1 /*- 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 3 * The Regents of the University of California. 4 * Copyright (c) 2008 Robert N. M. Watson 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 4. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 * 31 * @(#)udp_usrreq.c 8.6 (Berkeley) 5/23/95 32 */ 33 34 #include <sys/cdefs.h> 35 __FBSDID("$FreeBSD$"); 36 37 #include "opt_ipfw.h" 38 #include "opt_inet6.h" 39 #include "opt_ipsec.h" 40 41 #include <sys/param.h> 42 #include <sys/domain.h> 43 #include <sys/eventhandler.h> 44 #include <sys/jail.h> 45 #include <sys/kernel.h> 46 #include <sys/lock.h> 47 #include <sys/malloc.h> 48 #include <sys/mbuf.h> 49 #include <sys/priv.h> 50 #include <sys/proc.h> 51 #include <sys/protosw.h> 52 #include <sys/signalvar.h> 53 #include <sys/socket.h> 54 #include <sys/socketvar.h> 55 #include <sys/sx.h> 56 #include <sys/sysctl.h> 57 #include <sys/syslog.h> 58 #include <sys/systm.h> 59 60 #include <vm/uma.h> 61 62 #include <net/if.h> 63 #include <net/route.h> 64 65 #include <netinet/in.h> 66 #include <netinet/in_pcb.h> 67 #include <netinet/in_systm.h> 68 #include <netinet/in_var.h> 69 #include <netinet/ip.h> 70 #ifdef INET6 71 #include <netinet/ip6.h> 72 #endif 73 #include <netinet/ip_icmp.h> 74 #include <netinet/icmp_var.h> 75 #include <netinet/ip_var.h> 76 #include <netinet/ip_options.h> 77 #ifdef INET6 78 #include <netinet6/ip6_var.h> 79 #endif 80 #include <netinet/udp.h> 81 #include <netinet/udp_var.h> 82 83 #ifdef IPSEC 84 #include <netipsec/ipsec.h> 85 #include <netipsec/esp.h> 86 #endif 87 88 #include <machine/in_cksum.h> 89 90 #include <security/mac/mac_framework.h> 91 92 /* 93 * UDP protocol implementation. 94 * Per RFC 768, August, 1980. 95 */ 96 97 /* 98 * BSD 4.2 defaulted the udp checksum to be off. Turning off udp checksums 99 * removes the only data integrity mechanism for packets and malformed 100 * packets that would otherwise be discarded due to bad checksums, and may 101 * cause problems (especially for NFS data blocks). 102 */ 103 static int udp_cksum = 1; 104 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_RW, &udp_cksum, 105 0, "compute udp checksum"); 106 107 int udp_log_in_vain = 0; 108 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW, 109 &udp_log_in_vain, 0, "Log all incoming UDP packets"); 110 111 VNET_DEFINE(int, udp_blackhole) = 0; 112 SYSCTL_VNET_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_RW, 113 &VNET_NAME(udp_blackhole), 0, 114 "Do not send port unreachables for refused connects"); 115 116 u_long udp_sendspace = 9216; /* really max datagram size */ 117 /* 40 1K datagrams */ 118 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW, 119 &udp_sendspace, 0, "Maximum outgoing UDP datagram size"); 120 121 u_long udp_recvspace = 40 * (1024 + 122 #ifdef INET6 123 sizeof(struct sockaddr_in6) 124 #else 125 sizeof(struct sockaddr_in) 126 #endif 127 ); 128 129 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW, 130 &udp_recvspace, 0, "Maximum space for incoming UDP datagrams"); 131 132 VNET_DEFINE(struct inpcbhead, udb); /* from udp_var.h */ 133 VNET_DEFINE(struct inpcbinfo, udbinfo); 134 static VNET_DEFINE(uma_zone_t, udpcb_zone); 135 #define V_udpcb_zone VNET(udpcb_zone) 136 137 #ifndef UDBHASHSIZE 138 #define UDBHASHSIZE 128 139 #endif 140 141 VNET_DEFINE(struct udpstat, udpstat); /* from udp_var.h */ 142 SYSCTL_VNET_STRUCT(_net_inet_udp, UDPCTL_STATS, stats, CTLFLAG_RW, 143 &VNET_NAME(udpstat), udpstat, 144 "UDP statistics (struct udpstat, netinet/udp_var.h)"); 145 146 static void udp_detach(struct socket *so); 147 static int udp_output(struct inpcb *, struct mbuf *, struct sockaddr *, 148 struct mbuf *, struct thread *); 149 #ifdef IPSEC 150 #ifdef IPSEC_NAT_T 151 #define UF_ESPINUDP_ALL (UF_ESPINUDP_NON_IKE|UF_ESPINUDP) 152 #ifdef INET 153 static struct mbuf *udp4_espdecap(struct inpcb *, struct mbuf *, int); 154 #endif 155 #endif /* IPSEC_NAT_T */ 156 #endif /* IPSEC */ 157 158 static void 159 udp_zone_change(void *tag) 160 { 161 162 uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets); 163 uma_zone_set_max(V_udpcb_zone, maxsockets); 164 } 165 166 static int 167 udp_inpcb_init(void *mem, int size, int flags) 168 { 169 struct inpcb *inp; 170 171 inp = mem; 172 INP_LOCK_INIT(inp, "inp", "udpinp"); 173 return (0); 174 } 175 176 void 177 udp_init(void) 178 { 179 180 in_pcbinfo_init(&V_udbinfo, "udp", &V_udb, UDBHASHSIZE, UDBHASHSIZE, 181 "udp_inpcb", udp_inpcb_init, NULL, UMA_ZONE_NOFREE); 182 V_udpcb_zone = uma_zcreate("udpcb", sizeof(struct udpcb), 183 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 184 uma_zone_set_max(V_udpcb_zone, maxsockets); 185 EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL, 186 EVENTHANDLER_PRI_ANY); 187 } 188 189 /* 190 * Kernel module interface for updating udpstat. The argument is an index 191 * into udpstat treated as an array of u_long. While this encodes the 192 * general layout of udpstat into the caller, it doesn't encode its location, 193 * so that future changes to add, for example, per-CPU stats support won't 194 * cause binary compatibility problems for kernel modules. 