udp_usrreq.c revision 274266
1/*- 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 3 * The Regents of the University of California. 4 * Copyright (c) 2008 Robert N. M. Watson 5 * Copyright (c) 2010-2011 Juniper Networks, Inc. 6 * Copyright (c) 2014 Kevin Lo 7 * All rights reserved. 8 * 9 * Portions of this software were developed by Robert N. M. Watson under 10 * contract to Juniper Networks, Inc. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 4. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * @(#)udp_usrreq.c 8.6 (Berkeley) 5/23/95 37 */ 38 39#include <sys/cdefs.h> 40__FBSDID("$FreeBSD: stable/10/sys/netinet/udp_usrreq.c 274266 2014-11-08 02:53:55Z bryanv $"); 41 42#include "opt_ipfw.h" 43#include "opt_inet.h" 44#include "opt_inet6.h" 45#include "opt_ipsec.h" 46#include "opt_kdtrace.h" 47 48#include <sys/param.h> 49#include <sys/domain.h> 50#include <sys/eventhandler.h> 51#include <sys/jail.h> 52#include <sys/kernel.h> 53#include <sys/lock.h> 54#include <sys/malloc.h> 55#include <sys/mbuf.h> 56#include <sys/priv.h> 57#include <sys/proc.h> 58#include <sys/protosw.h> 59#include <sys/sdt.h> 60#include <sys/signalvar.h> 61#include <sys/socket.h> 62#include <sys/socketvar.h> 63#include <sys/sx.h> 64#include <sys/sysctl.h> 65#include <sys/syslog.h> 66#include <sys/systm.h> 67 68#include <vm/uma.h> 69 70#include <net/if.h> 71#include <net/route.h> 72 73#include <netinet/in.h> 74#include <netinet/in_kdtrace.h> 75#include <netinet/in_pcb.h> 76#include <netinet/in_systm.h> 77#include <netinet/in_var.h> 78#include <netinet/ip.h> 79#ifdef INET6 80#include <netinet/ip6.h> 81#endif 82#include <netinet/ip_icmp.h> 83#include <netinet/icmp_var.h> 84#include <netinet/ip_var.h> 85#include <netinet/ip_options.h> 86#ifdef INET6 87#include <netinet6/ip6_var.h> 88#endif 89#include <netinet/udp.h> 90#include <netinet/udp_var.h> 91#include <netinet/udplite.h> 92 93#ifdef IPSEC 94#include <netipsec/ipsec.h> 95#include <netipsec/esp.h> 96#endif 97 98#include <machine/in_cksum.h> 99 100#include <security/mac/mac_framework.h> 101 102/* 103 * UDP and UDP-Lite protocols implementation. 104 * Per RFC 768, August, 1980. 105 * Per RFC 3828, July, 2004. 106 */ 107 108/* 109 * BSD 4.2 defaulted the udp checksum to be off. Turning off udp checksums 110 * removes the only data integrity mechanism for packets and malformed 111 * packets that would otherwise be discarded due to bad checksums, and may 112 * cause problems (especially for NFS data blocks). 113 */ 114VNET_DEFINE(int, udp_cksum) = 1; 115SYSCTL_VNET_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_RW, 116 &VNET_NAME(udp_cksum), 0, "compute udp checksum"); 117 118int udp_log_in_vain = 0; 119SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW, 120 &udp_log_in_vain, 0, "Log all incoming UDP packets"); 121 122VNET_DEFINE(int, udp_blackhole) = 0; 123SYSCTL_VNET_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_RW, 124 &VNET_NAME(udp_blackhole), 0, 125 "Do not send port unreachables for refused connects"); 126 127u_long udp_sendspace = 9216; /* really max datagram size */ 128 /* 40 1K datagrams */ 129SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW, 130 &udp_sendspace, 0, "Maximum outgoing UDP datagram size"); 131 132u_long udp_recvspace = 40 * (1024 + 133#ifdef INET6 134 sizeof(struct sockaddr_in6) 135#else 136 sizeof(struct sockaddr_in) 137#endif 138 ); 139 140SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW, 141 &udp_recvspace, 0, "Maximum space for incoming UDP datagrams"); 142 143VNET_DEFINE(struct inpcbhead, udb); /* from udp_var.h */ 144VNET_DEFINE(struct inpcbinfo, udbinfo); 145VNET_DEFINE(struct inpcbhead, ulitecb); 146VNET_DEFINE(struct inpcbinfo, ulitecbinfo); 147static VNET_DEFINE(uma_zone_t, udpcb_zone); 148#define V_udpcb_zone VNET(udpcb_zone) 149 150#ifndef UDBHASHSIZE 151#define UDBHASHSIZE 128 152#endif 153 154VNET_PCPUSTAT_DEFINE(struct udpstat, udpstat); /* from udp_var.h */ 155VNET_PCPUSTAT_SYSINIT(udpstat); 156SYSCTL_VNET_PCPUSTAT(_net_inet_udp, UDPCTL_STATS, stats, struct udpstat, 157 udpstat, "UDP statistics (struct udpstat, netinet/udp_var.h)"); 158 159#ifdef VIMAGE 160VNET_PCPUSTAT_SYSUNINIT(udpstat); 161#endif /* VIMAGE */ 162#ifdef INET 163static void udp_detach(struct socket *so); 164static int udp_output(struct inpcb *, struct mbuf *, struct sockaddr *, 165 struct mbuf *, struct thread *); 166#endif 167 168#ifdef IPSEC 169#ifdef IPSEC_NAT_T 170#define UF_ESPINUDP_ALL (UF_ESPINUDP_NON_IKE|UF_ESPINUDP) 171#ifdef INET 172static struct mbuf *udp4_espdecap(struct inpcb *, struct mbuf *, int); 173#endif 174#endif /* IPSEC_NAT_T */ 175#endif /* IPSEC */ 176 177static void 178udp_zone_change(void *tag) 179{ 180 181 uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets); 182 uma_zone_set_max(V_udpcb_zone, maxsockets); 183} 184 185static int 186udp_inpcb_init(void *mem, int size, int flags) 187{ 188 struct inpcb *inp; 189 190 inp = mem; 191 INP_LOCK_INIT(inp, "inp", "udpinp"); 192 return (0); 193} 194 195static int 196udplite_inpcb_init(void *mem, int size, int flags) 197{ 198 struct inpcb *inp; 199 200 inp = mem; 201 INP_LOCK_INIT(inp, "inp", "udpliteinp"); 202 return (0); 203} 204 205void 206udp_init(void) 207{ 208 209 in_pcbinfo_init(&V_udbinfo, "udp", &V_udb, UDBHASHSIZE, UDBHASHSIZE, 210 "udp_inpcb", udp_inpcb_init, NULL, UMA_ZONE_NOFREE, 211 IPI_HASHFIELDS_2TUPLE); 212 V_udpcb_zone = uma_zcreate("udpcb", sizeof(struct udpcb), 213 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 214 uma_zone_set_max(V_udpcb_zone, maxsockets); 215 uma_zone_set_warning(V_udpcb_zone, "kern.ipc.maxsockets limit reached"); 216 EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL, 217 EVENTHANDLER_PRI_ANY); 218} 219 220void 221udplite_init(void) 222{ 223 224 in_pcbinfo_init(&V_ulitecbinfo, "udplite", &V_ulitecb, UDBHASHSIZE, 225 UDBHASHSIZE, "udplite_inpcb", udplite_inpcb_init, NULL, 226 UMA_ZONE_NOFREE, IPI_HASHFIELDS_2TUPLE); 227} 228 229/* 230 * Kernel module interface for updating udpstat. The argument is an index 231 * into udpstat treated as an array of u_long. While this encodes the 232 * general layout of udpstat into the caller, it doesn't encode its location, 233 * so that future changes to add, for example, per-CPU stats support won't 234 * cause binary compatibility problems for kernel modules. 