udp_usrreq.c revision 272645
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 272645 2014-10-06 13:32:30Z tuexen $"); 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 if (n == NULL) 311 return; 312 313 off += sizeof(struct udphdr); 314 315#ifdef IPSEC 316 /* Check AH/ESP integrity. */ 317 if (ipsec4_in_reject(n, inp)) { 318 m_freem(n); 319 IPSECSTAT_INC(ips_in_polvio); 320 return; 321 } 322#ifdef IPSEC_NAT_T 323 up = intoudpcb(inp); 324 KASSERT(up != NULL, ("%s: udpcb NULL", __func__)); 325 if (up->u_flags & UF_ESPINUDP_ALL) { /* IPSec UDP encaps. */ 326 n = udp4_espdecap(inp, n, off); 327 if (n == NULL) /* Consumed. */ 328 return; 329 } 330#endif /* IPSEC_NAT_T */ 331#endif /* IPSEC */ 332#ifdef MAC 333 if (mac_inpcb_check_deliver(inp, n) != 0) { 334 m_freem(n); 335 return; 336 } 337#endif /* MAC */ 338 if (inp->inp_flags & INP_CONTROLOPTS || 339 inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) { 340#ifdef INET6 341 if (inp->inp_vflag & INP_IPV6) 342 (void)ip6_savecontrol_v4(inp, n, &opts, NULL); 343 else 344#endif /* INET6 */ 345 ip_savecontrol(inp, &opts, ip, n); 346 } 347#ifdef INET6 348 if (inp->inp_vflag & INP_IPV6) { 349 bzero(&udp_in6, sizeof(udp_in6)); 350 udp_in6.sin6_len = sizeof(udp_in6); 351 udp_in6.sin6_family = AF_INET6; 352 in6_sin_2_v4mapsin6(udp_in, &udp_in6); 353 append_sa = (struct sockaddr *)&udp_in6; 354 } else 355#endif /* INET6 */ 356 append_sa = (struct sockaddr *)udp_in; 357 m_adj(n, off); 358 359 so = inp->inp_socket; 360 SOCKBUF_LOCK(&so->so_rcv); 361 if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) { 362 SOCKBUF_UNLOCK(&so->so_rcv); 363 m_freem(n); 364 if (opts) 365 m_freem(opts); 366 UDPSTAT_INC(udps_fullsock); 367 } else 368 sorwakeup_locked(so); 369} 370 371void 372udp_input(struct mbuf *m, int off) 373{ 374 int iphlen = off; 375 struct ip *ip; 376 struct udphdr *uh; 377 struct ifnet *ifp; 378 struct inpcb *inp; 379 uint16_t len, ip_len; 380 struct inpcbinfo *pcbinfo; 381 struct ip save_ip; 382 struct sockaddr_in udp_in; 383 struct m_tag *fwd_tag; 384 int cscov_partial; 385 uint8_t pr; 386 387 ifp = m->m_pkthdr.rcvif; 388 UDPSTAT_INC(udps_ipackets); 389 390 /* 391 * Strip IP options, if any; should skip this, make available to 392 * user, and use on returned packets, but we don't yet have a way to 393 * check the checksum with options still present. 394 */ 395 if (iphlen > sizeof (struct ip)) { 396 ip_stripoptions(m); 397 iphlen = sizeof(struct ip); 398 } 399 400 /* 401 * Get IP and UDP header together in first mbuf. 402 */ 403 ip = mtod(m, struct ip *); 404 if (m->m_len < iphlen + sizeof(struct udphdr)) { 405 if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == NULL) { 406 UDPSTAT_INC(udps_hdrops); 407 return; 408 } 409 ip = mtod(m, struct ip *); 410 } 411 uh = (struct udphdr *)((caddr_t)ip + iphlen); 412 pr = ip->ip_p; 413 cscov_partial = (pr == IPPROTO_UDPLITE) ? 1 : 0; 414 415 /* 416 * Destination port of 0 is illegal, based on RFC768. 417 */ 418 if (uh->uh_dport == 0) 419 goto badunlocked; 420 421 /* 422 * Construct sockaddr format source address. Stuff source address 423 * and datagram in user buffer. 424 */ 425 bzero(&udp_in, sizeof(udp_in)); 426 udp_in.sin_len = sizeof(udp_in); 427 udp_in.sin_family = AF_INET; 428 udp_in.sin_port = uh->uh_sport; 429 udp_in.sin_addr = ip->ip_src; 430 431 /* 432 * Make mbuf data length reflect UDP length. If not enough data to 433 * reflect UDP length, drop. 434 */ 435 len = ntohs((u_short)uh->uh_ulen); 436 ip_len = ntohs(ip->ip_len) - iphlen; 437 if (pr == IPPROTO_UDPLITE && (len == 0 || len == ip_len)) { 438 /* Zero means checksum over the complete packet. */ 439 if (len == 0) 440 len = ip_len; 441 cscov_partial = 0; 442 } 443 if (ip_len != len) { 444 if (len > ip_len || len < sizeof(struct udphdr)) { 445 UDPSTAT_INC(udps_badlen); 446 goto badunlocked; 447 } 448 if (pr == IPPROTO_UDP) 449 m_adj(m, len - ip_len); 450 } 451 452 /* 453 * Save a copy of the IP header in case we want restore it for 454 * sending an ICMP error message in response. 455 */ 456 if (!V_udp_blackhole) 457 save_ip = *ip; 458 else 459 memset(&save_ip, 0, sizeof(save_ip)); 460 461 /* 462 * Checksum extended UDP header and data. 463 */ 464 if (uh->uh_sum) { 465 u_short uh_sum; 466 467 if ((m->m_pkthdr.csum_flags & CSUM_DATA_VALID) && 468 !cscov_partial) { 469 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 470 uh_sum = m->m_pkthdr.csum_data; 471 else 472 uh_sum = in_pseudo(ip->ip_src.s_addr, 473 ip->ip_dst.s_addr, htonl((u_short)len + 474 m->m_pkthdr.csum_data + pr)); 475 uh_sum ^= 0xffff; 476 } else { 477 char b[9]; 478 479 bcopy(((struct ipovly *)ip)->ih_x1, b, 9); 480 bzero(((struct ipovly *)ip)->ih_x1, 9); 481 ((struct ipovly *)ip)->ih_len = (pr == IPPROTO_UDP) ? 