so, sopt);
 852                 }
 853 #endif
 854 #if defined(INET) && defined(INET6)
 855                 else
 856 #endif
 857 #ifdef INET
 858                 {
 859                         INP_WUNLOCK(inp);
 860                         error = ip_ctloutput(so, sopt);
 861                 }
 862 #endif
 863                 return (error);
 864         }
 865
 866         switch (sopt->sopt_dir) {
 867         case SOPT_SET:
 868                 switch (sopt->sopt_name) {
 869                 case UDP_ENCAP:
 870                         INP_WUNLOCK(inp);
 871                         error = sooptcopyin(sopt, &optval, sizeof optval,
 872                                             sizeof optval);
 873                         if (error)
 874                                 break;
 875                         inp = sotoinpcb(so);
 876                         KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
 877                         INP_WLOCK(inp);
 878 #ifdef IPSEC_NAT_T
 879                         up = intoudpcb(inp);
 880                         KASSERT(up != NULL, ("%s: up == NULL", __func__));
 881 #endif
 882                         switch (optval) {
 883                         case 0:
 884                                 /* Clear all UDP encap. */
 885 #ifdef IPSEC_NAT_T
 886                                 up->u_flags &= ~UF_ESPINUDP_ALL;
 887 #endif
 888                                 break;
 889 #ifdef IPSEC_NAT_T
 890                         case UDP_ENCAP_ESPINUDP:
 891                         case UDP_ENCAP_ESPINUDP_NON_IKE:
 892                                 up->u_flags &= ~UF_ESPINUDP_ALL;
 893                                 if (optval == UDP_ENCAP_ESPINUDP)
 894                                         up->u_flags |= UF_ESPINUDP;
 895                                 else if (optval == UDP_ENCAP_ESPINUDP_NON_IKE)
 896                                         up->u_flags |= UF_ESPINUDP_NON_IKE;
 897                                 break;
 898 #endif
 899                         default:
 900                                 error = EINVAL;
 901                                 break;
 902                         }
 903                         INP_WUNLOCK(inp);
 904                         break;
 905                 default:
 906                         INP_WUNLOCK(inp);
 907                         error = ENOPROTOOPT;
 908                         break;
 909                 }
 910                 break;
 911         case SOPT_GET:
 912                 switch (sopt->sopt_name) {
 913 #ifdef IPSEC_NAT_T
 914                 case UDP_ENCAP:
 915                         up = intoudpcb(inp);
 916                         KASSERT(up != NULL, ("%s: up == NULL", __func__));
 917                         optval = up->u_flags & UF_ESPINUDP_ALL;
 918                         INP_WUNLOCK(inp);
 919                         error = sooptcopyout(sopt, &optval, sizeof optval);
 920                         break;
 921 #endif
 922                 default:
 923                         INP_WUNLOCK(inp);
 924                         error = ENOPROTOOPT;
 925                         break;
 926                 }
 927                 break;
 928         }
 929         return (error);
 930 }
 931
 932 #ifdef INET
 933 #define UH_WLOCKED      2
 934 #define UH_RLOCKED      1
 935 #define UH_UNLOCKED     0
 936 static int
 937 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
 938     struct mbuf *control, struct thread *td)
 939 {
 940         struct udpiphdr *ui;
 941         int len = m->m_pkthdr.len;
 942         struct in_addr faddr, laddr;
 943         struct cmsghdr *cm;
 944         struct sockaddr_in *sin, src;
 945         int error = 0;
 946         int ipflags;
 947         u_short fport, lport;
 948         int unlock_udbinfo;
 949
 950         /*
 951          * udp_output() may need to temporarily bind or connect the current
 952          * inpcb.  As such, we don't know up front whether we will need the
 953          * pcbinfo lock or not.  Do any work to decide what is needed up
 954          * front before acquiring any locks.
 955          */
 956         if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
 957                 if (control)
 958                         m_freem(control);
 959                 m_freem(m);
 960                 return (EMSGSIZE);
 961         }
 962
 963         src.sin_family = 0;
 964         if (control != NULL) {
 965                 /*
 966                  * XXX: Currently, we assume all the optional information is
 967                  * stored in a single mbuf.