235 */ 236void 237kmod_udpstat_inc(int statnum) 238{ 239 240 counter_u64_add(VNET(udpstat)[statnum], 1); 241} 242 243int 244udp_newudpcb(struct inpcb *inp) 245{ 246 struct udpcb *up; 247 248 up = uma_zalloc(V_udpcb_zone, M_NOWAIT | M_ZERO); 249 if (up == NULL) 250 return (ENOBUFS); 251 inp->inp_ppcb = up; 252 return (0); 253} 254 255void 256udp_discardcb(struct udpcb *up) 257{ 258 259 uma_zfree(V_udpcb_zone, up); 260} 261 262#ifdef VIMAGE 263void 264udp_destroy(void) 265{ 266 267 in_pcbinfo_destroy(&V_udbinfo); 268 uma_zdestroy(V_udpcb_zone); 269} 270 271void 272udplite_destroy(void) 273{ 274 275 in_pcbinfo_destroy(&V_ulitecbinfo); 276} 277#endif 278 279#ifdef INET 280/* 281 * Subroutine of udp_input(), which appends the provided mbuf chain to the 282 * passed pcb/socket. The caller must provide a sockaddr_in via udp_in that 283 * contains the source address. If the socket ends up being an IPv6 socket, 284 * udp_append() will convert to a sockaddr_in6 before passing the address 285 * into the socket code. 286 */ 287static void 288udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off, 289 struct sockaddr_in *udp_in) 290{ 291 struct sockaddr *append_sa; 292 struct socket *so; 293 struct mbuf *opts = 0; 294#ifdef INET6 295 struct sockaddr_in6 udp_in6; 296#endif 297 struct udpcb *up; 298 299 INP_LOCK_ASSERT(inp); 300 301 /* 302 * Engage the tunneling protocol. 303 */ 304 up = intoudpcb(inp); 305 if (up->u_tun_func != NULL) { 306 (*up->u_tun_func)(n, off, inp); 307 return; 308 } 309 310 off += sizeof(struct udphdr); 311 312#ifdef IPSEC 313 /* Check AH/ESP integrity. */ 314 if (ipsec4_in_reject(n, inp)) { 315 m_freem(n); 316 IPSECSTAT_INC(ips_in_polvio); 317 return; 318 } 319#ifdef IPSEC_NAT_T 320 up = intoudpcb(inp); 321 KASSERT(up != NULL, ("%s: udpcb NULL", __func__)); 322 if (up->u_flags & UF_ESPINUDP_ALL) { /* IPSec UDP encaps. */ 323 n = udp4_espdecap(inp, n, off); 324 if (n == NULL) /* Consumed. */ 325 return; 326 } 327#endif /* IPSEC_NAT_T */ 328#endif /* IPSEC */ 329#ifdef MAC 330 if (mac_inpcb_check_deliver(inp, n) != 0) { 331 m_freem(n); 332 return; 333 } 334#endif /* MAC */ 335 if (inp->inp_flags & INP_CONTROLOPTS || 336 inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) { 337#ifdef INET6 338 if (inp->inp_vflag & INP_IPV6) 339 (void)ip6_savecontrol_v4(inp, n, &opts, NULL); 340 else 341#endif /* INET6 */ 342 ip_savecontrol(inp, &opts, ip, n); 343 } 344#ifdef INET6 345 if (inp->inp_vflag & INP_IPV6) { 346 bzero(&udp_in6, sizeof(udp_in6)); 347 udp_in6.sin6_len = sizeof(udp_in6); 348 udp_in6.sin6_family = AF_INET6; 349 in6_sin_2_v4mapsin6(udp_in, &udp_in6); 350 append_sa = (struct sockaddr *)&udp_in6; 351 } else 352#endif /* INET6 */ 353 append_sa = (struct sockaddr *)udp_in; 354 m_adj(n, off); 355 356 so = inp->inp_socket; 357 SOCKBUF_LOCK(&so->so_rcv); 358 if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) { 359 SOCKBUF_UNLOCK(&so->so_rcv); 360 m_freem(n); 361 if (opts) 362 m_freem(opts); 363 UDPSTAT_INC(udps_fullsock); 364 } else 365 sorwakeup_locked(so); 366} 367 368void 369udp_input(struct mbuf *m, int off) 370{ 371 int iphlen = off; 372 struct ip *ip; 373 struct udphdr *uh; 374 struct ifnet *ifp; 375 struct inpcb *inp; 376 uint16_t len, ip_len; 377 struct inpcbinfo *pcbinfo; 378 struct ip save_ip; 379 struct sockaddr_in udp_in; 380 struct m_tag *fwd_tag; 381 int cscov_partial; 382 uint8_t pr; 383 384 ifp = m->m_pkthdr.rcvif; 385 UDPSTAT_INC(udps_ipackets); 386 387 /* 388 * Strip IP options, if any; should skip this, make available to 389 * user, and use on returned packets, but we don't yet have a way to 390 * check the checksum with options still present. 391 */ 392 if (iphlen > sizeof (struct ip)) { 393 ip_stripoptions(m); 394 iphlen = sizeof(struct ip); 395 } 396 397 /* 398 * Get IP and UDP header together in first mbuf. 399 */ 400 ip = mtod(m, struct ip *); 401 if (m->m_len < iphlen + sizeof(struct udphdr)) { 402 if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == NULL) { 403 UDPSTAT_INC(udps_hdrops); 404 return; 405 } 406 ip = mtod(m, struct ip *); 407 } 408 uh = (struct udphdr *)((caddr_t)ip + iphlen); 409 pr = ip->ip_p; 410 cscov_partial = (pr == IPPROTO_UDPLITE) ? 1 : 0; 411 412 /* 413 * Destination port of 0 is illegal, based on RFC768. 414 */ 415 if (uh->uh_dport == 0) 416 goto badunlocked; 417 418 /* 419 * Construct sockaddr format source address. Stuff source address 420 * and datagram in user buffer. 421 */ 422 bzero(&udp_in, sizeof(udp_in)); 423 udp_in.sin_len = sizeof(udp_in); 424 udp_in.sin_family = AF_INET; 425 udp_in.sin_port = uh->uh_sport; 426 udp_in.sin_addr = ip->ip_src; 427 428 /* 429 * Make mbuf data length reflect UDP length. If not enough data to 430 * reflect UDP length, drop. 431 */ 432 len = ntohs((u_short)uh->uh_ulen); 433 ip_len = ntohs(ip->ip_len) - iphlen; 434 if (pr == IPPROTO_UDPLITE && (len == 0 || len == ip_len)) { 435 /* Zero means checksum over the complete packet. */ 436 if (len == 0) 437 len = ip_len; 438 cscov_partial = 0; 439 } 440 if (ip_len != len) { 441 if (len > ip_len || len < sizeof(struct udphdr)) { 442 UDPSTAT_INC(udps_badlen); 443 goto badunlocked; 444 } 445 if (pr == IPPROTO_UDP) 446 m_adj(m, len - ip_len); 447 } 448 449 /* 450 * Save a copy of the IP header in case we want restore it for 451 * sending an ICMP error message in response. 452 */ 453 if (!V_udp_blackhole) 454 save_ip = *ip; 455 else 456 memset(&save_ip, 0, sizeof(save_ip)); 457 458 /* 459 * Checksum extended UDP header and data. 460 */ 461 if (uh->uh_sum) { 462 u_short uh_sum; 463 464 if ((m->m_pkthdr.csum_flags & CSUM_DATA_VALID) && 465 !cscov_partial) { 466 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 467 uh_sum = m->m_pkthdr.csum_data; 468 else 469 uh_sum = in_pseudo(ip->ip_src.s_addr, 470 ip->ip_dst.s_addr, htonl((u_short)len + 471 m->m_pkthdr.csum_data + pr)); 472 uh_sum ^= 0xffff; 473 } else { 474 char b[9]; 475 476 bcopy(((struct ipovly *)ip)->ih_x1, b, 9); 477 bzero(((struct ipovly *)ip)->ih_x1, 9); 478 ((struct ipovly *)ip)->ih_len = (pr == IPPROTO_UDP) ? 