195 */ 196 void 197 kmod_udpstat_inc(int statnum) 198 { 199 200 (*((u_long *)&V_udpstat + statnum))++; 201 } 202 203 int 204 udp_newudpcb(struct inpcb *inp) 205 { 206 struct udpcb *up; 207 208 up = uma_zalloc(V_udpcb_zone, M_NOWAIT | M_ZERO); 209 if (up == NULL) 210 return (ENOBUFS); 211 inp->inp_ppcb = up; 212 return (0); 213 } 214 215 void 216 udp_discardcb(struct udpcb *up) 217 { 218 219 uma_zfree(V_udpcb_zone, up); 220 } 221 222 #ifdef VIMAGE 223 void 224 udp_destroy(void) 225 { 226 227 in_pcbinfo_destroy(&V_udbinfo); 228 uma_zdestroy(V_udpcb_zone); 229 } 230 #endif 231 232 /* 233 * Subroutine of udp_input(), which appends the provided mbuf chain to the 234 * passed pcb/socket. The caller must provide a sockaddr_in via udp_in that 235 * contains the source address. If the socket ends up being an IPv6 socket, 236 * udp_append() will convert to a sockaddr_in6 before passing the address 237 * into the socket code. 238 */ 239 static void 240 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off, 241 struct sockaddr_in *udp_in) 242 { 243 struct sockaddr *append_sa; 244 struct socket *so; 245 struct mbuf *opts = 0; 246 #ifdef INET6 247 struct sockaddr_in6 udp_in6; 248 #endif 249 #ifdef IPSEC 250 #ifdef IPSEC_NAT_T 251 #ifdef INET 252 struct udpcb *up; 253 #endif 254 #endif 255 #endif 256 257 INP_RLOCK_ASSERT(inp); 258 259 #ifdef IPSEC 260 /* Check AH/ESP integrity. */ 261 if (ipsec4_in_reject(n, inp)) { 262 m_freem(n); 263 V_ipsec4stat.in_polvio++; 264 return; 265 } 266 #ifdef IPSEC_NAT_T 267 #ifdef INET 268 up = intoudpcb(inp); 269 KASSERT(up != NULL, ("%s: udpcb NULL", __func__)); 270 if (up->u_flags & UF_ESPINUDP_ALL) { /* IPSec UDP encaps. */ 271 n = udp4_espdecap(inp, n, off); 272 if (n == NULL) /* Consumed. */ 273 return; 274 } 275 #endif /* INET */ 276 #endif /* IPSEC_NAT_T */ 277 #endif /* IPSEC */ 278 #ifdef MAC 279 if (mac_inpcb_check_deliver(inp, n) != 0) { 280 m_freem(n); 281 return; 282 } 283 #endif 284 if (inp->inp_flags & INP_CONTROLOPTS || 285 inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) { 286 #ifdef INET6 287 if (inp->inp_vflag & INP_IPV6) 288 (void)ip6_savecontrol_v4(inp, n, &opts, NULL); 289 else 290 #endif 291 ip_savecontrol(inp, &opts, ip, n); 292 } 293 #ifdef INET6 294 if (inp->inp_vflag & INP_IPV6) { 295 bzero(&udp_in6, sizeof(udp_in6)); 296 udp_in6.sin6_len = sizeof(udp_in6); 297 udp_in6.sin6_family = AF_INET6; 298 in6_sin_2_v4mapsin6(udp_in, &udp_in6); 299 append_sa = (struct sockaddr *)&udp_in6; 300 } else 301 #endif 302 append_sa = (struct sockaddr *)udp_in; 303 m_adj(n, off); 304 305 so = inp->inp_socket; 306 SOCKBUF_LOCK(&so->so_rcv); 307 if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) { 308 SOCKBUF_UNLOCK(&so->so_rcv); 309 m_freem(n); 310 if (opts) 311 m_freem(opts); 312 UDPSTAT_INC(udps_fullsock); 313 } else 314 sorwakeup_locked(so); 315 } 316 317 void 318 udp_input(struct mbuf *m, int off) 319 { 320 int iphlen = off; 321 struct ip *ip; 322 struct udphdr *uh; 323 struct ifnet *ifp; 324 struct inpcb *inp; 325 struct udpcb *up; 326 int len; 327 struct ip save_ip; 328 struct sockaddr_in udp_in; 329 #ifdef IPFIREWALL_FORWARD 330 struct m_tag *fwd_tag; 331 #endif 332 333 ifp = m->m_pkthdr.rcvif; 334 UDPSTAT_INC(udps_ipackets); 335 336 /* 337 * Strip IP options, if any; should skip this, make available to 338 * user, and use on returned packets, but we don't yet have a way to 339 * check the checksum with options still present. 340 */ 341 if (iphlen > sizeof (struct ip)) { 342 ip_stripoptions(m, (struct mbuf *)0); 343 iphlen = sizeof(struct ip); 344 } 345 346 /* 347 * Get IP and UDP header together in first mbuf. 348 */ 349 ip = mtod(m, struct ip *); 350 if (m->m_len < iphlen + sizeof(struct udphdr)) { 351 if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == 0) { 352 UDPSTAT_INC(udps_hdrops); 353 return; 354 } 355 ip = mtod(m, struct ip *); 356 } 357 uh = (struct udphdr *)((caddr_t)ip + iphlen); 358 359 /* 360 * Destination port of 0 is illegal, based on RFC768. 361 */ 362 if (uh->uh_dport == 0) 363 goto badunlocked; 364 365 /* 366 * Construct sockaddr format source address. Stuff source address 367 * and datagram in user buffer. 368 */ 369 bzero(&udp_in, sizeof(udp_in)); 370 udp_in.sin_len = sizeof(udp_in); 371 udp_in.sin_family = AF_INET; 372 udp_in.sin_port = uh->uh_sport; 373 udp_in.sin_addr = ip->ip_src; 374 375 /* 376 * Make mbuf data length reflect UDP length. If not enough data to 377 * reflect UDP length, drop. 378 */ 379 len = ntohs((u_short)uh->uh_ulen); 380 if (ip->ip_len != len) { 381 if (len > ip->ip_len || len < sizeof(struct udphdr)) { 382 UDPSTAT_INC(udps_badlen); 383 goto badunlocked; 384 } 385 m_adj(m, len - ip->ip_len); 386 /* ip->ip_len = len; */ 387 } 388 389 /* 390 * Save a copy of the IP header in case we want restore it for 391 * sending an ICMP error message in response. 392 */ 393 if (!V_udp_blackhole) 394 save_ip = *ip; 395 else 396 memset(&save_ip, 0, sizeof(save_ip)); 397 398 /* 399 * Checksum extended UDP header and data. 400 */ 401 if (uh->uh_sum) { 402 u_short uh_sum; 403 404 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 405 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 406 uh_sum = m->m_pkthdr.csum_data; 407 else 408 uh_sum = in_pseudo(ip->ip_src.s_addr, 409 ip->ip_dst.s_addr, htonl((u_short)len + 410 m->m_pkthdr.csum_data + IPPROTO_UDP)); 411 uh_sum ^= 0xffff; 412 } else { 413 char b[9]; 414 415 bcopy(((struct ipovly *)ip)->ih_x1, b, 9); 416 bzero(((struct ipovly *)ip)->ih_x1, 9); 417 ((struct ipovly *)ip)->ih_len = uh->uh_ulen; 418 uh_sum = in_cksum(m, len + sizeof (struct ip)); 419 bcopy(b, ((struct ipovly *)ip)->ih_x1, 9); 420 } 421 if (uh_sum) { 422 UDPSTAT_INC(udps_badsum); 423 m_freem(m); 424 return; 425 } 426 } else 427 UDPSTAT_INC(udps_nosum); 428 429 #ifdef IPFIREWALL_FORWARD 430 /* 431 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 432 */ 433 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 434 if (fwd_tag != NULL) { 435 struct sockaddr_in *next_hop; 436 437 /* 438 * Do the hack. 439 */ 440 next_hop = (struct sockaddr_in *)(fwd_tag + 1); 441 ip->ip_dst = next_hop->sin_addr; 442 uh->uh_dport = ntohs(next_hop->sin_port); 443 444 /* 445 * Remove the tag from the packet. We don't need it anymore. 