 968                  */
 969                 if (control->m_next) {
 970                         m_freem(control);
 971                         m_freem(m);
 972                         return (EINVAL);
 973                 }
 974                 for (; control->m_len > 0;
 975                     control->m_data += CMSG_ALIGN(cm->cmsg_len),
 976                     control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
 977                         cm = mtod(control, struct cmsghdr *);
 978                         if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0
 979                             || cm->cmsg_len > control->m_len) {
 980                                 error = EINVAL;
 981                                 break;
 982                         }
 983                         if (cm->cmsg_level != IPPROTO_IP)
 984                                 continue;
 985
 986                         switch (cm->cmsg_type) {
 987                         case IP_SENDSRCADDR:
 988                                 if (cm->cmsg_len !=
 989                                     CMSG_LEN(sizeof(struct in_addr))) {
 990                                         error = EINVAL;
 991                                         break;
 992                                 }
 993                                 bzero(&src, sizeof(src));
 994                                 src.sin_family = AF_INET;
 995                                 src.sin_len = sizeof(src);
 996                                 src.sin_port = inp->inp_lport;
 997                                 src.sin_addr =
 998                                     *(struct in_addr *)CMSG_DATA(cm);
 999                                 break;
1000
1001                         default:
1002                                 error = ENOPROTOOPT;
1003                                 break;
1004                         }
1005                         if (error)
1006                                 break;
1007                 }
1008                 m_freem(control);
1009         }
1010         if (error) {
1011                 m_freem(m);
1012                 return (error);
1013         }
1014
1015         /*
1016          * Depending on whether or not the application has bound or connected
1017          * the socket, we may have to do varying levels of work.  The optimal
1018          * case is for a connected UDP socket, as a global lock isn't
1019          * required at all.
1020          *
1021          * In order to decide which we need, we require stability of the
1022          * inpcb binding, which we ensure by acquiring a read lock on the
1023          * inpcb.  This doesn't strictly follow the lock order, so we play
1024          * the trylock and retry game; note that we may end up with more
1025          * conservative locks than required the second time around, so later
1026          * assertions have to accept that.  Further analysis of the number of
1027          * misses under contention is required.
1028          *
1029          * XXXRW: Check that hash locking update here is correct.
1030          */
1031         sin = (struct sockaddr_in *)addr;
1032         INP_RLOCK(inp);
1033         if (sin != NULL &&
1034             (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
1035                 INP_RUNLOCK(inp);
1036                 INP_WLOCK(inp);
1037                 INP_HASH_WLOCK(&V_udbinfo);
1038                 unlock_udbinfo = UH_WLOCKED;
1039         } else if ((sin != NULL && (
1040             (sin->sin_addr.s_addr == INADDR_ANY) ||
1041             (sin->sin_addr.s_addr == INADDR_BROADCAST) ||
1042             (inp->inp_laddr.s_addr == INADDR_ANY) ||
1043             (inp->inp_lport == 0))) ||
1044             (src.sin_family == AF_INET)) {
1045                 INP_HASH_RLOCK(&V_udbinfo);
1046                 unlock_udbinfo = UH_RLOCKED;
1047         } else
1048                 unlock_udbinfo = UH_UNLOCKED;
1049
1050         /*
1051          * If the IP_SENDSRCADDR control message was specified, override the
1052          * source address for this datagram.  Its use is invalidated if the
1053          * address thus specified is incomplete or clobbers other inpcbs.
1054          */
1055         laddr = inp->inp_laddr;
1056         lport = inp->inp_lport;
1057         if (src.sin_family == AF_INET) {
1058                 INP_HASH_LOCK_ASSERT(&V_udbinfo);
1059                 if ((lport == 0) ||
1060                     (laddr.s_addr == INADDR_ANY &&
1061                      src.sin_addr.s_addr == INADDR_ANY)) {
1062                         error = EINVAL;
1063                         goto release;
1064                 }
1065                 error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
1066                     &laddr.s_addr, &lport, td->td_ucred);
1067                 if (error)
1068                         goto release;
1069         }
1070
1071         /*
1072          * If a UDP socket has been connected, then a local address/port will
1073          * have been selected and bound.
