sys/netinet/udp_usrreq.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-3-Clause
   3  *
   4  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
   5  *      The Regents of the University of California.
   6  * Copyright (c) 2008 Robert N. M. Watson
   7  * Copyright (c) 2010-2011 Juniper Networks, Inc.
   8  * Copyright (c) 2014 Kevin Lo
   9  * All rights reserved.
  10  *
  11  * Portions of this software were developed by Robert N. M. Watson under
  12  * contract to Juniper Networks, Inc.
  13  *
  14  * Redistribution and use in source and binary forms, with or without
  15  * modification, are permitted provided that the following conditions
  16  * are met:
  17  * 1. Redistributions of source code must retain the above copyright
  18  *    notice, this list of conditions and the following disclaimer.
  19  * 2. Redistributions in binary form must reproduce the above copyright
  20  *    notice, this list of conditions and the following disclaimer in the
  21  *    documentation and/or other materials provided with the distribution.
  22  * 3. Neither the name of the University nor the names of its contributors
  23  *    may be used to endorse or promote products derived from this software
  24  *    without specific prior written permission.
  25  *
  26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  36  * SUCH DAMAGE.
  37  *
  38  *      @(#)udp_usrreq.c        8.6 (Berkeley) 5/23/95
  39  */
  40
  41 #include <sys/cdefs.h>
  42 __FBSDID("$FreeBSD$");
  43
  44 #include "opt_inet.h"
  45 #include "opt_inet6.h"
  46 #include "opt_ipsec.h"
  47 #include "opt_rss.h"
  48
  49 #include <sys/param.h>
  50 #include <sys/domain.h>
  51 #include <sys/eventhandler.h>
  52 #include <sys/jail.h>
  53 #include <sys/kernel.h>
  54 #include <sys/lock.h>
  55 #include <sys/malloc.h>
  56 #include <sys/mbuf.h>
  57 #include <sys/priv.h>
  58 #include <sys/proc.h>
  59 #include <sys/protosw.h>
  60 #include <sys/sdt.h>
  61 #include <sys/signalvar.h>
  62 #include <sys/socket.h>
  63 #include <sys/socketvar.h>
  64 #include <sys/sx.h>
  65 #include <sys/sysctl.h>
  66 #include <sys/syslog.h>
  67 #include <sys/systm.h>
  68
  69 #include <vm/uma.h>
  70
  71 #include <net/if.h>
  72 #include <net/if_var.h>
  73 #include <net/route.h>
  74 #include <net/route/nhop.h>
  75 #include <net/rss_config.h>
  76
  77 #include <netinet/in.h>
  78 #include <netinet/in_kdtrace.h>
  79 #include <netinet/in_pcb.h>
  80 #include <netinet/in_systm.h>
  81 #include <netinet/in_var.h>
  82 #include <netinet/ip.h>
  83 #ifdef INET6
  84 #include <netinet/ip6.h>
  85 #endif
  86 #include <netinet/ip_icmp.h>
  87 #include <netinet/icmp_var.h>
  88 #include <netinet/ip_var.h>
  89 #include <netinet/ip_options.h>
  90 #ifdef INET6
  91 #include <netinet6/ip6_var.h>
  92 #endif
  93 #include <netinet/udp.h>
  94 #include <netinet/udp_var.h>
  95 #include <netinet/udplite.h>
  96 #include <netinet/in_rss.h>
  97
  98 #include <netipsec/ipsec_support.h>
  99
 100 #include <machine/in_cksum.h>
 101
 102 #include <security/mac/mac_framework.h>
 103
 104 /*
 105  * UDP and UDP-Lite protocols implementation.
 106  * Per RFC 768, August, 1980.
 107  * Per RFC 3828, July, 2004.
 108  */
 109
 110 /*
 111  * BSD 4.2 defaulted the udp checksum to be off.  Turning off udp checksums
 112  * removes the only data integrity mechanism for packets and malformed
 113  * packets that would otherwise be discarded due to bad checksums, and may
 114  * cause problems (especially for NFS data blocks).
 115  */
 116 VNET_DEFINE(int, udp_cksum) = 1;
 117 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_VNET | CTLFLAG_RW,
 118     &VNET_NAME(udp_cksum), 0, "compute udp checksum");
 119
 120 VNET_DEFINE(int, udp_log_in_vain) = 0;
 121 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_VNET | CTLFLAG_RW,
 122     &VNET_NAME(udp_log_in_vain), 0, "Log all incoming UDP packets");
 123
 124 VNET_DEFINE(int, udp_blackhole) = 0;
 125 SYSCTL_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW,
 126     &VNET_NAME(udp_blackhole), 0,
 127     "Do not send port unreachables for refused connects");
 128
 129 u_long  udp_sendspace = 9216;           /* really max datagram size */
 130 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
 131     &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
 132
 133 u_long  udp_recvspace = 40 * (1024 +
 134 #ifdef INET6
 135                                       sizeof(struct sockaddr_in6)
 136 #else
 137                                       sizeof(struct sockaddr_in)
 138 #endif
 139                                       );        /* 40 1K datagrams */
 140
 141 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
 142     &udp_recvspace, 0, "Maximum space for incoming UDP datagrams");
 143
 144 VNET_DEFINE(struct inpcbhead, udb);             /* from udp_var.h */
 145 VNET_DEFINE(struct inpcbinfo, udbinfo);
 146 VNET_DEFINE(struct inpcbhead, ulitecb);
 147 VNET_DEFINE(struct inpcbinfo, ulitecbinfo);
 148 VNET_DEFINE_STATIC(uma_zone_t, udpcb_zone);
 149 #define V_udpcb_zone                    VNET(udpcb_zone)
 150
 151 #ifndef UDBHASHSIZE
 152 #define UDBHASHSIZE     128
 153 #endif
 154
 155 VNET_PCPUSTAT_DEFINE(struct udpstat, udpstat);          /* from udp_var.h */
 156 VNET_PCPUSTAT_SYSINIT(udpstat);
 157 SYSCTL_VNET_PCPUSTAT(_net_inet_udp, UDPCTL_STATS, stats, struct udpstat,
 158     udpstat, "UDP statistics (struct udpstat, netinet/udp_var.h)");
 159
 160 #ifdef VIMAGE
 161 VNET_PCPUSTAT_SYSUNINIT(udpstat);
 162 #endif /* VIMAGE */
 163 #ifdef INET
 164 static void     udp_detach(struct socket *so);
 165 static int      udp_output(struct inpcb *, struct mbuf *, struct sockaddr *,
 166                     struct mbuf *, struct thread *);
 167 #endif
 168
 169 static void
 170 udp_zone_change(void *tag)
 171 {
 172
 173         uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets);
 174         uma_zone_set_max(V_udpcb_zone, maxsockets);
 175 }
 176
 177 static int
 178 udp_inpcb_init(void *mem, int size, int flags)
 179 {
 180         struct inpcb *inp;
 181
 182         inp = mem;
 183         INP_LOCK_INIT(inp, "inp", "udpinp");
 184         return (0);
 185 }
 186
 187 static int
 188 udplite_inpcb_init(void *mem, int size, int flags)
 189 {
 190         struct inpcb *inp;
 191
 192         inp = mem;
 193         INP_LOCK_INIT(inp, "inp", "udpliteinp");
 194         return (0);
 195 }
 196
 197 void
 198 udp_init(void)
 199 {
 200
 201         /*
 202          * For now default to 2-tuple UDP hashing - until the fragment
 203          * reassembly code can also update the flowid.
 204          *
 205          * Once we can calculate the flowid that way and re-establish
 206          * a 4-tuple, flip this to 4-tuple.