446 */ 447 m_tag_delete(m, fwd_tag); 448 } 449 #endif 450 451 INP_INFO_RLOCK(&V_udbinfo); 452 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 453 in_broadcast(ip->ip_dst, ifp)) { 454 struct inpcb *last; 455 struct ip_moptions *imo; 456 457 last = NULL; 458 LIST_FOREACH(inp, &V_udb, inp_list) { 459 if (inp->inp_lport != uh->uh_dport) 460 continue; 461 #ifdef INET6 462 if ((inp->inp_vflag & INP_IPV4) == 0) 463 continue; 464 #endif 465 if (inp->inp_laddr.s_addr != INADDR_ANY && 466 inp->inp_laddr.s_addr != ip->ip_dst.s_addr) 467 continue; 468 if (inp->inp_faddr.s_addr != INADDR_ANY && 469 inp->inp_faddr.s_addr != ip->ip_src.s_addr) 470 continue; 471 if (inp->inp_fport != 0 && 472 inp->inp_fport != uh->uh_sport) 473 continue; 474 475 INP_RLOCK(inp); 476 477 /* 478 * Handle socket delivery policy for any-source 479 * and source-specific multicast. [RFC3678] 480 */ 481 imo = inp->inp_moptions; 482 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) && 483 imo != NULL) { 484 struct sockaddr_in group; 485 int blocked; 486 487 bzero(&group, sizeof(struct sockaddr_in)); 488 group.sin_len = sizeof(struct sockaddr_in); 489 group.sin_family = AF_INET; 490 group.sin_addr = ip->ip_dst; 491 492 blocked = imo_multi_filter(imo, ifp, 493 (struct sockaddr *)&group, 494 (struct sockaddr *)&udp_in); 495 if (blocked != MCAST_PASS) { 496 if (blocked == MCAST_NOTGMEMBER) 497 IPSTAT_INC(ips_notmember); 498 if (blocked == MCAST_NOTSMEMBER || 499 blocked == MCAST_MUTED) 500 UDPSTAT_INC(udps_filtermcast); 501 INP_RUNLOCK(inp); 502 continue; 503 } 504 } 505 if (last != NULL) { 506 struct mbuf *n; 507 508 n = m_copy(m, 0, M_COPYALL); 509 up = intoudpcb(last); 510 if (up->u_tun_func == NULL) { 511 if (n != NULL) 512 udp_append(last, 513 ip, n, 514 iphlen + 515 sizeof(struct udphdr), 516 &udp_in); 517 } else { 518 /* 519 * Engage the tunneling protocol we 520 * will have to leave the info_lock 521 * up, since we are hunting through 522 * multiple UDP's. 523 */ 524 525 (*up->u_tun_func)(n, iphlen, last); 526 } 527 INP_RUNLOCK(last); 528 } 529 last = inp; 530 /* 531 * Don't look for additional matches if this one does 532 * not have either the SO_REUSEPORT or SO_REUSEADDR 533 * socket options set. This heuristic avoids 534 * searching through all pcbs in the common case of a 535 * non-shared port. It assumes that an application 536 * will never clear these options after setting them. 537 */ 538 if ((last->inp_socket->so_options & 539 (SO_REUSEPORT|SO_REUSEADDR)) == 0) 540 break; 541 } 542 543 if (last == NULL) { 544 /* 545 * No matching pcb found; discard datagram. (No need 546 * to send an ICMP Port Unreachable for a broadcast 547 * or multicast datgram.) 548 */ 549 UDPSTAT_INC(udps_noportbcast); 550 goto badheadlocked; 551 } 552 up = intoudpcb(last); 553 if (up->u_tun_func == NULL) { 554 udp_append(last, ip, m, iphlen + sizeof(struct udphdr), 555 &udp_in); 556 } else { 557 /* 558 * Engage the tunneling protocol. 559 */ 560 (*up->u_tun_func)(m, iphlen, last); 561 } 562 INP_RUNLOCK(last); 563 INP_INFO_RUNLOCK(&V_udbinfo); 564 return; 565 } 566 567 /* 568 * Locate pcb for datagram. 569 */ 570 inp = in_pcblookup_hash(&V_udbinfo, ip->ip_src, uh->uh_sport, 571 ip->ip_dst, uh->uh_dport, 1, ifp); 572 if (inp == NULL) { 573 if (udp_log_in_vain) { 574 char buf[4*sizeof "123"]; 575 576 strcpy(buf, inet_ntoa(ip->ip_dst)); 577 log(LOG_INFO, 578 "Connection attempt to UDP %s:%d from %s:%d\n", 579 buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src), 580 ntohs(uh->uh_sport)); 581 } 582 UDPSTAT_INC(udps_noport); 583 if (m->m_flags & (M_BCAST | M_MCAST)) { 584 UDPSTAT_INC(udps_noportbcast); 585 goto badheadlocked; 586 } 587 if (V_udp_blackhole) 588 goto badheadlocked; 589 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0) 590 goto badheadlocked; 591 *ip = save_ip; 592 ip->ip_len += iphlen; 593 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0); 594 INP_INFO_RUNLOCK(&V_udbinfo); 595 return; 596 } 597 598 /* 599 * Check the minimum TTL for socket. 600 */ 601 INP_RLOCK(inp); 602 INP_INFO_RUNLOCK(&V_udbinfo); 603 if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) { 604 INP_RUNLOCK(inp); 605 goto badunlocked; 606 } 607 up = intoudpcb(inp); 608 if (up->u_tun_func == NULL) { 609 udp_append(inp, ip, m, iphlen + sizeof(struct udphdr), &udp_in); 610 } else { 611 /* 612 * Engage the tunneling protocol. 613 */ 614 615 (*up->u_tun_func)(m, iphlen, inp); 616 } 617 INP_RUNLOCK(inp); 618 return; 619 620 badheadlocked: 621 if (inp) 622 INP_RUNLOCK(inp); 623 INP_INFO_RUNLOCK(&V_udbinfo); 624 badunlocked: 625 m_freem(m); 626 } 627 628 /* 629 * Notify a udp user of an asynchronous error; just wake up so that they can 630 * collect error status. 