482 uh->uh_ulen : htons(ip_len); 483 uh_sum = in_cksum(m, len + sizeof (struct ip)); 484 bcopy(b, ((struct ipovly *)ip)->ih_x1, 9); 485 } 486 if (uh_sum) { 487 UDPSTAT_INC(udps_badsum); 488 m_freem(m); 489 return; 490 } 491 } else 492 UDPSTAT_INC(udps_nosum); 493 494 pcbinfo = get_inpcbinfo(pr); 495 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 496 in_broadcast(ip->ip_dst, ifp)) { 497 struct inpcb *last; 498 struct inpcbhead *pcblist; 499 struct ip_moptions *imo; 500 501 INP_INFO_RLOCK(pcbinfo); 502 pcblist = get_pcblist(pr); 503 last = NULL; 504 LIST_FOREACH(inp, pcblist, inp_list) { 505 if (inp->inp_lport != uh->uh_dport) 506 continue; 507#ifdef INET6 508 if ((inp->inp_vflag & INP_IPV4) == 0) 509 continue; 510#endif 511 if (inp->inp_laddr.s_addr != INADDR_ANY && 512 inp->inp_laddr.s_addr != ip->ip_dst.s_addr) 513 continue; 514 if (inp->inp_faddr.s_addr != INADDR_ANY && 515 inp->inp_faddr.s_addr != ip->ip_src.s_addr) 516 continue; 517 if (inp->inp_fport != 0 && 518 inp->inp_fport != uh->uh_sport) 519 continue; 520 521 INP_RLOCK(inp); 522 523 /* 524 * XXXRW: Because we weren't holding either the inpcb 525 * or the hash lock when we checked for a match 526 * before, we should probably recheck now that the 527 * inpcb lock is held. 528 */ 529 530 /* 531 * Handle socket delivery policy for any-source 532 * and source-specific multicast. [RFC3678] 533 */ 534 imo = inp->inp_moptions; 535 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) { 536 struct sockaddr_in group; 537 int blocked; 538 if (imo == NULL) { 539 INP_RUNLOCK(inp); 540 continue; 541 } 542 bzero(&group, sizeof(struct sockaddr_in)); 543 group.sin_len = sizeof(struct sockaddr_in); 544 group.sin_family = AF_INET; 545 group.sin_addr = ip->ip_dst; 546 547 blocked = imo_multi_filter(imo, ifp, 548 (struct sockaddr *)&group, 549 (struct sockaddr *)&udp_in); 550 if (blocked != MCAST_PASS) { 551 if (blocked == MCAST_NOTGMEMBER) 552 IPSTAT_INC(ips_notmember); 553 if (blocked == MCAST_NOTSMEMBER || 554 blocked == MCAST_MUTED) 555 UDPSTAT_INC(udps_filtermcast); 556 INP_RUNLOCK(inp); 557 continue; 558 } 559 } 560 if (last != NULL) { 561 struct mbuf *n; 562 563 n = m_copy(m, 0, M_COPYALL); 564 udp_append(last, ip, n, iphlen, &udp_in); 565 INP_RUNLOCK(last); 566 } 567 last = inp; 568 /* 569 * Don't look for additional matches if this one does 570 * not have either the SO_REUSEPORT or SO_REUSEADDR 571 * socket options set. This heuristic avoids 572 * searching through all pcbs in the common case of a 573 * non-shared port. It assumes that an application 574 * will never clear these options after setting them. 575 */ 576 if ((last->inp_socket->so_options & 577 (SO_REUSEPORT|SO_REUSEADDR)) == 0) 578 break; 579 } 580 581 if (last == NULL) { 582 /* 583 * No matching pcb found; discard datagram. (No need 584 * to send an ICMP Port Unreachable for a broadcast 585 * or multicast datgram.) 586 */ 587 UDPSTAT_INC(udps_noportbcast); 588 if (inp) 589 INP_RUNLOCK(inp); 590 INP_INFO_RUNLOCK(pcbinfo); 591 goto badunlocked; 592 } 593 udp_append(last, ip, m, iphlen, &udp_in); 594 INP_RUNLOCK(last); 595 INP_INFO_RUNLOCK(pcbinfo); 596 return; 597 } 598 599 /* 600 * Locate pcb for datagram. 601 */ 602 603 /* 604 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 605 */ 606 if ((m->m_flags & M_IP_NEXTHOP) && 607 (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) { 608 struct sockaddr_in *next_hop; 609 610 next_hop = (struct sockaddr_in *)(fwd_tag + 1); 611 612 /* 613 * Transparently forwarded. Pretend to be the destination. 614 * Already got one like this? 615 */ 616 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport, 617 ip->ip_dst, uh->uh_dport, INPLOOKUP_RLOCKPCB, ifp, m); 618 if (!inp) { 619 /* 620 * It's new. Try to find the ambushing socket. 621 * Because we've rewritten the destination address, 622 * any hardware-generated hash is ignored. 623 */ 624 inp = in_pcblookup(pcbinfo, ip->ip_src, 625 uh->uh_sport, next_hop->sin_addr, 626 next_hop->sin_port ? htons(next_hop->sin_port) : 627 uh->uh_dport, INPLOOKUP_WILDCARD | 628 INPLOOKUP_RLOCKPCB, ifp); 629 } 630 /* Remove the tag from the packet. We don't need it anymore. */ 631 m_tag_delete(m, fwd_tag); 632 m->m_flags &= ~M_IP_NEXTHOP; 633 } else 634 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport, 635 ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD | 636 INPLOOKUP_RLOCKPCB, ifp, m); 637 if (inp == NULL) { 638 if (udp_log_in_vain) { 639 char buf[4*sizeof "123"]; 640 641 strcpy(buf, inet_ntoa(ip->ip_dst)); 642 log(LOG_INFO, 643 "Connection attempt to UDP %s:%d from %s:%d\n", 644 buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src), 645 ntohs(uh->uh_sport)); 646 } 647 UDPSTAT_INC(udps_noport); 648 if (m->m_flags & (M_BCAST | M_MCAST)) { 649 UDPSTAT_INC(udps_noportbcast); 650 goto badunlocked; 651 } 652 if (V_udp_blackhole) 653 goto badunlocked; 654 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0) 655 goto badunlocked; 656 *ip = save_ip; 657 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0); 658 return; 659 } 660 661 /* 662 * Check the minimum TTL for socket. 663 */ 664 INP_RLOCK_ASSERT(inp); 665 if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) { 666 INP_RUNLOCK(inp); 667 m_freem(m); 668 return; 669 } 670 if (cscov_partial) { 671 struct udpcb *up; 672 673 up = intoudpcb(inp); 674 if (up->u_rxcslen > len) { 675 INP_RUNLOCK(inp); 676 m_freem(m); 677 return; 678 } 679 } 680 681 UDP_PROBE(receive, NULL, inp, ip, inp, uh); 682 udp_append(inp, ip, m, iphlen, &udp_in); 683 INP_RUNLOCK(inp); 684 return; 685 686badunlocked: 687 m_freem(m); 688} 689#endif /* INET */ 690 691/* 692 * Notify a udp user of an asynchronous error; just wake up so that they can 693 * collect error status. 