479 uh->uh_ulen : htons(ip_len); 480 uh_sum = in_cksum(m, len + sizeof (struct ip)); 481 bcopy(b, ((struct ipovly *)ip)->ih_x1, 9); 482 } 483 if (uh_sum) { 484 UDPSTAT_INC(udps_badsum); 485 m_freem(m); 486 return; 487 } 488 } else { 489 if (pr == IPPROTO_UDP) { 490 UDPSTAT_INC(udps_nosum); 491 } else { 492 /* UDPLite requires a checksum */ 493 /* XXX: What is the right UDPLite MIB counter here? */ 494 m_freem(m); 495 return; 496 } 497 } 498 499 pcbinfo = get_inpcbinfo(pr); 500 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 501 in_broadcast(ip->ip_dst, ifp)) { 502 struct inpcb *last; 503 struct inpcbhead *pcblist; 504 struct ip_moptions *imo; 505 506 INP_INFO_RLOCK(pcbinfo); 507 pcblist = get_pcblist(pr); 508 last = NULL; 509 LIST_FOREACH(inp, pcblist, inp_list) { 510 if (inp->inp_lport != uh->uh_dport) 511 continue; 512#ifdef INET6 513 if ((inp->inp_vflag & INP_IPV4) == 0) 514 continue; 515#endif 516 if (inp->inp_laddr.s_addr != INADDR_ANY && 517 inp->inp_laddr.s_addr != ip->ip_dst.s_addr) 518 continue; 519 if (inp->inp_faddr.s_addr != INADDR_ANY && 520 inp->inp_faddr.s_addr != ip->ip_src.s_addr) 521 continue; 522 if (inp->inp_fport != 0 && 523 inp->inp_fport != uh->uh_sport) 524 continue; 525 526 INP_RLOCK(inp); 527 528 /* 529 * XXXRW: Because we weren't holding either the inpcb 530 * or the hash lock when we checked for a match 531 * before, we should probably recheck now that the 532 * inpcb lock is held. 533 */ 534 535 /* 536 * Handle socket delivery policy for any-source 537 * and source-specific multicast. [RFC3678] 538 */ 539 imo = inp->inp_moptions; 540 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) { 541 struct sockaddr_in group; 542 int blocked; 543 if (imo == NULL) { 544 INP_RUNLOCK(inp); 545 continue; 546 } 547 bzero(&group, sizeof(struct sockaddr_in)); 548 group.sin_len = sizeof(struct sockaddr_in); 549 group.sin_family = AF_INET; 550 group.sin_addr = ip->ip_dst; 551 552 blocked = imo_multi_filter(imo, ifp, 553 (struct sockaddr *)&group, 554 (struct sockaddr *)&udp_in); 555 if (blocked != MCAST_PASS) { 556 if (blocked == MCAST_NOTGMEMBER) 557 IPSTAT_INC(ips_notmember); 558 if (blocked == MCAST_NOTSMEMBER || 559 blocked == MCAST_MUTED) 560 UDPSTAT_INC(udps_filtermcast); 561 INP_RUNLOCK(inp); 562 continue; 563 } 564 } 565 if (last != NULL) { 566 struct mbuf *n; 567 568 if ((n = m_copy(m, 0, M_COPYALL)) != NULL) { 569 UDP_PROBE(receive, NULL, last, ip, 570 last, uh); 571 udp_append(last, ip, n, iphlen, 572 &udp_in); 573 } 574 INP_RUNLOCK(last); 575 } 576 last = inp; 577 /* 578 * Don't look for additional matches if this one does 579 * not have either the SO_REUSEPORT or SO_REUSEADDR 580 * socket options set. This heuristic avoids 581 * searching through all pcbs in the common case of a 582 * non-shared port. It assumes that an application 583 * will never clear these options after setting them. 584 */ 585 if ((last->inp_socket->so_options & 586 (SO_REUSEPORT|SO_REUSEADDR)) == 0) 587 break; 588 } 589 590 if (last == NULL) { 591 /* 592 * No matching pcb found; discard datagram. (No need 593 * to send an ICMP Port Unreachable for a broadcast 594 * or multicast datgram.) 595 */ 596 UDPSTAT_INC(udps_noportbcast); 597 if (inp) 598 INP_RUNLOCK(inp); 599 INP_INFO_RUNLOCK(pcbinfo); 600 goto badunlocked; 601 } 602 UDP_PROBE(receive, NULL, last, ip, last, uh); 603 udp_append(last, ip, m, iphlen, &udp_in); 604 INP_RUNLOCK(last); 605 INP_INFO_RUNLOCK(pcbinfo); 606 return; 607 } 608 609 /* 610 * Locate pcb for datagram. 611 */ 612 613 /* 614 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 615 */ 616 if ((m->m_flags & M_IP_NEXTHOP) && 617 (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) { 618 struct sockaddr_in *next_hop; 619 620 next_hop = (struct sockaddr_in *)(fwd_tag + 1); 621 622 /* 623 * Transparently forwarded. Pretend to be the destination. 624 * Already got one like this? 625 */ 626 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport, 627 ip->ip_dst, uh->uh_dport, INPLOOKUP_RLOCKPCB, ifp, m); 628 if (!inp) { 629 /* 630 * It's new. Try to find the ambushing socket. 631 * Because we've rewritten the destination address, 632 * any hardware-generated hash is ignored. 633 */ 634 inp = in_pcblookup(pcbinfo, ip->ip_src, 635 uh->uh_sport, next_hop->sin_addr, 636 next_hop->sin_port ? htons(next_hop->sin_port) : 637 uh->uh_dport, INPLOOKUP_WILDCARD | 638 INPLOOKUP_RLOCKPCB, ifp); 639 } 640 /* Remove the tag from the packet. We don't need it anymore. */ 641 m_tag_delete(m, fwd_tag); 642 m->m_flags &= ~M_IP_NEXTHOP; 643 } else 644 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport, 645 ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD | 646 INPLOOKUP_RLOCKPCB, ifp, m); 647 if (inp == NULL) { 648 if (udp_log_in_vain) { 649 char buf[4*sizeof "123"]; 650 651 strcpy(buf, inet_ntoa(ip->ip_dst)); 652 log(LOG_INFO, 653 "Connection attempt to UDP %s:%d from %s:%d\n", 654 buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src), 655 ntohs(uh->uh_sport)); 656 } 657 UDPSTAT_INC(udps_noport); 658 if (m->m_flags & (M_BCAST | M_MCAST)) { 659 UDPSTAT_INC(udps_noportbcast); 660 goto badunlocked; 661 } 662 if (V_udp_blackhole) 663 goto badunlocked; 664 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0) 665 goto badunlocked; 666 *ip = save_ip; 667 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0); 668 return; 669 } 670 671 /* 672 * Check the minimum TTL for socket. 673 */ 674 INP_RLOCK_ASSERT(inp); 675 if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) { 676 INP_RUNLOCK(inp); 677 m_freem(m); 678 return; 679 } 680 if (cscov_partial) { 681 struct udpcb *up; 682 683 up = intoudpcb(inp); 684 if (up->u_rxcslen == 0 || up->u_rxcslen > len) { 685 INP_RUNLOCK(inp); 686 m_freem(m); 687 return; 688 } 689 } 690 691 UDP_PROBE(receive, NULL, inp, ip, inp, uh); 692 udp_append(inp, ip, m, iphlen, &udp_in); 693 INP_RUNLOCK(inp); 694 return; 695 696badunlocked: 697 m_freem(m); 698} 699#endif /* INET */ 700 701/* 702 * Notify a udp user of an asynchronous error; just wake up so that they can 703 * collect error status. 