631 */ 632 struct inpcb * 633 udp_notify(struct inpcb *inp, int errno) 634 { 635 636 /* 637 * While udp_ctlinput() always calls udp_notify() with a read lock 638 * when invoking it directly, in_pcbnotifyall() currently uses write 639 * locks due to sharing code with TCP. For now, accept either a read 640 * or a write lock, but a read lock is sufficient. 641 */ 642 INP_LOCK_ASSERT(inp); 643 644 inp->inp_socket->so_error = errno; 645 sorwakeup(inp->inp_socket); 646 sowwakeup(inp->inp_socket); 647 return (inp); 648 } 649 650 void 651 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 652 { 653 struct ip *ip = vip; 654 struct udphdr *uh; 655 struct in_addr faddr; 656 struct inpcb *inp; 657 658 faddr = ((struct sockaddr_in *)sa)->sin_addr; 659 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 660 return; 661 662 /* 663 * Redirects don't need to be handled up here. 664 */ 665 if (PRC_IS_REDIRECT(cmd)) 666 return; 667 668 /* 669 * Hostdead is ugly because it goes linearly through all PCBs. 670 * 671 * XXX: We never get this from ICMP, otherwise it makes an excellent 672 * DoS attack on machines with many connections. 673 */ 674 if (cmd == PRC_HOSTDEAD) 675 ip = NULL; 676 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 677 return; 678 if (ip != NULL) { 679 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 680 INP_INFO_RLOCK(&V_udbinfo); 681 inp = in_pcblookup_hash(&V_udbinfo, faddr, uh->uh_dport, 682 ip->ip_src, uh->uh_sport, 0, NULL); 683 if (inp != NULL) { 684 INP_RLOCK(inp); 685 if (inp->inp_socket != NULL) { 686 udp_notify(inp, inetctlerrmap[cmd]); 687 } 688 INP_RUNLOCK(inp); 689 } 690 INP_INFO_RUNLOCK(&V_udbinfo); 691 } else 692 in_pcbnotifyall(&V_udbinfo, faddr, inetctlerrmap[cmd], 693 udp_notify); 694 } 695 696 static int 697 udp_pcblist(SYSCTL_HANDLER_ARGS) 698 { 699 int error, i, n; 700 struct inpcb *inp, **inp_list; 701 inp_gen_t gencnt; 702 struct xinpgen xig; 703 704 /* 705 * The process of preparing the PCB list is too time-consuming and 706 * resource-intensive to repeat twice on every request. 707 */ 708 if (req->oldptr == 0) { 709 n = V_udbinfo.ipi_count; 710 req->oldidx = 2 * (sizeof xig) 711 + (n + n/8) * sizeof(struct xinpcb); 712 return (0); 713 } 714 715 if (req->newptr != 0) 716 return (EPERM); 717 718 /* 719 * OK, now we're committed to doing something. 720 */ 721 INP_INFO_RLOCK(&V_udbinfo); 722 gencnt = V_udbinfo.ipi_gencnt; 723 n = V_udbinfo.ipi_count; 724 INP_INFO_RUNLOCK(&V_udbinfo); 725 726 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 727 + n * sizeof(struct xinpcb)); 728 if (error != 0) 729 return (error); 730 731 xig.xig_len = sizeof xig; 732 xig.xig_count = n; 733 xig.xig_gen = gencnt; 734 xig.xig_sogen = so_gencnt; 735 error = SYSCTL_OUT(req, &xig, sizeof xig); 736 if (error) 737 return (error); 738 739 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 740 if (inp_list == 0) 741 return (ENOMEM); 742 743 INP_INFO_RLOCK(&V_udbinfo); 744 for (inp = LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n; 745 inp = LIST_NEXT(inp, inp_list)) { 746 INP_WLOCK(inp); 747 if (inp->inp_gencnt <= gencnt && 748 cr_canseeinpcb(req->td->td_ucred, inp) == 0) { 749 in_pcbref(inp); 750 inp_list[i++] = inp; 751 } 752 INP_WUNLOCK(inp); 753 } 754 INP_INFO_RUNLOCK(&V_udbinfo); 755 n = i; 756 757 error = 0; 758 for (i = 0; i < n; i++) { 759 inp = inp_list[i]; 760 INP_RLOCK(inp); 761 if (inp->inp_gencnt <= gencnt) { 762 struct xinpcb xi; 763 764 bzero(&xi, sizeof(xi)); 765 xi.xi_len = sizeof xi; 766 /* XXX should avoid extra copy */ 767 bcopy(inp, &xi.xi_inp, sizeof *inp); 768 if (inp->inp_socket) 769 sotoxsocket(inp->inp_socket, &xi.xi_socket); 770 xi.xi_inp.inp_gencnt = inp->inp_gencnt; 771 INP_RUNLOCK(inp); 772 error = SYSCTL_OUT(req, &xi, sizeof xi); 773 } else 774 INP_RUNLOCK(inp); 775 } 776 INP_INFO_WLOCK(&V_udbinfo); 777 for (i = 0; i < n; i++) { 778 inp = inp_list[i]; 779 INP_WLOCK(inp); 780 if (!in_pcbrele(inp)) 781 INP_WUNLOCK(inp); 782 } 783 INP_INFO_WUNLOCK(&V_udbinfo); 784 785 if (!error) { 786 /* 787 * Give the user an updated idea of our state. If the 788 * generation differs from what we told her before, she knows 789 * that something happened while we were processing this 790 * request, and it might be necessary to retry. 791 */ 792 INP_INFO_RLOCK(&V_udbinfo); 793 xig.xig_gen = V_udbinfo.ipi_gencnt; 794 xig.xig_sogen = so_gencnt; 795 xig.xig_count = V_udbinfo.ipi_count; 796 INP_INFO_RUNLOCK(&V_udbinfo); 797 error = SYSCTL_OUT(req, &xig, sizeof xig); 798 } 799 free(inp_list, M_TEMP); 800 return (error); 801 } 802 803 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0, 804 udp_pcblist, "S,xinpcb", "List of active UDP sockets"); 805 806 static int 807 udp_getcred(SYSCTL_HANDLER_ARGS) 808 { 809 struct xucred xuc; 810 struct sockaddr_in addrs[2]; 811 struct inpcb *inp; 812 int error; 813 814 error = priv_check(req->td, PRIV_NETINET_GETCRED); 815 if (error) 816 return (error); 817 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 818 if (error) 819 return (error); 820 INP_INFO_RLOCK(&V_udbinfo); 821 inp = in_pcblookup_hash(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port, 822 addrs[0].sin_addr, addrs[0].sin_port, 1, NULL); 823 if (inp != NULL) { 824 INP_RLOCK(inp); 825 INP_INFO_RUNLOCK(&V_udbinfo); 826 if (inp->inp_socket == NULL) 827 error = ENOENT; 828 if (error == 0) 829 error = cr_canseeinpcb(req->td->td_ucred, inp); 830 if (error == 0) 831 cru2x(inp->inp_cred, &xuc); 832 INP_RUNLOCK(inp); 833 } else { 834 INP_INFO_RUNLOCK(&V_udbinfo); 835 error = ENOENT; 836 } 837 if (error == 0) 838 