1074          *
1075          * If a UDP socket has not been connected to, then an explicit
1076          * destination address must be used, in which case a local
1077          * address/port may not have been selected and bound.
1078          */
1079         if (sin != NULL) {
1080                 INP_LOCK_ASSERT(inp);
1081                 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1082                         error = EISCONN;
1083                         goto release;
1084                 }
1085
1086                 /*
1087                  * Jail may rewrite the destination address, so let it do
1088                  * that before we use it.
1089                  */
1090                 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1091                 if (error)
1092                         goto release;
1093
1094                 /*
1095                  * If a local address or port hasn't yet been selected, or if
1096                  * the destination address needs to be rewritten due to using
1097                  * a special INADDR_ constant, invoke in_pcbconnect_setup()
1098                  * to do the heavy lifting.  Once a port is selected, we
1099                  * commit the binding back to the socket; we also commit the
1100                  * binding of the address if in jail.
1101                  *
1102                  * If we already have a valid binding and we're not
1103                  * requesting a destination address rewrite, use a fast path.
1104                  */
1105                 if (inp->inp_laddr.s_addr == INADDR_ANY ||
1106                     inp->inp_lport == 0 ||
1107                     sin->sin_addr.s_addr == INADDR_ANY ||
1108                     sin->sin_addr.s_addr == INADDR_BROADCAST) {
1109                         INP_HASH_LOCK_ASSERT(&V_udbinfo);
1110                         error = in_pcbconnect_setup(inp, addr, &laddr.s_addr,
1111                             &lport, &faddr.s_addr, &fport, NULL,
1112                             td->td_ucred);
1113                         if (error)
1114                                 goto release;
1115
1116                         /*
1117                          * XXXRW: Why not commit the port if the address is
1118                          * !INADDR_ANY?
1119                          */
1120                         /* Commit the local port if newly assigned. */
1121                         if (inp->inp_laddr.s_addr == INADDR_ANY &&
1122                             inp->inp_lport == 0) {
1123                                 INP_WLOCK_ASSERT(inp);
1124                                 INP_HASH_WLOCK_ASSERT(&V_udbinfo);
1125                                 /*
1126                                  * Remember addr if jailed, to prevent
1127                                  * rebinding.
1128                                  */
1129                                 if (prison_flag(td->td_ucred, PR_IP4))
1130                                         inp->inp_laddr = laddr;
1131                                 inp->inp_lport = lport;
1132                                 if (in_pcbinshash(inp) != 0) {
1133                                         inp->inp_lport = 0;
1134                                         error = EAGAIN;
1135                                         goto release;
1136                                 }
1137                                 inp->inp_flags |= INP_ANONPORT;
1138                         }
1139                 } else {
1140                         faddr = sin->sin_addr;
1141                         fport = sin->sin_port;
1142                 }
1143         } else {
1144                 INP_LOCK_ASSERT(inp);
1145                 faddr = inp->inp_faddr;
1146                 fport = inp->inp_fport;
1147                 if (faddr.s_addr == INADDR_ANY) {
1148                         error = ENOTCONN;
1149                         goto release;
1150                 }
1151         }
1152
1153         /*
1154          * Calculate data length and get a mbuf for UDP, IP, and possible
1155          * link-layer headers.  Immediate slide the data pointer back forward
1156          * since we won't use that space at this layer.
1157          */
1158         M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_DONTWAIT);
1159         if (m == NULL) {
1160                 error = ENOBUFS;
1161                 goto release;
1162         }
1163         m->m_data += max_linkhdr;
1164         m->m_len -= max_linkhdr;
1165         m->m_pkthdr.len -= max_linkhdr;
1166
1167         /*
1168          * Fill in mbuf with extended UDP header and addresses and length put
1169          * into network format.
1170          */
1171         ui = mtod(m, struct udpiphdr *);
1172         bzero(ui->ui_x1, sizeof(ui->ui_x1));    /* XXX still needed? */
1173         ui->ui_pr = IPPROTO_UDP;
1174         ui->ui_src = laddr;
1175         ui->ui_dst = faddr;
1176         ui->ui_sport = lport;
1177         ui->ui_dport = fport;
1178         ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
1179
1180         /*
1181          * Set the Don't Fragment bit in the IP header.