694 */ 695struct inpcb * 696udp_notify(struct inpcb *inp, int errno) 697{ 698 699 /* 700 * While udp_ctlinput() always calls udp_notify() with a read lock 701 * when invoking it directly, in_pcbnotifyall() currently uses write 702 * locks due to sharing code with TCP. For now, accept either a read 703 * or a write lock, but a read lock is sufficient. 704 */ 705 INP_LOCK_ASSERT(inp); 706 707 inp->inp_socket->so_error = errno; 708 sorwakeup(inp->inp_socket); 709 sowwakeup(inp->inp_socket); 710 return (inp); 711} 712 713#ifdef INET 714static void 715udp_common_ctlinput(int cmd, struct sockaddr *sa, void *vip, 716 struct inpcbinfo *pcbinfo) 717{ 718 struct ip *ip = vip; 719 struct udphdr *uh; 720 struct in_addr faddr; 721 struct inpcb *inp; 722 723 faddr = ((struct sockaddr_in *)sa)->sin_addr; 724 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 725 return; 726 727 /* 728 * Redirects don't need to be handled up here. 729 */ 730 if (PRC_IS_REDIRECT(cmd)) 731 return; 732 733 /* 734 * Hostdead is ugly because it goes linearly through all PCBs. 735 * 736 * XXX: We never get this from ICMP, otherwise it makes an excellent 737 * DoS attack on machines with many connections. 738 */ 739 if (cmd == PRC_HOSTDEAD) 740 ip = NULL; 741 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 742 return; 743 if (ip != NULL) { 744 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 745 inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport, 746 ip->ip_src, uh->uh_sport, INPLOOKUP_RLOCKPCB, NULL); 747 if (inp != NULL) { 748 INP_RLOCK_ASSERT(inp); 749 if (inp->inp_socket != NULL) { 750 udp_notify(inp, inetctlerrmap[cmd]); 751 } 752 INP_RUNLOCK(inp); 753 } 754 } else 755 in_pcbnotifyall(pcbinfo, faddr, inetctlerrmap[cmd], 756 udp_notify); 757} 758void 759udp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 760{ 761 762 return (udp_common_ctlinput(cmd, sa, vip, &V_udbinfo)); 763} 764 765void 766udplite_ctlinput(int cmd, struct sockaddr *sa, void *vip) 767{ 768 769 return (udp_common_ctlinput(cmd, sa, vip, &V_ulitecbinfo)); 770} 771#endif /* INET */ 772 773static int 774udp_pcblist(SYSCTL_HANDLER_ARGS) 775{ 776 int error, i, n; 777 struct inpcb *inp, **inp_list; 778 inp_gen_t gencnt; 779 struct xinpgen xig; 780 781 /* 782 * The process of preparing the PCB list is too time-consuming and 783 * resource-intensive to repeat twice on every request. 784 */ 785 if (req->oldptr == 0) { 786 n = V_udbinfo.ipi_count; 787 n += imax(n / 8, 10); 788 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb); 789 return (0); 790 } 791 792 if (req->newptr != 0) 793 return (EPERM); 794 795 /* 796 * OK, now we're committed to doing something. 797 */ 798 INP_INFO_RLOCK(&V_udbinfo); 799 gencnt = V_udbinfo.ipi_gencnt; 800 n = V_udbinfo.ipi_count; 801 INP_INFO_RUNLOCK(&V_udbinfo); 802 803 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 804 + n * sizeof(struct xinpcb)); 805 if (error != 0) 806 return (error); 807 808 xig.xig_len = sizeof xig; 809 xig.xig_count = n; 810 xig.xig_gen = gencnt; 811 xig.xig_sogen = so_gencnt; 812 error = SYSCTL_OUT(req, &xig, sizeof xig); 813 if (error) 814 return (error); 815 816 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 817 if (inp_list == 0) 818 return (ENOMEM); 819 820 INP_INFO_RLOCK(&V_udbinfo); 821 for (inp = LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n; 822 inp = LIST_NEXT(inp, inp_list)) { 823 INP_WLOCK(inp); 824 if (inp->inp_gencnt <= gencnt && 825 cr_canseeinpcb(req->td->td_ucred, inp) == 0) { 826 in_pcbref(inp); 827 inp_list[i++] = inp; 828 } 829 INP_WUNLOCK(inp); 830 } 831 INP_INFO_RUNLOCK(&V_udbinfo); 832 n = i; 833 834 error = 0; 835 for (i = 0; i < n; i++) { 836 inp = inp_list[i]; 837 INP_RLOCK(inp); 838 if (inp->inp_gencnt <= gencnt) { 839 struct xinpcb xi; 840 841 bzero(&xi, sizeof(xi)); 842 xi.xi_len = sizeof xi; 843 /* XXX should avoid extra copy */ 844 bcopy(inp, &xi.xi_inp, sizeof *inp); 845 if (inp->inp_socket) 846 sotoxsocket(inp->inp_socket, &xi.xi_socket); 847 xi.xi_inp.inp_gencnt = inp->inp_gencnt; 848 INP_RUNLOCK(inp); 849 error = SYSCTL_OUT(req, &xi, sizeof xi); 850 } else 851 INP_RUNLOCK(inp); 852 } 853 INP_INFO_WLOCK(&V_udbinfo); 854 for (i = 0; i < n; i++) { 855 inp = inp_list[i]; 856 INP_RLOCK(inp); 857 if (!in_pcbrele_rlocked(inp)) 858 INP_RUNLOCK(inp); 859 } 860 INP_INFO_WUNLOCK(&V_udbinfo); 861 862 if (!error) { 863 /* 864 * Give the user an updated idea of our state. If the 865 * generation differs from what we told her before, she knows 866 * that something happened while we were processing this 867 * request, and it might be necessary to retry. 