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 839 return (error); 840 } 841 842 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred, 843 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 844 udp_getcred, "S,xucred", "Get the xucred of a UDP connection"); 845 846 int 847 udp_ctloutput(struct socket *so, struct sockopt *sopt) 848 { 849 int error = 0, optval; 850 struct inpcb *inp; 851 #ifdef IPSEC_NAT_T 852 struct udpcb *up; 853 #endif 854 855 inp = sotoinpcb(so); 856 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 857 INP_WLOCK(inp); 858 if (sopt->sopt_level != IPPROTO_UDP) { 859 #ifdef INET6 860 if (INP_CHECK_SOCKAF(so, AF_INET6)) { 861 INP_WUNLOCK(inp); 862 error = ip6_ctloutput(so, sopt); 863 } else { 864 #endif 865 INP_WUNLOCK(inp); 866 error = ip_ctloutput(so, sopt); 867 #ifdef INET6 868 } 869 #endif 870 return (error); 871 } 872 873 switch (sopt->sopt_dir) { 874 case SOPT_SET: 875 switch (sopt->sopt_name) { 876 case UDP_ENCAP: 877 INP_WUNLOCK(inp); 878 error = sooptcopyin(sopt, &optval, sizeof optval, 879 sizeof optval); 880 if (error) 881 break; 882 inp = sotoinpcb(so); 883 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 884 INP_WLOCK(inp); 885 #ifdef IPSEC_NAT_T 886 up = intoudpcb(inp); 887 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 888 #endif 889 switch (optval) { 890 case 0: 891 /* Clear all UDP encap. */ 892 #ifdef IPSEC_NAT_T 893 up->u_flags &= ~UF_ESPINUDP_ALL; 894 #endif 895 break; 896 #ifdef IPSEC_NAT_T 897 case UDP_ENCAP_ESPINUDP: 898 case UDP_ENCAP_ESPINUDP_NON_IKE: 899 up->u_flags &= ~UF_ESPINUDP_ALL; 900 if (optval == UDP_ENCAP_ESPINUDP) 901 up->u_flags |= UF_ESPINUDP; 902 else if (optval == UDP_ENCAP_ESPINUDP_NON_IKE) 903 up->u_flags |= UF_ESPINUDP_NON_IKE; 904 break; 905 #endif 906 default: 907 error = EINVAL; 908 break; 909 } 910 INP_WUNLOCK(inp); 911 break; 912 default: 913 INP_WUNLOCK(inp); 914 error = ENOPROTOOPT; 915 break; 916 } 917 break; 918 case SOPT_GET: 919 switch (sopt->sopt_name) { 920 #ifdef IPSEC_NAT_T 921 case UDP_ENCAP: 922 up = intoudpcb(inp); 923 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 924 optval = up->u_flags & UF_ESPINUDP_ALL; 925 INP_WUNLOCK(inp); 926 error = sooptcopyout(sopt, &optval, sizeof optval); 927 break; 928 #endif 929 default: 930 INP_WUNLOCK(inp); 931 error = ENOPROTOOPT; 932 break; 933 } 934 break; 935 } 936 return (error); 937 } 938 939 static int 940 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr, 941 struct mbuf *control, struct thread *td) 942 { 943 struct udpiphdr *ui; 944 int len = m->m_pkthdr.len; 945 struct in_addr faddr, laddr; 946 struct cmsghdr *cm; 947 struct sockaddr_in *sin, src; 948 int error = 0; 949 int ipflags; 950 u_short fport, lport; 951 int unlock_udbinfo; 952 953 /* 954 * udp_output() may need to temporarily bind or connect the current 955 * inpcb. As such, we don't know up front whether we will need the 956 * pcbinfo lock or not. Do any work to decide what is needed up 957 * front before acquiring any locks. 958 */ 959 if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) { 960 if (control) 961 m_freem(control); 962 m_freem(m); 963 return (EMSGSIZE); 964 } 965 966 src.sin_family = 0; 967 if (control != NULL) { 968 /* 969 * XXX: Currently, we assume all the optional information is 970 * stored in a single mbuf. 971 */ 972 if (control->m_next) { 973 m_freem(control); 974 m_freem(m); 975 return (EINVAL); 976 } 977 for (; control->m_len > 0; 978 control->m_data += CMSG_ALIGN(cm->cmsg_len), 979 control->m_len -= CMSG_ALIGN(cm->cmsg_len)) { 980 cm = mtod(control, struct cmsghdr *); 981 if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0 982 || cm->cmsg_len > control->m_len) { 983 error = EINVAL; 984 break; 985 } 986 if (cm->cmsg_level != IPPROTO_IP) 987 continue; 988 989 switch (cm->cmsg_type) { 990 case IP_SENDSRCADDR: 991 if (cm->cmsg_len != 992 CMSG_LEN(sizeof(struct in_addr))) { 993 error = EINVAL; 994 break; 995 } 996 bzero(&src, sizeof(src)); 997 src.sin_family = AF_INET; 998 src.sin_len = sizeof(src); 999 src.sin_port = inp->inp_lport; 1000 src.sin_addr = 1001 *(struct in_addr *)CMSG_DATA(cm); 1002 break; 1003 1004 default: 1005 error = ENOPROTOOPT; 1006 break; 1007 } 1008 if (error) 1009 break; 1010 } 1011 m_freem(control); 1012 } 1013 if (error) { 1014 m_freem(m); 1015 return (error); 1016 } 1017 1018 /* 1019 * Depending on whether or not the application has bound or connected 1020 * the socket, we may have to do varying levels of work. The optimal 1021 * case is for a connected UDP socket, as a global lock isn't 1022 * required at all. 1023 * 1024 * In order to decide which we need, we require stability of the 1025 * inpcb binding, which we ensure by acquiring a read lock on the 1026 * inpcb. This doesn't strictly follow the lock order, so we play 1027 * the trylock and retry game; note that we may end up with more 1028 * conservative locks than required the second time around, so later 1029 * assertions have to accept that. Further analysis of the number of 1030 * misses under contention is required. 