1182          */
1183         if (inp->inp_flags & INP_DONTFRAG) {
1184                 struct ip *ip;
1185
1186                 ip = (struct ip *)&ui->ui_i;
1187                 ip->ip_off |= IP_DF;
1188         }
1189
1190         ipflags = 0;
1191         if (inp->inp_socket->so_options & SO_DONTROUTE)
1192                 ipflags |= IP_ROUTETOIF;
1193         if (inp->inp_socket->so_options & SO_BROADCAST)
1194                 ipflags |= IP_ALLOWBROADCAST;
1195         if (inp->inp_flags & INP_ONESBCAST)
1196                 ipflags |= IP_SENDONES;
1197
1198 #ifdef MAC
1199         mac_inpcb_create_mbuf(inp, m);
1200 #endif
1201
1202         /*
1203          * Set up checksum and output datagram.
1204          */
1205         if (udp_cksum) {
1206                 if (inp->inp_flags & INP_ONESBCAST)
1207                         faddr.s_addr = INADDR_BROADCAST;
1208                 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
1209                     htons((u_short)len + sizeof(struct udphdr) + IPPROTO_UDP));
1210                 m->m_pkthdr.csum_flags = CSUM_UDP;
1211                 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1212         } else
1213                 ui->ui_sum = 0;
1214         ((struct ip *)ui)->ip_len = sizeof (struct udpiphdr) + len;
1215         ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl;    /* XXX */
1216         ((struct ip *)ui)->ip_tos = inp->inp_ip_tos;    /* XXX */
1217         UDPSTAT_INC(udps_opackets);
1218
1219         if (unlock_udbinfo == UH_WLOCKED)
1220                 INP_HASH_WUNLOCK(&V_udbinfo);
1221         else if (unlock_udbinfo == UH_RLOCKED)
1222                 INP_HASH_RUNLOCK(&V_udbinfo);
1223         error = ip_output(m, inp->inp_options, NULL, ipflags,
1224             inp->inp_moptions, inp);
1225         if (unlock_udbinfo == UH_WLOCKED)
1226                 INP_WUNLOCK(inp);
1227         else
1228                 INP_RUNLOCK(inp);
1229         return (error);
1230
1231 release:
1232         if (unlock_udbinfo == UH_WLOCKED) {
1233                 INP_HASH_WUNLOCK(&V_udbinfo);
1234                 INP_WUNLOCK(inp);
1235         } else if (unlock_udbinfo == UH_RLOCKED) {
1236                 INP_HASH_RUNLOCK(&V_udbinfo);
1237                 INP_RUNLOCK(inp);
1238         } else
1239                 INP_RUNLOCK(inp);
1240         m_freem(m);
1241         return (error);
1242 }
1243
1244
1245 #if defined(IPSEC) && defined(IPSEC_NAT_T)
1246 /*
1247  * Potentially decap ESP in UDP frame.  Check for an ESP header
1248  * and optional marker; if present, strip the UDP header and
1249  * push the result through IPSec.
1250  *
1251  * Returns mbuf to be processed (potentially re-allocated) or
1252  * NULL if consumed and/or processed.
1253  */
1254 static struct mbuf *
1255 udp4_espdecap(struct inpcb *inp, struct mbuf *m, int off)
1256 {
1257         size_t minlen, payload, skip, iphlen;
1258         caddr_t data;
1259         struct udpcb *up;
1260         struct m_tag *tag;
1261         struct udphdr *udphdr;
1262         struct ip *ip;
1263
1264         INP_RLOCK_ASSERT(inp);
1265
1266         /*
1267          * Pull up data so the longest case is contiguous:
1268          *    IP/UDP hdr + non ESP marker + ESP hdr.