868 */ 869 INP_INFO_RLOCK(&V_udbinfo); 870 xig.xig_gen = V_udbinfo.ipi_gencnt; 871 xig.xig_sogen = so_gencnt; 872 xig.xig_count = V_udbinfo.ipi_count; 873 INP_INFO_RUNLOCK(&V_udbinfo); 874 error = SYSCTL_OUT(req, &xig, sizeof xig); 875 } 876 free(inp_list, M_TEMP); 877 return (error); 878} 879 880SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, 881 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0, 882 udp_pcblist, "S,xinpcb", "List of active UDP sockets"); 883 884#ifdef INET 885static int 886udp_getcred(SYSCTL_HANDLER_ARGS) 887{ 888 struct xucred xuc; 889 struct sockaddr_in addrs[2]; 890 struct inpcb *inp; 891 int error; 892 893 error = priv_check(req->td, PRIV_NETINET_GETCRED); 894 if (error) 895 return (error); 896 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 897 if (error) 898 return (error); 899 inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port, 900 addrs[0].sin_addr, addrs[0].sin_port, 901 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL); 902 if (inp != NULL) { 903 INP_RLOCK_ASSERT(inp); 904 if (inp->inp_socket == NULL) 905 error = ENOENT; 906 if (error == 0) 907 error = cr_canseeinpcb(req->td->td_ucred, inp); 908 if (error == 0) 909 cru2x(inp->inp_cred, &xuc); 910 INP_RUNLOCK(inp); 911 } else 912 error = ENOENT; 913 if (error == 0) 914 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 915 return (error); 916} 917 918SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred, 919 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 920 udp_getcred, "S,xucred", "Get the xucred of a UDP connection"); 921#endif /* INET */ 922 923int 924udp_ctloutput(struct socket *so, struct sockopt *sopt) 925{ 926 struct inpcb *inp; 927 struct udpcb *up; 928 int isudplite, error, optval; 929 930 error = 0; 931 isudplite = (so->so_proto->pr_protocol == IPPROTO_UDPLITE) ? 1 : 0; 932 inp = sotoinpcb(so); 933 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 934 INP_WLOCK(inp); 935 if (sopt->sopt_level != so->so_proto->pr_protocol) { 936#ifdef INET6 937 if (INP_CHECK_SOCKAF(so, AF_INET6)) { 938 INP_WUNLOCK(inp); 939 error = ip6_ctloutput(so, sopt); 940 } 941#endif 942#if defined(INET) && defined(INET6) 943 else 944#endif 945#ifdef INET 946 { 947 INP_WUNLOCK(inp); 948 error = ip_ctloutput(so, sopt); 949 } 950#endif 951 return (error); 952 } 953 954 switch (sopt->sopt_dir) { 955 case SOPT_SET: 956 switch (sopt->sopt_name) { 957 case UDP_ENCAP: 958 INP_WUNLOCK(inp); 959 error = sooptcopyin(sopt, &optval, sizeof optval, 960 sizeof optval); 961 if (error) 962 break; 963 inp = sotoinpcb(so); 964 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 965 INP_WLOCK(inp); 966#ifdef IPSEC_NAT_T 967 up = intoudpcb(inp); 968 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 969#endif 970 switch (optval) { 971 case 0: 972 /* Clear all UDP encap. */ 973#ifdef IPSEC_NAT_T 974 up->u_flags &= ~UF_ESPINUDP_ALL; 975#endif 976 break; 977#ifdef IPSEC_NAT_T 978 case UDP_ENCAP_ESPINUDP: 979 case UDP_ENCAP_ESPINUDP_NON_IKE: 980 up->u_flags &= ~UF_ESPINUDP_ALL; 981 if (optval == UDP_ENCAP_ESPINUDP) 982 up->u_flags |= UF_ESPINUDP; 983 else if (optval == UDP_ENCAP_ESPINUDP_NON_IKE) 984 up->u_flags |= UF_ESPINUDP_NON_IKE; 985 break; 986#endif 987 default: 988 error = EINVAL; 989 break; 990 } 991 INP_WUNLOCK(inp); 992 break; 993 case UDPLITE_SEND_CSCOV: 994 case UDPLITE_RECV_CSCOV: 995 if (!isudplite) { 996 INP_WUNLOCK(inp); 997 error = ENOPROTOOPT; 998 break; 999 } 1000 INP_WUNLOCK(inp); 1001 error = sooptcopyin(sopt, &optval, sizeof(optval), 1002 sizeof(optval)); 1003 if (error != 0) 1004 break; 1005 inp = sotoinpcb(so); 1006 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 1007 INP_WLOCK(inp); 1008 up = intoudpcb(inp); 1009 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1010 if ((optval != 0 && optval < 8) || (optval > 65535)) { 1011 INP_WUNLOCK(inp); 1012 error = EINVAL; 1013 break; 1014 } 1015 if (sopt->sopt_name == UDPLITE_SEND_CSCOV) 1016 up->u_txcslen = optval; 1017 else 1018 up->u_rxcslen = optval; 1019 INP_WUNLOCK(inp); 1020 break; 1021 default: 1022 INP_WUNLOCK(inp); 1023 error = ENOPROTOOPT; 1024 break; 1025 } 1026 break; 1027 case SOPT_GET: 1028 switch (sopt->sopt_name) { 1029#ifdef IPSEC_NAT_T 1030 case UDP_ENCAP: 1031 up = intoudpcb(inp); 1032 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1033 optval = up->u_flags & UF_ESPINUDP_ALL; 1034 INP_WUNLOCK(inp); 1035 error = sooptcopyout(sopt, &optval, sizeof optval); 1036 break; 1037#endif 1038 case UDPLITE_SEND_CSCOV: 1039 case UDPLITE_RECV_CSCOV: 1040 if (!isudplite) { 1041 INP_WUNLOCK(inp); 1042 error = ENOPROTOOPT; 1043 break; 1044 } 1045 up = intoudpcb(inp); 1046 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1047 if (sopt->sopt_name == UDPLITE_SEND_CSCOV) 1048 optval = up->u_txcslen; 1049 else 1050 optval = up->u_rxcslen; 1051 INP_WUNLOCK(inp); 1052 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1053 break; 1054 default: 1055 INP_WUNLOCK(inp); 1056 error = ENOPROTOOPT; 1057 break; 1058 } 1059 break; 1060 } 1061 return (error); 1062} 1063 1064#ifdef INET 1065#define UH_WLOCKED 2 1066#define UH_RLOCKED 1 1067#define UH_UNLOCKED 0 1068static int 1069udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr, 1070 struct mbuf *control, struct thread *td) 1071{ 1072 struct udpiphdr *ui; 1073 int len = m->m_pkthdr.len; 1074 struct in_addr faddr, laddr; 1075 struct cmsghdr *cm; 1076 struct inpcbinfo *pcbinfo; 1077 struct sockaddr_in *sin, src; 1078 int cscov_partial = 0; 1079 int error = 0; 1080 int ipflags; 1081 u_short fport, lport; 1082 int unlock_udbinfo; 1083 u_char tos; 1084 uint8_t pr; 1085 uint16_t cscov = 0; 1086 1087 /* 1088 * udp_output() may need to temporarily bind or connect the current 1089 * inpcb. As such, we don't know up front whether we will need the 1090 * pcbinfo lock or not. Do any work to decide what is needed up 1091 * front before acquiring any locks. 