704 */ 705struct inpcb * 706udp_notify(struct inpcb *inp, int errno) 707{ 708 709 /* 710 * While udp_ctlinput() always calls udp_notify() with a read lock 711 * when invoking it directly, in_pcbnotifyall() currently uses write 712 * locks due to sharing code with TCP. For now, accept either a read 713 * or a write lock, but a read lock is sufficient. 714 */ 715 INP_LOCK_ASSERT(inp); 716 717 inp->inp_socket->so_error = errno; 718 sorwakeup(inp->inp_socket); 719 sowwakeup(inp->inp_socket); 720 return (inp); 721} 722 723#ifdef INET 724static void 725udp_common_ctlinput(int cmd, struct sockaddr *sa, void *vip, 726 struct inpcbinfo *pcbinfo) 727{ 728 struct ip *ip = vip; 729 struct udphdr *uh; 730 struct in_addr faddr; 731 struct inpcb *inp; 732 733 faddr = ((struct sockaddr_in *)sa)->sin_addr; 734 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 735 return; 736 737 /* 738 * Redirects don't need to be handled up here. 739 */ 740 if (PRC_IS_REDIRECT(cmd)) 741 return; 742 743 /* 744 * Hostdead is ugly because it goes linearly through all PCBs. 745 * 746 * XXX: We never get this from ICMP, otherwise it makes an excellent 747 * DoS attack on machines with many connections. 748 */ 749 if (cmd == PRC_HOSTDEAD) 750 ip = NULL; 751 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 752 return; 753 if (ip != NULL) { 754 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 755 inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport, 756 ip->ip_src, uh->uh_sport, INPLOOKUP_RLOCKPCB, NULL); 757 if (inp != NULL) { 758 INP_RLOCK_ASSERT(inp); 759 if (inp->inp_socket != NULL) { 760 udp_notify(inp, inetctlerrmap[cmd]); 761 } 762 INP_RUNLOCK(inp); 763 } 764 } else 765 in_pcbnotifyall(pcbinfo, faddr, inetctlerrmap[cmd], 766 udp_notify); 767} 768void 769udp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 770{ 771 772 return (udp_common_ctlinput(cmd, sa, vip, &V_udbinfo)); 773} 774 775void 776udplite_ctlinput(int cmd, struct sockaddr *sa, void *vip) 777{ 778 779 return (udp_common_ctlinput(cmd, sa, vip, &V_ulitecbinfo)); 780} 781#endif /* INET */ 782 783static int 784udp_pcblist(SYSCTL_HANDLER_ARGS) 785{ 786 int error, i, n; 787 struct inpcb *inp, **inp_list; 788 inp_gen_t gencnt; 789 struct xinpgen xig; 790 791 /* 792 * The process of preparing the PCB list is too time-consuming and 793 * resource-intensive to repeat twice on every request. 794 */ 795 if (req->oldptr == 0) { 796 n = V_udbinfo.ipi_count; 797 n += imax(n / 8, 10); 798 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb); 799 return (0); 800 } 801 802 if (req->newptr != 0) 803 return (EPERM); 804 805 /* 806 * OK, now we're committed to doing something. 807 */ 808 INP_INFO_RLOCK(&V_udbinfo); 809 gencnt = V_udbinfo.ipi_gencnt; 810 n = V_udbinfo.ipi_count; 811 INP_INFO_RUNLOCK(&V_udbinfo); 812 813 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 814 + n * sizeof(struct xinpcb)); 815 if (error != 0) 816 return (error); 817 818 xig.xig_len = sizeof xig; 819 xig.xig_count = n; 820 xig.xig_gen = gencnt; 821 xig.xig_sogen = so_gencnt; 822 error = SYSCTL_OUT(req, &xig, sizeof xig); 823 if (error) 824 return (error); 825 826 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 827 if (inp_list == 0) 828 return (ENOMEM); 829 830 INP_INFO_RLOCK(&V_udbinfo); 831 for (inp = LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n; 832 inp = LIST_NEXT(inp, inp_list)) { 833 INP_WLOCK(inp); 834 if (inp->inp_gencnt <= gencnt && 835 cr_canseeinpcb(req->td->td_ucred, inp) == 0) { 836 in_pcbref(inp); 837 inp_list[i++] = inp; 838 } 839 INP_WUNLOCK(inp); 840 } 841 INP_INFO_RUNLOCK(&V_udbinfo); 842 n = i; 843 844 error = 0; 845 for (i = 0; i < n; i++) { 846 inp = inp_list[i]; 847 INP_RLOCK(inp); 848 if (inp->inp_gencnt <= gencnt) { 849 struct xinpcb xi; 850 851 bzero(&xi, sizeof(xi)); 852 xi.xi_len = sizeof xi; 853 /* XXX should avoid extra copy */ 854 bcopy(inp, &xi.xi_inp, sizeof *inp); 855 if (inp->inp_socket) 856 sotoxsocket(inp->inp_socket, &xi.xi_socket); 857 xi.xi_inp.inp_gencnt = inp->inp_gencnt; 858 INP_RUNLOCK(inp); 859 error = SYSCTL_OUT(req, &xi, sizeof xi); 860 } else 861 INP_RUNLOCK(inp); 862 } 863 INP_INFO_WLOCK(&V_udbinfo); 864 for (i = 0; i < n; i++) { 865 inp = inp_list[i]; 866 INP_RLOCK(inp); 867 if (!in_pcbrele_rlocked(inp)) 868 INP_RUNLOCK(inp); 869 } 870 INP_INFO_WUNLOCK(&V_udbinfo); 871 872 if (!error) { 873 /* 874 * Give the user an updated idea of our state. If the 875 * generation differs from what we told her before, she knows 876 * that something happened while we were processing this 877 * request, and it might be necessary to retry. 878 */ 879 INP_INFO_RLOCK(&V_udbinfo); 880 xig.xig_gen = V_udbinfo.ipi_gencnt; 881 xig.xig_sogen = so_gencnt; 882 xig.xig_count = V_udbinfo.ipi_count; 883 INP_INFO_RUNLOCK(&V_udbinfo); 884 error = SYSCTL_OUT(req, &xig, sizeof xig); 885 } 886 free(inp_list, M_TEMP); 887 return (error); 888} 889 890SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, 891 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0, 892 udp_pcblist, "S,xinpcb", "List of active UDP sockets"); 893 894#ifdef INET 895static int 896udp_getcred(SYSCTL_HANDLER_ARGS) 897{ 898 struct xucred xuc; 899 struct sockaddr_in addrs[2]; 900 struct inpcb *inp; 901 int error; 902 903 error = priv_check(req->td, PRIV_NETINET_GETCRED); 904 if (error) 905 return (error); 906 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 907 if (error) 908 return (error); 909 inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port, 910 addrs[0].sin_addr, addrs[0].sin_port, 911 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL); 912 if (inp != NULL) { 913 INP_RLOCK_ASSERT(inp); 914 if (inp->inp_socket == NULL) 915 error = ENOENT; 916 if (error == 0) 917 error = cr_canseeinpcb(req->td->td_ucred, inp); 918 if (error == 0) 919 cru2x(inp->inp_cred, &xuc); 920 INP_RUNLOCK(inp); 921 } else 922 error = ENOENT; 923 if (error == 0) 924 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 925 return (error); 926} 927 928SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred, 929 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 930 udp_getcred, "S,xucred", "Get the xucred of a UDP connection"); 931#endif /* INET */ 932 933int 934udp_ctloutput(struct socket *so, struct sockopt *sopt) 935{ 936 struct inpcb *inp; 937 struct udpcb *up; 938 int isudplite, error, optval; 939 940 error = 0; 941 isudplite = (so->so_proto->pr_protocol == IPPROTO_UDPLITE) ? 