1269          */
1270         minlen = off + sizeof(uint64_t) + sizeof(struct esp);
1271         if (minlen > m->m_pkthdr.len)
1272                 minlen = m->m_pkthdr.len;
1273         if ((m = m_pullup(m, minlen)) == NULL) {
1274                 V_ipsec4stat.in_inval++;
1275                 return (NULL);          /* Bypass caller processing. */
1276         }
1277         data = mtod(m, caddr_t);        /* Points to ip header. */
1278         payload = m->m_len - off;       /* Size of payload. */
1279
1280         if (payload == 1 && data[off] == '\xff')
1281                 return (m);             /* NB: keepalive packet, no decap. */
1282
1283         up = intoudpcb(inp);
1284         KASSERT(up != NULL, ("%s: udpcb NULL", __func__));
1285         KASSERT((up->u_flags & UF_ESPINUDP_ALL) != 0,
1286             ("u_flags 0x%x", up->u_flags));
1287
1288         /*
1289          * Check that the payload is large enough to hold an
1290          * ESP header and compute the amount of data to remove.
1291          *
1292          * NB: the caller has already done a pullup for us.
1293          * XXX can we assume alignment and eliminate bcopys?
1294          */
1295         if (up->u_flags & UF_ESPINUDP_NON_IKE) {
1296                 /*
1297                  * draft-ietf-ipsec-nat-t-ike-0[01].txt and
1298                  * draft-ietf-ipsec-udp-encaps-(00/)01.txt, ignoring
1299                  * possible AH mode non-IKE marker+non-ESP marker
1300                  * from draft-ietf-ipsec-udp-encaps-00.txt.
1301                  */
1302                 uint64_t marker;
1303
1304                 if (payload <= sizeof(uint64_t) + sizeof(struct esp))
1305                         return (m);     /* NB: no decap. */
1306                 bcopy(data + off, &marker, sizeof(uint64_t));
1307                 if (marker != 0)        /* Non-IKE marker. */
1308                         return (m);     /* NB: no decap. */
1309                 skip = sizeof(uint64_t) + sizeof(struct udphdr);
1310         } else {
1311                 uint32_t spi;
1312
1313                 if (payload <= sizeof(struct esp)) {
1314                         V_ipsec4stat.in_inval++;
1315                         m_freem(m);
1316                         return (NULL);  /* Discard. */
1317                 }
1318                 bcopy(data + off, &spi, sizeof(uint32_t));
1319                 if (spi == 0)           /* Non-ESP marker. */
1320                         return (m);     /* NB: no decap. */
1321                 skip = sizeof(struct udphdr);
1322         }
1323
1324         /*
1325          * Setup a PACKET_TAG_IPSEC_NAT_T_PORT tag to remember
1326          * the UDP ports. This is required if we want to select
1327          * the right SPD for multiple hosts behind same NAT.
1328          *
1329          * NB: ports are maintained in network byte order everywhere
1330          *     in the NAT-T code.
1331          */
1332         tag = m_tag_get(PACKET_TAG_IPSEC_NAT_T_PORTS,
1333                 2 * sizeof(uint16_t), M_NOWAIT);
1334         if (tag == NULL) {
1335                 V_ipsec4stat.in_nomem++;
1336                 m_freem(m);
1337                 return (NULL);          /* Discard. */
1338         }
1339         iphlen = off - sizeof(struct udphdr);
1340         udphdr = (struct udphdr *)(data + iphlen);
1341         ((uint16_t *)(tag + 1))[0] = udphdr->uh_sport;
1342         ((uint16_t *)(tag + 1))[1] = udphdr->uh_dport;
1343         m_tag_prepend(m, tag);
1344
1345         /*
1346          * Remove the UDP header (and possibly the non ESP marker)
1347          * IP header length is iphlen
1348          * Before:
1349          *   <--- off --->
1350          *   +----+------+-----+
1351          *   | IP |  UDP | ESP |
1352          *   +----+------+-----+
1353          *        <-skip->
1354          * After:
1355          *          +----+-----+
1356          *          | IP | ESP |
1357          *          +----+-----+
1358          *   <-skip->
1359          */
1360         ovbcopy(data, data + skip, iphlen);
1361         m_adj(m, skip);
1362
1363         ip = mtod(m, struct ip *);
1364         ip->ip_len -= skip;
1365         ip->ip_p = IPPROTO_ESP;
1366
1367         /*
1368          * We cannot yet update the cksums so clear any
1369          * h/w cksum flags as they are no longer valid.