1031 */ 1032 sin = (struct sockaddr_in *)addr; 1033 INP_RLOCK(inp); 1034 if (sin != NULL && 1035 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) { 1036 INP_RUNLOCK(inp); 1037 INP_INFO_WLOCK(&V_udbinfo); 1038 INP_WLOCK(inp); 1039 unlock_udbinfo = 2; 1040 } else if ((sin != NULL && ( 1041 (sin->sin_addr.s_addr == INADDR_ANY) || 1042 (sin->sin_addr.s_addr == INADDR_BROADCAST) || 1043 (inp->inp_laddr.s_addr == INADDR_ANY) || 1044 (inp->inp_lport == 0))) || 1045 (src.sin_family == AF_INET)) { 1046 if (!INP_INFO_TRY_RLOCK(&V_udbinfo)) { 1047 INP_RUNLOCK(inp); 1048 INP_INFO_RLOCK(&V_udbinfo); 1049 INP_RLOCK(inp); 1050 } 1051 unlock_udbinfo = 1; 1052 } else 1053 unlock_udbinfo = 0; 1054 1055 /* 1056 * If the IP_SENDSRCADDR control message was specified, override the 1057 * source address for this datagram. Its use is invalidated if the 1058 * address thus specified is incomplete or clobbers other inpcbs. 1059 */ 1060 laddr = inp->inp_laddr; 1061 lport = inp->inp_lport; 1062 if (src.sin_family == AF_INET) { 1063 INP_INFO_LOCK_ASSERT(&V_udbinfo); 1064 if ((lport == 0) || 1065 (laddr.s_addr == INADDR_ANY && 1066 src.sin_addr.s_addr == INADDR_ANY)) { 1067 error = EINVAL; 1068 goto release; 1069 } 1070 error = in_pcbbind_setup(inp, (struct sockaddr *)&src, 1071 &laddr.s_addr, &lport, td->td_ucred); 1072 if (error) 1073 goto release; 1074 } 1075 1076 /* 1077 * If a UDP socket has been connected, then a local address/port will 1078 * have been selected and bound. 1079 * 1080 * If a UDP socket has not been connected to, then an explicit 1081 * destination address must be used, in which case a local 1082 * address/port may not have been selected and bound. 1083 */ 1084 if (sin != NULL) { 1085 INP_LOCK_ASSERT(inp); 1086 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1087 error = EISCONN; 1088 goto release; 1089 } 1090 1091 /* 1092 * Jail may rewrite the destination address, so let it do 1093 * that before we use it. 1094 */ 1095 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); 1096 if (error) 1097 goto release; 1098 1099 /* 1100 * If a local address or port hasn't yet been selected, or if 1101 * the destination address needs to be rewritten due to using 1102 * a special INADDR_ constant, invoke in_pcbconnect_setup() 1103 * to do the heavy lifting. Once a port is selected, we 1104 * commit the binding back to the socket; we also commit the 1105 * binding of the address if in jail. 1106 * 1107 * If we already have a valid binding and we're not 1108 * requesting a destination address rewrite, use a fast path. 1109 */ 1110 if (inp->inp_laddr.s_addr == INADDR_ANY || 1111 inp->inp_lport == 0 || 1112 sin->sin_addr.s_addr == INADDR_ANY || 1113 sin->sin_addr.s_addr == INADDR_BROADCAST) { 1114 INP_INFO_LOCK_ASSERT(&V_udbinfo); 1115 error = in_pcbconnect_setup(inp, addr, &laddr.s_addr, 1116 &lport, &faddr.s_addr, &fport, NULL, 1117 td->td_ucred); 1118 if (error) 1119 goto release; 1120 1121 /* 1122 * XXXRW: Why not commit the port if the address is 1123 * !INADDR_ANY? 1124 */ 1125 /* Commit the local port if newly assigned. */ 1126 if (inp->inp_laddr.s_addr == INADDR_ANY && 1127 inp->inp_lport == 0) { 1128 INP_INFO_WLOCK_ASSERT(&V_udbinfo); 1129 INP_WLOCK_ASSERT(inp); 1130 /* 1131 * Remember addr if jailed, to prevent 1132 * rebinding. 1133 */ 1134 if (prison_flag(td->td_ucred, PR_IP4)) 1135 inp->inp_laddr = laddr; 1136 inp->inp_lport = lport; 1137 if (in_pcbinshash(inp) != 0) { 1138 inp->inp_lport = 0; 1139 error = EAGAIN; 1140 goto release; 1141 } 1142 inp->inp_flags |= INP_ANONPORT; 1143 } 1144 } else { 1145 faddr = sin->sin_addr; 1146 fport = sin->sin_port; 1147 } 1148 } else { 1149 INP_LOCK_ASSERT(inp); 1150 faddr = inp->inp_faddr; 1151 fport = inp->inp_fport; 1152 if (faddr.s_addr == INADDR_ANY) { 1153 error = ENOTCONN; 1154 goto release; 1155 } 1156 } 1157 1158 /* 1159 * Calculate data length and get a mbuf for UDP, IP, and possible 1160 * link-layer headers. Immediate slide the data pointer back forward 1161 * since we won't use that space at this layer. 1162 */ 1163 M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_DONTWAIT); 1164 if (m == NULL) { 1165 error = ENOBUFS; 1166 goto release; 1167 } 1168 m->m_data += max_linkhdr; 1169 m->m_len -= max_linkhdr; 1170 m->m_pkthdr.len -= max_linkhdr; 1171 1172 /* 1173 * Fill in mbuf with extended UDP header and addresses and length put 1174 * into network format. 1175 */ 1176 ui = mtod(m, struct udpiphdr *); 1177 bzero(ui->ui_x1, sizeof(ui->ui_x1)); /* XXX still needed? */ 1178 ui->ui_pr = IPPROTO_UDP; 1179 ui->ui_src = laddr; 1180 ui->ui_dst = faddr; 1181 ui->ui_sport = lport; 1182 ui->ui_dport = fport; 1183 ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr)); 1184 1185 /* 1186 * Set the Don't Fragment bit in the IP header. 1187 */ 1188 if (inp->inp_flags & INP_DONTFRAG) { 1189 struct ip *ip; 1190 1191 ip = (struct ip *)&ui->ui_i; 1192 ip->ip_off |= IP_DF; 1193 } 1194 1195 ipflags = 0; 1196 if (inp->inp_socket->so_options & SO_DONTROUTE) 1197 ipflags |= IP_ROUTETOIF; 1198 if (inp->inp_socket->so_options & SO_BROADCAST) 1199 ipflags |= IP_ALLOWBROADCAST; 1200 if (inp->inp_flags & INP_ONESBCAST) 1201 ipflags |= IP_SENDONES; 1202 1203 #ifdef MAC 1204 mac_inpcb_create_mbuf(inp, m); 1205 #endif 1206 1207 /* 1208 * Set up checksum and output datagram. 