udp_usrreq.c revision 272661
1132332Smarcel/*- 2132332Smarcel * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 3132332Smarcel * The Regents of the University of California. 4132332Smarcel * Copyright (c) 2008 Robert N. M. Watson 5132332Smarcel * Copyright (c) 2010-2011 Juniper Networks, Inc. 6132332Smarcel * Copyright (c) 2014 Kevin Lo 7132332Smarcel * All rights reserved. 8132332Smarcel * 9132332Smarcel * Portions of this software were developed by Robert N. M. Watson under 10132332Smarcel * contract to Juniper Networks, Inc. 11132332Smarcel * 12132332Smarcel * Redistribution and use in source and binary forms, with or without 13132332Smarcel * modification, are permitted provided that the following conditions 14132332Smarcel * are met: 15132332Smarcel * 1. Redistributions of source code must retain the above copyright 16132332Smarcel * notice, this list of conditions and the following disclaimer. 17132332Smarcel * 2. Redistributions in binary form must reproduce the above copyright 18132332Smarcel * notice, this list of conditions and the following disclaimer in the 19132332Smarcel * documentation and/or other materials provided with the distribution. 20132332Smarcel * 4. Neither the name of the University nor the names of its contributors 21132332Smarcel * may be used to endorse or promote products derived from this software 22132332Smarcel * without specific prior written permission. 23132332Smarcel * 24132332Smarcel * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25132332Smarcel * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26132332Smarcel * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27132332Smarcel * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28132332Smarcel * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29132332Smarcel * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30132332Smarcel * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31132332Smarcel * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32132332Smarcel * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33132332Smarcel * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34132332Smarcel * SUCH DAMAGE. 35132332Smarcel * 36132332Smarcel * @(#)udp_usrreq.c 8.6 (Berkeley) 5/23/95 37181059Smarcel */ 38181059Smarcel 39181059Smarcel#include <sys/cdefs.h> 40181059Smarcel__FBSDID("$FreeBSD: stable/10/sys/netinet/udp_usrreq.c 272661 2014-10-06 17:04:26Z tuexen $"); 41181059Smarcel 42181059Smarcel#include "opt_ipfw.h" 43132332Smarcel#include "opt_inet.h" 44181059Smarcel#include "opt_inet6.h" 45132332Smarcel#include "opt_ipsec.h" 46132332Smarcel#include "opt_kdtrace.h" 47132332Smarcel 48132332Smarcel#include <sys/param.h> 49132332Smarcel#include <sys/domain.h> 50132332Smarcel#include <sys/eventhandler.h> 51132332Smarcel#include <sys/jail.h> 52132332Smarcel#include <sys/kernel.h> 53132332Smarcel#include <sys/lock.h> 54132332Smarcel#include <sys/malloc.h> 55132332Smarcel#include <sys/mbuf.h> 56132332Smarcel#include <sys/priv.h> 57132332Smarcel#include <sys/proc.h> 58132332Smarcel#include <sys/protosw.h> 59132332Smarcel#include <sys/sdt.h> 60133802Sdavidxu#include <sys/signalvar.h> 61133802Sdavidxu#include <sys/socket.h> 62133802Sdavidxu#include <sys/socketvar.h> 63133802Sdavidxu#include <sys/sx.h> 64133802Sdavidxu#include <sys/sysctl.h> 65133802Sdavidxu#include <sys/syslog.h> 66133802Sdavidxu#include <sys/systm.h> 67133802Sdavidxu 68133802Sdavidxu#include <vm/uma.h> 69133802Sdavidxu 70133802Sdavidxu#include <net/if.h> 71133802Sdavidxu#include <net/route.h> 72133802Sdavidxu 73133802Sdavidxu#include <netinet/in.h> 74133802Sdavidxu#include <netinet/in_kdtrace.h> 75133802Sdavidxu#include <netinet/in_pcb.h> 76133802Sdavidxu#include <netinet/in_systm.h> 77158680Sdavidxu#include <netinet/in_var.h> 78158680Sdavidxu#include <netinet/ip.h> 79132332Smarcel#ifdef INET6 80132332Smarcel#include <netinet/ip6.h> 81132332Smarcel#endif 82132332Smarcel#include <netinet/ip_icmp.h> 83132332Smarcel#include <netinet/icmp_var.h> 84132332Smarcel#include <netinet/ip_var.h> 85132332Smarcel#include <netinet/ip_options.h> 86132332Smarcel#ifdef INET6 87132332Smarcel#include <netinet6/ip6_var.h> 88146818Sdfr#endif 89146818Sdfr#include <netinet/udp.h> 90146818Sdfr#include <netinet/udp_var.h> 91146818Sdfr#include <netinet/udplite.h> 92132332Smarcel 93132332Smarcel#ifdef IPSEC 94132332Smarcel#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 if (pr == IPPROTO_UDP) { 493 UDPSTAT_INC(udps_nosum); 494 } else { 495 /* UDPLite requires a checksum */ 496 /* XXX: What is the right UDPLite MIB counter here? */ 497 m_freem(m); 498 return; 499 } 500 } 501 502 pcbinfo = get_inpcbinfo(pr); 503 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 504 in_broadcast(ip->ip_dst, ifp)) { 505 struct inpcb *last; 506 struct inpcbhead *pcblist; 507 struct ip_moptions *imo; 508 509 INP_INFO_RLOCK(pcbinfo); 510 pcblist = get_pcblist(pr); 511 last = NULL; 512 LIST_FOREACH(inp, pcblist, inp_list) { 513 if (inp->inp_lport != uh->uh_dport) 514 continue; 515#ifdef INET6 516 if ((inp->inp_vflag & INP_IPV4) == 0) 517 continue; 518#endif 519 if (inp->inp_laddr.s_addr != INADDR_ANY && 520 inp->inp_laddr.s_addr != ip->ip_dst.s_addr) 521 continue; 522 if (inp->inp_faddr.s_addr != INADDR_ANY && 523 inp->inp_faddr.s_addr != ip->ip_src.s_addr) 524 continue; 525 if (inp->inp_fport != 0 && 526 inp->inp_fport != uh->uh_sport) 527 continue; 528 529 INP_RLOCK(inp); 530 531 /* 532 * XXXRW: Because we weren't holding either the inpcb 533 * or the hash lock when we checked for a match 534 * before, we should probably recheck now that the 535 * inpcb lock is held. 536 */ 537 538 /* 539 * Handle socket delivery policy for any-source 540 * and source-specific multicast. [RFC3678] 541 */ 542 imo = inp->inp_moptions; 543 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) { 544 struct sockaddr_in group; 545 int blocked; 546 if (imo == NULL) { 547 INP_RUNLOCK(inp); 548 continue; 549 } 550 bzero(&group, sizeof(struct sockaddr_in)); 551 group.sin_len = sizeof(struct sockaddr_in); 552 group.sin_family = AF_INET; 553 group.sin_addr = ip->ip_dst; 554 555 blocked = imo_multi_filter(imo, ifp, 556 (struct sockaddr *)&group, 557 (struct sockaddr *)&udp_in); 558 if (blocked != MCAST_PASS) { 559 if (blocked == MCAST_NOTGMEMBER) 560 IPSTAT_INC(ips_notmember); 561 if (blocked == MCAST_NOTSMEMBER || 562 blocked == MCAST_MUTED) 563 UDPSTAT_INC(udps_filtermcast); 564 INP_RUNLOCK(inp); 565 continue; 566 } 567 } 568 if (last != NULL) { 569 struct mbuf *n; 570 571 n = m_copy(m, 0, M_COPYALL); 572 udp_append(last, ip, n, iphlen, &udp_in); 573 INP_RUNLOCK(last); 574 } 575 last = inp; 576 /* 577 * Don't look for additional matches if this one does 578 * not have either the SO_REUSEPORT or SO_REUSEADDR 579 * socket options set. This heuristic avoids 580 * searching through all pcbs in the common case of a 581 * non-shared port. It assumes that an application 582 * will never clear these options after setting them. 583 */ 584 if ((last->inp_socket->so_options & 585 (SO_REUSEPORT|SO_REUSEADDR)) == 0) 586 break; 587 } 588 589 if (last == NULL) { 590 /* 591 * No matching pcb found; discard datagram. (No need 592 * to send an ICMP Port Unreachable for a broadcast 593 * or multicast datgram.) 594 */ 595 UDPSTAT_INC(udps_noportbcast); 596 if (inp) 597 INP_RUNLOCK(inp); 598 INP_INFO_RUNLOCK(pcbinfo); 599 goto badunlocked; 600 } 601 udp_append(last, ip, m, iphlen, &udp_in); 602 INP_RUNLOCK(last); 603 INP_INFO_RUNLOCK(pcbinfo); 604 return; 605 } 606 607 /* 608 * Locate pcb for datagram. 609 */ 610 611 /* 612 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 613 */ 614 if ((m->m_flags & M_IP_NEXTHOP) && 615 (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) { 616 struct sockaddr_in *next_hop; 617 618 next_hop = (struct sockaddr_in *)(fwd_tag + 1); 619 620 /* 621 * Transparently forwarded. Pretend to be the destination. 622 * Already got one like this? 623 */ 624 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport, 625 ip->ip_dst, uh->uh_dport, INPLOOKUP_RLOCKPCB, ifp, m); 626 if (!inp) { 627 /* 628 * It's new. Try to find the ambushing socket. 629 * Because we've rewritten the destination address, 630 * any hardware-generated hash is ignored. 631 */ 632 inp = in_pcblookup(pcbinfo, ip->ip_src, 633 uh->uh_sport, next_hop->sin_addr, 634 next_hop->sin_port ? htons(next_hop->sin_port) : 635 uh->uh_dport, INPLOOKUP_WILDCARD | 636 INPLOOKUP_RLOCKPCB, ifp); 637 } 638 /* Remove the tag from the packet. We don't need it anymore. */ 639 m_tag_delete(m, fwd_tag); 640 m->m_flags &= ~M_IP_NEXTHOP; 641 } else 642 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport, 643 ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD | 644 INPLOOKUP_RLOCKPCB, ifp, m); 645 if (inp == NULL) { 646 if (udp_log_in_vain) { 647 char buf[4*sizeof "123"]; 648 649 strcpy(buf, inet_ntoa(ip->ip_dst)); 650 log(LOG_INFO, 651 "Connection attempt to UDP %s:%d from %s:%d\n", 652 buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src), 653 ntohs(uh->uh_sport)); 654 } 655 UDPSTAT_INC(udps_noport); 656 if (m->m_flags & (M_BCAST | M_MCAST)) { 657 UDPSTAT_INC(udps_noportbcast); 658 goto badunlocked; 659 } 660 if (V_udp_blackhole) 661 goto badunlocked; 662 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0) 663 goto badunlocked; 664 *ip = save_ip; 665 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0); 666 return; 667 } 668 669 /* 670 * Check the minimum TTL for socket. 671 */ 672 INP_RLOCK_ASSERT(inp); 673 if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) { 674 INP_RUNLOCK(inp); 675 m_freem(m); 676 return; 677 } 678 if (cscov_partial) { 679 struct udpcb *up; 680 681 up = intoudpcb(inp); 682 if (up->u_rxcslen == 0 || up->u_rxcslen > len) { 683 INP_RUNLOCK(inp); 684 m_freem(m); 685 return; 686 } 687 } 688 689 UDP_PROBE(receive, NULL, inp, ip, inp, uh); 690 udp_append(inp, ip, m, iphlen, &udp_in); 691 INP_RUNLOCK(inp); 692 return; 693 694badunlocked: 695 m_freem(m); 696} 697#endif /* INET */ 698 699/* 700 * Notify a udp user of an asynchronous error; just wake up so that they can 701 * collect error status. 