 207          */
 208         in_pcbinfo_init(&V_udbinfo, "udp", &V_udb, UDBHASHSIZE, UDBHASHSIZE,
 209             "udp_inpcb", udp_inpcb_init, IPI_HASHFIELDS_2TUPLE);
 210         V_udpcb_zone = uma_zcreate("udpcb", sizeof(struct udpcb),
 211             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
 212         uma_zone_set_max(V_udpcb_zone, maxsockets);
 213         uma_zone_set_warning(V_udpcb_zone, "kern.ipc.maxsockets limit reached");
 214         EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL,
 215             EVENTHANDLER_PRI_ANY);
 216 }
 217
 218 void
 219 udplite_init(void)
 220 {
 221
 222         in_pcbinfo_init(&V_ulitecbinfo, "udplite", &V_ulitecb, UDBHASHSIZE,
 223             UDBHASHSIZE, "udplite_inpcb", udplite_inpcb_init,
 224             IPI_HASHFIELDS_2TUPLE);
 225 }
 226
 227 /*
 228  * Kernel module interface for updating udpstat.  The argument is an index
 229  * into udpstat treated as an array of u_long.  While this encodes the
 230  * general layout of udpstat into the caller, it doesn't encode its location,
 231  * so that future changes to add, for example, per-CPU stats support won't
 232  * cause binary compatibility problems for kernel modules.
 233  */
 234 void
 235 kmod_udpstat_inc(int statnum)
 236 {
 237
 238         counter_u64_add(VNET(udpstat)[statnum], 1);
 239 }
 240
 241 int
 242 udp_newudpcb(struct inpcb *inp)
 243 {
 244         struct udpcb *up;
 245
 246         up = uma_zalloc(V_udpcb_zone, M_NOWAIT | M_ZERO);
 247         if (up == NULL)
 248                 return (ENOBUFS);
 249         inp->inp_ppcb = up;
 250         return (0);
 251 }
 252
 253 void
 254 udp_discardcb(struct udpcb *up)
 255 {
 256
 257         uma_zfree(V_udpcb_zone, up);
 258 }
 259
 260 #ifdef VIMAGE
 261 static void
 262 udp_destroy(void *unused __unused)
 263 {
 264
 265         in_pcbinfo_destroy(&V_udbinfo);
 266         uma_zdestroy(V_udpcb_zone);
 267 }
 268 VNET_SYSUNINIT(udp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udp_destroy, NULL);
 269
 270 static void
 271 udplite_destroy(void *unused __unused)
 272 {
 273
 274         in_pcbinfo_destroy(&V_ulitecbinfo);
 275 }
 276 VNET_SYSUNINIT(udplite, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udplite_destroy,
 277     NULL);
 278 #endif
 279
 280 #ifdef INET
 281 /*
 282  * Subroutine of udp_input(), which appends the provided mbuf chain to the
 283  * passed pcb/socket.  The caller must provide a sockaddr_in via udp_in that
 284  * contains the source address.  If the socket ends up being an IPv6 socket,
 285  * udp_append() will convert to a sockaddr_in6 before passing the address
 286  * into the socket code.
 287  *
 288  * In the normal case udp_append() will return 0, indicating that you
 289  * must unlock the inp. However if a tunneling protocol is in place we increment
 290  * the inpcb refcnt and unlock the inp, on return from the tunneling protocol we
 291  * then decrement the reference count. If the inp_rele returns 1, indicating the
 292  * inp is gone, we return that to the caller to tell them *not* to unlock
 293  * the inp. In the case of multi-cast this will cause the distribution
 294  * to stop (though most tunneling protocols known currently do *not* use
 295  * multicast).
 296  */
 297 static int
 298 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off,
 299     struct sockaddr_in *udp_in)
 300 {
 301         struct sockaddr *append_sa;
 302         struct socket *so;
 303         struct mbuf *tmpopts, *opts = NULL;
 304 #ifdef INET6
 305         struct sockaddr_in6 udp_in6;
 306 #endif
 307         struct udpcb *up;
 308
 309         INP_LOCK_ASSERT(inp);
 310
 311         /*
 312          * Engage the tunneling protocol.
 313          */
 314         up = intoudpcb(inp);
 315         if (up->u_tun_func != NULL) {
 316                 in_pcbref(inp);
 317                 INP_RUNLOCK(inp);
 318                 (*up->u_tun_func)(n, off, inp, (struct sockaddr *)&udp_in[0],
 319                     up->u_tun_ctx);
 320                 INP_RLOCK(inp);
 321                 return (in_pcbrele_rlocked(inp));
 322         }
 323
 324         off += sizeof(struct udphdr);
 325
 326 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
 327         /* Check AH/ESP integrity. */
 328         if (IPSEC_ENABLED(ipv4) &&
 329             IPSEC_CHECK_POLICY(ipv4, n, inp) != 0) {
 330                 m_freem(n);
 331                 return (0);
 332         }
 333         if (up->u_flags & UF_ESPINUDP) {/* IPSec UDP encaps. */
 334                 if (IPSEC_ENABLED(ipv4) &&
 335                     UDPENCAP_INPUT(n, off, AF_INET) != 0)
 336                         return (0);     /* Consumed. */
 337         }
 338 #endif /* IPSEC */
 339 #ifdef MAC
 340         if (mac_inpcb_check_deliver(inp, n) != 0) {
 341                 m_freem(n);
 342                 return (0);
 343         }
 344 #endif /* MAC */
 345         if (inp->inp_flags & INP_CONTROLOPTS ||
 346             inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
 347 #ifdef INET6
 348                 if (inp->inp_vflag & INP_IPV6)
 349                         (void)ip6_savecontrol_v4(inp, n, &opts, NULL);
 350                 else
 351 #endif /* INET6 */
 352                         ip_savecontrol(inp, &opts, ip, n);
 353         }
 354         if ((inp->inp_vflag & INP_IPV4) && (inp->inp_flags2 & INP_ORIGDSTADDR)) {
 355                 tmpopts = sbcreatecontrol((caddr_t)&udp_in[1],
 356                         sizeof(struct sockaddr_in), IP_ORIGDSTADDR, IPPROTO_IP);
 357                 if (tmpopts) {
 358                         if (opts) {
 359                                 tmpopts->m_next = opts;
 360                                 opts = tmpopts;
 361                         } else
 362                                 opts = tmpopts;
 363                 }
 364         }
 365 #ifdef INET6
 366         if (inp->inp_vflag & INP_IPV6) {
 367                 bzero(&udp_in6, sizeof(udp_in6));
 368                 udp_in6.sin6_len = sizeof(udp_in6);
 369                 udp_in6.sin6_family = AF_INET6;
 370                 in6_sin_2_v4mapsin6(&udp_in[0], &udp_in6);
 371                 append_sa = (struct sockaddr *)&udp_in6;
 372         } else
 373 #endif /* INET6 */
 374                 append_sa = (struct sockaddr *)&udp_in[0];
 375         m_adj(n, off);
 376
 377         so = inp->inp_socket;
 378         SOCKBUF_LOCK(&so->so_rcv);
 379         if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) {
 380                 SOCKBUF_UNLOCK(&so->so_rcv);
 381                 m_freem(n);
 382                 if (opts)
 383                         m_freem(opts);
 384                 UDPSTAT_INC(udps_fullsock);
 385         } else
 386                 sorwakeup_locked(so);
 387         return (0);
 388 }
 389
 390 int
 391 udp_input(struct mbuf **mp, int *offp, int proto)
 392 {
 393         struct ip *ip;
 394         struct udphdr *uh;
 395         struct ifnet *ifp;
 396         struct inpcb *inp;
 397         uint16_t len, ip_len;
 398         struct inpcbinfo *pcbinfo;
 399         struct ip save_ip;
 400         struct sockaddr_in udp_in[2];
 401         struct mbuf *m;
 402         struct m_tag *fwd_tag;
 403         int cscov_partial, iphlen;
 404
 405         m = *mp;
 406         iphlen = *offp;
 407         ifp = m->m_pkthdr.rcvif;
 408         *mp = NULL;
 409         UDPSTAT_INC(udps_ipackets);
 410
 411         /*
 412          * Strip IP options, if any; should skip this, make available to
 413          * user, and use on returned packets, but we don't yet have a way to
 414          * check the checksum with options still present.
 415          */
 416         if (iphlen > sizeof (struct ip)) {
 417                 ip_stripoptions(m);
 418                 iphlen = sizeof(struct ip);
 419         }
 420
 421         /*
 422          * Get IP and UDP header together in first mbuf.