1209 */ 1210 if (udp_cksum) { 1211 if (inp->inp_flags & INP_ONESBCAST) 1212 faddr.s_addr = INADDR_BROADCAST; 1213 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr, 1214 htons((u_short)len + sizeof(struct udphdr) + IPPROTO_UDP)); 1215 m->m_pkthdr.csum_flags = CSUM_UDP; 1216 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 1217 } else 1218 ui->ui_sum = 0; 1219 ((struct ip *)ui)->ip_len = sizeof (struct udpiphdr) + len; 1220 ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl; /* XXX */ 1221 ((struct ip *)ui)->ip_tos = inp->inp_ip_tos; /* XXX */ 1222 UDPSTAT_INC(udps_opackets); 1223 1224 if (unlock_udbinfo == 2) 1225 INP_INFO_WUNLOCK(&V_udbinfo); 1226 else if (unlock_udbinfo == 1) 1227 INP_INFO_RUNLOCK(&V_udbinfo); 1228 error = ip_output(m, inp->inp_options, NULL, ipflags, 1229 inp->inp_moptions, inp); 1230 if (unlock_udbinfo == 2) 1231 INP_WUNLOCK(inp); 1232 else 1233 INP_RUNLOCK(inp); 1234 return (error); 1235 1236 release: 1237 if (unlock_udbinfo == 2) { 1238 INP_WUNLOCK(inp); 1239 INP_INFO_WUNLOCK(&V_udbinfo); 1240 } else if (unlock_udbinfo == 1) { 1241 INP_RUNLOCK(inp); 1242 INP_INFO_RUNLOCK(&V_udbinfo); 1243 } else 1244 INP_RUNLOCK(inp); 1245 m_freem(m); 1246 return (error); 1247 } 1248 1249 1250 #if defined(IPSEC) && defined(IPSEC_NAT_T) 1251 #ifdef INET 1252 /* 1253 * Potentially decap ESP in UDP frame. Check for an ESP header 1254 * and optional marker; if present, strip the UDP header and 1255 * push the result through IPSec. 1256 * 1257 * Returns mbuf to be processed (potentially re-allocated) or 1258 * NULL if consumed and/or processed. 1259 */ 1260 static struct mbuf * 1261 udp4_espdecap(struct inpcb *inp, struct mbuf *m, int off) 1262 { 1263 size_t minlen, payload, skip, iphlen; 1264 caddr_t data; 1265 struct udpcb *up; 1266 struct m_tag *tag; 1267 struct udphdr *udphdr; 1268 struct ip *ip; 1269 1270 INP_RLOCK_ASSERT(inp); 1271 1272 /* 1273 * Pull up data so the longest case is contiguous: 1274 * IP/UDP hdr + non ESP marker + ESP hdr. 1275 */ 1276 minlen = off + sizeof(uint64_t) + sizeof(struct esp); 1277 if (minlen > m->m_pkthdr.len) 1278 minlen = m->m_pkthdr.len; 1279 if ((m = m_pullup(m, minlen)) == NULL) { 1280 V_ipsec4stat.in_inval++; 1281 return (NULL); /* Bypass caller processing. */ 1282 } 1283 data = mtod(m, caddr_t); /* Points to ip header. */ 1284 payload = m->m_len - off; /* Size of payload. */ 1285 1286 if (payload == 1 && data[off] == '\xff') 1287 return (m); /* NB: keepalive packet, no decap. */ 1288 1289 up = intoudpcb(inp); 1290 KASSERT(up != NULL, ("%s: udpcb NULL", __func__)); 1291 KASSERT((up->u_flags & UF_ESPINUDP_ALL) != 0, 1292 ("u_flags 0x%x", up->u_flags)); 1293 1294 /* 1295 * Check that the payload is large enough to hold an 1296 * ESP header and compute the amount of data to remove. 1297 * 1298 * NB: the caller has already done a pullup for us. 1299 * XXX can we assume alignment and eliminate bcopys? 1300 */ 1301 if (up->u_flags & UF_ESPINUDP_NON_IKE) { 1302 /* 1303 * draft-ietf-ipsec-nat-t-ike-0[01].txt and 1304 * draft-ietf-ipsec-udp-encaps-(00/)01.txt, ignoring 1305 * possible AH mode non-IKE marker+non-ESP marker 1306 * from draft-ietf-ipsec-udp-encaps-00.txt. 1307 */ 1308 uint64_t marker; 1309 1310 if (payload <= sizeof(uint64_t) + sizeof(struct esp)) 1311 return (m); /* NB: no decap. */ 1312 bcopy(data + off, &marker, sizeof(uint64_t)); 1313 if (marker != 0) /* Non-IKE marker. */ 1314 return (m); /* NB: no decap. */ 1315 skip = sizeof(uint64_t) + sizeof(struct udphdr); 1316 } else { 1317 uint32_t spi; 1318 1319 if (payload <= sizeof(struct esp)) { 1320 V_ipsec4stat.in_inval++; 1321 m_freem(m); 1322 return (NULL); /* Discard. */ 1323 } 1324 bcopy(data + off, &spi, sizeof(uint32_t)); 1325 if (spi == 0) /* Non-ESP marker. */ 1326 return (m); /* NB: no decap. */ 1327 skip = sizeof(struct udphdr); 1328 } 1329 1330 /* 1331 * Setup a PACKET_TAG_IPSEC_NAT_T_PORT tag to remember 1332 * the UDP ports. This is required if we want to select 1333 * the right SPD for multiple hosts behind same NAT. 1334 * 1335 * NB: ports are maintained in network byte order everywhere 1336 * in the NAT-T code. 1337 */ 1338 tag = m_tag_get(PACKET_TAG_IPSEC_NAT_T_PORTS, 1339 2 * sizeof(uint16_t), M_NOWAIT); 1340 if (tag == NULL) { 1341 V_ipsec4stat.in_nomem++; 1342 m_freem(m); 1343 return (NULL); /* Discard. */ 1344 } 1345 iphlen = off - sizeof(struct udphdr); 1346 udphdr = (struct udphdr *)(data + iphlen); 1347 ((uint16_t *)(tag + 1))[0] = udphdr->uh_sport; 1348 ((uint16_t *)(tag + 1))[1] = udphdr->uh_dport; 1349 m_tag_prepend(m, tag); 1350 1351 /* 1352 * Remove the UDP header (and possibly the non ESP marker) 1353 * IP header length is iphlen 1354 * Before: 1355 * <--- off ---> 1356 * +----+------+-----+ 1357 * | IP | UDP | ESP | 1358 * +----+------+-----+ 1359 * <-skip-> 1360 * After: 1361 * +----+-----+ 1362 * | IP | ESP | 1363 * +----+-----+ 1364 * <-skip-> 1365 */ 1366 ovbcopy(data, data + skip, iphlen); 1367 m_adj(m, skip); 1368 1369 ip = mtod(m, struct ip *); 1370 ip->ip_len -= skip; 1371 ip->ip_p = IPPROTO_ESP; 1372 1373 /* 1374 * We cannot yet update the cksums so clear any 1375 * h/w cksum flags as they are no longer valid. 