1092 */ 1093 if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) { 1094 if (control) 1095 m_freem(control); 1096 m_freem(m); 1097 return (EMSGSIZE); 1098 } 1099 1100 src.sin_family = 0; 1101 INP_RLOCK(inp); 1102 tos = inp->inp_ip_tos; 1103 if (control != NULL) { 1104 /* 1105 * XXX: Currently, we assume all the optional information is 1106 * stored in a single mbuf. 1107 */ 1108 if (control->m_next) { 1109 INP_RUNLOCK(inp); 1110 m_freem(control); 1111 m_freem(m); 1112 return (EINVAL); 1113 } 1114 for (; control->m_len > 0; 1115 control->m_data += CMSG_ALIGN(cm->cmsg_len), 1116 control->m_len -= CMSG_ALIGN(cm->cmsg_len)) { 1117 cm = mtod(control, struct cmsghdr *); 1118 if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0 1119 || cm->cmsg_len > control->m_len) { 1120 error = EINVAL; 1121 break; 1122 } 1123 if (cm->cmsg_level != IPPROTO_IP) 1124 continue; 1125 1126 switch (cm->cmsg_type) { 1127 case IP_SENDSRCADDR: 1128 if (cm->cmsg_len != 1129 CMSG_LEN(sizeof(struct in_addr))) { 1130 error = EINVAL; 1131 break; 1132 } 1133 bzero(&src, sizeof(src)); 1134 src.sin_family = AF_INET; 1135 src.sin_len = sizeof(src); 1136 src.sin_port = inp->inp_lport; 1137 src.sin_addr = 1138 *(struct in_addr *)CMSG_DATA(cm); 1139 break; 1140 1141 case IP_TOS: 1142 if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) { 1143 error = EINVAL; 1144 break; 1145 } 1146 tos = *(u_char *)CMSG_DATA(cm); 1147 break; 1148 1149 default: 1150 error = ENOPROTOOPT; 1151 break; 1152 } 1153 if (error) 1154 break; 1155 } 1156 m_freem(control); 1157 } 1158 if (error) { 1159 INP_RUNLOCK(inp); 1160 m_freem(m); 1161 return (error); 1162 } 1163 1164 /* 1165 * Depending on whether or not the application has bound or connected 1166 * the socket, we may have to do varying levels of work. The optimal 1167 * case is for a connected UDP socket, as a global lock isn't 1168 * required at all. 1169 * 1170 * In order to decide which we need, we require stability of the 1171 * inpcb binding, which we ensure by acquiring a read lock on the 1172 * inpcb. This doesn't strictly follow the lock order, so we play 1173 * the trylock and retry game; note that we may end up with more 1174 * conservative locks than required the second time around, so later 1175 * assertions have to accept that. Further analysis of the number of 1176 * misses under contention is required. 1177 * 1178 * XXXRW: Check that hash locking update here is correct. 1179 */ 1180 pr = inp->inp_socket->so_proto->pr_protocol; 1181 pcbinfo = get_inpcbinfo(pr); 1182 sin = (struct sockaddr_in *)addr; 1183 if (sin != NULL && 1184 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) { 1185 INP_RUNLOCK(inp); 1186 INP_WLOCK(inp); 1187 INP_HASH_WLOCK(pcbinfo); 1188 unlock_udbinfo = UH_WLOCKED; 1189 } else if ((sin != NULL && ( 1190 (sin->sin_addr.s_addr == INADDR_ANY) || 1191 (sin->sin_addr.s_addr == INADDR_BROADCAST) || 1192 (inp->inp_laddr.s_addr == INADDR_ANY) || 1193 (inp->inp_lport == 0))) || 1194 (src.sin_family == AF_INET)) { 1195 INP_HASH_RLOCK(pcbinfo); 1196 unlock_udbinfo = UH_RLOCKED; 1197 } else 1198 unlock_udbinfo = UH_UNLOCKED; 1199 1200 /* 1201 * If the IP_SENDSRCADDR control message was specified, override the 1202 * source address for this datagram. Its use is invalidated if the 1203 * address thus specified is incomplete or clobbers other inpcbs. 1204 */ 1205 laddr = inp->inp_laddr; 1206 lport = inp->inp_lport; 1207 if (src.sin_family == AF_INET) { 1208 INP_HASH_LOCK_ASSERT(pcbinfo); 1209 if ((lport == 0) || 1210 (laddr.s_addr == INADDR_ANY && 1211 src.sin_addr.s_addr == INADDR_ANY)) { 1212 error = EINVAL; 1213 goto release; 1214 } 1215 error = in_pcbbind_setup(inp, (struct sockaddr *)&src, 1216 &laddr.s_addr, &lport, td->td_ucred); 1217 if (error) 1218 goto release; 1219 } 1220 1221 /* 1222 * If a UDP socket has been connected, then a local address/port will 1223 * have been selected and bound. 1224 * 1225 * If a UDP socket has not been connected to, then an explicit 1226 * destination address must be used, in which case a local 1227 * address/port may not have been selected and bound. 1228 */ 1229 if (sin != NULL) { 1230 INP_LOCK_ASSERT(inp); 1231 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1232 error = EISCONN; 1233 goto release; 1234 } 1235 1236 /* 1237 * Jail may rewrite the destination address, so let it do 1238 * that before we use it. 1239 */ 1240 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); 1241 if (error) 1242 goto release; 1243 1244 /* 1245 * If a local address or port hasn't yet been selected, or if 1246 * the destination address needs to be rewritten due to using 1247 * a special INADDR_ constant, invoke in_pcbconnect_setup() 1248 * to do the heavy lifting. Once a port is selected, we 1249 * commit the binding back to the socket; we also commit the 1250 * binding of the address if in jail. 1251 * 1252 * If we already have a valid binding and we're not 1253 * requesting a destination address rewrite, use a fast path. 1254 */ 1255 if (inp->inp_laddr.s_addr == INADDR_ANY || 1256 inp->inp_lport == 0 || 1257 sin->sin_addr.s_addr == INADDR_ANY || 1258 sin->sin_addr.s_addr == INADDR_BROADCAST) { 1259 INP_HASH_LOCK_ASSERT(pcbinfo); 1260 error = in_pcbconnect_setup(inp, addr, &laddr.s_addr, 1261 &lport, &faddr.s_addr, &fport, NULL, 1262 td->td_ucred); 1263 if (error) 1264 goto release; 1265 1266 /* 1267 * XXXRW: Why not commit the port if the address is 1268 * !INADDR_ANY? 1269 */ 1270 /* Commit the local port if newly assigned. */ 1271 if (inp->inp_laddr.s_addr == INADDR_ANY && 1272 inp->inp_lport == 0) { 1273 INP_WLOCK_ASSERT(inp); 1274 INP_HASH_WLOCK_ASSERT(pcbinfo); 1275 /* 1276 * Remember addr if jailed, to prevent 1277 * rebinding. 