702 */ 703struct inpcb * 704udp_notify(struct inpcb *inp, int errno) 705{ 706 707 /* 708 * While udp_ctlinput() always calls udp_notify() with a read lock 709 * when invoking it directly, in_pcbnotifyall() currently uses write 710 * locks due to sharing code with TCP. For now, accept either a read 711 * or a write lock, but a read lock is sufficient. 712 */ 713 INP_LOCK_ASSERT(inp); 714 715 inp->inp_socket->so_error = errno; 716 sorwakeup(inp->inp_socket); 717 sowwakeup(inp->inp_socket); 718 return (inp); 719} 720 721#ifdef INET 722static void 723udp_common_ctlinput(int cmd, struct sockaddr *sa, void *vip, 724 struct inpcbinfo *pcbinfo) 725{ 726 struct ip *ip = vip; 727 struct udphdr *uh; 728 struct in_addr faddr; 729 struct inpcb *inp; 730 731 faddr = ((struct sockaddr_in *)sa)->sin_addr; 732 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 733 return; 734 735 /* 736 * Redirects don't need to be handled up here. 737 */ 738 if (PRC_IS_REDIRECT(cmd)) 739 return; 740 741 /* 742 * Hostdead is ugly because it goes linearly through all PCBs. 743 * 744 * XXX: We never get this from ICMP, otherwise it makes an excellent 745 * DoS attack on machines with many connections. 746 */ 747 if (cmd == PRC_HOSTDEAD) 748 ip = NULL; 749 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 750 return; 751 if (ip != NULL) { 752 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 753 inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport, 754 ip->ip_src, uh->uh_sport, INPLOOKUP_RLOCKPCB, NULL); 755 if (inp != NULL) { 756 INP_RLOCK_ASSERT(inp); 757 if (inp->inp_socket != NULL) { 758 udp_notify(inp, inetctlerrmap[cmd]); 759 } 760 INP_RUNLOCK(inp); 761 } 762 } else 763 in_pcbnotifyall(pcbinfo, faddr, inetctlerrmap[cmd], 764 udp_notify); 765} 766void 767udp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 768{ 769 770 return (udp_common_ctlinput(cmd, sa, vip, &V_udbinfo)); 771} 772 773void 774udplite_ctlinput(int cmd, struct sockaddr *sa, void *vip) 775{ 776 777 return (udp_common_ctlinput(cmd, sa, vip, &V_ulitecbinfo)); 778} 779#endif /* INET */ 780 781static int 782udp_pcblist(SYSCTL_HANDLER_ARGS) 783{ 784 int error, i, n; 785 struct inpcb *inp, **inp_list; 786 inp_gen_t gencnt; 787 struct xinpgen xig; 788 789 /* 790 * The process of preparing the PCB list is too time-consuming and 791 * resource-intensive to repeat twice on every request. 792 */ 793 if (req->oldptr == 0) { 794 n = V_udbinfo.ipi_count; 795 n += imax(n / 8, 10); 796 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb); 797 return (0); 798 } 799 800 if (req->newptr != 0) 801 return (EPERM); 802 803 /* 804 * OK, now we're committed to doing something. 805 */ 806 INP_INFO_RLOCK(&V_udbinfo); 807 gencnt = V_udbinfo.ipi_gencnt; 808 n = V_udbinfo.ipi_count; 809 INP_INFO_RUNLOCK(&V_udbinfo); 810 811 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 812 + n * sizeof(struct xinpcb)); 813 if (error != 0) 814 return (error); 815 816 xig.xig_len = sizeof xig; 817 xig.xig_count = n; 818 xig.xig_gen = gencnt; 819 xig.xig_sogen = so_gencnt; 820 error = SYSCTL_OUT(req, &xig, sizeof xig); 821 if (error) 822 return (error); 823 824 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 825 if (inp_list == 0) 826 return (ENOMEM); 827 828 INP_INFO_RLOCK(&V_udbinfo); 829 for (inp = LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n; 830 inp = LIST_NEXT(inp, inp_list)) { 831 INP_WLOCK(inp); 832 if (inp->inp_gencnt <= gencnt && 833 cr_canseeinpcb(req->td->td_ucred, inp) == 0) { 834 in_pcbref(inp); 835 inp_list[i++] = inp; 836 } 837 INP_WUNLOCK(inp); 838 } 839 INP_INFO_RUNLOCK(&V_udbinfo); 840 n = i; 841 842 error = 0; 843 for (i = 0; i < n; i++) { 844 inp = inp_list[i]; 845 INP_RLOCK(inp); 846 if (inp->inp_gencnt <= gencnt) { 847 struct xinpcb xi; 848 849 bzero(&xi, sizeof(xi)); 850 xi.xi_len = sizeof xi; 851 /* XXX should avoid extra copy */ 852 bcopy(inp, &xi.xi_inp, sizeof *inp); 853 if (inp->inp_socket) 854 sotoxsocket(inp->inp_socket, &xi.xi_socket); 855 xi.xi_inp.inp_gencnt = inp->inp_gencnt; 856 INP_RUNLOCK(inp); 857 error = SYSCTL_OUT(req, &xi, sizeof xi); 858 } else 859 INP_RUNLOCK(inp); 860 } 861 INP_INFO_WLOCK(&V_udbinfo); 862 for (i = 0; i < n; i++) { 863 inp = inp_list[i]; 864 INP_RLOCK(inp); 865 if (!in_pcbrele_rlocked(inp)) 866 INP_RUNLOCK(inp); 867 } 868 INP_INFO_WUNLOCK(&V_udbinfo); 869 870 if (!error) { 871 /* 872 * Give the user an updated idea of our state. If the 873 * generation differs from what we told her before, she knows 874 * that something happened while we were processing this 875 * request, and it might be necessary to retry. 