1 : 0; 942 inp = sotoinpcb(so); 943 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 944 INP_WLOCK(inp); 945 if (sopt->sopt_level != so->so_proto->pr_protocol) { 946#ifdef INET6 947 if (INP_CHECK_SOCKAF(so, AF_INET6)) { 948 INP_WUNLOCK(inp); 949 error = ip6_ctloutput(so, sopt); 950 } 951#endif 952#if defined(INET) && defined(INET6) 953 else 954#endif 955#ifdef INET 956 { 957 INP_WUNLOCK(inp); 958 error = ip_ctloutput(so, sopt); 959 } 960#endif 961 return (error); 962 } 963 964 switch (sopt->sopt_dir) { 965 case SOPT_SET: 966 switch (sopt->sopt_name) { 967 case UDP_ENCAP: 968 INP_WUNLOCK(inp); 969 error = sooptcopyin(sopt, &optval, sizeof optval, 970 sizeof optval); 971 if (error) 972 break; 973 inp = sotoinpcb(so); 974 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 975 INP_WLOCK(inp); 976#ifdef IPSEC_NAT_T 977 up = intoudpcb(inp); 978 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 979#endif 980 switch (optval) { 981 case 0: 982 /* Clear all UDP encap. */ 983#ifdef IPSEC_NAT_T 984 up->u_flags &= ~UF_ESPINUDP_ALL; 985#endif 986 break; 987#ifdef IPSEC_NAT_T 988 case UDP_ENCAP_ESPINUDP: 989 case UDP_ENCAP_ESPINUDP_NON_IKE: 990 up->u_flags &= ~UF_ESPINUDP_ALL; 991 if (optval == UDP_ENCAP_ESPINUDP) 992 up->u_flags |= UF_ESPINUDP; 993 else if (optval == UDP_ENCAP_ESPINUDP_NON_IKE) 994 up->u_flags |= UF_ESPINUDP_NON_IKE; 995 break; 996#endif 997 default: 998 error = EINVAL; 999 break; 1000 } 1001 INP_WUNLOCK(inp); 1002 break; 1003 case UDPLITE_SEND_CSCOV: 1004 case UDPLITE_RECV_CSCOV: 1005 if (!isudplite) { 1006 INP_WUNLOCK(inp); 1007 error = ENOPROTOOPT; 1008 break; 1009 } 1010 INP_WUNLOCK(inp); 1011 error = sooptcopyin(sopt, &optval, sizeof(optval), 1012 sizeof(optval)); 1013 if (error != 0) 1014 break; 1015 inp = sotoinpcb(so); 1016 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 1017 INP_WLOCK(inp); 1018 up = intoudpcb(inp); 1019 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1020 if ((optval != 0 && optval < 8) || (optval > 65535)) { 1021 INP_WUNLOCK(inp); 1022 error = EINVAL; 1023 break; 1024 } 1025 if (sopt->sopt_name == UDPLITE_SEND_CSCOV) 1026 up->u_txcslen = optval; 1027 else 1028 up->u_rxcslen = optval; 1029 INP_WUNLOCK(inp); 1030 break; 1031 default: 1032 INP_WUNLOCK(inp); 1033 error = ENOPROTOOPT; 1034 break; 1035 } 1036 break; 1037 case SOPT_GET: 1038 switch (sopt->sopt_name) { 1039#ifdef IPSEC_NAT_T 1040 case UDP_ENCAP: 1041 up = intoudpcb(inp); 1042 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1043 optval = up->u_flags & UF_ESPINUDP_ALL; 1044 INP_WUNLOCK(inp); 1045 error = sooptcopyout(sopt, &optval, sizeof optval); 1046 break; 1047#endif 1048 case UDPLITE_SEND_CSCOV: 1049 case UDPLITE_RECV_CSCOV: 1050 if (!isudplite) { 1051 INP_WUNLOCK(inp); 1052 error = ENOPROTOOPT; 1053 break; 1054 } 1055 up = intoudpcb(inp); 1056 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1057 if (sopt->sopt_name == UDPLITE_SEND_CSCOV) 1058 optval = up->u_txcslen; 1059 else 1060 optval = up->u_rxcslen; 1061 INP_WUNLOCK(inp); 1062 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1063 break; 1064 default: 1065 INP_WUNLOCK(inp); 1066 error = ENOPROTOOPT; 1067 break; 1068 } 1069 break; 1070 } 1071 return (error); 1072} 1073 1074#ifdef INET 1075#define UH_WLOCKED 2 1076#define UH_RLOCKED 1 1077#define UH_UNLOCKED 0 1078static int 1079udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr, 1080 struct mbuf *control, struct thread *td) 1081{ 1082 struct udpiphdr *ui; 1083 int len = m->m_pkthdr.len; 1084 struct in_addr faddr, laddr; 1085 struct cmsghdr *cm; 1086 struct inpcbinfo *pcbinfo; 1087 struct sockaddr_in *sin, src; 1088 int cscov_partial = 0; 1089 int error = 0; 1090 int ipflags; 1091 u_short fport, lport; 1092 int unlock_udbinfo; 1093 u_char tos; 1094 uint8_t pr; 1095 uint16_t cscov = 0; 1096 1097 /* 1098 * udp_output() may need to temporarily bind or connect the current 1099 * inpcb. As such, we don't know up front whether we will need the 1100 * pcbinfo lock or not. Do any work to decide what is needed up 1101 * front before acquiring any locks. 1102 */ 1103 if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) { 1104 if (control) 1105 m_freem(control); 1106 m_freem(m); 1107 return (EMSGSIZE); 1108 } 1109 1110 src.sin_family = 0; 1111 INP_RLOCK(inp); 1112 tos = inp->inp_ip_tos; 1113 if (control != NULL) { 1114 /* 1115 * XXX: Currently, we assume all the optional information is 1116 * stored in a single mbuf. 1117 */ 1118 if (control->m_next) { 1119 INP_RUNLOCK(inp); 1120 m_freem(control); 1121 m_freem(m); 1122 return (EINVAL); 1123 } 1124 for (; control->m_len > 0; 1125 control->m_data += CMSG_ALIGN(cm->cmsg_len), 1126 control->m_len -= CMSG_ALIGN(cm->cmsg_len)) { 1127 cm = mtod(control, struct cmsghdr *); 1128 if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0 1129 || cm->cmsg_len > control->m_len) { 1130 error = EINVAL; 1131 break; 1132 } 1133 if (cm->cmsg_level != IPPROTO_IP) 1134 continue; 1135 1136 switch (cm->cmsg_type) { 1137 case IP_SENDSRCADDR: 1138 if (cm->cmsg_len != 1139 CMSG_LEN(sizeof(struct in_addr))) { 1140 error = EINVAL; 1141 break; 1142 } 1143 bzero(&src, sizeof(src)); 1144 src.sin_family = AF_INET; 1145 src.sin_len = sizeof(src); 1146 src.sin_port = inp->inp_lport; 1147 src.sin_addr = 1148 *(struct in_addr *)CMSG_DATA(cm); 1149 break; 1150 1151 case IP_TOS: 1152 if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) { 1153 error = EINVAL; 1154 break; 1155 } 1156 tos = *(u_char *)CMSG_DATA(cm); 1157 break; 1158 1159 default: 1160 error = ENOPROTOOPT; 1161 break; 1162 } 1163 if (error) 1164 break; 1165 } 1166 m_freem(control); 1167 } 1168 if (error) { 1169 INP_RUNLOCK(inp); 1170 m_freem(m); 1171 return (error); 1172 } 1173 1174 /* 1175 * Depending on whether or not the application has bound or connected 1176 * the socket, we may have to do varying levels of work. The optimal 1177 * case is for a connected UDP socket, as a global lock isn't 1178 * required at all. 1179 * 1180 * In order to decide which we need, we require stability of the 1181 * inpcb binding, which we ensure by acquiring a read lock on the 1182 * inpcb. This doesn't strictly follow the lock order, so we play 1183 * the trylock and retry game; note that we may end up with more 1184 * conservative locks than required the second time around, so later 1185 * assertions have to accept that. Further analysis of the number of 1186 * misses under contention is required. 