1278 */ 1279 if (prison_flag(td->td_ucred, PR_IP4)) 1280 inp->inp_laddr = laddr; 1281 inp->inp_lport = lport; 1282 if (in_pcbinshash(inp) != 0) { 1283 inp->inp_lport = 0; 1284 error = EAGAIN; 1285 goto release; 1286 } 1287 inp->inp_flags |= INP_ANONPORT; 1288 } 1289 } else { 1290 faddr = sin->sin_addr; 1291 fport = sin->sin_port; 1292 } 1293 } else { 1294 INP_LOCK_ASSERT(inp); 1295 faddr = inp->inp_faddr; 1296 fport = inp->inp_fport; 1297 if (faddr.s_addr == INADDR_ANY) { 1298 error = ENOTCONN; 1299 goto release; 1300 } 1301 } 1302 1303 /* 1304 * Calculate data length and get a mbuf for UDP, IP, and possible 1305 * link-layer headers. Immediate slide the data pointer back forward 1306 * since we won't use that space at this layer. 1307 */ 1308 M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT); 1309 if (m == NULL) { 1310 error = ENOBUFS; 1311 goto release; 1312 } 1313 m->m_data += max_linkhdr; 1314 m->m_len -= max_linkhdr; 1315 m->m_pkthdr.len -= max_linkhdr; 1316 1317 /* 1318 * Fill in mbuf with extended UDP header and addresses and length put 1319 * into network format. 1320 */ 1321 ui = mtod(m, struct udpiphdr *); 1322 bzero(ui->ui_x1, sizeof(ui->ui_x1)); /* XXX still needed? */ 1323 ui->ui_pr = pr; 1324 ui->ui_src = laddr; 1325 ui->ui_dst = faddr; 1326 ui->ui_sport = lport; 1327 ui->ui_dport = fport; 1328 ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr)); 1329 if (pr == IPPROTO_UDPLITE) { 1330 struct udpcb *up; 1331 uint16_t plen; 1332 1333 up = intoudpcb(inp); 1334 cscov = up->u_txcslen; 1335 plen = (u_short)len + sizeof(struct udphdr); 1336 if (cscov >= plen) 1337 cscov = 0; 1338 ui->ui_len = htons(plen); 1339 ui->ui_ulen = htons(cscov); 1340 /* 1341 * For UDP-Lite, checksum coverage length of zero means 1342 * the entire UDPLite packet is covered by the checksum. 1343 */ 1344 cscov_partial = (cscov == 0) ? 0 : 1; 1345 } else 1346 ui->ui_v = IPVERSION << 4; 1347 1348 /* 1349 * Set the Don't Fragment bit in the IP header. 1350 */ 1351 if (inp->inp_flags & INP_DONTFRAG) { 1352 struct ip *ip; 1353 1354 ip = (struct ip *)&ui->ui_i; 1355 ip->ip_off |= htons(IP_DF); 1356 } 1357 1358 ipflags = 0; 1359 if (inp->inp_socket->so_options & SO_DONTROUTE) 1360 ipflags |= IP_ROUTETOIF; 1361 if (inp->inp_socket->so_options & SO_BROADCAST) 1362 ipflags |= IP_ALLOWBROADCAST; 1363 if (inp->inp_flags & INP_ONESBCAST) 1364 ipflags |= IP_SENDONES; 1365 1366#ifdef MAC 1367 mac_inpcb_create_mbuf(inp, m); 1368#endif 1369 1370 /* 1371 * Set up checksum and output datagram. 1372 */ 1373 ui->ui_sum = 0; 1374 if (pr == IPPROTO_UDPLITE) { 1375 if (inp->inp_flags & INP_ONESBCAST) 1376 faddr.s_addr = INADDR_BROADCAST; 1377 if (cscov_partial) { 1378 if ((ui->ui_sum = in_cksum(m, sizeof(struct ip) + cscov)) == 0) 1379 ui->ui_sum = 0xffff; 1380 } else { 1381 if ((ui->ui_sum = in_cksum(m, sizeof(struct udpiphdr) + len)) == 0) 1382 ui->ui_sum = 0xffff; 1383 } 1384 } else if (V_udp_cksum) { 1385 if (inp->inp_flags & INP_ONESBCAST) 1386 faddr.s_addr = INADDR_BROADCAST; 1387 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr, 1388 htons((u_short)len + sizeof(struct udphdr) + pr)); 1389 m->m_pkthdr.csum_flags = CSUM_UDP; 1390 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 1391 } 1392 ((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len); 1393 ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl; /* XXX */ 1394 ((struct ip *)ui)->ip_tos = tos; /* XXX */ 1395 UDPSTAT_INC(udps_opackets); 1396 1397 if (unlock_udbinfo == UH_WLOCKED) 1398 INP_HASH_WUNLOCK(pcbinfo); 1399 else if (unlock_udbinfo == UH_RLOCKED) 1400 INP_HASH_RUNLOCK(pcbinfo); 1401 UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u); 1402 error = ip_output(m, inp->inp_options, NULL, ipflags, 1403 inp->inp_moptions, inp); 1404 if (unlock_udbinfo == UH_WLOCKED) 1405 INP_WUNLOCK(inp); 1406 else 1407 INP_RUNLOCK(inp); 1408 return (error); 1409 1410release: 1411 if (unlock_udbinfo == UH_WLOCKED) { 1412 INP_HASH_WUNLOCK(pcbinfo); 1413 INP_WUNLOCK(inp); 1414 } else if (unlock_udbinfo == UH_RLOCKED) { 1415 INP_HASH_RUNLOCK(pcbinfo); 1416 INP_RUNLOCK(inp); 1417 } else 1418 INP_RUNLOCK(inp); 1419 m_freem(m); 1420 return (error); 1421} 1422 1423 1424#if defined(IPSEC) && defined(IPSEC_NAT_T) 1425/* 1426 * Potentially decap ESP in UDP frame. Check for an ESP header 1427 * and optional marker; if present, strip the UDP header and 1428 * push the result through IPSec. 