876 */ 877 INP_INFO_RLOCK(&V_udbinfo); 878 xig.xig_gen = V_udbinfo.ipi_gencnt; 879 xig.xig_sogen = so_gencnt; 880 xig.xig_count = V_udbinfo.ipi_count; 881 INP_INFO_RUNLOCK(&V_udbinfo); 882 error = SYSCTL_OUT(req, &xig, sizeof xig); 883 } 884 free(inp_list, M_TEMP); 885 return (error); 886} 887 888SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, 889 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0, 890 udp_pcblist, "S,xinpcb", "List of active UDP sockets"); 891 892#ifdef INET 893static int 894udp_getcred(SYSCTL_HANDLER_ARGS) 895{ 896 struct xucred xuc; 897 struct sockaddr_in addrs[2]; 898 struct inpcb *inp; 899 int error; 900 901 error = priv_check(req->td, PRIV_NETINET_GETCRED); 902 if (error) 903 return (error); 904 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 905 if (error) 906 return (error); 907 inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port, 908 addrs[0].sin_addr, addrs[0].sin_port, 909 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL); 910 if (inp != NULL) { 911 INP_RLOCK_ASSERT(inp); 912 if (inp->inp_socket == NULL) 913 error = ENOENT; 914 if (error == 0) 915 error = cr_canseeinpcb(req->td->td_ucred, inp); 916 if (error == 0) 917 cru2x(inp->inp_cred, &xuc); 918 INP_RUNLOCK(inp); 919 } else 920 error = ENOENT; 921 if (error == 0) 922 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 923 return (error); 924} 925 926SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred, 927 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 928 udp_getcred, "S,xucred", "Get the xucred of a UDP connection"); 929#endif /* INET */ 930 931int 932udp_ctloutput(struct socket *so, struct sockopt *sopt) 933{ 934 struct inpcb *inp; 935 struct udpcb *up; 936 int isudplite, error, optval; 937 938 error = 0; 939 isudplite = (so->so_proto->pr_protocol == IPPROTO_UDPLITE) ? 1 : 0; 940 inp = sotoinpcb(so); 941 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 942 INP_WLOCK(inp); 943 if (sopt->sopt_level != so->so_proto->pr_protocol) { 944#ifdef INET6 945 if (INP_CHECK_SOCKAF(so, AF_INET6)) { 946 INP_WUNLOCK(inp); 947 error = ip6_ctloutput(so, sopt); 948 } 949#endif 950#if defined(INET) && defined(INET6) 951 else 952#endif 953#ifdef INET 954 { 955 INP_WUNLOCK(inp); 956 error = ip_ctloutput(so, sopt); 957 } 958#endif 959 return (error); 960 } 961 962 switch (sopt->sopt_dir) { 963 case SOPT_SET: 964 switch (sopt->sopt_name) { 965 case UDP_ENCAP: 966 INP_WUNLOCK(inp); 967 error = sooptcopyin(sopt, &optval, sizeof optval, 968 sizeof optval); 969 if (error) 970 break; 971 inp = sotoinpcb(so); 972 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 973 INP_WLOCK(inp); 974#ifdef IPSEC_NAT_T 975 up = intoudpcb(inp); 976 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 977#endif 978 switch (optval) { 979 case 0: 980 /* Clear all UDP encap. */ 981#ifdef IPSEC_NAT_T 982 up->u_flags &= ~UF_ESPINUDP_ALL; 983#endif 984 break; 985#ifdef IPSEC_NAT_T 986 case UDP_ENCAP_ESPINUDP: 987 case UDP_ENCAP_ESPINUDP_NON_IKE: 988 up->u_flags &= ~UF_ESPINUDP_ALL; 989 if (optval == UDP_ENCAP_ESPINUDP) 990 up->u_flags |= UF_ESPINUDP; 991 else if (optval == UDP_ENCAP_ESPINUDP_NON_IKE) 992 up->u_flags |= UF_ESPINUDP_NON_IKE; 993 break; 994#endif 995 default: 996 error = EINVAL; 997 break; 998 } 999 INP_WUNLOCK(inp); 1000 break; 1001 case UDPLITE_SEND_CSCOV: 1002 case UDPLITE_RECV_CSCOV: 1003 if (!isudplite) { 1004 INP_WUNLOCK(inp); 1005 error = ENOPROTOOPT; 1006 break; 1007 } 1008 INP_WUNLOCK(inp); 1009 error = sooptcopyin(sopt, &optval, sizeof(optval), 1010 sizeof(optval)); 1011 if (error != 0) 1012 break; 1013 inp = sotoinpcb(so); 1014 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 1015 INP_WLOCK(inp); 1016 up = intoudpcb(inp); 1017 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1018 if ((optval != 0 && optval < 8) || (optval > 65535)) { 1019 INP_WUNLOCK(inp); 1020 error = EINVAL; 1021 break; 1022 } 1023 if (sopt->sopt_name == UDPLITE_SEND_CSCOV) 1024 up->u_txcslen = optval; 1025 else 1026 up->u_rxcslen = optval; 1027 INP_WUNLOCK(inp); 1028 break; 1029 default: 1030 INP_WUNLOCK(inp); 1031 error = ENOPROTOOPT; 1032 break; 1033 } 1034 break; 1035 case SOPT_GET: 1036 switch (sopt->sopt_name) { 1037#ifdef IPSEC_NAT_T 1038 case UDP_ENCAP: 1039 up = intoudpcb(inp); 1040 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1041 optval = up->u_flags & UF_ESPINUDP_ALL; 1042 INP_WUNLOCK(inp); 1043 error = sooptcopyout(sopt, &optval, sizeof optval); 1044 break; 1045#endif 1046 case UDPLITE_SEND_CSCOV: 1047 case UDPLITE_RECV_CSCOV: 1048 if (!isudplite) { 1049 INP_WUNLOCK(inp); 1050 error = ENOPROTOOPT; 1051 break; 1052 } 1053 up = intoudpcb(inp); 1054 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1055 if (sopt->sopt_name == UDPLITE_SEND_CSCOV) 1056 optval = up->u_txcslen; 1057 else 1058 optval = up->u_rxcslen; 1059 INP_WUNLOCK(inp); 1060 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1061 break; 1062 default: 1063 INP_WUNLOCK(inp); 1064 error = ENOPROTOOPT; 1065 break; 1066 } 1067 break; 1068 } 1069 return (error); 1070} 1071 1072#ifdef INET 1073#define UH_WLOCKED 2 1074#define UH_RLOCKED 1 1075#define UH_UNLOCKED 0 1076static int 1077udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr, 1078 struct mbuf *control, struct thread *td) 1079{ 1080 struct udpiphdr *ui; 1081 int len = m->m_pkthdr.len; 1082 struct in_addr faddr, laddr; 1083 struct cmsghdr *cm; 1084 struct inpcbinfo *pcbinfo; 1085 struct sockaddr_in *sin, src; 1086 int cscov_partial = 0; 1087 int error = 0; 1088 int ipflags; 1089 u_short fport, lport; 1090 int unlock_udbinfo; 1091 u_char tos; 1092 uint8_t pr; 1093 uint16_t cscov = 0; 1094 1095 /* 1096 * udp_output() may need to temporarily bind or connect the current 1097 * inpcb. As such, we don't know up front whether we will need the 1098 * pcbinfo lock or not. Do any work to decide what is needed up 1099 * front before acquiring any locks. 