1187 * 1188 * XXXRW: Check that hash locking update here is correct. 1189 */ 1190 pr = inp->inp_socket->so_proto->pr_protocol; 1191 pcbinfo = get_inpcbinfo(pr); 1192 sin = (struct sockaddr_in *)addr; 1193 if (sin != NULL && 1194 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) { 1195 INP_RUNLOCK(inp); 1196 INP_WLOCK(inp); 1197 INP_HASH_WLOCK(pcbinfo); 1198 unlock_udbinfo = UH_WLOCKED; 1199 } else if ((sin != NULL && ( 1200 (sin->sin_addr.s_addr == INADDR_ANY) || 1201 (sin->sin_addr.s_addr == INADDR_BROADCAST) || 1202 (inp->inp_laddr.s_addr == INADDR_ANY) || 1203 (inp->inp_lport == 0))) || 1204 (src.sin_family == AF_INET)) { 1205 INP_HASH_RLOCK(pcbinfo); 1206 unlock_udbinfo = UH_RLOCKED; 1207 } else 1208 unlock_udbinfo = UH_UNLOCKED; 1209 1210 /* 1211 * If the IP_SENDSRCADDR control message was specified, override the 1212 * source address for this datagram. Its use is invalidated if the 1213 * address thus specified is incomplete or clobbers other inpcbs. 1214 */ 1215 laddr = inp->inp_laddr; 1216 lport = inp->inp_lport; 1217 if (src.sin_family == AF_INET) { 1218 INP_HASH_LOCK_ASSERT(pcbinfo); 1219 if ((lport == 0) || 1220 (laddr.s_addr == INADDR_ANY && 1221 src.sin_addr.s_addr == INADDR_ANY)) { 1222 error = EINVAL; 1223 goto release; 1224 } 1225 error = in_pcbbind_setup(inp, (struct sockaddr *)&src, 1226 &laddr.s_addr, &lport, td->td_ucred); 1227 if (error) 1228 goto release; 1229 } 1230 1231 /* 1232 * If a UDP socket has been connected, then a local address/port will 1233 * have been selected and bound. 1234 * 1235 * If a UDP socket has not been connected to, then an explicit 1236 * destination address must be used, in which case a local 1237 * address/port may not have been selected and bound. 1238 */ 1239 if (sin != NULL) { 1240 INP_LOCK_ASSERT(inp); 1241 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1242 error = EISCONN; 1243 goto release; 1244 } 1245 1246 /* 1247 * Jail may rewrite the destination address, so let it do 1248 * that before we use it. 1249 */ 1250 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); 1251 if (error) 1252 goto release; 1253 1254 /* 1255 * If a local address or port hasn't yet been selected, or if 1256 * the destination address needs to be rewritten due to using 1257 * a special INADDR_ constant, invoke in_pcbconnect_setup() 1258 * to do the heavy lifting. Once a port is selected, we 1259 * commit the binding back to the socket; we also commit the 1260 * binding of the address if in jail. 1261 * 1262 * If we already have a valid binding and we're not 1263 * requesting a destination address rewrite, use a fast path. 1264 */ 1265 if (inp->inp_laddr.s_addr == INADDR_ANY || 1266 inp->inp_lport == 0 || 1267 sin->sin_addr.s_addr == INADDR_ANY || 1268 sin->sin_addr.s_addr == INADDR_BROADCAST) { 1269 INP_HASH_LOCK_ASSERT(pcbinfo); 1270 error = in_pcbconnect_setup(inp, addr, &laddr.s_addr, 1271 &lport, &faddr.s_addr, &fport, NULL, 1272 td->td_ucred); 1273 if (error) 1274 goto release; 1275 1276 /* 1277 * XXXRW: Why not commit the port if the address is 1278 * !INADDR_ANY? 1279 */ 1280 /* Commit the local port if newly assigned. */ 1281 if (inp->inp_laddr.s_addr == INADDR_ANY && 1282 inp->inp_lport == 0) { 1283 INP_WLOCK_ASSERT(inp); 1284 INP_HASH_WLOCK_ASSERT(pcbinfo); 1285 /* 1286 * Remember addr if jailed, to prevent 1287 * rebinding. 1288 */ 1289 if (prison_flag(td->td_ucred, PR_IP4)) 1290 inp->inp_laddr = laddr; 1291 inp->inp_lport = lport; 1292 if (in_pcbinshash(inp) != 0) { 1293 inp->inp_lport = 0; 1294 error = EAGAIN; 1295 goto release; 1296 } 1297 inp->inp_flags |= INP_ANONPORT; 1298 } 1299 } else { 1300 faddr = sin->sin_addr; 1301 fport = sin->sin_port; 1302 } 1303 } else { 1304 INP_LOCK_ASSERT(inp); 1305 faddr = inp->inp_faddr; 1306 fport = inp->inp_fport; 1307 if (faddr.s_addr == INADDR_ANY) { 1308 error = ENOTCONN; 1309 goto release; 1310 } 1311 } 1312 1313 /* 1314 * Calculate data length and get a mbuf for UDP, IP, and possible 1315 * link-layer headers. Immediate slide the data pointer back forward 1316 * since we won't use that space at this layer. 1317 */ 1318 M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT); 1319 if (m == NULL) { 1320 error = ENOBUFS; 1321 goto release; 1322 } 1323 m->m_data += max_linkhdr; 1324 m->m_len -= max_linkhdr; 1325 m->m_pkthdr.len -= max_linkhdr; 1326 1327 /* 1328 * Fill in mbuf with extended UDP header and addresses and length put 1329 * into network format. 1330 */ 1331 ui = mtod(m, struct udpiphdr *); 1332 bzero(ui->ui_x1, sizeof(ui->ui_x1)); /* XXX still needed? */ 1333 ui->ui_pr = pr; 1334 ui->ui_src = laddr; 1335 ui->ui_dst = faddr; 1336 ui->ui_sport = lport; 1337 ui->ui_dport = fport; 1338 ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr)); 1339 if (pr == IPPROTO_UDPLITE) { 1340 struct udpcb *up; 1341 uint16_t plen; 1342 1343 up = intoudpcb(inp); 1344 cscov = up->u_txcslen; 1345 plen = (u_short)len + sizeof(struct udphdr); 1346 if (cscov >= plen) 1347 cscov = 0; 1348 ui->ui_len = htons(plen); 1349 ui->ui_ulen = htons(cscov); 1350 /* 1351 * For UDP-Lite, checksum coverage length of zero means 1352 * the entire UDPLite packet is covered by the checksum. 1353 */ 1354 cscov_partial = (cscov == 0) ? 0 : 1; 1355 } else 1356 ui->ui_v = IPVERSION << 4; 1357 1358 /* 1359 * Set the Don't Fragment bit in the IP header. 1360 */ 1361 if (inp->inp_flags & INP_DONTFRAG) { 1362 struct ip *ip; 1363 1364 ip = (struct ip *)&ui->ui_i; 1365 ip->ip_off |= htons(IP_DF); 1366 } 1367 1368 ipflags = 0; 1369 if (inp->inp_socket->so_options & SO_DONTROUTE) 1370 ipflags |= IP_ROUTETOIF; 1371 if (inp->inp_socket->so_options & SO_BROADCAST) 1372 ipflags |= IP_ALLOWBROADCAST; 1373 if (inp->inp_flags & INP_ONESBCAST) 1374 ipflags |= IP_SENDONES; 1375 1376#ifdef MAC 1377 mac_inpcb_create_mbuf(inp, m); 1378#endif 1379 1380 /* 1381 * Set up checksum and output datagram. 