1429 * 1430 * Returns mbuf to be processed (potentially re-allocated) or 1431 * NULL if consumed and/or processed. 1432 */ 1433static struct mbuf * 1434udp4_espdecap(struct inpcb *inp, struct mbuf *m, int off) 1435{ 1436 size_t minlen, payload, skip, iphlen; 1437 caddr_t data; 1438 struct udpcb *up; 1439 struct m_tag *tag; 1440 struct udphdr *udphdr; 1441 struct ip *ip; 1442 1443 INP_RLOCK_ASSERT(inp); 1444 1445 /* 1446 * Pull up data so the longest case is contiguous: 1447 * IP/UDP hdr + non ESP marker + ESP hdr. 1448 */ 1449 minlen = off + sizeof(uint64_t) + sizeof(struct esp); 1450 if (minlen > m->m_pkthdr.len) 1451 minlen = m->m_pkthdr.len; 1452 if ((m = m_pullup(m, minlen)) == NULL) { 1453 IPSECSTAT_INC(ips_in_inval); 1454 return (NULL); /* Bypass caller processing. */ 1455 } 1456 data = mtod(m, caddr_t); /* Points to ip header. */ 1457 payload = m->m_len - off; /* Size of payload. */ 1458 1459 if (payload == 1 && data[off] == '\xff') 1460 return (m); /* NB: keepalive packet, no decap. */ 1461 1462 up = intoudpcb(inp); 1463 KASSERT(up != NULL, ("%s: udpcb NULL", __func__)); 1464 KASSERT((up->u_flags & UF_ESPINUDP_ALL) != 0, 1465 ("u_flags 0x%x", up->u_flags)); 1466 1467 /* 1468 * Check that the payload is large enough to hold an 1469 * ESP header and compute the amount of data to remove. 1470 * 1471 * NB: the caller has already done a pullup for us. 1472 * XXX can we assume alignment and eliminate bcopys? 1473 */ 1474 if (up->u_flags & UF_ESPINUDP_NON_IKE) { 1475 /* 1476 * draft-ietf-ipsec-nat-t-ike-0[01].txt and 1477 * draft-ietf-ipsec-udp-encaps-(00/)01.txt, ignoring 1478 * possible AH mode non-IKE marker+non-ESP marker 1479 * from draft-ietf-ipsec-udp-encaps-00.txt. 1480 */ 1481 uint64_t marker; 1482 1483 if (payload <= sizeof(uint64_t) + sizeof(struct esp)) 1484 return (m); /* NB: no decap. */ 1485 bcopy(data + off, &marker, sizeof(uint64_t)); 1486 if (marker != 0) /* Non-IKE marker. */ 1487 return (m); /* NB: no decap. */ 1488 skip = sizeof(uint64_t) + sizeof(struct udphdr); 1489 } else { 1490 uint32_t spi; 1491 1492 if (payload <= sizeof(struct esp)) { 1493 IPSECSTAT_INC(ips_in_inval); 1494 m_freem(m); 1495 return (NULL); /* Discard. */ 1496 } 1497 bcopy(data + off, &spi, sizeof(uint32_t)); 1498 if (spi == 0) /* Non-ESP marker. */ 1499 return (m); /* NB: no decap. */ 1500 skip = sizeof(struct udphdr); 1501 } 1502 1503 /* 1504 * Setup a PACKET_TAG_IPSEC_NAT_T_PORT tag to remember 1505 * the UDP ports. This is required if we want to select 1506 * the right SPD for multiple hosts behind same NAT. 1507 * 1508 * NB: ports are maintained in network byte order everywhere 1509 * in the NAT-T code. 1510 */ 1511 tag = m_tag_get(PACKET_TAG_IPSEC_NAT_T_PORTS, 1512 2 * sizeof(uint16_t), M_NOWAIT); 1513 if (tag == NULL) { 1514 IPSECSTAT_INC(ips_in_nomem); 1515 m_freem(m); 1516 return (NULL); /* Discard. */ 1517 } 1518 iphlen = off - sizeof(struct udphdr); 1519 udphdr = (struct udphdr *)(data + iphlen); 1520 ((uint16_t *)(tag + 1))[0] = udphdr->uh_sport; 1521 ((uint16_t *)(tag + 1))[1] = udphdr->uh_dport; 1522 m_tag_prepend(m, tag); 1523 1524 /* 1525 * Remove the UDP header (and possibly the non ESP marker) 1526 * IP header length is iphlen 1527 * Before: 1528 * <--- off ---> 1529 * +----+------+-----+ 1530 * | IP | UDP | ESP | 1531 * +----+------+-----+ 1532 * <-skip-> 1533 * After: 1534 * +----+-----+ 1535 * | IP | ESP | 1536 * +----+-----+ 1537 * <-skip-> 1538 */ 1539 ovbcopy(data, data + skip, iphlen); 1540 m_adj(m, skip); 1541 1542 ip = mtod(m, struct ip *); 1543 ip->ip_len = htons(ntohs(ip->ip_len) - skip); 1544 ip->ip_p = IPPROTO_ESP; 1545 1546 /* 1547 * We cannot yet update the cksums so clear any 1548 * h/w cksum flags as they are no longer valid. 1549 */ 1550 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) 1551 m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID|CSUM_PSEUDO_HDR); 1552 1553 (void) ipsec4_common_input(m, iphlen, ip->ip_p); 1554 return (NULL); /* NB: consumed, bypass processing. */ 1555} 1556#endif /* defined(IPSEC) && defined(IPSEC_NAT_T) */ 1557 1558static void 1559udp_abort(struct socket *so) 1560{ 1561 struct inpcb *inp; 1562 struct inpcbinfo *pcbinfo; 1563 1564 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1565 inp = sotoinpcb(so); 1566 KASSERT(inp != NULL, ("udp_abort: inp == NULL")); 1567 INP_WLOCK(inp); 1568 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1569 INP_HASH_WLOCK(pcbinfo); 1570 in_pcbdisconnect(inp); 1571 inp->inp_laddr.s_addr = INADDR_ANY; 1572 INP_HASH_WUNLOCK(pcbinfo); 1573 soisdisconnected(so); 1574 } 1575 INP_WUNLOCK(inp); 1576} 1577 1578static int 1579udp_attach(struct socket *so, int proto, struct thread *td) 1580{ 1581 struct inpcb *inp; 1582 struct inpcbinfo *pcbinfo; 1583 int error; 1584 1585 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1586 inp = sotoinpcb(so); 1587 KASSERT(inp == NULL, ("udp_attach: inp != NULL")); 1588 error = soreserve(so, udp_sendspace, udp_recvspace); 1589 if (error) 1590 return (error); 1591 INP_INFO_WLOCK(pcbinfo); 1592 error = in_pcballoc(so, pcbinfo); 1593 if (error) { 1594 INP_INFO_WUNLOCK(pcbinfo); 1595 return (error); 1596 } 1597 1598 inp = sotoinpcb(so); 1599 inp->inp_vflag |= INP_IPV4; 