1100 */ 1101 if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) { 1102 if (control) 1103 m_freem(control); 1104 m_freem(m); 1105 return (EMSGSIZE); 1106 } 1107 1108 src.sin_family = 0; 1109 INP_RLOCK(inp); 1110 tos = inp->inp_ip_tos; 1111 if (control != NULL) { 1112 /* 1113 * XXX: Currently, we assume all the optional information is 1114 * stored in a single mbuf. 1115 */ 1116 if (control->m_next) { 1117 INP_RUNLOCK(inp); 1118 m_freem(control); 1119 m_freem(m); 1120 return (EINVAL); 1121 } 1122 for (; control->m_len > 0; 1123 control->m_data += CMSG_ALIGN(cm->cmsg_len), 1124 control->m_len -= CMSG_ALIGN(cm->cmsg_len)) { 1125 cm = mtod(control, struct cmsghdr *); 1126 if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0 1127 || cm->cmsg_len > control->m_len) { 1128 error = EINVAL; 1129 break; 1130 } 1131 if (cm->cmsg_level != IPPROTO_IP) 1132 continue; 1133 1134 switch (cm->cmsg_type) { 1135 case IP_SENDSRCADDR: 1136 if (cm->cmsg_len != 1137 CMSG_LEN(sizeof(struct in_addr))) { 1138 error = EINVAL; 1139 break; 1140 } 1141 bzero(&src, sizeof(src)); 1142 src.sin_family = AF_INET; 1143 src.sin_len = sizeof(src); 1144 src.sin_port = inp->inp_lport; 1145 src.sin_addr = 1146 *(struct in_addr *)CMSG_DATA(cm); 1147 break; 1148 1149 case IP_TOS: 1150 if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) { 1151 error = EINVAL; 1152 break; 1153 } 1154 tos = *(u_char *)CMSG_DATA(cm); 1155 break; 1156 1157 default: 1158 error = ENOPROTOOPT; 1159 break; 1160 } 1161 if (error) 1162 break; 1163 } 1164 m_freem(control); 1165 } 1166 if (error) { 1167 INP_RUNLOCK(inp); 1168 m_freem(m); 1169 return (error); 1170 } 1171 1172 /* 1173 * Depending on whether or not the application has bound or connected 1174 * the socket, we may have to do varying levels of work. The optimal 1175 * case is for a connected UDP socket, as a global lock isn't 1176 * required at all. 1177 * 1178 * In order to decide which we need, we require stability of the 1179 * inpcb binding, which we ensure by acquiring a read lock on the 1180 * inpcb. This doesn't strictly follow the lock order, so we play 1181 * the trylock and retry game; note that we may end up with more 1182 * conservative locks than required the second time around, so later 1183 * assertions have to accept that. Further analysis of the number of 1184 * misses under contention is required. 1185 * 1186 * XXXRW: Check that hash locking update here is correct. 1187 */ 1188 pr = inp->inp_socket->so_proto->pr_protocol; 1189 pcbinfo = get_inpcbinfo(pr); 1190 sin = (struct sockaddr_in *)addr; 1191 if (sin != NULL && 1192 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) { 1193 INP_RUNLOCK(inp); 1194 INP_WLOCK(inp); 1195 INP_HASH_WLOCK(pcbinfo); 1196 unlock_udbinfo = UH_WLOCKED; 1197 } else if ((sin != NULL && ( 1198 (sin->sin_addr.s_addr == INADDR_ANY) || 1199 (sin->sin_addr.s_addr == INADDR_BROADCAST) || 1200 (inp->inp_laddr.s_addr == INADDR_ANY) || 1201 (inp->inp_lport == 0))) || 1202 (src.sin_family == AF_INET)) { 1203 INP_HASH_RLOCK(pcbinfo); 1204 unlock_udbinfo = UH_RLOCKED; 1205 } else 1206 unlock_udbinfo = UH_UNLOCKED; 1207 1208 /* 1209 * If the IP_SENDSRCADDR control message was specified, override the 1210 * source address for this datagram. Its use is invalidated if the 1211 * address thus specified is incomplete or clobbers other inpcbs. 1212 */ 1213 laddr = inp->inp_laddr; 1214 lport = inp->inp_lport; 1215 if (src.sin_family == AF_INET) { 1216 INP_HASH_LOCK_ASSERT(pcbinfo); 1217 if ((lport == 0) || 1218 (laddr.s_addr == INADDR_ANY && 1219 src.sin_addr.s_addr == INADDR_ANY)) { 1220 error = EINVAL; 1221 goto release; 1222 } 1223 error = in_pcbbind_setup(inp, (struct sockaddr *)&src, 1224 &laddr.s_addr, &lport, td->td_ucred); 1225 if (error) 1226 goto release; 1227 } 1228 1229 /* 1230 * If a UDP socket has been connected, then a local address/port will 1231 * have been selected and bound. 1232 * 1233 * If a UDP socket has not been connected to, then an explicit 1234 * destination address must be used, in which case a local 1235 * address/port may not have been selected and bound. 1236 */ 1237 if (sin != NULL) { 1238 INP_LOCK_ASSERT(inp); 1239 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1240 error = EISCONN; 1241 goto release; 1242 } 1243 1244 /* 1245 * Jail may rewrite the destination address, so let it do 1246 * that before we use it. 1247 */ 1248 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); 1249 if (error) 1250 goto release; 1251 1252 /* 1253 * If a local address or port hasn't yet been selected, or if 1254 * the destination address needs to be rewritten due to using 1255 * a special INADDR_ constant, invoke in_pcbconnect_setup() 1256 * to do the heavy lifting. Once a port is selected, we 1257 * commit the binding back to the socket; we also commit the 1258 * binding of the address if in jail. 1259 * 1260 * If we already have a valid binding and we're not 1261 * requesting a destination address rewrite, use a fast path. 1262 */ 1263 if (inp->inp_laddr.s_addr == INADDR_ANY || 1264 inp->inp_lport == 0 || 1265 sin->sin_addr.s_addr == INADDR_ANY || 1266 sin->sin_addr.s_addr == INADDR_BROADCAST) { 1267 INP_HASH_LOCK_ASSERT(pcbinfo); 1268 error = in_pcbconnect_setup(inp, addr, &laddr.s_addr, 1269 &lport, &faddr.s_addr, &fport, NULL, 1270 td->td_ucred); 1271 if (error) 1272 goto release; 1273 1274 /* 1275 * XXXRW: Why not commit the port if the address is 1276 * !INADDR_ANY? 1277 */ 1278 /* Commit the local port if newly assigned. */ 1279 if (inp->inp_laddr.s_addr == INADDR_ANY && 1280 inp->inp_lport == 0) { 1281 INP_WLOCK_ASSERT(inp); 1282 INP_HASH_WLOCK_ASSERT(pcbinfo); 1283 /* 1284 * Remember addr if jailed, to prevent 1285 * rebinding. 