 423          */
 424         if (m->m_len < iphlen + sizeof(struct udphdr)) {
 425                 if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == NULL) {
 426                         UDPSTAT_INC(udps_hdrops);
 427                         return (IPPROTO_DONE);
 428                 }
 429         }
 430         ip = mtod(m, struct ip *);
 431         uh = (struct udphdr *)((caddr_t)ip + iphlen);
 432         cscov_partial = (proto == IPPROTO_UDPLITE) ? 1 : 0;
 433
 434         /*
 435          * Destination port of 0 is illegal, based on RFC768.
 436          */
 437         if (uh->uh_dport == 0)
 438                 goto badunlocked;
 439
 440         /*
 441          * Construct sockaddr format source address.  Stuff source address
 442          * and datagram in user buffer.
 443          */
 444         bzero(&udp_in[0], sizeof(struct sockaddr_in) * 2);
 445         udp_in[0].sin_len = sizeof(struct sockaddr_in);
 446         udp_in[0].sin_family = AF_INET;
 447         udp_in[0].sin_port = uh->uh_sport;
 448         udp_in[0].sin_addr = ip->ip_src;
 449         udp_in[1].sin_len = sizeof(struct sockaddr_in);
 450         udp_in[1].sin_family = AF_INET;
 451         udp_in[1].sin_port = uh->uh_dport;
 452         udp_in[1].sin_addr = ip->ip_dst;
 453
 454         /*
 455          * Make mbuf data length reflect UDP length.  If not enough data to
 456          * reflect UDP length, drop.
 457          */
 458         len = ntohs((u_short)uh->uh_ulen);
 459         ip_len = ntohs(ip->ip_len) - iphlen;
 460         if (proto == IPPROTO_UDPLITE && (len == 0 || len == ip_len)) {
 461                 /* Zero means checksum over the complete packet. */
 462                 if (len == 0)
 463                         len = ip_len;
 464                 cscov_partial = 0;
 465         }
 466         if (ip_len != len) {
 467                 if (len > ip_len || len < sizeof(struct udphdr)) {
 468                         UDPSTAT_INC(udps_badlen);
 469                         goto badunlocked;
 470                 }
 471                 if (proto == IPPROTO_UDP)
 472                         m_adj(m, len - ip_len);
 473         }
 474
 475         /*
 476          * Save a copy of the IP header in case we want restore it for
 477          * sending an ICMP error message in response.
 478          */
 479         if (!V_udp_blackhole)
 480                 save_ip = *ip;
 481         else
 482                 memset(&save_ip, 0, sizeof(save_ip));
 483
 484         /*
 485          * Checksum extended UDP header and data.
 486          */
 487         if (uh->uh_sum) {
 488                 u_short uh_sum;
 489
 490                 if ((m->m_pkthdr.csum_flags & CSUM_DATA_VALID) &&
 491                     !cscov_partial) {
 492                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
 493                                 uh_sum = m->m_pkthdr.csum_data;
 494                         else
 495                                 uh_sum = in_pseudo(ip->ip_src.s_addr,
 496                                     ip->ip_dst.s_addr, htonl((u_short)len +
 497                                     m->m_pkthdr.csum_data + proto));
 498                         uh_sum ^= 0xffff;
 499                 } else {
 500                         char b[9];
 501
 502                         bcopy(((struct ipovly *)ip)->ih_x1, b, 9);
 503                         bzero(((struct ipovly *)ip)->ih_x1, 9);
 504                         ((struct ipovly *)ip)->ih_len = (proto == IPPROTO_UDP) ?
 505                             uh->uh_ulen : htons(ip_len);
 506                         uh_sum = in_cksum(m, len + sizeof (struct ip));
 507                         bcopy(b, ((struct ipovly *)ip)->ih_x1, 9);
 508                 }
 509                 if (uh_sum) {
 510                         UDPSTAT_INC(udps_badsum);
 511                         m_freem(m);
 512                         return (IPPROTO_DONE);
 513                 }
 514         } else {
 515                 if (proto == IPPROTO_UDP) {
 516                         UDPSTAT_INC(udps_nosum);
 517                 } else {
 518                         /* UDPLite requires a checksum */
 519                         /* XXX: What is the right UDPLite MIB counter here? */
 520                         m_freem(m);
 521                         return (IPPROTO_DONE);
 522                 }
 523         }
 524
 525         pcbinfo = udp_get_inpcbinfo(proto);
 526         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
 527             in_broadcast(ip->ip_dst, ifp)) {
 528                 struct inpcb *last;
 529                 struct inpcbhead *pcblist;
 530
 531                 NET_EPOCH_ASSERT();
 532
 533                 pcblist = udp_get_pcblist(proto);
 534                 last = NULL;
 535                 CK_LIST_FOREACH(inp, pcblist, inp_list) {
 536                         if (inp->inp_lport != uh->uh_dport)
 537                                 continue;
 538 #ifdef INET6
 539                         if ((inp->inp_vflag & INP_IPV4) == 0)
 540                                 continue;
 541 #endif
 542                         if (inp->inp_laddr.s_addr != INADDR_ANY &&
 543                             inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
 544                                 continue;
 545                         if (inp->inp_faddr.s_addr != INADDR_ANY &&
 546                             inp->inp_faddr.s_addr != ip->ip_src.s_addr)
 547                                 continue;
 548                         if (inp->inp_fport != 0 &&
 549                             inp->inp_fport != uh->uh_sport)
 550                                 continue;
 551
 552                         INP_RLOCK(inp);
 553
 554                         if (__predict_false(inp->inp_flags2 & INP_FREED)) {
 555                                 INP_RUNLOCK(inp);
 556                                 continue;
 557                         }
 558
 559                         /*
 560                          * XXXRW: Because we weren't holding either the inpcb
 561                          * or the hash lock when we checked for a match
 562                          * before, we should probably recheck now that the
 563                          * inpcb lock is held.
 564                          */
 565
 566                         /*
 567                          * Handle socket delivery policy for any-source
 568                          * and source-specific multicast. [RFC3678]
 569                          */
 570                         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
 571                                 struct ip_moptions      *imo;
 572                                 struct sockaddr_in       group;
 573                                 int                      blocked;
 574
 575                                 imo = inp->inp_moptions;
 576                                 if (imo == NULL) {
 577                                         INP_RUNLOCK(inp);
 578                                         continue;
 579                                 }
 580                                 bzero(&group, sizeof(struct sockaddr_in));
 581                                 group.sin_len = sizeof(struct sockaddr_in);
 582                                 group.sin_family = AF_INET;
 583                                 group.sin_addr = ip->ip_dst;
 584
 585                                 blocked = imo_multi_filter(imo, ifp,
 586                                         (struct sockaddr *)&group,
 587                                         (struct sockaddr *)&udp_in[0]);
 588                                 if (blocked != MCAST_PASS) {
 589                                         if (blocked == MCAST_NOTGMEMBER)
 590                                                 IPSTAT_INC(ips_notmember);
 591                                         if (blocked == MCAST_NOTSMEMBER ||
 592                                             blocked == MCAST_MUTED)
 593                                                 UDPSTAT_INC(udps_filtermcast);
 594                                         INP_RUNLOCK(inp);
 595                                         continue;
 596                                 }
 597                         }
 598                         if (last != NULL) {
 599                                 struct mbuf *n;
 600
 601                                 if ((n = m_copym(m, 0, M_COPYALL, M_NOWAIT)) !=
 602                                     NULL) {
 603                                         if (proto == IPPROTO_UDPLITE)
 604                                                 UDPLITE_PROBE(receive, NULL, last, ip,
 605                                                     last, uh);
 606                                         else
 607                                                 UDP_PROBE(receive, NULL, last, ip, last,
 608                                                     uh);
 609                                         if (udp_append(last, ip, n, iphlen,
 610                                                 udp_in)) {
 611                                                 goto inp_lost;
 612                                         }
 613                                 }
 614                                 INP_RUNLOCK(last);
 615                         }
 616                         last = inp;
 617                         /*
 618                          * Don't look for additional matches if this one does
 619                          * not have either the SO_REUSEPORT or SO_REUSEADDR
 620                          * socket options set.  This heuristic avoids
 621                          * searching through all pcbs in the common case of a
 622                          * non-shared port.  It assumes that an application
 623                          * will never clear these options after setting them.