1370          */
1371         if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID)
1372                 m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID|CSUM_PSEUDO_HDR);
1373
1374         (void) ipsec4_common_input(m, iphlen, ip->ip_p);
1375         return (NULL);                  /* NB: consumed, bypass processing. */
1376 }
1377 #endif /* defined(IPSEC) && defined(IPSEC_NAT_T) */
1378
1379 static void
1380 udp_abort(struct socket *so)
1381 {
1382         struct inpcb *inp;
1383
1384         inp = sotoinpcb(so);
1385         KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
1386         INP_WLOCK(inp);
1387         if (inp->inp_faddr.s_addr != INADDR_ANY) {
1388                 INP_HASH_WLOCK(&V_udbinfo);
1389                 in_pcbdisconnect(inp);
1390                 inp->inp_laddr.s_addr = INADDR_ANY;
1391                 INP_HASH_WUNLOCK(&V_udbinfo);
1392                 soisdisconnected(so);
1393         }
1394         INP_WUNLOCK(inp);
1395 }
1396
1397 static int
1398 udp_attach(struct socket *so, int proto, struct thread *td)
1399 {
1400         struct inpcb *inp;
1401         int error;
1402
1403         inp = sotoinpcb(so);
1404         KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
1405         error = soreserve(so, udp_sendspace, udp_recvspace);
1406         if (error)
1407                 return (error);
1408         INP_INFO_WLOCK(&V_udbinfo);
1409         error = in_pcballoc(so, &V_udbinfo);
1410         if (error) {
1411                 INP_INFO_WUNLOCK(&V_udbinfo);
1412                 return (error);
1413         }
1414
1415         inp = sotoinpcb(so);
1416         inp->inp_vflag |= INP_IPV4;
1417         inp->inp_ip_ttl = V_ip_defttl;
1418
1419         error = udp_newudpcb(inp);
1420         if (error) {
1421                 in_pcbdetach(inp);
1422                 in_pcbfree(inp);
1423                 INP_INFO_WUNLOCK(&V_udbinfo);
1424                 return (error);
1425         }
1426
1427         INP_WUNLOCK(inp);
1428         INP_INFO_WUNLOCK(&V_udbinfo);
1429         return (0);
1430 }
1431 #endif /* INET */
1432
1433 int
1434 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f)
1435 {
1436         struct inpcb *inp;
1437         struct udpcb *up;
1438
1439         KASSERT(so->so_type == SOCK_DGRAM,
1440             ("udp_set_kernel_tunneling: !dgram"));
1441         inp = sotoinpcb(so);
1442         KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL"));
1443         INP_WLOCK(inp);
1444         up = intoudpcb(inp);
1445         if (up->u_tun_func != NULL) {
1446                 INP_WUNLOCK(inp);
1447                 return (EBUSY);
1448         }
1449         up->u_tun_func = f;
1450         INP_WUNLOCK(inp);
1451         return (0);
1452 }
1453
1454 #ifdef INET
1455 static int
1456 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1457 {
1458         struct inpcb *inp;
1459         int error;
1460
1461         inp = sotoinpcb(so);
1462         KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1463         INP_WLOCK(inp);
1464         INP_HASH_WLOCK(&V_udbinfo);
1465         error = in_pcbbind(inp, nam, td->td_ucred);
1466         INP_HASH_WUNLOCK(&V_udbinfo);
1467         INP_WUNLOCK(inp);
1468         return (error);
1469 }
1470
1471 static void
1472 udp_close(struct socket *so)
1473 {
1474         struct inpcb *inp;
1475
1476         inp = sotoinpcb(so);
1477         KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1478         INP_WLOCK(inp);
1479         if (inp->inp_faddr.s_addr != INADDR_ANY) {
1480                 INP_HASH_WLOCK(&V_udbinfo);
1481                 in_pcbdisconnect(inp);
1482                 inp->inp_laddr.s_addr = INADDR_ANY;
1483                 INP_HASH_WUNLOCK(&V_udbinfo);
1484                 soisdisconnected(so);
1485         }
1486         INP_WUNLOCK(inp);
1487 }
1488
1489 static int
1490 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1491 {