1286 */ 1287 if (prison_flag(td->td_ucred, PR_IP4)) 1288 inp->inp_laddr = laddr; 1289 inp->inp_lport = lport; 1290 if (in_pcbinshash(inp) != 0) { 1291 inp->inp_lport = 0; 1292 error = EAGAIN; 1293 goto release; 1294 } 1295 inp->inp_flags |= INP_ANONPORT; 1296 } 1297 } else { 1298 faddr = sin->sin_addr; 1299 fport = sin->sin_port; 1300 } 1301 } else { 1302 INP_LOCK_ASSERT(inp); 1303 faddr = inp->inp_faddr; 1304 fport = inp->inp_fport; 1305 if (faddr.s_addr == INADDR_ANY) { 1306 error = ENOTCONN; 1307 goto release; 1308 } 1309 } 1310 1311 /* 1312 * Calculate data length and get a mbuf for UDP, IP, and possible 1313 * link-layer headers. Immediate slide the data pointer back forward 1314 * since we won't use that space at this layer. 1315 */ 1316 M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT); 1317 if (m == NULL) { 1318 error = ENOBUFS; 1319 goto release; 1320 } 1321 m->m_data += max_linkhdr; 1322 m->m_len -= max_linkhdr; 1323 m->m_pkthdr.len -= max_linkhdr; 1324 1325 /* 1326 * Fill in mbuf with extended UDP header and addresses and length put 1327 * into network format. 1328 */ 1329 ui = mtod(m, struct udpiphdr *); 1330 bzero(ui->ui_x1, sizeof(ui->ui_x1)); /* XXX still needed? */ 1331 ui->ui_pr = pr; 1332 ui->ui_src = laddr; 1333 ui->ui_dst = faddr; 1334 ui->ui_sport = lport; 1335 ui->ui_dport = fport; 1336 ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr)); 1337 if (pr == IPPROTO_UDPLITE) { 1338 struct udpcb *up; 1339 uint16_t plen; 1340 1341 up = intoudpcb(inp); 1342 cscov = up->u_txcslen; 1343 plen = (u_short)len + sizeof(struct udphdr); 1344 if (cscov >= plen) 1345 cscov = 0; 1346 ui->ui_len = htons(plen); 1347 ui->ui_ulen = htons(cscov); 1348 /* 1349 * For UDP-Lite, checksum coverage length of zero means 1350 * the entire UDPLite packet is covered by the checksum. 1351 */ 1352 cscov_partial = (cscov == 0) ? 0 : 1; 1353 } else 1354 ui->ui_v = IPVERSION << 4; 1355 1356 /* 1357 * Set the Don't Fragment bit in the IP header. 1358 */ 1359 if (inp->inp_flags & INP_DONTFRAG) { 1360 struct ip *ip; 1361 1362 ip = (struct ip *)&ui->ui_i; 1363 ip->ip_off |= htons(IP_DF); 1364 } 1365 1366 ipflags = 0; 1367 if (inp->inp_socket->so_options & SO_DONTROUTE) 1368 ipflags |= IP_ROUTETOIF; 1369 if (inp->inp_socket->so_options & SO_BROADCAST) 1370 ipflags |= IP_ALLOWBROADCAST; 1371 if (inp->inp_flags & INP_ONESBCAST) 1372 ipflags |= IP_SENDONES; 1373 1374#ifdef MAC 1375 mac_inpcb_create_mbuf(inp, m); 1376#endif 1377 1378 /* 1379 * Set up checksum and output datagram. 1380 */ 1381 ui->ui_sum = 0; 1382 if (pr == IPPROTO_UDPLITE) { 1383 if (inp->inp_flags & INP_ONESBCAST) 1384 faddr.s_addr = INADDR_BROADCAST; 1385 if (cscov_partial) { 1386 if ((ui->ui_sum = in_cksum(m, sizeof(struct ip) + cscov)) == 0) 1387 ui->ui_sum = 0xffff; 1388 } else { 1389 if ((ui->ui_sum = in_cksum(m, sizeof(struct udpiphdr) + len)) == 0) 1390 ui->ui_sum = 0xffff; 1391 } 1392 } else if (V_udp_cksum) { 1393 if (inp->inp_flags & INP_ONESBCAST) 1394 faddr.s_addr = INADDR_BROADCAST; 1395 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr, 1396 htons((u_short)len + sizeof(struct udphdr) + pr)); 1397 m->m_pkthdr.csum_flags = CSUM_UDP; 1398 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 1399 } 1400 ((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len); 1401 ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl; /* XXX */ 1402 ((struct ip *)ui)->ip_tos = tos; /* XXX */ 1403 UDPSTAT_INC(udps_opackets); 1404 1405 if (unlock_udbinfo == UH_WLOCKED) 1406 INP_HASH_WUNLOCK(pcbinfo); 1407 else if (unlock_udbinfo == UH_RLOCKED) 1408 INP_HASH_RUNLOCK(pcbinfo); 1409 UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u); 1410 error = ip_output(m, inp->inp_options, NULL, ipflags, 1411 inp->inp_moptions, inp); 1412 if (unlock_udbinfo == UH_WLOCKED) 1413 INP_WUNLOCK(inp); 1414 else 1415 INP_RUNLOCK(inp); 1416 return (error); 1417 1418release: 1419 if (unlock_udbinfo == UH_WLOCKED) { 1420 INP_HASH_WUNLOCK(pcbinfo); 1421 INP_WUNLOCK(inp); 1422 } else if (unlock_udbinfo == UH_RLOCKED) { 1423 INP_HASH_RUNLOCK(pcbinfo); 1424 INP_RUNLOCK(inp); 1425 } else 1426 INP_RUNLOCK(inp); 1427 m_freem(m); 1428 return (error); 1429} 1430 1431 1432#if defined(IPSEC) && defined(IPSEC_NAT_T) 1433/* 1434 * Potentially decap ESP in UDP frame. Check for an ESP header 1435 * and optional marker; if present, strip the UDP header and 1436 * push the result through IPSec. 1437 * 1438 * Returns mbuf to be processed (potentially re-allocated) or 1439 * NULL if consumed and/or processed. 