 624                          */
 625                         if ((last->inp_socket->so_options &
 626                             (SO_REUSEPORT|SO_REUSEPORT_LB|SO_REUSEADDR)) == 0)
 627                                 break;
 628                 }
 629
 630                 if (last == NULL) {
 631                         /*
 632                          * No matching pcb found; discard datagram.  (No need
 633                          * to send an ICMP Port Unreachable for a broadcast
 634                          * or multicast datgram.)
 635                          */
 636                         UDPSTAT_INC(udps_noportbcast);
 637                         if (inp)
 638                                 INP_RUNLOCK(inp);
 639                         goto badunlocked;
 640                 }
 641                 if (proto == IPPROTO_UDPLITE)
 642                         UDPLITE_PROBE(receive, NULL, last, ip, last, uh);
 643                 else
 644                         UDP_PROBE(receive, NULL, last, ip, last, uh);
 645                 if (udp_append(last, ip, m, iphlen, udp_in) == 0)
 646                         INP_RUNLOCK(last);
 647         inp_lost:
 648                 return (IPPROTO_DONE);
 649         }
 650
 651         /*
 652          * Locate pcb for datagram.
 653          */
 654
 655         /*
 656          * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
 657          */
 658         if ((m->m_flags & M_IP_NEXTHOP) &&
 659             (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) {
 660                 struct sockaddr_in *next_hop;
 661
 662                 next_hop = (struct sockaddr_in *)(fwd_tag + 1);
 663
 664                 /*
 665                  * Transparently forwarded. Pretend to be the destination.
 666                  * Already got one like this?
 667                  */
 668                 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport,
 669                     ip->ip_dst, uh->uh_dport, INPLOOKUP_RLOCKPCB, ifp, m);
 670                 if (!inp) {
 671                         /*
 672                          * It's new.  Try to find the ambushing socket.
 673                          * Because we've rewritten the destination address,
 674                          * any hardware-generated hash is ignored.
 675                          */
 676                         inp = in_pcblookup(pcbinfo, ip->ip_src,
 677                             uh->uh_sport, next_hop->sin_addr,
 678                             next_hop->sin_port ? htons(next_hop->sin_port) :
 679                             uh->uh_dport, INPLOOKUP_WILDCARD |
 680                             INPLOOKUP_RLOCKPCB, ifp);
 681                 }
 682                 /* Remove the tag from the packet. We don't need it anymore. */
 683                 m_tag_delete(m, fwd_tag);
 684                 m->m_flags &= ~M_IP_NEXTHOP;
 685         } else
 686                 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport,
 687                     ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD |
 688                     INPLOOKUP_RLOCKPCB, ifp, m);
 689         if (inp == NULL) {
 690                 if (V_udp_log_in_vain) {
 691                         char src[INET_ADDRSTRLEN];
 692                         char dst[INET_ADDRSTRLEN];
 693
 694                         log(LOG_INFO,
 695                             "Connection attempt to UDP %s:%d from %s:%d\n",
 696                             inet_ntoa_r(ip->ip_dst, dst), ntohs(uh->uh_dport),
 697                             inet_ntoa_r(ip->ip_src, src), ntohs(uh->uh_sport));
 698                 }
 699                 if (proto == IPPROTO_UDPLITE)
 700                         UDPLITE_PROBE(receive, NULL, NULL, ip, NULL, uh);
 701                 else
 702                         UDP_PROBE(receive, NULL, NULL, ip, NULL, uh);
 703                 UDPSTAT_INC(udps_noport);
 704                 if (m->m_flags & (M_BCAST | M_MCAST)) {
 705                         UDPSTAT_INC(udps_noportbcast);
 706                         goto badunlocked;
 707                 }
 708                 if (V_udp_blackhole)
 709                         goto badunlocked;
 710                 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
 711                         goto badunlocked;
 712                 *ip = save_ip;
 713                 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
 714                 return (IPPROTO_DONE);
 715         }
 716
 717         /*
 718          * Check the minimum TTL for socket.
 719          */
 720         INP_RLOCK_ASSERT(inp);
 721         if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) {
 722                 if (proto == IPPROTO_UDPLITE)
 723                         UDPLITE_PROBE(receive, NULL, inp, ip, inp, uh);
 724                 else
 725                         UDP_PROBE(receive, NULL, inp, ip, inp, uh);
 726                 INP_RUNLOCK(inp);
 727                 m_freem(m);
 728                 return (IPPROTO_DONE);
 729         }
 730         if (cscov_partial) {
 731                 struct udpcb *up;
 732
 733                 up = intoudpcb(inp);
 734                 if (up->u_rxcslen == 0 || up->u_rxcslen > len) {
 735                         INP_RUNLOCK(inp);
 736                         m_freem(m);
 737                         return (IPPROTO_DONE);
 738                 }
 739         }
 740
 741         if (proto == IPPROTO_UDPLITE)
 742                 UDPLITE_PROBE(receive, NULL, inp, ip, inp, uh);
 743         else
 744                 UDP_PROBE(receive, NULL, inp, ip, inp, uh);
 745         if (udp_append(inp, ip, m, iphlen, udp_in) == 0)
 746                 INP_RUNLOCK(inp);
 747         return (IPPROTO_DONE);
 748
 749 badunlocked:
 750         m_freem(m);
 751         return (IPPROTO_DONE);
 752 }
 753 #endif /* INET */
 754
 755 /*
 756  * Notify a udp user of an asynchronous error; just wake up so that they can
 757  * collect error status.
 758  */
 759 struct inpcb *
 760 udp_notify(struct inpcb *inp, int errno)
 761 {
 762
 763         INP_WLOCK_ASSERT(inp);
 764         if ((errno == EHOSTUNREACH || errno == ENETUNREACH ||
 765              errno == EHOSTDOWN) && inp->inp_route.ro_nh) {
 766                 NH_FREE(inp->inp_route.ro_nh);
 767                 inp->inp_route.ro_nh = (struct nhop_object *)NULL;
 768         }
 769
 770         inp->inp_socket->so_error = errno;
 771         sorwakeup(inp->inp_socket);
 772         sowwakeup(inp->inp_socket);
 773         return (inp);
 774 }
 775
 776 #ifdef INET
 777 static void
 778 udp_common_ctlinput(int cmd, struct sockaddr *sa, void *vip,
 779     struct inpcbinfo *pcbinfo)
 780 {
 781         struct ip *ip = vip;
 782         struct udphdr *uh;
 783         struct in_addr faddr;
 784         struct inpcb *inp;
 785
 786         faddr = ((struct sockaddr_in *)sa)->sin_addr;
 787         if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
 788                 return;
 789
 790         if (PRC_IS_REDIRECT(cmd)) {
 791                 /* signal EHOSTDOWN, as it flushes the cached route */
 792                 in_pcbnotifyall(&V_udbinfo, faddr, EHOSTDOWN, udp_notify);
 793                 return;
 794         }
 795
 796         /*
 797          * Hostdead is ugly because it goes linearly through all PCBs.
 798          *
 799          * XXX: We never get this from ICMP, otherwise it makes an excellent
 800          * DoS attack on machines with many connections.
