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