1440 */ 1441static struct mbuf * 1442udp4_espdecap(struct inpcb *inp, struct mbuf *m, int off) 1443{ 1444 size_t minlen, payload, skip, iphlen; 1445 caddr_t data; 1446 struct udpcb *up; 1447 struct m_tag *tag; 1448 struct udphdr *udphdr; 1449 struct ip *ip; 1450 1451 INP_RLOCK_ASSERT(inp); 1452 1453 /* 1454 * Pull up data so the longest case is contiguous: 1455 * IP/UDP hdr + non ESP marker + ESP hdr. 1456 */ 1457 minlen = off + sizeof(uint64_t) + sizeof(struct esp); 1458 if (minlen > m->m_pkthdr.len) 1459 minlen = m->m_pkthdr.len; 1460 if ((m = m_pullup(m, minlen)) == NULL) { 1461 IPSECSTAT_INC(ips_in_inval); 1462 return (NULL); /* Bypass caller processing. */ 1463 } 1464 data = mtod(m, caddr_t); /* Points to ip header. */ 1465 payload = m->m_len - off; /* Size of payload. */ 1466 1467 if (payload == 1 && data[off] == '\xff') 1468 return (m); /* NB: keepalive packet, no decap. */ 1469 1470 up = intoudpcb(inp); 1471 KASSERT(up != NULL, ("%s: udpcb NULL", __func__)); 1472 KASSERT((up->u_flags & UF_ESPINUDP_ALL) != 0, 1473 ("u_flags 0x%x", up->u_flags)); 1474 1475 /* 1476 * Check that the payload is large enough to hold an 1477 * ESP header and compute the amount of data to remove. 1478 * 1479 * NB: the caller has already done a pullup for us. 1480 * XXX can we assume alignment and eliminate bcopys? 1481 */ 1482 if (up->u_flags & UF_ESPINUDP_NON_IKE) { 1483 /* 1484 * draft-ietf-ipsec-nat-t-ike-0[01].txt and 1485 * draft-ietf-ipsec-udp-encaps-(00/)01.txt, ignoring 1486 * possible AH mode non-IKE marker+non-ESP marker 1487 * from draft-ietf-ipsec-udp-encaps-00.txt. 1488 */ 1489 uint64_t marker; 1490 1491 if (payload <= sizeof(uint64_t) + sizeof(struct esp)) 1492 return (m); /* NB: no decap. */ 1493 bcopy(data + off, &marker, sizeof(uint64_t)); 1494 if (marker != 0) /* Non-IKE marker. */ 1495 return (m); /* NB: no decap. */ 1496 skip = sizeof(uint64_t) + sizeof(struct udphdr); 1497 } else { 1498 uint32_t spi; 1499 1500 if (payload <= sizeof(struct esp)) { 1501 IPSECSTAT_INC(ips_in_inval); 1502 m_freem(m); 1503 return (NULL); /* Discard. */ 1504 } 1505 bcopy(data + off, &spi, sizeof(uint32_t)); 1506 if (spi == 0) /* Non-ESP marker. */ 1507 return (m); /* NB: no decap. */ 1508 skip = sizeof(struct udphdr); 1509 } 1510 1511 /* 1512 * Setup a PACKET_TAG_IPSEC_NAT_T_PORT tag to remember 1513 * the UDP ports. This is required if we want to select 1514 * the right SPD for multiple hosts behind same NAT. 1515 * 1516 * NB: ports are maintained in network byte order everywhere 1517 * in the NAT-T code. 1518 */ 1519 tag = m_tag_get(PACKET_TAG_IPSEC_NAT_T_PORTS, 1520 2 * sizeof(uint16_t), M_NOWAIT); 1521 if (tag == NULL) { 1522 IPSECSTAT_INC(ips_in_nomem); 1523 m_freem(m); 1524 return (NULL); /* Discard. */ 1525 } 1526 iphlen = off - sizeof(struct udphdr); 1527 udphdr = (struct udphdr *)(data + iphlen); 1528 ((uint16_t *)(tag + 1))[0] = udphdr->uh_sport; 1529 ((uint16_t *)(tag + 1))[1] = udphdr->uh_dport; 1530 m_tag_prepend(m, tag); 1531 1532 /* 1533 * Remove the UDP header (and possibly the non ESP marker) 1534 * IP header length is iphlen 1535 * Before: 1536 * <--- off ---> 1537 * +----+------+-----+ 1538 * | IP | UDP | ESP | 1539 * +----+------+-----+ 1540 * <-skip-> 1541 * After: 1542 * +----+-----+ 1543 * | IP | ESP | 1544 * +----+-----+ 1545 * <-skip-> 1546 */ 1547 ovbcopy(data, data + skip, iphlen); 1548 m_adj(m, skip); 1549 1550 ip = mtod(m, struct ip *); 1551 ip->ip_len = htons(ntohs(ip->ip_len) - skip); 1552 ip->ip_p = IPPROTO_ESP; 1553 1554 /* 1555 * We cannot yet update the cksums so clear any 1556 * h/w cksum flags as they are no longer valid. 1557 */ 1558 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) 1559 m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID|CSUM_PSEUDO_HDR); 1560 1561 (void) ipsec4_common_input(m, iphlen, ip->ip_p); 1562 return (NULL); /* NB: consumed, bypass processing. */ 1563} 1564#endif /* defined(IPSEC) && defined(IPSEC_NAT_T) */ 1565 1566static void 1567udp_abort(struct socket *so) 1568{ 1569 struct inpcb *inp; 1570 struct inpcbinfo *pcbinfo; 1571 1572 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1573 inp = sotoinpcb(so); 1574 KASSERT(inp != NULL, ("udp_abort: inp == NULL")); 1575 INP_WLOCK(inp); 1576 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1577 INP_HASH_WLOCK(pcbinfo); 1578 in_pcbdisconnect(inp); 1579 inp->inp_laddr.s_addr = INADDR_ANY; 1580 INP_HASH_WUNLOCK(pcbinfo); 1581 soisdisconnected(so); 1582 } 1583 INP_WUNLOCK(inp); 1584} 1585 1586static int 1587udp_attach(struct socket *so, int proto, struct thread *td) 1588{ 1589 struct inpcb *inp; 1590 struct inpcbinfo *pcbinfo; 1591 int error; 1592 1593 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1594 inp = sotoinpcb(so); 1595 KASSERT(inp == NULL, ("udp_attach: inp != NULL")); 1596 error = soreserve(so, udp_sendspace, udp_recvspace); 1597 if (error) 1598 return (error); 1599 INP_INFO_WLOCK(pcbinfo); 1600 error = in_pcballoc(so, pcbinfo); 1601 if (error) { 1602 INP_INFO_WUNLOCK(pcbinfo); 1603 return (error); 1604 } 1605 1606 inp = sotoinpcb(so); 1607 inp->inp_vflag |= INP_IPV4; 1608 inp->inp_ip_ttl = V_ip_defttl; 1609 1610 