 801          */
 802         if (cmd == PRC_HOSTDEAD)
 803                 ip = NULL;
 804         else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
 805                 return;
 806         if (ip != NULL) {
 807                 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
 808                 inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport,
 809                     ip->ip_src, uh->uh_sport, INPLOOKUP_WLOCKPCB, NULL);
 810                 if (inp != NULL) {
 811                         INP_WLOCK_ASSERT(inp);
 812                         if (inp->inp_socket != NULL) {
 813                                 udp_notify(inp, inetctlerrmap[cmd]);
 814                         }
 815                         INP_WUNLOCK(inp);
 816                 } else {
 817                         inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport,
 818                                            ip->ip_src, uh->uh_sport,
 819                                            INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
 820                         if (inp != NULL) {
 821                                 struct udpcb *up;
 822                                 void *ctx;
 823                                 udp_tun_icmp_t func;
 824
 825                                 up = intoudpcb(inp);
 826                                 ctx = up->u_tun_ctx;
 827                                 func = up->u_icmp_func;
 828                                 INP_RUNLOCK(inp);
 829                                 if (func != NULL)
 830                                         (*func)(cmd, sa, vip, ctx);
 831                         }
 832                 }
 833         } else
 834                 in_pcbnotifyall(pcbinfo, faddr, inetctlerrmap[cmd],
 835                     udp_notify);
 836 }
 837 void
 838 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
 839 {
 840
 841         return (udp_common_ctlinput(cmd, sa, vip, &V_udbinfo));
 842 }
 843
 844 void
 845 udplite_ctlinput(int cmd, struct sockaddr *sa, void *vip)
 846 {
 847
 848         return (udp_common_ctlinput(cmd, sa, vip, &V_ulitecbinfo));
 849 }
 850 #endif /* INET */
 851
 852 static int
 853 udp_pcblist(SYSCTL_HANDLER_ARGS)
 854 {
 855         struct xinpgen xig;
 856         struct epoch_tracker et;
 857         struct inpcb *inp;
 858         int error;
 859
 860         if (req->newptr != 0)
 861                 return (EPERM);
 862
 863         if (req->oldptr == 0) {
 864                 int n;
 865
 866                 n = V_udbinfo.ipi_count;
 867                 n += imax(n / 8, 10);
 868                 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb);
 869                 return (0);
 870         }
 871
 872         if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
 873                 return (error);
 874
 875         bzero(&xig, sizeof(xig));
 876         xig.xig_len = sizeof xig;
 877         xig.xig_count = V_udbinfo.ipi_count;
 878         xig.xig_gen = V_udbinfo.ipi_gencnt;
 879         xig.xig_sogen = so_gencnt;
 880         error = SYSCTL_OUT(req, &xig, sizeof xig);
 881         if (error)
 882                 return (error);
 883
 884         NET_EPOCH_ENTER(et);
 885         for (inp = CK_LIST_FIRST(V_udbinfo.ipi_listhead);
 886             inp != NULL;
 887             inp = CK_LIST_NEXT(inp, inp_list)) {
 888                 INP_RLOCK(inp);
 889                 if (inp->inp_gencnt <= xig.xig_gen &&
 890                     cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
 891                         struct xinpcb xi;
 892
 893                         in_pcbtoxinpcb(inp, &xi);
 894                         INP_RUNLOCK(inp);
 895                         error = SYSCTL_OUT(req, &xi, sizeof xi);
 896                         if (error)
 897                                 break;
 898                 } else
 899                         INP_RUNLOCK(inp);
 900         }
 901         NET_EPOCH_EXIT(et);
 902
 903         if (!error) {
 904                 /*
 905                  * Give the user an updated idea of our state.  If the
 906                  * generation differs from what we told her before, she knows
 907                  * that something happened while we were processing this
 908                  * request, and it might be necessary to retry.
 909                  */
 910                 xig.xig_gen = V_udbinfo.ipi_gencnt;
 911                 xig.xig_sogen = so_gencnt;
 912                 xig.xig_count = V_udbinfo.ipi_count;
 913                 error = SYSCTL_OUT(req, &xig, sizeof xig);
 914         }
 915
 916         return (error);
 917 }
 918
 919 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist,
 920     CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
 921     udp_pcblist, "S,xinpcb",
 922     "List of active UDP sockets");
 923
 924 #ifdef INET
 925 static int
 926 udp_getcred(SYSCTL_HANDLER_ARGS)
 927 {
 928         struct xucred xuc;
 929         struct sockaddr_in addrs[2];
 930         struct epoch_tracker et;
 931         struct inpcb *inp;
 932         int error;
 933
 934         error = priv_check(req->td, PRIV_NETINET_GETCRED);
 935         if (error)
 936                 return (error);
 937         error = SYSCTL_IN(req, addrs, sizeof(addrs));
 938         if (error)
 939                 return (error);
 940         NET_EPOCH_ENTER(et);
 941         inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
 942             addrs[0].sin_addr, addrs[0].sin_port,
 943             INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL);
 944         NET_EPOCH_EXIT(et);
 945         if (inp != NULL) {
 946                 INP_RLOCK_ASSERT(inp);
 947                 if (inp->inp_socket == NULL)
 948                         error = ENOENT;
 949                 if (error == 0)
 950                         error = cr_canseeinpcb(req->td->td_ucred, inp);
 951                 if (error == 0)
 952                         cru2x(inp->inp_cred, &xuc);
 953                 INP_RUNLOCK(inp);
 954         } else
 955                 error = ENOENT;
 956         if (error == 0)
 957                 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
 958         return (error);
 959 }
 960
 961 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
 962     CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_MPSAFE,
 963     0, 0, udp_getcred, "S,xucred",
 964     "Get the xucred of a UDP connection");
 965 #endif /* INET */
 966
 967 int
 968 udp_ctloutput(struct socket *so, struct sockopt *sopt)
 969 {
 970         struct inpcb *inp;
 971         struct udpcb *up;
 972         int isudplite, error, optval;
 973
 974         error = 0;
 975         isudplite = (so->so_proto->pr_protocol == IPPROTO_UDPLITE) ? 1 : 0;
 976         inp = sotoinpcb(so);
 977         KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
 978         INP_WLOCK(inp);
 979         if (sopt->sopt_level != so->so_proto->pr_protocol) {
 980 #ifdef INET6
 981                 if (INP_CHECK_SOCKAF(so, AF_INET6)) {
 982                         INP_WUNLOCK(inp);
 983                         error = ip6_ctloutput(so, sopt);
 984                 }
 985 #endif
 986 #if defined(INET) && defined(INET6)
 987                 else
 988 #endif
 989 #ifdef INET
 990                 {
 991                         INP_WUNLOCK(inp);
 992                         error = ip_ctloutput(so, sopt);
 993                 }
 994 #endif
 995                 return (error);
 996         }
 997
 998         switch (sopt->sopt_dir) {
 999         case SOPT_SET:
1000                 switch (sopt->sopt_name) {
1001 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1002 #ifdef INET
1003                 case UDP_ENCAP:
1004                         if (!IPSEC_ENABLED(ipv4)) {
1005                                 INP_WUNLOCK(inp);
1006                                 return (ENOPROTOOPT);
1007                         }
1008                         error = UDPENCAP_PCBCTL(inp, sopt);
1009                         break;
1010 #endif /* INET */
1011 #endif /* IPSEC */
1012                 case UDPLITE_SEND_CSCOV:
1013                 case UDPLITE_RECV_CSCOV:
1014                         if (!isudplite) {
1015                                 INP_WUNLOCK(inp);
1016                                 error = ENOPROTOOPT;
1017                                 break;
1018                         }
1019                         INP_WUNLOCK(inp);
1020                         error = sooptcopyin(sopt, &optval, sizeof(optval),
1021                             sizeof(optval));
1022                         if (error != 0)
1023                                 break;
1024                         inp = sotoinpcb(so);
1025                         KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
1026                         INP_WLOCK(inp);
1027                         up = intoudpcb(inp);
1028                         KASSERT(up != NULL, ("%s: up == NULL", __func__));
1029                         if ((optval != 0 && optval < 8) || (optval > 65535)) {
1030                                 INP_WUNLOCK(inp);
1031                                 error = EINVAL;
1032                                 break;
1033                         }
1034                         if (sopt->sopt_name == UDPLITE_SEND_CSCOV)
1035                                 up->u_txcslen = optval;
1036                         else
1037                                 up->u_rxcslen = optval;
1038                         INP_WUNLOCK(inp);
1039                         break;
1040                 default:
1041                         INP_WUNLOCK(inp);
1042                         error = ENOPROTOOPT;
1043                         break;
1044                 }
1045                 break;
1046         case SOPT_GET:
1047                 switch (sopt->sopt_name) {
1048 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1049 #ifdef INET
1050                 case UDP_ENCAP:
1051                         if (!IPSEC_ENABLED(ipv4)) {
1052                                 INP_WUNLOCK(inp);
1053                                 return (ENOPROTOOPT);
1054                         }
1055                         error = UDPENCAP_PCBCTL(inp, sopt);
1056                         break;
1057 #endif /* INET */
1058 #endif /* IPSEC */
1059                 case UDPLITE_SEND_CSCOV:
1060                 case UDPLITE_RECV_CSCOV:
1061                         if (!isudplite) {
1062                                 INP_WUNLOCK(inp);
1063                                 error = ENOPROTOOPT;
1064                                 break;
1065                         }
1066                         up = intoudpcb(inp);
1067                         KASSERT(up != NULL, ("%s: up == NULL", __func__));
1068                         if (sopt->sopt_name == UDPLITE_SEND_CSCOV)
1069                                 optval = up->u_txcslen;
1070                         else
1071                                 optval = up->u_rxcslen;
1072                         INP_WUNLOCK(inp);
1073                         error = sooptcopyout(sopt, &optval, sizeof(optval));
1074                         break;
1075                 default:
1076                         INP_WUNLOCK(inp);
1077                         error = ENOPROTOOPT;
1078                         break;
1079                 }
1080                 break;
1081         }
1082         return (error);
1083 }
1084
1085 #ifdef INET
1086 static int
1087 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
1088     struct mbuf *control, struct thread *td)
1089 {
1090         struct udpiphdr *ui;
1091         int len = m->m_pkthdr.len;
1092         struct in_addr faddr, laddr;
1093         struct cmsghdr *cm;
1094         struct inpcbinfo *pcbinfo;
1095         struct sockaddr_in *sin, src;
1096         struct epoch_tracker et;
1097         int cscov_partial = 0;
1098         int error = 0;
1099         int ipflags;
1100         u_short fport, lport;
1101         u_char tos;
1102         uint8_t pr;
1103         uint16_t cscov = 0;
1104         uint32_t flowid = 0;
1105         uint8_t flowtype = M_HASHTYPE_NONE;
1106
1107         if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
1108                 if (control)
1109                         m_freem(control);
1110                 m_freem(m);
1111                 return (EMSGSIZE);
1112         }
1113
1114         src.sin_family = 0;
1115         sin = (struct sockaddr_in *)addr;
1116
1117         /*
1118          * udp_output() may need to temporarily bind or connect the current
1119          * inpcb.  As such, we don't know up front whether we will need the
1120          * pcbinfo lock or not.  Do any work to decide what is needed up
1121          * front before acquiring any locks.