1376 */ 1377 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) 1378 m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID|CSUM_PSEUDO_HDR); 1379 1380 (void) ipsec4_common_input(m, iphlen, ip->ip_p); 1381 return (NULL); /* NB: consumed, bypass processing. */ 1382 } 1383 #endif /* INET */ 1384 #endif /* defined(IPSEC) && defined(IPSEC_NAT_T) */ 1385 1386 static void 1387 udp_abort(struct socket *so) 1388 { 1389 struct inpcb *inp; 1390 1391 inp = sotoinpcb(so); 1392 KASSERT(inp != NULL, ("udp_abort: inp == NULL")); 1393 INP_INFO_WLOCK(&V_udbinfo); 1394 INP_WLOCK(inp); 1395 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1396 in_pcbdisconnect(inp); 1397 inp->inp_laddr.s_addr = INADDR_ANY; 1398 soisdisconnected(so); 1399 } 1400 INP_WUNLOCK(inp); 1401 INP_INFO_WUNLOCK(&V_udbinfo); 1402 } 1403 1404 static int 1405 udp_attach(struct socket *so, int proto, struct thread *td) 1406 { 1407 struct inpcb *inp; 1408 int error; 1409 1410 inp = sotoinpcb(so); 1411 KASSERT(inp == NULL, ("udp_attach: inp != NULL")); 1412 error = soreserve(so, udp_sendspace, udp_recvspace); 1413 if (error) 1414 return (error); 1415 INP_INFO_WLOCK(&V_udbinfo); 1416 error = in_pcballoc(so, &V_udbinfo); 1417 if (error) { 1418 INP_INFO_WUNLOCK(&V_udbinfo); 1419 return (error); 1420 } 1421 1422 inp = sotoinpcb(so); 1423 inp->inp_vflag |= INP_IPV4; 1424 inp->inp_ip_ttl = V_ip_defttl; 1425 1426 error = udp_newudpcb(inp); 1427 if (error) { 1428 in_pcbdetach(inp); 1429 in_pcbfree(inp); 1430 INP_INFO_WUNLOCK(&V_udbinfo); 1431 return (error); 1432 } 1433 1434 INP_WUNLOCK(inp); 1435 INP_INFO_WUNLOCK(&V_udbinfo); 1436 return (0); 1437 } 1438 1439 int 1440 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f) 1441 { 1442 struct inpcb *inp; 1443 struct udpcb *up; 1444 1445 KASSERT(so->so_type == SOCK_DGRAM, 1446 ("udp_set_kernel_tunneling: !dgram")); 1447 inp = sotoinpcb(so); 1448 KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL")); 1449 INP_WLOCK(inp); 1450 up = intoudpcb(inp); 1451 if (up->u_tun_func != NULL) { 1452 INP_WUNLOCK(inp); 1453 return (EBUSY); 1454 } 1455 up->u_tun_func = f; 1456 INP_WUNLOCK(inp); 1457 return (0); 1458 } 1459 1460 static int 1461 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 1462 { 1463 struct inpcb *inp; 1464 int error; 1465 1466 inp = sotoinpcb(so); 1467 KASSERT(inp != NULL, ("udp_bind: inp == NULL")); 1468 INP_INFO_WLOCK(&V_udbinfo); 1469 INP_WLOCK(inp); 1470 error = in_pcbbind(inp, nam, td->td_ucred); 1471 INP_WUNLOCK(inp); 1472 INP_INFO_WUNLOCK(&V_udbinfo); 1473 return (error); 1474 } 1475 1476 static void 1477 udp_close(struct socket *so) 1478 { 1479 struct inpcb *inp; 1480 1481 inp = sotoinpcb(so); 1482 KASSERT(inp != NULL, ("udp_close: inp == NULL")); 1483 INP_INFO_WLOCK(&V_udbinfo); 1484 INP_WLOCK(inp); 1485 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1486 in_pcbdisconnect(inp); 1487 inp->inp_laddr.s_addr = INADDR_ANY; 1488 soisdisconnected(so); 1489 } 1490 INP_WUNLOCK(inp); 1491 INP_INFO_WUNLOCK(&V_udbinfo); 1492 } 1493 1494 static int 1495 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 1496 { 1497 struct inpcb *inp; 1498 int error; 1499 struct sockaddr_in *sin; 1500 1501 inp = sotoinpcb(so); 1502 KASSERT(inp != NULL, ("udp_connect: inp == NULL")); 1503 INP_INFO_WLOCK(&V_udbinfo); 1504 INP_WLOCK(inp); 1505 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1506 INP_WUNLOCK(inp); 1507 INP_INFO_WUNLOCK(&V_udbinfo); 1508 return (EISCONN); 1509 } 1510 sin = (struct sockaddr_in *)nam; 1511 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); 1512 if (error != 0) { 1513 INP_WUNLOCK(inp); 1514 INP_INFO_WUNLOCK(&V_udbinfo); 1515 return (error); 1516 } 1517 error = in_pcbconnect(inp, nam, td->td_ucred); 1518 if (error == 0) 1519 soisconnected(so); 1520 INP_WUNLOCK(inp); 1521 INP_INFO_WUNLOCK(&V_udbinfo); 1522 return (error); 1523 } 1524 1525 static void 1526 udp_detach(struct socket *so) 1527 { 1528 struct inpcb *inp; 1529 struct udpcb *up; 1530 1531 inp = sotoinpcb(so); 1532 KASSERT(inp != NULL, ("udp_detach: inp == NULL")); 1533 KASSERT(inp->inp_faddr.s_addr == INADDR_ANY, 1534 ("udp_detach: not disconnected")); 1535 INP_INFO_WLOCK(&V_udbinfo); 1536 INP_WLOCK(inp); 1537 up = intoudpcb(inp); 1538 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1539 inp->inp_ppcb = NULL; 1540 in_pcbdetach(inp); 1541 in_pcbfree(inp); 1542 INP_INFO_WUNLOCK(&V_udbinfo); 1543 udp_discardcb(up); 1544 } 1545 1546 static int 1547 udp_disconnect(struct socket *so) 1548 { 1549 struct inpcb *inp; 1550 1551 inp = sotoinpcb(so); 1552 KASSERT(inp != NULL, ("udp_disconnect: inp == NULL")); 1553 INP_INFO_WLOCK(&V_udbinfo); 1554 INP_WLOCK(inp); 1555 if (inp->inp_faddr.s_addr == INADDR_ANY) { 1556 INP_WUNLOCK(inp); 1557 INP_INFO_WUNLOCK(&V_udbinfo); 1558 return (ENOTCONN); 1559 } 1560 1561 in_pcbdisconnect(inp); 1562 inp->inp_laddr.s_addr = INADDR_ANY; 1563 SOCK_LOCK(so); 1564 so->so_state &= ~SS_ISCONNECTED; /* XXX */ 1565 SOCK_UNLOCK(so); 1566 INP_WUNLOCK(inp); 1567 INP_INFO_WUNLOCK(&V_udbinfo); 1568 return (0); 1569 } 1570 1571 static int 1572 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr, 1573 struct mbuf *control, struct thread *td) 1574 { 1575 struct inpcb *inp; 1576 1577 inp = sotoinpcb(so); 1578 KASSERT(inp != NULL, ("udp_send: inp == NULL")); 1579 return (udp_output(inp, m, addr, control, td)); 1580 } 1581 1582 int 1583 udp_shutdown(struct socket *so) 1584 { 1585 struct inpcb *inp; 1586 1587 inp = sotoinpcb(so); 1588 KASSERT(inp != NULL, ("udp_shutdown: inp == NULL")); 1589 INP_WLOCK(inp); 1590 socantsendmore(so); 1591 INP_WUNLOCK(inp); 1592 return (0); 1593 } 1594 1595 struct pr_usrreqs udp_usrreqs = { 1596 .pru_abort = udp_abort, 1597 .pru_attach = udp_attach, 1598 .pru_bind = udp_bind, 1599 .pru_connect = udp_connect, 1600 .pru_control = in_control, 1601 .pru_detach = udp_detach, 1602 .pru_disconnect = udp_disconnect, 1603 .pru_peeraddr = in_getpeeraddr, 1604 .pru_send = udp_send, 1605 .pru_soreceive = soreceive_dgram, 1606 .pru_sosend = sosend_dgram, 1607 .pru_shutdown = udp_shutdown, 1608 .pru_sockaddr = in_getsockaddr, 1609 .pru_sosetlabel = in_pcbsosetlabel, 1610 .pru_close = udp_close, 1611 }; 1612