error = udp_newudpcb(inp); 1611 if (error) { 1612 in_pcbdetach(inp); 1613 in_pcbfree(inp); 1614 INP_INFO_WUNLOCK(pcbinfo); 1615 return (error); 1616 } 1617 1618 INP_WUNLOCK(inp); 1619 INP_INFO_WUNLOCK(pcbinfo); 1620 return (0); 1621} 1622#endif /* INET */ 1623 1624int 1625udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f) 1626{ 1627 struct inpcb *inp; 1628 struct udpcb *up; 1629 1630 KASSERT(so->so_type == SOCK_DGRAM, 1631 ("udp_set_kernel_tunneling: !dgram")); 1632 inp = sotoinpcb(so); 1633 KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL")); 1634 INP_WLOCK(inp); 1635 up = intoudpcb(inp); 1636 if (up->u_tun_func != NULL) { 1637 INP_WUNLOCK(inp); 1638 return (EBUSY); 1639 } 1640 up->u_tun_func = f; 1641 INP_WUNLOCK(inp); 1642 return (0); 1643} 1644 1645#ifdef INET 1646static int 1647udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 1648{ 1649 struct inpcb *inp; 1650 struct inpcbinfo *pcbinfo; 1651 int error; 1652 1653 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1654 inp = sotoinpcb(so); 1655 KASSERT(inp != NULL, ("udp_bind: inp == NULL")); 1656 INP_WLOCK(inp); 1657 INP_HASH_WLOCK(pcbinfo); 1658 error = in_pcbbind(inp, nam, td->td_ucred); 1659 INP_HASH_WUNLOCK(pcbinfo); 1660 INP_WUNLOCK(inp); 1661 return (error); 1662} 1663 1664static void 1665udp_close(struct socket *so) 1666{ 1667 struct inpcb *inp; 1668 struct inpcbinfo *pcbinfo; 1669 1670 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1671 inp = sotoinpcb(so); 1672 KASSERT(inp != NULL, ("udp_close: inp == NULL")); 1673 INP_WLOCK(inp); 1674 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1675 INP_HASH_WLOCK(pcbinfo); 1676 in_pcbdisconnect(inp); 1677 inp->inp_laddr.s_addr = INADDR_ANY; 1678 INP_HASH_WUNLOCK(pcbinfo); 1679 soisdisconnected(so); 1680 } 1681 INP_WUNLOCK(inp); 1682} 1683 1684static int 1685udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 1686{ 1687 struct inpcb *inp; 1688 struct inpcbinfo *pcbinfo; 1689 struct sockaddr_in *sin; 1690 int error; 1691 1692 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1693 inp = sotoinpcb(so); 1694 KASSERT(inp != NULL, ("udp_connect: inp == NULL")); 1695 INP_WLOCK(inp); 1696 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1697 INP_WUNLOCK(inp); 1698 return (EISCONN); 1699 } 1700 sin = (struct sockaddr_in *)nam; 1701 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); 1702 if (error != 0) { 1703 INP_WUNLOCK(inp); 1704 return (error); 1705 } 1706 INP_HASH_WLOCK(pcbinfo); 1707 error = in_pcbconnect(inp, nam, td->td_ucred); 1708 INP_HASH_WUNLOCK(pcbinfo); 1709 if (error == 0) 1710 soisconnected(so); 1711 INP_WUNLOCK(inp); 1712 return (error); 1713} 1714 1715static void 1716udp_detach(struct socket *so) 1717{ 1718 struct inpcb *inp; 1719 struct inpcbinfo *pcbinfo; 1720 struct udpcb *up; 1721 1722 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1723 inp = sotoinpcb(so); 1724 KASSERT(inp != NULL, ("udp_detach: inp == NULL")); 1725 KASSERT(inp->inp_faddr.s_addr == INADDR_ANY, 1726 ("udp_detach: not disconnected")); 1727 INP_INFO_WLOCK(pcbinfo); 1728 INP_WLOCK(inp); 1729 up = intoudpcb(inp); 1730 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1731 inp->inp_ppcb = NULL; 1732 in_pcbdetach(inp); 1733 in_pcbfree(inp); 1734 INP_INFO_WUNLOCK(pcbinfo); 1735 udp_discardcb(up); 1736} 1737 1738static int 1739udp_disconnect(struct socket *so) 1740{ 1741 struct inpcb *inp; 1742 struct inpcbinfo *pcbinfo; 1743 1744 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1745 inp = sotoinpcb(so); 1746 KASSERT(inp != NULL, ("udp_disconnect: inp == NULL")); 1747 INP_WLOCK(inp); 1748 if (inp->inp_faddr.s_addr == INADDR_ANY) { 1749 INP_WUNLOCK(inp); 1750 return (ENOTCONN); 1751 } 1752 INP_HASH_WLOCK(pcbinfo); 1753 in_pcbdisconnect(inp); 1754 inp->inp_laddr.s_addr = INADDR_ANY; 1755 INP_HASH_WUNLOCK(pcbinfo); 1756 SOCK_LOCK(so); 1757 so->so_state &= ~SS_ISCONNECTED; /* XXX */ 1758 SOCK_UNLOCK(so); 1759 INP_WUNLOCK(inp); 1760 return (0); 1761} 1762 1763static int 1764udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr, 1765 struct mbuf *control, struct thread *td) 1766{ 1767 struct inpcb *inp; 1768 1769 inp = sotoinpcb(so); 1770 KASSERT(inp != NULL, ("udp_send: inp == NULL")); 1771 return (udp_output(inp, m, addr, control, td)); 1772} 1773#endif /* INET */ 1774 1775int 1776udp_shutdown(struct socket *so) 1777{ 1778 struct inpcb *inp; 1779 1780 inp = sotoinpcb(so); 1781 KASSERT(inp != NULL, ("udp_shutdown: inp == NULL")); 1782 INP_WLOCK(inp); 1783 socantsendmore(so); 1784 INP_WUNLOCK(inp); 1785 return (0); 1786} 1787 1788#ifdef INET 1789struct pr_usrreqs udp_usrreqs = { 1790 .pru_abort = udp_abort, 1791 .pru_attach = udp_attach, 1792 .pru_bind = udp_bind, 1793 .pru_connect = udp_connect, 1794 .pru_control = in_control, 1795 .pru_detach = udp_detach, 1796 .pru_disconnect = udp_disconnect, 1797 .pru_peeraddr = in_getpeeraddr, 1798 .pru_send = udp_send, 1799 .pru_soreceive = soreceive_dgram, 1800 .pru_sosend = sosend_dgram, 1801 .pru_shutdown = udp_shutdown, 1802 .pru_sockaddr = in_getsockaddr, 1803 .pru_sosetlabel = in_pcbsosetlabel, 1804 .pru_close = udp_close, 1805}; 1806#endif /* INET */ 1807