1122          *
1123          * We will need network epoch in either case, to safely lookup into
1124          * pcb hash.
1125          */
1126         if (sin == NULL ||
1127             (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0))
1128                 INP_WLOCK(inp);
1129         else
1130                 INP_RLOCK(inp);
1131         NET_EPOCH_ENTER(et);
1132         tos = inp->inp_ip_tos;
1133         if (control != NULL) {
1134                 /*
1135                  * XXX: Currently, we assume all the optional information is
1136                  * stored in a single mbuf.
1137                  */
1138                 if (control->m_next) {
1139                         m_freem(control);
1140                         error = EINVAL;
1141                         goto release;
1142                 }
1143                 for (; control->m_len > 0;
1144                     control->m_data += CMSG_ALIGN(cm->cmsg_len),
1145                     control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
1146                         cm = mtod(control, struct cmsghdr *);
1147                         if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0
1148                             || cm->cmsg_len > control->m_len) {
1149                                 error = EINVAL;
1150                                 break;
1151                         }
1152                         if (cm->cmsg_level != IPPROTO_IP)
1153                                 continue;
1154
1155                         switch (cm->cmsg_type) {
1156                         case IP_SENDSRCADDR:
1157                                 if (cm->cmsg_len !=
1158                                     CMSG_LEN(sizeof(struct in_addr))) {
1159                                         error = EINVAL;
1160                                         break;
1161                                 }
1162                                 bzero(&src, sizeof(src));
1163                                 src.sin_family = AF_INET;
1164                                 src.sin_len = sizeof(src);
1165                                 src.sin_port = inp->inp_lport;
1166                                 src.sin_addr =
1167                                     *(struct in_addr *)CMSG_DATA(cm);
1168                                 break;
1169
1170                         case IP_TOS:
1171                                 if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) {
1172                                         error = EINVAL;
1173                                         break;
1174                                 }
1175                                 tos = *(u_char *)CMSG_DATA(cm);
1176                                 break;
1177
1178                         case IP_FLOWID:
1179                                 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1180                                         error = EINVAL;
1181                                         break;
1182                                 }
1183                                 flowid = *(uint32_t *) CMSG_DATA(cm);
1184                                 break;
1185
1186                         case IP_FLOWTYPE:
1187                                 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1188                                         error = EINVAL;
1189                                         break;
1190                                 }
1191                                 flowtype = *(uint32_t *) CMSG_DATA(cm);
1192                                 break;
1193
1194 #ifdef  RSS
1195                         case IP_RSSBUCKETID:
1196                                 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) {
1197                                         error = EINVAL;
1198                                         break;
1199                                 }
1200                                 /* This is just a placeholder for now */
1201                                 break;
1202 #endif  /* RSS */
1203                         default:
1204                                 error = ENOPROTOOPT;
1205                                 break;
1206                         }
1207                         if (error)
1208                                 break;
1209                 }
1210                 m_freem(control);
1211         }
1212         if (error)
1213                 goto release;
1214
1215         pr = inp->inp_socket->so_proto->pr_protocol;
1216         pcbinfo = udp_get_inpcbinfo(pr);
1217
1218         /*
1219          * If the IP_SENDSRCADDR control message was specified, override the
1220          * source address for this datagram.  Its use is invalidated if the
1221          * address thus specified is incomplete or clobbers other inpcbs.
1222          */
1223         laddr = inp->inp_laddr;
1224         lport = inp->inp_lport;
1225         if (src.sin_family == AF_INET) {
1226                 INP_HASH_LOCK_ASSERT(pcbinfo);
1227                 if ((lport == 0) ||
1228                     (laddr.s_addr == INADDR_ANY &&
1229                      src.sin_addr.s_addr == INADDR_ANY)) {
1230                         error = EINVAL;
1231                         goto release;
1232                 }
1233                 error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
1234                     &laddr.s_addr, &lport, td->td_ucred);
1235                 if (error)
1236                         goto release;
1237         }
1238
1239         /*
1240          * If a UDP socket has been connected, then a local address/port will
1241          * have been selected and bound.
1242          *
1243          * If a UDP socket has not been connected to, then an explicit
1244          * destination address must be used, in which case a local
1245          * address/port may not have been selected and bound.
1246          */
1247         if (sin != NULL) {
1248                 INP_LOCK_ASSERT(inp);
1249                 if (inp->inp_faddr.s_addr != INADDR_ANY) {
1250                         error = EISCONN;
1251                         goto release;
1252                 }
1253
1254                 /*
1255                  * Jail may rewrite the destination address, so let it do
1256                  * that before we use it.
1257                  */
1258                 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1259                 if (error)
1260                         goto release;
1261
1262                 /*
1263                  * If a local address or port hasn't yet been selected, or if
1264                  * the destination address needs to be rewritten due to using
1265                  * a special INADDR_ constant, invoke in_pcbconnect_setup()
1266                  * to do the heavy lifting.  Once a port is selected, we
1267                  * commit the binding back to the socket; we also commit the
1268                  * binding of the address if in jail.
1269                  *
1270                  * If we already have a valid binding and we're not
1271                  * requesting a destination address rewrite, use a fast path.
1272                  */
1273                 if (inp->inp_laddr.s_addr == INADDR_ANY ||
1274                     inp->inp_lport == 0 ||
1275                     sin->sin_addr.s_addr == INADDR_ANY ||
1276                     sin->sin_addr.s_addr == INADDR_BROADCAST) {
1277                         INP_HASH_LOCK_ASSERT(pcbinfo);
1278                         error = in_pcbconnect_setup(inp, addr, &laddr.s_addr,
1279                             &lport, &faddr.s_addr, &fport, NULL,
1280                             td->td_ucred);
1281                         if (error)
1282                                 goto release;
1283
1284                         /*
1285                          * XXXRW: Why not commit the port if the address is
1286                          * !INADDR_ANY?
1287                          */
1288                         /* Commit the local port if newly assigned. */
1289                         if (inp->inp_laddr.s_addr == INADDR_ANY &&
1290                             inp->inp_lport == 0) {
1291                                 INP_WLOCK_ASSERT(inp);
1292                                 /*
1293                                  * Remember addr if jailed, to prevent
1294                                  * rebinding.