1600 inp->inp_ip_ttl = V_ip_defttl; 1601 1602 error = udp_newudpcb(inp); 1603 if (error) { 1604 in_pcbdetach(inp); 1605 in_pcbfree(inp); 1606 INP_INFO_WUNLOCK(pcbinfo); 1607 return (error); 1608 } 1609 1610 INP_WUNLOCK(inp); 1611 INP_INFO_WUNLOCK(pcbinfo); 1612 return (0); 1613} 1614#endif /* INET */ 1615 1616int 1617udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f) 1618{ 1619 struct inpcb *inp; 1620 struct udpcb *up; 1621 1622 KASSERT(so->so_type == SOCK_DGRAM, 1623 ("udp_set_kernel_tunneling: !dgram")); 1624 inp = sotoinpcb(so); 1625 KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL")); 1626 INP_WLOCK(inp); 1627 up = intoudpcb(inp); 1628 if (up->u_tun_func != NULL) { 1629 INP_WUNLOCK(inp); 1630 return (EBUSY); 1631 } 1632 up->u_tun_func = f; 1633 INP_WUNLOCK(inp); 1634 return (0); 1635} 1636 1637#ifdef INET 1638static int 1639udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 1640{ 1641 struct inpcb *inp; 1642 struct inpcbinfo *pcbinfo; 1643 int error; 1644 1645 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1646 inp = sotoinpcb(so); 1647 KASSERT(inp != NULL, ("udp_bind: inp == NULL")); 1648 INP_WLOCK(inp); 1649 INP_HASH_WLOCK(pcbinfo); 1650 error = in_pcbbind(inp, nam, td->td_ucred); 1651 INP_HASH_WUNLOCK(pcbinfo); 1652 INP_WUNLOCK(inp); 1653 return (error); 1654} 1655 1656static void 1657udp_close(struct socket *so) 1658{ 1659 struct inpcb *inp; 1660 struct inpcbinfo *pcbinfo; 1661 1662 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1663 inp = sotoinpcb(so); 1664 KASSERT(inp != NULL, ("udp_close: inp == NULL")); 1665 INP_WLOCK(inp); 1666 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1667 INP_HASH_WLOCK(pcbinfo); 1668 in_pcbdisconnect(inp); 1669 inp->inp_laddr.s_addr = INADDR_ANY; 1670 INP_HASH_WUNLOCK(pcbinfo); 1671 soisdisconnected(so); 1672 } 1673 INP_WUNLOCK(inp); 1674} 1675 1676static int 1677udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 1678{ 1679 struct inpcb *inp; 1680 struct inpcbinfo *pcbinfo; 1681 struct sockaddr_in *sin; 1682 int error; 1683 1684 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1685 inp = sotoinpcb(so); 1686 KASSERT(inp != NULL, ("udp_connect: inp == NULL")); 1687 INP_WLOCK(inp); 1688 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1689 INP_WUNLOCK(inp); 1690 return (EISCONN); 1691 } 1692 sin = (struct sockaddr_in *)nam; 1693 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); 1694 if (error != 0) { 1695 INP_WUNLOCK(inp); 1696 return (error); 1697 } 1698 INP_HASH_WLOCK(pcbinfo); 1699 error = in_pcbconnect(inp, nam, td->td_ucred); 1700 INP_HASH_WUNLOCK(pcbinfo); 1701 if (error == 0) 1702 soisconnected(so); 1703 INP_WUNLOCK(inp); 1704 return (error); 1705} 1706 1707static void 1708udp_detach(struct socket *so) 1709{ 1710 struct inpcb *inp; 1711 struct inpcbinfo *pcbinfo; 1712 struct udpcb *up; 1713 1714 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1715 inp = sotoinpcb(so); 1716 KASSERT(inp != NULL, ("udp_detach: inp == NULL")); 1717 KASSERT(inp->inp_faddr.s_addr == INADDR_ANY, 1718 ("udp_detach: not disconnected")); 1719 INP_INFO_WLOCK(pcbinfo); 1720 INP_WLOCK(inp); 1721 up = intoudpcb(inp); 1722 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1723 inp->inp_ppcb = NULL; 1724 in_pcbdetach(inp); 1725 in_pcbfree(inp); 1726 INP_INFO_WUNLOCK(pcbinfo); 1727 udp_discardcb(up); 1728} 1729 1730static int 1731udp_disconnect(struct socket *so) 1732{ 1733 struct inpcb *inp; 1734 struct inpcbinfo *pcbinfo; 1735 1736 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1737 inp = sotoinpcb(so); 1738 KASSERT(inp != NULL, ("udp_disconnect: inp == NULL")); 1739 INP_WLOCK(inp); 1740 if (inp->inp_faddr.s_addr == INADDR_ANY) { 1741 INP_WUNLOCK(inp); 1742 return (ENOTCONN); 1743 } 1744 INP_HASH_WLOCK(pcbinfo); 1745 in_pcbdisconnect(inp); 1746 inp->inp_laddr.s_addr = INADDR_ANY; 1747 INP_HASH_WUNLOCK(pcbinfo); 1748 SOCK_LOCK(so); 1749 so->so_state &= ~SS_ISCONNECTED; /* XXX */ 1750 SOCK_UNLOCK(so); 1751 INP_WUNLOCK(inp); 1752 return (0); 1753} 1754 1755static int 1756udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr, 1757 struct mbuf *control, struct thread *td) 1758{ 1759 struct inpcb *inp; 1760 1761 inp = sotoinpcb(so); 1762 KASSERT(inp != NULL, ("udp_send: inp == NULL")); 1763 return (udp_output(inp, m, addr, control, td)); 1764} 1765#endif /* INET */ 1766 1767int 1768udp_shutdown(struct socket *so) 1769{ 1770 struct inpcb *inp; 1771 1772 inp = sotoinpcb(so); 1773 KASSERT(inp != NULL, ("udp_shutdown: inp == NULL")); 1774 INP_WLOCK(inp); 1775 socantsendmore(so); 1776 INP_WUNLOCK(inp); 1777 return (0); 1778} 1779 1780#ifdef INET 1781struct pr_usrreqs udp_usrreqs = { 1782 .pru_abort = udp_abort, 1783 .pru_attach = udp_attach, 1784 .pru_bind = udp_bind, 1785 .pru_connect = udp_connect, 1786 .pru_control = in_control, 1787 .pru_detach = udp_detach, 1788 .pru_disconnect = udp_disconnect, 1789 .pru_peeraddr = in_getpeeraddr, 1790 .pru_send = udp_send, 1791 .pru_soreceive = soreceive_dgram, 1792 .pru_sosend = sosend_dgram, 1793 .pru_shutdown = udp_shutdown, 1794 .pru_sockaddr = in_getsockaddr, 1795 .pru_sosetlabel = in_pcbsosetlabel, 1796 .pru_close = udp_close, 1797}; 1798#endif /* INET */ 1799