1382 */ 1383 ui->ui_sum = 0; 1384 if (pr == IPPROTO_UDPLITE) { 1385 if (inp->inp_flags & INP_ONESBCAST) 1386 faddr.s_addr = INADDR_BROADCAST; 1387 if (cscov_partial) { 1388 if ((ui->ui_sum = in_cksum(m, sizeof(struct ip) + cscov)) == 0) 1389 ui->ui_sum = 0xffff; 1390 } else { 1391 if ((ui->ui_sum = in_cksum(m, sizeof(struct udpiphdr) + len)) == 0) 1392 ui->ui_sum = 0xffff; 1393 } 1394 } else if (V_udp_cksum) { 1395 if (inp->inp_flags & INP_ONESBCAST) 1396 faddr.s_addr = INADDR_BROADCAST; 1397 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr, 1398 htons((u_short)len + sizeof(struct udphdr) + pr)); 1399 m->m_pkthdr.csum_flags = CSUM_UDP; 1400 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 1401 } 1402 ((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len); 1403 ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl; /* XXX */ 1404 ((struct ip *)ui)->ip_tos = tos; /* XXX */ 1405 UDPSTAT_INC(udps_opackets); 1406 1407 if (unlock_udbinfo == UH_WLOCKED) 1408 INP_HASH_WUNLOCK(pcbinfo); 1409 else if (unlock_udbinfo == UH_RLOCKED) 1410 INP_HASH_RUNLOCK(pcbinfo); 1411 UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u); 1412 error = ip_output(m, inp->inp_options, NULL, ipflags, 1413 inp->inp_moptions, inp); 1414 if (unlock_udbinfo == UH_WLOCKED) 1415 INP_WUNLOCK(inp); 1416 else 1417 INP_RUNLOCK(inp); 1418 return (error); 1419 1420release: 1421 if (unlock_udbinfo == UH_WLOCKED) { 1422 INP_HASH_WUNLOCK(pcbinfo); 1423 INP_WUNLOCK(inp); 1424 } else if (unlock_udbinfo == UH_RLOCKED) { 1425 INP_HASH_RUNLOCK(pcbinfo); 1426 INP_RUNLOCK(inp); 1427 } else 1428 INP_RUNLOCK(inp); 1429 m_freem(m); 1430 return (error); 1431} 1432 1433 1434#if defined(IPSEC) && defined(IPSEC_NAT_T) 1435/* 1436 * Potentially decap ESP in UDP frame. Check for an ESP header 1437 * and optional marker; if present, strip the UDP header and 1438 * push the result through IPSec. 1439 * 1440 * Returns mbuf to be processed (potentially re-allocated) or 1441 * NULL if consumed and/or processed. 1442 */ 1443static struct mbuf * 1444udp4_espdecap(struct inpcb *inp, struct mbuf *m, int off) 1445{ 1446 size_t minlen, payload, skip, iphlen; 1447 caddr_t data; 1448 struct udpcb *up; 1449 struct m_tag *tag; 1450 struct udphdr *udphdr; 1451 struct ip *ip; 1452 1453 INP_RLOCK_ASSERT(inp); 1454 1455 /* 1456 * Pull up data so the longest case is contiguous: 1457 * IP/UDP hdr + non ESP marker + ESP hdr. 1458 */ 1459 minlen = off + sizeof(uint64_t) + sizeof(struct esp); 1460 if (minlen > m->m_pkthdr.len) 1461 minlen = m->m_pkthdr.len; 1462 if ((m = m_pullup(m, minlen)) == NULL) { 1463 IPSECSTAT_INC(ips_in_inval); 1464 return (NULL); /* Bypass caller processing. */ 1465 } 1466 data = mtod(m, caddr_t); /* Points to ip header. */ 1467 payload = m->m_len - off; /* Size of payload. */ 1468 1469 if (payload == 1 && data[off] == '\xff') 1470 return (m); /* NB: keepalive packet, no decap. */ 1471 1472 up = intoudpcb(inp); 1473 KASSERT(up != NULL, ("%s: udpcb NULL", __func__)); 1474 KASSERT((up->u_flags & UF_ESPINUDP_ALL) != 0, 1475 ("u_flags 0x%x", up->u_flags)); 1476 1477 /* 1478 * Check that the payload is large enough to hold an 1479 * ESP header and compute the amount of data to remove. 1480 * 1481 * NB: the caller has already done a pullup for us. 1482 * XXX can we assume alignment and eliminate bcopys? 1483 */ 1484 if (up->u_flags & UF_ESPINUDP_NON_IKE) { 1485 /* 1486 * draft-ietf-ipsec-nat-t-ike-0[01].txt and 1487 * draft-ietf-ipsec-udp-encaps-(00/)01.txt, ignoring 1488 * possible AH mode non-IKE marker+non-ESP marker 1489 * from draft-ietf-ipsec-udp-encaps-00.txt. 1490 */ 1491 uint64_t marker; 1492 1493 if (payload <= sizeof(uint64_t) + sizeof(struct esp)) 1494 return (m); /* NB: no decap. */ 1495 bcopy(data + off, &marker, sizeof(uint64_t)); 1496 if (marker != 0) /* Non-IKE marker. */ 1497 return (m); /* NB: no decap. */ 1498 skip = sizeof(uint64_t) + sizeof(struct udphdr); 1499 } else { 1500 uint32_t spi; 1501 1502 if (payload <= sizeof(struct esp)) { 1503 IPSECSTAT_INC(ips_in_inval); 1504 m_freem(m); 1505 return (NULL); /* Discard. */ 1506 } 1507 bcopy(data + off, &spi, sizeof(uint32_t)); 1508 if (spi == 0) /* Non-ESP marker. */ 1509 return (m); /* NB: no decap. */ 1510 skip = sizeof(struct udphdr); 1511 } 1512 1513 /* 1514 * Setup a PACKET_TAG_IPSEC_NAT_T_PORT tag to remember 1515 * the UDP ports. This is required if we want to select 1516 * the right SPD for multiple hosts behind same NAT. 1517 * 1518 * NB: ports are maintained in network byte order everywhere 1519 * in the NAT-T code. 1520 */ 1521 tag = m_tag_get(PACKET_TAG_IPSEC_NAT_T_PORTS, 1522 2 * sizeof(uint16_t), M_NOWAIT); 1523 if (tag == NULL) { 1524 IPSECSTAT_INC(ips_in_nomem); 1525 m_freem(m); 1526 return (NULL); /* Discard. */ 1527 } 1528 iphlen = off - sizeof(struct udphdr); 1529 udphdr = (struct udphdr *)(data + iphlen); 1530 ((uint16_t *)(tag + 1))[0] = udphdr->uh_sport; 1531 ((uint16_t *)(tag + 1))[1] = udphdr->uh_dport; 1532 m_tag_prepend(m, tag); 1533 1534 /* 1535 * Remove the UDP header (and possibly the non ESP marker) 1536 * IP header length is iphlen 1537 * Before: 1538 * <--- off ---> 1539 * +----+------+-----+ 1540 * | IP | UDP | ESP | 1541 * +----+------+-----+ 1542 * <-skip-> 1543 * After: 1544 * +----+-----+ 1545 * | IP | ESP | 1546 * +----+-----+ 1547 * <-skip-> 1548 */ 1549 ovbcopy(data, data + skip, iphlen); 1550 m_adj(m, skip); 1551 1552 ip = mtod(m, struct ip *); 1553 ip->ip_len = htons(ntohs(ip->ip_len) - skip); 1554 ip->ip_p = IPPROTO_ESP; 1555 1556 /* 1557 * We cannot yet update the cksums so clear any 1558 * h/w cksum flags as they are no longer valid. 