1295                                  */
1296                                 if (prison_flag(td->td_ucred, PR_IP4))
1297                                         inp->inp_laddr = laddr;
1298                                 inp->inp_lport = lport;
1299                                 INP_HASH_WLOCK(pcbinfo);
1300                                 error = in_pcbinshash(inp);
1301                                 INP_HASH_WUNLOCK(pcbinfo);
1302                                 if (error != 0) {
1303                                         inp->inp_lport = 0;
1304                                         error = EAGAIN;
1305                                         goto release;
1306                                 }
1307                                 inp->inp_flags |= INP_ANONPORT;
1308                         }
1309                 } else {
1310                         faddr = sin->sin_addr;
1311                         fport = sin->sin_port;
1312                 }
1313         } else {
1314                 INP_LOCK_ASSERT(inp);
1315                 faddr = inp->inp_faddr;
1316                 fport = inp->inp_fport;
1317                 if (faddr.s_addr == INADDR_ANY) {
1318                         error = ENOTCONN;
1319                         goto release;
1320                 }
1321         }
1322
1323         /*
1324          * Calculate data length and get a mbuf for UDP, IP, and possible
1325          * link-layer headers.  Immediate slide the data pointer back forward
1326          * since we won't use that space at this layer.
1327          */
1328         M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT);
1329         if (m == NULL) {
1330                 error = ENOBUFS;
1331                 goto release;
1332         }
1333         m->m_data += max_linkhdr;
1334         m->m_len -= max_linkhdr;
1335         m->m_pkthdr.len -= max_linkhdr;
1336
1337         /*
1338          * Fill in mbuf with extended UDP header and addresses and length put
1339          * into network format.
1340          */
1341         ui = mtod(m, struct udpiphdr *);
1342         bzero(ui->ui_x1, sizeof(ui->ui_x1));    /* XXX still needed? */
1343         ui->ui_v = IPVERSION << 4;
1344         ui->ui_pr = pr;
1345         ui->ui_src = laddr;
1346         ui->ui_dst = faddr;
1347         ui->ui_sport = lport;
1348         ui->ui_dport = fport;
1349         ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
1350         if (pr == IPPROTO_UDPLITE) {
1351                 struct udpcb *up;
1352                 uint16_t plen;
1353
1354                 up = intoudpcb(inp);
1355                 cscov = up->u_txcslen;
1356                 plen = (u_short)len + sizeof(struct udphdr);
1357                 if (cscov >= plen)
1358                         cscov = 0;
1359                 ui->ui_len = htons(plen);
1360                 ui->ui_ulen = htons(cscov);
1361                 /*
1362                  * For UDP-Lite, checksum coverage length of zero means
1363                  * the entire UDPLite packet is covered by the checksum.
1364                  */
1365                 cscov_partial = (cscov == 0) ? 0 : 1;
1366         }
1367
1368         /*
1369          * Set the Don't Fragment bit in the IP header.
1370          */
1371         if (inp->inp_flags & INP_DONTFRAG) {
1372                 struct ip *ip;
1373
1374                 ip = (struct ip *)&ui->ui_i;
1375                 ip->ip_off |= htons(IP_DF);
1376         }
1377
1378         ipflags = 0;
1379         if (inp->inp_socket->so_options & SO_DONTROUTE)
1380                 ipflags |= IP_ROUTETOIF;
1381         if (inp->inp_socket->so_options & SO_BROADCAST)
1382                 ipflags |= IP_ALLOWBROADCAST;
1383         if (inp->inp_flags & INP_ONESBCAST)
1384                 ipflags |= IP_SENDONES;
1385
1386 #ifdef MAC
1387         mac_inpcb_create_mbuf(inp, m);
1388 #endif
1389
1390         /*
1391          * Set up checksum and output datagram.
1392          */
1393         ui->ui_sum = 0;
1394         if (pr == IPPROTO_UDPLITE) {
1395                 if (inp->inp_flags & INP_ONESBCAST)
1396                         faddr.s_addr = INADDR_BROADCAST;
1397                 if (cscov_partial) {
1398                         if ((ui->ui_sum = in_cksum(m, sizeof(struct ip) + cscov)) == 0)
1399                                 ui->ui_sum = 0xffff;
1400                 } else {
1401                         if ((ui->ui_sum = in_cksum(m, sizeof(struct udpiphdr) + len)) == 0)
1402                                 ui->ui_sum = 0xffff;
1403                 }
1404         } else if (V_udp_cksum) {
1405                 if (inp->inp_flags & INP_ONESBCAST)
1406                         faddr.s_addr = INADDR_BROADCAST;
1407                 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
1408                     htons((u_short)len + sizeof(struct udphdr) + pr));
1409                 m->m_pkthdr.csum_flags = CSUM_UDP;
1410                 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1411         }
1412         ((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len);
1413         ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl;    /* XXX */
1414         ((struct ip *)ui)->ip_tos = tos;                /* XXX */
1415         UDPSTAT_INC(udps_opackets);
1416
1417         /*
1418          * Setup flowid / RSS information for outbound socket.
1419          *
1420          * Once the UDP code decides to set a flowid some other way,
1421          * this allows the flowid to be overridden by userland.
1422          */
1423         if (flowtype != M_HASHTYPE_NONE) {
1424                 m->m_pkthdr.flowid = flowid;
1425                 M_HASHTYPE_SET(m, flowtype);
1426         }
1427 #ifdef  RSS
1428         else {
1429                 uint32_t hash_val, hash_type;
1430                 /*
1431                  * Calculate an appropriate RSS hash for UDP and
1432                  * UDP Lite.
1433                  *
1434                  * The called function will take care of figuring out
1435                  * whether a 2-tuple or 4-tuple hash is required based
1436                  * on the currently configured scheme.
1437                  *
1438                  * Later later on connected socket values should be
1439                  * cached in the inpcb and reused, rather than constantly
1440                  * re-calculating it.
1441                  *
1442                  * UDP Lite is a different protocol number and will
1443                  * likely end up being hashed as a 2-tuple until
1444                  * RSS / NICs grow UDP Lite protocol awareness.
1445                  */
1446                 if (rss_proto_software_hash_v4(faddr, laddr, fport, lport,
1447                     pr, &hash_val, &hash_type) == 0) {
1448                         m->m_pkthdr.flowid = hash_val;
1449                         M_HASHTYPE_SET(m, hash_type);
1450                 }
1451         }
1452
1453         /*
1454          * Don't override with the inp cached flowid value.
1455          *
1456          * Depending upon the kind of send being done, the inp
1457          * flowid/flowtype values may actually not be appropriate
1458          * for this particular socket send.
1459          *
1460          * We should either leave the flowid at zero (which is what is
1461          * currently done) or set it to some software generated
1462          * hash value based on the packet contents.