1492         struct inpcb *inp;
1493         int error;
1494         struct sockaddr_in *sin;
1495
1496         inp = sotoinpcb(so);
1497         KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1498         INP_WLOCK(inp);
1499         if (inp->inp_faddr.s_addr != INADDR_ANY) {
1500                 INP_WUNLOCK(inp);
1501                 return (EISCONN);
1502         }
1503         sin = (struct sockaddr_in *)nam;
1504         error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1505         if (error != 0) {
1506                 INP_WUNLOCK(inp);
1507                 return (error);
1508         }
1509         INP_HASH_WLOCK(&V_udbinfo);
1510         error = in_pcbconnect(inp, nam, td->td_ucred);
1511         INP_HASH_WUNLOCK(&V_udbinfo);
1512         if (error == 0)
1513                 soisconnected(so);
1514         INP_WUNLOCK(inp);
1515         return (error);
1516 }
1517
1518 static void
1519 udp_detach(struct socket *so)
1520 {
1521         struct inpcb *inp;
1522         struct udpcb *up;
1523
1524         inp = sotoinpcb(so);
1525         KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1526         KASSERT(inp->inp_faddr.s_addr == INADDR_ANY,
1527             ("udp_detach: not disconnected"));
1528         INP_INFO_WLOCK(&V_udbinfo);
1529         INP_WLOCK(inp);
1530         up = intoudpcb(inp);
1531         KASSERT(up != NULL, ("%s: up == NULL", __func__));
1532         inp->inp_ppcb = NULL;
1533         in_pcbdetach(inp);
1534         in_pcbfree(inp);
1535         INP_INFO_WUNLOCK(&V_udbinfo);
1536         udp_discardcb(up);
1537 }
1538
1539 static int
1540 udp_disconnect(struct socket *so)
1541 {
1542         struct inpcb *inp;
1543
1544         inp = sotoinpcb(so);
1545         KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1546         INP_WLOCK(inp);
1547         if (inp->inp_faddr.s_addr == INADDR_ANY) {
1548                 INP_WUNLOCK(inp);
1549                 return (ENOTCONN);
1550         }
1551         INP_HASH_WLOCK(&V_udbinfo);
1552         in_pcbdisconnect(inp);
1553         inp->inp_laddr.s_addr = INADDR_ANY;
1554         INP_HASH_WUNLOCK(&V_udbinfo);
1555         SOCK_LOCK(so);
1556         so->so_state &= ~SS_ISCONNECTED;                /* XXX */
1557         SOCK_UNLOCK(so);
1558         INP_WUNLOCK(inp);
1559         return (0);
1560 }
1561
1562 static int
1563 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1564     struct mbuf *control, struct thread *td)
1565 {
1566         struct inpcb *inp;
1567
1568         inp = sotoinpcb(so);
1569         KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1570         return (udp_output(inp, m, addr, control, td));
1571 }
1572 #endif /* INET */
1573
1574 int
1575 udp_shutdown(struct socket *so)
1576 {
1577         struct inpcb *inp;
1578
1579         inp = sotoinpcb(so);
1580         KASSERT(inp != NULL, ("udp_shutdown: inp == NULL"));
1581         INP_WLOCK(inp);
1582         socantsendmore(so);
1583         INP_WUNLOCK(inp);
1584         return (0);
1585 }
1586
1587 #ifdef INET
1588 struct pr_usrreqs udp_usrreqs = {
1589         .pru_abort =            udp_abort,
1590         .pru_attach =           udp_attach,
1591         .pru_bind =             udp_bind,
1592         .pru_connect =          udp_connect,
1593         .pru_control =          in_control,
1594         .pru_detach =           udp_detach,
1595         .pru_disconnect =       udp_disconnect,
1596         .pru_peeraddr =         in_getpeeraddr,
1597         .pru_send =             udp_send,
1598         .pru_soreceive =        soreceive_dgram,
1599         .pru_sosend =           sosend_dgram,
1600         .pru_shutdown =         udp_shutdown,
1601         .pru_sockaddr =         in_getsockaddr,
1602         .pru_sosetlabel =       in_pcbsosetlabel,
1603         .pru_close =            udp_close,
1604 };
1605 #endif /* INET */