1559 */ 1560 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) 1561 m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID|CSUM_PSEUDO_HDR); 1562 1563 (void) ipsec4_common_input(m, iphlen, ip->ip_p); 1564 return (NULL); /* NB: consumed, bypass processing. */ 1565} 1566#endif /* defined(IPSEC) && defined(IPSEC_NAT_T) */ 1567 1568static void 1569udp_abort(struct socket *so) 1570{ 1571 struct inpcb *inp; 1572 struct inpcbinfo *pcbinfo; 1573 1574 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1575 inp = sotoinpcb(so); 1576 KASSERT(inp != NULL, ("udp_abort: inp == NULL")); 1577 INP_WLOCK(inp); 1578 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1579 INP_HASH_WLOCK(pcbinfo); 1580 in_pcbdisconnect(inp); 1581 inp->inp_laddr.s_addr = INADDR_ANY; 1582 INP_HASH_WUNLOCK(pcbinfo); 1583 soisdisconnected(so); 1584 } 1585 INP_WUNLOCK(inp); 1586} 1587 1588static int 1589udp_attach(struct socket *so, int proto, struct thread *td) 1590{ 1591 struct inpcb *inp; 1592 struct inpcbinfo *pcbinfo; 1593 int error; 1594 1595 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1596 inp = sotoinpcb(so); 1597 KASSERT(inp == NULL, ("udp_attach: inp != NULL")); 1598 error = soreserve(so, udp_sendspace, udp_recvspace); 1599 if (error) 1600 return (error); 1601 INP_INFO_WLOCK(pcbinfo); 1602 error = in_pcballoc(so, pcbinfo); 1603 if (error) { 1604 INP_INFO_WUNLOCK(pcbinfo); 1605 return (error); 1606 } 1607 1608 inp = sotoinpcb(so); 1609 inp->inp_vflag |= INP_IPV4; 1610 inp->inp_ip_ttl = V_ip_defttl; 1611 1612 error = udp_newudpcb(inp); 1613 if (error) { 1614 in_pcbdetach(inp); 1615 in_pcbfree(inp); 1616 INP_INFO_WUNLOCK(pcbinfo); 1617 return (error); 1618 } 1619 1620 INP_WUNLOCK(inp); 1621 INP_INFO_WUNLOCK(pcbinfo); 1622 return (0); 1623} 1624#endif /* INET */ 1625 1626int 1627udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f) 1628{ 1629 struct inpcb *inp; 1630 struct udpcb *up; 1631 1632 KASSERT(so->so_type == SOCK_DGRAM, 1633 ("udp_set_kernel_tunneling: !dgram")); 1634 inp = sotoinpcb(so); 1635 KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL")); 1636 INP_WLOCK(inp); 1637 up = intoudpcb(inp); 1638 if (up->u_tun_func != NULL) { 1639 INP_WUNLOCK(inp); 1640 return (EBUSY); 1641 } 1642 up->u_tun_func = f; 1643 INP_WUNLOCK(inp); 1644 return (0); 1645} 1646 1647#ifdef INET 1648static int 1649udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 1650{ 1651 struct inpcb *inp; 1652 struct inpcbinfo *pcbinfo; 1653 int error; 1654 1655 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1656 inp = sotoinpcb(so); 1657 KASSERT(inp != NULL, ("udp_bind: inp == NULL")); 1658 INP_WLOCK(inp); 1659 INP_HASH_WLOCK(pcbinfo); 1660 error = in_pcbbind(inp, nam, td->td_ucred); 1661 INP_HASH_WUNLOCK(pcbinfo); 1662 INP_WUNLOCK(inp); 1663 return (error); 1664} 1665 1666static void 1667udp_close(struct socket *so) 1668{ 1669 struct inpcb *inp; 1670 struct inpcbinfo *pcbinfo; 1671 1672 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1673 inp = sotoinpcb(so); 1674 KASSERT(inp != NULL, ("udp_close: inp == NULL")); 1675 INP_WLOCK(inp); 1676 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1677 INP_HASH_WLOCK(pcbinfo); 1678 in_pcbdisconnect(inp); 1679 inp->inp_laddr.s_addr = INADDR_ANY; 1680 INP_HASH_WUNLOCK(pcbinfo); 1681 soisdisconnected(so); 1682 } 1683 INP_WUNLOCK(inp); 1684} 1685 1686static int 1687udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 1688{ 1689 struct inpcb *inp; 1690 struct inpcbinfo *pcbinfo; 1691 struct sockaddr_in *sin; 1692 int error; 1693 1694 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1695 inp = sotoinpcb(so); 1696 KASSERT(inp != NULL, ("udp_connect: inp == NULL")); 1697 INP_WLOCK(inp); 1698 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1699 INP_WUNLOCK(inp); 1700 return (EISCONN); 1701 } 1702 sin = (struct sockaddr_in *)nam; 1703 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); 1704 if (error != 0) { 1705 INP_WUNLOCK(inp); 1706 return (error); 1707 } 1708 INP_HASH_WLOCK(pcbinfo); 1709 error = in_pcbconnect(inp, nam, td->td_ucred); 1710 INP_HASH_WUNLOCK(pcbinfo); 1711 if (error == 0) 1712 soisconnected(so); 1713 INP_WUNLOCK(inp); 1714 return (error); 1715} 1716 1717static void 1718udp_detach(struct socket *so) 1719{ 1720 struct inpcb *inp; 1721 struct inpcbinfo *pcbinfo; 1722 struct udpcb *up; 1723 1724 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1725 inp = sotoinpcb(so); 1726 KASSERT(inp != NULL, ("udp_detach: inp == NULL")); 1727 KASSERT(inp->inp_faddr.s_addr == INADDR_ANY, 1728 ("udp_detach: not disconnected")); 1729 INP_INFO_WLOCK(pcbinfo); 1730 INP_WLOCK(inp); 1731 up = intoudpcb(inp); 1732 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1733 inp->inp_ppcb = NULL; 1734 in_pcbdetach(inp); 1735 in_pcbfree(inp); 1736 INP_INFO_WUNLOCK(pcbinfo); 1737 udp_discardcb(up); 1738} 1739 1740static int 1741udp_disconnect(struct socket *so) 1742{ 1743 struct inpcb *inp; 1744 struct inpcbinfo *pcbinfo; 1745 1746 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1747 inp = sotoinpcb(so); 1748 KASSERT(inp != NULL, ("udp_disconnect: inp == NULL")); 1749 INP_WLOCK(inp); 1750 if (inp->inp_faddr.s_addr == INADDR_ANY) { 1751 INP_WUNLOCK(inp); 1752 return (ENOTCONN); 1753 } 1754 INP_HASH_WLOCK(pcbinfo); 1755 in_pcbdisconnect(inp); 1756 inp->inp_laddr.s_addr = INADDR_ANY; 1757 INP_HASH_WUNLOCK(pcbinfo); 1758 SOCK_LOCK(so); 1759 so->so_state &= ~SS_ISCONNECTED; /* XXX */ 1760 SOCK_UNLOCK(so); 1761 INP_WUNLOCK(inp); 1762 return (0); 1763} 1764 1765static int 1766udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr, 1767 struct mbuf *control, struct thread *td) 1768{ 1769 struct inpcb *inp; 1770 1771 inp = sotoinpcb(so); 1772 KASSERT(inp != NULL, ("udp_send: inp == NULL")); 1773 return (udp_output(inp, m, addr, control, td)); 1774} 1775#endif /* INET */ 1776 1777int 1778udp_shutdown(struct socket *so) 1779{ 1780 struct inpcb *inp; 1781 1782 inp = sotoinpcb(so); 1783 KASSERT(inp != NULL, ("udp_shutdown: inp == NULL")); 1784 INP_WLOCK(inp); 1785 socantsendmore(so); 1786 INP_WUNLOCK(inp); 1787 return (0); 1788} 1789 1790#ifdef INET 1791struct pr_usrreqs udp_usrreqs = { 1792 .pru_abort = udp_abort, 1793 .pru_attach = udp_attach, 1794 .pru_bind = udp_bind, 1795 .pru_connect = udp_connect, 1796 .pru_control = in_control, 1797 .pru_detach = udp_detach, 1798 .pru_disconnect = udp_disconnect, 1799 .pru_peeraddr = in_getpeeraddr, 1800 .pru_send = udp_send, 1801 .pru_soreceive = soreceive_dgram, 1802 .pru_sosend = sosend_dgram, 1803 .pru_shutdown = udp_shutdown, 1804 .pru_sockaddr = in_getsockaddr, 1805 .pru_sosetlabel = in_pcbsosetlabel, 1806 .pru_close = udp_close, 1807}; 1808#endif /* INET */ 1809