1463          */
1464         ipflags |= IP_NODEFAULTFLOWID;
1465 #endif  /* RSS */
1466
1467         if (pr == IPPROTO_UDPLITE)
1468                 UDPLITE_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u);
1469         else
1470                 UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u);
1471         error = ip_output(m, inp->inp_options,
1472             INP_WLOCKED(inp) ? &inp->inp_route : NULL, ipflags,
1473             inp->inp_moptions, inp);
1474         INP_UNLOCK(inp);
1475         NET_EPOCH_EXIT(et);
1476         return (error);
1477
1478 release:
1479         INP_UNLOCK(inp);
1480         NET_EPOCH_EXIT(et);
1481         m_freem(m);
1482         return (error);
1483 }
1484
1485 static void
1486 udp_abort(struct socket *so)
1487 {
1488         struct inpcb *inp;
1489         struct inpcbinfo *pcbinfo;
1490
1491         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1492         inp = sotoinpcb(so);
1493         KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
1494         INP_WLOCK(inp);
1495         if (inp->inp_faddr.s_addr != INADDR_ANY) {
1496                 INP_HASH_WLOCK(pcbinfo);
1497                 in_pcbdisconnect(inp);
1498                 inp->inp_laddr.s_addr = INADDR_ANY;
1499                 INP_HASH_WUNLOCK(pcbinfo);
1500                 soisdisconnected(so);
1501         }
1502         INP_WUNLOCK(inp);
1503 }
1504
1505 static int
1506 udp_attach(struct socket *so, int proto, struct thread *td)
1507 {
1508         static uint32_t udp_flowid;
1509         struct inpcb *inp;
1510         struct inpcbinfo *pcbinfo;
1511         int error;
1512
1513         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1514         inp = sotoinpcb(so);
1515         KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
1516         error = soreserve(so, udp_sendspace, udp_recvspace);
1517         if (error)
1518                 return (error);
1519         INP_INFO_WLOCK(pcbinfo);
1520         error = in_pcballoc(so, pcbinfo);
1521         if (error) {
1522                 INP_INFO_WUNLOCK(pcbinfo);
1523                 return (error);
1524         }
1525
1526         inp = sotoinpcb(so);
1527         inp->inp_vflag |= INP_IPV4;
1528         inp->inp_ip_ttl = V_ip_defttl;
1529         inp->inp_flowid = atomic_fetchadd_int(&udp_flowid, 1);
1530         inp->inp_flowtype = M_HASHTYPE_OPAQUE;
1531
1532         error = udp_newudpcb(inp);
1533         if (error) {
1534                 in_pcbdetach(inp);
1535                 in_pcbfree(inp);
1536                 INP_INFO_WUNLOCK(pcbinfo);
1537                 return (error);
1538         }
1539
1540         INP_WUNLOCK(inp);
1541         INP_INFO_WUNLOCK(pcbinfo);
1542         return (0);
1543 }
1544 #endif /* INET */
1545
1546 int
1547 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f, udp_tun_icmp_t i, void *ctx)
1548 {
1549         struct inpcb *inp;
1550         struct udpcb *up;
1551
1552         KASSERT(so->so_type == SOCK_DGRAM,
1553             ("udp_set_kernel_tunneling: !dgram"));
1554         inp = sotoinpcb(so);
1555         KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL"));
1556         INP_WLOCK(inp);
1557         up = intoudpcb(inp);
1558         if ((up->u_tun_func != NULL) ||
1559             (up->u_icmp_func != NULL)) {
1560                 INP_WUNLOCK(inp);
1561                 return (EBUSY);
1562         }
1563         up->u_tun_func = f;
1564         up->u_icmp_func = i;
1565         up->u_tun_ctx = ctx;
1566         INP_WUNLOCK(inp);
1567         return (0);
1568 }
1569
1570 #ifdef INET
1571 static int
1572 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1573 {
1574         struct inpcb *inp;
1575         struct inpcbinfo *pcbinfo;
1576         int error;
1577
1578         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1579         inp = sotoinpcb(so);
1580         KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1581         INP_WLOCK(inp);
1582         INP_HASH_WLOCK(pcbinfo);
1583         error = in_pcbbind(inp, nam, td->td_ucred);
1584         INP_HASH_WUNLOCK(pcbinfo);
1585         INP_WUNLOCK(inp);
1586         return (error);
1587 }
1588
1589 static void
1590 udp_close(struct socket *so)
1591 {
1592         struct inpcb *inp;
1593         struct inpcbinfo *pcbinfo;
1594
1595         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1596         inp = sotoinpcb(so);
1597         KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1598         INP_WLOCK(inp);
1599         if (inp->inp_faddr.s_addr != INADDR_ANY) {
1600                 INP_HASH_WLOCK(pcbinfo);
1601                 in_pcbdisconnect(inp);
1602                 inp->inp_laddr.s_addr = INADDR_ANY;
1603                 INP_HASH_WUNLOCK(pcbinfo);
1604                 soisdisconnected(so);
1605         }
1606         INP_WUNLOCK(inp);
1607 }
1608
1609 static int
1610 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1611 {
1612         struct epoch_tracker et;
1613         struct inpcb *inp;
1614         struct inpcbinfo *pcbinfo;
1615         struct sockaddr_in *sin;
1616         int error;
1617
1618         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1619         inp = sotoinpcb(so);
1620         KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1621         INP_WLOCK(inp);
1622         if (inp->inp_faddr.s_addr != INADDR_ANY) {
1623                 INP_WUNLOCK(inp);
1624                 return (EISCONN);
1625         }
1626         sin = (struct sockaddr_in *)nam;
1627         error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1628         if (error != 0) {
1629                 INP_WUNLOCK(inp);
1630                 return (error);
1631         }
1632         NET_EPOCH_ENTER(et);
1633         INP_HASH_WLOCK(pcbinfo);
1634         error = in_pcbconnect(inp, nam, td->td_ucred);
1635         INP_HASH_WUNLOCK(pcbinfo);
1636         NET_EPOCH_EXIT(et);
1637         if (error == 0)
1638                 soisconnected(so);
1639         INP_WUNLOCK(inp);
1640         return (error);
1641 }
1642
1643 static void
1644 udp_detach(struct socket *so)
1645 {
1646         struct inpcb *inp;
1647         struct inpcbinfo *pcbinfo;
1648         struct udpcb *up;
1649
1650         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1651         inp = sotoinpcb(so);
1652         KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1653         KASSERT(inp->inp_faddr.s_addr == INADDR_ANY,
1654             ("udp_detach: not disconnected"));
1655         INP_INFO_WLOCK(pcbinfo);
1656         INP_WLOCK(inp);
1657         up = intoudpcb(inp);
1658         KASSERT(up != NULL, ("%s: up == NULL", __func__));
1659         inp->inp_ppcb = NULL;
1660         in_pcbdetach(inp);
1661         in_pcbfree(inp);
1662         INP_INFO_WUNLOCK(pcbinfo);
1663         udp_discardcb(up);
1664 }
1665
1666 static int
1667 udp_disconnect(struct socket *so)
1668 {
1669         struct inpcb *inp;
1670         struct inpcbinfo *pcbinfo;
1671
1672         pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol);
1673         inp = sotoinpcb(so);
1674         KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1675         INP_WLOCK(inp);
1676         if (inp->inp_faddr.s_addr == INADDR_ANY) {
1677                 INP_WUNLOCK(inp);
1678                 return (ENOTCONN);
1679         }
1680         INP_HASH_WLOCK(pcbinfo);
1681         in_pcbdisconnect(inp);
1682         inp->inp_laddr.s_addr = INADDR_ANY;
1683         INP_HASH_WUNLOCK(pcbinfo);
1684         SOCK_LOCK(so);
1685         so->so_state &= ~SS_ISCONNECTED;                /* XXX */
1686         SOCK_UNLOCK(so);
1687         INP_WUNLOCK(inp);
1688         return (0);
1689 }
1690
1691 static int
1692 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1693     struct mbuf *control, struct thread *td)
1694 {
1695         struct inpcb *inp;
1696
1697         inp = sotoinpcb(so);
1698         KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1699         return (udp_output(inp, m, addr, control, td));
1700 }
1701 #endif /* INET */
1702
1703 int
1704 udp_shutdown(struct socket *so)
1705 {
1706         struct inpcb *inp;
1707
1708         inp = sotoinpcb(so);
1709         KASSERT(inp != NULL, ("udp_shutdown: inp == NULL"));
1710         INP_WLOCK(inp);
1711         socantsendmore(so);
1712         INP_WUNLOCK(inp);
1713         return (0);
1714 }
1715
1716 #ifdef INET
1717 struct pr_usrreqs udp_usrreqs = {
1718         .pru_abort =            udp_abort,
1719         .pru_attach =           udp_attach,
1720         .pru_bind =             udp_bind,
1721         .pru_connect =          udp_connect,
1722         .pru_control =          in_control,
1723         .pru_detach =           udp_detach,
1724         .pru_disconnect =       udp_disconnect,
1725         .pru_peeraddr =         in_getpeeraddr,
1726         .pru_send =             udp_send,
1727         .pru_soreceive =        soreceive_dgram,
1728         .pru_sosend =           sosend_dgram,
1729         .pru_shutdown =         udp_shutdown,
1730         .pru_sockaddr =         in_getsockaddr,
1731         .pru_sosetlabel =       in_pcbsosetlabel,
1732         .pru_close =            udp_close,
1733 };
1734 #endif /* INET */